<div class="xblock xblock-public_view xblock-public_view-vertical" data-block-type="vertical" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="VerticalStudentView" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@vertical+block@9953a37a801b4d85a6a756c4172c03da" data-course-id="course-v1:MITx+HST.953x+3T2020">
<h2 class="hd hd-2 unit-title">Introduction</h2>
<div class="vert-mod">
<div class="vert vert-0" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@f34de1fce09844f299f6a817efa038e6">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@f34de1fce09844f299f6a817efa038e6" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<p>In late 2019, a group of researchers found widespread bias in a broadly deployed algorithm for predicting the medical complexity of patients, with the algorithm assigning lower risk scores to Black patients with a similar level of need as white patients. At the centre of this problem was the developers seemingly innocuous use of healthcare spending as a proxy for medical care needs - despite intentionally excluding race as a factor in the algorithm, the researchers failed to account for the systemic factors which lead Black patients to receive lower cost care.</p>
<p>This case is prototypical of the risks facing machine learning in healthcare. While these technologies have the potential to bring about revolutions in care cost and quality, they also have the potential to exacerbate inequities by improving care to a lesser degree for, or even directly harming, marginalized groups.</p>
<p><strong>OUTLINE:</strong></p>
<p>We will begin with a brief overview of the concept of fairness, and different ways in which it can be considered:</p>
<ol type="1">
<li value="1">What is fairness?</li>
<ul type="disc">
<li>In this section, we will explore the regulatory and ethical foundations of fairness through the lenses of disparate treatment and disparate outcome. A deep understanding of these concepts is important, as fairness is case-specific and can at times involve a conflict between principles.</li>
</ul>
</ol>
<p>We will then walk through the different avenues whereby bias can influence machine learning for healthcare. A specific project may intersect with one or all three.</p>
<ol type="1">
<li value="1">Fairness in Healthcare:</li>
<ul type="disc">
<li>In this section, we will explore the existing biases in healthcare in the United States, and the ways in which this bias already harms patients. It is important to understand this because models uncritically trained on this system will only act to reproduce it, without active efforts to the contrary.</li>
</ul>
<li>Fairness in Machine Learning:</li>
<ul type="disc">
<li>In this section, we will explore the issue of bias in machine learning more broadly, both in the abstract and as applied to case studies such as sentencing and loan applications. We will also touch on technical methods to interrogate and reduce bias.</li>
</ul>
<li>Fairness in Machine Learning for Healthcare:</li>
<ul type="disc">
<li>In this section, we will explore the intersection between healthcare bias and machine learning bias, with a particular emphasis upon the real-world deployment of algorithms. While some of the problems with both healthcare and machine learning can be solved in isolation, most require consideration of how these intersecting contexts overlap</li>
</ul>
</ol>
<p>It is not inevitable, however, that machine learning will be a bias-inducing force in healthcare. If done effectively and carefully, the technology has substantial potential as an equity-promoting force. This leads into our final section:</p>
<ol type="1">
<li value="5">Ensuring Healthcare Data Science Works for All:</li>
<ul type="disc">
<li>In this section, we will explore the potential for machine learning to act as an inequity-reducing force in healthcare by enabling active, data-driven interrogation of biased processes. We will explore recommendations for achieving this in practice, both at the systemic level and in the context of individual projects.</li>
</ul>
</ol>
<p>This workshop is a little different from the others in that there will be no coding exercises or follow-along practical components outside of the content questions. We encourage students to reflect actively on the content, and to explore further in the linked sources. The message we hope to put forth with this workshop is that fairness in machine learning for healthcare is difficult and complex, but ultimately worthwhile and an ethical imperative. We hope that this introduction will equip you with the skills and knowledge to understand how fairness relates to your own clinical data science work.</p>
</div>
</div>
</div>
</div>
<div class="xblock xblock-public_view xblock-public_view-vertical" data-block-type="vertical" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="VerticalStudentView" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@vertical+block@81fe8b3084294a92a96a452b1a779905" data-course-id="course-v1:MITx+HST.953x+3T2020">
<h2 class="hd hd-2 unit-title">Introduction to Fairness</h2>
<div class="vert-mod">
<div class="vert vert-0" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@289052a74289477984f632b24659786a">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@289052a74289477984f632b24659786a" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<p><strong>Intro:</strong></p>
<p>In striving to understand fairness and bias in the contexts of healthcare and machine learning, we must begin by understanding these concepts more generally. As defined by Mirriam-Webster, to be "fair" is to be "marked by impartiality and honest : free from self-interest, prejudice, or favoritism", and "bias" is "an inclination of temperament or outlook, <em>especially</em> a personal and sometimes unreasoned judgment". While these definitions track with the colloquial sense and usage of the term, their broadness and subjectivity makes them difficult to translate into an actionable form. Working actionably with fairness and bias, especially when legal and regulatory burdens are involved, requires further depth of analysis and formalization. </p>
<p>Within this section we will explore a framework examining multiple types of fairness, explore how this relates to United States law, examine the possibility for proxy variables to introduce bias in a "colorblind" system, and briefly touch upon related factors such as diversity, explanation, and feedback loops.</p>
<p><strong>What is Fairness?</strong></p>
<p>(Much is owed to <a href="https://fairmlbook.org/tutorial1.html" target="[object Object]">Barocas and Hardt (2017)</a>)</p>
<p><strong><em>Parity, Preference, Treatment, and Outcome</em></strong></p>
<p>An excellent approach to fairness in machine learning is formalized in the work of <a href="https://arxiv.org/pdf/1710.03184.pdf" target="[object Object]">Gajane and Pechenizkiy (2018)</a>, and builds upon two important questions:</p>
<ul>
<li><strong>Parity or preference?</strong></li>
<ul>
<li>Does fairness mean achieving statistical parity, or satisfying the preferences of individuals within various groups? (For an excellent further exploration of this, see <a href="https://arxiv.org/pdf/1707.00010.pdf" target="[object Object]">Zafar et al</a>)</li>
<li>For example, imagine an algorithm designed to allocate a gender-balanced group of 100 people either pizza or spaghetti for dinner, with only 50 of each dinner option available. If 60% of men prefer spaghetti and 60% of women prefer pizza, what is a fair way of allocating these foods?</li>
<ul>
<li>A parity-preserving approach would strive for a representative division regardless of preference, assigning 25 pizza and 25 spaghetti to each group.</li>
<li>A preference-preserving approach would strive to optimize for the preference of users, assigning 30 pizza and 20 spaghetti to the women, and assigning 20 pizza and 30 spaghetti to the men.</li>
</ul>
</ul>
<li><strong>Disparate treatment or disparate outcome?</strong></li>
<ul>
<li>Fairness oriented around avoiding <strong>disparate treatment</strong> is <em>procedural fairness</em>, and emphasises the concept of equality of opportunity. All individuals with similar characteristics should receive similar decisions based upon the system.</li>
<li>Fairness oriented around avoiding <strong>disparate impact </strong>is <em>distributive justice</em>, and emphasises the concept of equality of outcome. All groups of individuals (as defined by specific identifiable characteristics deemed non-permissible sources of discrimination between groups, such as race or sex) should receive similar outcomes from the system</li>
</ul>
</ul>
<p>Optimally, all four of these conditions might be equally satisfied by a system. There are, however, cases (such as that of affirmative action) in which these various conditions may come into tension. It is an open, context-specific, and complex ethical question to decide what is "right" in such circumstances. <strong>It is critical to realize that fairness is a value-laden concept that lacks a single universal definition or set of satisfiable criteria</strong>. Rather, there are a series of different approaches to defining fairness that may be most applicable in specific contexts</p>
<p><strong><em>Selected Approaches to Fairness:</em></strong></p>
<ul>
<li><strong>Fairness through unawareness:</strong></li>
<ul>
<li><em>"A predictor is said to achieve fairness through unawareness of protected attributes are not explicitly used in the prediction process"</em></li>
<li>PROS:</li>
<ul>
<li>Guaranteed not to make an explicit judgment based on a sensitive attribute</li>
</ul>
<li>CONS:</li>
<ul>
<li>Sometimes the attribute <em>is</em> important to the decision-making process (e.g. different symptoms based on the sex of a patient)</li>
<li>Unfairness (such as racism) <a href="https://www.brennancenter.org/sites/default/files/legacy/Justice/09%20Racial%20Blindsight.pdf" target="[object Object]">can</a><em> and <a href="https://www.amazon.com/Racism-without-Racists-Color-Blind-Persistence/dp/1442276231" target="[object Object]">does</a> happen in "color-blind settings"</em>, and unawareness can mask and hide this.</li>
</ul>
</ul>
<li><strong>Individual fairness:</strong></li>
<ul>
<li>Formal Definition: <img height="167" width="1127" src="/assets/courseware/v1/9b0840a85d54801c7be9f8b865121051/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Indiv_fairness_definition.png" alt="Formal definition of individual fairness" /></li>
<li>Simplified Definition:</li>
<ul>
<li>A predictor satisfies individual fairness if it produces similar outputs for similar individuals.</li>
</ul>
<li>PROS:</li>
<ul>
<li>Consistent between individuals</li>
</ul>
<li>CONS:</li>
<ul>
<li>"Similarity" can be difficult and nebulous to define, especially when multiple overlapping metrics are involved.</li>
<li>There is no way to "zoom out" and check whether group fairness is also satisfied under these definitions</li>
</ul>
</ul>
<li><strong>Group Fairness:</strong></li>
<ul>
<li>Formal Definition: <img height="211" width="878" src="/assets/courseware/v1/457a08b1993fec542420870ad86a130c/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Group_Fairness.png" alt="Formal definition of group fairness" /></li>
<li>Simplified Definition:</li>
<ul>
<li>A predictor satisfies group fairness if it predicts a particular outcome for individuals across groups with similar probability (e.g. a healthcare algorithm that assigns 5% of white patients as eligible for dialysis also assigns 5% of Black patients as eligible for dialysis)</li>
</ul>
<li>PROS:</li>
<ul>
<li>Satisfies notions of avoiding penalizing or harming a specific group</li>
<li>Aligns with concerns about group equity (e.g. the aforementioned algorithm ensuring that similar dialysis spending is granted to both Black and white patients)</li>
</ul>
<li>CONS:</li>
<ul>
<li>There is no requirement to pick the "most qualified" within each group (that is, group fairness can be achieved in a way that violated individual fairness, e.g. the aforementioned algorithm would satisfy this even if it randomly chooses 5% of black patients to send for dialysis)</li>
<li>This can be legitimately less accurate and potentially inappropriate if base rates of a label differ (e.g. if more Black patients are actually in need of dialysis, ensuring group parity would lead to those patients not receiving necessary treatment)</li>
</ul>
</ul>
<li><strong>Equality of Opportunity:</strong></li>
<ul>
<li>Formal Definition: <img height="206" width="864" src="/assets/courseware/v1/798c8c62507ad4e1aa334ab8ac56fd14/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Equality_of_Opportunity.png" alt="Formal definition of equality of opportunity" /></li>
<li>Simplified Definition:</li>
<ul>
<li>Equality of opportunity is satisfied if the probability of an outcome is the same across different classes (e.g. if a man has a 40% chance of being hired for a job, so does a woman with similar experience)</li>
</ul>
<li>PROS:</li>
<ul>
<li>The true positive rate is the same for all groups (also known as "calibration")</li>
</ul>
<li>CONS:</li>
<ul>
<li>If the base rates of the labels are different, there would be different false positive rates (and this could be inappropriate) (e.g. if a higher proportion of women are qualified for the job, more unqualified men may be hired).</li>
</ul>
</ul>
<li><strong>Counterfactual Fairness:</strong></li>
<ul>
<li>Formal Definition: <img height="290" width="1228" src="/assets/courseware/v1/1911b52e1fa3377621ab4e60f370c541/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Counterfactual_Fairness.png" alt="Formal definition of counterfactual fairness" /></li>
<li>Simplified Definition:</li>
<ul>
<li>A predictor satisfies counterfactual fairness if its decisions for a person who is a member of group X are the same as they would have been were that person a member of group Y (e.g. the algorithm makes the same decision for a Black woman as it would have if she were a white woman).</li>
</ul>
<li>PROS:</li>
<ul>
<li>Aligns with an intuitive and aspirational sense of fairness without being colorblind</li>
</ul>
<li>CONS:</li>
<ul>
<li>Lots of assumptions have to be made</li>
<li>Different factors are interrelated, and the world is too complex to build models that <em>truly </em>estimate the counterfactual (e.g. does a racial counter-factual also influence a person's inter-generational family wealth?)</li>
<li>Defining the "similarly situated" member of the non-minority group can be difficult (e.g. is the similar comparator between racial groups on an entrance exam someone who has achieved the same numerical score, or someone who is the same number of standard deviations above the population average in their racial group?)</li>
<li>Intersectional identities further the complexity (e.g. comparing a Black woman to a white man is different from both comparing her to a Black man and to a white woman due to the intersection of her identities)</li>
</ul>
</ul>
</ul>
<p><strong>Fairness and The Law:</strong></p>
<ul>
<li>Fairness is not only a moral imperative, but a legal one as well. Domains including credit, education, employment, housing are all subject to regulation under United States law, and organizations working in these domains have to meet certain standards of non-discrimination (which can vary by domain).</li>
<li>Importantly, these rules apply with respect to a defined set of <strong>"protected classes"</strong>, including: race, color, sex, religion, national origin, citizenship, age, pregnancy, familial status, disability status, veteran status, and genetic information. Discrimination with respect to, say, the left-handedness or right-handedness of a person would not be covered by one of these protected classes under the law.</li>
<li>In the context of machine learning for healthcare, various regulations necessitate non-discrimination in the provision and quality of healthcare.</li>
</ul>
<p><strong>Unfairness and Proxy Variables:</strong></p>
<ul>
<li>As alluded to in our previous discussion of "fairness through unawareness", <a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3347959" target="[object Object]">it is entirely possible for an algorithm that has zero knowledge of the protected characteristic (e.g. sex) to be unfair and discriminatory.</a> Sometimes this occurs through the inclusion of <strong>"proxy variables"</strong>, which are closely associated with the protected characteristic. The connection may be clear and explicit (e.g. the connection between hair length and gender), or it may be more subtle (e.g. the use of zip code as a proxy for race).</li>
<li>While such proxy discrimination can be deliberate and intentional, it can also arise unintentionally in the use of machine learning algorithms - especially given the accuracy with which certain model types are able to infer hidden characteristics from their associations with other variables.</li>
<li>An algorithm trained on a dataset that embeds systemic bias will learn and perpetuate its patterns even if blind to the protected class. For example, algorithm to predict hiring in a sexist company with a history of discrimination against women could simply learn "long hair" as a predictor of "not hire" (assuming the stereotype that women tend to have longer hair than men). Even if all such variables are purged, systemic bias may influence other aspects such as interviewers' subjective scores. There is concern that algorithms can hide such discrimination beneath a veil of algorithmic objectivity, and in fact make it more difficult to weed out.</li>
</ul>
<p><strong>Framing, and Other Factors:</strong></p>
<ul>
<li>Understanding the complexity of fairness as a concept is essential to dealing with it effectively in the healthcare machine learning context. A computer science approach can tend to be reductionist, and regard disparate situations as similar (such as the examples in Table 1). Each of these situations, however, brings different historical context and different impact upon those involved.</li>
<li>There are other factors to consider outside of the fairness / unfairness that might be captured in input and output pairs, such as:</li>
<ul>
<li><em>Diversity:</em></li>
<ul>
<li>Are there benefits from having a variety of people represented in the college context? Is the same true in triage? Can the sum of a group exceed its parts?</li>
<li>Are there risks to a "winner-take-all" market, wherein one group consistently is selected by an algorithm?</li>
</ul>
<li><em>Feedback Loops:</em></li>
<ul>
<li>The outputs of an algorithm can have further influence on its inputs. For example, an algorithm that deprioritizes women for organizational advancement due to a lower projected level of career success creates a self-fulfilling and cyclical prophecy.</li>
</ul>
<li><em>Explanation:</em></li>
<ul>
<li>To what extent are users entitled to an explanation for decisions? Are the burdens of explanation different in different contexts (e.g. not hiring someone vs. firing them)?</li>
</ul>
</ul>
</ul>
</div>
</div>
<div class="vert vert-1" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@244caabf45bb446aa31b3fdd5c9ca40e">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@244caabf45bb446aa31b3fdd5c9ca40e" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_244caabf45bb446aa31b3fdd5c9ca40e" class="problems-wrapper" role="group"
aria-labelledby="244caabf45bb446aa31b3fdd5c9ca40e-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@244caabf45bb446aa31b3fdd5c9ca40e" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@244caabf45bb446aa31b3fdd5c9ca40e/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="244caabf45bb446aa31b3fdd5c9ca40e-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@244caabf45bb446aa31b3fdd5c9ca40e-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@244caabf45bb446aa31b3fdd5c9ca40e-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>A sentencing algorithm is constructed with the primary goal that convicted persons with similar characteristics receive similar sentences. Which of the following definitions of fairness (according to Gajane and Pechenizkiy, 2018) does this align with?</p>
<div class="choicegroup capa_inputtype" id="inputtype_244caabf45bb446aa31b3fdd5c9ca40e_2_1">
<fieldset aria-describedby="status_244caabf45bb446aa31b3fdd5c9ca40e_2_1">
<div class="field">
<input type="radio" name="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1" id="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="244caabf45bb446aa31b3fdd5c9ca40e_2_1-choice_0-label" for="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_244caabf45bb446aa31b3fdd5c9ca40e_2_1"> Equality of Opportunity
</label>
</div>
<div class="field">
<input type="radio" name="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1" id="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="244caabf45bb446aa31b3fdd5c9ca40e_2_1-choice_1-label" for="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_244caabf45bb446aa31b3fdd5c9ca40e_2_1"> Group Fairness
</label>
</div>
<div class="field">
<input type="radio" name="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1" id="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1_choice_2" class="field-input input-radio" value="choice_2"/><label id="244caabf45bb446aa31b3fdd5c9ca40e_2_1-choice_2-label" for="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1_choice_2" class="response-label field-label label-inline" aria-describedby="status_244caabf45bb446aa31b3fdd5c9ca40e_2_1"> Individual Fairness
</label>
</div>
<div class="field">
<input type="radio" name="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1" id="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1_choice_3" class="field-input input-radio" value="choice_3"/><label id="244caabf45bb446aa31b3fdd5c9ca40e_2_1-choice_3-label" for="input_244caabf45bb446aa31b3fdd5c9ca40e_2_1_choice_3" class="response-label field-label label-inline" aria-describedby="status_244caabf45bb446aa31b3fdd5c9ca40e_2_1"> Counterfactual Fairness
</label>
</div>
<span id="answer_244caabf45bb446aa31b3fdd5c9ca40e_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_244caabf45bb446aa31b3fdd5c9ca40e_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_244caabf45bb446aa31b3fdd5c9ca40e_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_244caabf45bb446aa31b3fdd5c9ca40e" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_244caabf45bb446aa31b3fdd5c9ca40e">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="244caabf45bb446aa31b3fdd5c9ca40e-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="244caabf45bb446aa31b3fdd5c9ca40e-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="244caabf45bb446aa31b3fdd5c9ca40e-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-2" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@0bbfbf31f0d64922b027a4eead3bc436">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@0bbfbf31f0d64922b027a4eead3bc436" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_0bbfbf31f0d64922b027a4eead3bc436" class="problems-wrapper" role="group"
aria-labelledby="0bbfbf31f0d64922b027a4eead3bc436-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@0bbfbf31f0d64922b027a4eead3bc436" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@0bbfbf31f0d64922b027a4eead3bc436/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="0bbfbf31f0d64922b027a4eead3bc436-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@0bbfbf31f0d64922b027a4eead3bc436-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@0bbfbf31f0d64922b027a4eead3bc436-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>Which of the following is NOT a reason that counterfactual fairness is difficult to implement in practice?</p>
<div class="choicegroup capa_inputtype" id="inputtype_0bbfbf31f0d64922b027a4eead3bc436_2_1">
<fieldset aria-describedby="status_0bbfbf31f0d64922b027a4eead3bc436_2_1">
<div class="field">
<input type="radio" name="input_0bbfbf31f0d64922b027a4eead3bc436_2_1" id="input_0bbfbf31f0d64922b027a4eead3bc436_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="0bbfbf31f0d64922b027a4eead3bc436_2_1-choice_0-label" for="input_0bbfbf31f0d64922b027a4eead3bc436_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_0bbfbf31f0d64922b027a4eead3bc436_2_1"> Lack of data for direct comparisons between members of majority and minority groups
</label>
</div>
<div class="field">
<input type="radio" name="input_0bbfbf31f0d64922b027a4eead3bc436_2_1" id="input_0bbfbf31f0d64922b027a4eead3bc436_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="0bbfbf31f0d64922b027a4eead3bc436_2_1-choice_1-label" for="input_0bbfbf31f0d64922b027a4eead3bc436_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_0bbfbf31f0d64922b027a4eead3bc436_2_1"> Illegality of using race as a factor in evaluating decisions
</label>
</div>
<div class="field">
<input type="radio" name="input_0bbfbf31f0d64922b027a4eead3bc436_2_1" id="input_0bbfbf31f0d64922b027a4eead3bc436_2_1_choice_2" class="field-input input-radio" value="choice_2"/><label id="0bbfbf31f0d64922b027a4eead3bc436_2_1-choice_2-label" for="input_0bbfbf31f0d64922b027a4eead3bc436_2_1_choice_2" class="response-label field-label label-inline" aria-describedby="status_0bbfbf31f0d64922b027a4eead3bc436_2_1"> Difficulty disentangling intersections between multiple identity factors
</label>
</div>
<div class="field">
<input type="radio" name="input_0bbfbf31f0d64922b027a4eead3bc436_2_1" id="input_0bbfbf31f0d64922b027a4eead3bc436_2_1_choice_3" class="field-input input-radio" value="choice_3"/><label id="0bbfbf31f0d64922b027a4eead3bc436_2_1-choice_3-label" for="input_0bbfbf31f0d64922b027a4eead3bc436_2_1_choice_3" class="response-label field-label label-inline" aria-describedby="status_0bbfbf31f0d64922b027a4eead3bc436_2_1"> Complexity of modelling all factors present in the real world
</label>
</div>
<span id="answer_0bbfbf31f0d64922b027a4eead3bc436_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_0bbfbf31f0d64922b027a4eead3bc436_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_0bbfbf31f0d64922b027a4eead3bc436_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_0bbfbf31f0d64922b027a4eead3bc436" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_0bbfbf31f0d64922b027a4eead3bc436">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="0bbfbf31f0d64922b027a4eead3bc436-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="0bbfbf31f0d64922b027a4eead3bc436-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="0bbfbf31f0d64922b027a4eead3bc436-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-3" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@82842e89afe04bd0aa1877a5d870368a">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@82842e89afe04bd0aa1877a5d870368a" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_82842e89afe04bd0aa1877a5d870368a" class="problems-wrapper" role="group"
aria-labelledby="82842e89afe04bd0aa1877a5d870368a-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@82842e89afe04bd0aa1877a5d870368a" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@82842e89afe04bd0aa1877a5d870368a/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="82842e89afe04bd0aa1877a5d870368a-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@82842e89afe04bd0aa1877a5d870368a-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@82842e89afe04bd0aa1877a5d870368a-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>Which of the following is NOT a protected class under United States law?</p>
<div class="choicegroup capa_inputtype" id="inputtype_82842e89afe04bd0aa1877a5d870368a_2_1">
<fieldset aria-describedby="status_82842e89afe04bd0aa1877a5d870368a_2_1">
<div class="field">
<input type="radio" name="input_82842e89afe04bd0aa1877a5d870368a_2_1" id="input_82842e89afe04bd0aa1877a5d870368a_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="82842e89afe04bd0aa1877a5d870368a_2_1-choice_0-label" for="input_82842e89afe04bd0aa1877a5d870368a_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_82842e89afe04bd0aa1877a5d870368a_2_1"> Religion
</label>
</div>
<div class="field">
<input type="radio" name="input_82842e89afe04bd0aa1877a5d870368a_2_1" id="input_82842e89afe04bd0aa1877a5d870368a_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="82842e89afe04bd0aa1877a5d870368a_2_1-choice_1-label" for="input_82842e89afe04bd0aa1877a5d870368a_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_82842e89afe04bd0aa1877a5d870368a_2_1"> Pregnancy Status
</label>
</div>
<div class="field">
<input type="radio" name="input_82842e89afe04bd0aa1877a5d870368a_2_1" id="input_82842e89afe04bd0aa1877a5d870368a_2_1_choice_2" class="field-input input-radio" value="choice_2"/><label id="82842e89afe04bd0aa1877a5d870368a_2_1-choice_2-label" for="input_82842e89afe04bd0aa1877a5d870368a_2_1_choice_2" class="response-label field-label label-inline" aria-describedby="status_82842e89afe04bd0aa1877a5d870368a_2_1"> Disability Status
</label>
</div>
<div class="field">
<input type="radio" name="input_82842e89afe04bd0aa1877a5d870368a_2_1" id="input_82842e89afe04bd0aa1877a5d870368a_2_1_choice_3" class="field-input input-radio" value="choice_3"/><label id="82842e89afe04bd0aa1877a5d870368a_2_1-choice_3-label" for="input_82842e89afe04bd0aa1877a5d870368a_2_1_choice_3" class="response-label field-label label-inline" aria-describedby="status_82842e89afe04bd0aa1877a5d870368a_2_1"> Educational Background
</label>
</div>
<span id="answer_82842e89afe04bd0aa1877a5d870368a_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_82842e89afe04bd0aa1877a5d870368a_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_82842e89afe04bd0aa1877a5d870368a_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_82842e89afe04bd0aa1877a5d870368a" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_82842e89afe04bd0aa1877a5d870368a">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="82842e89afe04bd0aa1877a5d870368a-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="82842e89afe04bd0aa1877a5d870368a-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="82842e89afe04bd0aa1877a5d870368a-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
</div>
</div>
<div class="xblock xblock-public_view xblock-public_view-vertical" data-block-type="vertical" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="VerticalStudentView" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@vertical+block@d171b487f9f042439e24a05ab8fedfb9" data-course-id="course-v1:MITx+HST.953x+3T2020">
<h2 class="hd hd-2 unit-title">Fairness and Bias in Healthcare</h2>
<div class="vert-mod">
<div class="vert vert-0" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@0bbe06cbc5a44844b9fbc9912769bc31">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@0bbe06cbc5a44844b9fbc9912769bc31" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<p><strong>Intro:</strong></p>
<p>Despite aspirations to the contrary, it is empirically clear that the health system of the United States does not work equally well for everyone. This manifests upstream, with racism and systemic bias influencing social determinants of health to worsen the health status of certain groups. Rather than assuaging this disparity, there are numerous instances in which disparate treatment in the healthcare system actually widens these gaps. This all manifests in widely disparate outcomes.</p>
<p>Within this section, we will explore this topic of unfairness and bias through the lenses of disparate treatment, disparate outcomes, and disparities in medical knowledge generation.Those working with healthcare data need to understand the presence and nature of these biases if they are to ensure that machine learning systems do not work to amplify them.
<p>Before we start discussing those topics in depth, in the following video William Boag will be introducing a project that was conducted at MIT dealing with racial disparities in End-of-Life Care.
</div>
</div>
<div class="vert vert-1" data-id="block-v1:MITx+HST.953x+3T2020+type@video+block@c473c2fec8e94b558877c270f4365e49">
<div class="xblock xblock-public_view xblock-public_view-video xmodule_display xmodule_VideoBlock" data-block-type="video" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@video+block@c473c2fec8e94b558877c270f4365e49" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Video"}
</script>
<h3 class="hd hd-2">Video</h3>
<div
id="video_c473c2fec8e94b558877c270f4365e49"
class="video closed"
data-metadata='{"transcriptAvailableTranslationsUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@c473c2fec8e94b558877c270f4365e49/handler/transcript/available_translations", "saveStateUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@c473c2fec8e94b558877c270f4365e49/handler/xmodule_handler/save_user_state", "streams": "1.00:3PTIRJIiabE", "transcriptLanguages": {"en": "English"}, "completionPercentage": 0.95, "autoplay": false, "start": 0.0, "ytApiUrl": "https://www.youtube.com/iframe_api", "sources": [], "saveStateEnabled": false, "captionDataDir": null, "duration": 0.0, "autohideHtml5": false, "publishCompletionUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@c473c2fec8e94b558877c270f4365e49/handler/publish_completion", "prioritizeHls": false, "transcriptLanguage": "en", "recordedYoutubeIsAvailable": true, "autoAdvance": false, "ytTestTimeout": 1500, "transcriptTranslationUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@c473c2fec8e94b558877c270f4365e49/handler/transcript/translation/__lang__", "generalSpeed": 1.0, "lmsRootURL": "https://openlearninglibrary.mit.edu", "speed": null, "ytMetadataEndpoint": "", "completionEnabled": false, "savedVideoPosition": 0.0, "showCaptions": "true", "end": 0.0, "poster": null}'
data-bumper-metadata='null'
data-autoadvance-enabled="False"
data-poster='null'
tabindex="-1"
>
<div class="focus_grabber first"></div>
<div class="tc-wrapper">
<div class="video-wrapper">
<span tabindex="0" class="spinner" aria-hidden="false" aria-label="Loading video player"></span>
<span tabindex="-1" class="btn-play fa fa-youtube-play fa-2x is-hidden" aria-hidden="true" aria-label="Play video"></span>
<div class="video-player-pre"></div>
<div class="video-player">
<div id="c473c2fec8e94b558877c270f4365e49"></div>
<h4 class="hd hd-4 video-error is-hidden">No playable video sources found.</h4>
<h4 class="hd hd-4 video-hls-error is-hidden">
Your browser does not support this video format. Try using a different browser.
</h4>
</div>
<div class="video-player-post"></div>
<div class="closed-captions"></div>
<div class="video-controls is-hidden">
<div>
<div class="vcr"><div class="vidtime">0:00 / 0:00</div></div>
<div class="secondary-controls"></div>
</div>
</div>
</div>
</div>
<div class="focus_grabber last"></div>
<h3 class="hd hd-4 downloads-heading sr" id="video-download-transcripts_c473c2fec8e94b558877c270f4365e49">Downloads and transcripts</h3>
<div class="wrapper-downloads" role="region" aria-labelledby="video-download-transcripts_c473c2fec8e94b558877c270f4365e49">
<div class="wrapper-download-transcripts">
<h4 class="hd hd-5">Transcripts</h4>
<ul class="list-download-transcripts">
<li class="transcript-option">
<a class="btn btn-link" href="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@c473c2fec8e94b558877c270f4365e49/handler/transcript/download" data-value="srt">Download SubRip (.srt) file</a>
</li>
<li class="transcript-option">
<a class="btn btn-link" href="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@c473c2fec8e94b558877c270f4365e49/handler/transcript/download" data-value="txt">Download Text (.txt) file</a>
</li>
</ul>
</div>
</div>
</div>
</div>
</div>
<div class="vert vert-2" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@79a345f2e9b041428c267a45d1378f6b">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@79a345f2e9b041428c267a45d1378f6b" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<p><strong>Disparate Treatment in the Healthcare System:</strong></p>
<p>Disparate treatment in healthcare arises from a wide range of origins, <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3797360/#:~:text=Research%20suggests%20that%20implicit%20bias,ethnicity%2C%20gender%20or%20other%20characteristics." target="[object Object]">including direct implicit bias on the part of healthcare practitioners </a>and structural disparities in the construction and delivery of healthcare services. A brief, non-exhaustive list of examples of such disparities follows:</p>
<ul type="disc">
<li><em><strong>Disparities in Treatment:</strong></em></li>
<ul type="circle">
<li>African Americans, both <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4829078/" target="[object Object]">children</a> and <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4976905/" target="[object Object]">adults </a>are less likely to receive appropriate pain treatment when in acute distress, even when controlling for age, sex, and time of treatment.</li>
<li>Up to half of medical students<a href="https://www.pnas.org/content/early/2016/03/30/1516047113.abstract" target="[object Object]"> in one study</a> hold the physiologically false belief that Black and white patients experience pain differently </li>
</ul>
<li><em><strong>Disparities in Diagnosis:</strong></em></li>
<ul type="circle">
<li>Women presenting to the emergency department with coronary syndromes <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2043078/" target="[object Object]">are less likely than men with the same symptoms to be admitted to acute care</a>.</li>
<li>Black patients experiencing symptoms of major depression are <a href="https://ps.psychiatryonline.org/doi/10.1176/appi.ps.201800223" target="[object Object]">more likely to receive a potentially inappropriate diagnosis of schizophrenia </a></li>
<li><a href="https://www.nejm.org/doi/full/10.1056/NEJMp1915891" target="[object Object]">Medical students often only learn to diagnose important and common dermatologic findings on light colored skin</a>, with darker skin being treated as an afterthought.</li>
</ul>
<li><em><strong>Structural Disparities:</strong></em></li>
<ul type="circle">
<li><a href="https://www.healthaffairs.org/doi/abs/10.1377/hlthaff.2017.0338" target="[object Object]">54% of rural counties</a> do not have a hospital with obstetrics services </li>
<li>Elderly LGBT individuals experience isolation, and <a href="https://www.lgbtagingcenter.org/resources/resource.cfm?r=30" target="[object Object]">a lack of access to culturally competent healthcare and social service providers </a></li>
</ul>
</ul>
<p>Understanding these treatment disparities is important to machine learning in that many models implicitly use existing healthcare processes as their "gold standard" for recommendations. A model that attempts to predict diagnostic decisions for patients, for example, will itself be biased if it is generated based on biased physician decisions.</p>
<p><strong>Disparate Outcomes in the Healthcare System:</strong></p>
<p>Healthcare disparities, along with broader disparities in social and government services, contribute to wildly disparate health outcomes for different groups in American society. A similarly non-exhaustive list of these disparities follows:</p>
<ul type="disc">
<li><em><strong>Life Expectancy Gaps:</strong></em></li>
<ul type="circle">
<li>The life expectancy at birth for <a href="https://www.cdc.gov/nchs/data/hus/2017/015.pd" target="[object Object]">white Americans is 4.4 years longer than for Black Americans</a>.</li>
<li>Life expectancy varies <a href="https://www.ncbi.nlm.nih.gov/books/NBK425844/" target="[object Object]">by up to 7 years between states</a>.</li>
</ul>
<li><em><strong>Infant Mortality Gaps:</strong></em></li>
<ul type="circle">
<li><a href="https://www.cdc.gov/nchs/data/databriefs/db285.pdf" target="[object Object]">Infant mortality is twice as high for Black Americans as compared to white Americans</a>, at each level of socioeconomic status </li>
<li>American Indian and Alaska Native infants are <a href="https://minorityhealth.hhs.gov/omh/browse.aspx?lvl=4&lvlid=38#:~:text=American%20Indian%2FAlaska%20Native%20infants%20are%202.7%20times%20more%20likely,to%20non%2DHispanic%20white%20mothers." target="[object Object]">2.7x as likely</a> to die from accidents before one year of age</li>
</ul>
<li><em><strong>Chronic Health Problem Gaps:</strong></em></li>
<ul type="circle">
<li>Puerto Ricans are <a href="https://minorityhealth.hhs.gov/omh/browse.aspx?lvl=4&lvlid=60" target="[object Object]">72% more likely</a> to develop asthma than non-Hispanic whites </li>
<li>LGBT youth are <a href="https://www.sciencedirect.com/science/article/abs/pii/0197007091900806" target="[object Object]">more likely to be homeless </a>and <a href="https://jamanetwork.com/journals/jamapediatrics/fullarticle/346930)" target="[object Object]">2-3x more likely to attempt suicide</a>.</li>
</ul>
</ul>
<p>These vast prior differences in health outcomes between populations cannot be ignored in healthcare data science research, particularly for those models that seek to predict patient outcomes. <a href="https://www.nejm.org/doi/full/10.1056/NEJMms2004740" target="[object Object]">As recent scholarship has argued</a>, accounting for these disparities is not necessarily so simple as adding a "race factor" to account for and adjust results - as such interventions may act only to set disparities in stone.</p>
<p><strong>Disparities in Medical Knowledge Generation:</strong></p>
<p>Another substantial source of unfairness in healthcare is the biased nature of the medical knowledge generation system, which broadly holds a <a href="https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1001918" target="[object Object]">single (typically white, and typically male)</a> prototype for the presentation and course of disease.</p>
<ul type="disc">
<li><em><strong>Framingham Scores:</strong></em></li>
<ul type="circle">
<li>A classic example of disparity in knowledge generation arises from the Framingham risk study, which generated cardiovascular risk scores from long-term examination of the predominantly white, working class town of Framingham Massachusetts. Perhaps unsurprisingly, these scores <a href="https://pubmed.ncbi.nlm.nih.gov/26134404/" target="[object Object]">perform more poorly when applied to Black patients</a>.</li>
</ul>
<li><em><strong>Clinical Trials:</strong></em></li>
<ul type="circle">
<li>Clinical trials for drugs and treatments that go on to influence the care of a diverse range of patients from around the world are <a href="https://equityhealthj.biomedcentral.com/articles/10.1186/s12939-019-0954-x" target="[object Object]">often tested on a predominantly white and male population</a>, with <a href="https://equityhealthj.biomedcentral.com/articles/10.1186/s12939-019-0954-x" target="[object Object]">women </a>and<a href="https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1001918" target="[object Object]"> people of color</a> frequently excluded from or underrepresented in studies.</li>
</ul>
</ul>
<p>In the machine learning context, these disparities can represent a classic example of <a href="https://towardsdatascience.com/understanding-dataset-shift-f2a5a262a766?gi=ca27abd80675" target="[object Object]">dataset shift</a>, as the training population (predominantly white and male) can vary substantially in its characteristics as compared to the (broadly diverse) testing population. It is always important to interrogate the demographic makeup of data source populations, and where possible to seek to improve the diversity of datasets (as, for example, <a href="https://www.nejm.org/doi/full/10.1056/NEJMsr1809937" target="[object Object]">the "All of Us" researchers are attempting to do</a>).</p>
<p><strong>Trust:</strong></p>
<p>From sordid medical experiments performed on slaves, to the infamous "Tuskegee Syphilis Study" wherein <a href="https://books.google.ca/books/about/Medical_Apartheid.html?id=apGhwRt6A7QC" target="[object Object]">African American men were not given treatment for advanced stage syphilis in order to study the disease's effects</a>, medicine has routinely exploited Black patients in the name of "research". Similarly, the <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4695779/" target="[object Object]">psychiatric pathologization of homosexuality</a> provided a long-standing pretext for anti-gay bigotry. Understandably, it follows that <a href="https://pubmed.ncbi.nlm.nih.gov/19175244/" target="[object Object]">African Americans</a> and LGBT individuals (<a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5278794/" target="[object Object]">particularly those with intersectional identities</a>) have high levels of mistrust toward the medical establishment. It is important for those seeking to implement data science in healthcare to understand this history, and to understand the critiques of those who fear that these technologies may act simply to <a href="https://science.sciencemag.org/content/366/6464/421" target="[object Object]">"automate racism"</a> and further disparities. Trust in healthcare data science is something that must be built and earned, rather than assumed.</p>
</div>
</div>
<div class="vert vert-3" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@7813e171d1fb40e5977e55b29087d5bb">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@7813e171d1fb40e5977e55b29087d5bb" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_7813e171d1fb40e5977e55b29087d5bb" class="problems-wrapper" role="group"
aria-labelledby="7813e171d1fb40e5977e55b29087d5bb-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@7813e171d1fb40e5977e55b29087d5bb" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@7813e171d1fb40e5977e55b29087d5bb/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="7813e171d1fb40e5977e55b29087d5bb-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@7813e171d1fb40e5977e55b29087d5bb-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@7813e171d1fb40e5977e55b29087d5bb-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>True or False: Existing biases within healthcare can become embedded in, and thus impact the performance of, machine learning models</p>
<div class="choicegroup capa_inputtype" id="inputtype_7813e171d1fb40e5977e55b29087d5bb_2_1">
<fieldset aria-describedby="status_7813e171d1fb40e5977e55b29087d5bb_2_1">
<div class="field">
<input type="radio" name="input_7813e171d1fb40e5977e55b29087d5bb_2_1" id="input_7813e171d1fb40e5977e55b29087d5bb_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="7813e171d1fb40e5977e55b29087d5bb_2_1-choice_0-label" for="input_7813e171d1fb40e5977e55b29087d5bb_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_7813e171d1fb40e5977e55b29087d5bb_2_1"> True
</label>
</div>
<div class="field">
<input type="radio" name="input_7813e171d1fb40e5977e55b29087d5bb_2_1" id="input_7813e171d1fb40e5977e55b29087d5bb_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="7813e171d1fb40e5977e55b29087d5bb_2_1-choice_1-label" for="input_7813e171d1fb40e5977e55b29087d5bb_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_7813e171d1fb40e5977e55b29087d5bb_2_1"> False
</label>
</div>
<span id="answer_7813e171d1fb40e5977e55b29087d5bb_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_7813e171d1fb40e5977e55b29087d5bb_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_7813e171d1fb40e5977e55b29087d5bb_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_7813e171d1fb40e5977e55b29087d5bb" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_7813e171d1fb40e5977e55b29087d5bb">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="7813e171d1fb40e5977e55b29087d5bb-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="7813e171d1fb40e5977e55b29087d5bb-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="7813e171d1fb40e5977e55b29087d5bb-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-4" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@89e69d9be2f24c838cd122a214b4a9fc">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@89e69d9be2f24c838cd122a214b4a9fc" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_89e69d9be2f24c838cd122a214b4a9fc" class="problems-wrapper" role="group"
aria-labelledby="89e69d9be2f24c838cd122a214b4a9fc-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@89e69d9be2f24c838cd122a214b4a9fc" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@89e69d9be2f24c838cd122a214b4a9fc/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="89e69d9be2f24c838cd122a214b4a9fc-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@89e69d9be2f24c838cd122a214b4a9fc-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@89e69d9be2f24c838cd122a214b4a9fc-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>True or False: Adding a corrective "race factor" solves all of the problems associated with differential performance of machine learning models</p>
<div class="choicegroup capa_inputtype" id="inputtype_89e69d9be2f24c838cd122a214b4a9fc_2_1">
<fieldset aria-describedby="status_89e69d9be2f24c838cd122a214b4a9fc_2_1">
<div class="field">
<input type="radio" name="input_89e69d9be2f24c838cd122a214b4a9fc_2_1" id="input_89e69d9be2f24c838cd122a214b4a9fc_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="89e69d9be2f24c838cd122a214b4a9fc_2_1-choice_0-label" for="input_89e69d9be2f24c838cd122a214b4a9fc_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_89e69d9be2f24c838cd122a214b4a9fc_2_1"> True
</label>
</div>
<div class="field">
<input type="radio" name="input_89e69d9be2f24c838cd122a214b4a9fc_2_1" id="input_89e69d9be2f24c838cd122a214b4a9fc_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="89e69d9be2f24c838cd122a214b4a9fc_2_1-choice_1-label" for="input_89e69d9be2f24c838cd122a214b4a9fc_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_89e69d9be2f24c838cd122a214b4a9fc_2_1"> False
</label>
</div>
<span id="answer_89e69d9be2f24c838cd122a214b4a9fc_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_89e69d9be2f24c838cd122a214b4a9fc_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_89e69d9be2f24c838cd122a214b4a9fc_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_89e69d9be2f24c838cd122a214b4a9fc" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_89e69d9be2f24c838cd122a214b4a9fc">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="89e69d9be2f24c838cd122a214b4a9fc-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="89e69d9be2f24c838cd122a214b4a9fc-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="89e69d9be2f24c838cd122a214b4a9fc-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-5" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@dd47d2e88b7f4f24adbcfc9e64ac0c38">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@dd47d2e88b7f4f24adbcfc9e64ac0c38" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_dd47d2e88b7f4f24adbcfc9e64ac0c38" class="problems-wrapper" role="group"
aria-labelledby="dd47d2e88b7f4f24adbcfc9e64ac0c38-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@dd47d2e88b7f4f24adbcfc9e64ac0c38" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@dd47d2e88b7f4f24adbcfc9e64ac0c38/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="dd47d2e88b7f4f24adbcfc9e64ac0c38-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@dd47d2e88b7f4f24adbcfc9e64ac0c38-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@dd47d2e88b7f4f24adbcfc9e64ac0c38-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>True or false: Racial disparities in healthcare can be solely explained by disparities in socioeconomic status.</p>
<div class="choicegroup capa_inputtype" id="inputtype_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1">
<fieldset aria-describedby="status_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1">
<div class="field">
<input type="radio" name="input_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1" id="input_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1-choice_0-label" for="input_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1"> True
</label>
</div>
<div class="field">
<input type="radio" name="input_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1" id="input_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1-choice_1-label" for="input_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1"> False
</label>
</div>
<span id="answer_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_dd47d2e88b7f4f24adbcfc9e64ac0c38_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_dd47d2e88b7f4f24adbcfc9e64ac0c38_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_dd47d2e88b7f4f24adbcfc9e64ac0c38" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_dd47d2e88b7f4f24adbcfc9e64ac0c38">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="dd47d2e88b7f4f24adbcfc9e64ac0c38-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="dd47d2e88b7f4f24adbcfc9e64ac0c38-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="dd47d2e88b7f4f24adbcfc9e64ac0c38-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
</div>
</div>
<div class="xblock xblock-public_view xblock-public_view-vertical" data-block-type="vertical" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="VerticalStudentView" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@vertical+block@cfdd9f01e6a94798beda9c4a480f8dd0" data-course-id="course-v1:MITx+HST.953x+3T2020">
<h2 class="hd hd-2 unit-title">Fairness and Bias in Machine Learning</h2>
<div class="vert-mod">
<div class="vert vert-0" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@872ad23dc92b46e4acca14645dbbf9b1">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@872ad23dc92b46e4acca14645dbbf9b1" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<p><strong>Intro:</strong></p>
<p>Bias and fairness are huge topics in machine learning, even outside of the healthcare context. Understanding the ways in which algorithms can act as unfair classifiers, or participate in socio-technical systems that perpetuate unfairness and inequity, has become an increasingly prominent area of research.</p>
<p><strong>Unfairness and Bias in Data</strong></p>
<p>Unfair differential performance of an algorithm may arise from the nature of the data that the algorithm has been created from.</p>
<p>For an accessible overview of unfairness in relation to data, please watch this brief video by Irene Chen et al (2018)</p>
<p></p>
</div>
</div>
<div class="vert vert-1" data-id="block-v1:MITx+HST.953x+3T2020+type@video+block@b602754d17d84ed2bdf689c5e6ee1f1b">
<div class="xblock xblock-public_view xblock-public_view-video xmodule_display xmodule_VideoBlock" data-block-type="video" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@video+block@b602754d17d84ed2bdf689c5e6ee1f1b" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Video"}
</script>
<h3 class="hd hd-2">Irene Chen et al - Why Is My Classifier Discriminatory?</h3>
<div
id="video_b602754d17d84ed2bdf689c5e6ee1f1b"
class="video closed"
data-metadata='{"transcriptAvailableTranslationsUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@b602754d17d84ed2bdf689c5e6ee1f1b/handler/transcript/available_translations", "saveStateUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@b602754d17d84ed2bdf689c5e6ee1f1b/handler/xmodule_handler/save_user_state", "streams": "1.00:qROvkENF6Wg", "transcriptLanguages": {"en": "English"}, "completionPercentage": 0.95, "autoplay": false, "start": 0.0, "ytApiUrl": "https://www.youtube.com/iframe_api", "sources": [], "saveStateEnabled": false, "captionDataDir": null, "duration": 0.0, "autohideHtml5": false, "publishCompletionUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@b602754d17d84ed2bdf689c5e6ee1f1b/handler/publish_completion", "prioritizeHls": false, "transcriptLanguage": "en", "recordedYoutubeIsAvailable": true, "autoAdvance": false, "ytTestTimeout": 1500, "transcriptTranslationUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@b602754d17d84ed2bdf689c5e6ee1f1b/handler/transcript/translation/__lang__", "generalSpeed": 1.0, "lmsRootURL": "https://openlearninglibrary.mit.edu", "speed": null, "ytMetadataEndpoint": "", "completionEnabled": false, "savedVideoPosition": 0.0, "showCaptions": "true", "end": 0.0, "poster": null}'
data-bumper-metadata='null'
data-autoadvance-enabled="False"
data-poster='null'
tabindex="-1"
>
<div class="focus_grabber first"></div>
<div class="tc-wrapper">
<div class="video-wrapper">
<span tabindex="0" class="spinner" aria-hidden="false" aria-label="Loading video player"></span>
<span tabindex="-1" class="btn-play fa fa-youtube-play fa-2x is-hidden" aria-hidden="true" aria-label="Play video"></span>
<div class="video-player-pre"></div>
<div class="video-player">
<div id="b602754d17d84ed2bdf689c5e6ee1f1b"></div>
<h4 class="hd hd-4 video-error is-hidden">No playable video sources found.</h4>
<h4 class="hd hd-4 video-hls-error is-hidden">
Your browser does not support this video format. Try using a different browser.
</h4>
</div>
<div class="video-player-post"></div>
<div class="closed-captions"></div>
<div class="video-controls is-hidden">
<div>
<div class="vcr"><div class="vidtime">0:00 / 0:00</div></div>
<div class="secondary-controls"></div>
</div>
</div>
</div>
</div>
<div class="focus_grabber last"></div>
<h3 class="hd hd-4 downloads-heading sr" id="video-download-transcripts_b602754d17d84ed2bdf689c5e6ee1f1b">Downloads and transcripts</h3>
<div class="wrapper-downloads" role="region" aria-labelledby="video-download-transcripts_b602754d17d84ed2bdf689c5e6ee1f1b">
<div class="wrapper-download-transcripts">
<h4 class="hd hd-5">Transcripts</h4>
<ul class="list-download-transcripts">
<li class="transcript-option">
<a class="btn btn-link" href="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@b602754d17d84ed2bdf689c5e6ee1f1b/handler/transcript/download" data-value="srt">Download SubRip (.srt) file</a>
</li>
<li class="transcript-option">
<a class="btn btn-link" href="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@b602754d17d84ed2bdf689c5e6ee1f1b/handler/transcript/download" data-value="txt">Download Text (.txt) file</a>
</li>
</ul>
</div>
</div>
</div>
</div>
</div>
<div class="vert vert-2" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@5591c4b53d0c4049b1d60e0b95fd8409">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@5591c4b53d0c4049b1d60e0b95fd8409" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<ul type="disc">
<li><em><strong>Error due to variance</strong></em> arises when there is a lack of data for a specific group, leading to a poor estimate. It is solved by gathering additional data.</li>
<ul type="disc">
<li><img src="/assets/courseware/v1/b30b68bf80c1cc442b098fb372ca1ae0/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Variance.png" alt="Graph of bias due to variance" width="557" height="265" /></li>
</ul>
<li><em><strong>Error due to bias</strong> </em>arises when there are legitimately different data functions between groups. It may be solved by changing the model class.</li>
<ul type="disc">
<li><img src="/assets/courseware/v1/24887feba73a6640d7a605335a0d410c/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Bias_due_to_variance.png" alt="Graph showing two lines with different basic structures sloping away from each other" width="538" height="246" /></li>
</ul>
<li><em><strong>Error due to noise</strong></em> arises when there is a greater degree of noise for the minority group on a given set of features. It may be solved by adding additional features.</li>
<ul type="disc">
<li><img src="/assets/courseware/v1/ab6e21ca7ba9dc31c6930d030994b5cb/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Bias_due_to_noise.png" alt="Image shows two groups of dots, with the orange set of dots following the same trajectory but with much more noise" width="545" height="213" /></li>
</ul>
</ul>
<p>One excellent method for assessing the nature of a dataset and whether it is suitable for a specific application arises from the concept of "datasheets for datasets", developed by <a href="https://arxiv.org/pdf/1803.09010.pdf" target="[object Object]">Timnit Gebru et al</a>. The method involves 50 relevant questions around the motivation, composition, collection process, cleaning process, uses, distribution, and maintenance of a machine learning model.</p>
<p><strong>Unfairness Arising from Models</strong></p>
<p>One significant source of criticism for complex machine learning models (particularly deep learning models) is the tendency of such models toward a lack of interpretability and the "black box" effect wherein it can be difficult to tell how a specific set of inputs generated a specific set of outputs. This obfuscation can hide a wide range of sources of unfairness and bias that might be present within the model itself, or within the intentions of those generating and deploying the model.</p>
<p>The machine learning pipeline can be vulnerable to bias at each of its many steps (<a href="http://harinisuresh.com/img/bias_framework.pdf" target="[object Object]">Harini Suresh et al</a>):</p>
<ul type="disc">
<li><em><strong>Data Collection</strong></em> refers to the initial gathering of the data in the real world</li>
<li><em><strong>Data Preparation</strong></em> refers to the decisions made in structuring and pre-processing the data for deployment, as well as the designation of train, test, and validation sets.</li>
<li><em><strong>Model Development</strong></em> refers to the process of choosing and designing the model, as well as tuning its parameters</li>
<li><em><strong>Model Evaluation</strong></em> refers to the process of evaluating the performance of the model against either a reserved dataset or a benchmark dataset</li>
<li><em><strong>Model Postprocessing</strong></em> refers to the decisions made in better aligning the developed model to its target use cases</li>
<li><em><strong>Model Deployment</strong></em> refers to the real-world steps to deploying the model in its operative environment.</li>
</ul>
<p>Each of the decisions made at each of these points implicitly embeds certain values and will have certain impacts. <a href="https://arxiv.org/pdf/1908.09635.pdf" target="[object Object]">A wide range of methods</a> have been proposed for addressing fairness in model construction at each of these levels.</p>
<p><strong>Unfairness Arising from Socio-Technical Context:</strong></p>
<p>Focusing too directly on "biased data" or "poor model design", however, can be insufficient to describe all relevant socio-technical considerations that play a role in machine learning-driven unfairness or bias. Increasingly, fairness researchers are taking a process-oriented view, and emphasizing that the machine learning pipeline from beginning to end consists of a series of value-driven decisions. We will explore this topic through the framework of <a href="http://harinisuresh.com/img/bias_framework.pdf" target="[object Object]">Harini Suresh and John V. Guttag (2020)</a>.</p>
<p><strong>Types of Bias:</strong></p>
<ul type="disc">
<li><em><strong>Historical Bias</strong></em> arises when the structure of the world is misaligned with the values or objectives to be encoded or propagated by a model. Importantly, this bias exists in the broader context of the world and is not solved by perfect model design.</li>
<li><em><strong>Representation Bias</strong></em> arises when the training population under-represents a specific portion of the use case population, and thus does not perform well for that portion of the population</li>
<li><em><strong>Measurement Bias</strong></em> arises when the chosen set of features (typically proxy features) does not align well with the actual desired characteristics that the model is attempting to capture.</li>
<li><em><strong>Aggregation Bias</strong></em> arises when distinct populations are inappropriately combined, and a heterogeneous group is reduced to a single model.</li>
<li><em><strong>Evaluation Bias</strong></em> arises when the testing or benchmark population does not accurately represent the use case population (or when the metrics being used to evaluate the model are not aligned with the way the model is supposed to be used)</li>
<li><em><strong>Deployment Bias</strong> </em>arises when the model is used or interpreted in inappropriate ways after model deployment.</li>
</ul>
<p>The diversity of machine learning practitioners is another important component of the sociotechnical context. Women and BIPOC individuals<a href="https://www.theguardian.com/technology/2019/apr/16/artificial-intelligence-lack-diversity-new-york-university-study" target="[object Object]"> are broadly underrepresented in the machine learning world</a>, and this has implications for the field's "blind spots", and the questions that are and are not asked during the development process.</p>
<p></p>
</div>
</div>
<div class="vert vert-3" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@9257631fd96a45f39c16eebbafecff75">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@9257631fd96a45f39c16eebbafecff75" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_9257631fd96a45f39c16eebbafecff75" class="problems-wrapper" role="group"
aria-labelledby="9257631fd96a45f39c16eebbafecff75-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@9257631fd96a45f39c16eebbafecff75" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@9257631fd96a45f39c16eebbafecff75/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="9257631fd96a45f39c16eebbafecff75-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@9257631fd96a45f39c16eebbafecff75-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@9257631fd96a45f39c16eebbafecff75-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>True or False: Deep learning models are unbiased because they learn directly from the data, without any opportunity for human biases to sneak in</p>
<div class="choicegroup capa_inputtype" id="inputtype_9257631fd96a45f39c16eebbafecff75_2_1">
<fieldset aria-describedby="status_9257631fd96a45f39c16eebbafecff75_2_1">
<div class="field">
<input type="radio" name="input_9257631fd96a45f39c16eebbafecff75_2_1" id="input_9257631fd96a45f39c16eebbafecff75_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="9257631fd96a45f39c16eebbafecff75_2_1-choice_0-label" for="input_9257631fd96a45f39c16eebbafecff75_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_9257631fd96a45f39c16eebbafecff75_2_1"> True
</label>
</div>
<div class="field">
<input type="radio" name="input_9257631fd96a45f39c16eebbafecff75_2_1" id="input_9257631fd96a45f39c16eebbafecff75_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="9257631fd96a45f39c16eebbafecff75_2_1-choice_1-label" for="input_9257631fd96a45f39c16eebbafecff75_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_9257631fd96a45f39c16eebbafecff75_2_1"> False
</label>
</div>
<span id="answer_9257631fd96a45f39c16eebbafecff75_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_9257631fd96a45f39c16eebbafecff75_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_9257631fd96a45f39c16eebbafecff75_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_9257631fd96a45f39c16eebbafecff75" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_9257631fd96a45f39c16eebbafecff75">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="9257631fd96a45f39c16eebbafecff75-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="9257631fd96a45f39c16eebbafecff75-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="9257631fd96a45f39c16eebbafecff75-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-4" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@57cf59be573049f694b5ea016aa3e469">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@57cf59be573049f694b5ea016aa3e469" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_57cf59be573049f694b5ea016aa3e469" class="problems-wrapper" role="group"
aria-labelledby="57cf59be573049f694b5ea016aa3e469-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@57cf59be573049f694b5ea016aa3e469" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@57cf59be573049f694b5ea016aa3e469/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="57cf59be573049f694b5ea016aa3e469-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@57cf59be573049f694b5ea016aa3e469-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@57cf59be573049f694b5ea016aa3e469-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>A model is being created to predict whether pets will prefer chicken or salmon in their dinner based on the hair length of the pet. Assume that long haired cats tend to prefer salmon, while short haired cats tend to prefer chicken, and all dogs prefer chicken regardless of their hair length. What would be the primary source of bias in the performance of this model if it is trained on both cat and dog data?</p>
<div class="choicegroup capa_inputtype" id="inputtype_57cf59be573049f694b5ea016aa3e469_2_1">
<fieldset aria-describedby="status_57cf59be573049f694b5ea016aa3e469_2_1">
<div class="field">
<input type="radio" name="input_57cf59be573049f694b5ea016aa3e469_2_1" id="input_57cf59be573049f694b5ea016aa3e469_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="57cf59be573049f694b5ea016aa3e469_2_1-choice_0-label" for="input_57cf59be573049f694b5ea016aa3e469_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_57cf59be573049f694b5ea016aa3e469_2_1"> Error due to variance
</label>
</div>
<div class="field">
<input type="radio" name="input_57cf59be573049f694b5ea016aa3e469_2_1" id="input_57cf59be573049f694b5ea016aa3e469_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="57cf59be573049f694b5ea016aa3e469_2_1-choice_1-label" for="input_57cf59be573049f694b5ea016aa3e469_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_57cf59be573049f694b5ea016aa3e469_2_1"> Error due to noise
</label>
</div>
<div class="field">
<input type="radio" name="input_57cf59be573049f694b5ea016aa3e469_2_1" id="input_57cf59be573049f694b5ea016aa3e469_2_1_choice_2" class="field-input input-radio" value="choice_2"/><label id="57cf59be573049f694b5ea016aa3e469_2_1-choice_2-label" for="input_57cf59be573049f694b5ea016aa3e469_2_1_choice_2" class="response-label field-label label-inline" aria-describedby="status_57cf59be573049f694b5ea016aa3e469_2_1"> Error due to bias
</label>
</div>
<div class="field">
<input type="radio" name="input_57cf59be573049f694b5ea016aa3e469_2_1" id="input_57cf59be573049f694b5ea016aa3e469_2_1_choice_3" class="field-input input-radio" value="choice_3"/><label id="57cf59be573049f694b5ea016aa3e469_2_1-choice_3-label" for="input_57cf59be573049f694b5ea016aa3e469_2_1_choice_3" class="response-label field-label label-inline" aria-describedby="status_57cf59be573049f694b5ea016aa3e469_2_1"> Deployment error
</label>
</div>
<span id="answer_57cf59be573049f694b5ea016aa3e469_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_57cf59be573049f694b5ea016aa3e469_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_57cf59be573049f694b5ea016aa3e469_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_57cf59be573049f694b5ea016aa3e469" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_57cf59be573049f694b5ea016aa3e469">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="57cf59be573049f694b5ea016aa3e469-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="57cf59be573049f694b5ea016aa3e469-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="57cf59be573049f694b5ea016aa3e469-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-5" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@91853fb9ad724e8c89b94a8f060f20aa">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@91853fb9ad724e8c89b94a8f060f20aa" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_91853fb9ad724e8c89b94a8f060f20aa" class="problems-wrapper" role="group"
aria-labelledby="91853fb9ad724e8c89b94a8f060f20aa-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@91853fb9ad724e8c89b94a8f060f20aa" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@91853fb9ad724e8c89b94a8f060f20aa/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="91853fb9ad724e8c89b94a8f060f20aa-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@91853fb9ad724e8c89b94a8f060f20aa-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@91853fb9ad724e8c89b94a8f060f20aa-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>Many models used in an attempt to understand and predict crime use "arrest" as a proxy variable for crime, despite the fact that minority communities are often more highly policed and thus have higher arrest rates for an equivalent amount of crime. What type of bias is shown in this example?</p>
<div class="choicegroup capa_inputtype" id="inputtype_91853fb9ad724e8c89b94a8f060f20aa_2_1">
<fieldset aria-describedby="status_91853fb9ad724e8c89b94a8f060f20aa_2_1">
<div class="field">
<input type="radio" name="input_91853fb9ad724e8c89b94a8f060f20aa_2_1" id="input_91853fb9ad724e8c89b94a8f060f20aa_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="91853fb9ad724e8c89b94a8f060f20aa_2_1-choice_0-label" for="input_91853fb9ad724e8c89b94a8f060f20aa_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_91853fb9ad724e8c89b94a8f060f20aa_2_1"> Aggregation bias
</label>
</div>
<div class="field">
<input type="radio" name="input_91853fb9ad724e8c89b94a8f060f20aa_2_1" id="input_91853fb9ad724e8c89b94a8f060f20aa_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="91853fb9ad724e8c89b94a8f060f20aa_2_1-choice_1-label" for="input_91853fb9ad724e8c89b94a8f060f20aa_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_91853fb9ad724e8c89b94a8f060f20aa_2_1"> Measurement bias
</label>
</div>
<div class="field">
<input type="radio" name="input_91853fb9ad724e8c89b94a8f060f20aa_2_1" id="input_91853fb9ad724e8c89b94a8f060f20aa_2_1_choice_2" class="field-input input-radio" value="choice_2"/><label id="91853fb9ad724e8c89b94a8f060f20aa_2_1-choice_2-label" for="input_91853fb9ad724e8c89b94a8f060f20aa_2_1_choice_2" class="response-label field-label label-inline" aria-describedby="status_91853fb9ad724e8c89b94a8f060f20aa_2_1"> Evaluation bias
</label>
</div>
<div class="field">
<input type="radio" name="input_91853fb9ad724e8c89b94a8f060f20aa_2_1" id="input_91853fb9ad724e8c89b94a8f060f20aa_2_1_choice_3" class="field-input input-radio" value="choice_3"/><label id="91853fb9ad724e8c89b94a8f060f20aa_2_1-choice_3-label" for="input_91853fb9ad724e8c89b94a8f060f20aa_2_1_choice_3" class="response-label field-label label-inline" aria-describedby="status_91853fb9ad724e8c89b94a8f060f20aa_2_1"> Deployment bias
</label>
</div>
<span id="answer_91853fb9ad724e8c89b94a8f060f20aa_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_91853fb9ad724e8c89b94a8f060f20aa_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_91853fb9ad724e8c89b94a8f060f20aa_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_91853fb9ad724e8c89b94a8f060f20aa" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_91853fb9ad724e8c89b94a8f060f20aa">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="91853fb9ad724e8c89b94a8f060f20aa-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="91853fb9ad724e8c89b94a8f060f20aa-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="91853fb9ad724e8c89b94a8f060f20aa-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
</div>
</div>
<div class="xblock xblock-public_view xblock-public_view-vertical" data-block-type="vertical" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="VerticalStudentView" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@vertical+block@66e998b806ed4000990915bee1bfdf36" data-course-id="course-v1:MITx+HST.953x+3T2020">
<h2 class="hd hd-2 unit-title">Fairness and Bias in Machine Learning for Healthcare</h2>
<div class="vert-mod">
<div class="vert vert-0" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@4ef7ce2452384875902f7f4a4188b1f3">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@4ef7ce2452384875902f7f4a4188b1f3" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<p><strong>Intro:</strong></p>
<p>Fairness in machine learning for healthcare exists at the intersection of both fairness in healthcare and fairness in machine learning, and in addition to taking on characteristics of both it has its own emergent complexities. Machine learning in healthcare has the potential to:</p>
<ul type="disc">
<li><em><strong>Uncritically reproduce existing inequities in healthcare</strong></em></li>
<ul type="circle">
<li>A model might give outputs that directly track the existing inequities in the healthcare system</li>
</ul>
<li><em><strong>Exacerbate existing inequities in healthcare</strong></em></li>
<ul type="circle">
<li>A model might increase the size and scale of inequities, and can interfere with efforts at reform by hiding implicit biases behind a veneer of algorithmic objectivity</li>
</ul>
<li><em><strong>Introduce novel inequities in healthcare</strong></em></li>
<ul type="circle">
<li>Changes in healthcare practice brought on by the introduction of models may introduce novel inequities into healthcare. One clear example of this was the introduction of melanoma-detection algorithms that are <a href="https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2752995" target="[object Object]">trained predominantly on images of light skin</a>.</li>
</ul>
</ul>
<p>Within this section, we will explore these topics through a case-study based approach, and outline several considerations relevant to machine learning in healthcare.</p>
<p><strong>Considerations at the Intersection of ML and Healthcare</strong></p>
<ul type="disc">
<li>The machine learning for health roadmap is complex:</li>
<ul type="circle">
<li>Machine learning for healthcare occurs at the intersection of two complex and highly fields (medicine and computer science). Close collaboration between experts of both disciplines is required to ensure that machine learning for healthcare does not simply produce "solutions in search of problems", and that it is actually aligned with relevant patient needs and clinical contexts. The high stakes life-or-death nature of medicine also places certain burdens upon rigor in evaluation and reliability in deployment of these systems. For an excellent in-depth overview of responsible machine learning in healthcare, see <a href="https://www.nature.com/articles/s41591-019-0548-6?draft=collection#auth-2" target="[object Object]">Wiens et al (2019)</a>.</li>
</ul>
<li>Healthcare data is sensitive</li>
<ul type="circle">
<li>Information disclosed by patients to healthcare providers is often highly sensitive and intimate, including such topics as HIV status, drug use, and sexuality. A <a href="https://link.springer.com/article/10.1186/s40537-017-0110-7" target="[object Object]">particularly high burden on data safety and security</a>, as well as anonymization, arises when machine learning is applied in the healthcare context</li>
</ul>
<li>Deployment can be tricky</li>
<ul type="circle">
<li>Real-world healthcare environments are complex and constantly changing, with the clinical reality rarely aligning fully with the sanitized picture presented in health datasets. Machine learning that is aiming for clinical deployment must <a href="https://www.frontiersin.org/articles/10.3389/fdata.2018.00007/full" target="[object Object]">manage complex human-computer interaction and process design considerations</a> in order to find success.</li>
</ul>
</ul>
<p><strong>Research at the Intersection of Fairness in ML and Healthcare</strong></p>
<p>For real-world examples of fairness in machine learning and healthcare, we turn to a lecture from MIT PhD student Willie Boag</p>
<p><img src="/assets/courseware/v1/abdd55ec679d6aae2f9a3a236e47e075/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Willie_1.png" alt="Slide outlining paper "Can AI Help Reduce Disparities in General Medical and Mental Health Care?" width="1251" height="654" /></p>
<p><img src="/assets/courseware/v1/5d1b981d90e2d0d3d31573d5ef25b7b9/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Willie_2.png" alt="Slide outlining unpublished work showing that users have misplaced trust in model recommendations" width="1117" height="568" /></p>
<p><img src="/assets/courseware/v1/e7f84bd05fda16735e5cc269e318d379/asset-v1:MITx+HST.953x+3T2020+type@asset+block/Willie_3.png" alt="Slide outlining challenges of real-world implementation of anti-bias software " width="1272" height="556" /></p>
<p></p>
</div>
</div>
<div class="vert vert-1" data-id="block-v1:MITx+HST.953x+3T2020+type@video+block@bf8404a7ce0343f1b8749ada093ac968">
<div class="xblock xblock-public_view xblock-public_view-video xmodule_display xmodule_VideoBlock" data-block-type="video" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@video+block@bf8404a7ce0343f1b8749ada093ac968" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Video"}
</script>
<h3 class="hd hd-2">Video</h3>
<div
id="video_bf8404a7ce0343f1b8749ada093ac968"
class="video closed"
data-metadata='{"transcriptAvailableTranslationsUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@bf8404a7ce0343f1b8749ada093ac968/handler/transcript/available_translations", "saveStateUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@bf8404a7ce0343f1b8749ada093ac968/handler/xmodule_handler/save_user_state", "streams": "1.00:wUCJnt-wYLc", "transcriptLanguages": {"en": "English"}, "completionPercentage": 0.95, "autoplay": false, "start": 0.0, "ytApiUrl": "https://www.youtube.com/iframe_api", "sources": [], "saveStateEnabled": false, "captionDataDir": null, "duration": 0.0, "autohideHtml5": false, "publishCompletionUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@bf8404a7ce0343f1b8749ada093ac968/handler/publish_completion", "prioritizeHls": false, "transcriptLanguage": "en", "recordedYoutubeIsAvailable": true, "autoAdvance": false, "ytTestTimeout": 1500, "transcriptTranslationUrl": "/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@bf8404a7ce0343f1b8749ada093ac968/handler/transcript/translation/__lang__", "generalSpeed": 1.0, "lmsRootURL": "https://openlearninglibrary.mit.edu", "speed": null, "ytMetadataEndpoint": "", "completionEnabled": false, "savedVideoPosition": 0.0, "showCaptions": "true", "end": 0.0, "poster": null}'
data-bumper-metadata='null'
data-autoadvance-enabled="False"
data-poster='null'
tabindex="-1"
>
<div class="focus_grabber first"></div>
<div class="tc-wrapper">
<div class="video-wrapper">
<span tabindex="0" class="spinner" aria-hidden="false" aria-label="Loading video player"></span>
<span tabindex="-1" class="btn-play fa fa-youtube-play fa-2x is-hidden" aria-hidden="true" aria-label="Play video"></span>
<div class="video-player-pre"></div>
<div class="video-player">
<div id="bf8404a7ce0343f1b8749ada093ac968"></div>
<h4 class="hd hd-4 video-error is-hidden">No playable video sources found.</h4>
<h4 class="hd hd-4 video-hls-error is-hidden">
Your browser does not support this video format. Try using a different browser.
</h4>
</div>
<div class="video-player-post"></div>
<div class="closed-captions"></div>
<div class="video-controls is-hidden">
<div>
<div class="vcr"><div class="vidtime">0:00 / 0:00</div></div>
<div class="secondary-controls"></div>
</div>
</div>
</div>
</div>
<div class="focus_grabber last"></div>
<h3 class="hd hd-4 downloads-heading sr" id="video-download-transcripts_bf8404a7ce0343f1b8749ada093ac968">Downloads and transcripts</h3>
<div class="wrapper-downloads" role="region" aria-labelledby="video-download-transcripts_bf8404a7ce0343f1b8749ada093ac968">
<div class="wrapper-download-transcripts">
<h4 class="hd hd-5">Transcripts</h4>
<ul class="list-download-transcripts">
<li class="transcript-option">
<a class="btn btn-link" href="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@bf8404a7ce0343f1b8749ada093ac968/handler/transcript/download" data-value="srt">Download SubRip (.srt) file</a>
</li>
<li class="transcript-option">
<a class="btn btn-link" href="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@video+block@bf8404a7ce0343f1b8749ada093ac968/handler/transcript/download" data-value="txt">Download Text (.txt) file</a>
</li>
</ul>
</div>
</div>
</div>
</div>
</div>
<div class="vert vert-2" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@b831449e90584e8395077e8cb324af7c">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@b831449e90584e8395077e8cb324af7c" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<p><strong> Research Community:</strong></p>
<p>There is a dedicated research community committed to exploring and addressing these issue of fairness in machine learning and healthcare. These conferences are excellent sources of further reading, and may be worthwhile venues for your own future work!</p>
<ul type="disc">
<li><em><strong>ML for Health Venues:</strong></em></li>
<ul type="circle">
<li>Machine learning for healthcare (<a href="https://www.mlforhc.org/">https://www.mlforhc.org/</a>)</li>
<li>ACM conference on health, inference, and learning (ACM-CHIL) (<a href="https://www.chilconference.org/">https://www.chilconference.org/</a>)</li>
<li>ML for Health / Fair ML for Health NeurIPS Workshops (<a href="https://ml4health.github.io/2019/index.html">https://ml4health.github.io/2019/index.html</a>), (<a href="https://www.fairmlforhealth.com/home">https://www.fairmlforhealth.com/home</a>)</li>
</ul>
<li><em><strong>ML Fairness Venues:</strong></em></li>
<ul type="circle">
<li>ACM Conference on Fairness, Accountability, and Transparency (ACM FAT) (<a href="https://facctconference.org/">https://facctconference.org/</a>)</li>
<li>ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW) (<a href="https://cscw.acm.org/2019/">https://cscw.acm.org/2019/</a>)</li>
</ul>
</ul>
<p></p>
</div>
</div>
<div class="vert vert-3" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@e815926d9ded43b6ad844f4b451d1354">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@e815926d9ded43b6ad844f4b451d1354" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_e815926d9ded43b6ad844f4b451d1354" class="problems-wrapper" role="group"
aria-labelledby="e815926d9ded43b6ad844f4b451d1354-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@e815926d9ded43b6ad844f4b451d1354" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@e815926d9ded43b6ad844f4b451d1354/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="e815926d9ded43b6ad844f4b451d1354-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@e815926d9ded43b6ad844f4b451d1354-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@e815926d9ded43b6ad844f4b451d1354-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>True or False: At worst, machine learning can simply reproduce the existing inequities in healthcare.</p>
<div class="choicegroup capa_inputtype" id="inputtype_e815926d9ded43b6ad844f4b451d1354_2_1">
<fieldset aria-describedby="status_e815926d9ded43b6ad844f4b451d1354_2_1">
<div class="field">
<input type="radio" name="input_e815926d9ded43b6ad844f4b451d1354_2_1" id="input_e815926d9ded43b6ad844f4b451d1354_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="e815926d9ded43b6ad844f4b451d1354_2_1-choice_0-label" for="input_e815926d9ded43b6ad844f4b451d1354_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_e815926d9ded43b6ad844f4b451d1354_2_1"> True
</label>
</div>
<div class="field">
<input type="radio" name="input_e815926d9ded43b6ad844f4b451d1354_2_1" id="input_e815926d9ded43b6ad844f4b451d1354_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="e815926d9ded43b6ad844f4b451d1354_2_1-choice_1-label" for="input_e815926d9ded43b6ad844f4b451d1354_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_e815926d9ded43b6ad844f4b451d1354_2_1"> False
</label>
</div>
<span id="answer_e815926d9ded43b6ad844f4b451d1354_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_e815926d9ded43b6ad844f4b451d1354_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_e815926d9ded43b6ad844f4b451d1354_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_e815926d9ded43b6ad844f4b451d1354" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_e815926d9ded43b6ad844f4b451d1354">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="e815926d9ded43b6ad844f4b451d1354-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="e815926d9ded43b6ad844f4b451d1354-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="e815926d9ded43b6ad844f4b451d1354-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-4" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@743f8ef2716e402a95e03e6e90a32953">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@743f8ef2716e402a95e03e6e90a32953" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_743f8ef2716e402a95e03e6e90a32953" class="problems-wrapper" role="group"
aria-labelledby="743f8ef2716e402a95e03e6e90a32953-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@743f8ef2716e402a95e03e6e90a32953" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@743f8ef2716e402a95e03e6e90a32953/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="743f8ef2716e402a95e03e6e90a32953-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@743f8ef2716e402a95e03e6e90a32953-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@743f8ef2716e402a95e03e6e90a32953-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>Which of the following is NOT a reason that the intersection of machine learning and healthcare is difficult?</p>
<div class="choicegroup capa_inputtype" id="inputtype_743f8ef2716e402a95e03e6e90a32953_2_1">
<fieldset aria-describedby="status_743f8ef2716e402a95e03e6e90a32953_2_1">
<div class="field">
<input type="radio" name="input_743f8ef2716e402a95e03e6e90a32953_2_1" id="input_743f8ef2716e402a95e03e6e90a32953_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="743f8ef2716e402a95e03e6e90a32953_2_1-choice_0-label" for="input_743f8ef2716e402a95e03e6e90a32953_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_743f8ef2716e402a95e03e6e90a32953_2_1"> A deep level of domain knowledge in both computer science and medicine must be present
</label>
</div>
<div class="field">
<input type="radio" name="input_743f8ef2716e402a95e03e6e90a32953_2_1" id="input_743f8ef2716e402a95e03e6e90a32953_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="743f8ef2716e402a95e03e6e90a32953_2_1-choice_1-label" for="input_743f8ef2716e402a95e03e6e90a32953_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_743f8ef2716e402a95e03e6e90a32953_2_1"> Medical data often includes sensitive information that must be protected
</label>
</div>
<div class="field">
<input type="radio" name="input_743f8ef2716e402a95e03e6e90a32953_2_1" id="input_743f8ef2716e402a95e03e6e90a32953_2_1_choice_2" class="field-input input-radio" value="choice_2"/><label id="743f8ef2716e402a95e03e6e90a32953_2_1-choice_2-label" for="input_743f8ef2716e402a95e03e6e90a32953_2_1_choice_2" class="response-label field-label label-inline" aria-describedby="status_743f8ef2716e402a95e03e6e90a32953_2_1"> Healthcare workflows are complex and can be difficult to integrate new tools into
</label>
</div>
<div class="field">
<input type="radio" name="input_743f8ef2716e402a95e03e6e90a32953_2_1" id="input_743f8ef2716e402a95e03e6e90a32953_2_1_choice_3" class="field-input input-radio" value="choice_3"/><label id="743f8ef2716e402a95e03e6e90a32953_2_1-choice_3-label" for="input_743f8ef2716e402a95e03e6e90a32953_2_1_choice_3" class="response-label field-label label-inline" aria-describedby="status_743f8ef2716e402a95e03e6e90a32953_2_1"> It is illegal to perform machine learning on HIPPA-protected data
</label>
</div>
<span id="answer_743f8ef2716e402a95e03e6e90a32953_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_743f8ef2716e402a95e03e6e90a32953_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_743f8ef2716e402a95e03e6e90a32953_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_743f8ef2716e402a95e03e6e90a32953" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_743f8ef2716e402a95e03e6e90a32953">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="743f8ef2716e402a95e03e6e90a32953-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="743f8ef2716e402a95e03e6e90a32953-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="743f8ef2716e402a95e03e6e90a32953-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-5" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@c8fd0449aa414901ade8e680e4d958e0">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@c8fd0449aa414901ade8e680e4d958e0" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_c8fd0449aa414901ade8e680e4d958e0" class="problems-wrapper" role="group"
aria-labelledby="c8fd0449aa414901ade8e680e4d958e0-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@c8fd0449aa414901ade8e680e4d958e0" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@c8fd0449aa414901ade8e680e4d958e0/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="c8fd0449aa414901ade8e680e4d958e0-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@c8fd0449aa414901ade8e680e4d958e0-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@c8fd0449aa414901ade8e680e4d958e0-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>True or False: Users of machine learning systems can have excessively high levels of trust in recommendations they receive</p>
<div class="choicegroup capa_inputtype" id="inputtype_c8fd0449aa414901ade8e680e4d958e0_2_1">
<fieldset aria-describedby="status_c8fd0449aa414901ade8e680e4d958e0_2_1">
<div class="field">
<input type="radio" name="input_c8fd0449aa414901ade8e680e4d958e0_2_1" id="input_c8fd0449aa414901ade8e680e4d958e0_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="c8fd0449aa414901ade8e680e4d958e0_2_1-choice_0-label" for="input_c8fd0449aa414901ade8e680e4d958e0_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_c8fd0449aa414901ade8e680e4d958e0_2_1"> True
</label>
</div>
<div class="field">
<input type="radio" name="input_c8fd0449aa414901ade8e680e4d958e0_2_1" id="input_c8fd0449aa414901ade8e680e4d958e0_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="c8fd0449aa414901ade8e680e4d958e0_2_1-choice_1-label" for="input_c8fd0449aa414901ade8e680e4d958e0_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_c8fd0449aa414901ade8e680e4d958e0_2_1"> False
</label>
</div>
<span id="answer_c8fd0449aa414901ade8e680e4d958e0_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_c8fd0449aa414901ade8e680e4d958e0_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_c8fd0449aa414901ade8e680e4d958e0" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_c8fd0449aa414901ade8e680e4d958e0">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="c8fd0449aa414901ade8e680e4d958e0-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="c8fd0449aa414901ade8e680e4d958e0-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="c8fd0449aa414901ade8e680e4d958e0-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
</div>
</div>
<div class="xblock xblock-public_view xblock-public_view-vertical" data-block-type="vertical" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="VerticalStudentView" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@vertical+block@91b9db98552947d0945bea84336adc84" data-course-id="course-v1:MITx+HST.953x+3T2020">
<h2 class="hd hd-2 unit-title">Ensuring Healthcare Machine Learning Works for All</h2>
<div class="vert-mod">
<div class="vert vert-0" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@c7a2c0a6eec74d14a2cf34043294997f">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@c7a2c0a6eec74d14a2cf34043294997f" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<p><strong>Intro:</strong></p>
<p>Despite the before-discussed concerns regarding fairness in healthcare machine learning, it must not be regarded as a foregone conclusion that this technology will act to increase inequity. Rather, concerted efforts must be taken to ensure that data science acts as an equity-promoting force that improves outcomes for all without increasing disparities. Succeeding in this won't be easy, however. In this section, we will explore preliminary work around the potential for AI to act as a positive force for fairness in healthcare, and we will explore structural and systemic changes that are necessary to ensure that this potential is realized.</p>
<p><strong>Can machine learning act as an equity-promoting force?</strong></p>
<p>Recent scholarship has indicated that there may be substantial potential for data science to in fact <a href="https://www.nature.com/articles/s41591-019-0649-2?draft=collection" target="[object Object]">counter existing biases in healthcare </a>and to <a href="https://journalofethics.ama-assn.org/article/can-ai-help-reduce-disparities-general-medical-and-mental-health-care/2019-02" target="[object Object]">act as an equity-promoting force</a>. At the clinical level, machine learning may be used to audit healthcare datasets and processes , <a href="https://arxiv.org/abs/1902.03731" target="[object Object]">rendering implicit biases measurable and explicit so they might be directly and deliberately addressed</a>. While the "black-box" tendencies of algorithms may be concerning, some have <a href="https://arxiv.org/abs/1606.03490" target="[object Object]">argued that human decision-making is in general no-more explainable</a> - especially when it involves implicit biases that physicians may be unlikely to admit to.</p>
<p>At the system level, there is enthusiasm about the potential for data science to help address disparities in the clinical research pipeline by mobilizing large amounts of routine data from across broad populations. Programs such as <a href="https://www.nejm.org/doi/full/10.1056/NEJMsr1809937" target="[object Object]">the "All of Us" research program</a> are attempting to explicitly improve the inclusion of traditionally under-represented groups in biological information databases.</p>
<p><strong>Recommendations for ML to achieve this goal</strong></p>
<p>It is important for the machine learning community to realize its limitations as it pursues the goal of being an equity-promoting force. The design and implementation of algorithms is not a "silver bullet" that will magically solve all sources of unfairness and inequity in healthcare - it must be treated as a supplement to, rather than replacement for, broader efforts aimed at improving healthcare fairness. In Table 1 we outline a series of recommendations adapted from McCoy et al (forthcoming) for structural reforms in health data science that will act to help ensure that the field works equally for all.</p>
<p></p>
</div>
</div>
<div class="vert vert-1" data-id="block-v1:MITx+HST.953x+3T2020+type@html+block@f62a8ac22e1149c6b68d80c1ec277e97">
<div class="xblock xblock-public_view xblock-public_view-html xmodule_display xmodule_HtmlBlock" data-block-type="html" data-has-score="False" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@html+block@f62a8ac22e1149c6b68d80c1ec277e97" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "HTMLModule"}
</script>
<style type="text/css">
.tg {border-collapse:collapse;border-spacing:0;}
.tg td{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
overflow:hidden;padding:10px 5px;word-break:normal;}
.tg th{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;}
.tg .tg-0lax{text-align:left;vertical-align:top}
</style>
<table class="tg">
<thead>
<tr>
<th class="tg-0lax"> <br><span style="font-weight:bold">Area of Emphasis</span> </th>
<th class="tg-0lax"> <br><span style="font-weight:bold">Recommendations</span> </th>
</tr>
</thead>
<tbody>
<tr>
<td class="tg-0lax"> <br>Ensure Health Data Science is Equitable By Design </td>
<td class="tg-0lax"> <br> <br> <br>Develop pipelines for the promotion of diverse teams in all aspects of health data science<br> <br>Ensure the inclusion of data from a broad range of groups, in a broad range of contexts<br> <br>Incorporate global partners to ensure health data science promotes global health equity. </td>
</tr>
<tr>
<td class="tg-0lax"> <br>Encourage Public and Open Health Data Science Research </td>
<td class="tg-0lax"> <br> <br> <br>Fund both direct HDS research and research into ethical aspects of HDS<br> <br>Harmonize ethical oversight between public and private research domains </td>
</tr>
<tr>
<td class="tg-0lax"> <br>Ensure Adequate Access to Health Information Technology Infrastructure </td>
<td class="tg-0lax"> <br> <br> <br>Ensure all are included in the datasets by funding health data gathering infrastructure in underserved communities<br> <br>Develop HDS products with an awareness of the broad range of health IT contexts for deployment </td>
</tr>
<tr>
<td class="tg-0lax"> <br>Ensure Health Data Science is Clinically Effective and Impactful </td>
<td class="tg-0lax"> <br> <br> <br>Ensure the presence of multidisciplinary teams that represent both clinical and data science perspectives<br> <br>Promote pathways for interdisciplinary training<br> <br>Hold HDS innovations to the same standards as other healthcare interventions, including requirements for prospective validation and clear demonstration of impact </td>
</tr>
<tr>
<td class="tg-0lax"> <br>Audit Health Data Science on Ethical Metrics </td>
<td class="tg-0lax"> <br> <br> <br>Mandate assessments of the performance of novel HDS technology for impacts on marginalized and intersectional groups. <br> <br>Record racial data necessary to perform these audits in an ongoing fashion </td>
</tr>
<tr>
<td class="tg-0lax"> <br>Mandate Transparency in Data Collection, Analysis, and Usage </td>
<td class="tg-0lax"> <br> <br> <br>Build patient trust by ensuring that protocols for the collection, analysis, and usage of data are transparent and open </td>
</tr>
<tr>
<td class="tg-0lax"> <br>Promote Inclusive and Interoperable Data Policy </td>
<td class="tg-0lax"> <br> <br> <br>Ensure the existence of clear and ethical methods for ensuring the sharing of data between different sources while protecting patient rights and privacy<br> <br>Ensure that global partners are included, so that interoperability barriers do not hinder inclusive global collaboration </td>
</tr>
</tbody>
</table>
</div>
</div>
<div class="vert vert-2" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@98d6a04afe354f66aa7002b57b76678a">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@98d6a04afe354f66aa7002b57b76678a" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_98d6a04afe354f66aa7002b57b76678a" class="problems-wrapper" role="group"
aria-labelledby="98d6a04afe354f66aa7002b57b76678a-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@98d6a04afe354f66aa7002b57b76678a" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@98d6a04afe354f66aa7002b57b76678a/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="98d6a04afe354f66aa7002b57b76678a-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@98d6a04afe354f66aa7002b57b76678a-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@98d6a04afe354f66aa7002b57b76678a-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>True or false: Machine learning can only act as an inequity-promoting force in healthcare</p>
<div class="choicegroup capa_inputtype" id="inputtype_98d6a04afe354f66aa7002b57b76678a_2_1">
<fieldset aria-describedby="status_98d6a04afe354f66aa7002b57b76678a_2_1">
<div class="field">
<input type="radio" name="input_98d6a04afe354f66aa7002b57b76678a_2_1" id="input_98d6a04afe354f66aa7002b57b76678a_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="98d6a04afe354f66aa7002b57b76678a_2_1-choice_0-label" for="input_98d6a04afe354f66aa7002b57b76678a_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_98d6a04afe354f66aa7002b57b76678a_2_1"> True
</label>
</div>
<div class="field">
<input type="radio" name="input_98d6a04afe354f66aa7002b57b76678a_2_1" id="input_98d6a04afe354f66aa7002b57b76678a_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="98d6a04afe354f66aa7002b57b76678a_2_1-choice_1-label" for="input_98d6a04afe354f66aa7002b57b76678a_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_98d6a04afe354f66aa7002b57b76678a_2_1"> False
</label>
</div>
<span id="answer_98d6a04afe354f66aa7002b57b76678a_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_98d6a04afe354f66aa7002b57b76678a_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_98d6a04afe354f66aa7002b57b76678a_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_98d6a04afe354f66aa7002b57b76678a" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_98d6a04afe354f66aa7002b57b76678a">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="98d6a04afe354f66aa7002b57b76678a-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="98d6a04afe354f66aa7002b57b76678a-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="98d6a04afe354f66aa7002b57b76678a-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
<div class="vert vert-3" data-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@5a6c9d90b59c42b69d941dc1662eced5">
<div class="xblock xblock-public_view xblock-public_view-problem xmodule_display xmodule_ProblemBlock" data-block-type="problem" data-has-score="True" data-runtime-class="LmsRuntime" data-runtime-version="1" data-init="XBlockToXModuleShim" data-request-token="f826b20243c511efaedb16ffcfe1c469" data-graded="False" data-usage-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@5a6c9d90b59c42b69d941dc1662eced5" data-course-id="course-v1:MITx+HST.953x+3T2020">
<script type="json/xblock-args" class="xblock-json-init-args">
{"xmodule-type": "Problem"}
</script>
<div id="problem_5a6c9d90b59c42b69d941dc1662eced5" class="problems-wrapper" role="group"
aria-labelledby="5a6c9d90b59c42b69d941dc1662eced5-problem-title"
data-problem-id="block-v1:MITx+HST.953x+3T2020+type@problem+block@5a6c9d90b59c42b69d941dc1662eced5" data-url="/courses/course-v1:MITx+HST.953x+3T2020/xblock/block-v1:MITx+HST.953x+3T2020+type@problem+block@5a6c9d90b59c42b69d941dc1662eced5/handler/xmodule_handler"
data-problem-score="0"
data-problem-total-possible="1"
data-attempts-used="0"
data-content="
<h3 class="hd hd-3 problem-header" id="5a6c9d90b59c42b69d941dc1662eced5-problem-title" aria-describedby="block-v1:MITx+HST.953x+3T2020+type@problem+block@5a6c9d90b59c42b69d941dc1662eced5-problem-progress" tabindex="-1">
Multiple Choice
</h3>
<div class="problem-progress" id="block-v1:MITx+HST.953x+3T2020+type@problem+block@5a6c9d90b59c42b69d941dc1662eced5-problem-progress"></div>
<div class="problem">
<div>
<div class="wrapper-problem-response" tabindex="-1" aria-label="Question 1" role="group"><p>True or False: Effective data science can solve all of medicine's problems with bias and discrimination </p>
<div class="choicegroup capa_inputtype" id="inputtype_5a6c9d90b59c42b69d941dc1662eced5_2_1">
<fieldset aria-describedby="status_5a6c9d90b59c42b69d941dc1662eced5_2_1">
<div class="field">
<input type="radio" name="input_5a6c9d90b59c42b69d941dc1662eced5_2_1" id="input_5a6c9d90b59c42b69d941dc1662eced5_2_1_choice_0" class="field-input input-radio" value="choice_0"/><label id="5a6c9d90b59c42b69d941dc1662eced5_2_1-choice_0-label" for="input_5a6c9d90b59c42b69d941dc1662eced5_2_1_choice_0" class="response-label field-label label-inline" aria-describedby="status_5a6c9d90b59c42b69d941dc1662eced5_2_1"> True
</label>
</div>
<div class="field">
<input type="radio" name="input_5a6c9d90b59c42b69d941dc1662eced5_2_1" id="input_5a6c9d90b59c42b69d941dc1662eced5_2_1_choice_1" class="field-input input-radio" value="choice_1"/><label id="5a6c9d90b59c42b69d941dc1662eced5_2_1-choice_1-label" for="input_5a6c9d90b59c42b69d941dc1662eced5_2_1_choice_1" class="response-label field-label label-inline" aria-describedby="status_5a6c9d90b59c42b69d941dc1662eced5_2_1"> False
</label>
</div>
<span id="answer_5a6c9d90b59c42b69d941dc1662eced5_2_1"/>
</fieldset>
<div class="indicator-container">
<span class="status unanswered" id="status_5a6c9d90b59c42b69d941dc1662eced5_2_1" data-tooltip="Not yet answered.">
<span class="sr">unanswered</span><span class="status-icon" aria-hidden="true"/>
</span>
</div>
</div><div class="solution-span">
<span id="solution_5a6c9d90b59c42b69d941dc1662eced5_solution_1"/>
</div></div>
</div>
<div class="action">
<input type="hidden" name="problem_id" value="Multiple Choice" />
<div class="submit-attempt-container">
<button type="button" class="submit btn-brand" data-submitting="Submitting" data-value="Submit" data-should-enable-submit-button="True" aria-describedby="submission_feedback_5a6c9d90b59c42b69d941dc1662eced5" >
<span class="submit-label">Submit</span>
</button>
<div class="submission-feedback" id="submission_feedback_5a6c9d90b59c42b69d941dc1662eced5">
<span class="sr">Some problems have options such as save, reset, hints, or show answer. These options follow the Submit button.</span>
</div>
</div>
<div class="problem-action-buttons-wrapper">
</div>
</div>
<div class="notification warning notification-gentle-alert
is-hidden"
tabindex="-1">
<span class="icon fa fa-exclamation-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="5a6c9d90b59c42b69d941dc1662eced5-problem-title">
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification warning notification-save
is-hidden"
tabindex="-1">
<span class="icon fa fa-save" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="5a6c9d90b59c42b69d941dc1662eced5-problem-title">None
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
<div class="notification general notification-show-answer
is-hidden"
tabindex="-1">
<span class="icon fa fa-info-circle" aria-hidden="true"></span>
<span class="notification-message" aria-describedby="5a6c9d90b59c42b69d941dc1662eced5-problem-title">Answers are displayed within the problem
</span>
<div class="notification-btn-wrapper">
<button type="button" class="btn btn-default btn-small notification-btn review-btn sr">Review</button>
</div>
</div>
</div>
"
data-graded="False">
<p class="loading-spinner">
<i class="fa fa-spinner fa-pulse fa-2x fa-fw"></i>
<span class="sr">Loading…</span>
</p>
</div>
</div>
</div>
</div>
</div>
© All Rights Reserved