Exploring and Promoting Diagnostic Transparency and Explainability in Online Symptom Checkers

Tsai, Chwei-Shyong; You, You; Gui, Xinning; Kou, Yubo; Carroll, John M.

doi:10.1145/3411764.3445101

Cited by 42 publications

(41 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…HCI studies on XAI have found explanations to improve user understanding of AI systems [12,22,32], and somewhat mixed results on enhancing user trust [22], satisfaction [32] and willingness to adopt AI systems [80]. Moreover, explanations provide additional information that can be utilized to assist the task that people perform.…”

Section: Ai Explanations For Human-ai Interactionmentioning

confidence: 99%

Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation

Lai

Carton

Bhatnagar

et al. 2022

CHI Conference on Human Factors in Computing Systems

View full text Add to dashboard Cite

Despite impressive performance in many benchmark datasets, AI models can still make mistakes, especially among out-of-distribution examples. It remains an open question how such imperfect models can be used effectively in collaboration with humans. Prior work has focused on AI assistance that helps people make individual high-stakes decisions, which is not scalable for a large amount of relatively low-stakes decisions, e.g., moderating social media comments. Instead, we propose conditional delegation as an alternative paradigm for human-AI collaboration where humans create rules to indicate trustworthy regions of a model. Using content moderation as a testbed, we develop novel interfaces to assist humans in creating conditional delegation rules and conduct a randomized experiment with two datasets to simulate in-distribution and outof-distribution scenarios. Our study demonstrates the promise of conditional delegation in improving model performance and provides insights into design for this novel paradigm, including the effect of AI explanations.

show abstract

Section: Ai Explanations For Human-ai Interactionmentioning

confidence: 99%

Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation

Lai

Carton

Bhatnagar

et al. 2022

CHI Conference on Human Factors in Computing Systems

View full text Add to dashboard Cite

show abstract

“…Moreover, users may be unclear about the technology behind web-based symptom checkers. Research suggests that web-based symptom checkers’ artificial intelligence (AI) systems are neither transparent nor comprehensible to users, which may undermine trust in such tools [ 10 ]. Nevertheless, despite hesitancy and concerns regarding the accuracy, AI-powered symptom checkers have been perceived as useful for diagnosis by users [ 11 ].…”

Section: Introductionmentioning

confidence: 99%

Health Information Seeking From an Intelligent Web-Based Symptom Checker: Cross-sectional Questionnaire Study

Carmona¹,

Chittamuru²,

Kravitz³

et al. 2022

J Med Internet Res

View full text Add to dashboard Cite

Background The ever-growing amount of health information available on the web is increasing the demand for tools providing personalized and actionable health information. Such tools include symptom checkers that provide users with a potential diagnosis after responding to a set of probes about their symptoms. Although the potential for their utility is great, little is known about such tools’ actual use and effects. Objective We aimed to understand who uses a web-based artificial intelligence–powered symptom checker and its purposes, how they evaluate the experience of the web-based interview and quality of the information, what they intend to do with the recommendation, and predictors of future use. Methods Cross-sectional survey of web-based health information seekers following the completion of a symptom checker visit (N=2437). Measures of comprehensibility, confidence, usefulness, health-related anxiety, empowerment, and intention to use in the future were assessed. ANOVAs and the Wilcoxon rank sum test examined mean outcome differences in racial, ethnic, and sex groups. The relationship between perceptions of the symptom checker and intention to follow recommended actions was assessed using multilevel logistic regression. Results Buoy users were well-educated (1384/1704, 81.22% college or higher), primarily White (1227/1693, 72.47%), and female (2069/2437, 84.89%). Most had insurance (1449/1630, 88.89%), a regular health care provider (1307/1709, 76.48%), and reported good health (1000/1703, 58.72%). Three types of symptoms—pain (855/2437, 35.08%), gynecological issues (293/2437, 12.02%), and masses or lumps (204/2437, 8.37%)—accounted for almost half (1352/2437, 55.48%) of site visits. Buoy’s top three primary recommendations split across less-serious triage categories: primary care physician in 2 weeks (754/2141, 35.22%), self-treatment (452/2141, 21.11%), and primary care in 1 to 2 days (373/2141, 17.42%). Common diagnoses were musculoskeletal (303/2437, 12.43%), gynecological (304/2437, 12.47%) and skin conditions (297/2437, 12.19%), and infectious diseases (300/2437, 12.31%). Users generally reported high confidence in Buoy, found it useful and easy to understand, and said that Buoy made them feel less anxious and more empowered to seek medical help. Users for whom Buoy recommended “Waiting/Watching” or “Self-Treatment” had strongest intentions to comply, whereas those advised to seek primary care had weaker intentions. Compared with White users, Latino and Black users had significantly more confidence in Buoy (P<.05), and the former also found it significantly more useful (P<.05). Latino (odds ratio 1.96, 95% CI 1.22-3.25) and Black (odds ratio 2.37, 95% CI 1.57-3.66) users also had stronger intentions to discuss recommendations with a provider than White users. Conclusions Results demonstrate the potential utility of a web-based health information tool to empower people to seek care and reduce health-related anxiety. However, despite encouraging results suggesting the tool may fulfill unmet health information needs among women and Black and Latino adults, analyses of the user base illustrate persistent second-level digital divide effects.

show abstract

“…The core dimensions of the negative wellbeing outcomes theme was related to a lack of explanation of diagnostic decision making within the results report. This has also been identified in previous reports in relation to Artificial Intelligence (AI), which highlighted that users of online symptom checkers wish to be provided an explanation for the results reached based upon their personal data (103). Ensuring that users are aware of how results of digital assessments were reached may potentially increase trust, and encourage users to follow personalized triage recommendations (104).…”

Section: Theme Combination Example Feedback Comment Frequencymentioning

confidence: 69%

Learnings from user feedback of a novel digital mental health assessment

et al. 2022

View full text Add to dashboard Cite

Digital mental health interventions (DMHI) have the potential to address barriers to face-to-face mental healthcare. In particular, digital mental health assessments offer the opportunity to increase access, reduce strain on services, and improve identification. Despite the potential of DMHIs there remains a high drop-out rate. Therefore, investigating user feedback may elucidate how to best design and deliver an engaging digital mental health assessment. The current study aimed to understand 1304 user perspectives of (1) a newly developed digital mental health assessment to determine which features users consider to be positive or negative and (2) the Composite International Diagnostic Interview (CIDI) employed in a previous large-scale pilot study. A thematic analysis method was employed to identify themes in feedback to three question prompts related to: (1) the questions included in the digital assessment, (2) the homepage design and reminders, and (3) the assessment results report. The largest proportion of the positive and negative feedback received regarding the questions included in the assessment (n = 706), focused on the quality of the assessment (n = 183, 25.92% and n = 284, 40.23%, respectively). Feedback for the homepage and reminders (n = 671) was overwhelmingly positive, with the largest two themes identified being positive usability (i.e., ease of use; n = 500, 74.52%) and functionality (i.e., reminders; n = 278, 41.43%). The most frequently identified negative theme in results report feedback (n = 794) was related to the report content (n = 309, 38.92%), with users stating it was lacking in-depth information. Nevertheless, the most frequent positive theme regarding the results report feedback was related to wellbeing outcomes (n = 145, 18.26%), with users stating the results report, albeit brief, encouraged them to seek professional support. Interestingly, despite some negative feedback, most users reported that completing the digital mental health assessment has been worthwhile (n = 1,017, 77.99%). Based on these findings, we offer recommendations to address potential barriers to user engagement with a digital mental health assessment. In summary, we recommend undertaking extensive co-design activities during the development of digital assessment tools, flexibility in answering modalities within digital assessment, customizable additional features such as reminders, transparency of diagnostic decision making, and an actionable results report with personalized mental health resources.

show abstract

Exploring and Promoting Diagnostic Transparency and Explainability in Online Symptom Checkers

Cited by 42 publications

References 53 publications

Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation

Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation

Health Information Seeking From an Intelligent Web-Based Symptom Checker: Cross-sectional Questionnaire Study

Learnings from user feedback of a novel digital mental health assessment

Contact Info

Product

Resources

About