Dataset Information

The Impact of Explanations on Layperson Trust in Artificial Intelligence-Driven Symptom Checker Apps: Experimental Study.

ABSTRACT:

Background

Artificial intelligence (AI)-driven symptom checkers are available to millions of users globally and are advocated as a tool to deliver health care more efficiently. To achieve the promoted benefits of a symptom checker, laypeople must trust and subsequently follow its instructions. In AI, explanations are seen as a tool to communicate the rationale behind black-box decisions to encourage trust and adoption. However, the effectiveness of the types of explanations used in AI-driven symptom checkers has not yet been studied. Explanations can follow many forms, including why-explanations and how-explanations. Social theories suggest that why-explanations are better at communicating knowledge and cultivating trust among laypeople.

Objective

The aim of this study is to ascertain whether explanations provided by a symptom checker affect explanatory trust among laypeople and whether this trust is impacted by their existing knowledge of disease.

Methods

A cross-sectional survey of 750 healthy participants was conducted. The participants were shown a video of a chatbot simulation that resulted in the diagnosis of either a migraine or temporal arteritis, chosen for their differing levels of epidemiological prevalence. These diagnoses were accompanied by one of four types of explanations. Each explanation type was selected either because of its current use in symptom checkers or because it was informed by theories of contrastive explanation. Exploratory factor analysis of participants' responses followed by comparison-of-means tests were used to evaluate group differences in trust.

Results

Depending on the treatment group, two or three variables were generated, reflecting the prior knowledge and subsequent mental model that the participants held. When varying explanation type by disease, migraine was found to be nonsignificant (P=.65) and temporal arteritis, marginally significant (P=.09). Varying disease by explanation type resulted in statistical significance for input influence (P=.001), social proof (P=.049), and no explanation (P=.006), with counterfactual explanation (P=.053). The results suggest that trust in explanations is significantly affected by the disease being explained. When laypeople have existing knowledge of a disease, explanations have little impact on trust. Where the need for information is greater, different explanation types engender significantly different levels of trust. These results indicate that to be successful, symptom checkers need to tailor explanations to each user's specific question and discount the diseases that they may also be aware of.

Conclusions

System builders developing explanations for symptom-checking apps should consider the recipient's knowledge of a disease and tailor explanations to each user's specific need. Effort should be placed on generating explanations that are personalized to each user of a symptom checker to fully discount the diseases that they may be aware of and to close their information gap.

SUBMITTER: Woodcock C

PROVIDER: S-EPMC8600426 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:BACKGROUND:Patients are increasingly seeking Web-based symptom checkers to obtain diagnoses. However, little is known about the characteristics of the patients who use these resources, their rationale for use, and whether they find them accurate and useful. OBJECTIVE:The study aimed to examine patients' experiences using an artificial intelligence (AI)-assisted online symptom checker. METHODS:An online survey was administered between March 2, 2018, through March 15, 2018, to US users of the Isabel Symptom Checker within 6 months of their use. User characteristics, experiences of symptom checker use, experiences discussing results with physicians, and prior personal history of experiencing a diagnostic error were collected. RESULTS:A total of 329 usable responses was obtained. The mean respondent age was 48.0 (SD 16.7) years; most were women (230/304, 75.7%) and white (271/304, 89.1%). Patients most commonly used the symptom checker to better understand the causes of their symptoms (232/304, 76.3%), followed by for deciding whether to seek care (101/304, 33.2%) or where (eg, primary or urgent care: 63/304, 20.7%), obtaining medical advice without going to a doctor (48/304, 15.8%), and understanding their diagnoses better (39/304, 12.8%). Most patients reported receiving useful information for their health problems (274/304, 90.1%), with half reporting positive health effects (154/302, 51.0%). Most patients perceived it to be useful as a diagnostic tool (253/301, 84.1%), as a tool providing insights leading them closer to correct diagnoses (231/303, 76.2%), and reported they would use it again (278/304, 91.4%). Patients who discussed findings with their physicians (103/213, 48.4%) more often felt physicians were interested (42/103, 40.8%) than not interested in learning about the tool's results (24/103, 23.3%) and more often felt physicians were open (62/103, 60.2%) than not open (21/103, 20.4%) to discussing the results. Compared with patients who had not previously experienced diagnostic errors (missed or delayed diagnoses: 123/304, 40.5%), patients who had previously experienced diagnostic errors (181/304, 59.5%) were more likely to use the symptom checker to determine where they should seek care (15/123, 12.2% vs 48/181, 26.5%; P=.002), but they less often felt that physicians were interested in discussing the tool's results (20/34, 59% vs 22/69, 32%; P=.04). CONCLUSIONS:Despite ongoing concerns about symptom checker accuracy, a large patient-user group perceived an AI-assisted symptom checker as useful for diagnosis. Formal validation studies evaluating symptom checker accuracy and effectiveness in real-world practice could provide additional useful information about their benefit.

Project description:BackgroundSymptom checkers are digital tools assisting laypersons in self-assessing the urgency and potential causes of their medical complaints. They are widely used but face concerns from both patients and health care professionals, especially regarding their accuracy. A 2015 landmark study substantiated these concerns using case vignettes to demonstrate that symptom checkers commonly err in their triage assessment.ObjectiveThis study aims to revisit the landmark index study to investigate whether and how symptom checkers' capabilities have evolved since 2015 and how they currently compare with laypersons' stand-alone triage appraisal.MethodsIn early 2020, we searched for smartphone and web-based applications providing triage advice. We evaluated these apps on the same 45 case vignettes as the index study. Using descriptive statistics, we compared our findings with those of the index study and with publicly available data on laypersons' triage capability.ResultsWe retrieved 22 symptom checkers providing triage advice. The median triage accuracy in 2020 (55.8%, IQR 15.1%) was close to that in 2015 (59.1%, IQR 15.5%). The apps in 2020 were less risk averse (odds 1.11:1, the ratio of overtriage errors to undertriage errors) than those in 2015 (odds 2.82:1), missing >40% of emergencies. Few apps outperformed laypersons in either deciding whether emergency care was required or whether self-care was sufficient. No apps outperformed the laypersons on both decisions.ConclusionsTriage performance of symptom checkers has, on average, not improved over the course of 5 years. It decreased in 2 use cases (advice on when emergency care is required and when no health care is needed for the moment). However, triage capability varies widely within the sample of symptom checkers. Whether it is beneficial to seek advice from symptom checkers depends on the app chosen and on the specific question to be answered. Future research should develop resources (eg, case vignette repositories) to audit the capabilities of symptom checkers continuously and independently and provide guidance on when and to whom they should be recommended.

Project description:Antimalarial drugs are becoming less effective due to the emergence of drug resistance. Resistance has been reported for all available malaria drugs, including artemisinin, thus creating a perpetual need for alternative drug candidates. The traditional drug discovery approach of high throughput screening (HTS) of large compound libraries for identification of new drug leads is time-consuming and resource intensive. While virtual in silico screening is a solution to this problem, however, the generalization of the models is not ideal. Artificial intelligence (AI), utilizing either structure-based or ligand-based approaches, has demonstrated highly accurate performances in the field of chemical property prediction. Leveraging the existing data, AI would be a suitable alternative to blind-search HTS or fingerprint-based virtual screening. The AI model would learn patterns within the data and help to search for hit compounds efficiently. In this work, we introduce DeepMalaria, a deep-learning based process capable of predicting the anti-Plasmodium falciparum inhibitory properties of compounds using their SMILES. A graph-based model is trained on 13,446 publicly available antiplasmodial hit compounds from GlaxoSmithKline (GSK) dataset that are currently being used to find novel drug candidates for malaria. We validated this model by predicting hit compounds from a macrocyclic compound library and already approved drugs that are used for repurposing. We have chosen macrocyclic compounds as these ligand-binding structures are underexplored in malaria drug discovery. The in silico pipeline for this process also consists of additional validation of an in-house independent dataset consisting mostly of natural product compounds. Transfer learning from a large dataset was leveraged to improve the performance of the deep learning model. To validate the DeepMalaria generated hits, we used a commonly used SYBR Green I fluorescence assay based phenotypic screening. DeepMalaria was able to detect all the compounds with nanomolar activity and 87.5% of the compounds with greater than 50% inhibition. Further experiments to reveal the compounds' mechanism of action have shown that not only does one of the hit compounds, DC-9237, inhibits all asexual stages of Plasmodium falciparum, but is a fast-acting compound which makes it a strong candidate for further optimization.

Project description:BackgroundArtificial intelligence (AI) has shown potential to improve diagnostics of various diseases, especially for early detection of skin cancer. Studies have yet to investigate the clear application of AI technology in clinical practice or determine the added value for younger user groups. Translation of AI-based diagnostic tools can only be successful if they are accepted by potential users. Young adults as digital natives may offer the greatest potential for successful implementation of AI into clinical practice, while at the same time, representing the future generation of skin cancer screening participants.ObjectiveWe conducted an anonymous online survey to examine how and to what extent individuals are willing to accept AI-based mobile apps for skin cancer diagnostics. We evaluated preferences and relative influences of concerns, with a focus on younger age groups.MethodsWe recruited participants below 35 years of age using three social media channels-Facebook, LinkedIn, and Xing. Descriptive analysis and statistical tests were performed to evaluate participants' attitudes toward mobile apps for skin examination. We integrated an adaptive choice-based conjoint to assess participants' preferences. We evaluated potential concerns using maximum difference scaling.ResultsWe included 728 participants in the analysis. The majority of participants (66.5%, 484/728; 95% CI 0.631-0.699) expressed a positive attitude toward the use of AI-based apps. In particular, participants residing in big cities or small towns (P=.02) and individuals that were familiar with the use of health or fitness apps (P=.02) were significantly more open to mobile diagnostic systems. Hierarchical Bayes estimation of the preferences of participants with a positive attitude (n=484) revealed that the use of mobile apps as an assistance system was preferred. Participants ruled out app versions with an accuracy of ≤65%, apps using data storage without encryption, and systems that did not provide background information about the decision-making process. However, participants did not mind their data being used anonymously for research purposes, nor did they object to the inclusion of clinical patient information in the decision-making process. Maximum difference scaling analysis for the negative-minded participant group (n=244) showed that data security, insufficient trust in the app, and lack of personal interaction represented the dominant concerns with respect to app use.ConclusionsThe majority of potential future users below 35 years of age were ready to accept AI-based diagnostic solutions for early detection of skin cancer. However, for translation into clinical practice, the participants' demands for increased transparency and explainability of AI-based tools seem to be critical. Altogether, digital natives between 18 and 24 years and between 25 and 34 years of age expressed similar preferences and concerns when compared both to each other and to results obtained by previous studies that included other age groups.