JMIR Publications

ABSTRACT

Background:

Symptom checker apps are patient-facing decision support systems aimed at providing advice to laypersons on whether, where and how to seek healthcare (triage advice). Such triage advice can improve laypersons’ self-assessment prior to seeking medical care and thus, ultimately improve medical outcomes and disburden the healthcare system by making patient journeys more efficient. Past research focused on evaluating how accurate the advice of symptom checker apps is. However, in order to support decision making for the better, such apps need to not only provide accurate but also trustworthy advice, as users will not act upon advice from decision aids they do not trust. To date only very few studies addressed the question to what extent laypersons trust symptom checker app advice or which factors moderate their trust. Studies on decision support systems showed that the framing of the technology affects users’ trust in automated systems. For instance, with laypersons’ trust in healthcare professionals being generally high, some symptom checker apps use an anthropomorphic framing displaying icons or pictures of physicians on the user interface to make it look more like a healthcare professional. Others emphasize the expertise of technical systems using icons symbolizing, for instance, an artificial intelligence (AI).

Objective:

Our study aims at identifying factors influencing laypersons’ trust in advice provided by symptom checker apps. Primarily, we investigate whether designs using an anthropomorphic framing or framing the app as an AI increase users’ trust compared to no such framing. Furthermore, we explored whether users’ trust is influenced by sociodemographic factors, users’ decisional certainty in their own triage appraisal, and the level of urgency of the advice provided by the app.

Methods:

Through an online survey we recruited US residents with no professional medical training. Participants had to first appraise the urgency of a fictitious patient description (case vignette). Subsequently, a decision aid (mock symptom checker app) provided triage advice contradicting the participants’ stand-alone appraisal, and then participants had to appraise the vignette again. Participants were randomized into three groups, two experimental groups using a visual framing (anthropomorphic vs. AI) and a neutral one without such framing. We employed the Trust in Automated Systems Survey to determine subjective trust, and the proportion of participants following the app’s advice (follow rates) as a proxy for behavioral trust. We used an ANOVA to test whether framing affects trust. Additionally, we used linear and logistic regression to determine the influence of other independent variables on users’ trust.

Results:

Most participants (77.7% of N = 494) followed the decision aid’s advice, regardless of whether this advice was of higher or lower urgency than the participant’s own assessment. Neither anthropomorphic framing nor framing as AI increased subjective or behavioral trust compared to the no-frame condition. Even participants who were extremely certain in their own decision commonly changed it in favor of the symptom checker’s advice. Propensity to trust and eHealth literacy were associated with increased subjective trust in the symptom checker, whereas sociodemographic variables showed no such link with both subjective and behavioral trust.

Conclusions:

As most participants trusted the mock app’s advice, even when very certain of their own assessment, the question arises whether laypersons use such symptom checkers as substitute rather than as aid in their own decision making. Contrary to our expectation, the high trust in health care professionals did not result in a trust evoking effect of anthropomorphic framing, and neither did the emphasis of artificial intelligence. Thus, other factors may be responsible for evoking a high level of trust, for example how symptom checkers substantiate their triage recommendation.