CoChat

Psychological Factors Influencing Appropriate Reliance on AI-enabled Clinical Decision Support Systems: Experimental Web-Based Study Among Dermatologists

Open access

Published version of a preprint

The DOI matches its Crossref record; 75 of the 78 references it cites were found in a registry or an index; 16 of the 84 sources its claims cite were checked for support.

Abstract

Background: Artificial intelligence (AI)–enabled decision support systems are critical tools in medical practice; however, their reliability is not absolute, necessitating human oversight for final decision-making. Human reliance on such systems can vary, influenced by factors such as individual psychological factors and physician experience. Objective: This study aimed to explore the psychological factors influencing subjective trust and reliance on medical AI’s advice, specifically examining relative AI reliance and relative self-reliance to assess the appropriateness of reliance. Methods: A survey was conducted with 223 dermatologists, which included lesion image classification tasks and validated questionnaires assessing subjective trust, propensity to trust technology, affinity for technology interaction, control beliefs, need for cognition, as well as queries on medical experience and decision confidence. Results: A 2-tailed t test revealed that participants’ accuracy improved significantly with AI support (t222=−3.3; P<.001; Cohen d=4.5), but only by an average of 1% (1/100). Reliance on AI was stronger for correct advice than for incorrect advice (t222=4.2; P<.001; Cohen d=0.1). Notably, participants demonstrated a mean relative AI reliance of 10.04% (139/1384) and a relative self-reliance of 85.6% (487/569), indicating a high level of self-reliance but a low level of AI reliance. Propensity to trust technology influenced AI reliance, mediated by trust (indirect effect=0.024, 95% CI 0.008-0.042; P<.001), and medical experience negatively predicted AI reliance (indirect effect=–0.001, 95% CI –0.002 to −0.001; P<.001). Conclusions: The findings highlight the need to design AI support systems in a way that assists less experienced users with a high propensity to trust technology to identify potential AI errors, while encouraging experienced physicians to actively engage with system recommendations and potentially reassess initial decisions.

Overview

CoChat Summary

## TL;DR

This preregistered web-based experiment with 223 dermatologists found that participants relied more on correct than incorrect AI advice, but overall showed low AI reliance and high self-reliance. Trust in the system and medical experience were the main predictors of reliance: higher trust increased reliance, while greater medical experience reduced trust and AI reliance.

## Key findings

  • Among 223 dermatologists, first-response accuracy was 63% (15/24) and final accuracy with AI support was 65% (15/24), an average increase of 1% (SD 4.5%).
  • Reliance was much more self-directed than AI-directed: RAIR was 10.04% (139/1384) and RSR was 85.6% (487/569).
  • Participants relied less strongly on incorrect AI advice than on correct advice; the difference in WoA was significant (t\_222 = 4.2; P < .001).
  • Trust predicted reliance (β = 0.05; P < .001), explaining 5.8% of the variance in reliance (R² = 0.1); trust was positively related to RAIR and negatively related to RSR.
  • Propensity to trust technology had a significant indirect effect on reliance through trust (indirect effect = 0.024; 95% CI 0.008-0.042), while medical experience showed a significant negative indirect effect on reliance through trust (indirect effect = −0.001; 95% CI −0.002 to −0.001).
  • Affinity for technology interaction, control beliefs in dealing with technology, and need for cognition were not significant predictors of trust or reliance.

## Methods

  • Web-based preregistered repeated-measures experiment conducted from June to September 2023 with 223 professional dermatologists (out of 256 responses) in Germany; mean experience was 15.5 years (SD 11.2; range 0-50).
  • Participants classified 24 skin lesion image cases as benign or malignant in a 2-step task: an initial unaided judgment, then a revised judgment after seeing AI advice said to be 80% accurate; 5 of 24 AI recommendations were deliberately incorrect and randomized.
  • Main analyses included a t test for reliance differences between correct vs incorrect AI advice, regression and moderation analyses for trust/reliance, and PROCESS-based mediation analyses for psychological factors and medical experience; the planned structural equation model was dropped because model fit was unacceptable.

## Limitations

  • The paper states that the AI only provided the final classification, so participants could not engage with explanations or other background information about the model’s validity.
  • Confidence was measured with a proxy based on slider position and dichotomized, which the paper notes may be less sensitive than directly asking participants about confidence.
  • AI-misclassified cases were randomly selected, which the paper says may have introduced uncontrolled variation in case difficulty.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.2196/58660

Cites

75 of 75 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Example of the 2-step classification task.
Figure 1 Example of the 2-step classification task.CC BY · PubMed Central
Figure 2 Mediation model for hypothesis 4 regarding the weight of advice. AI: artificial intelligence.
Figure 2 Mediation model for hypothesis 4 regarding the weight of advice. AI: artificial intelligence.CC BY · PubMed Central
Figure 3 Mediation model for hypothesis 5 regarding the weight of advice. AI: artificial intelligence.
Figure 3 Mediation model for hypothesis 5 regarding the weight of advice. AI: artificial intelligence.CC BY · PubMed Central

Verification

Claims the paper makes

9 Valid · 0 Not valid · 40 Not checked · 78 Own results and interpretations

28 more claims

Create a free account to see all 127 claims and to verify the 68 references still open.

See all claims — free account
  • Probably not a substantial claim. 59
  • It may state two separate claims as one. 49
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

5 of 8 · the rest aren’t loaded

3 more findings

Create a free account to see every finding of this check.

See all findings — free account

Citations

Cites

Show all 75 references

Research works printed but not linked: 5

  1. Agrawal, V, arXiv
  2. HandrichMAlexa, you freak me out – identifying drivers of innovation resistance and adoption of intelligent personal assistantsProceedings of Second International Conference on Intelligent System 20212021ICIS 2021December 12-15, 2021Austin, TX
  3. Beier, G, 1999, Report Psychologie, 9, 684
  4. JessupSASchneiderTRAlarconGMRyanTJCapiolaAThe measurement of the propensity to trust automationProceedings of the 11th International Conference on Virtual, Augmented and Mixed Reality. Applications and Case Studies2019VAMR 2019July 26-31, 2019Orlando, FL
  5. Papenmeier, A, arXiv

Cited by

See all 31

Metadata

Authors
Küper, A., Lodde, G. C., Livingstone, E., Schadendorf, D., & Krämer, N.
Affiliations
  • University of Duisburg-Essen (DE)
  • Essen University Hospital (DE)
Corresponding author
Küper, Alisa
Venue
Journal of Medical Internet Research
Published
2025-04-04
Volume
27
Pages
e58660
ISSN
1438-8871 (ISSN-L)
OpenAlex
W4408250977
Cited by
31
References
78
Citations by year
2026: 27 · 2025: 4
FWCI
7.41
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Journal of Medical Internet Research (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Universitätsbibliographie, Universität Duisburg-Essen (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Deutsche Forschungsgemeinschaft
  • Universität Duisburg-Essen
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Social Sciences › Human Factors and Ergonomics › Human-Automation Interaction and Safety
  • Health Sciences › Health Professions › Health Information Management › Electronic Health Records Systems
Keywords
dermatologists, clinical experience, self-reliance, control beliefs, need for cognition
MeSH terms
Dermatologists, Adult, Artificial Intelligence, Female, Humans, Male, Middle Aged, Surveys and Questionnaires, Decision Support Systems, Clinical, Internet, Trust

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.