Study purpose: Jutzi et al. (2020, Frontiers in Medicine) conducted the first comprehensive survey of patient attitudes toward artificial intelligence in melanoma diagnostics, comparing views between patients with a history of melanoma and those without, to understand acceptance barriers before clinical AI deployment.
Survey approach: An anonymous online questionnaire was distributed through German university hospital partnerships and melanoma support groups from November 2019 to January 2020, targeting 298 respondents including 154 with a previous melanoma diagnosis.
Core finding: 94% of respondents supported AI use in medicine, but only 41% were amenable to AI as a stand-alone diagnostic system, while 94% endorsed AI as an assistance system for physicians - reflecting a near-universal preference for human oversight over AI diagnostic autonomy.
Clinical translation insight: Understanding patient perspectives before deploying AI is essential for designing acceptable clinical workflows; this survey identifies specific patient preferences, concerns, and use-case preferences that should inform how AI is introduced into melanoma diagnostics.
Survey development: The questionnaire was designed de novo (no validated instrument existed for this purpose), informed by a literature review and dermato-oncology expert feedback. Five lay volunteers with no AI background tested comprehensibility before distribution, and the survey required German language proficiency.
Distribution channels: The survey was distributed via email to university hospital dermato-oncology partnerships and melanoma support groups (for patient recruitment), and advertised on social media, specifically to enable comparison between melanoma-experienced and melanoma-naive respondents.
Participant demographics: Of 298 respondents, 73.2% were female, 51.7% had a previous melanoma diagnosis, 41.3% were aged 46-60 years, and 66.4% held a university admission qualification or degree - demographics not representative of the general German population, skewed toward higher-educated and cancer-affected individuals.
Statistical analysis: Descriptive statistics with 95% confidence intervals were calculated for all primary outcomes. Wilcoxon signed rank and chi-squared tests compared responses between prespecified subgroups (age, sex, education, melanoma history).
High general acceptance: 88% of respondents had prior awareness of AI, 94% supported AI use in medicine, and 88% were willing to share their anonymized health data to further AI development - indicating a highly informed and pro-AI attitude among this melanoma-adjacent population.
Assistance vs. stand-alone divide: While 94% endorsed AI as an assistance system for physicians, only 41% accepted its use as a stand-alone diagnostic system and 56% were willing to use AI for home-based screening (e.g., smartphone apps), showing a clear hierarchy of acceptance depending on physician involvement.
Top concerns: The three most frequently cited concerns were data protection (risk of insurance or employer misuse, hacker exposure), impersonality (sterile consultations, reduced physician-patient relationship), and susceptibility to errors (physicians becoming dependent on AI and losing diagnostic skills).
Top expectations: Patients expected AI to deliver faster diagnoses, reduce pathologist waiting times, enable earlier detection, reduce diagnostic errors and unnecessary biopsies, and provide more objective and less subjective lesion assessments than human observers.
Higher acceptance among melanoma patients: Respondents with a previous melanoma diagnosis were significantly more likely to support AI in medicine (97% vs. 91%, p=0.03) and more open to home-based AI screening apps (66% vs. 46%, p=0.004) compared to those without a melanoma history.
Independent assessment preference: Melanoma patients preferred an application scenario where the physician and AI classify the lesion independently and a biopsy is taken if either recommends it - a higher sensitivity approach. Non-melanoma respondents more often preferred the physician simply incorporating AI results into their own diagnosis.
Sex differences: More female than male participants indicated they would have a lesion removed if either the physician or the AI recommended it (83-93% vs. 70-83%, p<0.02), suggesting sex influences risk tolerance in the context of AI-physician disagreement.
Confidence in AI vs. physicians: When AI and physician recommendations conflicted, 89% would follow the physician's recommendation for biopsy even if AI disagreed, but if AI had proven superior accuracy, 91% would follow AI. Accuracy-based trust was the primary determinant of confidence allocation.
Physician-assistance model preferred: The near-universal (94%) preference for AI as an assistance rather than stand-alone system establishes a clear design requirement for clinical AI: systems should present AI recommendations as one input to physician decision-making, not autonomous diagnoses.
Sensitivity-specificity tradeoff guidance: The melanoma patient preference for independent AI-physician assessment with biopsy if either recommends it suggests patients would accept a high-sensitivity (low false negative) AI even at the cost of more unnecessary biopsies - an important input for algorithm threshold calibration.
Home screening application design: The 56% willingness to use home smartphone screening apps, combined with concerns about algorithm quality and misclassification anxiety, suggests that smartphone melanoma apps require rigorous clinical validation and clear patient communication about limitations before deployment.
Communication requirements: The concern about AI decision non-traceability and missing transparency directly maps to the interpretability requirements in CLEAR Derm guidelines - patients want to understand why AI flagged a lesion, reinforcing the clinical and ethical case for explainable AI in dermoscopy.
Non-representative sample: The survey population was predominantly female (73%), highly educated (66% university level), and drawn disproportionately from melanoma support groups - not representative of the general German population or international patient populations with different cultural attitudes toward AI.
Social desirability bias: Online survey participants who actively sought out a melanoma-related AI survey likely have greater pre-existing interest in and positive disposition toward AI compared to typical clinic patients, potentially inflating positive attitude percentages.
German-language only: The German-only survey limits generalizability to other cultural and linguistic contexts; cultural attitudes toward AI, physician authority, and data privacy vary significantly across countries and would affect patient acceptance patterns.
Future directions: Larger, demographically representative surveys including patients across multiple countries, languages, and socioeconomic backgrounds are needed. Longitudinal surveys tracking attitude change as AI tools become more visible in clinical settings, and intervention studies testing whether patient education about AI changes acceptance, would provide actionable insights for deployment planning.