AI skin cancer diagnostics: scoping review finds little evidence of clinical or economic value

Human idea, AI creation10 Nov 2026Oncology

A scoping review of 33 studies found that evaluations of AI for skin cancer mostly report process and user-experience outcomes. Few studies examined clinical, prognostic, system-level or economic outcomes, and most were done in controlled settings.

Abstract dark blue cell with glowing cyan core and yellow highlights dissolving into pixels on a dark background — Image: human idea, AI creation

Summarised study

Ocker-Serger L, Livingstone E, Lodde G, Stockfleth E, Abu Rached N, Schadendorf D, et al. From algorithm performance to clinical value in AI-assisted skin cancer diagnostics. Front Digit Health. 2026;8:1896521. doi:10.3389/fdgth.2026.1896521. PMID: 42935948. PubMed

Key findings

Of 4,021 records identified, 33 studies were eligible for inclusion, while 389 publications were excluded for reporting only technical performance metrics.
Process and experiential outcomes, such as user acceptance, workflow efficiency and diagnostic confidence, were reported in 28/33 studies.
Just five studies examined clinical and prognostic endpoints: four examining prognostic value, and one on patient management.
Three studies assessed system-level and economic endpoints.
The majority of evaluations took place under controlled experimental conditions, and fewer studies examined prospective implementation in real-world practice.

Background

AI has shown strong diagnostic performance for detecting skin cancer and is increasingly being built into clinical workflows¹. However, it is unclear whether these gains lead to outcomes that matter for patients and health systems. Value-oriented evidence of this kind had not previously been synthesised systematically¹.

Methods

The authors performed a scoping review following PRISMA-ScR¹. They searched Web of Science, Scopus and PubMed/MEDLINE for studies published from January 2015 to October 2025¹.

Eligible studies assessed AI technologies for diagnosing, triaging, risk-stratifying or managing skin cancer¹. Studies also had to report endpoints other than measures of technical accuracy¹. Outcomes were grouped into three domains¹:

clinical and prognostic outcomes
process and experiential outcomes
economic and system-level outcomes

Results

Of 4,021 records, 33 studies were included. A further 389 papers were left out for reporting technical performance alone¹.
Process and experiential outcomes dominated (28/33 studies). These included user acceptance, efficiency of workflow and confidence in diagnosis¹.
Only five studies reported clinical and prognostic outcomes. Four addressed prognostic utility and one addressed patient management¹.
Three studies reported system-level and economic outcomes¹.
Most evaluations took place in controlled experimental settings. Prospective studies of real-world implementation were less frequent¹.

Interpretation

The authors note that evaluation of AI skin cancer diagnostics remains largely performance-oriented and concentrated on experience-related and process-related endpoints¹.

From a value-based healthcare viewpoint, the evidence sits mostly at intermediate points in the diagnostic value chain¹. AI support may improve workflows, clinician confidence and decision making¹. However, few studies have examined downstream effects on patient-relevant clinical and prognostic outcomes, or on system-level and economic outcomes¹.

The authors call for three kinds of research: prospective real-world studies, evaluation frameworks driven by outcomes, and economic assessments of how AI-supported decisions affect clinical outcomes and health systems¹. They conclude that, despite encouraging effects on diagnostic processes, the evidence of clinical and system-level value is still limited¹.

References

Ocker-Serger L, Livingstone E, Lodde G, Stockfleth E, Abu Rached N, Schadendorf D, et al. From algorithm performance to clinical value in AI-assisted skin cancer diagnostics. Front Digit Health. 2026;8:1896521. doi:10.3389/fdgth.2026.1896521. PMID: 42935948. PubMed

This article is not medical advice. Always check the source and local prescribing information before making treatment decisions. Patients should seek a personal assessment by a licensed physician.

Medically reviewed by Drs van den Borne, Stijn on 10 Nov 2026