You are using an old and unsupported browser. Most core functionality will not work. Please upgrade to a supported browser - Google Chrome


Agreement between clinician - and model-generated melanoma risk (EPOSTER: 4mins)

Presentation Description

Background: Improvements in clinical information systems has seen a growing use of risk prediction models in chronic disease management. For example, cardiovascular risk prediction models have been adopted into systems to assist with the risk stratification of patients and subsequent management of hypertension. We identified 28 published melanoma risk prediction models in a systematic review however none have been integrated into clinical systems to assist clinicians in estimating melanoma risk. We aimed to assess whether unassisted clinician-generated melanoma risk predictions agree with model-generated melanoma risk predictions.

Method: We used a cross-sectional design. Participants were recruited through GPs Down Under, a Facebook group comprising over 6000 authenticated general practitioners (GPs) from Australia and New Zealand. GP participants completed an online survey with questions on: (1) their overall melanoma risk in both absolute and relative terms, and (2) melanoma risk factors as identified in a validated melanoma risk prediction model to enable the calculation of absolute and relative risk. The relation between clinician- and model- generated melanoma risk prediction (both absolute and relative melanoma risk) was assessed using Pearson correlation coefficients and correlation plots.

Results: 136 of the 150 GP respondents completed the online survey between June to August 2019. The Pearson correlation coefficient for clinician- and model-generated melanoma risk prediction was 0.20 (95% CI 0.03 to 0.36) for remaining lifetime absolute melanoma risk and 0.60 (95% CI 0.48 to 0.70) for relative melanoma risk. There was a tendency for participants to overestimate risk when it is low, and underestimate risk when it is high.

Discussion: This is the first study to compare clinician-generated melanoma risk assessments against a well-validated and prospectively evaluated model. It showed poor correlation between clinician-reported against model-generated melanoma risk.

Implications for practice: Further work is needed on understanding the clinical impact of risk discordance.