Algorithm-based advice taking and clinical judgement: impact of advice distance and algorithm information.

Area of Science:

• Clinical decision-making
• Medical informatics
• Human-computer interaction

Background:

• Evidence-based algorithms can enhance judgments but are underutilized.
• Humans often discount advice, especially when it conflicts with their own opinions (egocentric advice discounting).
• Previous research on advice discounting used low-stakes tasks and student participants.

Purpose of the Study:

• To investigate advice discounting principles within the clinical domain using the judge-advisor framework.
• To assess how general practitioners (GPs) integrate algorithmic advice from a cancer risk calculator into their clinical judgment.
• To determine if informing GPs about the algorithm's performance impacts their advice-taking behavior.

Main Methods:

• Utilized realistic patient scenarios and algorithmic advice from a validated cancer risk calculator.
• Recruited general practitioners (GPs) as participants.
• Measured the weight GPs assigned to their own estimates versus the algorithmic advice, with half receiving information on the algorithm's accuracy.

Main Results:

• GPs, on average, weighed their estimates and the algorithm equally.
• GPs retained initial estimates 29% of the time and fully updated them 27% of the time.
• Informing GPs about the algorithm's derivation, validation, and accuracy did not significantly alter their updating behavior. A slight discount was observed when advice significantly differed from their own estimate.

Conclusions:

• GPs demonstrate a balanced approach, integrating algorithmic advice with their own judgment.
• The findings present a more optimistic view of expert advice-taking than previously suggested by egocentric discounting models.
• Experts can effectively weigh and move towards algorithmic recommendations, even when they challenge initial assessments, suggesting potential for improved clinical decision support.