Boosting medical diagnostics by pooling independent judgments

Area of Science:

• Decision Sciences
• Medical Diagnostics
• Artificial Intelligence

Background:

• Collective intelligence enhances group problem-solving, with potential applications in medicine, economics, and politics.
• Understanding the conditions for effective collective intelligence in real-world scenarios, particularly medical diagnostics, remains limited.

Purpose of the Study:

• To investigate the conditions under which combining multiple doctors' judgments surpasses the performance of the best individual doctor in medical diagnostics.
• To identify key factors influencing collective intelligence in breast and skin cancer detection using real-world data.

Main Methods:

• A simulation study utilizing large-scale, real-world datasets.
• Analysis of over 20,000 diagnoses made by more than 140 doctors in breast and skin cancer detection.
• Examination of how diagnostic accuracy similarity impacts group performance compared to individual experts.

Main Results:

• Collective intelligence, where group decisions outperform the best individual, is significantly enhanced when doctors exhibit similar diagnostic accuracy.
• When diagnostic accuracy varies widely among doctors, aggregation of judgments does not consistently outperform the top performer.
• The positive effect of accuracy similarity is robust across various group sizes and performance levels, explained by its impact on overruling individual decisions.

Conclusions:

• Similarity in diagnostic accuracy is a critical determinant for successful collective intelligence in medical diagnosis.
• Findings suggest that optimizing group composition based on accuracy similarity can lead to more effective diagnostic decision-making.
• This research provides a foundation for developing improved strategies for complex real-world decision-making processes.