Diagnoses: When Are Several Opinions Better Than One?
Study investigates conditions for the emergence of collective intelligence
Methods of collective intelligence can result in considerably more accurate medical diagnoses, but only under certain conditions. A study headed by researchers at the Max Planck Institute for Human Development has investigated how group composition affects the outcomes of collective decision making. The results have been published in the online edition of the Proceedings of the National Academy of Sciences of the United States of America (PNAS).
The accuracy of medical decisions can be improved by combining several independent opinions. Studies conducted at the Max Planck Institute for Human Development and the Leibniz-Institute of Freshwater Ecology and Inland Fisheries have already found evidence for the benefits of collective intelligence in the context of skin and breast cancer diagnostics. In a follow-up study, the researchers have now examined how the diagnostic accuracy of individual doctors affects the collective diagnostic outcome. “Collective intelligence is a promising approach to making better decisions. We were interested in which conditions have to be met for the group’s decision to be better than that of the best individual in the group,” says Ralf Kurvers, lead author of the study and researcher in the Center for Adaptive Rationality at the Max Planck Institute for Human Development.
The study shows that the diagnostic accuracy of the doctors whose diagnoses are combined has to be similar. Only then can the collective outperform the best individual in the group. If, in contrast, doctors’ levels of accuracy differ too much, combining their decisions leads to worse diagnostic outcomes. This effect holds across different group sizes and different performance levels of the best group member. “It is not the case that groups always make the best decisions. If individual abilities differ too much within the group, it makes more sense to rely on the best diagnostician in the group,” says Ralf Kurvers.
For their study, the researchers used two large data sets available from previous studies on breast and skin cancer diagnostics. They were thus able to draw on more than 20,000 diagnoses made by more than 140 doctors to determine individual diagnostic accuracy. They used this information to identify the conditions under which diagnoses made using collective intelligence rules are more accurate than the diagnoses of the best individual. Specifically, they applied the choose-the-most-confident rule and the majority rule. The choose-the-most-confident rule adopts the diagnosis of the doctor who has the highest confidence in his/her diagnosis; the majority rule takes the diagnosis given by the most doctors.
“Our findings represent another major step in understanding how collective intelligence emerges,” says co-author Max Wolf, who investigates collective intelligence in natural settings at the Leibniz-Institute of Freshwater Ecology and Inland Fisheries. The new findings underline how important the diagnostic accuracy of individual doctors is for the overall outcome. Diagnostic accuracy should therefore be a key criterion for assembling groups in medical diagnostics—for example, in the context of independent double reading of mammograms. In future work, the researchers plan to find out what information is needed to gauge a doctor’s diagnostic accuracy as quickly as possible.
Kurvers, R. H. J. M., Herzog, S. M., Hertwig, R., Krause, J., Carney, P. A., Bogart, A., Argenziano, G., Zalaudek, I., & Wolf, M. (2016). Boosting medical diagnostics by pooling independent judgments. Proceedings of the National Academy of Sciences of the United States of America. Advance online publication. doi:10.1073/pnas.1601827113
Kurvers, R. H. J. M, Krause, J., Argenziano, G., Zalaudek, I., & Wolf, M. (2015). Detection accuracy of collective intelligence assessments for skin cancer diagnosis. JAMA Dermatology, 151(12), 1– 8. doi:10.1001/jamadermatol.2015.3149