HOW MANY MORE PATIENTS WILL WE TREAT BY DECREASING THE THRESHOLD?

Purpose: Most physicians are afraid to harm non-diseased patients by unnecessary treatment. The absence of perfect reference tests in daily practice forces physicians to use multiple tests to diagnose a specific disease. The threshold theory describes how to find a pivotal disease probability balancing disutility and benefit of treating versus not treating. The ‘disutility’ of unnecessary treatment tends to be overestimated so that the threshold increases. In the past, we studied the relation between post-test probabilities and thresholds in three real life cohorts. This research aims at acquiring better insight in general into the distribution of disease probabilities after multiple testing and the effect of lowering the treatment threshold on the number of treated diseased and non-diseased patients.

Method: Simulation of patient distributions after applying 10 independent tests. Calculation of the number of treated diseased and not-diseased patients with different prevalences and test accuracies at different thresholds.

Result: Lowering the threshold from 95% to 50% and 5% in simulations with a prevalence of 50%, 10% and 1% after tests with high accuracy has nearly no influence on the number of patients treated because the diseased and non-diseased patients groups are nearly perfectly separated. In real situations, using tests with lower accuracy, lowering the threshold will increase the number of treated patients substantially. More diseased patients are treated than non-diseased down to a threshold of 5% and prevalence of 10%. Under these values it is the opposite, but nearly all diseased patients are treated. 

Conclusion: After 10 high accuracy tests lowering the threshold has nearly no influence on the number of treated patients. Using low accuracy tests, lowering the threshold increases the number of treated patients substantially and more diseased than non-diseased patients will be treated down to threshold of 5% and a prevalence of 10%.