DOES MEDICAL EDUCATION PROVIDE A SHARED UNDERSTANDING OF THE TREATMENT THRESHOLD PROBABILITY FOR SUSPECTED STREP THROAT?

Purpose: Assuming physicians have a threshold probability at which they’d give antibiotics for a suspected strep throat, to measure that threshold in four ways, comparing the means, variabilities, and correlations among the methods and as a function of medical experience.

Method(s): A web survey was promoted among convenience samples of primary care clinicians and residents, medical and physician assistant students, undergraduate students, and patients.  Respondents provided judgments from which 4 measures of their treatment threshold probability for sore throat could be calculated: direct statement of threshold, judgment of 4 utilities (u(TN), u(TP),u( FN), and u( FP)), judgment of 2 utility differences (u(TP) - u(FN) and u(TN) - u(FP)), and person tradeoff judgments (number of people experiencing the less severe error that would be equal to one person experiencing the more severe error). Wording variants and question presentation orders were randomized. Survey asked demographics, parenting experience, and clinical experience of strep throat and of bad outcomes associated with misses and with unnecessary antibiotics. 

Result(s): 950 started survey and 735 (77.4%) finished. Each threshold method was noisy, with responses ranging  from 0 to 100%. Responses were characterized as invalid, doubtful, worrisome, and reasonable. Those who judged only some methods before quitting made more unreasonable responses. Intercorrelations among measures ranged from -0.03 to 0.39 when all responses were included, and from 0.61 to 0.75 when only reasonable responses were considered. Mean directly stated treatment threshold probabilities (0.59) were higher than the mean thresholds calculated from component judgments (each 0.43). Thresholds showed a U trend over medical education, with 3^rdyear medical and PA students having the lowest thresholds (method means 0.26 - 0.41), while patients (means 0.45 - 0.61) and practicing clinicians (means 0.42 - 0.62) had the highest thresholds.

Data did not support that medical education makes individuals agree more with each other about a threshold value.  

Physicians reported familiarity with the concept of a treatment threshold probability, but few reported explicitly comparing a patient’s disease probability to a treatment threshold.

Conclusion(s): Treatment threshold probability judgments are very noisy. Different methods yield different thresholds. Agreement increases when obvious and suspected mistakes are discarded, suggesting people may have an implicit threshold. Perhaps more explicit instruction and communication regarding recommended treatment thresholds could build on this.