Decision Modeling of Disagreements

Abstract

Purpose. To identify core disagreements between pediatric hematologists who would treat children with idiopathic thrombocytopenic purpura (ITP) on initial presentation (“Treaters”) and those who would treat by observation (“Observers”), to determine whether each group’s preferred stance was consistent with each individual’s detailed perceptions, and to identify key variables in each stance. Methods. A decision model was constructed in collaboration with experts, and a detailed questionnaire was presented to a nationally representative committee of 25 pediatric hematologists. A full decision tree was specified for each respondent. Results. Nineteen (76%) experts responded; based on preference for initial treatment, 9 were Treaters and 10 Observers. Of the 30 probability/effectiveness variables, 8—almost all concerning treatment effectiveness—had at least one statistically-significant difference between the 2 groups regarding low, best, or high estimates. To convince Observers that treatment is effective would take a clinical trial with between 39 000 and 87 000 participants; to convince Treaters that treatment is not effective enough, between 97 000 and 114 000 participants. Observers’ calculated numbers needed to treat (NNTs) of about 150 000 are more consistent ( P = 0.0023) with their elicited maximum NTTs of about 500. Conclusion. Physicians not specifically trained provided enough data to specify complete individual decision models. From the estimates provided, no practical clinical trial could convince hematologists who would treat children on initial presentation with ITP just to simply observe them or could convince those who would just observe to instead treat with available agents. Perceived burdens could be better characterized, perhaps by including parental perceptions and preferences.