Do General Practitioners’ risk assessments and antibiotic prescribing decisions follow the STARWAVe clinical prediction rule?

Problem

When children present with cough in primary care, prognostic uncertainty can lead to defensive antibiotic prescribing (“treat, just in case”). To combat this, a clinical prediction rule called “STARWAVe” was developed and validated. STARWAVe uses seven clinical factors (Short illness duration, Temperature, Age, Recession, Wheeze, Asthma, Vomiting) to differentiate children at “very low” (0.3%, ≤1 factor present), “normal” (1.5%, 2/3 factors present) and “high” (11.8%, ≥4 factors present) risk of deterioration. In so doing, it aims to reduce prognostic uncertainty and unnecessary prescribing in non-high-risk cases. Providing STARWAVe as a decision aid to General Practitioners (GPs) could improve risk assessment and prescribing decisions. However, risk scores are merely probabilities and could be ignored, especially if they contradict the decision-maker's intuitive assessment of risk. We aimed to compare GPs’ intuitive risk assessments and prescribing decisions to those of STARWAVe. We also explored two different methods of eliciting risk estimates, and assessed the impact of parental worry.

Approach

188 UK GPs were randomly assigned to view 32 (of 64) vignettes depicting children with cough. Vignettes comprised the seven STARWAVe factors, varied in a fractional factorial design. Per vignette, GPs estimated risk of deterioration in one of two ways (self-generated percentage vs. category selection; randomly assigned) and indicated how they would manage the patient (“prescribe antibiotics”, “review within 24hrs” and/or “admit for paediatric assessment”; GPs could tick all that applied). GPs then saw an additional vignette, suggesting that “the parent is quite concerned”. Using mixed-effects regression, we measured the influence of STARWAVe factors, risk elicitation method, and parental concern on risk estimates and prescribing decisions.

Findings

Relative to STARWAVe, GPs underestimated risk in 15% of cases (877/6016) and overestimated it in 30% (1776/6016). Consistent with STARWAVe, younger patient age increased GPs’ risk estimates, as did fever, recession, wheeze, asthma, and vomiting (ORs>1.49, ps<0.001). Inconsistent with STARWAVe, shorter illness duration reduced them (OR=0.83, p=0.003).The prescribing rate was low (13%, 797/6016), with GPs preferring to admit high risk cases for paediatric assessment (70%, 2074/2972) and/or review within 24hrs (34%, 1014/2972), vs. prescribe (13%, 390/2972). Still, 51% of prescriptions were unnecessary relative to GPs’ own risk estimates (407/797), and 69% relative to STARWAVe risk estimates (550/797). Factors increasing the odds of an unnecessary prescription were long illness duration, asthma, fever, and wheeze (ORs>1.54, ps<0.05).Risk estimates and prescribing odds were higher when GPs expressed risk as a percentage (vs. category selection; ORs>1.52, ps<0.042). Parental concern increased GPs’ risk estimates (OR=2.42, p<.001) but not prescribing odds (OR=0.74, p=0.378).

Consequences

Relative to STARWAVe, GPs sometimes overestimated risk of deterioration and prescribed unnecessarily. Systematically, they misinterpreted illness duration, prescribing for longer rather than shorter illnesses. Providing STARWAVe as a decision aid necessitates that GPs are aware of and agree with its assessment of clinical factors.