Diagnostic accuracy of FEVER-Pain and Centor criteria for bacterial throat infection in adults with sore throat: a secondary analysis of a randomised controlled trial

Background: Sore throat is a common and self-limiting condition. There remains ambiguity in stratifying patients to immediate, delayed or no antibiotic prescriptions. NICE recommends two clinical prediction rules (CPRs), FeverPAIN and CENTOR, to guide decision-making. Aim: Describe the diagnostic accuracy of CPRs in identifying streptococcal throat infections. Design and Setting: Adults presenting to UK primary care with sore throat, who did not require immediate antibiotics. Methods: As part of the Treatment Options without Antibiotics for Sore Throat (TOAST) trial, 565 participants, aged ≥ 18, were recruited on day of presentation to general practice. Physicians could opt to give delayed prescriptions. CPR scores were not part of the trial protocol but calculated post-hoc from baseline assessments. Diagnostic accuracy was calculated by comparing scores to throat swab cultures. Results: 81/502 (16.1%) of patients had Group A, C or G streptococcus cultured on throat swab. Overall diagnostic accuracy of both CPRs was poor: area under receiver-operating-characteristics curve 0.62 for Centor; 0.59 for FeverPAIN. Post-test probability of a positive or negative test was 27.3% (95% confidence intervals: 6.0-61.0%) and 84.1% (80.6-87.2%) for FeverPAIN ≥4, versus 25.7% (16.2-37.2%) and 85.5% (81.8-88.7%) for Centor ≥3. Higher CPR scores were associated with increased delayed antibiotic prescriptions (2 = 8.42, p=0.004 for FeverPAIN ≥4; 2 = 32.0, p<0.001 for Centor ≥3). Conclusions: In those that do not require immediate antibiotics in primary care, neither CPR provide a reliable way of diagnosing streptococcal throat infection. However, clinicians were more likely to give delayed prescriptions to those with higher scores.