The scrutiny of identifying community-acquired pneumonia episodes quantified bias in absolute effect estimation in a population-based pneumococcal vaccination trial

Objectives To determine the accurateness of detecting community-acquired pneumonia (CAP) in the Community-Acquired Pneumonia immunization Trial in Adults (CAPiTA), a community-based, double-blind, randomized placebo-controlled trial in which the needed to treat (NNT) for prevention of vaccine-type pneumococcal CAP was 1,007 [95% confidence interval (CI): 613, 2,646]. Study Design and Setting Study participants developing pneumonia were identified in 58 participating hospitals by research nurses (RNs) using local-adapted protocols. In addition, general practitioner (GP) records were screened for hospital referrals for suspected pneumonia. Two independent reviewers determined reasons for not identifying pneumonia episodes, and the NNT adjusted for missed episodes was estimated. Results Of 2,183 hospital referrals with suspected pneumonia detected in GP records, 232 (11%) were admitted outside established screening routes and 102 (5%) were not suspected of pneumonia on admission. Of the remaining 1,849 episodes, 1,374 (63% of all episodes and 74% of identifiable episodes) were identified by RNs. Several causes of missing episodes were identified. After adjustment for missed episodes, the NNT reduced to 634 (95% CI: 386, 1,675). Conclusion With the screening procedure, 63% of suspected pneumonia episodes were identified, and the estimated NNT reduced from 1,007 to 634. Root cause analysis of unidentified episodes provides guidance for improving pneumonia detection in future trials.