The validity of test-negative design for assessment of typhoid conjugate vaccine protection: comparison of estimates by different study designs using data from a cluster-randomised controlled trial.

BACKGROUND Typhoid fever remains a substantial public health challenge in low-income and middle-income countries. By 2023, typhoid conjugate vaccines (TCVs) had been introduced in six countries globally, with more than 50 million doses distributed. Now that TCVs are being deployed, there is a need for observational studies to assess vaccine effectiveness in the field. We aimed to evaluate the validity of different observational study designs in estimating vaccine protection. METHODS We compared different observational and experimental study designs for assessing vaccine effectiveness by re-analysing data from the TyVAC Bangladesh trial, a participant-blinded and observer-blinded cluster-randomised controlled trial done in Mirpur, Dhaka, Bangladesh. 150 geographical clusters were randomly assigned (1:1) to receive either TCV or Japanese encephalitis vaccine. Eligible children aged 9 months to 15 years were offered a single dose of the vaccine randomly assigned to their cluster of residence, and baseline vaccination was done between April 15 and May 15, 2018. We compared estimates of vaccine effectiveness from the cluster-randomised controlled trial analysis-which assessed the risk of blood-culture-confirmed typhoid fever among recipients of TCV versus recipients of Japanese encephalitis vaccine-with estimates from cohort study and test-negative case-control study design (TND) analyses, which compared recipients of TCV with non-vaccinees in the 75 geographical clusters where TCV was administered. We further conducted negative-control exposure (NCE) and negative-control outcome (NCO) analyses as bias indicators. FINDINGS 41 344 (67%) of 62 025 age-eligible children in the study area received the TCV or Japanese encephalitis vaccine during the baseline vaccination campaign. Among the 62 025 age-eligible children, 5582 blood-culture specimens were collected by passive surveillance, including 2546 (46%) specimens from the 75 TCV clusters. The estimated vaccine efficacy was 89% (95% CI 81-93) in the cluster-randomised controlled trial analysis, 79% (70-86) by the cohort design, 88% (79-93) by the TND when pan-negatives were used as test-negative controls, and 90% (75-96) by the TND when specimens positive for pathogens other than Salmonella enterica serotype Typhi were used as test-negative controls. Using NCE analysis, Japanese encephalitis vaccination was associated with an increased risk of typhoid fever compared with non-vaccinees in the 75 Japanese encephalitis clusters in the cohort design (incidence rate ratio 1·98 [95% CI 1·56-2·52]), but no significant association between Japanese encephalitis vaccination and typhoid fever was found with the TND. Similarly, an increased risk of non-typhoid infections was observed in the cohort NCO analyses when comparing vaccinees with non-vaccinees in both Japanese encephalitis vaccine clusters and TCV clusters, but not in the TND NCO analyses. INTERPRETATION Our findings suggests that the TND provides reliable estimates of TCV effectiveness, whereas the cohort design can bias vaccine effectiveness estimates, possibly due to unmeasured confounding effects, such as health-care-seeking behaviours. NCE and NCO approaches are useful tools for identifying such biases. FUNDING The Bill & Melinda Gates Foundation.