Rethinking cholera diagnostic test performance, interpretation, and evaluation: a field-based latent-class analysis in Bangladesh

BACKGROUND: Accurate and reliable diagnostics, including rapid diagnostic tests (RDTs), are crucial components of cholera control programmes, although their estimated performance has varied greatly across studies. The aim of this study was to assess cholera diagnostics performance accounting for possible sources of variability, including reference assay choice and patient-level and sampling characteristics, and the implications on result interpretation and test performance evaluation. METHODS: We enrolled all individuals aged 1 year and older presenting with suspected cholera seeking care at two health-care facilities in Sitakunda, Bangladesh. All participants (or, if younger than 18 years, their legal guardians) provided written informed consent and were given a short structured questionnaire on patient history and demographics, alongside a rectal swab or stool sample. All stool samples were tested with the CholKit Rapid Diagnostic Test (CholKit RDT; Incepta, Dhaka, Bangladesh), and a subset of samples (all positive RDTs and a random subset of approximately half of negative RDTs) were tested by PCR and culture. Test performance was estimated using a latent-class Bayesian framework accounting for imperfect test performance, incomplete PCR and culture testing, and time-varying changes in cholera incidence. Patient-level factor effects (including age age 1-4 years vs ≥5 years and previous antibiotic use) and sampling factor effects (season and testing delays) were estimated, and simulations were used to assess the bias in RDT performance estimates for sensitivity and specificity with 95% credible intervals (CrIs) when using traditional reference assays. FINDINGS: Between Jan 24, 2021, and Aug 31, 2022, we enrolled 3744 individuals with suspected cholera. Of the 3744 overall participants, 1918 (51·2%) were male and 1826 (48·8%) were female; 1095 (29·2%) were aged 1-4 years and 2649 (70·8%) were 5 years and older. Among the suspected cases of cholera, 692 (18·5%) participants tested positive by the CholKit RDT. Among the RDT-positive samples, 573 (82·8%) also tested positive by PCR, and 450 (65·0%) tested positive by culture. For RDT, PCR, and culture, we estimated a sensitivity of 93·5% (95% CrI 91·3-95·4), 90·3% (88·4-92·1), and 73·7% (70·8-76·5), respectively; and a specificity of 97·3% (96·7-97·8) and 97·2% (96·6-97·8) for RDT and PCR, respectively. Culture specificity was assumed perfect at 100%. We found that younger age (1-4 years), antibiotic use, and testing delays decreased culture sensitivity, but RDT performance remained relatively constant. The RDT positive predictive value ranged from 80% in participants 5 years and older, varying greatly across seasons. Simulations of field trials demonstrated underestimation of RDT sensitivity in low prevalence settings when evaluated against PCR, and underestimation of specificity in high prevalence settings regardless of the reference assay. INTERPRETATION: Our results provide potential mechanisms leading to the heterogeneous cholera RDT performance estimates in previous studies, including the use of culture as a reference assay. Across various patient and sampling characteristics, CholKit RDT had high performance in this cholera-endemic setting, supporting its use for cholera surveillance and control. Accounting for epidemiologic context is crucial both for individual-level clinical test interpretation, and for the future evaluation of diagnostics such as RDTs. FUNDING: The Gates Foundation.