Latent class analysis for the diagnosis of Clostridioides difficile infection

Background Clostridioides difficile infection (CDI) is an opportunistic disease that lacks a gold standard test. Nucleic acid amplification tests (NAATs) such as real-time PCR demonstrate excellent an limit of detection (LOD) whereas antigenic methods are able to detect free toxin. Latent class analysis (LCA) provides an unbiased statistical approach to resolving true disease. Methods A cross-sectional study was conducted with suspected CDI patients (n=96). Four commercial real-time PCR tests, toxin antigen detection by enzyme immunoassay (EIA), toxigenic culture, and fecal calprotectin were performed. CDI clinical diagnosis was determined by consensus majority of three experts. LCA was performed using laboratory and clinical variables independent of any gold standard. Results Six LCA models were generated to determine CDI probability using four variables including toxin EIA, toxigenic culture, clinical diagnosis, and fecal calprotectin levels. Three defined zones as a function of real-time PCR cycle threshold (Ct) were identified using LCA: CDI likely (>90% probability), equivocal (<90% and >10%), CDI unlikely (<10%). A single model comprising toxigenic culture, clinical diagnosis, and toxin EIA showed the best fitness. The following Ct cut-offs for four commercial test platforms were obtained using this model to delineate three CDI probability zones: [GeneXpert ® : 24.00, 33.61], [Simplexa ® 28.97, 36.85], [Elite MGB ® 30.18, 37.43], and [BD Max ™ 27.60, 34.26]. Conclusion The clinical implication of applying LCA to CDI is to report Ct values assigned to probability zones based on the commercial real-time PCR platform. A broad range of equivocation suggests clinical judgement is essential to the confirmation of CDI.