Derivation and validation of a diagnostic test for irritable bowel syndrome using latent class analysis

The accuracy of symptom-based diagnostic criteria for irritable bowel syndrome (IBS) is modest.

To derive and validate a new test that utilises latent class analysis.

Symptom, colonoscopy, and histology data were collected from 1981 patients and 360 patients in two cohorts referred to secondary care for investigation of their gastrointestinal symptoms in Canada and the UK, respectively. Latent class analysis was used to identify naturally occurring clusters in patient-reported symptoms in the Canadian dataset, and the latent class model derived from this was then applied to the UK dataset in order to validate it. Sensitivity, specificity, and positive and negative likelihood ratios (LRs) were calculated for the latent class models.

In the Canadian cohort, the model had a sensitivity of 44.7% (95% CI 40.0-50.0) and a specificity of 85.3% (95% CI 83.4-87.0). Positive and negative LRs were 3.03 (95% CI 2.57-3.56) and 0.65 (95% CI 0.59-0.71) respectively. A maximum positive LR of 3.93 was achieved following construction of a receiver operating characteristic curve. The performance in the UK cohort was similar, with a sensitivity and specificity of 52.5% (95% CI 42.2-62.7) and 84.3% (95% CI 79.3-88.6), respectively. Positive and negative LRs were 3.35 (95% CI 2.38-4.70) and 0.56 (95% CI 0.45-0.68), respectively, with a maximum positive LR of 4.15.

A diagnostic test for IBS, utilising patient-reported symptoms incorporated into a latent class model, performs as accurately as symptom-based criteria. It has potential for improvement via addition of clinical markers, such as coeliac serology and faecal calprotectin.