Editorial: latent class analysis to improve confidence in the diagnosis of IBS

Irritable bowel syndrome (IBS) is a chronic, multifactorial gastrointestinal disorder with no reliable biomarker. Rome is currently considered the gold standard of symptom-based criteria for IBS.¹ However, these criteria have only modest diagnostic sensitivity and specificity when administered alone. Therefore, in clinical practice it is recommended that Rome criteria be combined with a targeted physical, screening for alarm features and limited testing.^1,2 Indeed, the addition of screening for alarm features produced a positive predictive value of 100%, with sensitivity of 65% and specificity of 100% in an earlier version of the Rome criteria.³ Despite this, the diagnosis of IBS in clinical practice is commonly considered one of exclusion, resulting in a large number of unnecessary, invasive and costly tests and procedures (e.g. cholecystectomy or appendectomy).⁴ This tendency reflects clinicians’ and patients’ concerns about overlooking potentially serious diagnoses and appears to vary based on a clinician’s level of expertise in IBS.⁵

In their study, Sood et al. address the importance of establishing a noninvasive, accurate, patient-reported diagnostic tool for IBS.⁶ The authors collected data on patient-reported gastrointestinal symptoms associated with IBS as well as nine extra-intestinal items from the Patient Health Questionnaire 15, which are common in GI patients meeting criteria for IBS.⁷ The researchers then applied latent class analysis to identify IBS in two separate secondary care cohorts in Canada (n = 1981) and the UK (n = 360), using symptom-based criteria and absence of organic lower GI disease after colonoscopy as the reference standard. In the Canadian cohort, this model revealed sensitivity of 44.7% and specificity of 85.3% with a maximum positive likelihood ratio (LR) of 3.93 after construction of a receiver operating characteristic (ROC) curve. Results in the UK cohort were similar with sensitivity of 52.5%, specificity of 84.3%, and maximum LR of 4.15 after construction of ROC curve.

The model is an important step towards a more definitive symptom-based diagnostic criterion for IBS. The idea to use statistical modelling to diagnose IBS was first described by Kruis et al. in 1984,⁸ whose scoring system used logistic regression combined with laboratory testing to produce impressive diagnostic accuracy for IBS.⁹ At the time, however, Kruis’s model was not easily applied due to technological limitations and thus was not used widely. Current smartphone and Internet computing capabilities allow for statistical modelling, such as the one proposed by Sood et al., to be readily and easily applied.

The model presented by Sood et al. could be improved by the addition of laboratory tests. In its current form, the sensitivity and specificity are not optimal, and may be below what many might consider ideal for clinical practice. The authors acknowledge that the accuracy of their tool may be improved by including basic laboratory tests such as C-reactive protein and faecal calprotectin¹⁰ as well as emerging IBS biomarkers.¹¹ Although adding these and other tests would make the model more complex, it would improve the overall performance of this promising diagnostic tool and may also appeal more widely to patients and clinicians who worry about the chances, however unlikely, of missing an organic disease.