Introduction
There are no prediction models for a diagnosis of inflammatory bowel disease (IBD) in primary care. Our aim was to develop an IBD risk prediction tool to reduce the length of time patients have undiagnosed IBD symptoms and improve IBD clinical outcomes.
Aims & Methods
We developed and internally validated a risk prediction tool for the diagnosis of IBD, ulcerative colitis (UC) and Crohn’s disease (CD).
A population-based retrospective open cohort study using Clinical Practice Research Datalink (CPRD) database was undertaken between 1st January 2010 and 31st December 2019 of all the patients aged 18 years or older. Patients were followed from first presentation with lower gastrointestinal (GI) symptoms potentially related to IBD to IBD diagnosis. Candidate predictors were chosen based on clinical and substantive knowledge. Cox proportional hazards regression with backward elimination was used for model development of IBD, UC and CD risk prediction models. The predictive performance of the models was assessed using discrimination (C-statistic). Internal validation using 1000 bootstrap sampling repetitions was used to estimate the optimism-corrected measures of predictive performance.
Results
Of 2,056,398 patients with lower GI symptoms: 11,192 patients had an IBD diagnosis (7033 UC (62.8%) and 4138 CD (37.0%)). The final IBD model included demographic factors (age, sex, ethnicity, smoking, index of multiple deprivation, body mass index, Charlson comorbidity index, loperamide use) co-existing conditions (anxiety, depression, irritable bowel syndrome (IBS), haemorrhoids), extraintestinal (EIM) manifestations (mouth ulcers, ophthalmic, primary sclerosing cholangitis, dermatological), laboratory investigations (low haemoglobin (Hb), low Mean Corpuscular volume (MCV), low albumin, low ferritin, raised C-reactive protein (CRP), raised Erythrocyte Sedimentation Rate (ESR), raised calprotectin level). The UC model included all above except mouth ulcers, IBS, ophthalmic EIM. Depression and ferritin were excluded in the CD model though low vitamin B12 level was included. The risk score for each patient was the sum of individual risk factors. In the development dataset, models showed moderate discrimination. For the IBD prediction tool: C-statistic 0.71 (95% CI 0.70-0.71),UC prediction tool: C-statistic 0.70 (95% CI 0.69-0.70) and CD prediction tool: C-statistic 0.75 (95% CI 0.74-0.76). The C-statistics of the models showed that the IBD, UC and CD prediction tools reliably differentiated patients with and without IBD,UC and CD respectively.
Conclusion
A risk score of patient demographics, symptoms and investigations performs well for IBD,UC and CD and may help in prioritising suspected IBD referrals in symptomatic subjects in primary care.
Disclosure
No conflict of interest to declare