Likelihood ratios of clinical, laboratory and image data of pancreatic cancer: Bayesian approach

Abstract

PURPOSE: The diagnosis of pancreatic cancer (PC) is most frequently established in advanced stages. The aim of this study is to estimate the likelihood ratios (LRs) of diagnostic data with regards to PC that could be used to approach an earlier diagnosis. METHODS: A case-control study of 300 patients - 150 histological diagnosed cases of PC and 150 age-matched controls hospitalized for study of jaundice, abdominal pain, weight loss and/or chronic pancreatitis - was conducted. Bayesian probabilities in the form of LRs were estimated for PC predictions. RESULTS: Probability of PC was associated with jaundice [odds ratio (OR) 2.89; confidence interval (CI) 1.71-4.85], glycemic disturbance (OR 5.64; CI 2.36-13.46), tobacco index >20 (OR 2.11; CI 1.08-4.09) and tumour marker CA 19-9 (OR 9.33; CI 1.36-63.95). Computed tomography showed the highest test performance with regards to PC when comparing with other diagnostic tests. LRs for variables relevant to PC were estimated, among the most relevant: jaundice LR + 1.92, CA 19-9 LR + 5.36 and computed tomography LR + 4.15. The prediction model with an endoscopic retrograde cholangiopancreatography at a tertiary referral hospital determined a 67% probability of detecting PC. CONCLUSIONS: Through a Bayesian approach we can combine clinical, laboratory and imaging data to approximate to an earlier diagnosis of PC.