OC-098 How commonly is gastric cancer missed at endoscopy: a UK primary care based study

Abstract

Background: Meta-analysis of published single hospital series including 1977 subjects suggests that 14% of gastric cancer (GC) subjects have had an upper gastrointestinal endoscopy (EGD) up to 3 years previously that failed to diagnose their GC (50% in the 12 months before diagnosis and 50% 1 to 3 years before diagnosis). Methods: All patients with GC in the THIN general practice database covering 5 million UK subjects were examined. A nested case-control study was performed with cases subjects who underwent EGD 1-5 years prior to their EGD that diagnosed GC and controls subjects who did not undergo EGD 1-5 years prior to their diagnostic EGD. Results: 5473 GC were available for analysis (3402 males (62%), mean age 71 years), with follow up of 46779 subject years. 169 (3.1%, 98 males (58%), mean age 71 years) had an EGD which failed to diagnose GC between 1 and 5 years prior to diagnosis of GC, out of whom 128 (2.3%) had EGD between 1-3 years before and 41 (0.7%) had an EGD between 3-5 years before diagnosis. There were 56 primary care consultations with symptoms pertaining to oesophago-gastric cancer between 1-5 years prior to diagnosis of GC (dyspepsia (n) 51, anaemia 9, weight loss 12, dysphagia 3), of which all underwent EGD. No subject who had an EGD that did not diagnose cancer was on proton pump inhibitor (PPI) therapy in the year prior to EGD and 49 (0.9%) of subjects were on PPI therapy in the year prior to being diagnosed with GC. Logistic regression analysis of subjects who had an EGD that failed to diagnose cancer and those that did not, failed to identify any specific predictive factors (age 1.0 (0.99-1.01), p=0.99), sex 0.84 (0.61-1.14), p=0.26), related to an EGD that failed to diagnose GC. Conclusion: Missing GC at EGD is, reassuringly, less than half as common as previous much smaller studies in secondary care have suggested. PPI therapy does not contribute to missing GC at EGD. Advances in endoscopy and selection bias in previous studies may account for these differences.