A PPI test does not improve diagnosis of GERD

Clinical Question

Is a response to a proton pump inhibitor a useful diagnostic strategy to determine gastroesophageal reflux?

Bottom Line

Of patients who present with typical reflux symptoms, 70% will actually have gastroesophageal reflux disease (GERD). A 2-week trial to determine the response to a proton pump inhibitor (PPI) will not identify a significant proportion of patients who have true GERD. If a patient has typical reflux symptoms, begin treatment, since an early lack of response does not rule out true GERD. (LOE = 1c)


Aanen MC, Weusten BL, Numans ME, De Wit NJ, Baron A, Smout AJ. Diagnostic value of the proton pump inhibitor test for gastro-oesophageal reflux disease in primary care. Aliment Pharmacol Ther 2006;24:1377-1384.  [PMID:17059519]

Study Design

Diagnostic test evaluation




Outpatient (primary care)


A short course of a PPI is often used to determine whether a patient with typical reflux symptoms truly has GERD. To evaluate the diagnostic accuracy of the PPI test, the researchers enrolled 74 patients (average age = 51 years) with typical reflux symptoms. Approximately 1 in 5 patients smoked and 74% drank alcohol. Following 24-hour esophageal pH monitoring, all 74 patients then took esomeprazole 40 mg (Nexium) for 2 weeks, recording daily whether symptoms were suppressed. A positive response to the PPI test was considered to be a symptom association probability less than 95%, a symptom index of less than 50%, and a symptom severity index of less than 10%. These various outcomes are commonly used to evaluate GERD. The prevalence of GERD was 70% using the symptom association probability score, meaning that 70% of patients with refluxlike symptoms had GERD. The sensitivity of the PPI test was 91% (95% CI, 78%-96%) and the specificity was 26% (10%-49%). In this high prevalence population, the positive predictive value was 75% (62%-85%) and the negative predictive value was 54% (22%-81%). Using these results, the likelihood ratio for the PPI test was 1.2, similar to the likelihood ratio for the presence of typical symptoms.