Statistical analysis

The primary analysis was to identify the optimal threshold for pepsin that achieves desired sensitivity and specificity for classifying GERD versus no GERD as evaluated by receiver operating characteristic curves. Adjusted analyses considered age, body mass index, and sex in addition to pepsin in a logistic regression model to evaluate potential improvements in performance when using the predicted probability of being a case from the model. Models were fit comparing all GERD to no GERD, as well as a separate sensitivity model for comparing no GERD to NERD alone. The area under the curve (AUC) and 95% confidence interval was calculated for each model. The optimal threshold identified for pepsin (unadjusted) and the predicted probability (adjusted) were identified from the Phase 1 data, and then applied to the external Phase 2 validation data and summarized using sensitivity and specificity. Pepsin values measured as 16 to 24.9 ng/mL were imputed as 20.5 ng/mL, with sensitivity analyses to evaluate model performance if values were imputed as 16 or 24.9 ng/mL. The Phase 2 study also included a secondary analysis to assess quality control of device. This secondary analysis calculated the intraclass correlation coefficient assuming a two-way mixed-effects model for single measure absolute agreement since three pepsin values were calculated for all Phase 2 subjects. Comparisons of continuous measures between different groupings used a two-sample t-test. All figures created and analyses were conducted using R v3.6.3.