Simplifying Diagnosis of Bile Acid Diarrhea With Clinical and Biochemical Measurements on Blood and Single Stool Sample

Abstract

Background & aims: Diagnosis of bile acid diarrhea (BAD) has been based on 48-hour fecal BA excretion; serum 7αC4 (C4) has been used to screen for BAD. Optimal diagnostic cutoffs for C4 and biochemical measurements in a single stool sample are unknown. We sought to examine the relationship between total BA concentration (TBAc) and percent primary BA (%PBA) in a single stool sample and serum C4 in patients with and without BAD and explore performance characteristics of stool consistency and biochemical (serum C4 and single-stool BA) parameters for diagnosis of BAD compared with gold standard 48-hour fecal BA.

Methods: Based on data from patients with BAD, irritable bowel syndrome with diarrhea (IBS-D), and healthy control subjects, we assessed correlations among stool and serum measurements. Machine learning models (based on data from 30 patients with BAD, 8 patients with IBS-D, and 26 healthy control subjects) were trained on 25 bootstrapped random samples, the superior model was identified, and optimal cutoffs of biological measurements to diagnose BAD were summarized.

Results: There were correlations between serum C4 and %PBA (R = 0.284, P < .001), and between %PBA and TBAc (R = 0.49, P < .001). Using a %PBA of 1.05% (25th percentile in BAD), the %PBA distinguished BAD from IBS-D (odds ratio, 3.06; 95% confidence interval, 1.35-7.46; P = .01). The multivariate logistic regression model had superior balance of variance and bias. Optimal cutoffs for predicting BAD using logistic regression were 4.5% PBA (P = .023) and 1.88 μmol/g TBAc (P = .016). Serum C4 >24 ng/mL and PBA >4.6% individually had 57% and 75.8% positive predictive value, respectively, but together had a 90.1% positive predictive value. Stool consistency was less informative.

Conclusions: New diagnostic cutoffs based on serum C4 and single-stool TBAc and % PBA provide potential alternatives for diagnosing BAD. Further validation is warranted.

Keywords: Logistic Regression; Machine Learning; Primary; Stool Consistency; Total Concentration.