Background: The performance of a classification algorithm eventually reaches a point of diminishing returns, where the additional sample added does not improve the results. Thus, there is a need to determine an optimal sample size that maximizes performance while accounting for computational burden or budgetary concerns. Objective: This study aimed to determine optimal sample sizes and the relationships between sample size and dataset-level characteristics over a variety of binary classification algorithms. Methods: A total of 16 large open-source datasets were collected, each containing a binary clinical outcome. Furthermore, 4 machine learning algorithms were assessed: XGBoost (XGB), random forest (RF), logistic regression (LR), and neural networks (NNs). For each dataset, the cross-validated area under the curve (AUC) was calculated at increasing sample sizes, and learning curves were fit. Sample sizes needed to reach the observed full–dataset AUC minus 2 points (0.02) were calculated from the fitted learning curves and compared across the datasets and algorithms. Dataset–level characteristics, minority class proportion, full–dataset AUC, number of features, type of features, and degree of nonlinearity were examined. Negative binomial regression models were used to quantify relationships between these characteristics and expected sample sizes within each algorithm. A total of 4 multivariable models were constructed, which selected the best-fitting combination of dataset–level characteristics. Results: Among the 16 datasets (full-dataset sample sizes ranging from 70,000-1,000,000), median sample sizes were 9960 (XGB), 3404 (RF), 696 (LR), and 12,298 (NN) to reach AUC stability. For all 4 algorithms, more balanced classes (multiplier: 0.93-0.96 for a 1% increase in minority class proportion) were associated with decreased sample size. Other characteristics varied in importance across algorithms—in general, more features, weaker features, and more complex relationships between the predictors and the response increased expected sample sizes. In multivariable analysis, the top selected predictors were minority class proportion among all 4 algorithms assessed, full–dataset AUC (XGB, RF, and NN), and dataset nonlinearity (XGB, RF, and NN). For LR, the top predictors were minority class proportion, percentage of strong linear features, and number of features. Final multivariable sample size models had high goodness-of-fit, with dataset–level predictors explaining a majority (66.5%-84.5%) of the total deviance in the data among all 4 models. Conclusions: The sample sizes needed to reach AUC stability among 4 popular classification algorithms vary by dataset and method and are associated with dataset–level characteristics that can be influenced or estimated before the start of a research study.

INTRODUCTION: Spontaneous bacterial peritonitis (SBP) bacteriology has changed over time. Reappraisal of primary SBP prophylaxis showed an increased rate of resistance in patients on primary prophylaxis with resultant discontinuation of this prophylaxis throughout the Veterans Affairs (VA). We aimed to re-evaluate the risk-benefit ratio of secondary SBP prophylaxis (SecSBPPr). METHODS: Using validated International Classification of Diseases -9/10 codes, we used the VA Corporate Data Warehouse and the Non-VA National TriNetX database to identify patients in 2 different large US systems who survived their first SBP diagnosis (with chart review from 2 VA centers) between 2009 and 2019. We evaluated the prevalence of SecSBPPr and compared outcomes between those who started on SecSBPPr vs not. RESULTS: We identified 4,673 veterans who survived their index SBP episode; 54.3% of whom were prescribed SecSBPPr. Multivariable analysis showed higher SBP recurrence risk in those on vs off SecSBPPr (hazards ratio 1.63 [1.40–1.91], P < 0.001). This was accompanied by higher fluoroquinolone resistance odds in SecSBPPr patients (odds ratio = 4.32 [1.36–15.83], P = 0.03). In TriNetX, we identified 6,708 patients who survived their index SBP episode; 48.6% were on SecSBPPr. Multivariable analysis similarly showed SecSBPPr increased SBP recurrence risk (hazards ratio 1.68 [1.33–1.80], P < 0.001). Both data sets showed higher SBP recurrence trends over time in SecSBPPr patients. Results remained consistent at 6-month and 2-year timepoints. DISCUSSION: In 2 national data sets of >11,000 patients with SBP, we found that SecSBPPr was prescribed in roughly half of patients. When initiated, SecSBPPr, compared with no prophylaxis after SBP, increased the risk of SBP recurrence in multivariable analysis by 63%–68%, and this trend worsened over time. SecSBPPr should be reconsidered in cirrhosis.

We report that exhaled breath condensate (EBC) identifies cellular metabolic dysregulation in patients with radiation-induced lung injury. In this pilot study, untargeted metabolomics revealed a striking metabolic signature in EBC from patients with radiation-induced lung fibrosis compared to patients with lung cancer, at-risk smokers, and healthy volunteers. Patients with radiation-induced fibrosis exhibit specific changes in tricarboxylic acid (TCA) cycle energy metabolism that may be required to support the increased energy demands of fibroproliferation.

Despite negative outcomes, hepatic encephalopathy (HE) is not included in the Model for End-Stage Liver Disease (MELD) scores, including MELD 3.0. In a National Veterans Affairs database, we studied the additive mortality predictive impact of a documented inpatient overt HE diagnosis on MELD 3.0 and MELD-sodium (MELD-Na) scores in 75,327 veterans (26.8% with HE, 22% 90-day and 37% 1-year mortality). The addition of HE to MELD 3.0 and MELD-Na significantly improved their mortality prediction with interactions at 90 days and 1 year on multivariable regression. A documented inpatient HE episode is additive to the mortality prediction of MELD 3.0 by 4–5 extra points and could enhance future MELD iterations.

INTRODUCTION: Antibiotic overuse and subsequent antibiotic resistance lead to worse infection outcomes in cirrhosis. Secondary spontaneous bacterial peritonitis prophylaxis (SecSBBPr) is associated with higher SBP recurrence, but impact on non-SBP infections is unclear. METHODS: We studied patients with cirrhosis and SBP who were given SecSBPPr or not between 2009 and 2019 in 2 complementary national cohorts (Veterans Affairs Corporate Data Warehouse [VA-CDW] and non-VA TriNetX). Development of total non-SBP infections and specifically urinary tract infections (UTIs), bacteremia, pneumonia, and C. difficile using validated codes over 2 years was compared between those on SecSBPPr vs not. Multivariable regression for non-SBP infections was performed. RESULTS: VA-CDW: Of 4,673 veterans with index SBP, 2,539 (54.3%) were started on SecSBPPr. In total, 1,406 (30.1%) developed non-SBP infections (13.5% UTI, 12.4% pneumonia, 8.5% bacteremia, and 6.8% C. difficile). On multivariable regression, SecSBPPr was significantly associated with any non-SBP infection (odds ratio [OR] 1.26, 95% confidence interval [CI] 1.10–1.44, P < 0.0001) and UTI (OR 1.21, 95% CI 1.01–1.45, P = 0.036). TriNetX: Of 6,708 patients with index SBP, 3,261 (48.6%) were started on SecSBPPr. In total, 1,932 (28.8%) patients developed non-SBP infections (13.4% UTI, 12.9% pneumonia, 8.6% bacteremia, and 5.9% C. difficile). On multivariable regression, SecSBPPr was significantly associated with any non-SBP infection (OR 1.33, 95% CI 1.12–1.59, P < 0.0001), UTI (OR 1.35, 95% CI 1.07–1.71, P = 0.010), pneumonia (OR 1.35, 95% CI 1.06–1.72, P = 0.017), and bacteremia (OR 1.47, 95% CI 1.10–1.97, P = 0.009). DISCUSSION: In 2 diverse US-based national cohorts of patients with cirrhosis and SBP, use of SecSBPPr was associated with a higher risk of non-SBP infections, especially urinary tract infections.

Background: Terlipressin is the only FDA-approved vasoconstrictor for hepatorenal syndrome (HRS). The CONFIRM study is the largest trial of terlipressin versus placebo. Novel predictors of HRS response are required to enrich patient selection and optimize outcomes. Methods: Samples at treatment initiation were tested using (a) liquid chromatography–mass spectrometry of 1594 plasma/1420 urine metabolites (Metabolon Inc.), (b) aptamer-based array of 7289 plasma proteins (SomaScan), and (c) 14 plasma/urine pre-specified assays. The CONFIRM trial’s original definition of HRS response [2 serum creatinine (SCr) 2 h] was used as the primary outcome. Results: In all, 115 patients [79 terlipressin-treated (TT) and 36 placebo-treated (PT)] provided samples. Baseline characteristics, outcomes, and 2:1 TT:PT allocation were preserved from the original 300-patient trial. A total of 36 out of 116 (31.0%) patients achieved HRS reversal. HRS reversal was associated with lower SCr (p=0.001), cystatin C (p=0.005), angiopoietin-2 (p=0.04), and beta-2 microglobulin (p=0.006). In metabolite analysis, PT had the most significant differences in HRS reversal [n=26 plasma, n=50 urine, including lower urine levels of those centered on sulfated secondary bile acids (microbiome-derived), N-acetylated amino acids, catechols (both uremic toxins), and phosphocholines (cell membrane integrity)], with fewer in TT (n=1 plasma, n=2 urine), and in all patients (n=3 plasma, n=7 urine). There were no significant aptamers associated with HRS reversal after false-discovery correction. Conclusions: SCr, cystatin C, angiopoietin-2, and beta-2 microglobulin were associated with HRS reversal. Protein and metabolite signals centered on microbiome function and uremic toxins appeared more robust in PT patients, likely selecting a subgroup that may recover without terlipressin. Use of novel biomarkers may enrich for terlipressin response.