Publications

AIMS: To develop machine-learning algorithms for predicting the risk of a hospitalization or emergency department (ED) visit for opioid use disorder (OUD) (i.e. OUD acute events) in Pennsylvania Medicaid enrollees in the Opioid Use Disorder Centers of Excellence (COE) program and to evaluate the fairness of model performance across racial groups.

METHODS: We studied 20 983 United States Medicaid enrollees aged 18 years or older who had COE visits between April 2019 and March 2021. We applied multivariate logistic regression, least absolute shrinkage and selection operator models, random forests, and eXtreme Gradient Boosting (XGB), to predict OUD acute events following the initial COE visit. Our models included predictors at the system, patient, and regional levels. We assessed model performance using multiple metrics by racial groups. Individuals were divided into a low, medium and high-risk group based on predicted risk scores.

RESULTS: The training (n = 13 990) and testing (n = 6993) samples displayed similar characteristics (mean age 38.1 ± 9.3 years, 58% male, 80% White enrollees) with 4% experiencing OUD acute events at baseline. XGB demonstrated the best prediction performance (C-statistic = 76.6% [95% confidence interval = 75.6%-77.7%] vs. 72.8%-74.7% for other methods). At the balanced cutoff, XGB achieved a sensitivity of 68.2%, specificity of 70.0%, and positive predictive value of 8.3%. The XGB model classified the testing sample into high-risk (6%), medium-risk (30%), and low-risk (63%) groups. In the high-risk group, 40.7% had OUD acute events vs. 16.5% and 5.0% in the medium- and low-risk groups. The high- and medium-risk groups captured 44% and 26% of individuals with OUD events. The XGB model exhibited lower false negative rates and higher false positive rates in racial/ethnic minority groups than White enrollees.

CONCLUSIONS: New machine-learning algorithms perform well to predict risks of opioid use disorder (OUD) acute care use among United States Medicaid enrollees and improve fairness of prediction across racial and ethnic groups compared with previous OUD-related models.

BACKGROUND: Narcolepsy is linked to adverse cardiovascular disease (CVD) outcomes, but few studies have examined its associations with subclinical CVD, including in children. We assessed the relationship between narcolepsy and subclinical CVD outcomes, including hypertension, hyperlipidemia, diabetes, and nonalcoholic fatty liver disease/nonalcoholic steatohepatitis.

METHODS AND RESULTS: We conducted a retrospective cohort study using MarketScan Commercial and Medicare Supplemental databases from January 1, 2005 to December 31, 2021. Patients included N=22 293 diagnosed with narcolepsy (NT1 and NT2) and N=63 709 propensity-score-matched without. Patients with narcolepsy were identified as those with ≥2 outpatient insurance claims for narcolepsy (type 1 or type 2) within a 1-year interval with 1 claim being nondiagnostic. Main outcomes were diagnosis of hypertension, hyperlipidemia, diabetes, and nonalcoholic fatty liver disease/nonalcoholic steatohepatitis following index date, as well as a composite measure for CVD and major adverse cardiovascular events. Compared with propensity-score-matched patients without narcolepsy, patients with narcolepsy had an increased risk for hypertension (hazard ratio [HR], 1.40 [95% CI, 1.34-1.47]), hyperlipidemia (HR, 1.41 [95% CI, 1.35-1.47]), diabetes (HR, 1.50 [95% CI, 1.38-1.64), nonalcoholic fatty liver disease/nonalcoholic steatohepatitis (HR, 1.48 [95% CI, 1.28-1.73]), CVD composite (HR,1.61 [95% CI, 1.35-1.47]), and major adverse cardiovascular events (HR,1.69 [95% CI, 1.43-2.00]). Results remained significant following adjustment for narcolepsy medications including stimulants, wake-promoting agents, and oxybates. Results stratified by age groups showed similar findings, including heightened risk for those <25 years old.

CONCLUSIONS: Narcolepsy is associated with greater risk of subclinical CVD even in patients as early as childhood. Detection of these outcomes early in the course of narcolepsy could help reduce the burden of adverse cardiovascular events later in life.

OBJECTIVE: To compare the effectiveness of a modified surface gelatin sponge to a plain collagen sponge for hemostasis of parenchymal hepatic bleeding.

STUDY DESIGN: Prospective, randomized trial of two hemostatic agents.

ANIMALS: A total of 45 dogs undergoing elective liver surgery were randomly allocated into two groups: 22 in the adhesive gelatin (AG) group and 23 in the plain collagen (PC) group. A total of 20 patients per group underwent liver biopsy to create a uniformly sized bleeding surface, with the remaining patients (AG = 2, PC = 3) undergoing liver lobectomy.

METHODS: Evaluation of hemostatic effectiveness and tissue adhesion of each sponge type was performed by the operating surgeon using structured scoring systems. Hemostatic parameters were primarily evaluated at the liver biopsy site to maintain homogeneity of bleeding surface size.

RESULTS: For the liver biopsy group (n = 40), 5 min after hemostatic sponge application, 10/20 dogs were bleeding in the PC group, compared to 2/20 in AG group (p = .0138). The PC bleeding was significantly higher than AG across the 3 to 6 min evaluation period (p < .001). When surgeons tested the adhesion of the sponge across the whole cohort (n = 45), AG scored 2 (of 3) against 1 for PC (p < .001). In group PC, 5/23 sponges dislodged during abdominal lavage and preparations for closure and had to be replaced due to recurrence of bleeding, compared with no AG sponges dislodging (p = .042). There were no further complications related to the use of either sponge.

CONCLUSION: In the dogs with hepatic parenchymal incision, use of an adhesive gelatin sponge improved intraoperative attachment and haemostatic effectiveness, compared to a collagen sponge.

CLINICAL SIGNIFICANCE: Based on our clinical experience in these cases, adhesive gelatin sponges could be considered an effective option when selecting a hemostatic agent for liver surgery in dogs.

Enrollment in Medicare Advantage (MA) plans rose to over 50% of eligible Medicare patients in 2023. Payments to MA plans incorporate risk scores that are largely based on patient diagnoses from the prior year, which incentivizes MA plans to code diagnoses more intensively. We estimated coding inflation rates for individual MA contracts using a method that allows for differential selection into contracts based on patient health. We illustrate the method using data on MA risk scores and health conditions from the most recent year available, 2014. This approach could also be used beginning in 2022, when Medicare transitioned to MA risk scores based on MA Encounter records. Several existing methods assess coding intensity, but this study's approach is novel in its use of plan-level mortality rates to infer plan-level coding intensity. We found an enrollment-weighted mean coding inflation rate of 8.4%, with rates ranging from 3.4% to 12.7% for the largest 8 MA insurers and from 1.1% to 22.2% for the largest 20 MA contracts in 2014. We found higher coding intensity for health plans that were HMOs, provider-owned, large, older, or had high star ratings. Approximately 68.1% of MA enrollees were in contracts with coding inflation rates larger than Medicare's coding intensity adjustment.