A machine learning framework for optimizing obesity care by simulating clinical trajectories and targeted interventions.

OBJECTIVE: This study aimed to determine the important clinical management bottlenecks that contribute to underuse of weight loss surgery (WLS) and assess risk factors for attrition at each of them. METHODS: A multistate conceptual model of progression from primary care to WLS was developed and used to study all adults who were seen by a primary care provider (PCP) and eligible for WLS from 2016 to 2017 at a large institution. Outcomes were progression from each state to each subsequent state in the model: PCP visit, endocrine weight management referral, endocrine weight management visit, WLS referral, WLS visit, and WLS. RESULTS: Beginning with an initial PCP visit, the respective 2-year Kaplan-Meier estimate for each outcome was 35% (n = 2063), 15.6% (n = 930), 6.3% (n = 400), 4.7% (n = 298), and 1.0% (n = 69) among 5876 eligible patients. Individual providers and clinics differed significantly in their referral practices. Female patients, younger patients, those with higher BMI, and those seen by trainees were more likely to progress. A simulated intervention to increase referrals among PCPs would generate about 49 additional WLS procedures over 3 years. CONCLUSIONS: This study discovered novel insights into the specific dynamics underlying low WLS use rates. This methodology permits in silico testing of interventions designed to optimize obesity care prior to implementation.

Intermittent claudication treatment patterns in the commercially insured non-Medicare population.

OBJECTIVE: Despite published guidelines and data for Medicare patients, it is uncertain how younger patients with intermittent claudication (IC) are treated. Additionally, the degree to which treatment patterns have changed over time with the expansion of endovascular interventions and outpatient centers is unclear. Our goal was to characterize IC treatment patterns in the commercially insured non-Medicare population. METHODS: The IBM MarketScan Commercial Database, which includes more than 8 billion US commercial insurance claims, was queried for patients newly diagnosed with IC from 2007 to 2016. Patient demographics, medication profiles, and open/endovascular interventions were evaluated. Time trends were modeled using simple linear regression and goodness-of-fit was assessed with coefficients of determination (R2). A patient-centered cohort sample and a procedure-focused dataset were analyzed. RESULTS: Among 152,935,013 unique patients in the database, there were 300,590 patients newly diagnosed with IC. The mean insurance coverage was 4.4 years. The median patients age was 58 years and 56% of patients were male. The prevalence of statin use was 48% among patients at the time of IC diagnosis and increased to 52% among patients after one year from diagnosis. Interventions were performed in 14.3%, of whom 20% and 6% underwent two or more and three or more interventions, respectively. The median time from diagnosis to intervention decreased from 230 days in 2008 days to 49 days in 2016 (R2 = 0.98). There were 16,406 inpatient and 102,925 ambulatory interventions for IC over the study period. Among ambulatory interventions, 7.9% were performed in office-based/surgical centers. The proportion of atherectomies performed in the ambulatory setting increased from 9.7% in 2007 to 29% in 2016 (R2 = 0.94). In office-based/surgical centers, 57.6% of interventions for IC used atherectomy in 2016. Atherectomy was used in ambulatory interventions by cardiologists in 22.6%, surgeons in 15.2%, and radiologists in 13.6% of interventions. Inpatient atherectomy rates remained stable over the study period. Open and endovascular tibial interventions were performed in 7.9% and 7.8% of ambulatory and inpatient IC interventions, respectively. Tibial bypasses were performed in 8.2% of all open IC interventions. CONCLUSIONS: There has been shorter time to intervention in the treatment of younger, commercially insured patients with IC, with many receiving multiple interventions. Statin use was low. Ambulatory procedures, especially in office-based/surgical centers, increasingly used atherectomy, which was not observed in inpatient settings.

Development and validation of machine learning models to predict gastrointestinal leak and venous thromboembolism after weight loss surgery: an analysis of the MBSAQIP database.

BACKGROUND: Postoperative gastrointestinal leak and venous thromboembolism (VTE) are devastating complications of bariatric surgery. The performance of currently available predictive models for these complications remains wanting, while machine learning has shown promise to improve on traditional modeling approaches. The purpose of this study was to compare the ability of two machine learning strategies, artificial neural networks (ANNs), and gradient boosting machines (XGBs) to conventional models using logistic regression (LR) in predicting leak and VTE after bariatric surgery. METHODS: ANN, XGB, and LR prediction models for leak and VTE among adults undergoing initial elective weight loss surgery were trained and validated using preoperative data from 2015 to 2017 from Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program database. Data were randomly split into training, validation, and testing populations. Model performance was measured by the area under the receiver operating characteristic curve (AUC) on the testing data for each model. RESULTS: The study cohort contained 436,807 patients. The incidences of leak and VTE were 0.70% and 0.46%. ANN (AUC 0.75, 95% CI 0.73-0.78) was the best-performing model for predicting leak, followed by XGB (AUC 0.70, 95% CI 0.68-0.72) and then LR (AUC 0.63, 95% CI 0.61-0.65, p < 0.001 for all comparisons). In detecting VTE, ANN, and XGB, LR achieved similar AUCs of 0.65 (95% CI 0.63-0.68), 0.67 (95% CI 0.64-0.70), and 0.64 (95% CI 0.61-0.66), respectively; the performance difference between XGB and LR was statistically significant (p = 0.001). CONCLUSIONS: ANN and XGB outperformed traditional LR in predicting leak. These results suggest that ML has the potential to improve risk stratification for bariatric surgery, especially as techniques to extract more granular data from medical records improve. Further studies investigating the merits of machine learning to improve patient selection and risk management in bariatric surgery are warranted.