Dataset Information

Using machine learning to predict risk of incident opioid use disorder among fee-for-service Medicare beneficiaries: A prognostic study.

ABSTRACT: Objective: To develop and validate a machine-learning algorithm to improve prediction of incident OUD diagnosis among Medicare beneficiaries with ?1 opioid prescriptions.

Methods: This prognostic study included 361,527 fee-for-service Medicare beneficiaries, without cancer, filling ?1 opioid prescriptions from 2011-2016. We randomly divided beneficiaries into training, testing, and validation samples. We measured 269 potential predictors including socio-demographics, health status, patterns of opioid use, and provider-level and regional-level factors in 3-month periods, starting from three months before initiating opioids until development of OUD, loss of follow-up or end of 2016. The primary outcome was a recorded OUD diagnosis or initiating methadone or buprenorphine for OUD as proxy of incident OUD. We applied elastic net, random forests, gradient boosting machine, and deep neural network to predict OUD in the subsequent three months. We assessed prediction performance using C-statistics and other metrics (e.g., number needed to evaluate to identify an individual with OUD [NNE]). Beneficiaries were stratified into subgroups by risk-score decile.

Results: The training (n = 120,474), testing (n = 120,556), and validation (n = 120,497) samples had similar characteristics (age ?65 years = 81.1%; female = 61.3%; white = 83.5%; with disability eligibility = 25.5%; 1.5% had incident OUD). In the validation sample, the four approaches had similar prediction performances (C-statistic ranged from 0.874 to 0.882); elastic net required the fewest predictors (n = 48). Using the elastic net algorithm, individuals in the top decile of risk (15.8% [n = 19,047] of validation cohort) had a positive predictive value of 0.96%, negative predictive value of 99.7%, and NNE of 104. Nearly 70% of individuals with incident OUD were in the top two deciles (n = 37,078), having highest incident OUD (36 to 301 per 10,000 beneficiaries). Individuals in the bottom eight deciles (n = 83,419) had minimal incident OUD (3 to 28 per 10,000).

Conclusions: Machine-learning algorithms improve risk prediction and risk stratification of incident OUD in Medicare beneficiaries.

SUBMITTER: Lo-Ciganic WH

PROVIDER: S-EPMC7367453 | biostudies-literature | 2020

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Using machine learning to predict risk of incident opioid use disorder among fee-for-service Medicare beneficiaries: A prognostic study.

Lo-Ciganic Wei-Hsuan WH Huang James L JL Zhang Hao H HH Weiss Jeremy C JC Kwoh C Kent CK Donohue Julie M JM Gordon Adam J AJ Cochran Gerald G Malone Daniel C DC Kuza Courtney C CC Gellad Walid F WF

PloS one 20200717 7

<h4>Objective</h4>To develop and validate a machine-learning algorithm to improve prediction of incident OUD diagnosis among Medicare beneficiaries with ≥1 opioid prescriptions.<h4>Methods</h4>This prognostic study included 361,527 fee-for-service Medicare beneficiaries, without cancer, filling ≥1 opioid prescriptions from 2011-2016. We randomly divided beneficiaries into training, testing, and validation samples. We measured 269 potential predictors including socio-demographics, health status, ...[more]

PMID: 32678860

Similar Datasets

Project description:BackgroundPrevious studies have identified disparities in readmissions among Medicare beneficiaries hospitalized for the Hospital Readmissions Reduction Program's (HRRP's) priority conditions. Evidence suggests timely follow-up is associated with reduced risk of readmission, but it is unknown whether timely follow-up reduces disparities in readmission.ObjectiveTo assess whether follow-up within 7 days after discharge from a hospitalization reduces risk of readmission and mitigates identified readmission disparities.DesignA retrospective cohort study using Cox proportional hazards models to estimate the associations between sociodemographic characteristics (race and ethnicity, dual-eligibility status, rurality, and area social deprivation), follow-up, and readmission. Mediation analysis was used to examine if disparities in readmission were mitigated by follow-up.ParticipantsWe analyzed data from 749,402 Medicare fee-for-service beneficiaries hospitalized for acute myocardial infarction, chronic obstructive pulmonary disease, heart failure, or pneumonia, and discharged home between January 1 and December 1, 2018.Main measureAll-cause unplanned readmission within 30 days after discharge.Key resultsPost-discharge follow-up within 7 days of discharge was associated with a substantially lower risk of readmission (HR: 0.52, 95% CI: 0.52-0.53). Across all four HRRP conditions, beneficiaries with dual eligibility and beneficiaries living in areas with high social deprivation had a higher risk of readmission. Non-Hispanic Black beneficiaries had higher risk of readmission after hospitalization for pneumonia relative to non-Hispanic Whites. Mediation analysis suggested that 7-day follow-up mediated 21.2% of the disparity in the risk of readmission between dually and non-dually eligible beneficiaries and 50.7% of the disparity in the risk of readmission between beneficiaries living in areas with the highest and lowest social deprivation. Analysis suggested that after hospitalization for pneumonia, 7-day follow-up mediated nearly all (97.5%) of the increased risk of readmission between non-Hispanic Black and non-Hispanic White beneficiaries.ConclusionsImproving rates of follow-up could be a strategy to reduce readmissions for all beneficiaries and reduce disparities in readmission based on sociodemographic characteristics.

Project description:ObjectivesLoss of appetite in older adults can lead to malnutrition, weight loss, frailty, and death, but little is known about its epidemiology in the United States (US). The objective of this study was to estimate the annual prevalence and incidence of anorexia in older adults with Medicare fee-for-service (FFS) health insurance.DesignRetrospective and observational analysis of administrative health insurance claims data.SettingThis study included Medicare FFS claims from all settings (eg, hospital inpatient/outpatient, office, assisted living facility, skilled nursing facility, hospice, rehabilitation facility, home).ParticipantsThis study included all individuals aged 65 to 115 years old with continuous Medicare FFS medical coverage (Parts A and/or B) for at least one 12-month period from October 1, 2015, to September 30, 2021 (ie, approximately 30 million individuals each year).InterventionNot applicable.MeasurementsAnorexia was identified using medical claims with the ICD-10 diagnosis code "R63.0: Anorexia". This study compared individuals with anorexia to a control group without anorexia with respect to demographics, comorbidities using the Charlson Comorbidity Index (CCI), Claims-based Frailty Index (CFI), and annual mortality. The annual prevalence and incidence of anorexia were estimated for each 12-month period from October 1, 2015, to September 30, 2021.ResultsThe number of individuals with anorexia ranged from 317,964 to 328,977 per year, a mean annual prevalence rate of 1.1%. The number of individuals newly diagnosed with anorexia ranged from 243,391 to 281,071 per year, a mean annual incidence rate of 0.9%. Individuals with anorexia had a mean (±standard deviation) age of 80.5±8.7 years (vs 74.9±7.5 years without anorexia; p<.001), 64.4% were female (vs 53.8%; p<.001), and 78.4% were White (vs 83.2%; p<.001). The most common CCI comorbidities for those with anorexia were chronic pulmonary disease (39.4%), dementia (38.3%), and peripheral vascular disease (38.0%). Median (interquartile range [IQR]) CCI with anorexia was 4 [5] (vs 1 [3] without anorexia; p<.001). The annual mortality rate among those with anorexia was 22.3% (vs 4.1% without anorexia; relative risk 5.49 [95% confidence interval, 5.45-5.53]).ConclusionApproximately 1% of all adults aged 65-115 years old with Medicare FFS insurance are diagnosed with anorexia each year based on ICD-10 codes reported in claims. These individuals have a higher comorbidity burden and an increased risk of annual mortality compared to those without a diagnosis of anorexia. Further analyses are needed to better understand the relationship between anorexia, comorbidities, frailty, mortality, and other health outcomes.

Project description:BackgroundThe amount of total knee arthroplasty (TKA) procedures performed in the United States has been increasing steadily and is projected to reach 3 million procedures annually by 2030 in patients aged ≥65 years. A rise in TKA procedures will increase spending on osteoarthritis treatments, which is currently the second highest category of spending for Medicare patients. Because TKA procedures account for a substantial amount of total osteoarthritis spending, payers and providers are examining methods to reduce spending on the procedure while improving clinical outcomes. Customized individually made implants have been shown to improve clinical outcomes, such as physical function and limb alignment, compared with off-the-shelf implants; however, the economic impact of customized implants has yet to be established.ObjectiveTo analyze TKA episode expenditures among Medicare fee-for-service (FFS) members who received a customized or an off-the-shelf implant.MethodsMembers undergoing a TKA procedure using the customized implant technology were identified in the Medicare FFS database and were propensity matched (1:5) to a cohort of members who received off-the-shelf implants. Reimbursement for the initial procedure (ie, customized and off-the-shelf procedure), a preoperative computed tomography scan, and 12-month postoperative healthcare utilization were analyzed. The overall episode expenditures were used to construct a budget impact model to calculate the per-member per-month (PMPM) spending for Medicare FFS beneficiaries.ResultsThe average total episode spending was significantly lower among the customized implant cohort ($18,585) compared with the off-the-shelf implant cohort ($20,280; a $1695 difference; P <.0001). This savings resulted in $0.08 PMPM savings for the Medicare FFS program when a portion (10%) of eligible members received the customized implant technology. A sensitivity analysis, which varied with the customized implant market penetration and the percent of customized implant-eligible procedures, indicated that the savings could be as great as $0.28 PMPM.ConclusionThe results of this analysis demonstrate that the use of customized implants in TKA procedures can achieve substantial savings versus off-the-shelf procedures for the Medicare FFS program, and this savings is primarily driven by a lower average initial procedure cost and lower postoperative spendings for inpatient services and skilled-nursing facility costs.

Project description:BackgroundHospitalizations involving opioid use disorder (OUD) have been increasing among Medicare beneficiaries of all ages. With rising OUD-related acute care use comes the need to understand where post-acute care is provided and the capacities for OUD treatment in those settings. Our objective was to describe hospitalized Medicare beneficiaries with OUD, their post-acute care locations, and all-cause mortality and readmissions stratified by post-acute care location.MethodsWe conducted a retrospective cohort study of acute hospitalizations using 2016-2018 Medicare Provider Analysis and Review (MedPAR) files linked to Medicare enrollment data and the Residential History File (RHF) for 100% of Medicare fee-for-service beneficiaries. The RHF which provides a person-level chronological history of health service utilization and locations of care was used to identify hospital discharge locations. We used ICD-10 codes for opioid dependence or "abuse" to identify OUD diagnoses from the MedPAR file. We conducted logistic regression to identify factors associated with discharge to an institutional setting versus home adjusting for demographics, comorbidities, and hospital stay characteristics.ResultsOur analysis included 459,763 hospitalized patients with OUD. Of these, patients aged < 65 years and those dually enrolled in Medicaid comprised the majority (59.1%). OUD and opioid overdose were primary diagnoses in 14.3% and 6.2% of analyzed hospitalizations, respectively. We found that 70.3% of hospitalized patients with OUD were discharged home, 15.8% to a skilled nursing facility (SNF), 9.6% to a non-SNF institutional facility, 2.5% home with home health services, and 1.8% died in-hospital. Within 30 days of hospital discharge, rates of readmissions and mortality were 29.7% and 3.9%; respectively, with wide variation across post-acute locations. Factors associated with greater odds of discharge to institutional settings were older age, female sex, non-Hispanic White race and ethnicity, dual enrollment, longer hospital stay, more comorbidities, intensive care use, surgery, and primary diagnoses including opioid or other drug overdoses, fractures, and septicemia.ConclusionsMore than one-quarter (25.8%) of hospitalized Medicare beneficiaries with OUD received post-acute care in a setting other than home. High rates and wide variation in all-cause readmissions and mortality within 30 days post-discharge emphasize the need for improved post-acute care for people with OUD.

Dataset Information

Using machine learning to predict risk of incident opioid use disorder among fee-for-service Medicare beneficiaries: A prognostic study.

Publications

Using machine learning to predict risk of incident opioid use disorder among fee-for-service Medicare beneficiaries: A prognostic study.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets