• Epidemiology and outcomes of depression following cardiovascular events in elderly Medicare beneficiaries

      Blanchette, Christopher Michael; Simoni-Wastila, Linda (2007)
      Objective. To assess and compare elderly Medicare beneficiaries' occurrence of depression and associated treatment following a thrombotic cardiovascular event (TCE) for (1) annual prevalence rates, (2) healthcare services utilization outcomes in the first twelve months following a TCE, and (3) the occurrence of recurrent TCEs or death in the first twelve months following a TCE. Methods. Elders enrolled in the 1997 to 2002 Medicare Current Beneficiary Survey with a TCE (International Classification of Diseases, Ninth Revision (ICD-9) codes 410, 411, 413, 414, 415, 433--438, 452, or 453). Depression (ICD-9 codes 296.2, 296.3, 296.5, 296.6, 298.0, 300.4, 308.0, 309.0, 309.1, 309.4, or 311) was assessed by a claim within six months after the TCE. Demographic and descriptive characteristics were assessed. Prevalence rates of depression and associated antidepressant utilization rates by class were calculated. Time to first healthcare service use, recurrent TCE, and death were assessed using Cox-proportional hazard models. Counts of office visits were assessed using negative binomial regression models. Results. The sample included 7,051 elders with a TCE. The prevalence rate of depression was 7.6% across the study period. Close to 70% of elders with a depression claim were using an antidepressant in the year of depression diagnosis and 53% were using SSRI antidepressants. A depression claim was associated with 51% sooner hospitalization (95% CI = 1.31, 1.76), 56% sooner emergency department visit (95% CI = 1.29, 1.90), 19% sooner outpatient hospital visit (95% CI = 1.03, 1.38). Depression was associated with a shorter time to death (p = 0.008) in the unadjusted analysis; however not associated with time to death or recurrent events in adjusted analysis. Antidepressant use was not associated with any outcome. Conclusions. Prevalence rates of post-TCE depression were much lower than rates reported in previous studies. Depression is associated with more acute healthcare services and sooner during the first twelve months following a TCE; however not associated with time to recurrent event or death. Antidepressant use has no effect on outcomes.
    • Healthcare Resource Utilization and Costs Associated with Antiretroviral Regimen Complexity

      Fleming, Sean P.; Simoni-Wastila, Linda; 0000-0001-5332-0006 (2022)
      Background Single tablet antiretroviral regimens (STR) offer one-pill, once-a-day dosing for people living with HIV. In commercially insured populations under 50 years of age, these formulations have been associated with better adherence and lower healthcare resource utilization and costs compared to multi-tablet regimens (MTR). The objectives of this dissertation were to detail patterns of STR and MTR use and the associated healthcare utilization, incremental costs, and overall budget impact among Medicare beneficiaries living with HIV. We relied on Andersen’s Behavioral Model, Generalized Estimating Equations (GEE), instrumental variable regression, and budget impact modeling to answer these questions. Methods Using a 5% random sample of Medicare Fee-for-Service beneficiaries from the Chronic Conditions Data Warehouse (CCW) we identified Medicare beneficiaries living with HIV from January 1, 2014 through December 31, 2019. We used generalized estimating equation to estimate trends in antiretroviral and other healthcare resource utilization. We estimated 1-year incremental total direct medical costs with an instrumental variable approach via two-stage least squares regression. We utilized these estimates of utilization and costs to build a five-year budget impact model. Results Among 8,316 Medicare beneficiaries contributing 282,258 person-months of observation, STR receipt was 24% more likely in 2019 (57%) than it was in 2014 (27%), and integrase-based regimens were predominant among both STR (80%) and MTR (74%) by 2019. Among 7,044 beneficiaries who initiated a new ART regimen, the instrumented STR variable was associated with statistically significant lower 1-year incremental total direct medical costs compared to MTR (-$13,487.70, (95% Confidence Interval: -$16,750.77 to -$10.224.63), p<.001). Our budget impact model predicted $1.8 billion in additional costs in year 5 under a scenario where current trends in ART utilization and list prices continue compared to a scenario where they remain at 2019 levels. The excess costs were driven by $2.3 billion in greater ART spending among the 147,953 beneficiaries living with HIV despite reductions in all other healthcare resource utilization categories and related savings. Conclusion These findings demonstrate the adoption and predominance of STR and integrase-based regimens among Medicare beneficiaries living with HIV. STRs were associated with significantly lower 1-year total direct medical costs compared to MTR among an understudied population. This study improves upon prior study designs and demonstrates the usefulness of prescriber preference IVs. However, the magnitude of the reduction in HcRU associated with STR may not be adequate enough to justify increases in ART costs as was observed during the study period. Though still small in number, the costs of caring for people living with HIV in Medicare will likely continue to grow as a proportion of total spending and will be markedly impacted by any policies aimed at controlling expenditures on prescription drugs.
    • Impact of Prescription Drug Monitoring Program Implementation and Rigor on Prescription Opioid Utilization in Medicare

      Moyo, Patience; Simoni-Wastila, Linda; 0000-0003-1323-4554 (2017)
      Background: Prescription drug monitoring programs (PDMPs) are central to the federal and state policy responses to address prescription drug abuse. PDMPs are state-run electronic databases used to track the prescribing and dispensing of controlled prescription medications. Despite their prominence, there is limited and mixed evidence of PDMP effectiveness, particularly among vulnerable populations. This study aimed to evaluate the influence of PDMP implementation and program rigor on prescription opioid utilization among disabled and older adults. Methods: A retrospective study using 2007-2012 Medicare claims and PDMP state laws from the Prescription Drug Abuse Policy System was designed to quantify associations between PDMP status or rigor and state- and individual-level opioid utilization (opioid volume, days supplied, daily morphine equivalents, number of prescriptions, daily dose ≥120mg), accounting for sociodemographic characteristics and state controlled substance laws. A PDMP composite score was developed from the total number of best practices adopted by each state (range: 0-14), classifying states according to the median score ("high PDMP rigor" and "low PDMP rigor"). Generalized linear, negative binomial, and modified Poisson regression models adjusting for clustering were applied. Results: From 2007-2012, the number of states operating PDMPs rose from 27 to 44. PDMP implementation was associated with reduced opioid volume (-2.36kg/month, 95% CI -3.44, -1.28) compared to non-PDMP states. Observed reductions were stronger in disabled adults than older adults. Annual prescription rates per 10,000 opioid-users were lower in states with low PDMP rigor (-578 [95% CI: -1006, -151]) or high rigor (-687 [95% CI: -1081, -293]) than non-PDMP states. At the individual level, PDMPs of any rigor were associated with decreased opioid utilization. There was no significant evidence that estimated associations between states with low and high rigor PDMPs were different. Conclusions: Findings suggest PDMP rigor has limited impact on individual-level opioid utilization among Medicare beneficiaries. Further studies are needed to elucidate which PDMP characteristics add value rather than adding operating cost and effort with little return.
    • Influence of Comorbid Depression on Mortality among SSDI-eligible Medicare Beneficiaries with Chronic Obstructive Pulmonary Disease

      Qian, Jingjing; Simoni-Wastila, Linda (2012)
      Background: Chronic obstructive pulmonary disease (COPD) is a condition with high mortality and morbidity. Comorbid depression can place COPD patients at increased risk of adverse outcomes. Although both COPD and depression are associated with significant morbidity, to date few studies addressing COPD-related outcomes have included individuals who receive Social Security Disability Insurance (SSDI). Objectives: To examine the influence of comorbid depression on mortality among a nationally-representative sample of Medicare beneficiaries suffering from COPD by SSDI-eligibility status. Methods: This retrospective cohort study used a 5% random sample of the 2006-2008 Chronic Condition Warehouse administrative data. The study cohort included 93,019 Medicare beneficiaries diagnosed with COPD who lived through 2006 and were continuously enrolled in Medicare Parts A, B, and D. Two-year (2007-2008) all-cause mortality was the study outcome. Comorbid depression was measured in 2006-2008. SSDI-eligibility was defined using the original reason for Medicare entitlement. Multivariable generalized estimating equations models estimated the association between SSDI-eligibility and depression, as well as the modification effect of SSDI-eligibility on their relationship. Survival analyses using extended Cox proportional hazards models further estimated risk of death from depression and antidepressant treatment among beneficiaries aged 65 and older (n=75,699) by SSDI-eligibility. Results: About two-fifths (39.4%) of beneficiaries with COPD had a depression diagnosis in 2006-2008; of those, 79.5% received antidepressant treatment. SSDI-eligibility was not only associated with a 12% (95%CI=10%,15%) higher likelihood of depression but also modified factors in regard to depression diagnosis and receipt of antidepressant treatment. COPD beneficiaries with a baseline depression diagnosis had a higher risk of death (HR=1.13; 95%CI=1.09, 1.18) in non-SSDI-eligible beneficiaries. Those who received antidepressant treatment had reduced risk of death, with greater benefits on mortality in SSDI-eligible than non-SSDI-eligible beneficiaries. Conclusions: This study provides the first evidence suggesting that SSDI-eligibility is not only associated with higher likelihood of having a depression diagnosis, but also is a significant effect modifier of the relationship between antidepressant treatment and mortality in Medicare beneficiaries with COPD. Findings demonstrate the benefits of antidepressant treatment on mortality in both SSDI-eligible and non-SSDI-eligible beneficiaries. In practice, clinicians should consider timely antidepressant treatment to improve outcomes for this population.
    • Management of Traumatic Brain Injury with Statins among Older Medicare Beneficiaries

      Khokhar, Bilal; Simoni-Wastila, Linda; 0000-0003-0143-1390 (2016)
      Background: Traumatic brain injury (TBI) is a major health concern for older adults aged 65 and older. Older TBI patients are at increased risk of primary injury (in-hospital and all-cause mortality) and secondary injury (stroke, depression, and Alzheimer's disease and related dementias (ADRD)). There is limited research regarding optimal pharmacotherapeutic options and management of TBI patients; however, several studies have highlighted statins, used to treat hyperlipidemia, as potential pharmacologic agents to reduce inflammation and improve impaired cerebral blood flow associated with primary and secondary injury. The objectives of the study are to: 1) quantify statin utilization, and 2) determine the associations between statin use and primary and secondary injury among TBI patients. Methods: Statin use (atorvastatin, fluvastatin, lovastatin, pravastatin, rosuvastatin, and simvastatin), primary injury, and secondary injury were examined among Medicare beneficiaries hospitalized with a TBI between 2006 and 2010. Logistic regression was used to investigate the relationship between pre-TBI statin use and in-hospital mortality, while discrete time analysis was used to investigate the relationship between statin use following TBI and all-cause mortality and secondary injury. Results: Among the 75,698 beneficiaries who met study criteria, 37,874 (50.0%) beneficiaries used a statin at least once during the study period. The most common statin used was simvastatin, followed by atorvastatin. Fluvastatin was the least used statin. Pre-TBI use of atorvastatin (odds ratio (OR) 0.88; 95% confidence interval (CI) 0.82, 0.96), simvastatin (OR 0.84; 95% CI 0.79, 0.91), and rosuvastatin (OR 0.79; 95% CI 0.67, 0.94) were associated with significant decreases in the risk of in-hospital mortality. Any statin use was associated with reduced all-cause mortality following TBI-hospitalization discharge. Atorvastatin and simvastatin use also were associated with reductions in all secondary injury outcomes. Conclusion: Tens of thousands of older adults are hospitalized annually with TBI and experience disabling primary and secondary injury; findings from these analyses have salient implications for reducing the risk of TBI complications among older adults. The evidence generated suggests that preemptive use of statins may decrease the risk of in-hospital and all-cause mortality, as well as reduce the likelihood of stroke, depression, and ADRD.
    • Metabolic Syndrome and Second Generation Antipsychotics Utilization and Expenditure

      Yang, Hui-Wen Keri; Simoni-Wastila, Linda (2010)
      Objectives: Studies concerning metabolic effects associated with second generation antipsychotics (SGAs) have failed to examine the impact of psychiatric comorbidities or medications. This study aimed to determine the association of metabolic syndrome (MetS) and SGAs, and the incremental contributions of select psychiatric comorbidities and polypharmacy to MetS and expenditures of antipsychotic users with MetS. Methods: Descriptive and regression analyses were applied in a large administrative claims database of antipsychotic users to examine the association between SGAs (aripiprazole, ziprasidone, risperidone, quetiapine, and olanzapine) and MetS, as well as the effects of psychiatric comorbidity and polypharmacy. Select psychiatric comorbidities included schizophrenia, bipolar, depression, and other psychiatric disorders. Psychiatric polypharmacy was defined as concomitant use of antipsychotics with psychiatric drugs with metabolic effects (selective serotonin reuptake inhibitors [SSRIs], tricyclic antidepressants [TCAs], other antidepressants, and mood stabilizers). Outcomes of interest included MetS and total medical expenditures (all-cause, non-psychiatric and psychiatric-related). Results: Of 50,128 antipsychotic users, prevalence of MetS was lower in SGA users than non-SGA users (7.6% vs. 12.8%; p<.0001). Non-SGA users were older and had higher prevalence of MetS components. However, SGA users exhibited more indicators of psychiatric severity, evidenced through higher prevalence of psychiatric disorders and higher psychiatric polypharmacy. Multivariable regression analysis showed lower odds of MetS in SGA users (OR=0.86; p<.001) than non-SGA users. Concomitant use of antidepressants including SSRIs and TCAs significantly increased the odds of MetS (OR=1.26 and 1.29, respectively), as did diagnoses of schizophrenia, bipolar or depression (OR=1.22, 1.18, 1.12, respectively) (all p<.001). Among MetS patients, total medical expenditures were higher in SGA users than non-SGA users (median $15,077 vs. $7,776, respectively; p<.0001). Controlling for select psychiatric comorbidities and polypharmacy in antipsychotic users with MetS, SGA use increased total medical expenditures by 16.6 % (p<.0001). Psychiatric comorbidity and polypharmacy also significantly contributed to total medical expenditures by 23-30% (p<.001). Conclusion: Psychiatric comorbidity and polypharmacy increased the odds of MetS in antipsychotic users and contributed to higher expenditures in antipsychotic users with MetS. Findings suggest besides metabolic effects, clinicians need to consider patients' psychiatric comorbidity and polypharmacy burdens when prescribing SGAs.
    • NSAIDs and cardiovascular adverse events: Is increasing risk associated with increasing Cox-II selectivity ratios?

      Slagle, Ashley Fenstermacher; Simoni-Wastila, Linda (2008)
      Background. Recent studies have suggested there is an increased risk of cardiovascular (CV) events with some or all of the more Cox-II selective NSAIDs. One theory for this increased risk proposes that a shift in the balance between prostacyclin, an antithrombotic vasodilator that is reduced by Cox-II inhibition, and thromboxane A2, a prothrombotic vasoconstrictor that is reduced by Cox-I inhibition will theoretically increase the risk for CV and thrombotic events. This study attempts to specifically evaluate this theory by examining the association between CV events and the relative Cox-II to Cox-I selectivity of NSAIDs. Methods. Using 2000-2001 MarketScan claims data, cross-sectional and longitudinal analyses were performed. Univariate and bivariate analyses were performed to describe and compare new, adult users of NSAIDs. Propensity score matching was used to create a subset of these users, which was balanced across numerous covariates. Finally, multivariate, time to event analyses were performed using the matched sample to test the association between relative Cox-II to Cox-I selectivity ratio and possible adverse cardiovascular events. Results. Patients who were older, male, and had higher baseline cardiovascular risk were significantly more likely to receive a higher Cox-II selective drug than a lower Cox-II selective drug. In the propensity score matched sample, adjusted for confounders, the hazard of CV event was not significantly associated with increasing Cox-II selectivity ratios. Conclusion. This study shows that patients with an increased baseline risk of CV events are significantly more likely to receive a higher Cox-II selective NSAID. After controlling for these baseline differences, no significant association was found between selectivity ratios and CV events. Therefore, using Cox-II selectivity ratios to rank order NSAIDs in terms of risk of CV events is not appropriate and practitioners should not rely on selectivity ratios in making prescribing decisions. Additional studies should evaluate other potential mechanisms contributing to the CV risk of NSAIDs, while further study, in a larger sample and for a longer duration, should be performed to confirm the absence of an association between CV adverse events and Cox-II selectivity ratios.
    • Optimizing Pain Management in Medically Complex Long-Term Care Residents

      Kuzucan, Aida; Simoni-Wastila, Linda; 0000-0003-0893-7028 (2021)
      Problem statement: While much needed clinical research has emphasized appropriate opioid stewardship in the general population, the needs of long-term nursing home care (LTC) residents remain largely ignored. Methods: This dissertation identified emerging trends in opioid therapy and initial opioid dosing patterns among LTC Medicare beneficiaries using Medicare Parts A, B and D claims, the Minimum Data Set 3.0 (MDS) and LTCFocus datasets. Aim 1 is a repeat cross-section study using resident and facility adjusted generalized estimating equations (GEE) to examine patterns of opioid use alone and in conjunction with pain-adjuvant medications among general, hospice, cancer, non-cancer chronic pain and dementia-related LTC stays from 2011 to 2015. Aim 2 identifies common patterns of average morphine equivalent daily dosing (MEDD) across six 30-day intervals starting with the first opioid prescription using latent class growth modeling (LCGM). Multivariate multinomial regression quantifies associations between different opioid use patterns over time and resident characteristics. Aim 3 accesses the odds of falls among residents with the highest probability of belonging to each of the commonly identified opioid dosing patterns with a facility clustered GEE model. Results: From 2011 to 2015, adjusted analyses found no constant significant changes in dose, duration, or frequency of opioid use. Increased use of anticonvulsant and skeletal muscle relaxants in opioid-related stays, particularly among residents with dementia, were found. LCGM identified four common opioid dosing patterns; extended high, short-term, intermittent and restart. Almost half of LTC residents received extended high opioid dosing. Multinomial regression found significant associations between sex, race, U.S. geographical region, pain diagnosis and receipt of other pain treatments with receipt of extended high dose therapy. Fall odds were found to be similar in the extended high and short-term groups. Models did find increased odds of falls in groups with less opportunity to develop tolerance (i.e., the restart and intermittent groups). Findings were not consistently significant in stratified analyses. Conclusions: Opioid use varies by resident characteristics. Opioid dosing varies over the course of therapy. More research on what factors lead to decisions regarding pain treatment and the impact of opioid dosing strategies on health-related outcomes are warranted.
    • Patterns, Factors and Outcomes associated with Gabapentin use in Combination with Opioids and Benzodiazepines among Social Security Disability Insurance (SSDI)-eligible Medicare Beneficiaries

      Olopoenia, Abisola; Simoni-Wastila, Linda (2021)
      Background: Little is known about the patterns, factors, and public health outcomes associated with concurrent utilization of gabapentin, opioids, and benzodiazepines (GABA+OP+BZD) Objective: To examine the patterns, factors, and public health outcomes associated with concurrent utilization of GABA+OP+BZD among Social Security Disability Insurance (SSDI) eligible beneficiaries. Methods: Using a 5% sample of 2013-2016 Medicare data, we utilized a retrospective cohort design to examine the following patterns of concurrent utilization: monotherapy, dual therapy, tri-therapy, switching, augmentation, discontinuation, and continuation. Similarly, a retrospective cohort design was utilized to examine the sociodemographic and clinical factors associated with the longest concurrent medication utilization episode, defined based on the overlap of prescriptions for GABA+OP+BZD. We used a nested case control design to examine the association between concurrent utilization of GABA+OP+BZD and adverse outcomes (respiratory depression, substance and opioid related overdose, and adverse drug-related events) among disabled beneficiaries with acute pain [AP], chronic pain [CP], and mental health conditions [MH]. Results: Among disabled beneficiaries, gabapentin initiators were significantly more likely to become dual and tri-therapy users (p<0.01) and to augment therapy (50.1%) when compared to opioid (28.7%) and benzodiazepine (38.7%) users; the majority augmented within 2-months after initiating therapy. Back pain [AOR(95%CI): 1.23(1.07-1.41)], chronic pain [1.27 (1.07-1.51)], mental health [1.16 (1.02-1.33)], opioid dose [1.05 (1.03-1.06)] and duration [1.07 (1.06-1.07)], and benzodiazepine duration [1.06 (1.05-1.06)] were positive predictors of having longest concurrent use involving GABA+OP+BZD. Concurrent GABA+OP+BZD use was associated with increased odds of respiratory depression [AP: 1.35 (1.19-1.52), CP:1.24 (1.11-1.38) and MH: 1.16 (1.02-1.32)], opioid related overdose [AP: 1.43 (1.04-1.98), CP: 1.47 (1.07-2.00) and MH: 1.44 (1.04-2.00)], substance related overdose[AP: 1.77 (1.26-2.50), CP: 1.70 (1.24-2.34) and MH: 1.92 (1.31-2.82)] and adverse drug related events[AP: 1.36 (1.22-1.50), CP: 1.23 (1.10-1.36) and MH: 1.15 (1.02-1.30)]. Conclusion: Our study provides the first evidence of patterns, factors, and outcomes associated with concurrent utilization of GABA+OP+BZD. Given noted adverse outcomes associated with GABA+OP+BZD, it is imperative that the benefits and risks of co-prescribing these medications be examined comprehensively, especially for those at the greatest risk of being prescribed these medications.
    • Pharmacological Treatment for Serious Mental Illness: Geographic Variation and Association with Preventable Hospitalizations

      HUANG, TING-YING; Simoni-Wastila, Linda (2016)
      Background: The number of older adults with serious mental illness (SMI), including bipolar disorder, schizophrenia, and depressive disorders, is expected to increase. Yet, SMI treatment access and its association with adverse outcomes in this specific population are not well established. This study aims to quantify SMI pharmacological treatment initiation among older adults with SMI, analyze its geographic variation, and examine its association with preventable hospitalizations. Methods: Using 2006-2012 Medicare administrative and claims data, this retrospective cohort study identified fee-for-service beneficiaries newly-diagnosed with SMI. Pharmacological treatment initiation was defined as any prescription fill for medications indicated for the newly-diagnosed SMI in the 12 months after diagnosis, with no use in the 6 months before initiation. The crude and adjusted regional pharmacological treatment incidences were summarized at the hospital referral region level and examined with spatial clustering using local indicators of spatial autocorrelation (LISA). Preventable hospitalizations were measured by the count of hospital or emergency department admissions related to ambulatory care-sensitive conditions (e.g., diabetes, cardiovascular, respiratory disease) during the same follow-up period and compared between SMI treatment pharmacological initiators and nonusers. Generalized linear mixed models with random intercepts were conducted to generate all estimates, adjusting for beneficiary demographics, comorbidities, health services utilization, regional physician supply, and spatial clustering of regional SMI pharmacological treatment incidences. Results: Of the 38,607 beneficiaries aged 65 and older identified with newly-diagnosed SMI in 2008-2012, 64.8% initiated pharmacological treatment after diagnosis. The sample was predominantly female (74.0%) and white (85.1%), with a mean age of 78.5 years. LISA results visualized highly-localized regional pharmacological treatment incidences, with hot spots clustering in the Midwest and upper Pacific West and cold spots in the West South Central and lower New England regions after adjustment. Compared with nonusers, SMI pharmacological treatment initiators showed a 12% reduced risk for preventable hospitalizations (RtR 0.88, 95% CI 0.84-0.93). Conclusions: Findings suggest the majority of older adults with SMI receive pharmacological treatment after diagnosis. Clustering of regional SMI pharmacological treatment incidences implies locally-shared physician practice styles in treating SMI. Timely SMI pharmacological treatment initiation plays an important role in managing risks for preventable adverse outcomes.
    • Use of Machine Learning To Predict COPD Treatments and Exacerbations in Medicare Older Adults: A Comparison of Multiple Approaches

      Le, Tham Thi; Simoni-Wastila, Linda (2021)
      Background: Multiple comorbidities, suboptimal adherence to maintenance medications (MMs), and exacerbations remain clinically important problems among older adults with chronic obstructive pulmonary disease (COPD). To better understand comorbidity profiles and to facilitate risk-based strategies for disease management, this dissertation quantified the prevalence and newly diagnosed rates of comorbidities, and validated predictive models of COPD medication non-adherence and exacerbations in the older Medicare population. Methods: Comorbidities were quantified in COPD beneficiaries and compared with matched non-COPD individuals using multivariable logistic regression. In a cohort of COPD beneficiaries with prevalent and new MM use, logistic and LASSO regressions were used to cross-validate the prediction of one-year non-adherence to MMs using different sets of predictors. A time-varying design was applied to assess improvement in predicting COPD exacerbations of the super learner versus component approaches (logistic regression, elastic net regression, random forest, gradient boosting, and neural network). Results: COPD beneficiaries had significantly increased odds of 40 measured comorbidities relative to matched non-COPD controls. The best-performing models in predicting MM non-adherence were those including initial MM adherence as a predictor, with validated Area Under the ROC Curves (AUC: 0.871-0.881). In predicting COPD exacerbations there were time-varying estimates of predictive accuracy and associations between predictors and the exacerbation outcome. Super learner performed slightly better (AUC: 0.650-0.761) than individual machine learning methods. Conclusions: Comorbidity burden is substantial and increases over time among Medicare older adults with COPD. Generated models achieved good and average discrimination in predicting COPD medication non-adherence and exacerbations, respectively. COPD hospitalization, oxygen supplementation, COPD treatment adherence, and numbers of inpatient visits were the most important predictors of COPD medication non-adherence and exacerbations. Super learner demonstrates a slight improvement compared to component methods, suggesting potential usability in augmenting prediction. Validated models with good discrimination can be adopted using friendly tools to optimizing resources for risk-based management and interventions of COPD.