Publications

OBJECTIVE: To determine whether (a) quality in schizophrenia care varies by race/ethnicity and over time and (b) these patterns differ across counties within states.

DATA SOURCES: Medicaid claims data from California, Florida, New York, and North Carolina during 2002-2008.

STUDY DESIGN: We studied black, Latino, and white Medicaid beneficiaries with schizophrenia. Hierarchical regression models, by state, quantified person and county effects of race/ethnicity and year on a composite quality measure, adjusting for person-level characteristics.

PRINCIPAL FINDINGS: Overall, our cohort included 164,014 person-years (41-61 percent non-whites), corresponding to 98,400 beneficiaries. Relative to whites, quality was lower for blacks in every state and also lower for Latinos except in North Carolina. Temporal improvements were observed in California and North Carolina only. Within each state, counties differed in quality and disparities. Between-county variation in the black disparity was larger than between-county variation in the Latino disparity in California, and smaller in North Carolina; Latino disparities did not vary by county in Florida. In every state, counties differed in annual changes in quality; by 2008, no county had narrowed the initial disparities.

CONCLUSIONS: For Medicaid beneficiaries living in the same state, quality and disparities in schizophrenia care are influenced by county of residence for reasons beyond patients' characteristics.

BACKGROUND: Quality improvement efforts are frequently tied to patients achieving ≥80% medication adherence. However, there is little empirical evidence that this threshold optimally predicts important health outcomes.

OBJECTIVE: To apply machine learning to examine how adherence to oral hypoglycemic medications is associated with avoidance of hospitalizations, and to identify adherence thresholds for optimal discrimination of hospitalization risk.

METHODS: A retrospective cohort study of 33,130 non-dual-eligible Medicaid enrollees with type 2 diabetes. We randomly selected 90% of the cohort (training sample) to develop the prediction algorithm and used the remaining (testing sample) for validation. We applied random survival forests to identify predictors for hospitalization and fit survival trees to empirically derive adherence thresholds that best discriminate hospitalization risk, using the proportion of days covered (PDC).

OUTCOMES: Time to first all-cause and diabetes-related hospitalization.

RESULTS: The training and testing samples had similar characteristics (mean age, 48 y; 67% female; mean PDC=0.65). We identified 8 important predictors of all-cause hospitalizations (rank in order): prior hospitalizations/emergency department visit, number of prescriptions, diabetes complications, insulin use, PDC, number of prescribers, Elixhauser index, and eligibility category. The adherence thresholds most discriminating for risk of all-cause hospitalization varied from 46% to 94% according to patient health and medication complexity. PDC was not predictive of hospitalizations in the healthiest or most complex patient subgroups.

CONCLUSIONS: Adherence thresholds most discriminating of hospitalization risk were not uniformly 80%. Machine-learning approaches may be valuable to identify appropriate patient-specific adherence thresholds for measuring quality of care and targeting nonadherent patients for intervention.

Accountable care using global payment with performance bonuses has shown promise in controlling spending growth and improving care. This study examined how an early model, the Alternative Quality Contract (AQC) established in 2009 by Blue Cross Blue Shield of Massachusetts (BCBSMA), has affected care for mental illness. We compared spending and use for enrollees in AQC organizations that did and did not accept financial risk for mental health with enrollees not participating in the contract. Compared with BCBSMA enrollees in organizations not participating in the AQC, we found that enrollees in participating organizations were slightly less likely to use mental health services and, among mental health services users, small declines were detected in total health care spending, but no change was found in mental health spending. The declines in probability of use of mental health services and in total health spending among mental health service users attributable to the AQC were concentrated among enrollees in organizations that accepted financial risk for behavioral health. Interviews with AQC organization leaders suggested that the contractual arrangements did not meaningfully affect mental health care delivery in the program's initial years, but organizations are now at varying stages of efforts to improve mental health integration.

BACKGROUND: Use of analgesics has been associated with lower risk of ovarian cancer, but, to date, very few studies have explored the association between analgesics and ovarian cancer survival.

METHODS: We examined the relationship between self-reported prediagnostic use of aspirin, ibuprofen, and acetaminophen and overall survival (OS), progression-free survival (PFS), ascites at the time of primary treatment, and persistence of disease after primary treatment among 699 women diagnosed with epithelial ovarian carcinoma. The associations between use of these medications and OS and PFS were estimated using Cox proportional hazards models. We utilized unconditional logistic regression models to estimate associations between medication use and presence of ascites and persistence of disease.

RESULTS: Prediagnostic intake of aspirin, both low-dose and regular-dose, ibuprofen, and acetaminophen was not associated with any of the outcomes of interest.

CONCLUSIONS: Our results indicate a lack of association between prediagnostic intake of selected analgesics and OS, PFS, presence of ascites at the time of primary treatment, and persistence of disease after primary treatment.

IMPACT: Prediagnostic intake of analgesics may not be associated with ovarian cancer outcomes.

BACKGROUND: Health insurance claims data may play an important role for health care systems and payers in monitoring the nonmedical use of prescription opioids (NMPO) among patients. However, these systems require valid methods for identifying NMPO if they are to target individuals for intervention. Limited efforts have been made to define NMPO using administrative data available to health systems and payers. We conducted a systematic review of publications that defined and measured NMPO within health insurance claims databases in order to describe definitions of NMPO and identify areas for improvement.

METHODS: We searched 8 electronic databases for articles that included terms related to NMPO and health insurance claims. A total of 2613 articles were identified in our search. Titles, abstracts, and article full texts were assessed according to predetermined inclusion/exclusion criteria. Following article selection, we extracted general information, conceptual and operational definitions of NMPO, methods used to validate operational definitions of NMPO, and rates of NMPO.

RESULTS: A total of 7 studies met all inclusion criteria. A range of conceptual NMPO definitions emerged, from concrete concepts of abuse to qualified definitions of probable misuse. Operational definitions also varied, ranging from variables that rely on diagnostic codes to those that rely on opioid dosage and/or filling patterns. Quantitative validation of NMPO definitions was reported in 3 studies (e.g., receiver operating curves or logistic regression), with each study indicating adequate validity. Three studies reported qualitative validation, using face and content validity. One study reported no validation efforts. Rates of NMPO among the studies' populations ranged from 0.75% to 10.32%.

CONCLUSIONS: Disparate definitions of NMPO emerged from the literature, with little uniformity in conceptualization and operationalization. Validation approaches were also limited, and rates of NMPO varied across studies. Future research should prospectively test and validate a construct of NMPO to disseminate to payers and health officials.

OBJECTIVES: To examine the association between statin use and objectively assessed decline in gait speed in community-dwelling older adults.

DESIGN: Longitudinal cohort study.

SETTING: Health, Aging and Body Composition (Health ABC) Study.

PARTICIPANTS: Two thousand five participants aged 70-79 at baseline with medication and gait speed data at 1998-99, 1999-2000, 2001-02, and 2002-03.

MEASUREMENTS: The independent variables were any statin use and their standardized daily doses (low, moderate, high) and lipophilicity. The primary outcome measure was decline in gait speed of 0.1 m/s or more in the following year of statin use. Multivariable generalized estimating equations were used, adjusting for demographic characteristics, health-related behaviors, health status, and access to health care.

RESULTS: Statin use increased from 16.2% in 1998-99 to 25.6% in 2002-03. The overall proportions of those with decline in gait speed of 0.1 m/s or more increased from 22.2% in 1998 to 23.9% in 2003. Statin use was not associated with decline in gait speed of 0.1 m/s or more (adjusted odds ratio (AOR) = 0.90, 95% confidence interval (CI) = 0.77-1.06). Similar nonsignificant trends were also seen with the use of hydrophilic or lipophilic statins. Users of low-dose statins were found to have a 22% lower risk of decline in gait speed than nonusers (AOR = 0.78, 95% CI = 0.61-0.99), which was mainly driven by the results from 1999-2000 follow-up.

CONCLUSION: These results suggest that statin use did not increase decline in gait speed in community-dwelling older adults.

OBJECTIVES: Use of buprenorphine - an effective treatment for opioid use disorders (OUDs) - has increased rapidly in recent years and is often financed by Medicaid. We investigated predictors of buprenorphine treatment, patterns of care, and quality of care in a large state Medicaid program.

METHODS: Data from Pennsylvania Medicaid from 2007 to 2012 provided information regarding diagnoses, demographic characteristics, enrollment, and use of inpatient and outpatient services, and prescription drugs. We identified adult enrollees using buprenorphine, and examined prevalence of OUD diagnosis and patterns of use (duration and dose) and quality of care (physician visits, receipt of behavioral health counseling, urine drug screens, and other prescription drug use). We use a mixed logistic regression model to examine enrollee characteristics associated with buprenorphine use.

RESULTS: The share of enrollees with OUD filling prescriptions for buprenorphine increased from 2985 (9.8%) to 12,691 (25.2%) from 2007 to 2012. Between 26.2 and 32.0% of enrollees using buprenorphine had no diagnosis of OUD, depending on the year. Only 60.1% of enrollees with buprenorphine use received at least one urine drug screen, 41.0% had behavioral health counseling services, and 34.7 and 38.0% had other opioid and benzodiazepine claims, respectively, concomitant with buprenorphine use. Quality of care was lower among those with no OUD diagnosis recorded. The mean daily doses of buprenorphine decreased over time. We found wide variation in likelihood of buprenorphine use among those with OUD based upon age, sex, and race.

CONCLUSIONS: Increases in buprenorphine treatment in a Medicaid population were observed across time; however, increases varied by age, sex, and rate, and the quality of care received seemed to be generally poor. The quality of the provision of buprenorphine treatment occurring in Medicaid populations should be further explored.

BACKGROUND: Quality improvement efforts are frequently tied to patients achieving ≥80% medication adherence. However, there is little empirical evidence that this threshold optimally predicts important health outcomes.

OBJECTIVE: To apply machine learning to examine how adherence to oral hypoglycemic medications is associated with avoidance of hospitalizations, and to identify adherence thresholds for optimal discrimination of hospitalization risk.

METHODS: A retrospective cohort study of 33,130 non-dual-eligible Medicaid enrollees with type 2 diabetes. We randomly selected 90% of the cohort (training sample) to develop the prediction algorithm and used the remaining (testing sample) for validation. We applied random survival forests to identify predictors for hospitalization and fit survival trees to empirically derive adherence thresholds that best discriminate hospitalization risk, using the proportion of days covered (PDC).

OUTCOMES: Time to first all-cause and diabetes-related hospitalization.

RESULTS: The training and testing samples had similar characteristics (mean age, 48 y; 67% female; mean PDC=0.65). We identified 8 important predictors of all-cause hospitalizations (rank in order): prior hospitalizations/emergency department visit, number of prescriptions, diabetes complications, insulin use, PDC, number of prescribers, Elixhauser index, and eligibility category. The adherence thresholds most discriminating for risk of all-cause hospitalization varied from 46% to 94% according to patient health and medication complexity. PDC was not predictive of hospitalizations in the healthiest or most complex patient subgroups.

CONCLUSIONS: Adherence thresholds most discriminating of hospitalization risk were not uniformly 80%. Machine-learning approaches may be valuable to identify appropriate patient-specific adherence thresholds for measuring quality of care and targeting nonadherent patients for intervention.

BACKGROUND: Academic medical centers (AMCs) have increasingly adopted conflict of interest policies governing physician-industry relationships; it is unclear how policies impact prescribing.

OBJECTIVES: To determine whether 9 American Association of Medical Colleges (AAMC)-recommended policies influence psychiatrists' antipsychotic prescribing and compare prescribing between academic and nonacademic psychiatrists.

RESEARCH DESIGN: We measured number of prescriptions for 10 heavily promoted and 9 newly introduced/reformulated antipsychotics between 2008 and 2011 among 2464 academic psychiatrists at 101 AMCs and 11,201 nonacademic psychiatrists. We measured AMC compliance with 9 AAMC recommendations. Difference-in-difference analyses compared changes in antipsychotic prescribing between 2008 and 2011 among psychiatrists in AMCs compliant with ≥ 7/9 recommendations, those whose institutions had lesser compliance, and nonacademic psychiatrists.

RESULTS: Ten centers were AAMC compliant in 2008, 30 attained compliance by 2011, and 61 were never compliant. Share of prescriptions for heavily promoted antipsychotics was stable and comparable between academic and nonacademic psychiatrists (63.0%-65.8% in 2008 and 62.7%-64.4% in 2011). Psychiatrists in AAMC-compliant centers were slightly less likely to prescribe these antipsychotics compared with those in never-compliant centers (relative odds ratio, 0.95; 95% CI, 0.94-0.97; P < 0.0001). Share of prescriptions for new/reformulated antipsychotics grew from 5.3% in 2008 to 11.1% in 2011. Psychiatrists in AAMC-compliant centers actually increased prescribing of new/reformulated antipsychotics relative to those in never-compliant centers (relative odds ratio, 1.39; 95% CI, 1.35-1.44; P < 0.0001), a relative increase of 1.1% in probability.

CONCLUSIONS: Psychiatrists exposed to strict conflict of interest policies prescribed heavily promoted antipsychotics at rates similar to academic psychiatrists and nonacademic psychiatrists exposed to less strict or no policies.