Understanding the characteristics of Latino individuals with first-time DUI offenses to facilitate effective interventions.

Literature shows that Latinos who drink are more likely to experience alcohol-related consequences and less likely to seek care for alcohol misuse than Whites. We aim to understand characteristics, consumption patterns, and openness to treatment among Latino first-time offenders driving under the influence. Latino participants were significantly younger (29.0 years) than non-Latinos (37.7 years). In adjusted models, Latino participants were significantly more likely than non-Latinos to binge drink, but there were no significant group differences in amount of alcohol consumed in a typical week. There was no significant difference in incidence of alcohol-related consequences, readiness to change drinking, and driving behaviors in this sample.

Standardized assessment of medication reconciliation in post-acute care.

BACKGROUND: Medication reconciliation (MR) facilitates safety during transitions of care, which occur frequently across post-acute care (PAC) settings. Under the intent of the IMPACT Act of 2014, the Centers for Medicare & Medicaid Services contracted with the RAND Corporation to develop and test standardized assessment data elements (SADEs) that assess the MR process. METHODS: We employed an iterative process that incorporated stakeholder input and three rounds of testing to identify, refine, and evaluate MR SADEs. Testing took place in 186 PAC sites (57 home health agencies, 28 inpatient rehabilitation facilities, 28 long-term care hospitals, and 73 skilled nursing facilities). There were 2951 patients in the final test. Novel MR SADEs, based on the Joint Commission's framework, were refined. The final SADEs assessed whether: patient was taking high-risk medications; an indication was noted for each medication class; discrepancies were identified; patient or family/caregiver was involved in addressing discrepancies; discrepancies were communicated to physician (or designee) within 24 h; recommended physician actions regarding discrepancies were implemented within 24 h after physician response; and the reconciled list was communicated to patient, prescriber, and/or pharmacy. Two assessors per facility collected data for each patient. Analyses described completion time, data missingness, and interrater reliability, as well as feedback on assessor burden. RESULTS: Time to complete the MR SADEs was 3.2 min. Missing data were <5%. Interrater reliability was moderate to high (κ: 0.42 [whether a reconciled list was communicated to prescribers] to 0.89 [identifying patients taking hypoglycemics]). For identifying high-risk medication classes, interrater reliability was high (κ: 0.72-0.89). There were minimal differences by setting. CONCLUSIONS: This is the first set of MR SADEs that have been assessed across the PAC settings. Results demonstrate feasibility, based on missing data and completion time, and moderate to strong reliability, based on interrater comparisons, of assessing MR.

Who receives post-acute care? Characteristics of national population and field test sample.

BACKGROUND: Each year millions of Medicare beneficiaries in the United States receive post-acute care (PAC) in skilled nursing facilities (SNFs), inpatient rehabilitation facilities (IRFs), long-term care hospitals (LTCHs), and home health agencies (HHA). We describe, overall and by PAC setting, the national population of facilities and patients, evaluate the representativeness of a national field test sample, and describe patient characteristics in the national field test sample. METHODS: We analyzed the 2016 Provider of Service file, 2016 patient assessment data reported by PAC providers to Centers for Medicare & Medicaid, and data collected from PAC providers participating in a national field test. National data included 27,234 PAC settings and 5,033,820 beneficiaries receiving PAC. The national field test sample consisted of 143 facilities across 14 markets with 25-30 patients sampled from each facility (n = 3669). We describe PAC facility and patient characteristics for both the national and field test sample. RESULTS: Nationally, PAC facilities were more likely for-profit versus not for-profit, have an average nurse-to-bed ratio between 1:10 to 1:1 (lowest in SNFs) and be in metropolitan versus other areas. PAC patients were more likely to be white, female, and 75-89 years of age; heart failure as a primary medical condition tended to be more common than stroke or sepsis. There was limited variability across setting types. In the national field test, patients in LTCHs demonstrated a greater likelihood of cognitive impairment, positive depression screening, bowel and bladder appliance use, higher rates of medication drug classes taken, and use of therapeutic diets and IV medications. CONCLUSION: The national field test facility and patient samples were fairly representative of the national population overall and across settings with a few exceptions. Moreover, differences according to PAC setting on patient characteristics in the national field test aligned with general differences in patient populations.

Development and testing of a standardized pain interview assessment for use in post-acute care.

OBJECTIVES: Pain is highly prevalent among patients in post-acute care (PAC) settings and can affect quality of life, treatment outcomes, and transitions in care. Routine, accurate assessment of pain across settings is important for pain management and care planning; however, existing PAC assessment instruments do not assess patient pain in a standardized manner. METHODS: We developed and tested a set of pain interview data elements for use across PAC settings (skilled nursing facilities, inpatient rehabilitation facilities, long term care hospitals, home health agencies) as part of a larger effort undertaken by the Centers for Medicare & Medicaid Services to develop standardized assessment data elements to meet the requirements of the IMPACT Act of 2014. The interview assessed six pain constructs: presence; frequency; interference with sleep; interference with rehabilitation therapies [if applicable]; interference with daily activities; worst pain; and pain relief from treatments/medications). A total of 3031 PAC patients at 143 PAC settings (across 14 U.S. geographic/metropolitan areas in 10 states) participated in a national field test of standardized data elements from November 2017 to August 2018. We assessed item response distributions, time to complete interviews, inter-assessor agreement, and, for a subset of patients, change in responses between admission and discharge assessments. We also conducted focus groups with nurse assessors about their experiences administering the items. RESULTS: For patients reporting any pain, average time to complete the pain interview was 3.1 min (SD = 1.3), and interrater reliability was excellent for all data elements (kappa range: 0.95-0.99). Findings were similar across types of PAC settings. Qualitative data from nurses emphasized ease of administration and high perceived clinical utility. CONCLUSION: Findings provide support for feasibility of implementing a standardized pain interview assessment in PAC settings. This tool can support tracking of patient needs across settings and interoperability of data in electronic medical records.

Standardized assessment of cognitive function of post-acute care patients.

BACKGROUND: The assessment of cognitive function in post-acute care (PAC) settings is important for understanding an individual's condition and care needs, developing better person-directed care plans, predicting resource needs and understanding case mix. Therefore, we tested the feasibility and reliability of cognitive function assessments, including the Brief Interview for Mental Status (BIMS), Confusion Assessment Method (CAM©), Expression and Understanding, and Behavioral Signs and Symptoms for patients in PAC under the intent of the IMPACT Act of 2014. METHODS: We conducted a national test of assessments of four standardized cognitive function data elements among patients in PAC. One hundred and forty-three PAC settings (57 home health agencies, 28 inpatient rehabilitation facilities, 28 long-term care hospitals, and 73 Skilled Nursing Facilities) across 14 U.S. markets from November 2017 to August 2018. At least one of four cognitive function data elements were assessed in 3026 patients. We assessed descriptive statistics, percent of missing data, time to complete, and interrater reliability between paired research nurse and facility staff assessors, and assessor feedback. RESULTS: The BIMS, CAM©, Expression and Understanding, and Behavioral Signs and Symptoms demonstrated low rates of missing data (less than 2%), high percent agreement, and substantial support from assessors. The prevalence of Behavioral Signs and Symptoms was low in our sample of PAC settings. CONCLUSION: Findings provide support for feasibility of implementing standardized assessment of all our cognitive function data elements for patients in PAC settings. The BIMS and CAM© were adopted into federal Quality Reporting Programs in the fiscal year/calendar year 2020 final rules. Future work could consider implementing additional cognitive items that assess areas not covered by the BIMS and CAM©.

Standardized assessment of cognitive function, mood, and pain among patients who are unable to communicate.

BACKGROUND: Assessments of patients have sought to increase the patient voice through direct patient interviews and performance-based testing. However, some patients in post-acute care (PAC) are unable to communicate and cannot participate in interviews or structured cognitive tests. Therefore, we tested the feasibility and reliability of observational assessments of cognitive function, mood, and pain for patients who are unable to communicate in PAC settings. METHODS: We conducted a national test of observational assessments of cognitive function, mood, and pain in 143 PAC facilities (57 home health agencies, 28 Inpatient Rehabilitation Facilities, 28 Long-Term Care Hospitals, and 73 Skilled Nursing Facilities) across 14 U.S. markets from November 2017 to August 2018. For the 548 patients identified as unable to make themselves understood, we assessed descriptive statistics, percent of missing data, time to complete, and inter-rater reliability (IRR) between paired research nurse and facility staff assessors. RESULTS: Most sampled non-communicative patients were administered all three observational assessments. Among assessed patients, overall missing data was high for some items within the Staff Assessment for Mental Status (2.9% to 33.5%) and Staff Assessment of Patient Mood (12.4% to 44.3%), but not the Observational Assessment of Pain or Distress (0.0% to 4.4%). Average time to complete the data elements ranged from 2.4 to 3.5 min and IRR was good to excellent for all items (kappa range: 0.74-0.98). CONCLUSION: The three observational data elements had acceptable reliability. Although results revealed varying feasibility, there was support for feasibility overall in terms of implementing a standardized observational assessment of pain for patients in PAC settings. Additional work is needed for the Staff Assessment for Mental Status and the Staff Assessment of Patient Mood to improve the observable nature of these data elements and enhance instructions and training for standardizing the assessments.

Developing standardized patient assessment data elements for Medicare post-acute care assessments.

BACKGROUND: To support interoperability and care planning across provider types, the Improving Medicare Post-Acute Care Transformation Act of 2014 (IMPACT Act) requires the submission of standardized patient assessment data using the assessment instruments provided by the Centers for Medicare & Medicaid Services (CMS). CMS was tasked with developing standardized assessment data elements (SADEs) within clinical categories named in the IMPACT Act. METHOD: We used environmental scans, subject matter expert, and stakeholder input to identify candidate SADEs; tested candidate data elements in alpha testing; revised SADEs and training protocols based on alpha analyses and stakeholder feedback; tested SADEs across post-acute care (PAC) settings in a national field test that included 3121 patients across 143 home health agencies, inpatient rehabilitation facilities, long-term care hospitals, and skilled nursing facilities in 14 markets across the United States; and analyzed data and stakeholder input from national testing. Field testing measured the time required for assessment, percent completion, and inter-rater reliability. We analyzed qualitative feedback from stakeholder focus groups and technical expert panels. We also obtained survey and focus group feedback from data collectors. RESULTS: We developed a mixed-method, multi-stakeholder procedure to identify and gather input on SADE for cross-setting use. This process yielded feasible and reliable SADEs for PAC settings that assess pain, cognitive status, mood, and medication reconciliation. The success of this work depended on working iteratively with diverse stakeholders and providing qualitative as well as quantitative evidence. CONCLUSIONS: The procedures applied in this project for developing and adopting SADEs for PAC, as well as the challenges and strategies to overcome challenges, should be considered in future item and quality measure development.

Standardized assessment of depression symptoms in post-acute care: A screening threshold approach.

BACKGROUND: Depression symptoms have impacts on quality of life, rehabilitation and treatment adherence, and resource utilization among patients in post-acute care (PAC) settings. The PHQ-2 and PHQ-9 are instruments for the assessment of depression, previously used in PAC settings, that have tradeoffs in terms of measurement depth versus respondent/assessor burden. Therefore, the present study tested a gateway version of the protocol (PHQ-2 to 9). METHODS: In 143 PAC settings in 14 U.S. markets across 10 states from November 2017 to August 2018, facility and research nurses administered the PHQ-2 to communicative patients (n = 3010). Nurses administered the full PHQ-9 if the patient screened positive for either of the two cardinal symptoms assessed by the PHQ-2 (depressed mood and anhedonia). We assessed the prevalence and frequency of depression symptoms using the PHQ-2 to 9, associations between depression screening results and patient characteristics and clinical conditions, and feasibility indicators. RESULTS: More than 1 in 4 patients (28%) screened positive on the PHQ-2. Only 6% of those completing the full PHQ-9 had a score indicating "minimal" severity. The average score (M = 11.9) met the threshold for moderate depression. Positive PHQ-2 screening was associated with age, female gender, disposition at discharge, septicemia/severe sepsis, and dependence for ADLs of toileting and lying to sitting mobility. Age was also associated with full PHQ-9 scores; patients ages 45-64 had the highest mean score. Length of stay was not associated with PHQ-2 screening results or full PHQ-2 to 9 scores. Missing data were minimal (<2.4%). The average time to complete was 2.3 min. Interrater reliability and percent agreement were excellent. CONCLUSIONS: These findings suggest the feasibility of a gateway scoring approach to standardized assessment of depression symptoms among PAC patients, and that depression symptoms are relatively common among this inpatient population.

Causal inference under interference with prognostic scores for dynamic group therapy studies.

Group therapy is a common treatment modality for behavioral health conditions. Patients often enter and exit groups on an ongoing basis, leading to dynamic therapy groups. Examining the effect of high versus low session attendance on patient outcomes is a research question of interest. However, there are several challenges to identifying causal effects in this setting, including the lack of randomization, interference among patients, and the interrelatedness of patient participation. Dynamic therapy groups motivate a unique causal inference scenario, as the treatment statuses are completely defined by the patient attendance record for the therapy session, which is also the structure inducing interference. We adopt the Rubin causal model framework to define the causal effect of high versus low session attendance of group therapy at both the individual patient and peer levels. We propose a strategy to identify individual, peer, and total effects of high attendance versus low attendance on patient outcomes by the prognostic score stratification. We examine performance of our approach via simulation and apply it to data from a group cognitive behavioral therapy trial for treating depression among patients in a substance use disorders treatment setting.

Appropriate analyses of bimodal substance use frequency outcomes: a mixture model approach.

Background: In addiction research, outcome measures are often characterized by bimodal distributions. One mode can be for individuals with low substance use and the other mode for individuals with high substance use. Applying standard statistical procedures to bimodal data may result in invalid inference. Mixture models are appropriate for bimodal data because they assume that the sampled population is composed of several underlying subpopulations.Objectives: To introduce a novel mixture modeling approach to analyze bimodal substance use frequency data.Methods: We reviewed existing models used to analyze substance use frequency outcomes and developed multiple alternative variants of a finite mixture model. We applied all methods to data from a randomized controlled study in which 30-day alcohol abstinence was the primary outcome. Study data included 73 individuals (38 men and 35 women). Models were implemented in the software packages SAS, Stata, and Stan.Results: Shortcomings of existing approaches include: 1) inability to model outcomes with multiple modes, 2) invalid statistical inferences, including anti-conservative p-values, 3) sensitivity of results to the arbitrary choice to model days of substance use versus days of substance abstention, and 4) generation of predictions outside the range of common substance use frequency outcomes. Our mixture model variants avoided all of these shortcomings.Conclusions: Standard models of substance use frequency outcomes can be problematic, sometimes overstating treatment effectiveness. The mixture models developed improve the analysis of bimodal substance use frequency.

Longitudinal effects of social network changes on drinking outcomes for individuals with a first-time DUI.

BACKGROUND: Social networks are important predictors of alcohol-related outcomes, especially among those with a DUI where riskier social networks are associated with increased risk of drinking and driving. Social networks are increasingly a target for intervention; however, no studies have examined and measured whether longitudinal changes in social networks are associated with reductions in impaired driving. OBJECTIVE: The current study first examines longitudinal changes in social networks among participants receiving services following a first-time DUI, and then examines the association between network change and drinking outcomes at 4- and 10-month follow-up. METHODS: The study surveyed a subsample of participants (N = 94) enrolled in a clinical trial of individuals randomized to cognitive behavioral therapy (CBT) or usual care (UC) on an iPad using EgoWeb 2.0-an egocentric social network data collection software-about pre-DUI and post-DUI networks and their short- and long-term drinking behaviors. RESULTS: Participants were 65% male, 48% Hispanic, and an average of 32.5 years old. Overall, participants significantly reduced the proportion of network members with whom they drank from 0.41 to 0.30 (p = .001) and with whom they drank more alcohol than they wanted to from 0.15 to 0.07 (p = .0001) from two weeks prior to the DUI (measured at baseline) to 4-month follow-up. Furthermore, decreases in proportion of drinking partners over time were associated with reduced drinks per week, self-reported driving after drinking, and intentions to drive after drinking at 4-month follow-up. Participants who reported decreases in proportion of drinking partners also reported significantly less binge drinking at 10-month follow-up. Finally, increases in emotional support were associated with decreases in binge drinking at 4-month follow-up. The study found no differences in the changes in composition of networks between CBT and UC groups. CONCLUSIONS: These results suggest that individuals receiving services in DUI programs significantly reduced risky network members over time and that these social network changes were associated with reduced drinking and other indicators of risk for DUI recidivism. Clinical interventions that target reductions in risky network members may improve outcomes for those enrolled in a DUI program.

Social Networks of Clients in First-Time DUI Programs.

OBJECTIVE: Social networks play an important role in the development of and recovery from problem drinking behaviors; however, few studies have measured the social networks of individuals convicted of driving under the influence (DUI) or assessed the relationship between social network characteristics and risk for DUI relapse and recidivism. The goal of this study is to describe the social network characteristics of a first-time DUI population in the 2 weeks before the DUI incident; examine demographic differences in social network characteristics by age, ethnicity, and gender; and assess the relationship between social network characteristics and risk factors for DUI. METHOD: We collected personal (egocentric) social network survey data from 94 participants (65% male) enrolled in a randomized clinical trial comparing the effects of cognitive behavioral therapy with usual care for individuals convicted of a first-time DUI. Multivariate models were used to assess the relationship between pre-DUI personal network characteristics and risk factors for DUI measured at baseline interview. RESULTS: Results indicate that the proportion of drinking partners in one's personal network was positively associated with drinks per week, binge drinking, alcohol use, marijuana use, and alcohol-related consequences. Several dimensions of personal network support were inversely associated with risk factors for DUI. CONCLUSIONS: The pre-DUI composition of personal networks has a strong relationship to baseline risk factors for DUI; networks composed of more risky individuals (e.g., drinking partners) were associated with greater substance use and drinking and driving behaviors. Networks with greater levels of social support were associated with lower likelihood of self-reported driving after drinking and intentions to drive after drinking. Interventions that target positive and negative aspects of personal networks may enhance clinical treatments.

Group Cohesion and Climate in Cognitive Behavioral Therapy for Individuals with a First-Time DUI.

Few studies have examined group cohesion and climate in the substance use disorder treatment literature. We examined whether group cohesion and climate are associated with increased self-efficacy outcomes and reduced drinks per week, binge drinking and DUI behaviors, in a sample of individuals with a first-time DUI receiving either cognitive behavioral therapy (CBT) or usual care. Additionally, we examined whether CBT moderates these relationships. Group measures and drinking outcomes were not significantly associated. This study is the first to provide an in-depth analysis on group processes in DUI settings, and as such, provides important insights into how group processes may differ in a mandated DUI context.

Study design to evaluate a group-based therapy for support persons of adults on buprenorphine/naloxone.

BACKGROUND: Opioid use disorders (OUDs) have devastating effects on individuals, families, and communities. While medication treatments for OUD save lives and are increasingly utilized, rates of treatment dropout are very high. In addition, most existing medication treatments for OUD may often neglect the impact of untreated OUD on relationships and ignore the potential role support persons (SPs) could have on encouraging long-term recovery, which can also impact patient treatment retention. METHODS/DESIGN: The current study adapts Community Reinforcement and Family Training (CRAFT) for use with SPs (family member, spouse or friend) of patients using buprenorphine/naloxone (buprenorphine) in an outpatient community clinic setting. The study will evaluate whether the adapted intervention, also known as integrating support persons into recovery (INSPIRE), is effective in increasing patient retention on buprenorphine when compared to usual care. We will utilize a two-group randomized design where patients starting or restarting buprenorphine will be screened for support person status and recruited with their support person if eligible. Support persons will be randomly assigned to the INSPIRE intervention, which will consist of 10 rolling group sessions led by two facilitators. Patients and SPs will each be assessed at baseline, 3 months post-baseline, and 12 months post-baseline. Patient electronic medical record data will be collected at six and 12 months post-baseline. We will examine mechanisms of intervention effectiveness and also conduct pre/post-implementation surveys with clinic staff to assess issues that would affect sustainability. DISCUSSION: Incorporating the patient's support system may be an important way to improve treatment retention in medication treatments for OUD. If SPs can serve to support patient retention, this study would significantly advance work to help support the delivery of effective treatments that prevent the devastating consequences associated with OUD. Trial registration This study was registered with ClinicalTrials.gov, NCT04239235. Registered 27 January 2020, https://clinicaltrials.gov/ct2/show/NCT04239235 .

Shifting From Passive Quality Reporting to Active Nudging to Influence Consumer Choice of Health Plan.

Comparative quality information on health plan and provider performance is increasingly available in the form of quality report cards, but consumers rarely make use of these passively provided decision support tools. In 2012-2013, the Centers for Medicare & Medicaid Services (CMS) initiated quality-based nudges designed to encourage beneficiaries to move into higher quality Medicare Advantage (MA) plans. We assess the impacts of CMS' targeted quality-based nudges with longitudinal analysis of 2009-2014 MA plan enrollment trends. Nudges are associated with 17% reductions in enrollment in the lowest-performing plans and 3% increases in enrollment in the highest performing plans (annually, p < .01 for both), occurring at the time of nudge implementation and relative to trends for plans with moderate performance that were not targeted by nudges. These findings suggest that quality-based nudges can successfully steer consumers into higher quality plans and provide opportunities for purchasers and payers to increase consumers' use of quality information.

Randomized Clinical Trial Examining Cognitive Behavioral Therapy for Individuals With a First-Time DUI Offense.

BACKGROUND: Driving under the influence (DUI) programs are a unique setting to reduce disparities in treatment access to those who may not otherwise access treatment. Providing evidence-based therapy in these programs may help prevent DUI recidivism. METHODS: We conducted a randomized clinical trial of 312 participants enrolled in 1 of 3 DUI programs in California. Participants were 21 and older with a first-time DUI offense who screened positive for at-risk drinking in the past year. Participants were randomly assigned to a 12-session manualized cognitive behavioral therapy (CBT) or usual care (UC) group and then surveyed 4 and 10 months later. We conducted intent-to-treat analyses to test the hypothesis that participants receiving CBT would report reduced impaired driving, alcohol consumption (drinks per week, abstinence, and binge drinking), and alcohol-related negative consequences. We also explored whether race/ethnicity and gender moderated CBT findings. RESULTS: Participants were 72.3% male and 51.7% Hispanic, with an average age of 33.2 (SD = 12.4). Relative to UC, participants receiving CBT had lower odds of driving after drinking at the 4- and 10-month follow-ups compared to participants receiving UC (odds ratio [OR] = 0.37, p = 0.032, and OR = 0.29, p = 0.065, respectively). This intervention effect was more pronounced for females at 10-month follow-up. The remaining 4 outcomes did not significantly differ between UC versus CBT at 4- and 10-month follow-ups. Participants in both UC and CBT reported significant within-group reductions in 2 of 5 outcomes, binge drinking and alcohol-related consequences, at 10-month follow-up (p < 0.001). CONCLUSIONS: In the short-term, individuals receiving CBT reported significantly lower rates of repeated DUI than individuals receiving UC, which may suggest that learning cognitive behavioral strategies to prevent impaired driving may be useful in achieving short-term reductions in impaired driving.

Rates and Impact of Adherence to Recommended Care for Unhealthy Alcohol Use.

BACKGROUND: Unhealthy alcohol use is a major worldwide health problem. Yet few studies have assessed provider adherence to the alcohol-related care recommended in clinical practice guidelines, nor links between adherence to recommended care and outcomes. OBJECTIVES: To describe quality of care for unhealthy alcohol use and its impacts on drinking behavior RESEARCH DESIGN: Prospective observational cohort study of quality of alcohol care for the population of patients screening positive for unhealthy alcohol use in a large Veterans Affairs health system. PARTICIPANTS: A total of 719 patients who screened positive for unhealthy alcohol use at one of 11 primary care practices and who completed baseline and 6-month telephone interviews. MAIN MEASURES: Using administrative encounter and medical record data, we assessed three composite and 21 individual process-based measures of care delivered across primary and specialty care settings. We assessed self-reported daily alcohol use using telephone interviews at baseline and 6-month follow-up. KEY RESULTS: The median proportion of patients who received recommended care across measures was 32.8% (range < 1% for initiating pharmacotherapy to 93% for depression screening). There was negligible change in drinking for the study population between baseline and 6 months. In covariate-adjusted analyses, no composites were significantly associated with changes in heavy drinking days or drinks per week, and just one of nine individual measures tested was significantly associated. In a subsample of patients drinking above recommended weekly limits prior to screening, two of nine individual measures were significantly associated. CONCLUSIONS: This study shows wide variability in receipt of recommended care for unhealthy alcohol use. Receipt of recommended interventions for reducing drinking was frequently not associated with decreased drinking. Results suggest deficits in provision of comprehensive alcohol care and in understanding how to improve population-based drinking outcomes.

Sample selection in the face of design constraints: Use of clustering to define sample strata for qualitative research.

OBJECTIVE: To sample 40 physician organizations stratified on the basis of longitudinal cost of care measures for qualitative interviews in order to describe the range of care delivery structures and processes that are being deployed to influence the total costs of caring for patients. DATA SOURCES: Three years of physician organization-level total cost of care data (n = 156 in California) from the Integrated Healthcare Association's value-based pay-for-performance program. STUDY DESIGN: We fit total cost of care data using mixture and K-means clustering algorithms to segment the population of physician organizations into sampling strata based on 3-year cost trajectories (ie, cost curves). PRINCIPAL FINDINGS: A mixture of multivariate normal distributions can classify physician organization cost curves into clusters defined by total cost level, shape, and within-cluster variation. K-means clustering does not accommodate differing levels of within-cluster variation and resulted in more clusters being allocated to unstable cost curves. A mixture of regressions approach focuses overly on anomalous trajectories and is sensitive to model coding. CONCLUSIONS: Statistical clustering can be used to form sampling strata when longitudinal measures are of primary interest. Many clustering algorithms are available; the choice of the clustering algorithm can strongly impact the resulting strata because various algorithms focus on different aspects of the observed data.

Reducing disparities requires multiple strategies.

Disparities in care are a complex issue requiring multiple strategies to solve, including approaches to improve the measurement of quality and reporting stratified performance estimates.