Haili Sun

Implementation of the Patient-Centered Medical Home in the Veterans Health Administration: Associations With Patient Satisfaction, Quality of Care, Staff Burnout, and Hospital and Emergency Department Use

IMPORTANCE In 2010, the Veterans Health Administration (VHA) began implementing the patient-centered medical home (PCMH) model. The Patient Aligned Care Team (PACT) initiative aims to improve health outcomes through team-based care, improved access, and care management. To track progress and evaluate outcomes at all VHA primary care clinics, we developed and validated a method to assess PCMH implementation. OBJECTIVES To create an index that measures the extent of PCMH implementation, describe variation in implementation, and examine the association between the implementation index and key outcomes. DESIGN, SETTING, AND PARTICIPANTS We conducted an observational study using data on more than 5.6 million veterans who received care at 913 VHA hospital-based and community-based primary care clinics and 5404 primary care staff from (1) VHA clinical and administrative databases, (2) a national patient survey administered to a weighted random sample of veterans who received outpatient care from June 1 to December 31, 2012, and (3) a survey of all VHA primary care staff in June 2012. Composite scores were constructed for 8 core domains of PACT: access, continuity, care coordination, comprehensiveness, self-management support, patient-centered care and communication, shared decision making, and team-based care. MAIN OUTCOMES AND MEASURES Patient satisfaction, rates of hospitalization and emergency department use, quality of care, and staff burnout. RESULTS Fifty-three items were included in the PACT Implementation Progress Index (Pi2). Compared with the 87 clinics in the lowest decile of the Pi2, the 77 sites in the top decile exhibited significantly higher patient satisfaction (9.33 vs 7.53; P < .001), higher performance on 41 of 48 measures of clinical quality, lower staff burnout (Maslach Burnout Inventory emotional exhaustion subscale, 2.29 vs 2.80; P = .02), lower hospitalization rates for ambulatory care-sensitive conditions (4.42 vs 3.68 quarterly admissions for veterans 65 years or older per 1000 patients; P < .001), and lower emergency department use (188 vs 245 visits per 1000 patients; P < .001). CONCLUSIONS AND RELEVANCE The extent of PCMH implementation, as measured by the Pi2, was highly associated with important outcomes for both patients and providers. This measure will be used to track the effectiveness of implementing PACT over time and to elucidate the correlates of desired health outcomes.

Alcohol Screening Scores and Medication Nonadherence

Context Is alcohol misuse associated with medication nonadherence? Contribution This study of primary care patients attending 7 Veterans Affairs clinics found a graded, linear decrease in adherence to statins and hypertension medications with increasing levels of alcohol misuse. Caution Alcohol misuse was measured with a brief screening questionnaire that was mailed to patients. Adherence was measured by pharmacy refills. Implication Alcohol misuse may be associated with increased risk for medication nonadherence. The Editors Daily medications are the cornerstone of chronic disease management. Medications to treat hypertension, hyperlipidemia, and diabetespotent risk factors for cardiovascular diseaseare common and are often prescribed for asymptomatic patients to prevent future disease. However, nonadherence to medications is common (1) and is associated with poor outcomes, increased health care costs (2, 3), and death (4). Many studies have examined patient characteristics associated with nonadherence, but most identified risk factors for nonadherence are not modifiable. Alcohol misuse is common, has been associated with medication nonadherence, and is modifiable (57). However, research on alcohol misuse and medication adherence has been largely limited to patients with HIV (811) and a few studies of diabetes (3, 12, 13). One recent study found both a temporal and a doseresponse relationship between alcohol consumption and medication adherence (8) but used a lengthy interview measure of alcohol use that is not practical for busy clinical settings. Therefore, it remains unclear whether brief validated alcohol screening questionnaires used in clinical practice could identify patients at risk for nonadherence due to alcohol misuse. We examined whether primary care outpatient scores on a brief, scaled, alcohol screening questionnairethe Alcohol Use Disorder Identification TestConsumption (AUDIT-C)were associated with medication nonadherence. Specifically, we evaluated the association between increasing scores on the AUDIT-C (score range, 0 to 12) and adherence to oral medications commonly used for hypertension, hyperlipidemia, and diabetes. We hypothesized that higher AUDIT-C scores would be associated with an increased risk for medication nonadherence. Methods Participants and Setting We used data collected from the Ambulatory Care Quality Improvement Project (ACQUIP) cohort in this study (14). In brief, ACQUIP enrolled 36821 active patients from the general internal medicine clinics of 7 Veterans Affairs (VA) medical centers nationwide, including facilities in Seattle, Washington; West Los Angeles, California; Birmingham, Alabama; Little Rock, Arkansas; San Francisco, California; Richmond, Virginia; and White River Junction, Vermont. The ACQUIP initially surveyed all VA sites and selected these 7 sites (from 60 respondents) on the basis of geographic diversity; well-established systems for assigning patients to firms; and an experienced, interested investigator to lead the study. The ACQUIP was a randomized trial testing the effect of an audit and feedback quality-improvement intervention; there was no detectable effect of the intervention on primary outcomes, including alcohol misuse (14). Patients were eligible for ACQUIP if they had at least 1 visit to a primary care facility in the past year and had a primary care provider. The ACQUIP sent questionnaires (ACQUIP Health Checklist) at enrollment (1997 to 2000), and the institutional review board considered participant response to the survey to be consent for study participation. The survey assessed demographic characteristics, alcohol misuse, other health behaviors, and psychiatric and medical conditions. Patients who did not respond were mailed up to 3 additional surveys. The date the survey was received by the study team was considered the index date for all participants. Survey data were linked to electronic records, including pharmacy, diagnosis, and death records. Participants who died during follow-up were excluded. The institutional review board at each participating VA site approved ACQUIP, and the University of Washington Division of Human Subjects approved the secondary analyses that we present in this article. Pharmacy Data and Medication Cohorts Pharmacy data were retrieved electronically as part of the ACQUIP protocol from December 1995 to May 2000. Each prescription filled generated 1 record containing the drug name, the quantity and date dispensed, and the number of days supplied. These data are nearly identical to national VA pharmacy data (15), which have been used in several studies of medication adherence and pharmacoepidemiology (16, 17). We identified 3 nonexclusive cohorts of patients with increasing medication regimen complexity: a statin cohort, consisting of all patients prescribed a statin medication for hypercholesterolemia; an oral hypoglycemic cohort, with all patients who were prescribed either a sulfonylurea or metformin for blood glucose control; and a hypertension treatment cohort, consisting of all patients with self-reported hypertension who were prescribed at least 1 of 6 classes of antihypertensive drugs (-blockers, angiotensin-converting enzyme inhibitors, -blockers, calcium-channel blockers, thiazide-type diuretics, or nonthiazide diuretics) and a group consisting of other antihypertension medications usually used as fourth- or fifth-line agents (such as hydralazine). We considered patients medication users and included them in 1 of the cohorts if they received both 1 or more fills of the drug class within 2 years before the index date and 1 or more fills in the year after the index date. We used these criteria to minimize potential dropout bias by ensuring that patients were still engaged in care and obtaining medications from the VA. We excluded glitazones and angiotensin-receptor blockers from analyses because few patients were prescribed these medications, which were on a restricted formulary at the time of the study. In addition, we excluded patients in the oral hypoglycemic cohort if they had an active prescription for insulin other than neutral protamine Hagedorn, in order to remove patients who transitioned from oral medication to insulin during the study. Alcohol Misuse and AUDIT-C We assessed alcohol misuse with the AUDIT-C from the ACQUIP Health Checklist. The AUDIT-C assesses frequency and typical quantity of drinking during the past year, as well as the frequency of heavy episodic drinking (6 drinks per occasion) by using 3 questions (18). Each of the 3 questions is scored 0 to 4, for a total combined score of 0 to 12. The AUDIT-C is reliable (19) and has been validated as a screening test for the spectrum of alcohol misuse, including risky drinking and alcohol-use disorders on the basis of Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, criteria (18, 20, 21). A score of 4 or more is considered positive for alcohol misuse in male VA patients, but the AUDIT-C score has also been shown to be a scaled measure of risk for alcohol-related symptoms (22) and medical complications often associated with alcohol misuse (2326). To provide adequate precision in estimates and allow comparison with previous analyses (23, 24), we grouped AUDIT-C scores into 5 categories: nondrinkers (score, 0); low-level alcohol use (score range, 1 to 3); and mild (score range, 4 to 5), moderate (score range, 6 to 7), and severe (score range, 8 to 12) alcohol misuse. Medication Adherence We created an individual measure of refill adherence, which was previously validated within the VA and ACQUIP, for each patient and medication class. This measure is similar to a medicationpossession ratio, and it accounts for overstocking and medication gaps, correlates better with physiologic outcomes when compared with previous measures, and is described in detail elsewhere (27). From this measure, we derived a proportion of days covered that reflected the number of days during the observation period that medication was available (17). We considered all medications within a medication type (statin, oral hypoglycemics, and antihypertensive medications) to be equivalent for purposes of adherence. We calculated adherence separately for 2 different periods: 90 days and 1 year starting from the index date. We assessed at 1 year because it is a traditional measurement of adherence (16, 17). We also assessed at 90 days because refill adherence for this period has been correlated with outcomes (27). On the basis of previous medication adherence literature (16, 17), we considered patients in all medication cohorts to be adherent if they had medication available for at least 80% of the observation period. In other words, for the 90-day observation period, nonadherent patients would not have medication available for at least 18 days; for the 1 year-period, they would be without medication for at least 73 days. When more than 1 medication was used (for example, for diabetes or hypertension), the proportions of days covered were averaged, and we considered patients to be adherent if they had at least 80% of the drug regimen for diabetes or hypertension available for the observation period. A person who met the definition of a user for 2 drug classes but only maintained complete fills of 1 drug with no fills of the other drug therefore would have an average adherence of 0.5 and would be considered nonadherent to the overall regimen. Covariates Race was based on a combination of self-report from the ACQUIP Health Checklist and the electronic record. We determined sex, education, and marital status from the ACQUIP Health Checklist. We calculated a drug count from the number of oral drugs that patients obtained during the year before the index date to adjust for total medication regimen complexity. We classified smoking status as current, former, or never. We assessed depression with the Mental Health Inventory (score range, 5 to 30); scores gre

Predicting risk of hospitalization or death among patients receiving primary care in the Veterans Health Administration.

Background:Statistical models that identify patients at elevated risk of death or hospitalization have focused on population subsets, such as those with a specific clinical condition or hospitalized patients. Most models have limitations for clinical use. Our objective was to develop models that identified high-risk primary care patients. Methods:Using the Primary Care Management Module in the Veterans Health Administration (VHA)’s Corporate Data Warehouse, we identified all patients who were enrolled and assigned to a VHA primary care provider on October 1, 2010. The outcome variable was the occurrence of hospitalization or death during the subsequent 90 days and 1 year. We extracted predictors from 6 categories: sociodemographics, medical conditions, vital signs, prior year use of health services, medications, and laboratory tests and then constructed multinomial logistic regression models to predict outcomes for over 4.6 million patients. Results:In the predicted 95th risk percentiles, observed 90-day event rates were 19.6%, 6.2%, and 22.6%, respectively, for hospitalization, death, and either hospitalization or death, compared with population averages of 2.7%, 0.7%, and 3.4%, respectively; 1-year event rates were 42.3%, 19.4%, and 51.3%, respectively, compared with population averages of 8.2%, 2.6%, and 10.8%, respectively. The C-statistics for 90-day outcomes were 0.83, 0.86, and 0.81, respectively, for hospitalization, death, and either hospitalization or death and were 0.81, 0.85, and 0.79, respectively, for 1-year outcomes. Conclusions:Prediction models using electronic clinical data accurately identified patients with elevated risk for hospitalization or death. This information can enhance the coordination of care for patients with complex clinical conditions.

Prevalence, comorbidity, and prognosis of mental health among US veterans

OBJECTIVES We evaluated the association of mental illnesses with clinical outcomes among US veterans and evaluated the effects of Primary Care-Mental Health Integration (PCMHI). METHODS A total of 4 461 208 veterans were seen in the Veterans Health Administrations patient-centered medical homes called Patient Aligned Care Teams (PACT) in 2010 and 2011, of whom 1 147 022 had at least 1 diagnosis of depression, posttraumatic stress disorder (PTSD), substance use disorder (SUD), anxiety disorder, or serious mental illness (SMI; i.e., schizophrenia or bipolar disorder). We estimated 1-year risk of emergency department (ED) visits, hospitalizations, and mortality by mental illness category and by PCMHI involvement. RESULTS A quarter of all PACT patients reported 1 or more mental illnesses. Depression, SMI, and SUD were associated with increased risk of hospitalization or death. PTSD was associated with lower odds of ED visits and mortality. Having 1 or more contact with PCMHI was associated with better outcomes. CONCLUSIONS Mental illnesses are associated with poor outcomes, but integrating mental health treatment in primary care may be associated with lower risk of those outcomes.

Alcohol screening scores predict risk of subsequent fractures.

The Alcohol Use Disorders Identification Test-Consumption (AUDIT-C; 0–12 points) was included on health surveys in a cohort of 32,622 general medicine outpatients from seven US Department of Veterans Affairs (VA) hospitals. Cox proportional hazards models were used to estimate the risk of fracture (mean follow-up = 1.6 years) by AUDIT-C category. After adjusting for confounders, AUDIT-C scores of 8–9 and 10–12 were associated with significantly increased risks for subsequent fractures, HR (95% CI) = 1.37 (1.03 to 1.83) and 1.79 (1.38 to 2.33) respectively. These results can be used to provide feedback to patients linking their alcohol screening scores to medical outcomes—a critical component of evidence-based brief counseling for alcohol misuse. The studys limitations are noted.

AUDIT-C Alcohol Screening Results and Postoperative Inpatient Health Care Use.

BACKGROUND Alcohol screening scores ≥5 on the Alcohol Use Disorders Identification Test-Consumption (AUDIT-C) up to a year before surgery have been associated with postoperative complications, but the association with postoperative health care use is unknown. This study evaluated whether AUDIT-C scores in the year before surgery were associated with postoperative hospital length of stay, total ICU days, return to the operating room, and hospital readmission. STUDY DESIGN This cohort study included male Veterans Affairs patients who completed the AUDIT-C on mailed surveys (October 2003 through September 2006) and were hospitalized for nonemergent noncardiac major operations in the following year. Postoperative health care use was evaluated across 4 AUDIT-C risk groups (scores 0, 1 to 4, 5 to 8, and 9 to 12) using linear or logistic regression models adjusted for sociodemographics, smoking status, surgical category, relative value unit, and time from AUDIT-C to surgery. Patients with AUDIT-C scores indicating low-risk drinking (scores 1 to 4) were the referent group. RESULTS Adjusted analyses revealed that among eligible surgical patients (n = 5,171), those with the highest AUDIT-C scores (ie, 9 to 12) had longer postoperative hospital length of stay (5.8 [95% CI, 5.0-6.7] vs 5.0 [95% CI, 4.7-5.3] days), more ICU days (4.5 [95% CI, 3.2-5.8] vs 2.8 [95% CI, 2.6-3.1] days), and increased probability of return to the operating room (10% [95% CI, 6-13%] vs 5% [95% CI, 4-6%]) in the 30 days after surgery, but not increased hospital readmission within 30 days postdischarge, relative to the low-risk group. CONCLUSIONS AUDIT-C screening results could be used to identify patients at risk for increased postoperative health care use who might benefit from preoperative alcohol interventions.

Frequency of serum low-density lipoprotein cholesterol measurement and frequency of results ≤100 mg/dl among patients who had coronary events (Northwest VA Network Study) ∗

This population-based, cross-sectional analysis targeted all veterans with coronary heart disease (CHD) who were active patients in primary care or cardiology clinics in the Veterans Health Administration Northwest Network from July 1998 to June 1999. We report guideline compliance rates, including whether low-density lipoprotein (LDL) was measured, and if measured, whether the LDL was < or=100 mg/dl. In addition, we utilized multivariate logistic regression to determine patient characteristics associated with LDL measurements and levels. Of 13,891 active patients with CHD, 5,552 (40.0%) did not have a current LDL measurement. Of those with LDL measurements, 39.1% were at the LDL goal of < or =100 mg/dl, whereas 26.5% had LDL > or =130 mg/dl. Male gender, younger age, history of angioplasty or coronary artery bypass grafting, current hypertension, diabetes mellitus, and angina pectoris were associated with increased likelihood of LDL measurement. Older age and current diabetes and angina were associated with increased likelihood of LDL being < or =100 mg/dl, if measured. Although these rates of guideline adherence in the CHD population compare well to previously published results, they continue to be unacceptably low for optimal clinical outcomes. Attention to both LDL measurement and treatment (if elevated) is warranted.

Association between Alcohol Screening Results and Hospitalizations for Trauma in Veterans Affairs Outpatients

Background: Alcohol consumption is a risk factor for traumatic injury, but it is unknown whether responses to alcohol screening questionnaires administered routinely in primary care are associated with subsequent hospitalization for traumatic injury. Objective: We evaluated the association between alcohol screening scores and the risk for subsequent hospitalizations for trauma among Veterans Affairs (VA) general medicine patients. Method: This study included VA outpatients (n = 32,623) at seven sites who returned mailed surveys (1997–1999). Alcohol Use Disorders Identification Test Consumption (AUDIT-C) scores grouped patients into six drinking categories representing nondrinkers, screen-negative drinkers, and drinkers who screened positive for mild, moderate, severe, and very severe alcohol misuse (scores 0, 1–3, 4–5, 6–7, 8–9, 10–12, respectively). VA administrative and Medicare data identified primary discharge diagnoses for trauma. Cox proportional hazard models were used to estimate the risk of trauma-related hospitalization for each drinking group adjusted for demographics, smoking, and comorbidity. Results: Compared with screen-negative drinkers, patients with severe and very severe alcohol misuse (AUDIT-C 8–9 and ≥10) were at significantly increased risk for trauma-related hospitalization over the follow-up period (adjusted hazard ratios AUDIT-C: 8–9 2.06, 95% confidence interval (CI) 1.31– 3.24 and AUDIT-C≥10 2.13, 95% CI 1.32–3.42). Conclusions: Patients with severe and very severe alcohol misuse had a twofold increased risk of hospital admission for trauma compared to drinkers without alcohol misuse. Scientific Significance: Alcohol screening scores could be used to provide feedback to patients regarding risk of trauma-related hospitalization. Findings could be used by providers during brief alcohol-related interventions with patients with alcohol misuse.

Predicting Risk of Hospitalization or Death Among Patients With Heart Failure in the Veterans Health Administration

Patients with heart failure (HF) are at high risk of hospitalization or death. The objective of this study was to develop prediction models to identify patients with HF at highest risk for hospitalization or death. Using clinical and administrative databases, we identified 198,460 patients who received care from the Veterans Health Administration and had ≥1 primary or secondary diagnosis of HF that occurred within 1 year before June 1, 2009. We then tracked their outcomes of hospitalization and death during the subsequent 30 days and 1 year. Predictor variables chosen from 6 clinically relevant categories of sociodemographics, medical conditions, vital signs, use of health services, laboratory tests, and medications were used in multinomial regression models to predict outcomes of hospitalization and death. In patients who were in the ≥95th predicted risk percentile, observed event rates of hospitalization or death within 30 days and 1 year were 27% and 80% respectively, compared to population averages of 5% and 31%, respectively. The c-statistics for the 30-day outcomes were 0.82, 0.80, and 0.80 for hospitalization, death, and hospitalization or death, respectively, and 0.82, 0.76, and 0.77, respectively, for 1-year outcomes. In conclusion, prediction models using electronic health records can accurately identify patients who are at highest risk for hospitalization or death. This information can be used to assist care managers in selecting patients for interventions to decrease their risk of hospitalization or death.

Use of Administrative Claims Models to Assess 30-Day Mortality among Veterans Health Administration Hospitals

Background:The Centers for Medicare and Medicaid Services (CMS) publicly reports hospital-specific risk-standardized, 30-day, all-cause, mortality rates (RSMRs) for all hospitalizations among fee-for-service Medicare beneficiaries for acute myocardial infarction (AMI), heart failure (HF), and pneumonia at non-Federal hospitals. Objective:To examine the performance of the statistical models used by CMS among veterans at least 65 years of age hospitalized for AMI, HF, and pneumonia in Veterans Health Administration (VHA) hospitals. Research Design:Cross-sectional analysis of VHA administrative claims data between October 1, 2006 and September 30, 2009. Subjects:Thirteen thousand forty-six veterans hospitalized for AMI among 123 VHA hospitals; 26,379 veterans hospitalized for HF among 124 VHA hospitals; and 31,126 veterans hospitalized for pneumonia among 124 VHA hospitals. Measures:Hospital-specific RSMR for AMI, HF, and pneumonia hospitalizations calculated using hierarchical generalized linear models. Results:Median number of AMI hospitalizations per VHA hospital was 87. Average AMI RSMR was 14.3% [95% confidence interval (CI), 13.9%–14.6%] with modest heterogeneity among VHA hospitals (RSMR range: 8.4%–20.3%). The c-statistic for the AMI RSMR statistical model was 0.79. Median number of HF hospitalizations was 188. Average HF RSMR was 10.1% (95% CI, 9.9%–10.4%) with modest heterogeneity (RSMR range: 6.1%–14.9%). The c-statistic for the HF RSMR statistical model was 0.73. Median number of pneumonia hospitalizations was 221.5. Average pneumonia RSMR was 13.0% (95% CI, 12.7%–13.3%) with modest heterogeneity (RSMR range: 9.0%–18.4%). The c-statistic for the pneumonia RSMR statistical model was 0.72. Conclusions:The statistical models used by CMS to estimate RSMRs for AMI, HF, and pneumonia hospitalizations at non-Federal hospitals demonstrate similar discrimination when applied to VHA hospitals.