Trends in Quality During Medical Home Transformation

Setting

HPMG is a multispecialty group practice within an integrated health system that includes a health plan, several hospitals, and a wide range of other health care services. It provides care to 400,000 active patients in the metropolitan area of Minneapolis and St Paul, most of whom receive their primary care in the 21 clinics studied here. Importantly for generalizability, only 60% of HPMG patients have HealthPartners insurance; the other 40% either are covered by a variety of private or public insurance products or are uninsured (5%). The demographic characteristics of the entire patient population are very similar to those of the population in the metropolitan area. HPMG has had a paperless electronic health record (EHR) system in place in both primary and specialty care settings for 6 years.

Primary care in HPMG is provided by a mixture of pediatric, internal medicine, and family physicians, as well as a smaller number of nurse practitioners and physician assistants. Physicians are paid largely based on productivity, while the medical group is reimbursed on a fee-for-service basis from most payers (including the HealthPartners health plan). The medical group has been working on improving quality, costs, and patient experience since at least 2000, with a particular focus on primary care redesign and on use of information technology to support a variety of innovative communication methods with patients and population health improvement strategies. Except for implementing advanced access scheduling in 2000 and the EHR in 2002–2004, there have been no abrupt changes in approach to care worth noting, and performance measures during that time showed no dramatic change.

Clinic-Level Measures

Medical Group–Level Measures

Covariates

We used linkages to clinic administrative and clinical data to ascertain covariates for use in adjusting estimates of trends over time for each clinic. Picker patient satisfaction data for the clinics were linked to the following year-specific, clinic-level variables related to patient case mix: number of active patients, proportion of active patients covered by Medicaid, and proportions that were female, white, and aged 65 years or older. We also linked the EHR-based optimal diabetes quality measure for each clinic to year-specific, patient-level variables—sex, Medicaid status, age, race, English as primary language, and number of medications prescribed—and to a year-specific clinic-level variable—the number of active patients at the clinic.

Similarly, at the medical group level, demographic information was available for the health plan members who were attributed to each contracted medical group. We linked Consumer Choice data to the following year-specific, medical group–level variables related to patient case mix: number of patient visits per week, proportion covered by Medicaid, proportion female, mean age of patients, mean number of primary care visits per year, and mean number of medications prescribed. The clinical indicator quality measures were linked to administrative year-specific, patient-level variables of sex, Medicaid status, age, number of primary care visits per year, and number of medications prescribed for the medical group–level analysis.

Analysis

We summarized each patient satisfaction and quality measure at the clinic or medical group level, and then computed means, standard deviations, and ranges to describe central tendency and variation. For Picker items and the EHR-based optimal diabetes care measure for the clinic-level analysis, we used generalized linear mixed models with a logit link to predict each binary measure from the available patient covariates, clinic covariates, or both; a linear term for time; and a random term to account for patient nesting in clinics. We translated the time coefficient from these models into a model-predicted adjusted yearly change in each measure. For Consumer Choice items and the clinical indicator quality measures for the medical group–level analysis, we used generalized linear mixed models (identity link for the generic drug use measure, logit link for all others) to predict each measure from the available patient covariates, medical group covariates, or both; a linear term for time; an interaction between medical group (HPMG vs not) and time; and a random term to account for the nested data. Coefficients from the interaction model were translated into HPMG and non-HPMG model-predicted adjusted yearly changes for each measure. We used linear time trends based on visual inspection of trends, selecting them because of the limited number of time points for some data sources for estimating higher-order trends.

We examined the magnitude and direction of the covariance between the random slope and intercept terms to assess whether clinics or medical groups that started lower on a measure showed greater or lesser improvement over time on the measure.

Clinic and Medical Group Characteristics

Of the 21 clinics in the HPMG group, 6 were located in center city areas of the Twin Cities and 15 were suburban (Table 1). They had an average of 9.3 primary care physicians plus 2 nurse practitioners or physician assistants. About two-thirds of their patients had commercial insurance, 12% had Medicare, and the rest had various state program coverage or were self-pay. Although patient demographics varied by clinic, on average, 69% of patients were white and 96% were English speaking.

Of the 34 contracted non-HPMG medical groups serving as the comparison in medical group analyses, about one-half were also in the Minneapolis/St Paul metropolitan area, while the other one-half were in Minnesota but outside that area, and 2 were in Wisconsin and North Dakota (data not shown). Only 5% of the comparison medical groups were single-site groups, while 40% had 2 to 5 clinics and 27% had more than 15 clinics. Forty percent were primary care groups and 60% were multispecialty with considerable diversity in the proportions of their patients that had various insurance arrangements.

The few patient-level descriptors available for the analysis comparing the HPMG vs non-HPMG medical group suggested similar profiles, with 58% female patients in the HPMG group and 57% across the non-HPMG medical groups, a median age of 42 vs 43 years, a median of 2 primary care visits per year, and a median of 11 prescription medications in 2009.

HPMG Clinic Trends Over Time

Responses on the Picker satisfaction survey are shown in Table 2 along with unadjusted mean rates and clinic score ranges in 2006 and 2009. Although the changes were not large for measures that were already at fairly high levels, rates for all of the items rose by 0.6 to 4.3 percentage points from 2006 to 2009, primarily as a result of increases for the clinics with the lowest rates. The yearly increase in satisfaction after adjustment for clinic-level patient case mix in each year was positive for all 7 items but significant for only 3 of them. A survey used by HPMG before use of the Picker survey showed no change from 2003 to 2005 in “able to get appointment,” but improvement in “treated with respect” and “timely test results” (data not shown).

Trends for the quality measure of optimal diabetes care assessed from the EHR showed an increasing rate over time, with 11.2% of patients meeting the measure in 2005 (clinic range, 4.8%–19.9%), and 22.3% meeting the measure in 2009 (clinic range, 15.8%–31.1%) (data not shown). After adjusting for patient-level characteristics and clinic size in each year, the model-predicted increase in the rate of optimal diabetes care was 2.3% per year (P <.001).

Comparison of HPMG Trends with Secular Trends in Minnesota

Table 3 shows Consumer Choice satisfaction rates in 2005 and 2009, comparing HPMG scores with the average for the 19 non-HPMG medical groups having at least 153 member responses per year. (More-detailed data for the latter are given in Supplemental Table 1, available online at http://www.annfammed.org/content/9/6/515/suppl/DC1.) The percentage of patients choosing the top rating for satisfaction ranged from 30% to 47% for HPMG patients and from 28% to 64% for non-HPMG patients across the 7 items in 2005. Nearly all satisfaction items improved over time for both groups, but the magnitude of the HPMG improvements was comparatively greater. When adjusting for differences in the patient case mix at the level of the medical group in each year, the yearly increase in satisfaction rates was significantly higher for HPMG than for non-HPMG medical groups for 4 of 7 items that address access and general satisfaction. The pattern of increases for these 4 items indicated that HPMG started with lower rates in 2005 and then caught up to the non-HPMG rates by 2009. For the HPMG clinics, a comparison with the survey results in 2000 indicated no change from 2000 to 2005 in the items “satisfied with clinic” (37.8% in 2000), “recommend clinic” (48.1%), and “advice on staying healthy” (35.0%), and slow improvement in “convenient appointment” (27.3% in 2000) and “doctor of your choice” (28.4%) (data not shown).

Table 4 shows quality measures from clinical indicators, allowing comparison of trends over time between HPMG and the 20 to 35 non-HPMG medical groups that had data for these measures over the varying time intervals for which each measure had adequate data. (More-detailed data for the latter are given in Supplemental Table 2, available online at http://www.annfammed.org/content/9/6/515/suppl/DC1.) Changes from the baseline year to the last year (both of which varied depending on the measure) were all positive for both HPMG and non-HPMG groups. After adjusting for patient-level variables and medical group size in each year, the yearly change for each measure was positive for all 4 measures. The magnitude of yearly change suggested that HPMG exceeded the secular trend for the optimal diabetes care, optimal CAD care, and adult preventive services measures, but these difference in trends between groups were not significant. For generic drug use, the non-HPMG yearly increase was significantly greater than the HPMG yearly increase; non-HPMG groups started lower than HPMG groups and then nearly caught up. The very large sample size for the generic drug use analysis yielded a model with all highly significant terms, so the clinical significance of the difference in improvement for HPMG and non-HPMG groups needs to be assessed in addition to the statistical significance.