Strategies and Factors Associated With Top Performance in Primary Care for Diabetes

Insights From a Mixed Methods Study

Leif I. Solberg, MD; Kevin A. Peterson, MD, MPH; Helen Fu, RN, PhD; Milton Eder, PhD; Rachel Jacobsen, MPH, RD; Caroline S. Carlin, PhD

Disclosures

Ann Fam Med. 2021;19(2):110-116. 

In This Article

Methods

The Understanding Infrastructure Transformation Effects on Diabetes study recruited 451 (77%) of the 586 primary care practices that submitted standardized data to Minnesota Community Measurement (MNCM), a nonprofit organization that has collected and publicly reported standardized medical performance measures since 2006. State law has required clinics to submit diabetes performance data since 2011.[6] Quantitative analyses of this data set have been published.[7,8]

Quantitative Data

Study participation depended on having a clinic leader complete a 112-question survey in 2017 regarding the presence of various care management processes to support high-quality care for patients with chronic medical conditions. The survey was created and tested for reliability by the National Committee for Quality Assurance and has been widely used.[9–11] We received completed surveys from 416 clinics (92% of participating clinics). Respondents reported whether each care management process was present, and the score was based on the percentage of eligible processes present. We identified questions that specifically addressed care management processes for diabetes so we could analyze an overall score and a subscore specific to diabetes. Example questions included the following:

  • Does your clinic use checklists of tests or interventions that are needed for prevention or monitoring of chronic illness?

  • Does your clinic have nonphysician staff who are specially trained and designated to educate patients in managing their chronic illness?

  • Does your clinic provide guideline-based reminders for services a diabetes patient should receive that appear when seeing the patient?

  • Does your clinic provide or refer patients to formal support programs to assist in self-management for chronic conditions?

Respondents were also asked to rate their clinic's priority for improving diabetes care in the next year on a scale of 0 (not a priority) to 10 (highest priority). Another key quantitative data set for the present study was obtained from MNCM for performance data collected in 2016 and reported in 2017. For diabetes, these measures included the percentage of patients with diabetes at each clinic with controlled diabetes (glycated hemoglobin <8%), controlled hypertension (blood pressure <140/90 mm Hg), controlled hyperlipidemia (low-density lipoprotein cholesterol <100 mg/dL or taking a statin), taking aspirin unless contraindicated, and who were nonsmokers. We used an MNCM composite (all or none) measure counting each patient with diabetes controlled or not controlled depending on whether they met all 5 measures.

The MNCM data did not include information on the race/ethnicity, wealth, income, or educational status of each clinic's patients. Therefore, we matched patient addresses to 2015 five-year average American Community Survey data to identify patient characteristics by zip code.[12] Each characteristic was averaged across patients within a clinic, and clinics were ranked by quartile for each population factor. Clinics were also identified as being located in the Twin Cities metropolitan area or in other parts of the state.

Clinic Selection

To identify clinics for interviews, we matched pairs of clinics with high and low scores on the MNCM composite performance measure but similar average propensity scores, as calculated by a patient-level regression that predicted achievement of the composite measure adjusted for age, sex, presence of vascular disease or depression, presence of type 1 diabetes, insurance type, and average wealth, income/education, and race/ethnicity in the patient's zip code, allowing us to identify clinics performing better or worse than might have been predicted by the characteristics of their patient population. Of 9 metropolitan and 19 nonmetropolitan pairs of clinics with similar expected performance on the diabetes measure but actual performance at least 2 quartiles apart, we selected 10 pairs of clinics with balanced metropolitan/nonmetropolitan locations and from varied types of medical groups.

We invited leaders of clinics that completed the earlier survey on care management processes to participate in a 20- to 30-minute interview regarding their approach to diabetes care. Three clinics (2 identified as performing lower than expected) declined participation, and we were unable to identify equivalent replacements, resulting in interviews at 17 clinics (Table 1), 9 from the metropolitan area and 8 from nonmetropolitan areas. Typical of Minnesota primary care clinics, 15 were part of 8 large medical groups (11–62 clinics), 1 was a solo clinic, and 1 was in a 3-clinic organization. Of the 31 leaders interviewed, there were 5 physicians, 3 nurse practitioners, 10 clinic managers, and 13 clinical supervisors. Clinicians and nonclinicians were usually both present.

Qualitative Data Collection

Interviews were performed by 1 of 3 experienced coauthors (L.S. and K.P. physicians; M.E. anthropologist) following a semistructured interview guide with open-ended questions and probes regarding the following:

  • The factors that contributed most to their MNCM diabetes scores

  • The strategies that were most successful in improving care for patients with diabetes

  • The most important barriers and facilitators to improving diabetes care

  • The most important help that they received or would like to receive from their organization

  • Their awareness of comparative performance scores from MNCM

After obtaining consent, we conducted 20- to 30-minute interviews at the clinic site with all participants simultaneously, with recording to assure accuracy and completeness. All interview data were transcribed verbatim by a professional service that removed personnel identifiers and replaced clinic names with codes. We used NVivo version 12 (QSR International) for data management.

Analysis

The analysis team consisted of 5 coauthors, each of whom reviewed the interviews independently and then met 2 to 4 times per month over a period of 8 months to achieve consensus. We used grounded theory, with the initial coding framework constructed as a directed content analysis from the domains and questions used in the interviews—strategies, facilitators, and barriers.[13] The framework and individual codes were then modified by consensus as we went through the interviews together, using a combination of individual reviews followed by group discussion to clarify and standardize codes across all interviews. This was followed by agreeing on observations and then conducting a summative analysis for themes, combining frequency counts with observations in a constant comparative approach.[14] These steps were undertaken for high- and low-performance groups and comparing them. We used a similar approach for the middle-quartile clinics to test the themes developed from the original high/low comparison. Differences of opinion were discussed until consensus was achieved. A detailed codebook with definitions, a quantitative summary of clinic comments for each code, and an audit trail ensured rigorous analysis, but we did not check results with interviewees.[15] The study was approved by the University of Minnesota Institutional Review Board.

processing....