The Association between Quality of Care and the Intensity of Diabetes Disease Management Programs FREE

—The Editors

Persons with diabetes continue to receive suboptimal care (1 - 6). To improve quality, many health systems have implemented disease management programs for diabetes and other chronic conditions (7 - 9). These programs typically incorporate population-based strategies, such as disease registries, clinical guidelines, performance feedback, physician reminders, self-management support for patients, and targeted case management for high-risk patients (10).

Evidence for the effectiveness of disease management comes primarily from small efficacy trials (10 - 20). Such studies consistently found improved processes of diabetes care; however, improvements in outcomes (such as control of cardiovascular disease risk factors) were less consistent ((12), (17 - 18), (20 - 22)). Furthermore, most studies evaluated only 1 or 2 strategies (instead of multicomponent programs) in selected clinical settings. It is unclear how well findings from these smaller studies apply to entire patient populations.

Many components of disease management focus on improving processes of care. Early performance measurement projects, such as the Health Plan Employer Data Information System (23) and the Diabetes Quality Improvement Program (5), emphasized the importance of such processes as annual retinal screening or hemoglobin A_1c determination. Particularly for health plans, process measures are more readily available than are outcomes data. However, if disease management is to improve patient outcomes, it must also improve intermediate outcomes, such as hemoglobin A_1c levels, systolic blood pressure, and serum low-density lipoprotein (LDL) cholesterol levels.

Translating Research into Action for Diabetes (TRIAD) (24) is a multicenter study of diabetes care in managed care. The TRIAD study's central hypothesis is that health care systems features can affect quality of care. Here, we examine how 3 disease management strategies vary in intensity across physician groups and whether physician groups with more intensive disease management have higher quality of diabetes care. We assess quality by processes of care, by levels of intermediate outcomes, and by current clinical management of these outcomes.

The TRIAD study's sampling frame, methods, key hypotheses, and power calculations are detailed elsewhere (24). The study comprised 6 collaborating translational research centers that were partnered with 10 managed care health plans in 7 states. Of the 10 plans, 7 contracted with 1 to 26 physician groups (total, 68 groups), whereas 4 plans directly contracted with individual physicians.

A standard algorithm was applied to automated pharmacy, laboratory utilization, and inpatient and outpatient diagnostic data (25) to identify all community-dwelling patients with diabetes who were 18 years of age and older and who had been continuously enrolled in the TRIAD health plan for at least 18 months. The study cohort was randomly sampled from this population.

Sampled patients were recruited between July 2000 and October 2001 by using computer-assisted telephone interviews or written surveys that were conducted in English or Spanish. Eligibility was confirmed if patients verified that they had had diabetes for at least 12 months and had received most of their diabetes care through the TRIAD health plan. Permission was sought from all respondents to request copies of their outpatient medical records for the previous 18 months.

All health plan and physician group directors received mailed surveys (Appendix Figure) during the same interval. Face-to-face or telephone interviews were used to complete and clarify responses. Each director was offered $100 for completing the survey.

The TRIAD study was reviewed and approved by the institutional review boards of each research center and by the Centers for Disease Control and Prevention (CDC). Informed consent was obtained from all survey respondents.

Patient surveys included questions on health status, diabetes duration, current diabetes treatment, and demographic characteristics. Of 13 086 contacted and eligible persons, 11 927 (91%) completed the survey (56.6% by computer-assisted telephone interview and 43.4% by written survey) (Figure). We were unable to contact many individuals. Using a practice that is endorsed by the Council of American Survey Research Organizations (26), we assumed that persons whom we could not contact or for whom we could not confirm eligibility had the same eligibility rate as those contacted. Under that assumption, the response rate was 69%.

Of 11 927 patients who completed a survey, 8661 (73%) consented to medical record review and subsequently had charts available for review. Centrally trained reviewers used standardized data collection software to abstract process measures, most recent levels of hemoglobin A_1c, upper limits of normal for hemoglobin A_1c measurements that were recorded, serum LDL cholesterol levels, systolic blood pressure, current medications, and comorbid conditions. Interrater reliability (κ) for the main quality measures ranged from 0.86 to 0.94.

All 10 health plan directors and 52 of 68 physician group directors completed surveys. Surveys assessed organizational age, size, structure, profit status, insurance products, contracting arrangements, history of involvement with managed care, and detailed information on the organization's use of diabetes disease management strategies. Of the physician groups that did not respond (443 participants), 11 existed solely for the purpose of contracting with plans and had no diabetes disease management. These physician groups were assumed to have no care management strategies and were included in the analyses, as were patient groups (1150 participants) from the 4 health plans that contracted directly with physicians. The remaining 5 groups (159 participants) did not respond and were excluded from analyses (Figure). Consequently, the resulting sample included a total of 8661 survey respondents with charts available for review and data from 63 physician groups and 4 additional health plans (Figure). Mean duration of diabetes, body mass index, and health status did not meaningfully differ between persons whose medical records and physician group variables were available to the study team and those whose records were unavailable.

The primary predictors were 3 measures of the intensity of disease management strategies: physician reminders, performance feedback, and structured care. These were calculated for physician groups and health plans from multiple survey items. A detailed description of the methods used to calculate composite intensity scores is provided in Appendix 2. Selected item-level responses for physician groups in the most intense versus least intense tercile of each strategy are displayed in Table 1.

The physician reminders intensity score was derived from 2 questions, which detailed the types and content of the reminders physicians received. Groups whose use of reminders represented the upper tercile of intensity were found to have reminded physicians about 4 care processes on average. Most groups in the upper tercile delivered reminders electronically at the point of care.

Performance feedback intensity was obtained by tallying responses to a checklist of possible diabetes process and outcome feedback items. A total of 86% and 82% of groups in the upper tercile included levels of hemoglobin A_1c and serum LDL cholesterol, respectively, in feedback to physicians (Table 1). Physician feedback focused on many of the same elements of care as reminders.

The use of formal case management, diabetes guidelines, patient reminders, and diabetes education correlated highly in physician groups (Pearson correlation coefficients ranged from 0.63 to 0.88); therefore, we could not look at these approaches independently. Consequently, we combined the 4 approaches into a single composite score for “structured diabetes care management.” Use of formal case management was assessed by the proportion of patients with diabetes who were enrolled, the number of case managers per 10 000 patients, the extent to which the program targeted high-risk patients, and a checklist of case management activities. The clinical guidelines were scored to reflect the extent of implementation. The highest score was assigned to physician groups that incorporated guidelines into automated physician or patient reminders. Patient reminder intensity incorporated the number, type, and frequency of reminders sent. On average, physician groups in the upper tercile had diabetes education as a covered benefit, whereas those in the lowest tercile generally did not have these programs.

Because of the differing numbers of questions and wide range of possible values within each intensity score, each question was z-transformed to a mean of 0.0 and standard deviation near or equal to 1.0 to facilitate comparison. Scores for each of the care management composites were computed as the mean of its z-transformed responses. A combined measure of disease management intensity was constructed for each plan and group by adding the 3 composite scores together. The distribution of z-transformed composite scores for the 67 groups is shown in Appendix Table 1.

Study end points included the following 8 processes of care: levels of hemoglobin A_1c, serum lipids, and urine albumin; dilated retinal and foot examinations; recommendations for aspirin therapy and influenza vaccine; and counseling for smokers to encourage them to quit (n = 1264). The prevalence of the first 3 indicators was obtained exclusively from chart review; we used evidence from either the medical record or the patient survey to determine which of the remaining 5 measures had been performed. Each quality measure was the percentage of patients who received the process during the 12 months preceding the patient survey.

The second set of end points included the most recently recorded hemoglobin A_1c, serum LDL cholesterol, and systolic blood pressure values in the previous 12 months. These intermediate outcomes were analyzed as continuous and binary variables (hemoglobin A_1c level <8.0%, systolic blood pressure <140 mm Hg, and serum LDL cholesterol level <3.35 mmol/L [<130 mg/dL]).

We defined a third set of end points to reflect medical management of the 3 intermediate outcomes (Table 2). For each intermediate outcome, physician groups were assessed by the proportion of their patients with the condition who were at or below target levels or above target levels and currently receiving more medications (reflecting greater effort to manage the outcome) (27). Number of medications was dichotomized as use of 2 or more oral agents or insulin for diabetes; 1 or more lipid-lowering agents for hypercholesterolemia; and 2 or more antihypertensive agents for hypertension.

We used the patient survey to obtain the following model covariates: age, sex, ethnicity, education, income, health status from the Short Form-12 mental and physical component scores (28), duration of diabetes, and a 4-level treatment variable (diet-controlled, oral agents alone, oral agents and insulin, or insulin alone). The following 2 additional covariates, which described comorbid conditions and current therapies, were obtained from the medical record: the Charlson index (29 - 30) and an additional measure of cardiovascular comorbidity that indicated presence of previous myocardial infarction, stroke, or coronary or carotid artery revascularization. All models were adjusted for intensity of care management and clustering at the health plan level.

Both multiple imputation and modeling were performed with SAS statistical software, version 9.1.3 (SAS Institute Inc., Cary, North Carolina). Initial analyses examined distributions for predictors, outcomes, and covariates across physician groups. Fewer than 1% of patients had missing data for any key process-of-care variables. Hemoglobin A_1c, serum LDL cholesterol, and systolic blood pressure values were unavailable for 9%, 24%, and 5% of patients, respectively; these patients were excluded only from analyses of the missing end point. Short Form-12 scores and income were missing for approximately 11% of participants; values for all other covariates were missing in fewer than 5% of participants. Missing values for covariates (but not predictors or end points) were imputed 5 times. Each covariate was predicted as a function of all other covariates by using a Markov chain Monte Carlo multiple imputation algorithm (SAS PROC MI and PROC MIANALYZE functions).

Separate models assessed associations between each of the 3 disease management intensity scores and the combined intensity score with each process indicator, intermediate outcome level, and medication use variable. Because of the clustering of participants within physician groups and plans, hierarchical mixed-effects models were used to account for intraplan and physician group correlation. We used the SAS GLIMMIX procedure with penalized quasi-likelihood estimation for dichotomous outcomes and the SAS PROC MIXED procedure with restricted maximum likelihood estimates for continuous outcomes.

Our results are presented as model-based predicted probabilities of receiving each process of care in physician groups in the highest versus lowest tercile of intensity for each of the 3 disease management strategies. We used terciles of the distribution of disease management intensity to estimate interpretable differences in performance across the intensity distribution. In models predicting intermediate outcome levels and proportion of patients with controlled disease or receiving more medications, we did not adjust for diabetes treatment and health status because these 2 variables could mediate associations of disease management intensity with these outcomes.

We serially removed each of the health plans from the models, and we removed 1150 participants who did not receive care from physician groups to examine whether observed differences in processes of care were unduly affected by data from a single plan or from assigning a value of 0 to management indicators for physician group care, respectively. In separate sensitivity analyses, we examined whether our findings would be altered if we controlled for all 3 care strategies simultaneously or if we used cut-points for intermediate outcome control that were more stringent (hemoglobin A_1c level <7.0%; serum LDL cholesterol level <2.60 mmol/L [<100 mg/dL], and systolic blood pressure <130 mm Hg) or less stringent (hemoglobin A_1c level, 9.5%; serum LDL cholesterol level, 4.15 mmol/L [160 mg/dL]; and systolic blood pressure, 160 mm Hg). We also adjusted all hemoglobin A_1c analyses by the upper limit of normal for the assay used. None of these sensitivity analyses substantially changed the results (data not shown).

The study was funded by the CDC and the National Institute of Diabetes and Digestive and Kidney Diseases. The study was designed and governed by a 7-member steering committee; the CDC has 1 representative on this committee. Analysts from the CDC provided us with statistical assistance, reviewed the manuscript, and provided feedback to us.

Sample characteristics across physician groups are presented in Table 3. Nearly every characteristic varied greatly across physician groups, particularly age, sex, ethnicity, and prevalence of cardiovascular disease. Intensity of disease management strategies also varied substantially across physician groups (Table 1).

Rates for diabetes process of care were high but varied across physician groups (Table 4). Median levels of the 3 intermediate outcomes indicated that more than 50% of patients were above American Diabetes Association guideline target levels for each intermediate outcome (31).

As measured by the 8 process indicators, intensity of disease management was strongly associated with quality (Table 5). Of 24 adjusted associations (3 disease management strategies by 8 processes of care), 17 associations significantly favored physician groups with increased disease management intensity. Adjusted differences in predicted probabilities between upper and lower terciles ranged from 5% to 15%. Greater intensity of structured care management was associated with significantly higher levels for 6 of 8 processes, with differences ranging from 7% to 11%. More intense physician performance feedback was also associated with higher rates for 6 of 8 process indicators, with significant differences of 5% to 10%. Increased intensity of physician reminders was associated with higher rates for 5 of 8 process measures, with differences ranging from 8% to 15%. The combined intensity measure was also significantly related to 6 of the 8 processes of care. No strategy was significantly associated with rates of providing advice to stop smoking (among smokers).

Greater intensity of structured care management was associated with adjusted systolic blood pressure levels that were 3 mm Hg (95% CI, 1 to 5 mm Hg) higher than those for lower intensity groups. The only other significant association was between greater intensity of physician reminders and slightly lower levels of serum LDL cholesterol (0.05 mmol/L [2 mg/dL]). Otherwise, disease management intensity and levels of the 3 intermediate outcomes were not associated (Table 5). Models predicting dichotomized outcomes (that is, below target levels) confirmed the lack of association of the 3 strategies or the combined measure with intermediate outcome control (data not shown).

For the end points of medication management, only 1 of the 12 associations showed a significant difference in the expected direction. Greater use of physician reminders was associated with a 7% (CI, 1% to 13%) increase in number of patients who had serum LDL cholesterol levels at target or who were using a cholesterol-lowering medication. Paradoxically, the only other significant finding suggested that persons in physician groups with the most intense, structured diabetes care management were 4% less likely to have a well-controlled hemoglobin A_1c level or to be receiving 2 or more medications or insulin (Table 5; Appendix Table 2).

Although disease management is widely implemented in managed care, its effectiveness has not been carefully examined. A lack of valid comparison groups and competing secular influences make it difficult to evaluate programs within single organizations. The TRIAD study makes it possible to estimate the impact of disease management by collecting detailed, standardized data across multiple physician groups that implement these strategies differently. In addition, the TRIAD health plans' standardized information allows us to control for health plan actions while we examine the influence of care management at the level of physician group.

Our findings indicate strong associations between the intensity of the 3 disease management strategies and better care processes. These were noted despite relatively high process scores among all physician groups (32). The magnitude of observed differences was moderate (generally ranging from 5% to 15%) but similar to differences achieved in efficacy trials of individual disease management interventions ((12), (16)). However, none of these disease management programs seemed to affect intermediate outcomes in a clinically meaningful way, and they were not associated with apparent differences in the amount of medications used to control these outcomes.

Our findings are consistent with previous reports. Keating and colleagues (33) studied diabetic patients in 1 market by using a summary quality score that included 4 processes and 2 measures of intermediate outcome control. Across 135 physician practices, they found that practice accounted for only a small proportion of variation in quality. Nonsignificant trends toward higher quality were found for practices that received diabetes performance report cards and those that routinely enrolled patients in disease management programs. Fleming and colleagues (34) found that practices that used multiple care management approaches were often in the upper quartile of the distribution in a similar quality composite score. In both of these reports, the quality scores were dominated by process measures, and the influence of diabetes care management on processes versus control of intermediate outcomes cannot be separated. The latter report also indicated that the relative influence of distinct care management practices, such as the use of reminders versus performance feedback at the health plan level, cannot be directly compared (34).

In our analysis, we must carefully consider the absence of positive associations between disease management and intermediate outcome levels or responsive adjustments to treatment regimens. Accreditation organizations gave early emphasis to simple process-of-care measures that were easy to assess with administrative data (for example, whether a hemoglobin A_1c test was done). Such approaches ignored intermediate outcome levels and their clinical treatments (for example, whether there was management change in response to a high hemoglobin A_1c level). Disease management programs developed in response to accreditation undoubtedly focused more heavily on improving processes than outcomes of care. Our findings support the need for refinements in disease management that shift the focus toward direct measurement and feedback of intermediate outcomes and toward measurement of clinical processes of care that are more directly associated with improved outcomes (35 - 36).

Improving intermediate outcomes is more challenging than altering simple care processes. Process improvement can be more readily applied to entire populations with diabetes; however, intermediate outcome control requires identification of patients with elevated levels, targeted interventions, and support of self-management. In addition, control of intermediate outcomes requires the active participation of primary care physicians who may yet lack sufficient knowledge, decision support, or time to appropriately support patients in achieving control (37).

Health plans also apply disease management strategies. Because there were only 10 TRIAD health plans, we cannot say whether use of these strategies at the health plan level would be associated with the same pattern of findings reported here at the physician group level.

Several additional limitations of this study deserve mention. Because these cross-sectional analyses were conducted after generally 3 or fewer years of disease management, it may be too early to detect effects on intermediate outcome levels. Our analytical sample included only 63 physician groups that were not randomly selected from a larger, population-based listing. Furthermore, disease management strategies were not randomly allocated across groups. A larger physician group sample could have increased sensitivity to modest associations by increasing the statistical power as well as increasing the variability in disease management intensity. We used “more medication” as a surrogate for more aggressive care, but our cross-sectional analyses could not be used to determine whether providers changed therapy in response to suboptimal control of an intermediate outcome. Because some patients did not provide consent, medical records could not be obtained for approximately 30% of the entire TRIAD sample. However, survey data indicated that patients with missing medical records were quite similar to the other participants in terms of demographic characteristics, duration of diabetes, and self-reported health status.

The TRIAD study suggests that the disease management programs implemented in these physician groups were associated with higher levels of care processes, but they did not seem to influence the intermediate outcomes that affect risk for complications or the medical management of these outcomes. As disease management programs give greater attention to improving intermediate outcomes and as process measures become more clinically relevant, these programs may be more likely to meet their long-term goal of improved health.

Hawaii Translational Research Center and Pacific Health Research Institute: Principal Investigator: J. David Curb, MD, MPH. Co-Principal Investigator: Beth Waitzfelder, PhD. Co-Investigators: Richard Chung, MD; R. Adams Dudley, MD, MBA; Chien-Wen Tseng, MD, MPH; Thomas Vogt, MD, MPH. Analyst: Qimei He, PhD. Project Coordinator: Suzanne Firrell.

Indiana University Translational Research Center: Principal Investigator: David G. Marrero, PhD. Co-Principal Investigator: Ronald T. Ackermann, MD, MPH. Co-Investigators: Matthew J. Bair, MD, MS; Ed Brizendine, MS; Aaron E. Carroll, MD; Gilbert C. Liu, MD, MS; Paris Roach, MD; Changyu Shen, PhD; Morris Weinberger, PhD; Madelyn L. Wheeler, MS, RD, CD, FADA, CDE. Project Coordinator: Susanna R. Williams, MSPH.

Kaiser Foundation Research Institute: Principal Investigator and Study Chairman: Joe V. Selby, MD, MPH. Co-Principal Investigator: Andrew J. Karter, PhD. Co-Investigators: Assiamira Ferrara, MD, PhD; Julie Schmittdiel, PhD. Senior Analyst: Connie Uratsu. Project Director: Bix E. Swain, MS.

University of California, Los Angeles: Principal Investigator: Carol M. Mangione, MD, MSPH. Co-Principal Investigator: Arleen F. Brown, MD, PhD. Co-Investigators: O. Kenrik Duru, MD; Susan Ettner, PhD; Shaista Malik, MD, PhD; Martin F. Shapiro, MD, PhD. Senior Analyst: Neil Steers, PhD. Project Director: Chester Brown Jr., MHA. Senior Administrator: Carole Nagy, BA.

University of Medicine and Dentistry of New Jersey: Principal Investigator: Norman Lasser, MD, PhD. Co-Principal Investigator: Stephen H. Schneider, MD. Co-Investigator and Project Director: Dorothy A. Caputo, MA, APRN, BC-ADM, CDE. Co-Investigators: Jesse C. Crosson, PhD; Stephen Crystal, PhD; David S. Kountz, MD; Shou-En Lu, PhD; Monika M. Safford, MD; Weichung (Joe) Shih, PhD; Leslie-Faith Morritt Taub, DNSc, ANP-C, GNP-BC.

Central Activities Coordinating Center: TRIAD Central Program Specialist: Gabrielle J. Davis, BS, CHES. Central Program Assistant: Lucyna Lis, MA. Program Coordinator: Sonja Ross, BS, MHS.

University of Michigan Health System: Principal Investigator: William H. Herman, MD, MPH. Co-Investigators: Catherine Kim, MD, MPH; Michele Heisler, MD; Susan Johnson, MD; Kingsley Onyemere, MD. Research Associates: Ray Burke, MA; Laura McEwen, PhD, MPH; Bahman Tabaei, MPH. Project Director: Jennifer Goewey, MHA.

Department of Veterans Affairs: Principal Investigator: Eve A. Kerr, MD, MPH. Co-Principal Investigator: Rodney A. Hayward, MD. Co-Investigators: Sarah Krein, PhD; John Piette, PhD. Project Managers: Mary Hogan, PhD, RN; Fatima Makki, MPH, MSW. Data Manager: Jennifer Davis, MPH.

National Institute of Diabetes and Digestive and Kidney Diseases: Co-Investigator: Sanford A. Garfield, PhD.

Centers for Disease Control and Prevention: Lead Consultant: K.M. Venkat Narayan, MD, MSc, MBA, MRCP. Collaborating Investigators: Gloria Beckles, MD, MSc; Patrick Boyle, PhD; Michael Engelgau, MD, MS; Tiffany Gary, PhD; Linda Geiss, MS; Robert Gerzoff, MS; Edward W. Gregg, PhD; Betsy L. (Cadwell) Gunnels; Roberta H. Hilsdon, BBA, AAS; Henry Kahn, MD, FACP; Jinan Saaddine, MD; Mark Stevens, MSPH, MA; Theodore Thompson, MS; Ed Tierney, MPH; Roldofo Valdez, PhD; David F. Williamson, PhD; Ping Zhang, PhD. Management and Program Analyst: Shay Clayton. Preventive Effectiveness Fellow/Health Service Researcher: Rui Li, PhD. Program Administrator: Bernice Moore, MBA. Central Administrative Data Coordinator: Dori Bilik, MBA.

We describe how we constructed the physician performance feedback, intensity of physician reminders, and intensity of diabetes disease management scores. Scores were calculated for each health plan (HP) and physician group (PG).

To compute the physician performance feedback composite score, we counted the number of items checked on a list of 15 possible diabetes processes and outcomes that could be given back to physicians. After we averaged the counts over PGs and HPs separately, each group and plan count was then z-transformed.

The physician reminders composite was constructed from 2 multi-item questions. One question described the type of reminder, and the second described its content.

The PG/HP survey listed 6 types of reminders that organizations could send their clinicians. Each PG/HP was assigned a score between 0 and 5 that corresponded to the most intense reminder checked (no reminders sent, 0; preprinted guidelines only sent, 1; flowsheets sent, 2; flags on medical records sent, 3; mailed list of patients, 4; customized alerts on medical records, 5).

The PG/HP surveys included 9 possible items that are commonly included in reminders. Each PG/HP reminder content score corresponded to the number of reminder items checked.

The sums for both the type and content scores were z-transformed, as was done for the feedback score. The 2 z-transformed scores were averaged to give the final score for composite reminders.

The comprehensive disease management composite combined 4 domains: disease management programs, intensity of diabetes guideline use, intensity of patient reminders, and diabetes education.

We considered 3 aspects of disease management: program implementation, intensity of disease management, and formal case management.

A plan or group's score on this indicator was the number of years that it had had a diabetes disease management program multiplied by the proportion of patients with diabetes enrolled in each PG/HP. Scores were z-transformed.

A plan or group's score was the number of 9 possible disease management strategies included in the disease management program. Scores were z-transformed.

This indicator had 2 components: the number of diabetes case managers per 10 000 patients and the intensity of case management. To obtain the score for intensity of case management, each PG/HP was assigned a score between 0 and 3 that corresponded to the most intense of the following strategies checked (no patients assigned to case managers, 0; assignments made by primary care physician referral or patient self-referral, 1; most costly patients, recently discharged patients, or patients at high risk for micro- or macrovascular complications assigned, 2; all patients assigned, 3). The intensity score and the number of case managers per 10 000 patients in each PG/HP were z-transformed and the 2 components summed.

The disease management score was the mean of the z-transformed program implementation, disease management intensity, and case management scores.

Each PG/HP received a score between 0 and 4 that corresponded to the most intense strategy for implementing guidelines checked (no guidelines used, 0; guidelines given to patients in written form or as educational talks, 1; guidelines sent in written form to physicians, 2; guidelines provided to physicians in computerized form, 3; guidelines incorporated into physician reminders or automated patient reminders, 4). Scores were z-transformed.

We combined 2 multi-item indicators for this measure. The first was a weighted patient reminders indicator, which summed 6 possible reminders that were weighted by how often the reminders were sent to patients. The second indicator was a score between 0 and 4 that corresponded to the most intense of 4 possible strategies for sending reminders (no reminders sent, 0; generic newsletters or HP newsletters sent, 1; letters sent to specific patients, 2; letters sent from primary care physicians to specific patients, 3; personal telephone calls made to patients, 4). Both reminder scores were z-transformed, and the mean of the z-transformed scores was the intensity of patient reminders composite score.

This composite was constructed from the following 3 items: whether or not the PG/HP sponsored group or individual diabetes education classes; the number of full-time equivalent certified diabetes educators that were employed by the PG/HP; and whether the PG/HP had a formal diabetes education program. Each of these 3 indicators was z-transformed; the diabetes education composite score for each PG/HP was the mean of the 3 z-transformed scores.

The scores for disease management, guidelines, patient reminders, and education were highly correlated; pairwise Pearson r values ranged from 0.63 to 0.88. To avoid multicollinearity, we combined these scores into an overall structured diabetes care management comprehensive diabetes disease management score. The mean of the z-transformed indicators within each composite (the 3 indicators in the disease management composite; the 1 indicator in the guidelines composite; the 2 indicators in the patient reminders composite; and the 3 indicators in the education composite) was the structured diabetes care management score for each PG/HP.