The American Journal of Managed Care June 2011
Disease Management Programs in Type 2 Diabetes: Quality of Care
Objectives: To determine whether disease management programs (DMPs) for type 2 diabetes mellitus (T2DM) can improve some processes of care and intermediate outcomes.
Study Design: Two cross-sectional registries of patients with T2DM were used for data extraction before (previous cohort) and after (recent cohort) introduction of DMPs in Germany (N = 78,110).
Methods: In the recent cohort, 15,293 patients were treated within the DMPs and 9791 were not. Processes of care, medications, and intermediate outcomes (achievement of treatment targets for low-density lipoprotein [LDL] cholesterol, blood pressure, and glycosylated hemoglobin [A1C]) were analyzed using multi- variable, multilevel logistic regression, adjusting for patient case-mix and physician-level clustering to derive odds ratios and 95% confidence intervals (CIs).
Results: Availability of structured diabetes education and of lipid, blood pressure, and A1C measurements increased over time. In DMP patients, availability was significantly higher for blood pressure and A1C but not for lipid measurements. Prescription of angiotensin-converting enzyme inhibitors, oral antidiabetic drugs, and insulin increased over time and was more common in DMP patients. Statin prescription increased over time but was not influenced by DMP status. Intermediate outcomes improved over time, but DMPs had no influence on intermediate outcomes except for reaching LDL cholesterol targets (odds ratio 1.12 [95% CI 1.06, 1.19] in favor of DMPs).
Conclusions: While there may be some unmeasured confounding, our data suggest that improvement in processes of care by DMPs, as implemented in Germany, only partially translates into improvement of intermediate outcomes.
(Am J Manag Care. 2011;17(6):393-403)
A 2-level cross-sectional study investigating registries of patients with type 2 diabetes mellitus in 2003 and 2007 found that disease management programs (DMPs) were not effectively influencing intermediate outcomes.
- Processes of care improved over time.
- Intermediate outcomes (target achievement for lipids, blood pressure, and glycosylated hemoglobin) improved over time.
- However, there were no relevant differences in target achievements between patients who participated in DMPs and patients who did not participate in DMPs.
Previous studies have shown that, after implementation of guidelines and organizational improvement efforts, process outcomes but not necessarily patient outcomes are improved.3,9 The effects on patient outcomes may often be less clear because they have rarely been assessed.3 The official journal of the German Medical Association stated, based on data either unpublished or provided from health insurance companies, that patients who participate in DMPs receive better care than the ones that do not.10 Although DMPs for type 2 diabetes have been substantially subsidized, studies investigating their effectiveness, which are even required by law (Social Code Book V, paragraph 137f), are by and large missing. Inherent methodologic problems of such studies have been discussed intensely, but the stakeholders have been reluctant to support the suggested necessary randomized trials.11 Interestingly, randomized trials examining the effectiveness of diabetes DMPs have been performed in various countries implementing them, such as (among others) China, Thailand, South Korea, Canada, and the United Kingdom (reviewed by Pimouguet et al12), although these trials are limited by small numbers and examined A1C measurements only.
It is likely that improvements in quality of care are confounded by secular trends, rendering historical control groups unsuitable. Prospective nonrandomized studies are problematic due to the relatively high probability of selection bias causing confounding. Randomized trials, generallybelieved to be the gold standard for prospective controlled studies, are difficult to perform, at least under the conditions of German diabetes DMPs, since criteria for the inclusion of patients enforce preselection. For example, an integral part of the DMPs is the willingness and motivation of the patient to actively pursue treatment goals—the participating physician confirms that he or she includes only patients fulfilling these criteria. It can therefore be expected that basic treatment measures such as weight loss, physical exercise, dietary changes, and smoking cessation are more frequent in patients participating in DMPs compared with patients who are not, therefore improving outcomes. On the other hand, it could be that patients with higher risk or more comorbidities are encouraged to participate in DMPs, therefore resulting in underestimation of DMPs’ potentially positive effect.
There is evidence suggesting that quality of diabetes care improves over time (secular trends).13 The purpose of the present study was to examine (1) whether secular trends show improved processes of care and intermediate outcomes between the years 2002 and 2003 and 2006 and 2007 and (2) whether patients within DMPs receive better processes of care and achieve intermediate outcomes more often compared with non-DMP patients. Data from 2 large crosssectional diabetes registries, the DUTY14 and the LUTZ15 registries, were used. Primary intermediate outcome parameters were proportions of subjects reaching target values for lipid levels, blood pressure, and glycemia.
Study Design and Participants
The study is a 2-level cross-sectional trial in outpatients with type 2 diabetes performed in the years 2002-2003 (“previous cohort,” before the introduction of DMPs, N = 53,026) and 2006-2007 (“recent cohort,” after the introduction of DMPs, N = 25,084) in Germany. In the previous cohort we used the data from the DUTY registry (Diabetes mellitus needs unrestricted evaluation of patient data to yield treatment progress), while in the recent cohort we used data from the LUTZ registry (Lipidmanagement und Therapieziel-Erreichung; lipid management and achievement of treatment targets). Study protocols were approved by the Ethics Committee of the Bavarian Chamber of Physicians. The study designs have been published before.14-19
In short, in the DUTY registry 6700 office-based physicians (general practitioners, internists, and diabetologists) were approached to participate in the study. Each physician was asked to recruit 20 consecutive patients with type 1 or type 2 diabetes under his or her treatment. The diagnosis of diabetes was established by the reporting physician. Reports on 59,075 patients were received. Of these patients, 89.8% had type 2 diabetes, 5.7% type 1 diabetes, and in 4.5% the type of diabetes was not identified. For the present evaluation, only patients with type 2 diabetes were considered. Thus, data of 53,026 patients were analyzed, obtained from 3096 physicians.
In the LUTZ registry, 6551 office-based physicians (general practitioners, internists, cardiologists, and diabetologists) were invited to participate, irrespective of whether they took part in DMPs or not. They were requested to include 6 consecutive patients with diabetes and/or coronary heart disease (CHD). In practices that participated in a DMP, this sample was to be balanced (3 patients in any DMP, 3 non-DMP patients). The large majority of physicians (93.1%) took part in a DMP. Of those, 61.1% participated in DMPs for CHD and diabetes, 10.3% for diabetes alone, and 2.5% for CHD alone; in 25.7% of the cases the type of DMP was unspecified. Practices that did not participate in a DMP accounted for 6.7% (0.4% not reported). A total of 45,873 patients were documented in the registry. Only the ones with the diagnosis of type 2 diabetes were included (N = 25,084). Of these patients, 15,293 were treated within diabetes DMPs and 9791 were not (Figure 1). Among the patients in the diabetes DMPs, 21.5% were also in a DMP for CHD, and 8.9% of the ones not in a diabetes DMP were in a DMP for CHD.
Description of Disease Management Programs
The German type 2 diabetes DMPs are multifaceted and patient-centered interventions that are implemented within the nationwide statutory health insurance system by primary care physicians. Participation is voluntary for both patients and physicians. Medical services in the DMPs include a defined frequency of physician visits, rules for referral to a diabetologist, regular foot and eye examinations, physician counseling regarding lifestyle changes (eg, nutrition, smoking, and exercise), participation in diabetes educationcourses, and agreement on target values for A1C and blood pressure between the physician and the patient. Further elements of the DMPs are documentation of the course of disease and treatment every 3 to 6 months as well as automated reminders for physicians and patients.20 For more information on the DMP contents, a typical contract between health insurers and associations of statutory health insurance physicians can be accessed online: http://www.kvno.de/downloads/dmp_diab2_vertrag.pdf.
Definition of Variables
We defined general clinical management measures as processes of care (determinations of LDL cholesterol, blood pressure, A1C values, documentation of smoking status, providing structured diabetes education), medications (prescription of statins, angiotensin-converting enzyme [ACE] inhibitors, oral antidiabetic agents, insulin, and thrombocyte aggregation inhibitors), and intermediate outcomes, which were defined as achieving the recommended target values as specified by American Diabetes Association guidelines,21 namely, proportions of patients achieving LDL cholesterol concentrations of <100 mg/dL; systolic and diastolic blood pressure of <130 and <80 mm Hg, respectively; and A1C of <7%. Nonavailability of the data was considered to indicate that targets were not achieved.
Coronary heart disease was defined as a history of angina (stable or unstable), myocardial infarction, percutaneous transluminal coronary angioplasty, or coronary artery bypass graft surgery. Cerebrovascular disease was defined as history of stroke or transient ischemic attack. Individuals were considered to have peripheral arterial disease if they had a history of intermittent claudication, defined as posterior calf pain on walking relieved by rest and/or prior limb arterial revascularization. Diagnoses of CHD, cerebrovascular disease, and peripheral arterial disease were made based on the physicians’ clinical judgment. The presence of atherosclerotic disease was assumed when 1 or more diagnoses such as CHD, cerebrovascular disease, or peripheral arterial disease were present.
We conducted univariate analyses to describe patient characteristics and reported means and SDs for continuous variables, and frequencies and percentages for categorical variables. We calculated P values for differences between the previous and the recent cohort and between DMP and non-DMP patients (within the recent cohort), using t tests or Fisher’s exact test where appropriate.
In a multivariable, multilevel model we investigated the influence of time (secular trends) on processes of care and intermediate outcomes and the influence of DMPs. We adjusted for patient case-mix variables using sex (as a categorical variable), age (as a continuous variable), body mass index (as a continuous variable), and concomitant atherosclerotic disease (as a categorical variable). For each process and outcome measure, these logistic regression models were used to calculate estimated odds ratios (ORs) and 95% confidence intervals (CIs) comparing in a first step the influence of time (recent vs previous cohort) and in a second step the influence of DMPs. In this second model, additional adjustments were made for time (as an ordinal variable). In all multivariable models we used generalized estimating equations to investigate the additional effects (beyond case-mix bias) of physician-level clustering on significance testing. The model thus contained 2 levels, the physician’s practice and the patient. It was constructed this way to account forthe correlation between patients within practices and to take advantage of the information contained in these correlations. Formal inferences of significance were based on the Wald test.
We used the statistical software Stata version 9 (StataCorp, College Station, Texas). We used Stata’s xtgee command to model panel data. All reported P values are 2-sided and considered significant at <.05.