Our program goal is to improve adherence to evidence-based guidelines and to reduce clinical inertia in primary care physicians, and to therefore improve the management of glycemic control and cardio-vascular risk factors in persons with diabetes.

One hypothesis is that an advanced quality improvement program (AQIP) significantly improves clinical outcomes in persons with type 2 diabetes compared to a usual quality improvement program (UQIP). Subgroup analyses can analyze the effect of the program in the two intervention arms using cut-off values. The second hypothesis is that persons with type 2 diabetes who make use of a DCT will have significantly better outcomes compared to non-users of the DCT, regardless of their intervention arm.

The study is an open pragmatic cluster randomized trial with before/after measurements and two intervention arms. A cluster design is necessary because randomization is performed on a practice level, the intervention happens on the physician level, but a large part of the data are analyzed at the patient level. The implementation period of the trial is 18 months.

All 379 active primary care physicians (PCPs) in the project region are invited to participate in the project. These PCPs work in a semi-rural setting with 357,000 inhabitants and serve predominantly Caucasian patients with type 2 diabetes mellitus. PCPs provide care for approximately 80% of patients with type 2 diabetes, and are often the sole providers of care. The only inclusion criterion for the providers is agreeing to recruit all patients with type 2 diabetes mellitus to prevent selection bias. In addition, PCPs will be asked to screen more systematically for new type 2 diabetes mellitus patients during the seven months after registration begins. Diabetes is defined in accordance with the 2003 ADA criteria 35 with PCPs making the final diagnosis.

Only patients with type 2 diabetes mellitus will be included in the study, regardless of age. Patients who cannot provide informed consent will be excluded from the study.

The UK Medical Research Council (MRC) framework for the development and evaluation of complex interventions for randomized control trials (RCT) is used as a theoretical guide to designing the intervention 6. The MRC framework allows for the development of a high-quality study design, execution, generalizability of the results, and outlines five key phases for intervention development: a preclinical/theoretical phase, a modeling phase, a phase of exploratory trials prior to the randomized controlled trial (RCT), the trial itself, and long-term implementation 36. All phases except phase two (exploratory trial phase) and phase five (long-term implementation phase) are incorporated here. A detailed overview is provided in Table 1.

This phase involves exploration of the relevant theory and evidence to refine the underlying hypotheses, conceptual model, interventions, and indicators. We have previously performed a review of systematic reviews for this purpose 37. A total of 21 systematic reviews (1989–2006) were included in the review and represented 185 diabetes care programs. Conceptual background, goals, settings, type of programs, type and number of interventions, type and number of indicators, and (cost) effectiveness were evaluated in both the 21 systematic reviews and the individual diabetes care programs. The program is further built on the CCM 89 and principles of integrated care. As there is no unambiguous definition of integrated care, we further build on the definitions of Ellrodt and colleagues 38, Mur-Veeman and colleagues 39, and the Disease Management Association of America (DMAA) 40. We consider integrated care as 'an organizational process of continuous coordination of evidence-based and relevant interventions across the entire health care delivery system and care continuum that seeks to maximize quality of care tailored to the needs of every individual patient while minimizing costs'.

Besides exploring relevant theory and evidence, the local context in terms of existing national and regional governmental policies, characteristics of the region, and perceived barriers to high-quality diabetes care were extensively studied with regard to their impact on the content and execution of the study protocol. We have previously organized stakeholder interviews, including a representative group of 18 Belgian opinion leaders and experts in diabetes care 41.

In the modeling phase, we delineated the components of our complex intervention and the underlying mechanisms by which they influence the outcomes. We sought to understand the pathways by which the problem is caused and sustained, including all barriers to high-quality diabetes care. We also explored whether the pathways were amenable to change, and if so, at which points. Finally, we estimated potential for improvement in both process and primary outcomes. This analysis produced the best achievable combination of intervention components, implementation strategies, and intensities of care delivery, as well as the identification of feasible and valid outcome measures.

Two separate groups are defined: the first group will receive a usual quality improvement program (UQIP), and a second group will receive an advanced quality improvement program (AQIP). Physicians can make use of program services on a voluntary basis.

The UQIP arm will aim to improve adherence to evidence-based guidelines and to reduce the rate of clinical inertia in PCPs. The term 'usual' is applied because these interventions address the principal factors contributing to clinical inertia (physician, patient, and office system factors) and represent standard requirements for what is considered quality of diabetes care in most health care systems according to international clinical guidelines 42, and theoretical frameworks on quality of diabetes care in particular 43. The first intervention arm is innovative to the Belgian healthcare system and adds to available insights from the international literature on how to address clinical inertia in diabetes care.

The AQIP arm will receive similar interventions, but will also include supplementary and experimental interventions that extensively focus on behavior changes in patients and providers. Interventions that focus on the patient aim at a more active involvement of the patient in his/her treatment regimen, with a special focus on lifestyle attitude changes. Improvements in 'patient empowerment' will further decrease clinical inertia by increasing the patient's willingness to intensify his/her treatment 4445. Interventions that target the PCP focus on improvements in communication patterns with patients, interdisciplinary shared care, and involving PCPs in community campaigns. This multi-factorial approach, with a focus on patient, provider, and organizational aspects of care, is fully in line with the latest insights and findings on high-quality chronic care, and high-quality diabetes care in particular 9194647484950515253.

The differences between the AQIP and the UQIP are outlined further below and in Table 2.

We will use two classification schemes to incorporate all six dimensions of the CCM based on the classification scheme from Shojania and colleagues 54, who defined eleven distinct categories of quality improvement interventions adapted from the Cochrane Effective Practice and Organization Of Care (EPOC) group 55. These categories are: patient education, promotion of self-management, clinician education, audit and feedback, case management, team changes, electronic patient registry, clinician reminders, facilitated relay of clinical information to clinicians, patient reminder systems, and continuous quality improvement. Five interventions are not included in the service program as they are either integrated in other interventions of the program (e.g., the patient reminder system is integrated with physician reminder system) or because of complexity in the Belgian primary health care system (case management, audit, electronic patient registry, and facilitated relay of clinical information to clinicians). The different implementation strategies are derived from an overview by Grol and Wensing 56, who have summarized thirteen important theories and models related to the implementation of change to improve diabetes care. These theories/models relate to individual professionals/patients, the social context, and the organizational and economic context.

Feedback interventions, provided by a program manager to the physicians, will include summaries of clinical performance of diabetes care delivered to individual patients over a three-month period. AQIP physicians will receive ongoing benchmarking feedback, whereas the UQIP will only receive benchmarking feedback at the start and end of the project. Feedback includes the percentage of a physician's patients who achieve target levels for glycosylated hemoglobin, LDL, total cholesterol and triglycerides, systolic/diastolic blood pressure, an eye and foot examination, aspirin and statin prescriptions, anti-hypertensive medication, smoking status, and weight loss.

Clinician reminders for physicians of the AQIP are combined with quarterly feedback by the program manager and reminders to make use of the DCT if treatment targets are not met. Physicians are asked to remind their patients about upcoming appointments. Patients are asked by the physicians to make use of a diabetes passport in which the appointments are noted together with important treatment results. Physicians of the UQIP do not receive clinician/patient reminders nor do they receive reminders on the use of the DCT.

Continuous quality improvement will be assured by an iterative process for assessing quality problems in the implementation of the project, developing solutions to those problems, testing their impacts, and then reassessing the need for further action. For this purpose an interdisciplinary quality assurance team will be established that includes a diabetologist, four PCPs, two nurses, internists, dieticians, and pharmacists. The quality assurance team will be asked to monitor the implementation of the project, as well as evaluate outcome indicators of the project. Meetings will be organized on a regular basis with individual members of the quality assurance team.

The project funding agency requires a sample size of at least one-third of the potential PCPs (n = 379), which would capture roughly 2,500 patients with type 2 diabetes mellitus. This sample size allows 80% power (type II error: 0.20) to detect a 20% relative difference between the intervention arms in the proportion of patients achieving a 10% improvement in any one of the following: blood pressure, total cholesterol, or HbA1c (type I error: 0.05; assumed intracluster coefficient 0.6; 67 for calculation methods).

After recruitment, a researcher not involved in the study and blind to the identity of the practices will perform a randomization (by computer-generated numbers) stratified by practice size (solo/duo/group practice) and the presence or absence of an electronic medical recording system. To minimize the possibility of selection bias, all patients within a cluster will be included. Blinding will be ensured for the participating patients, but is not possible at the physician level.

These practices have no pre-existing registers of diabetic patients. Patients with type 2 diabetes mellitus will be identified using physician memory, searching computerized records, and laboratory lists of patients with increased glycemia or registered glycosylated hemoglobin. Baseline data will be collected over a seven-month period. PCPs will be asked to perform a complete examination and blood analysis at the patient's first visit and to complete a paper form. Identified patients without a visit during the first three months of the project will be invited to participate. The completeness of data capture will be double-checked by a data monitor. Final data will be collected over a seven-month period, with call-backs for non-compliant patients. Patient data sheets include socio-demographic and biomedical data. PCPs will need to indicate whether diabetes is treated by the PCP or in a diabetes clinic.

The primary endpoints of the study are the proportion of patients reaching ADA targets for three clinical outcomes: HbA1c < 7%; SBD ≤ 130 mmHg; and LDL-C < 100 mg/dl. Secondary endpoints are the mean improvements in individual values of 12 validated parameters: HbA1c, LDL-C, HDL-C, Total Cholesterol, SBP, DBP, weight, physical exercise, healthy diet, smoking status, and statin and anti-platelet therapy.

Statistical analyses will be performed using an intent-to-treat approach with a multilevel model. Linear and generalized linear mixed models will be used to account for the clustered nature of the data, i.e., patients clustered within PCPs, and repeated assessments clustered within patients. Such models measure how outcomes change over time within patients and whether these changes depend on patient and/or PCP's characteristics, such as the intervention program or DCT use (see hypothesis two). DCT use is defined as having at least one consultation with a member of the team besides the health psychologist and the traveling educator, which are only available for AQIP patients.

We will use generalized estimating equations (GEE), an extension of the quasi-likelihood approach, to test for differences in the evolution of the primary and secondary endpoints for all patients and within the intervention arms. For binary variables, we use the exponential inverse transformation to obtain the 95% confidence interval for the odds ratio.

Subgroup analyses (see hypothesis one) can distinguish intervention effects using different cut-off values. For HbA1C, three subgroups are defined: patients with HbA1c < 7%; HbA1c ≥ 7% and < 8%; and HbA1c ≥ 8%. For SBP, four subgroups are defined: patients with SBP ≤ 130 mmHg; SBP > 130 mmHg and ≤ 140 mmHg; SBP > 140 mmHg and ≤ 160 mmHg; and SBP > 160 mmHg. For LDL-C, four subgroups are defined: patients with LDL-C < 100 mg/dl; LDL-C ≥ 100 mg/dl and < 115 mg/dl; LDL-C ≥ 115 mg/dl and < 130 mg/dl; and LDL-C > 130 mg/dl.

Linear mixed models with subject-specific intercepts and slopes are used to test whether subject-specific evolutions are related to initial parameters. HbA1c will be transformed logarithmically to meet the parametric assumptions of the statistical models. All analyses will be performed using SAS, version 9.