Impact of a Program to Improve Adherence to Diabetes Guidelines by Primary Care Physicians

RESEARCH DESIGN AND METHODS

Background and study objectives

The Diabetes Research and Training Center at Indiana University has previously conducted surveys of the state’s primary care physicians to assess reported quality of diabetes care as well as attitudes toward diabetes care guidelines (2,3,7,20, 23). Several physicians from one county later requested that we work with physicians in their county to help them improve their care of diabetic patients. We initiated a project with several objectives: 1) to assess, through chart audits, the current state of care in relation to ADA guidelines; and 2) to determine whether a multifaceted intervention (consisting of repeated audit and feedback, development of local consensus guidelines, provision of physician and patient education, and provision of practice aids) would improve physician adherence to diabetes guidelines. Our hypotheses were that adherence to diabetes guidelines would improve after feedback was given to PCPs about their baseline performance and after development of local consensus guidelines, and that further improvements would be seen after implementation of the remaining interventions and repeated audit and feedback.

Health care system and demographics of the study county

The study county is rural (population 33,866) (31) and located ∼50 miles from Indianapolis. A small hospital (88 beds) and all physician offices are located in the county seat (population 16,662) (31). Primary care is provided by seven physicians (three internists, two general practitioners, and two family practitioners), each of whom maintains an independent practice. The PCPs are all men, with a mean number of years in practice of 21 (range 9–34) at the start of the study.

Other physicians in the county include a general surgeon, an obstetrician, a pediatrician, and a podiatrist. There were no ophthalmologists in the county during the 3 years of the project; however, there were several optometrists. Resources at the county hospital include one RN/Certified Diabetes Educator who conducts diabetes classes and one-on-one education for inpatients or outpatients referred by a physician. Several hospital staff dietitians, none with special expertise in diabetes, provide dietary counseling for referred patients.

Steps of the project

The project involved several sequential steps, to be described more fully below. Initially, the physician group developed consensus guidelines and agreed to adopt them. Chart audits were done on a randomly selected group of patients with diabetes, covering the year before the adoption of the guidelines. Information was provided to each physician about his own performance in terms of adherence to the guidelines, as well as information about performance of the group as a whole. A year after adoption of the guidelines and feedback of baseline performance data to the PCPs, a second chart audit was done; these results were again presented to the PCPs. A series of interventions, directed at physicians, patients, and the healthcare system, were conducted during the second year of the project. A third chart audit was conducted to encompass the final year.

The project was approved by the Institutional Review Board of Indiana University-Purdue University at Indianapolis. The board deemed that informed consent of patients was not required, since investigators were blinded to all identifying patient data.

Development and adoption of consensus guidelines

A guidelines group was formed, consisting of two PCPs (one internist and one family practitioner), an Indiana University endocrinologist who served as facilitator, the hospital Certified Diabetes Educator, and a nurse from the hospital’s Quality Assurance department. The group used the Standards of Care of the ADA for that year (32) as a starting point for developing the guidelines. The endocrinologist presented the evidence for each guideline. The local practitioners discussed each guideline and by consensus either agreed with the guideline or modified it to reflect prevailing practice styles and beliefs. The endocrinologist did not direct the discussion process, other than to occasionally review evidence. In most cases, modifications to the ADA guidelines made them slightly less broad. For example, in several cases the guideline group thought a guideline should be optional for patients over the age of 75. Additionally, the group believed that no microalbumin screening, beyond a yearly urinalysis, should be required for patients already treated with renoprotective antihypertensive drugs. Occasionally the group’s guidelines were more stringent than the ADA’s (for example, expecting that the PCP should do an annual eye exam in addition to referring the patient to an eye care specialist). The group developed guidelines in eight general areas: self-monitoring of blood glucose (SMBG), HbA_1c monitoring, screening for eye complications, screening for foot complications, screening for renal complications, lipid screening and treatment, blood pressure screening and treatment, and smoking assessment and cessation counseling. The consensus guidelines are available from the authors on request.

After the guidelines group developed the guidelines, an evening meeting was scheduled with all seven PCPs. The two PCP members of the guidelines group presented the guidelines to their colleagues, and after discussion the group agreed to adopt the guidelines. The guidelines were then distributed in paper form to all PCPs. Subsequent newsletters related to the project included the guidelines, as well.

Chart audits and feedback of performance data

Each PCP provided a list of medical record numbers of patients with diabetes (generated from International Classification of Diseases, Ninth Revision code 250.xx) who were seen within the prior 6 months. From these lists, a random sample of approximately half of each physician’s patients or 30 patients, whichever was larger, was selected for chart audit. Three nursing employees of the hospital were hired and trained by the endocrinologist to abstract charts using audit forms. The full chart audit was begun after interrater agreement for five charts was >95%. The initial audit included 275 charts of patients seen at least once during the year before the adoption of the consensus guidelines. A second audit was conducted a year later, after adoption of the guidelines, feedback to PCPs about their performance on the baseline audit, and dissemination of practice aids. The final audit was conducted 2 years later, after a year of interventions directed at physicians, patients, and the health care system.

After each chart audit, feedback was provided to PCPs as follows: demographic and clinical data about patients were tabulated, and data related to adherence to each guideline were analyzed. An evening meeting was scheduled for study staff and the PCPs. Each physician was provided with a document showing his own data as well as pooled data for the physician group. The endocrinologist discussed group data and answered questions about methodology or trends. For year 1 and year 2 audits, a similar process was used, but each PCP’s document included his own and pooled data for the baseline year and for the year in question.

Additional interventions

Practice aids.

In response to requests from the PCPs, a number of practice aids were developed and disseminated. These included chart stickers for diabetic patients’ charts, which were brightly colored and included cartoons of an eye, a kidney, and a foot. These stickers were meant to trigger the clinic staff to instruct patients to remove their shoes when they were put in the examination room, and to remind PCPs to recommend eye and kidney screening tests. Chart flow sheets were also developed, which facilitated tracking of laboratory results and examinations.

Linked physician and patient education sessions.

A series of targeted sessions were held with PCPs, covering various topics related to the guidelines. These sessions were led by specialists from Indiana University School of Medicine and were intended to be evidence-based and practical, with ample time for informal discussion. Within a few weeks of the physician session, an education session on the same topic was held for patients and their families. For example, after physicians talked about dyslipidemia, a session for laypersons entitled, “High cholesterol and diabetes: what can you do?” was held. Project staff provided information for PCP offices to mail to all patients with diabetes about these sessions, and the sessions were also advertised by flyer and in the local newspaper. Patient sessions were conducted by Indiana University specialists or local health care providers; ∼80 people attended each patient session. In addition to providing information about the medical topic, presenters emphasized steps patients could take at their physician’s office to improve their diabetes care, such as requesting certain tests or examinations.

Computerized nutritional support.

A barrier to diabetes care noted by the participating physicians was the inability to develop individualized meal plans for patients. In response to this need, a computer system was installed in the local hospital that enabled the hospital diabetes educator and registered dietitian to assist patients in constructing meal plans. The system, Computer Planned Menus, was developed by our group and previously used as part of successful diabetes interventions (33). The program uses a patient’s food preferences to generate daily menus that follow a specific diet prescription, generate shopping lists, and provide complete nutrient analyses.

Analyses

We assessed comparability of baseline characteristics of patients with at least one follow-up audit and those with no data after baseline. Age, sex, race, blood pressure, LDL cholesterol, insulin use (yes/no), diabetes complications, comorbidities, and guideline adherence variables were compared. Means of continuous variables were tested with two sample t tests and categorical variables with χ² tests. If distributional assumptions were not met for continuous variables, the Wilcoxon’s sign-rank test was used.

A chart audit was done if a patient had at least one visit during the year in question. For each adherence guideline studied, we calculated the proportion of audited patients for whom the guideline was followed. We tested effects on adherence at year 1 and year 2 in separate models, because different interventions were applied during these two intervals. Comparisons between baseline and year 1 adherence included only patient charts audited at both of those times. To test change in adherence over time, we used a logistic regression model including physician as a random effect to account for correlation of adherence outcome between patients with the same physician. Compound symmetry was assumed for the structure of the variance-covariance matrix of observations from the same individual at multiple times. We used the same methods to test change from baseline to year 2 adherence, including only patient charts with data in both audits.

Changes in blood pressure from baseline to year 2 were assessed with paired t tests. Because of a non-normal distribution, changes in LDL cholesterol and HbA_1c were tested with sign-rank tests for differences from baseline to year 2. All analyses were conducted with SAS version 8.2 (SAS Institute, Cary, NC).

RESULTS

Demographic and clinical variables for the 275 patients included in the baseline sample are as follows. The mean age was 61 years, and the sample was overwhelmingly Caucasian. Approximately 40% of the patients used insulin, and 46% had at least one diabetic microvascular complication. All patients had at least one of nine audited comorbid conditions, with hypertension (55%), osteoarthritis (35%), and coronary artery disease (31%) being the most prevalent.

There was a decline in charts that could be audited in the two subsequent years. One PCP opted not to participate in the project after the baseline chart audit, and for the remaining PCPs, there was attrition in charts available for audit due to patient death, relocation, or loss to follow-up. There were no significant differences in demographic or clinical variables between the three audits, with the exception of LDL cholesterol, which was significantly lower in subjects whose charts were not included in subsequent audits (median 104 vs. 121 mg/dl; P = 0.032). Because the physician who dropped out of the project had somewhat lower baseline adherence than the other physicians, adherence to several guidelines was lower in the subset of patients whose charts could not be audited after the baseline period compared with patients whose charts were assessed in subsequent years.

As shown in Table 1, during the year before adoption of the guidelines, adherence to the guidelines was generally well under 80%. Adherence was 15% for foot examinations, 20% for glycohemoglobin testing, 23% for eye exam referrals, 33% for urine protein or microalbumin testing, 35% for smoking cessation counseling, and 44% for lipid testing. Performance was better for SMBG use for insulin-using patients (73%) and blood pressure monitoring (78%). Because of very low rates of documentation of smoking status, smoking cessation counseling was dropped from subsequent chart audits.

Changes in rates of adherence between the baseline year and year 1 (after adoption of guidelines and feedback of performance data to PCPs) are shown in the center columns of Table 1. Statistically significant improvements were seen in blood pressure screening (71 vs. 83%), comprehensive foot exams (19 vs. 42%), glycohemoglobin testing (26 vs. 37%), and annual PCP eye exams (38 vs. 46%). A trend toward improvement that did not reach statistical significance was seen in eye care referrals (25 vs. 33%; P = 0.059). No significant changes were seen in SMBG use (77 vs. 84%), lipid testing (46% in both audits), or the combined outcomes of urine protein or microalbumin testing (36 vs. 38%) (all P > 0.48).

Changes in rates of adherence between the baseline year and year 2 (after adoption of guidelines, feedback of baseline and year 1 performance data, and all interventions) are shown in the righthand columns of Table 1. Statistically significant improvements in blood pressure monitoring (70 vs. 92%) and comprehensive foot examinations (22 vs. 47%) were seen again. The year 1 improvements in PCP eye exams, eye exam referrals, and glycohemoglobin monitoring were not sustained, with adherence rates reverting to near the baseline levels. Areas of adherence that had not improved at year 1 remained at baseline levels.

Changes in physiologic variables were examined between baseline and year 2. Blood pressure values were available on 98% of patients on whom chart audits were done at both baseline and year 2, but only 39% of patients in both audits had HbA_1c data available, and only 28% had lipid profile values. There were no statistically significant improvements in glycohemoglobin or in mean systolic or mean diastolic blood pressure. However, median LDL cholesterol improved from 127 to 111 mg/dl (P = 0.011).

CONCLUSIONS

Our finding of low levels of adherence by PCPs to a broad array of diabetes guidelines is consistent with findings by other researchers (2–17). Prior studies have demonstrated successful interventions to improve adherence to diabetes guidelines by PCPs. For example, we showed that reminders programmed into a computerized medical record system improved rates of ophthalmology referral, HbA_1c measurement, and nephropathy screening in an academic primary care clinic (29). Demakis et al. (26) demonstrated similar benefits from computerized reminders in Veterans Affairs ambulatory care clinics staffed by resident physicians. Peters and Davidson (28) significantly improved HbA_1c levels, rates of foot and retinal exams, and frequency of laboratory testing in a managed care setting. Their intervention consisted of a computerized tracking and recall system and patient follow-up by nurses who utilized protocols. Similarly, Clark et al. (30) showed striking improvements in physiologic measures (blood pressure, lipids, and HbA_1c), process measures, and satisfaction with care in a managed care system through use of a multifaceted intervention that included an enhanced data management system, use of nonphysician providers to perform some examinations, and use of protocols and standing orders.

Successful interventions seem to have in common several permissive factors. First, they occur in “closed” systems with standard processes for scheduling, recordkeeping, and carrying out orders. Second, there is generally a hierarchy such that all physicians are influenced by mandates or incentives for improvement. Removing “routine” aspects of care (ordering laboratory studies, referring patients to eye care professionals) from the purview of busy PCPs and allowing nurses or others to be responsible for these behaviors seems to benefit patient care. Finally, information systems can track data, identify high-risk patients, provide reminders, generate standing orders, and track outcomes far more efficiently than paper charts and busy physicians.

Although managed care and very large group practices are increasingly common, physicians in many areas of the country, including the rural Midwest, continue to practice independently with low levels of clinical support and without use of computer systems for patient care. We describe a “real-life” intervention to improve diabetes care by a group of such PCPs that met with mixed success. Promising improvements in five of eight areas were seen in the first year, when there was a high level of awareness by the PCPs of the guidelines and of their baseline suboptimal performance. However, these changes were not well maintained over time, despite labor-intensive and varied interventions targeted at multiple arenas (PCPs, patients, and the health care system).

There are several possible explanations for the lack of sustained effects in most areas. One is that we were unable to effectively assess the “ dose” of the intervention delivered. In general, we provided multiple tools but could not mandate use of the tools by either physicians or patients. For example, because we were blinded to audited patients’ identities, we did not know whether audited patients attended the patient educational sessions and could not correlate attendance with subsequent measures of diabetes care. Similarly, we know that use of computerized nutritional support was initially popular but then declined in frequency, but we do not know whether or how often patients whose charts were audited used the service.

A second possibility is that with the loss of patients from each audit period, we may have lost statistical power to detect improvements. However, examination of the magnitude of change in nonsignificant comparisons did not suggest this; the absolute proportions were quite similar. Other possibilities are that the feedback process lost its power to motivate physicians, or that physicians became complacent in year 2 after seeing the improvements that had occurred in year 1.

Our impression, based on targeted interviews with each PCP at the end of the project, is that physicians continued to recognize the importance of the guidelines, but that with time the large barriers to adherence again overcame the initial forces for change. The primary barrier identified by physicians in the interviews was their perception that patients would not comply with recommendations based on some of the intermediate outcomes of the study. For example, PCPs believed that most patients would not comply with medical nutrition therapy for elevated lipids or with insulin therapy or intensification of insulin therapy for elevated HbA_1c values. A second barrier identified by the physicians was lack of time to carry out multiple diabetes interventions in a brief visit, especially in patients with other medical issues to address. A final issue raised was poor resource accessibility, such as the lack of local ophthalmologists and difficulty getting timely communications from optometrists or distant ophthalmologists. In addition, PCPs believed that insurers were increasingly requiring patients to use out-of-town laboratories, rather than the county hospital laboratory. These perceived barriers are remarkably similar to those delineated by Helseth et al. (34) in their survey research with primary care physicians.

It is interesting that the two areas in which we saw sustained improvements (blood pressure measurement and foot examinations) were behaviors controlled or initiated by nonphysician staff who were responsible for blood pressure measurements and for instructing patients to remove their shoes while waiting for the PCP in the exam room. These results are consistent with those of Peters and Davidson (28) and Litzelman (35) and would suggest that interventions targeted at PCP office staff may be more fruitful than those targeted solely at physicians.

The results of our project suggest that changes in diabetes care are difficult to effect in busy primary care environments, especially when physicians work independently (with no one to mandate or reward change) and without computer support for data organization and reminders. Interventions other than those meant primarily to educate physicians and patients need to be developed and tested. These interventions must target the true barriers that exist, which are less likely to be physician knowledge and more likely to be competing goals that must be addressed in limited time, often with inadequate resources. Increased use of computerized medical records and interventions targeted at nonphysician staff may be most effective.