Impact of Population Management With Direct Physician Feedback on Care of Patients With Type 2 Diabetes

Patients and physicians at the intervention site received population management, whereas patients and physicians at the control clinic site continued with usual care. Within each clinic site, diabetes registries were created by screening the clinic population for diabetes-related billing claims (International Classification of Diseases, 9th Revision [ICD-9] codes 250.00–250.90). Potential diagnoses were then confirmed by detailed chart review. The PopMan clinical software was used to rank patients within the intervention site diabetes registry according to either highest HbA_1c or cholesterol levels. Once identified, the overall diabetes management of each intervention patient was reviewed. Appropriate testing intervals and goal risk factor levels for HbA_1c, cholesterol, and blood pressure were determined according to the ADA Clinical Practice Recommendations published in the year preceding the study (23). Applying these evidence-based criteria to the patients identified by the PopMan ranking program and to a similar, matched cohort of patients from the control clinic (23), we compared the proportion of patients with changes in the management of three diabetes-related metabolic risk factors—hyperglycemia, hypertension, and hyperlipidemia—over a 3-month interval. The study was approved by the Massachusetts General Hospital/Partners Health Care System Institutional Review Board. In accordance with section 903C of the Public Health Service Act (42 U.S.C.299a-1), confidentiality of all patient information was strictly maintained.

The intervention clinical site is an academically affiliated community health center serving a predominately working class community. A total of 910 patients were included in the diabetes registry at this site. Working approximately one-half day per week over the course of 1 year (from July 2001 to June 2002), the population manager (a family nurse-practitioner) used the PopMan software to identify patients in this registry with the highest HbA_1c and cholesterol levels. Although patients were initially selected on the basis of elevated HbA_1c or cholesterol level, multiple aspects of diabetes management were then assessed in these patients by applying the ADA Clinical Practice Recommendations for hyperglycemia, hypertension, and hyperlipidemia management (23). In each intervention patient, the following were assessed: interval since last HbA_1c, cholesterol, and blood pressure measurement; most recent measured value; and current medical regimen for each risk factor. Patient-specific recommendations were then summarized in an e-mail sent to the patient’s PCP. The text of this e-mail was also entered as a clinical note in the electronic medical record (EMR) used for all patient care at this clinical site.

For each patient reviewed by the population manager within the defined study period, we characterized the nature of the recommendations made and assessed changes in clinical management over the subsequent 3 months in response to the recommendation. Changes in management were identified by reviewing the EMR for subsequent clinic visit notes, patient phone calls or letters, and laboratory testing results. Our primary outcome was percentage of specific recommendations that were followed. Recommendations were grouped into two categories: 1) testing (for HbA_1c, blood pressure, or cholesterol) and 2) changes in therapy (to initiate or increase therapy for hyperglycemia, hypertension, or hyperlipidemia). For intervention patients evaluated more than once over the course of the study, only recommendations from the first e-mail were analyzed. For all intervention patients, we also recorded physician explanations for not following specific recommendations and tabulated potential barriers to effective care noted by the PCPs in the EMR.

To assess the impact of the population manager intervention beyond what changes may have occurred without her specific recommendations, we analyzed a comparison cohort in the control clinic. The comparison cohort consisted of diabetic patients from a second primary care practice affiliated with the same academic medical center. From an overall population of 2,240 diabetic patients in this clinic, we defined a group of control patients clinically similar to the 149 intervention patients by matching in a 1:1 ratio within 5 years of age, 0.5% of HbA_1c, and 10 mg/dl of total cholesterol. Our goal was to identify two groups of patients with comparable metabolic control who would generate similar overall management recommendations in the aggregate. For these control patients, we chose 1 March 2002 as the assessment date. Looking back from this date, we applied the same care guidelines used by the population manager to create a set of hypothetical recommendations for hyperglycemia, hypertension, and hyperlipidemia management applicable to the control cohort. We then looked forward 3 months from 1 March 2002 to determine whether these hypothetical recommendations were followed in the absence of prompting from a population manager. This study design allows a controlled test of management changes attributable to population management beyond changes that might be expected to occur in the course of usual care alone.

For all patients in either the intervention or the control group, medical records were reviewed and the following recommendations were made: 1) repeat HbA_1c testing if most recent results are more than 6 months out of date and repeat testing for either blood pressure or cholesterol if last measurements are more than a year out of date; 2) increase therapeutic regimen for glycemic, blood pressure, or cholesterol control if any of these levels are above the goal (HbA_1c >7%, blood pressure >135/80 mmHg, cholesterol >200 mg/dl, or LDL >130 mg/dl) and if the patient is either not on therapy or on an unchanged dose of therapy for more than 6 months.

Therefore, a patient selected because of elevated cholesterol level, for example, might be found to have an outdated HbA_1c result, untreated hyperlipidemia, and normal blood pressure. After reviewing the medical record for current therapy and medication allergies, the population manager would send an e-mail to the patient’s PCP recommending repeat HbA_1c testing and initiation of lipid-lowering therapy.

All analyses were performed using SAS statistical software (SAS Institute, Cary, NC) (24). Baseline comparison of clinic, physician, and patient characteristics used Student’s t test for normally distributed continuous variables and χ² tests for proportions. Within the intervention group, χ² tests were used to compare proportions of recommendations followed, comparing the categories of testing and therapeutic change recommendations. For the primary outcome comparing changes in evidence-based management between the intervention and control groups, we used generalized estimating equations to account for possible effects of physician clustering (25). For all tests, a P value of <0.05 was considered statistically significant.

The intervention clinic was staffed by 19 full- and part-time internists representing 9.8 full-time equivalent (FTE) physicians (defined as eight clinical sessions per week) and two nurse practitioners (Table 1). A total of 40 internists (16.8 FTEs) and six nurse practitioners staffed the control clinic. The total number of patients with diabetes per FTE and the frequency of visits per FTE were similar at the two clinic sites.

The population manager identified 149 patients (16.4% of the diabetic patient population at the intervention site) with the highest HbA_1c and cholesterol levels. The mean age was 58 years (SD 14), 53% of the patients were women, the mean HbA_1c was 9.7% (SD 2.7%, range 5–15%), and the mean total cholesterol was 211 mg/dl (SD 51, range 112–343). The 149 patients in the matched control cohort were statistically similar with regard to age, sex, HbA_1c, and total cholesterol values at baseline, and testing rates for HbA_1c and LDL cholesterol in the prior year (Table 2).

We assessed recommendations made by the population manager and corresponding changes in care for the 149 intervention patients. Of these patients, 110 (74%) had a follow-up visit with their PCP in the subsequent 3-month period. Overall, the population manager made 216 management recommendations (1.4 recommendations per patient) related to testing or therapy for hyperglycemia, hypertension, or hyperlipidemia.

In the subset of 110 patients with 3-month PCP follow-up visits, significantly more testing recommendations were followed by the PCPs than therapeutic change recommendations (P = 0.001) (Table 3). After reviewing PCP progress notes subsequent to the population manager’s e-mail message, we found that explanations for not following a particular recommendation were more common for therapy recommendations than for testing recommendations (P = 0.04). Of the 24 therapy recommendations that were not followed, common explanations included trend of recent improvement (n = 4), pending laboratory results (n = 4), continued trial of lifestyle changes (n = 3), and patient nonadherence with current regimen (n = 3).

Physicians noted barriers to clinical care in the EMR for 93 patients in the intervention cohort (62%). The four most common barriers to effective care were: 1) mental health problems (including depression or anxiety) in 52 patients (35% of the intervention cohort); 2) prior history of poor adherence to care plans (39 patients, 26%); 3) ongoing or prior substance abuse (27 patients, 18%); and 4) serious comorbidity limiting effectiveness of diabetes care (14 patients, 9%). A total of 41 patients (28%) had more than one barrier reported in the EMR. In an exploratory analysis, we did not find any relationship between the proportion of recommendations followed by physicians and the presence of patient-level adherence barriers (P = 0.9 comparing proportion of recommendations followed in the intervention group between patients with and without adherence barriers).

We tested the hypothesis that recommendations made by the population manager resulted in changes in management beyond what may have occurred during usual care. We applied identical testing and therapy management criteria to patients in the control cohort without forwarding the resulting recommendations to the control PCPs. Changes in management were then monitored for the 3-month follow-up period after control recommendations were made. Of the 149 control patients, 117 patients (79%) had a follow-up visit with their PCP in the subsequent 3-month period. A total of 204 control recommendations were made (1.4 recommendations per patient). There were no significant differences comparing intervention and control groups in the proportion of patients with PCP follow-up in 3 months (P = 0.7) or in the number of recommendations made per patient (P = 0.7).

Overall, a significantly greater proportion of evidence-based guideline practices were followed in the intervention group than in the control group (59 vs. 45%, P = 0.02; Table 4), suggesting that population management improved diabetes management beyond what changes may have occurred during usual care over a 3-month period. Although not statistically significant when analyzed separately, both testing rates (78 vs. 71%, P = 0.4) and changes in therapy (36 vs. 30%, P = 0.3) were greater in the intervention group compared with the control group.

We compared changes in the process of care for patients receiving population management with those in a matched cohort of patients without population management in similar academic primary care practices. We found that population management with patient-specific provider feedback led to increased rates of testing and therapy changes over a 3-month follow-up period when compared to usual care. Over a longer period of observation, it is likely that such evidence-based changes in management would result in an overall improvement in metabolic control.

A secondary goal of this study was to assess the prevalence of potential patient-related barriers to effective management. It is straightforward to audit the EMRs of a cohort of patients and find flaws relative to evidence-based management goals. The assumption that pointing out these flaws will lead to improved metabolic control fails to account for the multitude of “real world” barriers that interfere with evidence-based management. Physicians reported barriers to effective clinical care for most patients in the intervention cohort. Among these patients with poor glycemic and/or lipid control, there was a very high prevalence of depression and other mental health problems, substance abuse, and poor prior adherence to care plans.

Population management represents a novel approach to reducing clinical inertia and increasing patient engagement with the medical system. Distinct from case management, a time-intensive approach that relies on directing substantial resources toward individual patients with the greatest needs, population management seeks to more efficiently harness existing resources to improve the overall level of care for a given patient population. The population manager who reviewed the diabetes registry in our intervention site was a nurse practitioner who was familiar with medical therapies for diabetes. However, many of her recommendations could have been made by a less experienced provider, such as a licensed practical nurse (LPN) or other clinic staff member, following explicit algorithms for testing and visit frequency.

Our findings of a modest but significant impact on overall care parameters suggest that population management is a valuable approach to care. Our method of sending recommendations via e-mail directly to PCPs represents a relatively simple intervention. Strategies to implement changes in care for patients with poor metabolic control that do not require actions by PCPs, such as sending letters and laboratory testing requisitions directly to identified patients and scheduling follow-up appointments with PCPs, nutritionists, or diabetes educators as indicated, may have an even greater impact and should be studied. Future interventions to improve metabolic control in populations of patients with type 2 diabetes must also address the high prevalence of mental health and substance abuse problems and barriers to medication adherence and healthy lifestyle changes.

This study is part of the MGH Diabetes Mellitus: Putting Evidence into Practice Project, which is supported by grants from the Aetna Quality Care Research Fund and Pfizer Pharmaceuticals and by the MGH Primary Care Operations Improvement and Clinical Research Programs. R.W.G. received support from National Research Service Award 2-T3211001-14 and J.B.M. is supported by a career development award from the ADA.

We thank Nancy Wong, BS, Jessica A. Sandberg, MPH, and Jo Ann David-Kasdan, RN, MS, for assistance with data abstraction and analysis.