The article by Hayward and his colleagues1 is of interest to those trying to improve clinical guidelines and measures as we move into the rich array of electronic healthcare information that will be readily available in the near future. Currently, both clinical guidelines and their related performance measures are limited by the fact that guidelines must be simple to be remembered and measures can use only claims data, augmented where necessary by basic, yet costly, reviews of paper charts. This situation leads to oversimplification of guidelines and measures as well as potential misuse if they are overinterpreted and pushed beyond their limits. Indeed any general guideline related to control of blood pressure, glycosylated hemoglobin, or any other biologic parameter could cause harm if every patient was pushed therapeutically to comply with that target. It is important to note that the National Committee for Quality Assurance (NCQA) includes a strong advisory that 100% performance is not the goal and that clinical judgment should be used in applying a measure.

Indeed, if 100% compliance were the goal for a measure, it would mean there was absolutely no controversy about the measure or variability in whether patients should be at the threshold goals. There aren’t many issues like that in medicine, but for the ones that come close (eg, aspirin after acute myocardial infarction), compliance is close to perfect, and quality measurement is unlikely to drive improvement. Current guidelines and measures are designed to provide a starting point for the evaluation and improvement of quality of care in a population—not to dictate the care of any individual patient. They are only tools for management.

Although physicians should be aware of guidelines and measures, they need also to apply more nuanced approaches when seeing individual patients. To imply that clinicians would knowingly put patients in harm so they could perform marginally better than other physicians on a clinical performance measure provides a rather dim view of medical practice. The average levels of blood pressure control in persons with diabetes indicate that control is far from optimal overall.2 Thus, it is likely that improved decision making by clinicians and enhanced care of patients will result from providing data to benchmark widely used guidelines related to blood pressure control. Indeed, a reasonable interpretation of the data presented by Hayward et al is that even a highly regarded academic medical center with a sophisticated electronic medical record system may not have reached optimal performance levels for more than half of their patients with diabetes.

Many, but not all, of the widely used evidence-based guidelines (The Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure, American Diabetes Association, American Association of Clinical Endocrinologists) recommend 130/80 mm Hg as the most desirable level of control of blood pressure in most patients with diabetes, specifying diabetes as an equivalent to established cardiovascular disease. At the time the measure was developed, the recommendation of the technical subcommittee of the National Diabetes Quality Improvement Alliance, which was only 1 of the groups providing input to NCQA, to use “adequate doses of 3 drugs” as an alternative to reaching the threshold level was simply not feasible using the relatively simple chart reviews and limited claims data available. This is in contrast to the full electronic medical record data available to Hayward et al. That said, the suggestion by the authors of the current article to exclude patients whose systolic blood pressure is not at threshold but whose diastolic blood pressure is less than 70 mm Hg should first be reviewed by the scientific groups overseeing the evidence-based guidelines on which the Healthcare Effectiveness Data and Information Set (HEDIS) measures are based, and if changes are made in the guidelines, considered in turn by the measurement advisory committees and the Committee on Performance Measurement that oversee HEDIS measures.

At a more fundamental level, a previous article in this journal by one of us (LGP) pointed out the inherent problems in reference standards, guidelines, and measures that use a single threshold and suggested a much more robust set of approaches that could be used in settings with full electronic health record data systems.3 Approaches currently under study by NCQA include assessment of adequate dosing of medications using expanded order entry and pharmacy data, or using blood pressure levels integrated over time and adjusted for baseline patient risk with clinical data available in the electronic medical record. Equally interesting are efforts to link in electronic data systems much more highly evolved clinical guidelines and decision support tools with measurement and feedback in the “real time” of the clinician encounter. An example of an even more forward-looking approach is that of David Eddy and his group at Archimedes4: to create and apply at the time of the patient encounter patient-specific guidelines and clinical decision support using sophisticated mathematical modeling of both the individual patient’s own clinical parameters and the results of clinical trials. Measures then could be linked to this process to determine whether and how physicians use (or don’t use) such information to shape clinical treatment and the results of that treatment in reducing a patient’s overall cardiovascular risk.

However, all of these advances require fully developed and interoperable electronic clinical data systems.5 Until these systems are in widespread use, we will be limited to making rather small and stepwise improvements in current guidelines and their related measures.

From the National Committee for Quality Assurance (LGP), Washington, DC; and Partners Healthcare (THL), Boston, MA.

Dr Pawlson is the Executive Vice President for the National Committee for Quality Asurance, the organization that develops performance measures. Dr Lee reports serving on the advisory board to Bridges to Excellence. The authors (LGP, THL) report no other relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (LGP); drafting of the manuscript (LGP, THL); critical revision of the manuscript for important intellectual content (LGP, THL); and supervision (LGP).

L. Gregory Pawlson, MD, MPH, Executive Vice President, National Committee for Quality Assurance, 100 13th St, Washington, DC 20005. E-mail: pawlson@ncqa.org.

1. Choe HM, Bernstein SJ, Standiford C, Hayward RA.

 New diabetes HEDIS blood pressure quality measure: potential for overtreatment. 

2. National Committee for Quality Assurance.

2009 State of Health Care Quality Report.

 Washington, DC: NCQA; October 2009. http://www.ncqa.org/tabid/203/Default.aspx. Accessed December 14, 2009.

The past as prologue: future directions in clinical performance measurement in ambulatory care. 

4. Heikes KE, Eddy DM, Arondekar B, Schlessinger L. 

Diabetes Risk Calculator: a simple tool for detecting undiagnosed diabetes and prediabetes. 

Improving CRC screening requires innovative approaches: can electronic medical records help? 

Articles

EDITORIAL

Although physicians should be aware of guidelines and measures, they need to apply more nuanced approaches when seeing individual patients.

Clinical Guidelines and Performance Measures

: To evaluate the risk of adverse events (AEs) and the healthcare costs for elderly patients receiving specific potentially inappropriate medications (PIMs).

: Patients 65 years and older who started 1 of 23 PIMs were matched with control subjects who were not receiving PIMs. The following 4 AEs and PIMs were evaluated: delirium or hallucinations with Beers high-severity (BHS) anticholinergics, delirium or hallucinations with BHS narcotics (meperidine hydrochloride or pentazocine lactate or pentazocine hydrochloride), extrapyramidal effects with trimethobenzamide hydrochloride, and falls or fractures with BHS sedative hypnotics. The risk of having the AE of interest within 360 days and the annual healthcare costs were examined.

: Patients receiving BHS sedative hypnotics were significantly more likely to have a fall or fracture than controls (hazard ratio, 1.22; 95% confidence interval [CI], 1.10-1.35). Patients receiving BHS anticholinergics did not have higher risk of delirium or hallucinations than controls (hazard ratio, 1.03; 95% CI, 0.91-1.16). Delirium or hallucinations occurred at a higher rate among patients receiving BHS narcotics, and extrapyramidal effects occurred at a higher rate among patients receiving trimethobenzamide; however, too few events occurred to assess statistical significance. For all PIMs evaluated, annual adjusted medical and total healthcare costs were significantly higher for patients exposed to PIMs than for controls.

: The use of certain BHS PIMs in the elderly may increase AEs or healthcare costs.

	Despite widespread adoption of the Beers list of potentially inappropriate medications (PIMs) in the elderly, outcomes data are limited. Clinical and economic outcomes associated with the use of specific PIMs in the elderly were evaluated.

		Patients using Beers high-severity sedative hypnotics had increased risk of falls or fractures.

		Increased risk of delirium or hallucinations was not observed for patients using highseverity anticholinergics.

		Beers high-severity narcotics and trimethobenzamide hydrochloride may increase the rate of adverse events.

		For all PIMs evaluated, medical costs were significantly higher for patients exposed to PIMs than for control subjects.

	Developed in 1997 to improve the quality of pharmaceutical care, criteria by Beers

 have been widely used to identify potentially inappropriate medication (PIM) use in the elderly. In 2002, the Beers criteria were updated to incorporate new information from scientific literature.

 Severity ratings (high vs low) were assigned to each of the medications. Because some medications listed in the Beers criteria may be justified for certain patients, Zhan et al

 convened an expert panel to classify the Beers criteria into the following 3 categories: (1) drugs that should always be avoided by the elderly, (2) drugs that are rarely appropriate for the elderly, and (3) drugs that have some indications but are often misused.

	Clinicians and managed care organizations (MCOs) have adopted the Beers criteria to help identify and target elderly members who may be at risk of adverse events (AEs) due to their use of PIMs. Starting in 2006, the National Committee for Quality Assurance

 has included a Healthcare Effectiveness Data and Information Set (HEDIS) performance measure for MCOs to evaluate the percentage of elderly members who received PIMs. The specifications for this HEDIS measure are based on the Beers criteria.

	Despite the wide adoption of the Beers criteria within managed care, outcomes data associated with using these PIMs in the elderly are limited. While MCOs are tasked with developing clinical programs to decrease the use of PIMs in the elderly, it is unknown which specific PIMs are most beneficial to target to improve clinical outcomes and reduce healthcare costs. Because the Beers criteria were developed solely based on an expert consensus panel and have not been studied in controlled clinical trials, it has not been proven that reducing PIM use in the elderly prevents adverse clinical outcomes. Although retrospective claims analyses have examined healthcare expenditures or AEs among elderly patients receiving PIMs in the community setting,

 these studies grouped together patients with different types of PIMs and did not thoroughly examine outcomes associated with specific PIMs. Because outcomes may vary for different PIMs, studies evaluating specific PIMs are necessary to build stronger evidence that the use of PIMs leads to adverse outcomes in the elderly.

	We tested the hypothesis that the use of specific Beers high-severity (BHS) medications designated as “always avoid” or “rarely appropriate” in the elderly would result in increased AEs and healthcare costs. Using retrospective claims data, we evaluated the risk of selected AEs and healthcare costs for elderly patients receiving specific BHS PIMs versus comparable elderly patients not receiving PIMs.

	This was a retrospective cohort study using electronic pharmacy and medical claims from an MCO based in the western United States. To focus the analysis on a limited number of medications that would have a higher likelihood of causing AEs, medications were evaluated if they had a severity rating of “high” using the Beers criteria2 and were classified as “always avoid” or “rarely appropriate” by Zhan et al.

 Based on potential adverse outcomes that could occur in the elderly as reported in the Beers criteria

 these medications fell within the following 5 AE categories: delirium or hallucinations related to BHS anticholinergic use, delirium or hallucinations related to BHS narcotic use, extrapyramidal AEs related to trimethobenzamide hydrochloride use, falls or fractures related to BHS sedative hypnotic use, and hypoglycemia related to chlorpropamide use. Because only 40 patients receiving chlorpropamide were identified, the last category was not included in the evaluation. A total of 23 medications and 4 AE categories were evaluated (

	Patients 65 years and older were identified if they started 1 of 23 PIMs listed in Table 1 (with no prior use during the 180-day preperiod) during the identification period from January 2003 through June 2005. The date of the first prescription of the PIM during the identification period was defined as the index date. Patients 65 years and older who were not receiving the PIM or another PIM within the same AE category were selected as potential control subjects. For controls, an index date was randomly selected within the identification period. 

 gives additional inclusion and exclusion criteria that were applied for the different cohorts. The final cohorts were obtained by matching controls with patients exposed to a PIM on a 1:1 basis using the propensity score method.

 Independent variables used to calculate the propensity score for matching are given in 

. Matching was performed in a stepwise fashion; pairs were matched using the lowest window (propensity score, 0.0000001), and then the window was gradually extended by a factor of 10 until reaching 0.1.

	The primary outcome was the risk of having the AE of interest during a postperiod of up to 360 days for exposed patients versus controls. The AEs of interest were delirium or hallucinations for the BHS anticholinergics, delirium or hallucinations for the BHS narcotics, extrapyramidal effects for trimethobenzamide, and falls or fractures for the BHS sedative hypnotics. Patients were designated as having the AE of interest if they had an inpatient hospitalization, an emergency department visit, or 2 medical claims with a primary diagnosis code representing the AE (Table 1).

	Patients were followed up from the index date until the first date when one of the following conditions occurred: (1) the patient discontinued the index medication (defined as a gap of >30 days past the end of supply date for the last prescription for the index medication and the end of the postperiod), (2) the patient in the anticholinergic or narcotic medication evaluation filled a BHS anticholinergic or narcotic other than the index medication, (3) the patient in the control group for trimethobenzamide filled trimethobenzamide, (4) the patient in the sedative hypnotic evaluation filled a BHS sedative hypnotic other than the index medication, (5) the patient reached the end of the 360-day postperiod, or (6) the patient had an AE of interest.

	Pharmacy ingredient costs, medical charges, and total healthcare costs (pharmacy plus medical) were measured for exposed patients versus controls over the 360-day postperiod. Adjusted costs were estimated for the cohorts after controlling for potential confounding factors.

	Propensity scores were estimated using logistic regression analysis. Proportional hazards regression was used to analyze the relative risk of having the target AE for exposed patients versus controls. Patients who did not have the target AE were censored at their end of follow-up.

	Costs were evaluated using t tests for means, Wilcoxon rank sum tests for medians, and generalized linear models (GLMs) with a log link and gamma distribution for adjusted costs. Adjustment variables included in the GLMs were age, sex, health plan type, geographic state, Charlson Comorbidity Index

 (calculated during the preperiod using a method adapted for electronic claims databases

), and preperiod costs. Statistical analysis was performed using SAS version 9.1 (SAS Institute, Inc, Cary, NC). All comparisons were 2-sided with a .05 level of significance.

	Among 37,358 pairs of patients exposed to BHS anticholinergics and controls, baseline characteristics were similar except that exposed patients filled more days supply for anticholinergics during the preperiod (

). Exposed patients also had higher preperiod medical costs than

	Patients exposed to BHS anticholinergics experienced 15.60 cases of delirium or hallucinations per 1000 person-years compared with 15.18 cases per 1000 person-years for controls (

). The risk of delirium or hallucinations was not significantly different between patients exposed to BHS anticholinergics compared with controls (hazard ratio, 1.03; 95% confidence interval [CI], 0.91-1.16).

	Compared with controls, exposed patients had lower adjusted pharmacy costs ($773 vs $796) but higher adjusted medical costs ($17,154 vs $15,214) and higher adjusted total healthcare costs ($18,398 vs $16,482) during the postperiod (

 <.001 for all). These results are given in 

	The 395 patients who met the identification criteria for exposure to BHS narcotics were matched with controls having similar characteristics (Table 3). Exposed patients had higher preperiod medical costs than controls.

	Delirium or hallucinations occurred at a rate of 16.07 cases per 1000 person-years among exposed patients and 10.75 cases per 1000 person-years among controls (Table 4). However, the number of events (5 for exposed patients and 4 for controls) was too few to assess whether there were statistical differences between cohorts.

	While adjusted postperiod pharmacy costs were similar between cohorts, adjusted postperiod medical costs and adjusted postperiod total healthcare costs were significantly higher for exposed patients than for controls ($26,370 vs $20,952 for medical costs and $27,447 vs $22,471 for total healthcare costs; 

 = .03 for both). These results are given in Table 5.

	There were 1085 patients exposed to trimethobenzamide and 1085 controls. Clinical characteristics were well matched between the 2 cohorts (Table 3). Patients exposed to trimethobenzamide had higher preperiod medical costs than controls.

	Four patients exposed to trimethobenzamide and 1 control experienced extrapyramidal effects (4.09 vs 0.93 cases per 1000 person-years) (Table 4). Statistical testing and regression analysis were not performed because the number of events was too few to make meaningful comparisons.

	Compared with controls, exposed patients had similar adjusted pharmacy costs ($907 vs $916; 

 = .86) and significantly higher adjusted medical costs ($27,722 vs $16,647; 

 <.001) and adjusted total healthcare costs ($29,144 vs $18,034; 

	For 13,542 pairs of patients receiving BHS sedative hypnotics and controls, most of the clinical variables were similar; however, the exposed cohort had a higher percentage of patients with a medical condition that may increase the risk of falls or fractures than controls (17.3% vs 16.0%) (Table 3). Exposed patients also had higher preperiod medical costs than controls. Exposed patients had more cases of fall or fracture per 1000 person-years (69.10 vs 56.36) and were significantly more likely to have a fall or fracture than controls (hazard ratio, 1.22; 95% CI, 1.10-1.35; 

 <.001). These results are given in Table 4.

	Adjusted pharmacy costs were similar for patients exposed to BHS sedative hypnotics compared with controls (Table 5). However, exposed patients had significantly higher adjusted medical and total healthcare costs than controls ($20,537 vs $15,835 for adjusted medical costs and $21,807 vs $17,154 for adjusted total healthcare costs; 

	In this evaluation of BHS PIMs categorized as “always avoid” or “rarely appropriate” for the elderly, the risk of developing an AE was different depending on the category of PIM. Patients receiving BHS sedative hypnotics were 22% more likely to have a fall or fracture than controls. In contrast, patients receiving BHS anticholinergics were not found to have a higher risk of delirium or hallucinations than controls. While delirium or hallucinations occurred at a higher rate among patients receiving BHS narcotics compared with controls and extrapyramidal effects occurred at a higher rate among patients receiving trimethobenzamide compared with controls, too few events occurred to assess statistical significance. For all 4 types of BHS PIMs, annual adjusted medical and total healthcare costs were significantly higher for exposed patients than for controls.

	While these results are consistent with other retrospective claims analyses that have shown an association between PIM use in the elderly and increased adverse clinical outcomes

 this study adds to the current literature because it examines patients receiving specific PIMs instead of a heterogeneous mix of patients receiving different types of PIMs. In an era of limited healthcare resources, these results may help MCOs prioritize which specific medications and patients to target for interventions designed to reduce the use of PIMs in the elderly.

	The fact that elderly patients receiving BHS sedative hypnotics had an increased risk of falls or fractures is not surprising because the use of sedative hypnotics has long been implicated as contributing to falls and fractures among the elderly.

 Additional analysis of 790 patients within this study who were exposed to a BHS sedative hypnotic and who had a fall or fracture found that the mean (SD) event-related medical costs per patient over the 3-month period was $10,339 ($22,406) (median, $1821). These findings lend support to the current Medicare Part D design in which certain sedative hypnotics (eg, benzodiazepines) are a benefit exclusion. For non—Part D plans, these results justify the need to conduct interventions to minimize the use of BHS sedative hypnotics among the elderly and help prevent high medical costs associated with falls or fractures.

	With such a large sample size (37,358 pairs), it is surprising that patients receiving BHS anticholinergics did not have increased risk of developing delirium or hallucinations. While delirium or hallucinations would be expected with BHS anticholinergics in theory, perhaps the way that patients actually take these medications in the real-world setting minimizes the number of observed events. Patients may take these medications intermittently or at low doses and, as a result, may never experience such events. Another possible explanation is that delirium or hallucinations may be overlooked by patients or caregivers and may never be reported to the patient’s physician. It is also possible that these events were reported but the provider did not code them in the encounter. Delirium and hallucinations may be underrecognized or undercoded in the claims database.

	Although patients receiving BHS anticholinergics did not seem to have an increased risk of delirium or hallucinations, they had higher total postperiod healthcare costs. The disparity between AEs and healthcare costs for the BHS anticholinergics may be owing to these medications causing other AEs that were not evaluated in this analysis. While this study focused on evaluating the risk of delirium and hallucinations, which was the primary AE of concern as reported in the literature,10 it is possible that BHS anticholinergics could increase the risk of other costly events (eg, falls and fractures) that were not studied within this analysis. Further evaluation of different AE outcomes among patients exposed and not exposed to BHS anticholinergics may help explain some of the factors contributing to increased healthcare costs.

WEB EXCLUSIVE

Clinical and economic outcomes associated with the use of specific potentially inappropriate medications in the elderly were evaluated.

Clinical and Economic Outcomes Associated With Potentially Inappropriate Prescribing in the Elderly

 To examine the relationships among patient characteristics, labor inputs, and improvement in glycosylated hemoglobin (A1C) level in a successful primary care—based diabetes disease management program (DDMP).

 We performed subanalyses to examine the relationships among patient characteristics, labor inputs, and improvement in A1C level within a randomized controlled trial. Control patients received usual care, while intervention patients received usual care plus a comprehensive DDMP.

 The primary outcome was improvement in A1C level over 12 months stratified by intervention status and patient characteristics. Process outcomes included the number of actions or contacts with patients, time spent with patients, and number of glucose medication titrations or additions.

One hundred ninety-three of 217 enrolled patients (88.9%) had complete 12-month followup data. Patients in the intervention group had significantly greater improvement in A1C level than the control group (−2.1% vs −1.2%, 

 = .007). In multivariate analysis, no significant differences were observed in improvement in A1C level when stratified by age, race/ethnicity, income, or insurance status, and no interaction effect was observed between any covariate and intervention status. Among intervention patients, we observed similar labor inputs regardless of age, race/ethnicity, sex, education, or whether goal A1C level was achieved.

 Among intervention patients in a successful DDMP, improvement in A1C level was achieved regardless of age, race/ethnicity, sex, income, education, or insurance status. Labor inputs were similar regardless of age, race/ethnicity, sex, or education and may reflect the nondiscriminatory nature of providing algorithm-based disease management care.

Incorporation of a diabetes disease management program (DDMP) into the primary care setting can aid in addressing diabetes care disparities, reflecting the nondiscriminatory nature of algorithm-based care.

Agency for Healthcare Research and Quality—defined priority populations experience worse quality and access to care than their respective reference groups.

Little is known about the ability of DDMPs to deliver unbiased care. 

Our DDMP demonstrated significant reduction in glycosylated hemoglobin (A1C) level among participants regardless of race/ethnicity, sex, education, or insurance status.

Labor inputs were similar regardless of age, race/ethnicity, sex, education, or whether goal A1C level was achieved.

More than 20 years ago, the US Department of Health and Human Services issued their 

Report of the Secretary’s Task Force on Black and Minority Health

 This study documented many health disparities and sparked efforts to improve the health and healthcare of vulnerable populations and to raise awareness and concern for health disparities research.

 Disparate trends in care across various groups have been noted, including racial/ethnic minorities, as well as women, children, older adults, and other potentially vulnerable populations. Despite this research and advocacy, health disparities persist across a wide spectrum of disease processes, clinical settings, and populations within the United States. According to the 2007 National Healthcare Disparities Report,

 some disparate trends are improving, such as rates of cholesterol control among racial/ethnic minorities; however, many are in major need of improvement, particularly in the areas of diabetes care and chronic disease management.

The Agency for Healthcare Research and Quality (AHRQ) has identified several priority groups that serve as targets for health disparities research and quality improvement. These groups include racial/ethnic minorities, persons with low incomes, women, children, older adults, those with disabilities, and rural inhabitants.

 In general, members of these priority groups along with those who have less education and no private insurance are more likely to receive poorer quality of care and to experience greater barriers in terms of access to care compared with their respective counterparts.

 For example, compared with white patients,

 Hispanic and lower-income patients with diabetes mellitus are less likely to receive the recommended services for diabetes care such as timely measurement of lipid levels and glycosylated hemoglobin (A1C) levels, and black patients have higher rates of many of the complications of diabetes such as admissions for lower extremity amputations.

 Although studies performed by Trivedi et al

 have demonstrated that quality improvement measures can lead to reduced disparities in some diabetes-related processes of care, it has been more difficult to influence outcome measures such as attainment of goal blood glucose control (as measured by A1C level).

 have also suggested that disparities in diabetes care may be related to patient characteristics, provider characteristics, or system-level factors.

Disease management programs have been shown to be successful in improving care for patients with diabetes,

 and there is growing interest in the usefulness of such models in addressing healthcare disparities.

 However, little is known about the specific effect of disease management programs on disparate trends and particularly on their ability to provide unbiased care and to improve patient outcomes equally across priority groups. Results of a successful primary care—delivered algorithm-based diabetes disease management program (DDMP) have been previously reported.

 Evaluation of this program within a randomized controlled trial demonstrated significant improvements in A1C level, blood pressure, aspirin use, patient satisfaction, and other measures.

 The program was also found to require modest labor inputs.

 In the present study, we performed a subanalysis of this program to examine the relationships among patient characteristics, labor inputs, and improvement in A1C level, with a particular focus on evaluating the potential usefulness of programs such as this to influence the care and outcomes for patients considered to be within an AHRQ-defined priority group and for patients with less education or no private insurance.

We performed a randomized controlled trial of a DDMP among 217 patients with type 2 diabetes mellitus and poor glucose control (A1C level ≥8%) in an academic primary care practice. Details of the trial have been previously published.

 Patients in the intervention group received care from their primary care provider supplemented by a DDMP. The program included intensive management from clinical pharmacists and from a diabetes care coordinator who provided diabetes education, applied algorithms for treating glucose level abnormalities and decreasing cardiovascular risk, used an electronic registry to proactively identify patients in need of additional care, and addressed barriers to care. Treatment algorithms provided guidance on when to contact patients and information about titration or addition of new medications for glucose control and cardiovascular risk reduction. Control patients received a one-time disease management session from pharmacists, followed by usual care from their primary care provider. The primary outcome was improvement in A1C level over 12 months. The study was performed and subsequently analyzed with approval from the University of North Carolina at Chapel Hill and Vanderbilt University institutional review boards. Written consent was obtained from all study participants.

To estimate resources required for the DDMP, the study team documented process measures for all intervention patients. These “labor inputs” included the following: (1) minutes spent by the disease management team in direct contact with a patient or in activities related to patient care, (2) number of contacts with a patient, and (3) number of patientspecific glucose medication titrations or additions. These measures captured the labor of the DDMP and did not include any actions taken by the patient’s primary care providers.

All analyses were performed using commercially available statistical software (Stata 8.0; StataCorp LP, College Station, TX), and a 2-sided significance level of 5% was used for statistical inference. Baseline differences between intervention and control groups were determined using Wilcoxon rank sum test for continuous outcome variables and 

 test for categorical outcome variables. We examined the relationship between patient characteristics and improvement in the median A1C level using multivariate quantile regression analysis. Quantile regression modeling was chosen for this subanalysis because we report the median changes for each outcome of interest. The median reporting was required because of additional data stratification and subsequent changes in sample sizes. This should be contrasted with how the program was evaluated in previous publications.

 For each patient characteristic (age, race/ethnicity, sex, education, income, and nsurance status), we performed a separate model that included the study status, the patient characteristic in question, and an interaction term to evaluate the potential modifying effect of intervention status and the patient characteristic in question on change in A1C level. Patient characteristics were dichotomized based on priority criteria of the 

2007 National Healthcare Disparities Report

 (with age dichotomized at 65 years [older adults], race/ethnicity as white vs nonwhite, and household annual income at $10,000). Education was dichotomized at less than high school and insurance as private versus nonprivate. We used Wilcoxon rank sum test to examine the relationship between each of the labor inputs (median time spent with patients, median number of actions or contacts with patients, or median number of glucose medication additions or titrations) and each of the aforementioned patient characteristics. We also used Wilcoxon rank sum test to analyze the relationship between each of 3 labor inputs and attainment of goal A1C level (≤7%) at 12 months of follow-up. Finally, we examined the relationship between each of 3 labor inputs and improvement in the median A1C level using multivariate quantile regression models for each labor input, with each model adjusting for the aforementioned patient characteristics.

From February 2001 to April 2002, 217 patients were enrolled. Complete follow-up data were available for 193 patients (88.9%) at 12 months. Baseline demographics revealed diversity, with substantial sociodemographically vulnerable participants having poor diabetes control. Most nonwhite participants were African American. Intervention and control patients generally had similar characteristics (

). The A1C levels improved significantly more for intervention than  control patients during the trial (

). From baseline to 12 months, control patients had a median 1.2% decrease in A1C level, whereas intervention patients had a median 2.1% decrease (difference, 0.9%; 95% confidence interval, 0.6%-1.3%; 

There was no significant difference in improvement in the median A1C level when stratified by age, race/ethnicity, sex, income, or insurance status (Figure and 

). There was slightly greater improvement in the median A1C level among women compared with men (−1.2% vs −0.5%) and among those with less than a high school education compared with those with at least a high school education (−1.1% vs −0.8%). Results of multivariate quantile regression analysis did not reveal any significant interaction effect on improvement in A1C level between participants’ intervention status and each of the respective patient characteristics of interest.

Among intervention patients, there were no differences in labor inputs (median time spent with patients, median number of actions or contacts with patients, or median number of glucose medication titrations or additions) when stratified by age, race/ethnicity, sex, or education (

). The disease management team had significantly more contacts and spent more time with patients having a household annual income less than $10,000 or no private insurance. There were no observed differences in each of 3 labor inputs when stratified by patients obtaining the goal A1C level of 7% or less at the end of the trial. In multivariate analysis, there was no significant relationship between each of 3 labor inputs and improvement in the median A1C level when adjusted for age, race/ethnicity, sex, education, income, and insurance status.

In this subanalysis of our DDMP, a successful reduction in blood glucose (A1C) level was achieved irrespective of race/ethnicity and other important sociodemographic factors. Our findings were not influenced by the observed differences at baseline in terms of age and insurance status between the control and intervention groups. Also, none of the priority sociodemographic factors analyzed had a significant modifying effect with DDMP participation on improvement in A1C level. The time spent by the disease management team for intervention patients was provided similarly regardless of age, race/ethnicity, sex, or education. Improvement in A1C level did not require additional labor inputs compared with patients who did not show significant improvement when adjusted for patient characteristics. Our findings suggest that DDMPs may serve as a tool for providing equal quality of care and for addressing the current disparate trends in care often experienced by many of the AHRQ priority groups with diabetes.

The etiology of racial/ethnic and other socioeconomic disparities in diabetes and other chronic illness care is likely multifactorial. Disparities in provider behavior may be an important component. Sequist and colleagues

 recently conducted a provocative study in which they reported on variations in diabetes-related outcomes at the level of individual physicians. They examined a 14-site primary care group practice that used an integrated electronic medical record system and found significant within-physician variation between the percentages of white and black patients who were able to achieve recommended levels of glycemic, blood pressure, and lipid control. Patient sociodemographic factors accounted for a relative range of 13% to 38% of the observed racial/ethnic differences, whereas within-physician effects (ie, racial/ethnic disparities observed within the same physician panel) accounted for 68% to 75% of the observed differences. A study by Schulman and colleagues

 that surveyed 720 physicians eliciting recommendations regarding cardiac catheterization found that sex and race/ethnicity independently influenced how physicians managed their patients, with many physicians surveyed being less likely to recommend needed catheterizations among women or nonwhite patients.

One reason for the success of our program in improving care regardless of socioeconomic status may be the ability of disease management to eliminate potential disparities in provider-level behavior. The nondiscriminatory nature of providing algorithm-based care may serve as a means by which health systems could ensure delivery of effective and unbiased care, while accounting for and addressing the frequent special needs that arise when caring for traditionally vulnerable patient populations. As we observed in our program, this could be achieved through management practices that allowed for patient-centered approaches to care as individualized needs become evident (described herein). The use of a team approach to care with midlevel providers and a care coordinator working together with primary care physicians may also help to prevent provider-level disparities.

Our disease management team spent slightly more time among patients with lower income or no private insurance. This additional time may have been related to addressing economic barriers to attaining medications and other clinical care. More research is needed to better understand the resources required to achieve successful outcomes across all priority groups. In the present analysis, we also found that women and participants with less education showed slightly greater benefit from the intervention. Greater improvement in glycemic control among patients with less education and among women is more likely to reflect differences by intervention status rather than by education or sex. This was supported by the absence of an interaction effect between these variables and patient status with A1C level and should be contrasted with the previously reported analysis showing that patients having lower literacy had significantly more improvement in glucose control compared with those having higher literacy.

 Significant improvement in glycemic control based on literacy level may have been related to the fact that our intervention included several approaches to address literacy, including the use of simplified communication, the “teach back technique” for confirming patient understanding, lowerliteracy materials, and a care coordinator who confronted many social barriers.

 An approach in which individualized physician care of patients with diabetes is supplemented with an adequately tailored and focused chronic DDMP may be an important approach in eliminating disparate trends.

Several limitations were inherent in our study. Our sample size was small, and we may have lacked adequate power to detect differences based on patient characteristics alone. In addition, our documented labor inputs characterize time spent with patients by the disease management team and do not include time spent by primary care providers or the patients or utilization of clinical services or other resources. Furthermore, the time spent by the management team included some actions that were considered “outside” of the algorithm, such as discussion of certain medication changes with providers. As previously reported,

 discussion with providers was required for 112 of 500 drug titrations and for 233 of 348 new drug additions. The time required by such actions would be difficult to predict if our program was repeated. The generalizability of our findings is also reduced by the fact that our study was conducted at a single-site academic institution and did not have adequate representation of several racial/ethnic minorities such as Latinos. Furthermore, our participants at entry had poorly controlled diabetes, and it is arguable that the magnitude of our findings would be lessened if repeated in a population with better baseline control. Finally, although we believe that the algorithm-based care was a major factor contributing to the balanced improvement in A1C level across the priority groups, it remains a possibility that our findings were influenced by regression toward the mean or by the fact that our study did not find many clinically significant racial/ethnic disparities on baseline evaluation of our study population. One may argue that, if such disparities were in fact observed, our results may have differed in their magnitude. There may be other factors associated with the intervention that could have contributed to the similar improvements. For example, we focused on patient self-management skills and provided assistance with obtaining medications. These efforts combined with aggressive treatment using algorithmbased care might have had synergistic effects among certain individuals.

We do not go so far as to suggest that our program has eliminated racial/ethnic disparities but are confident that none were created among participants who received our intervention. A disease management program integrated into primary care can be a successful model for improving care among vulnerable patients with significant barriers. Emphasis on algorithm-based treatment and on team approaches to care can result in improvements regardless of race/ethnicity, sex, education, or other characteristics and represents an important tool for addressing healthcare disparities. Future research about the role of disease management to address health disparities in diabetes and other chronic illnesses is clearly indicated.

: From the Division of General Internal Medicine (ROW), Meharry Medical College, Nashville, TN; the Division of General Medicine and Epidemiology (DAD, RMM, MPP), and the Division of Pharmacy Practice and Experiential Education (RMM), University of North Carolina at Chapel Hill; and the Division of General Internal Medicine and Public Health (CYO, RLR), Vanderbilt University Medical Center, Nashville, TN.

This study was completed with support from the Robert Wood Johnson Clinical Scholars Program, the University of North Carolina (UNC) Program on Health Outcomes, the UNC Division of General Internal Medicine, the Vanderbilt Center for Health Services Research, and the Vanderbilt Diabetes Research and Training Center (P60 DK020593), by grant NCRR NIH 5R25RR017577 from Meharry CRECD program (Dr White), and by grant  R25RR017577-08S1 from ARA Supplement (Dr White).

Dr Rothman reports serving on the Clear Health Communication Fellowship Advisory Board for Pfizer, Inc. The other authors (ROW, DAD, RMM, CYO, MPP) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (DAD, RMM, MPP, RLR); acquisition of data (DAD, RMM, MPP); analysis and interpretation of data (ROW, DAD, MPP, RLR); drafting of the manuscript (ROW, CYO, RLR); critical revision of the manuscript for important intellectual content (ROW, DAD, CYO, MPP, RLR); statistical analysis (ROW, RLR); provision of study materials or patients (RLR); obtaining funding (RLR); administrative, technical, or logistic support (RMM, MPP); and supervision (RLR).

Richard O. White, MD, Division of General Internal Medicine, Meharry Medical College, 1005 Dr DB Todd Jr Blvd, Nashville, TN 37208. E-mail: rwhite@mmc.edu.

1. Task Force on Black and Minority Health.

Report of the Secretary’s Task Force on Black and Minority Health.

 Washington, DC: US Dept of Health and Human Services; 1985.

 Unequal treatment: the Institute of Medicine report and its public health implications. 

The IOM report Unequal Treatment: lessons for clinical practice. 

 Report of the Secretary’s Task Force on Black and Minority Health: a summary and a presentation of health data with regard to blacks. 

5. Agency for Healthcare Research and Quality

2007 National Healthcare Disparities Report.

 Rockville, MD: US Dept of Health and Human Services; 2008.

 Unequal treatment: confronting racial and ethnic disparities in health care.

 Primary language and receipt of recommended health care among Hispanics in the United States. 

 Diabetes mellitus and nontraumatic lower extremity amputation in black and white Americans: the National Health and Nutrition Examination Survey Epidemiologic Follow-up Study, 1971-1992. 

9. Trivedi AN, Zaslavsky AM, Schneider EC, Ayanian JZ.

 Trends in the quality of care and racial disparities in Medicare managed care. 

10. Chin MH, Cook S, Drum ML, et al; Midwest Cluster Health Disparities Collaborative.

 Improving diabetes care in Midwest community health centers with the Health Disparities Collaborative [published correction appears in 

11. Sequist TD, Adams A, Zhang F, Ross-Degnan D, Ayanian JZ.

 Effect of quality improvement on racial disparities in diabetes care. 

Diabetes health disparities: a systematic review of health care interventions. 

13. Gaede P, Vedel P, Larsen N, Jensen GV, Parving HH, Pedersen O.

 Multifactorial intervention and cardiovascular disease in patients with type 2 diabetes. 

14. Knight K, Badamgarav E, Henning JM, et al.

 A systematic review of diabetes disease management programs. 

15. Norris SL, Nichols PJ, Caspersen CJ, et al.

 The effectiveness of disease and case management for people with diabetes: a systematic review. 

16. Rothman RL, Malone R, Bryant B, et al. 

A randomized trial of a primary care—based disease management program to improve cardiovascular risk factors and glycated hemoglobin levels in patients with diabetes. 

17. Wagner EH, Glasgow RE, Davis C, et al.

 Quality improvement in chronic illness care: a collaborative approach. 

 Exploring disease management programs for diabetes mellitus: proposal of a novel hybrid model. 

Mandated diabetes registries will benefit persons with diabetes. 

20. Rothman RL, DeWalt DA, Malone R, et al.

 Influence of patient literacy on the effectiveness of a primary care—based diabetes disease management program. 

 Labor characteristics and program costs of a successful diabetes disease management program. 

22. Sequist TD, Fitzmaurice GM, Marshall R, Shaykevich S, Safran DG, Ayanian JZ

. Physician performance and racial disparities in diabetes mellitus care. 

23. Schulman KA, Berlin JA, Harless W, et al. 

The effect of race and sex on physicians’ recommendations for cardiac catheterization [published correction appears in 

MANAGERIAL

A subanalysis of a successful algorithm-driven primary care–based diabetes disease management program examines the relationships among patient characteristics, labor inputs, and improvement in A1C level.

Leveling the Field: Addressing Health Disparities Through Diabetes Disease Management

To compare the mortality rate of patients with type 2 diabetes who were enrolled in the German diabetes disease management program (DMP) with the mortality rate of those who were not enrolled.

 This observational study was part of the ELSID study (Evaluation of a Large Scale Implementation of disease management programs) in Germany.

 Participants had type 2 diabetes and were either enrolled or not enrolled in the DMP. The DMP provides systems-based, multifaceted, and patient-centered interventions. To reduce imbalances between the groups, a matched sample was created using sex, age, retirement status, federal state, pharmacy-based cost groups, and diagnostic-cost groups as matching criteria. Cox proportional hazards regression model and the Kaplan-Meier method were used to assess overall mortality. The observation period was 3 years beginning on January 1, 2006.

 A total of 11,079 patients were included in the analysis. As of January 1, 2006, 2300 patients were enrolled in the DMP and 8779 were receiving routine care. There were 1927 matched pairs of patients in the DMP group and the non-DMP group. The overall mortality rate was 11.3% in the DMP and 14.4% in the non-DMP group (log-rank test 

 We found an association between participation in the  German diabetes DMP and reduced mortality. This reduced mortality cannot be attributed directly to the DMP. However, further research should evaluate whether a primary care—based DMP contributes to increased life expectancy in patients with diabetes.

Patients with type 2 diabetes who participate in the German primary care—based diabetes disease management program (DMP) receive systems-based, multifaceted, and patientcentered interventions.

Participation in the DMP was associated with reduced mortality after 3 years.

Although this reduced mortality cannot be attributed directly to the DMP, extension of the approach within primary care may contribute to increased life expectancy for patients with diabetes.

Management of diabetes and associated problems is one of the main challenges for healthcare systems worldwide. Patients with type 2 diabetes are likely to have multiple conditions (eg, coronary heart disease), and their life expectancy is shorter than that of individuals without diabetes.

 In the United States, more than 60% of patients with diabetes die from cardiovascular causes.

Interventions to improve the health status and life expectancy of patients with diabetes should focus on the patients and their individual problems as well as on a restructuring of care. Complex interventions that go beyond the adjustment of clinical parameters are required. The patient needs to be considered holistically in light of his or her individual circumstances. According to the literature, multifaceted interventions by multiprofessional teams with additional patient education and the enhancement of the role of practice nurses have been associated with improved diabetes care as well as patient outcomes.

Diabetes disease management programs (DMPs) delivered in primary care settings were introduced into Germany nationwide in 2003.

 Using compulsory requirements determined on behalf of the German Ministry of Health, sickness funds arrange contracts with primary care physicians. Participation for doctors and patients is voluntary, but participating doctors are obliged to keep within the conditions of the program. Primary care physicians and the wider practice teams, mostly in small- to medium-size practices in Germany, have a central role in performing and coordinating the provision of care to enrolled patients with diabetes.

 The German DMPs have been designed to improve the quality of care for patients with chronic diseases, reduce complications, improve patientoriented outcomes, and lower costs. Currently about 2.7 million patients with type 2 diabetes are enrolled.

The diabetes DMP includes the implementation and audit of evidence- based clinical guidelines using quality indicators and quality assurance measures, with feedback to participants on their level of performance. It also includes regular recalls for patients and shared individual goal setting by the patient and the physician, with consideration of the individual circumstances and risk profiles. This shared goal setting is based on emphasizing both coordination and continuity of care and the physician’s knowledge of each patient.

 Patients are offered lifestyle advice with the aim of enabling them to achieve behavioral changes in diet and physical activity in support of their own self-management. The doctor-patient relationship is strengthened within the program, and continuity of care is guaranteed as patients are obliged to visit their physician regularly, either quarterly or semiannually. There also is an enhanced role for doctors’ assistants within the DMP. Patients enrolled in the program are more likely to receive diabetes care according to the chronic care model

Statutory nationwide evaluation of all DMPs is mandatory for all sickness funds. However, patients enrolled in a DMP are not compared with patients who are not. Although reducing mortality is a patient-relevant outcome of interventions and new models in diabetes care, it often is not measured in trials.

Patients with type 2 diabetes in a German disease management program had a lower mortality rate after 3 years than those not in the program.

January 2010

Clinical Guidelines and Performance Measures

Clinical and Economic Outcomes Associated With Potentially Inappropriate Prescribing in the Elderly

Leveling the Field: Addressing Health Disparities Through Diabetes Disease Management

Is There a Survival Benefit Within a German Primary Care-Based Disease Management Program?

Physician Utilization by Insurance Type Among Youth With Type 2 Diabetes

A Pay-for-Performance Program for Diabetes Care in Taiwan: A Preliminary Assessment

Mail-Order Pharmacy Use and Adherence to Diabetes-Related Medications

New Diabetes HEDIS Blood Pressure Quality Measure: Potential for Overtreatment

'All-or-None' (Bundled) Process and Outcome Indicators of Diabetes Care

Increasing Copayments and Adherence to Diabetes, Hypertension, and Hyperlipidemic Medications

The Effect of a PPO Pay-for-Performance Program on Patients With Diabetes