Health plans are increasingly offering payment incentives to motivate providers to strive for improving the quality of patient care based on established criteria such as those set forth by the National Committee for Quality Assurance (NCQA).1,2
The NCQA accredits healthcare organizations and manages the Health Plan Employer Data and Information Set, a tool for measuring performance in key areas, including diabetes management.3 In partnership with the American Diabetes Association (ADA),4 which has established specific treatment targets for long-term glycemic control in patients with diabetes, the NCQA supports high-quality care for patients through the Diabetes Recognition Program (DRP). The DRP recognizes physicians and practices that are providing highquality diabetes care as determined by 10 key measures, which include hemoglobin (Hb)A1c control, blood pressure (BP) control, low-density lipoprotein cholesterol (LDL-C) control, eye examinations, foot examinations, nephropathy assessment, and smoking cessation advice or treatment.5 For a physician to achieve DRP recognition, ≥40% of patients must meet the recommended HbA1c goal, ≥36% of patients must meet the recommended LDL-C goal, and ≥35% of patients must meet the recommended BP goal.3
Building on previous research,3,6,7 the present study was designed to examine the relationship between treatment by a DRP-certified physician and health-related outcomes for diabetic patients, including prescription utilization, medical resource use, and healthcare expenditures, and to compare the impact of treatment by DRP-certified physicians versus non–DRP-certified physicians on patient outcomes.
Methods
This was a retrospective claims-based analysis using medical, pharmacy, and enrollment information from a large US database of commercially insured patients. This administrative healthcare claims database included electronic pharmacy and medical claims and enrollment data for commercially insured patients identified between January 1, 2004, and December 31, 2007 (identification period), from a large US managed care provider affiliated with OptumInsight. For 2007, data were available relating to approximately 14 million individuals with medical and pharmacy benefit coverage. The plan provides fully insured coverage for professional (eg, physician), facility (eg, hospital), and outpatient prescription medication services. Because this study involved analysis of preexisting, deidentified data, it was exempt from Institutional Review Board approval.
Physician Samples
All physicians in the database between January 1, 2007, and November 30, 2007, who were treating adult patients with type 2 diabetes (Appendix A) whose claims data were also included in the database were identified for inclusion in this analysis. Treating physicians included family physicians, internists, and endocrinologists identified from a large and diverse group of physician specialties at the primary, secondary, and tertiary care levels. Our rationale in selecting physicians was based on identifying physician specialties that were most likely to be in charge of managing our targeted patient population. The identification process revealed that the majority of the physicians were either family practice physicians or internists. To be included in the study sample, physicians in the database were required to be either DRP-certified by the NCQA or not DRP-certified by the NCQA in any of the existing recognition programs as of May 2009 (see http://recognition.ncqa.org/index.aspx).
Each of the DRP-certified physicians was matched directly in a 1:1 ratio to a non–DRP-certified physician based on 4 variables: physician specialty, state of physician practice, size of potential patient population (defined as the population density of the postal zip code in which the physician practices), and number of nonpediatric patients with type 2 diabetes (similar number ± 3 patients) in the analytic period.
Patient Samples
Study patients were commercial enrollees with evidence of type 2 diabetes treated by a DRP-certified or non–DRP-certified physician identified in the database (8011 patients; 3836 treated by a DRP-certified physician and 4175 treated by a non–DRP-certified physician). The index date was defined as the date of the patient’s first claim for an outpatient diagnosis of diabetes (International Classification of Diseases, Ninth Revision [ICD-9]: 249.xx, 250.xx, 357.2, 362.0x, 366.41, 648.0x, 996.57, V45.85, V53.91, V58.67) by a DRP-certified or non–DRP-certified physician (ie, index physician) during the study period.
Patients were included if they were continuously eligible during the 6-month preindex and 12-month postindex periods. Furthermore, patients could not have had any visits with an index physician or with any DRPcertified physician during the preindex period. Patients with evidence of gestational diabetes or polycystic ovarian syndrome were excluded from the analysis.
Medication Use
Medication use was defined as the number of prescriptions per patient for oral antihyperglycemic, statin, or antihypertensive medications. These medication classes were selected because of their impact on the biometric measures that are included in the DRP certification requirements.
Medical Resource Utilization
The numbers of diabetes-related office visits, outpatient visits, emergency department visits, and inpatient admissions per patient were calculated. Diabetes-related visits were defined as visits with evidence of diabetes or diabetes-related complications (see Appendix B for ICD-9 codes), such as hypertension, dyslipidemia, ischemic heart disease, atherosclerosis, peripheral vascular disease, aortic aneurysm, congestive heart failure, myocardial infarction, stroke (with and without transient ischemic attack), coronary artery bypass graft surgery, angioplasty, nephropathy, neuropathy, retinopathy, foot ulcers, ketoacidosis, skin infection/skin ulcers, and lower-extremity amputations.
Expenditures
Expenditures were defined as combined health plan–paid and patient-paid dollar amounts per patient for medical, pharmacy, ambulatory, emergency, outpatient, and inpatient utilization that were diabetes-related, and all-cause total expenditures. Diabetes-related pharmacy utilization included oral and injectable hypoglycemic drugs. Expenditures (based on annualized US dollar estimates from 2006-2008) were adjusted using the annual medical care component of the Consumer Price Index to adjust for inflation between 2006 and 2008.
Statistical Analyses
All study variables, including preindex and postindex utilization, clinical, and expenditure measures, were first analyzed descriptively. Per-patient per-month (PPPM) measures were used to assist with the interpretation of within-group utilization and expenditure changes resulting from the difference in follow-up for the preindex (6 months) and postindex (12 months) periods.
Multivariate analyses were performed to compare the impact of specified independent variables between the DRP-certified and non–DRP-certified physician groups on utilization and expenditure measures on a per-patient per-year (PPPY) basis. PPPY was selected because the multivariate analysis focused on the 12-month postindex period only.
Differences in oral antihyperglycemic, antihypertensive, and statin drug use, as well as diabetes-related medical utilization between the DRP-certified and non– DRP-certified physicians were determined using Poisson regression. Predicted means for medication use and healthcare utilization counts were calculated using the parameters from the Poisson model and the mean of the independent variables.
Expenditures were examined using a generalized linear model (GLM) with a gamma distribution, with log link used to assess the incremental costs associated with DRP designation. The gamma distribution and log link account for the skewed distribution of costs. For dichotomous independent variables (eg, preindex hypertension), coefficients from the GLM specification for patients with nonzero costs represent the ratio of expected costs in one cohort versus another. For continuous independent variables (eg, age), coefficients from the GLM specification for patients with nonzero costs represent the ratio of expected costs for each unit increase (ie, per year of age). This method avoids potential difficulties introduced by transformation (eg, calculating the log of the costs) and retransformation of the dependent variable. 7 Predicted costs were calculated using the parameters from the GLM model and the mean of the independent variables.
All utilization and expenditure models were adjusted for age, gender, and preindex Deyo-Charlson-Quan comorbidity score.8 Preindex oral and injectable antihyperglycemic agent use, preindex diabetes diagnosis, and preindex all-cause total expenditures were also included in the models for postindex oral antihyperglycemic agent use.
Similar preindex disease-specific covariates and preindex all-cause total expenditures were also added to the models for postindex antihypertensive and statin drug use. Preindex all-cause total healthcare expenditures were also included in the diabetes-related postindex medical utilization models for office visits, outpatient visits, emergency department visits, and inpatient visits. Preindex diabetes-related office visits, emergency department visits, and inpatient visits were also included in the diabetes-related expenditure model, whereas preindex oral antihyperglycemic agent use, as well as preindex all-cause office visits, emergency department visits, and inpatient visits were also included in the allcause total expenditure model.
Results
Demographic Characteristics
The study sample consisted of 8011 patients (3836 treated by a DRP-certified physician and 4175 treated by a non–DRP-certified physician). Most of the patients resided in either the Midwest (44.26%) or South (41.60%). Males outnumbered females (53.5% vs 46.5%), and the mean (± standard deviation [SD]) age of the study population was 54.6 (± 11.05) years. The prevalence of comorbidities was high: 45% of patients had hypertension; 46% had diabetes (ie, they were newly diagnosed with diabetes at the time of study initiation and had not been diagnosed with diabetes during the preindex period); and 43% had dyslipidemia in the preindex period. The Deyo-Charlson-Quan comorbidity score (mean ± SD) was significantly (P = .01) higher for patients treated by the non–DRP-certified group (0.98 ± 1.36) than for those treated by the DRP-certified group (0.91 ± 1.24; Table 1).
No significant differences existed between patients treated by DRP-certified physicians and those treated by non–DRP-certified physicians with regard to gender, age, geographic distribution, or presence of preindex hypertension, diabetes, or dyslipidemia (Table 1).
Medication Use
Although there were no significant differences in preindex use of oral antihyperglycemic agents, antihypertensive agents, and statin medications (Table 2), univariate analysis showed that patients managed by DRPcertified physicians were more likely to receive prescriptions for oral antihyperglycemic agents than those managed by non–DRP-certified physicians (Table 3; mean prescriptions PPPM, 0.49 vs 0.46, respectively; P = .02) and statins (mean PPPM, 0.24 vs 0.22, respectively; P = .005) in the postindex period.
Medical Resource Utilization
Univariate analysis showed that the 2 cohorts did not significantly differ with regard to diabetes-related office visits, outpatient visits, emergency department visits, or inpatient visits during the preindex period (Table 2). However, the DRP-certified cohort had fewer emergency department visits (Table 3; mean PPPM, 0.003 vs 0.006, respectively; P <.001) and inpatient visits than the non–DRP-certified cohort (mean PPPM, 0.007 vs 0.008, respectively; P = .008), and greater use of office visits (mean PPPM, 0.388 vs 0.372, respectively; P = .03) during the postindex period.
Expenditures
Preindex expenditures did not differ significantly between the 2 cohorts (Table 2), but the postindex diabetes- related expenditures and all-cause total expenditures were significantly higher for the non–DRP-certified cohort (P = .02 and P = .02, respectively; Table 3). Specifically, diabetes-related expenditures (mean ± SD) were $241.40 ± $960.35 PPPM for the DRP-certified cohort and $315.04 ± $1693.25 PPPM for the non– DRP-certified cohort.
Multivariate analyses (Table 4) showed that patients treated by DRP-certified physicians had greater oral antihyperglycemic agent use (mean prescriptions PPPY, 5.84 vs 5.52, respectively; P <.001), lower antihypertensive agent use (mean PPPY, 7.46 vs 7.60, respectively; P = .02), and greater statin drug use (mean PPPY, 2.81 vs 2.68, respectively; P = .003) compared with the non– DRP-certified group. Patients treated by DRP-certified physicians had more diabetes-related office visits (mean PPPY, 4.69 vs 4.44, respectively; P <.001) and outpatient visits (mean PPPY, 0.93 vs 0.85, respectively; P <.001) visits. By contrast, patients treated by noncertified physicians had more diabetes-related emergency department visits (mean PPPY, 0.07 vs 0.04, respectively; P <.001) and inpatient visits (mean PPPY, 0.10 vs 0.08, respectively; P = .02).
Patients managed by non–DRP-certified physicians had greater diabetes-related expenditures than those managed by DRP-certified physicians (mean costs PPPY, $4097 vs $3424, respectively; P = .03). The mean allcause total expenditures for patients treated by DRPcertified physicians was not significantly lower (mean PPPY, $10,627 vs $11,221, respectively; P = .22) than for patients treated by non–DRP-certified physicians.
Discussion
Under pay-for-performance models in which financial incentives have been linked to the provision of care, healthcare quality has typically been measured using process-of-care and outcomes-based measures.9 Diabetes has been a target for payment reform models for primary care, because of its high prevalence among primary care patients, the presence of a recognized set of clinical practice guidelines, and the availability of clinical markers used to measure improved glycemic control.
The literature describing the use of retrospective claims analyses to study outcomes of care in pay-forperformance models for diabetes patients is limited. The goal of this study was to compare outcome measures between DRP-certified physicians and non–DRPcertified physicians.
The study revealed several key findings. First, differences in medication use were found between the DRPcertified and non–DRP-certified groups. Oral antihyperglycemic agent use and statin drug use were higher in the DRP-certified group compared with the noncertified group. This finding suggests that DRP-certified physicians prescribe guideline-recommended medications more frequently than non–DRP-certified physicians. However, antihypertensive agent use was lower in the DRP-certified group, suggesting that this is an area where DRP-certified physicians must improve their adherence to ADA standards of care—specifically, the guideline recommending that individuals with diabetes be prescribed angiotensin-converting enzyme inhibitors or angiotensin II receptor blockers for their renal protective effects and to treat hypertension.4
Second, the most compelling finding was that medical utilization among patients managed by a DRP-certified physician was lower for emergency department visits and inpatient visits but higher for office and outpatient visits compared with patients managed by non– DRP-certified physicians.
Emergency department and inpatient visits are high cost drivers and have been a primary focus of outcomebased measures in pay-for-performance models.9 In a nationwide evaluation of pay-for-performance patientcentered medical home models among UnitedHealthcare members with diabetes and other health conditions, results showed a 29% reduction in emergency department visits, 11% fewer inpatient visits, and 6% fewer office visits at the end of the pilot programs.10
The direction of the results of the present study are similar with regard to emergency department and inpatient visits, and are encouraging. However, the finding that the Deyo-Charlson-Quan comorbidity score was significantly lower for patients managed by the DRP-certified group than for those managed by the non–DRPcertified group may suggest that patients in the latter group had greater severity of illness, which could have accounted for their higher healthcare resource utilization. Although the nature of the reductions in emergency department and inpatient visits is unclear (eg, some of these visits might have been avoided as a result of greater primary care intervention), the trends are suggestive of more positive outcomes for DRP-certified physicians with regard to these high cost drivers.
Third, we found that diabetes-related expenditures were lower in the DRP-certified cohort than in the non–DRP-certified cohort. This finding is consistent with the reduction in emergency department and inpatient visits that was observed for the DRP-certified group and appears to provide initial confirmation that DRP-certified physicians could bring about savings in total healthcare expenditures compared with non– DRP-certified physicians.
Limitations
Although these results suggest that improvement in healthcare utilization can occur for patients managed by DRP-certified physicians, retrospective analyses are characterized by inherent limitations, namely, the uncontrolled structure of the investigation and the lack of clinical validation. Furthermore, interpretation of these findings must take into account some of the limitations associated with claims data, including coding variation between providers, time lag between receipt of service and claims processing, and missing data—especially in fields not relevant to reimbursement.
Claims are designed for purposes of payment, not research. The degree to which claims data can accurately capture an individual’s medical history is limited. Furthermore, the data are subject to possible coding errors, coding for the purpose of rule-out rather than actual disease, and undercoding. Although pharmacy claims provide information on prescriptions that were filled, medications may not have been taken as prescribed. Pharmacy claims also do not reflect prescriptions obtained outside of the plan (eg, physician samples). Most important, patients’ severity of disease and their individual clinical need for tests and treatment cannot be measured using claims data, and therefore, are not taken into account when using performance-based metrics.11
Other limitations specific to this study must also be considered. This study did not match patient for patient between cohorts (only physician cohorts were matched). Although multivariate analyses were conducted, unobservable confounding factors were not accounted for, such as disease severity and length of time with illness (ie, the longer the duration of illness, the more likely patients are to experience diabetes-related complications, which have an obvious effect on outcomes).
In addition, our physician sample was diverse rather than uniform. The length of time physicians have been in practice could not be assessed from claims data or from the DRP registry; however, because the development of the DRP stemmed from the original Bridges to Excellence 2001 pilot program, there may be a greater likelihood that physicians who have been in practice longer are DRP-certified. Therefore, physician experience may affect outcomes.
It was not discernible from the claims data whether providers in either group used the services of a certified diabetes educator in their practice or whether patients made any visits to certified diabetes educators. Patient consultation with a certified diabetes educator would have had the potential to influence our results. Furthermore, multivariate analyses were not conducted for measures that involved small patient sample sizes.
Other variables that may have been better suited for inclusion than the ones used in this study, such as disease severity and medication adherence, may have offered more insight into the differences between the 2 patient groups. In addition, the data used for this study came from a commercially insured managed care population, and may not be applicable to patients in non–managed care settings or to Medicare and Medicaid populations.
Most important, misclassification of providers in the noncertified cohort may have occurred, because this study relied on matching lists of physicians in the health plan and corresponding NCQA recognition programs. Therefore, the noncertified cohort may have included physicians who were in the process of seeking DRP certification, which may explain some of the comparable clinical and resource use findings— although this would have biased against the significant cost and treatment findings between the groups. This was addressed by selecting physician cohorts in the database with the most recent data (2007-2008) to when the NCQA lists were obtained (beginning of 2009). Furthermore, it was not possible to classify physicians who provided quality care but did not have DRP certification. This would also have biased against demonstrating any differences.
Conclusions
This study builds on previous research evaluating the effect of offering incentives to providers—through payments or certified “recognition” of improvements in clinical quality—on patient outcomes. It represents an important contribution to the literature, because it is one of the first studies using claims data from a large US database of commercially insured patients to examine the relationship between treatment by DRP-certified physicians and health-related outcomes for patients with type 2 diabetes. These outcomes include prescription utilization, medical resource use, healthcare expenditures, and clinical markers. Our findings provide insights on how to leverage multiple data sources for optimal provider performance measurement.
An accurate assessment of the impact of provider certification on patient outcomes may require combining the measures derived from claims data sources with more detailed data from electronic medical records to better explain and understand the results. Regardless of the factors that contributed to these study results, these findings suggest a potential advantage in expenditures associated with NCQA recognition of DRP-certified physicians, with room for improvement in medication use among patients managed by both DRP-certified and non–DRPcertified physicians.
References
- National Committee for Quality Assurance. Integrated Healthcare Association California pay for performance program: 2006 P4P guidelines. 2005 MY clinical domain. www.allhealth.org/publications/pub_4.pdf. Accessed May 2, 2011.
- Institute of Medicine of the National Academies. Rewarding provider performance: aligning incentives in Medicare. Washington, DC: National Academies Press; 2007.
- Ragucci KR, Fermo JD, Wessell AM, Chumney EC. Effectiveness of pharmacistadministered diabetes mellitus education and management services. Pharmacotherapy. 2005;12:1809-1816.
- American Diabetes Association. Standards of medical care in diabetes—2011. Diabetes Care. 2011;34(suppl 1):S11-S61.
- National Committee for Quality Assurance. Diabetes Recognition Program. www.ncqa.org/tabid/139/Default.aspx. Accessed May 2, 2011.
- Oglesby AK, Secnik K, Barron J, et al. The association between diabetes related medical costs and glycemic control: a retrospective analysis. Cost Eff Resour Alloc. 2006;4:1.
- Manning WG. The logged dependent variable, heteroscedasticity, and the retransformation problem. J Health Econ. 1998;17:283-295.
- Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43:1130-1139.
- Rosenthal MB. Beyond pay for performance—emerging models for provider-payment reform. N Engl J Med. 2008;59:1197-1200.
- Ho S. AHRQ 2009 Annual Conference Research to Reform—Achieving Health System Change. Slide presentation from the AHRQ 2009 annual conference. www.ahrq.gov/about/annualconf09/ho.htm. Accessed May 2, 2011.
- Walter LC, Davidowitz NP, Heineken PA, Covinsky KE. Pitfalls of converting practice guidelines into quality measures: lessons learned from a VA performance measure. JAMA. 2004;291:2466-2470.