Measurement of glycemic control in diabetic patients—an evaluation of risk adjustment using a primary care registry

Corresponding author: Jay H. Shubrook, DO, Ohio University Heritage College of Osteopathic Medicine, Grosvenor Hall, Athens, OH 45701.

E-mail address: shubrook@ohio.edu.

This article was designed to be viewed and distributed as a PDF. Please download the PDF for easiest reading.

The role that glycemic control plays in the reduction of complications from diabetes mellitus alone, or in combina- tion with lipid and blood pressure control, has been well established in type 2 diabetes over the past decade.1,2 Stud- ies show a 1% reduction in hemoglobin A1c has been associated with a 30% reduction in the rate of microvascular complications.1,3 Despite having the knowledge of these benefits, the rate of glycemic control nationally has remained unchanged from 1988 to 2002. However, the pro- portion of diabetics with a hemoglobin A1c between 6% and 8% increased from 34.2% to 47.0% during this time frame.4 Because of the disparity between current treatment recommendations and actual glycemic control of diabetes mellitus, payers in the United States have developed pro- grams linking payment to specific levels of glycemic con- trol. Programs such as the National Committee for Quality Assurance, Bridges to Excellence Program, and the Center for Medicaid and Medicare Services Physician Quality Re- porting Initiative are linking payment to reporting levels of glycemic control.5,6 Because these payment programs use unadjusted rates of control, they implicitly infer that control of hemoglobin A1c is fully attributable to the physician or practice. In clinical practice, however, physicians acknowl- edge that many factors contribute to glycemic goal achieve- ment including patient and system factors. The same issues have been raised for hospital performance measures over the past two decades. Research has resulted in a consensus that outcome performance measures—that quantify the re- sult of the interaction between health care and patients— need to be risk-adjusted for factors the patient brings to the interaction.7 As a result, most hospital performance reports or internal hospital outcomes measures are currently risk adjusted.8-10 As ambulatory measurements of diabetes man- agement increase, specifically those that focus on interme- diate outcomes such as glycemic, blood pressure, or lipemic control, it becomes more important to evaluate methods of isolating the portion of performance attributable to the pro- vider and practice. This becomes even more critical because these outcomes are increasingly tied to provider ranking and reimbursement. Recent studies have begun to investigate which factors affect intermediate outcomes in diabetes mel- litus. Evaluation of the association between glycemic con- trols and chronic illness with complexity (CIC) demon- strated varying associations between glycemic control and CIC in diabetic patients depending on the domain of com- plexity, including nondiabetes physical illness, diabetes- related complications, and mental illness/substance abuse conditions.11 A study investigating the association between patient appointment adherence and glycemic control dem- onstrated that adherence was a strong predictor of diabetes metabolic control, independent of the number of visits.12 Although there have been studies associating chronic dis- ease with glycemic control and patient adherence to glyce- mic control, there have been few that have looked at com- bining patient factors and system factors to determine the effect of nonphysician glycemic management. It is impor- tant to control for patient and system factors that also affect the outcome. In this study, we used a diabetes mellitus registry sponsored by the American Osteopathic Associa- tion (AOA) to evaluate the use of patient factors (as repre- sented by comorbid and severity of disease) and systematic factors (as represented by insurance and missed visits) on glycemic control across a national sample of internal med- icine and family medicine residency programs.

‌Table 1 Attributes included in analysis of glycemic control

Age category Patient age at the time of the

abstracted visit

Gender Patient gender

Diabetes mellitus Medications used to control diabetes at medications the most recent visit stratified by

oral, insulin, or both

Insurance type Insurance as of the most recent visit

including Medicare, Medicaid, commercial, self-pay, and other

Race/ethnicity Race/ethnicity stratified by Caucasian,

African American, Latino, and other Category of Number of scheduled visits the patient

missed visits missed expressed as a percentage of

the total visits over the study year, with patients missing >20% of visits classified as missing visits.

Hypertension Presence of diagnosed hypertension

treated with medication Albuminuria or Presence of albuminuria or previously

nephropathy diagnosed nephropathy

‌Methods

The study used data from the AOA Clinical Assessment Program (AOA-CAP), a web-based registry developed to provide osteopathic internal medicine and family medicine residency programs with measures to improve the quality of patient care and a training tool to enhance the core compe- tencies of practice-based learning and systems-based prac- tice.13 Standardized data and case definitions are used to ensure that residencies collect information in a consistent manner. The registry collects patient-level data that are then analyzed using measures recommended by the National Quality Forum and Physicians Consortium for Performance Improvement.14,15 As part of the residency accreditation requirements, programs contribute data to the diabetes mel- litus measure set on a random sample of type 1 or type 2 diabetic patients who had at least 2 visits for diabetes within the previous 12 months. The accreditation requirement is based on reporting, not performance. Patients with diet- controlled diabetes, or those lacking information on inde- pendent variables including race, haemoglobin A1c level, albuminuria screen, missed visits, hypertension, or insur- ance type were excluded from analysis. Data outcome of interest in this analysis was glucose control as measured by hemoglobin A1c of <7% on the most recent patient visit with laboratory information. The variables tested for asso- ciation with the outcome of interest are listed in Table 1.

Information regarding the number of missed appoint- ments during the study period was collected and patients were dichotomized by the percentage of missed visits (no show), with those missing >20% of scheduled visits clas- sified as nonadherent by the investigators. x2 tests were used to evaluate bivariate associations for dichotomous variables and t-tests were used for continuous variables. We used

logistic regression to evaluate the association between glu- cose control and independent variables and report adjusted odds ratios and 95% confidence intervals using Wald sta- tistics. All analysis was completed using SAS version 9.1 (SAS Institute, Cary, NC).

‌Results

A total of 4751 diabetes cases were available for analysis. The population was predominantly female (55%) with a mean age of 56.9 years (Table 1). The majority of patients (86.2%) were receiving oral medications or a combination of oral and insulin, with only 13.6% receiving insulin alone (Table 2). Government insurance programs accounted for

Table 2 Characteristics of study population

Characteristics of cohort (n = 4715)

#20% of total visits

>20% of total visits Mean HbA1c % Glucose control

<7%

7 to <8%

>8%

Hypertension

Albuminuria or nephropathy

79.9 (3766)

20.1 (949)

44.8 (2113)

23.7 (1115)

31.5 (1487)

79.8 (3760)

47.2 (2227)

>53% of the payment source for these patients. Distribution

of race showed that 54% of the population was Caucasian, 23% were African American, and the remaining patients were either Latino or other nonspecified ethnicities. The patients missed >20% of the scheduled visits. The mean haemoglobin A1c in the study group was 7.6%, with 44.8% of patients having their most recent reading as <7%. Table 3 shows the associations between risk factors and glycemic control. We found that age, medication types, insurance type other than commercial, race with the exception of “other,” increased frequency of missed visits, and albumin- uria/nephropathy all had statistically significant associations with glucose control when evaluated on a bivariate basis. Using a multivariable model, only age >60 years, oral medications without insulin, missing <20% of visits, lack of albuminuria or nephropathy, and Caucasian race were associated with a statistically significant increased likeli- hood of glycemic control. The discrimination of the model, with inclusion of all variables, was moderate, with a C-sta- tistic of 0.663. A mixed effects model was used to evaluate the effect of individual programs on the outcome of interest. Associations between the covariates did not change, whereas the variables listed in the multivariable model re- mained statistically significant.

‌Discussion

This evaluation furthers research into determinants of gly- cemic control of diabetic patients in the community setting by including measures of patient demographics, comorbid disease, and system factors such as insurance type and patient adherence. We were able to evaluate the independent effect of (1) comorbid illness and (2) severity of illness as intermediate outcome of glycemic control in type 1 and type 2 diabetics. In our analysis use of insulin, younger age, self-pay or other insurance type, non-Caucasian race, miss- ing >20% of visits, and presence of albuminuria were all associated with significantly higher risk of poor glycemic control. Although the clinical significance of these findings is intriguing, they have significant implications on how physician performance is judged. These risk factors serve as a starting point for discussions regarding which factors may be appropriate for inclusion in risk adjustment models to determine physician attribution to the control of diabetics within their practice. Performance assessment using out- comes measures is rapidly expanding in health care both in quality improvement and pay for performance models.

Since the early 1900s, when Ernest Codman first sug- gested that outcomes should be measured in hospitalized patients, there have been discussions and research directed at the development of methods of removing the risk of the outcome the patient brings to the interaction.16 These efforts have led to the development of risk adjustment systems that take into account patient factors such as demographics, the severity of underlying disease, and the presence of comor- bid illness that may influence the ultimate outcome of pa-

‌Table 3 Bivariate and multivariate associations between hemoglobin A1c <7% and risk factors Associations between glycemic control and risk factors (n = 4715)

Glycemic control in cohort with attribute (% <7%)

Bivariate comparison to patients

without attribute Multivariable comparison

p-value for bivariate comparison Adjusted odds ratio

tients presenting to the hospital. Although there may be arguments regarding the precision of these models, the fact that they are necessary to evaluate a hospital performance is unquestioned. The impact that risk adjustment can have on hospital rank, used as measures of improvement for absolute performance, was demonstrated in a study of hospital rank- ing in primary cesarean section rates in the Cleveland area.17 The findings of this study demonstrated a dramatic change in ranking, with one hospital moving from an outlier status having a risk adjustment.

The science of risk adjustment of hospital care has ad- vanced dramatically over the past several decades to the point where the federal government and others are publicly distributing risk-adjusted reports.10 However, it is impera- tive to keep performance measurement in the context of improving care and not forcing physicians to “cherry pick” their patients to improve performance measurement. The development of risk adjustment of ambulatory measures has just begun. These measures should include standard meth- ods of defining of risk adjustment systems including iden- tification of dimensions of risk for the outcome, testing associations between dimensions of risk and the outcome of

interest, development of multivariable predictive models using these dimensions, and then clinical and statistical validation of these models.

Several articles have demonstrated association among comorbid disease, patient compliance with visits, and eth- nicity as determinants of glycemic control in diabetic pa- tients.11,12 Our findings replicate the effect these variables have had on glycemic control and allow evaluation of the effect they have independently on control. By evaluating both comorbid illness and patient factors, we demonstrate that after multivariable adjustment, both factors can remain predictive of glycemic control. When evaluating hospital outcomes risk adjustment models, it is important to select risk factors that make statistical and clinical sense.

The measurement of outcomes in the ambulatory envi- ronment requires a reassessment of what is considered a dimension of risk compared with what has traditionally been used in the inpatient environment. The exposure of a patient to a hospital within a clinical entity, such as in the case of myocardial infarction, is finite in time and usually unaffected by factors other than the patient’s age, severity of illness, and burden of comorbid disease. Although not con-

sidered in inpatient risk adjustment methods, factors such as insurance status and patient motivation are important if we are using the physician as the unit of analysis and thus implying the physician’s control of the outcome. The locus of control for some of these risk factors can be attributed to the patient, the physician, or the system. For example, the association between insulin use and poor glycemic control may be the result of prescription inadequacy by the physi- cian, poor acceptance and adherence by the patient, or the inability to acquire the necessary instruments and medica- tion because of a lack of monetary resources as a system issue.

Several previous studies have raised issues regarding methods of measuring physician contribution to glycemic control even after case-mix adjustment.18,19 None of these studies used measures of patient adherence, although one suggested that only 4% of the variance found in profile mea- sures after case-mix adjustment was attributable to the physician.18 On the other hand, there is evidence that a well-designed systematic intervention to improve the care of diabetic patients does have an effect on glycemic control. A recent meta-regression analysis demonstrated the effect a variety of systematic changes in delivering care to people with diabetes has on their hemoglobin A1c, thus implying that there are opportunities to improve the care and out- comes of these patients.20

The need for risk adjustment when evaluating glycemic control in diabetics and attributing the level of control to a physician practice is apparent. The dialogue in terms of selection of dimensions of risk has yet to begin. As orga- nizations measuring quality, along with payers, move for- ward with measure development, some checks and balances need to be put in place so that the art of measurement does not move ahead of science. The impact of poorly designed measures implemented in a pay-for-performance setting may not be under control of the physician. These practices are more likely those that provide care for underserved populations. In this study we demonstrate that: (1) Patient factors associated with financial issues have a significant impact on glycemic control, (2) having self-pay as an in- surance status reduces the likelihood of maintaining glyce- mic control by 35%, and (3) missing >20% of visits re- duced the same likelihood by 17% independent of all other risk factors within the model.

Strengths of this study include the voluntary nature of data entry. Residency requirements are for contribution of data only. They are not based on levels of performance. This lack of consequences should offer a safer environment for ascertaining true practice patterns. Weaknesses of this study include generalizability because the populations represented here are generally underserved and under the care of phy- sicians in training. We chose a hemoglobin A1c of <7% as a marker of control; there has been recent evidence that intensive therapy with a goal of 6% was associated with increased mortality.21 The target in this study is less impor- tant than the associations. Analysis of the associations when using 7.5% as a goal revealed identical findings as when

using 7.0% as the outcome of interest in the current study. The intermediate outcome of glucose control in diabetes is affected by patient factors, such as adherence, and system factors, such as insurance and physician factors. It is im- portant to adjust for these factors to determine physician attribution and to identify which factors present the largest opportunity gap to improve care.

‌References

1. ‌UK prospective Diabetes Study (UKPDS) Group: Intensive blood- glucose control with sulphonylureas or insulin compared with conven- tional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33) UK Prospective Diabetes Study (UKPDS) Group. Lancet 352:837-853, 1998

2. Gaede P, Vedel P, Larsen N, et al: Multifactorial intervention and cardiovascular disease in patients with type 2 diabetes. N Engl J Med 348:383-393, 2003

3. ‌Diabetes Control and Complications Trial Research Group: The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus The Diabetes Control and Complications Trial Research Group. N Engl J Med 329:977-986, 1993

4. ‌Saaddine JB, Cadwell B, Gregg EW, et al: Improvements in diabetes processes of care and intermediate outcomes: United States, 1988- 2002. Ann Intern Med 144:465-474, 2006

5. National Committee for Quality Assurance: Bridges to Excellence® rewarding quality across the healthcare system [internet]. Available at: http://www.ncqa.org/tabid/431/Default.aspx. Accessed March 7, 2012.

6. Centers for Medicare and Medicaid Services: Physician quality report- ing initiative [internet]. Available at: http://www.cms.gov/site-search/ search-results.html?q=pqri%202010. Accessed March 7, 2012.

7. ‌Iezzoni L: Risk Adjustment for Measuring Health Care Outcomes. 3rd ed. Chicago, IL: Health Administration Press, 2003

8. Grover FL, Shroyer AL, Hammermeister K, et al: A decade’s experi- ence with quality improvement in cardiac surgery using the Veterans Affairs and society of thoracic surgeons national databases. Ann Surg 234:464-472, 2001

9. ‌Anderson HV, Shaw RE, Brindis RG, et al: Risk-adjusted mortality analysis of percutaneous coronary interventions by American College of Cardiology/American Heart Association guidelines recommenda- tions. Am J Cardiol 99:189-196, 2007

10. United States Department of Health and Human Services: Hospital Compare—A quality tool for adults, including people with Medicare. 2010. Available at: http://www.hospitalcompare.hhs.gov. Accessed August 20, 2012.

11. ‌Meduru P, Helmer D, Rajan M, et al: Chronic illness with complexity: implications for performance measurement of optimal glycemic con- trol. J Gen Intern Med 22(Suppl 3):408-418, 2007

12. ‌Schectman JM, Schorling JB, Voss JD: Appointment adherence and disparities in outcomes among patients with diabetes. J Gen Intern Med 23:1685-1687, 2008

13. ‌Rider EA, Nawotniak RH, Smith G: A Practical Guide to Teach- ing and Assessing the ACGME Core Competencies. 2007. HCPro, Inc., Marblehead, MA.

14. National Quality Forum: Standardizing ambulatory care performance measures. Available at: http://www.qualityforum.org/Measures_List. aspx. Accessed March 7, 2012.

15. ‌Physician Consortium for Performance Improvement. Chicago, IL, American Medical Association, 2012.

16. ‌2003 Ernest A. Codman Awards presented for excellence in outcomes measurement. Jt Comm Perspect 24:8-9, 2004

17. Aron DC, Harper DL, Shepardson LB, et al: Impact of risk-adjusting cesarean delivery rates when reporting hospital performance. JAMA 279:1968-1972, 1998

    
    

18. ‌Hofer TP, Hayward RA, Greenfield S, et al: The unreliability of individual physician “report cards” for assessing the costs and quality of care of a chronic disease. JAMA 281:2098-2150, 1999

19. Greenfield S, Kaplan SH, Kahn R, et al: Profiling care provided by different groups of physicians: effects of patient case-mix (bias) and physician-level clustering on quality assessment results. Ann Intern Med 136:111-121, 2002

20. ‌Shojania KG, Ranji SR, McDonald KM, et al: Effects of quality improvement strategies for type 2 diabetes on glycemic control: a meta-regression analysis. JAMA 296:427-440, 2006‌

21. Action to Control Cardiovascular Risk in Diabetes Study Group; Gerstein HC, Miller ME, Byington RP, et al: Effects of intensive glucose lowering in type 2 diabetes. N Engl J Med 358:2545-2559, 2008