Facility‐level variation in diabetes and blood pressure control in patients with diabetes: Findings from the Veterans Affairs national database

Abstract

Background

Intensive glycemic and blood pressure (BP) control in diabetic patients is associated with improved cardiovascular outcomes.

Hypothesis

We hypothesized that there is suboptimal glycemic and BP control with significant facility‐level variation in patients with diabetes.

Methods

We identified patients with diabetes receiving care in 130 facilities in the Veterans Affairs Health Care System. We assessed facility‐level rates of glycemic (hemoglobin [Hb]A1c <7%), BP (BP <140/90 mmHg), and combined glycemic and BP control (HbA1c <7% and BP <140/90 mmHg), and their facility‐level variation in using median rate ratios (MRR).

Results

Among 1 103 302 patients with diabetes, 50.2% participants had an HbA1c <7%, 77.5% had a BP <140/90 mmHg, and 39.8% had both, HbA1c <7% and BP <140/90 mmHg. Median facility‐level rates were 50.3% (interquartile range [IQR], 47.9%–52.4%) for glycemic control, 78.4% (IQR, 75.2%–80.0%) for BP control, and 39.9% (IQR, 38.14%–42.34%) for combined glycemic and BP control. Unadjusted MRR for glycemic control was 1.61 (95% confidence interval [CI]: 1.51‐1.70) which decreased to 1.16 (95% CI: 1.14‐1.19) after adjusting for patient and facility‐level variables, indicating a 16% variation in glycemic control between 2 identical patients receiving care at 2 random facilities. Unadjusted MRR for BP control was 1.49 (95% CI: 1.41‐1.56), which decreased to 1.25 (95% CI: 1.21‐1.28), whereas unadjusted MRR for combined glycemic and BP control was 1.59 (95% CI: 1.50‐1.68), which decreased to 1.15 (95% CI: 1.13‐1.17) after adjustment.

Conclusions

Facility‐level rates for BP control and glycemic control remain low with significant facility‐level variation. Much of this is explained by patient and facility‐level variables although 16%, 25%, and 15% variation in glycemic, BP, and combined glycemic and BP control remains unexplained.

1. INTRODUCTION

Patients with diabetes mellitus (DM) have a 2‐fold greater absolute risk of cardiovascular disease (CVD) compared to those without DM.1 Intensive glycemic and blood pressure (BP) control is associated with improved cardiovascular outcomes and reduced microvascular complications.2, 3, 4, 5 Therefore, the American Diabetes Association (ADA) recommends a target hemoglobin A1c (HbA1c) lower than 7% and a target BP lower than 140/90 mmHg in most patients with diabetes.6 Although data for glycemic and BP control are available, it is not well known how these vary from 1 facility to another.7 In an ideal healthcare system, there would be minimal to modest variation in their treatment.

Prior studies have shown significant variation in the use of statin therapy among patients with diabetes.8, 9 In this study, we examined the frequency and facility‐level variation in glycemic and BP control in DM patients receiving care in primary care facilities in the entire Veterans Affairs (VA) Health Care System. Furthermore, we assessed the degree of variation attributable to patient‐related variables.

2. METHODS

2.1. Patient population

We identified patients with DM ages 18 to 75 years with a primary care visit in 130 VA healthcare facilities or their associated community‐based outpatient clinics between October 2013 and September 2014. Cohort details have been described previously.9 We identified patients as having DM if any of the following were documented: 2 outpatient or 1 inpatient diagnosis code indicating DM, from International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) (250.xx, 357.2, 366.41), filled prescription for DM medication, or any fasting glucose ≥126 mg/dL, hemoglobin A1C >6.5%, or ≥2 outpatient blood glucose readings >200 mg/dL on 2 separate days.10, 11, 12 Based on chart review of 100 random patients, we found a positive predictive value of 94% for the diagnosis of diabetes using the algorithm described above compared to actual chart review. We excluded patients if they had a history of metastatic cancer in the preceding 5 years or a history of hospice care in the preceding year.10

2.2. Covariates

We identified patient demographics (age, gender, race) and past medical history of hypertension from the VA administrative data sources. We identified insulin use through filled prescriptions. ICD‐9‐CM diagnosis and procedure codes were used along with procedure terminology codes to identify presence of CVDs (ischemic heart disease, peripheral arterial disease, or ischemic cerebrovascular disease) as described in prior studies.9, 13, 14, 15

We used patient‐level variables to calculate Diagnostic Cost Group (DCG) relative risk scores (RRS), which serve as an indirect measure of a patient's overall illness burden. DCG RRS has been used and validated in prior studies as a surrogate marker for disease burden.9, 14, 15 A patient with a DCG RRS score of 1.5, for example, reflects a 50% greater expected cost of care compared with an average patient (DCG RRS score = 1), reflecting a 50% greater disease burden.

We also assessed facility and system of care variables such as receipt of care from a physician vs advanced practice provider (APP) (ie, a nurse practitioner or a physician assistant), teaching vs nonteaching facility, the number of primary care visits, and the number of endocrinology visits in 12 months preceding the index primary care visit.

For glycemic control, we determined the most recent HbA1c values in the 12 months preceding the patient's index primary care visit. We used BP readings on the patient's index visit to determine BP control.

2.3. Statistical analysis

We compared patient‐, provider‐, and facility‐related characteristics between patients with 1 HbA1c <7% and HbA1c ≥7% and 2 BP < 140/90 mmHg and BP ≥140/90 mmHg. The χ² test was used to analyze categorical variables (example, history of CVD, hypertension), whereas the t test was used for continuous variables (example, age, mean number of primary care or endocrinology visits in the 12 months prior to the index visit). We then assessed 1 median facility‐level rates for glycemic control (HbA1c <7%), 2 median facility‐level rates of BP control (BP <140/90 mmHg), and 3 median facility‐level rates of combined glycemic and BP control (HbA1c <7% and BP <140/90 mmHg). We constructed a multivariable hierarchical regression model to determine median rate ratio (MRR) to assess the magnitude of facility‐level variation in glycemic control, BP control, and the combination of glycemic and BP control. These hierarchical models adjusted for clustering of patients within facilities, and modeled each individual facility as a random effect and patient characteristics as filter effects within each facility.16 This allows patients with similar baseline characteristics from different facilities to be compared with each other by controlling for confounding. The MRR can be interpreted as the likelihood that 2 random facilities would differ in treatment of identical patients. For example, an MRR of 1 suggests no facility‐level variation, whereas an MRR of 1.50 suggests a 50% probability of differing treatment for identical DM patients receiving care at 2 random facilities. This methodology to study variation in care has been described before.8, 9, 13, 17 MRRs for glycemic control were initially derived from an unadjusted model, followed by a model adjusting for patient variables (age, gender, race [white vs others], history of hypertension and CVD, use of insulin, DCG RSS) and then for patient and nonpatient variables (provider type [physician vs APP]), receipt of care at a teaching vs nonteaching facility, and the number of primary‐care and endocrinology visits in the prior 12 months). The resultant MRR from these adjustments indicate how much of the variation in glycemic control was due to patient and nonpatient variables described above. Similar analyses were then performed for the outcome of BP control and the combined outcome of glycemic and BP control. Finally, we performed sensitivity analysis for HbA1c <8% and HbA1c <9% to evaluate for glycemic control and its facility‐level variation when less stringent cutoff values were used.

SAS version 9.1.3 (SAS Institute, Inc., Cary, NC) and Stata version 11 (StataCorp, College Station, TX) were used for analysis. The approval for protocol was obtained from the institutional review boards at Baylor College of Medicine and the Michael E. DeBakey VA Medical Center.

3. RESULTS

3.1. Baseline data

The cohort with DM patients had a total of 1 483 164 patients. After excluding for patients less than 18 years old or greater than 75 years old (n = 345 604) and patients with metastatic cancer and/or hospice care (n = 34 258), a total of 1 103 302 patients were included in the study. Most patients were white (68.9%) and were predominantly male (95.6%). CVD was present in 35.7% of the patients, whereas 33.7% of the patients were using insulin. Overall, 41% of the patients received care at a teaching facility and 20.6% of the patients received care from an APP.

Compared to patients who had HbA1c ≥7%, patients with HbA1c <7% were younger, and less likely to be white and male (P < 0.0001 for all) (Table 1). They were also less likely to have hypertension (82.7% vs 84.5%), though slightly more likely to have CVD (50.2% vs 49.8%, P < 0.0001 for both). Insulin use was more common in patients with HbA1c ≥7% (52.4% vs 15.1%, P < 0.0001). Both groups had similar DCG RSS (illness burden). A similar proportion of patients in each group were receiving care at a teaching facility and from an APP. The group with HbA1c <7% had a lower number of mean primary care (4.8 ± 4.91 vs 5.4 ± 5.37) and endocrinology visits in the year preceding the index primary care visit (0.3 ± 1.25 vs 0.11 ± 0.67, P < 0.0001 for both).

Table 1.

Baseline characteristics of the study population by glycemic control

Variables	HbA1c <7%, n = 516 626 (50.2%)	HbA1c ≥7%, n = 512 683 (49.8%)	P Value
Age, y, mean ± SD	63.4 ± 8.03	62.7 ± 8.25	<0.0001
White race, %	68.6	69.2	<0.0001
Male sex, %	95.0	96.2	<0.0001
History of hypertension, %	82.7	84.5	<0.0001
History of CVD, %	34.2	37.3	<0.0001
DCG RRS, mean ± SD	1.7 ± 2.56	1.7 ± 2.51	0.44
Insulin use, %	15.1	52.4	<0.0001
Receiving care at a teaching facility, %	40.9	41.1	00.13
Advanced practice primary care provider (NP or PA), %	21.1	20.2	<0.0001
No. of primary care visits in the 12 months prior to the index primary care visit, mean ± SD	4.8 ± 4.91	5.4 ± 5.37	<0.0001
No. of endocrinology clinic visits in the 12 months prior to the index primary care visit, mean ± SD	0.11 ± 0.67	0.3 ± 1.25	<0.0001

Abbreviations: CVD, cardiovascular disease; DCG RRS, diagnostic cost group relative risk score (marker of illness burden of patients); HbA1c, hemoglobin A1c; NP, nurse practitioner; PA, physician assistant; SD, standard deviation.

Patients with BP <140/90 mmHg were more likely to be white (69.9% vs 64.6%), whereas age and proportion of males was similar in both groups (Table 2). CVD was less prevalent in patients with BP <140/90 mmHg, whereas disease burden in patients, as per DCG RSS, was higher in the group with BP ≥140/90 mmHg (1.8 ± 2.79 vs 1.68 ± 2.58), as was the proportion of patients receiving care at teaching facilities (43.2% vs 40.4%, P < 0.0001 for all). APPs were caring for 20.9% of the patients in the BP <140/90 mmHg group and 20% in the BP ≥140/90 mmHg group (P < 0.0001). There were fewer primary care visits in the year preceding the index primary care visit for the group with BP <140/90 mmHg (5.0 ± 5.09 vs 5.2 ± 5.29, P < 0.0001). Statistical significance was reached despite small differences in many instances due to the large sample size.

Table 2.

Baseline characteristics of the study population by blood pressure control

Study Variables	BP <140/90 mmHg, n = 245 005 (22.5%)	BP ≥ 140/90 mmHg, n = 845 372 (77.5%)	P Value
Age, y, mean ± SD	63.1 ± 8.26	62.9 ± 8.06	<0.0001
White race, %	69.9	64.6	<0.0001
Male sex, %	95.4	95.8	<0.0001
History of CVD, %	36.1	35.1	<0.0001
DCG RRS, mean ± SD	1.68 ± 2.58	1.8 ± 2.79	<0.0001
Advanced practice primary care provider (NP or PA), %	20.9	20.0	<0.0001
Receiving care at a teaching facility, %	40.4	43.2	<0.0001
No. of primary care visits in the 12 months prior to the index primary care visit, mean ± SD	5.0 ± 5.09	5.2 ± 5.29	<0.0001

Abbreviations: BP, blood pressure; CVD, cardiovascular disease; DCG RRS, diagnostic cost group relative risk score (marker of illness burden of patients); NP, nurse practitioner; PA, physician assistant; SD, standard deviation.

3.2. Glycemic control and facility‐level variation

Approximately half the patients (50.2%) had an HbA1c <7% with the median facility‐level rate of glycemic control (HbA1c <7%) being 50.3% (interquartile range [IQR], 47.9%–52.4%) (Table 3). The unadjusted MRR was 1.61 (IQR, 1.51–1.70) which was substantially attenuated to 1.18 (IQR, 1.15–1.20) when adjusted for patient‐dependent variables. It further attenuated to 1.16 (IQR, 1.14–1.19) when facility‐dependent variables were added to the adjustment model. This suggests that there was an unexplained 16% probability of 2 identical patients attaining varying glycemic control at 2 different facilities.

Table 3.

Facility‐level variation in glycemic and blood pressure control among patients with diabetes

Measures of Glycemic and BP Control	Facility‐Level Rates of Glycemic or BP Control, Median (Interquartile Range)	MRR (95% CI), Unadjusted	MRR (95% CI), Adjusteda	MRR (95% CI), Fully Adjustedb
Glycemic control (HbA1c <7%)	50.27% (47.93–52.41%)	1.61 (1.51‐1.70)	1.18 (1.15‐1.20)	1.16 (1.14 ‐ 1.19)
BP control (<140/90 mmHg)	78.36% (75.22–80.03%)	1.49 (1.41‐1.56)	1.24 (1.20‐1.28)	1.25 (1.21 ‐ 1.28)
Glycemic control (HbA1c <7%) and BP control (<140/90 mmHg)	39.90% (38.14–42.34%)	1.59 (1.50‐1.68)	1.15 (1.13‐1 .17)	1.15 (1.13‐ 1.17)

Abbreviations: APP, advanced practice provider (nurse practitioner or a physician assistant); BP, blood pressure; CI, confidence interval; HbA1c: hemoglobin A1c; MRR, median rate ratio.

^aAdjusted for patient's age, gender, race (whites vs others), diagnostic cost group relative risk score (marker of illness burden of patients), insulin use, history of hypertension, and history of cardiovascular disease.

^bAdjusted for patient's age, gender, race (whites vs others), diagnostic cost group relative risk score (marker of illness burden of patients), use of insulin, history of hypertension, history of cardiovascular disease, APP vs physician provider, teaching vs nonteaching facility, number of primary care visits 1 year prior to the index primary care visit, and the number of endocrinology visits 1 year prior to the index primary care visit.

3.3. BP control and Facility‐level variation

Approximately 77.5% of patients had a BP <140/90 mmHg on their index primary care visit. Median facility‐level BP control was 78.4% (IQR, 75.2%–80.0%). Unadjusted MRR was 1.49 (IQR, 1.41–1.56) that attenuated mildly to 1.24 (IQR, 1.20–1.28) when adjusted for patient‐level variables. Further adjustment for provider‐ and facility‐level variables did not change the extent of facility‐level variation in BP control (MRR 1.25 [IQR, 1.21–1.28]).

Only 39.8% of patients had an HbA1c <7% and a BP <140/90 mmHg. Median facility‐level of combined glycemic and BP control (BP <140/90 mmHg and HbA1c <7%) was 39.9% (IQR, 38.1%–42.3%) (Table 3), with an unadjusted MRR of 1.59 (IQR, 1.50–1.68). When adjusted for patient‐dependent variables, this was reduced to 1.15 (IQR, 1.13–1.17) and did not change when further adjusted for provider‐ and facility‐dependent variables as well.

A greater number of patients met the less stringent criteria for glycemic control (median facility level for HbA1c <8% was 74.7% and for HbA1c <9% was 87.9%) (see Supporting Table 1 in the online version of this article). MRR for HbA1c <8% was 1.52 (IQR, 1.44–1.59), which when adjusted for patient‐level variables attenuated to 1.17 (IQR, 1.14–1.19) with no further attenuation when adjusted for non–patient‐dependent variables as well. MRR for HbA1c <9% was 1.37 (IQR, 1.31–1.42), which when adjusted for patient‐level variables attenuated to 1.16 (IQR, 1.14–1.18) but increased to 1.18 (IQR, 1.16–1.20) when further adjusted for nonpatient variables as well.

4. DISCUSSION

Our study showed that only 50.2% the patients met the recommended target of HbA1c <7%.6 Around 77.5% of the patients on their index primary care visit had a BP <140/90 mmHg. The proportion of patients who had optimal glycemic and BP control (HbA1c <7% and BP <140/90 mmHg) was only 39.8%. The facility‐level variation for glycemic control, BP control, and the combination of both in fully adjusted model were 16%, 25%, and 15% respectively.

The UK Prospective Diabetes Study (UKPDS) and Diabetes Control and Complications Trial (DCCT) both used HbA1c of 7% as the cutoff to compare outcomes.3, 18 The results from these trials did not only show later onset and progression of microvascular complications of diabetes with improved glycemic control, but also demonstrated a linear relationship between these complications and HbA1c levels.19 Nevertheless, 49.8% of the patients in our analyses had HbA1c ≥7%. Considering that 37.3% of the patients with HbA1c ≥7% in our study also had CVD, it puts them at high risk for cardiovascular mortality. Our study does not take into account patients unable to tolerate strict glucose control due to hypoglycemic episodes. Many patients who achieved glycemic goal of <7% may have been in the early stages of their diabetes. We attempted to control for this by including insulin use as a marker of long‐standing diabetes. Although insulin use could account for long‐standing diabetes (and likely more resistant diabetes), it may not fully account for this important variable.

To our knowledge, few studies have examined facility‐level variation in diabetes management. O'Connor et al. in 2008 studied the variation at clinic, patient, and provider level on 2589 patients cared for by 120 physicians in 18 clinics, and concluded that variation was most attributable to patient variables.20 Another study, conducted on 12 110 VA patients concluded that the variation was greatest at the facility level, but did not take into consideration patient variables.21 Our results show that most of this variation is related to differences in patient‐level characteristics.

Our study found the facility‐level variation of HbA1c <7% to be 61%, which was reduced to 18% after patient variables were taken into account. It further reduced to 16% (IQR, 14%–19%) after adjustment was made for provider‐ and facility‐related variables. This implies a significant variation in achieving goal HbA1c among patients with diabetes at different facilities, although mostly accounted for by patient‐level variables. There was a very modest 16% residual variation that could not be accounted for by patient‐level variables and the facility/system of care characteristics that we adjusted for. This residual variation in care likely lies at the level of individual providers in terms of their aggressiveness in managing these patients.

The ADA recommends a BP <140/90 mmHg for most patients and even lower BP (<130/80 mmHg) for those with higher risk of CVD.6 This is based on clinical trials that have shown the benefit of lower BP on CVD outcomes in patients with diabetes.22 Our study showed that 77.8% of the patients had a BP <140/90 mmHg on their index primary care visit, indicating that a reasonable proportion of patients had well‐controlled blood pressures.

To the best of our knowledge, there are no data relating to facility‐level variation in blood pressure control among diabetic patients. Our study showed a 49% facility‐level variation, which was attenuated to 25% when adjusted for patient, facility, and some provider‐related variables. This represents a moderate level variation in blood pressure management patterns at different facilities irrespective of patient variables, and is likely driven by practice patterns of individual providers at these facilities.

Less than half (39.8%) the patients had optimal glycemic (HbA1c <7%) and BP control (last documented BP <140/90 mmHg). These low numbers, however, seem to be primarily driven by those for HbA1c <7% with BP readings making a smaller contribution, reiterating the need to address the high rates of poor glycemic control. There was an estimated 59% facility‐level variation, which was reduced to 15% when patient‐, provider‐, and facility‐related variables were considered.

4.1. Study limitations

Our study has several limitations. We could not account for patients visiting their provider for an acute illness or those undergoing temporary treatment with medications such as steroids, that affect glycemic and BP control. Some patients are not candidates for intensive diabetes or BP management, and this decision could have been made by individual providers after discussion with the patients. We did, however, account for several important comorbidities including history of CVD and the overall illness burden of the patients. Index visit BP readings may not be an accurate reflection of longitudinal BP control, although this should lead to a random error across facilities decreasing the variation toward null. Finally, participants were from the VA healthcare system, making our study findings potentially not applicable to the general population. The strength of our study lies in its large sample size.

5. CONCLUSION

The study showed that only 50.2% of the patients had an HbA1c <7%, whereas 77.8% had a BP <140/90 mmHg on their index clinic visit. There was a significant facility‐level variation in both glycemic control and BP control among diabetes patients. Although most of this is explained by patient‐level variables, about a 16% facility‐level variation in HbA1c control and importantly, a more significant 25% variation in glycemic control remains unexplained. Efforts are needed to identify reasons behind facility‐level variation to improve glycemic and BP control in patients with diabetes and to reduce variations in care.

Conflicts of interest

The authors declare no potential conflicts of interest.

Supporting information

Table S1. Facility‐level variation in glycemic control among patients with diabetes using less stringent HbA1C cut‐offs

Rehman H., Akeroyd J. M., Ramsey D. et al. Facility‐level variation in diabetes and blood pressure control in patients with diabetes: Findings from the Veterans Affairs national database. Clin Cardiol. 2017;40:1055–1060. 10.1002/clc.22769