Variation in Diabetes Management: A National Assessment of Primary Care Providers

Abstract

Background:

Glucose control is monitored primarily through ordering HbA1c levels, which is problematic in patients with glycemic variability. Herein, we report on the management of these patients by board-certified primary care providers (PCPs) in the United States.

Methods:

We measured provider practice in a representative sample of 156 PCPs. All providers cared for simulated patients with diabetes presenting with symptoms of glycemic variability. Provider responses were reviewed by trained clinicians against evidence-based care standards and accepted standard of care protocols.

Results:

Care varied widely—overall quality of care averaged 51.3%±10.6%—with providers performing just over half the evidence-based practices necessary for their cases. More worryingly, provider identified the underlying etiology of the poor glycemic control only 36.3% of the time. HbA1c was routinely ordered in 91.3% of all cases but often (59.5%) inappropriately. Ordering other tests of glycemic control (done in 15% of cases) led to significant increases in identifying the etiology of the hyperglycemia. Correctly modifying their patient’s treatment was more likely to occur if doctors first identified the underlying etiology (65.9% vs 49.0%, P<0.001). We conservatively estimated a US $65/patient/visit in unnecessary testing and US $389 annually in additional care costs when the etiology was missed, translating potentially into millions of dollars of wasteful spending.

Conclusion:

Despite established evidence that HbA1c misses short-term changes in diabetes, we found PCPs consistently ordered HbA1c, rarely using other available blood tests. However, if the factors leading to poor glycemic control were recognized, PCPs were more likely to correctly alter their patient’s hypoglycemic therapy.

Introduction

With over 80 000 deaths a year, diabetes is the seventh leading cause of death in the United States.¹ Patients primarily die from complications related to diabetes, including cardiovascular disease, kidney failure, and Alzheimer’s disease.² The cost of diabetes care and its complications amounts to over US $237 billion in direct medical costs and US $90 billion in reduced productivity.³ With an over twofold increase in diabetes prevalence in the past two decades and an estimated 1.5 million new cases expected annually,⁴ it is clinically and economically critical for this condition to be managed appropriately.

Although there is no cure for diabetes, curtailing its effects requires steady diligence and on-going monitoring by both the patient and their healthcare provider. Numerous studies show that improving glycemic control for patients with type 2 diabetes is associated with an overall reduction in both mortality⁵ and cost of healthcare.^6,7 The benchmark for measuring glycemic control in a patient with diabetes is through the glycated hemoglobin (HbA1c) test.⁸ However, achieving glycemic control has proven difficult. In a recent study, more than 20% of patients with diabetes do not meet their HbA1c targets of <8%, and nearly half do not meet their HbA1c targets of <7%.⁹ Many patients with diabetes (such as postprandial hyperglycemia, patients on medications, or ones with other comorbidities) have even more complex needs. HbA1c testing is less helpful in these patients because it does not detect wide short-term swings in glucose levels,⁸ which has been implicated in complications from diabetes including cardiovascular disease.¹⁰

Using standardized patients, we evaluated the quality of clinical care patients in general, and diagnostic accuracy and glucose control specifically. We investigated three case types where glucose control is particularly difficult and the common glycemic indices of serum blood sugar and HbA1c less helpful: patients with postprandial hyperglycemic excursions; patients recently placed on medications impacting their glucose control in the short term; and patients with 2 comorbidities, anemia secondary to trauma and gestational diabetes. For these more complex patients, we set out to (1) better understand comprehensive diabetes management in a national sample of primary care providers (PCPs) and (2) determine the extent to which PCPs relied on the use of HbA1c results in a manner that would not be considered reasonable and necessary.

Methods

From November 2018 to January 2019, our team conducted the GLUCAR (GLUcose Control using 1,5-AG Randomized controlled trial) study, a prospective, cross-sectional study of the evaluation and care of online cases with diabetes. To control for patient variability and focus on physician practice, we used Clinical Performance and Value (CPV) vignettes. CPVs are simulated patients, frequently used to measure provider practice.

The study was carried out among a national sample of PCPs. We asked the PCPs, who were board certified in either internal or family medicine, to care for 3 types of patients with diabetes who presented with poor glycemic control: (1) patients with postprandial excursions, (2) patients placed on a new drug (prednisone) affecting their glucose control, and (3) patients with a comorbidity (ie, anemia and gestational diabetes) (see Table S1 for case details).

Ethics

This study was conducted in accordance with ethical standards, approved by the Advarra Institutional Review Board, Columbia, MD, and listed in clinicaltrials.gov (NCT03765164). We obtained informed consent from all study participants. Participation was voluntary.

Physician Selection

We randomly recruited study participants from a list of over 25 000 practicing PCPs. The recruitment lists were sourced from relevant physician contact files, including workforce databases, list serves, and rosters of medical associations, hospitals, professional organizations, and national conferences. We used a 12-item screening questionnaire to determine a physician’s eligibility in the study. Providers needed to (1) be physicians either board certified in internal or family medicine, (2) primarily practice in primary care, (3) have between 2 and 40 years of postresidency or postfellowship practice, (4) have an active adult patient panel of at least 1500 patients, and (5) have at least 15% of their panel receiving diabetes care.

Data Collection

We used two instruments to gather data: a survey and CPV vignettes.

Physician Survey

Once eligibility was established and consent obtained, we used results from the 12-item screening questionnaire to stratify a final participation cohort by demographic characteristics: regional geography, age, gender, number of years in practice, practice size, practice locale (urban/suburban/rural), and match the participants to the national demographics of the primary care workforce (see Table S2). We asked questions about their practice such as which glycemic control test(s) did they routinely use, and their practice type (single specialty, multispecialty, hospital-based, etc.).

Clinical Performance and Value-Standardized Patient Vignettes

CPV vignettes are patient simulations that have been validated against standardized patients to reflect actual clinical care.^11,12 CPV simulations have been used extensively in a variety of studies over many years to evaluate and compare clinical practice.^13,14 In a CPV, physicians take a history and order laboratory tests, imaging tests, and procedures just as they would in an actual patient visit. For the GLUCAR study, these open-ended queries in the CPVs are divided into 4 domains of care: (1) taking a history, (2) performing a physical, (3) ordering diagnostic workup, and (4) making a diagnosis with a treatment plan and follow-up. Each vignette has between 61 and 81 explicit criteria that are scored. Scoring is reported as a percentage of the items correct and done by 2 physicians, working independently, using predetermined criteria, with a third physician adjudicating in the case of a disagreement on any of the individual criterion. Because all physicians are caring for the same set of patients, CPV vignettes adjust for case-mix variation and provide a clear measurement of clinical practice variation.¹⁵ For this study, we created 6 CPV cases, with 2 CPVs each in 1 of the 3 abovementioned patient case types.

The care provided by PCPs for these CPV cases was reported as an overall score plus a care score in three specific clinical subdomains: ordering diagnostic workup, making the diagnosis, and developing and outlining a treatment plan.

Analysis

The primary outcomes were (1) the overall variation in provider practice as measured by the CPVs, (2) diagnostic accuracy of the underlying etiology for changes in glycemic control, (3) the efficacy of HbA1c and other glycemic control testing in making the diagnosis, and (4) whether physicians effectively treated patients with poor glycemic control by appropriately modifying the hypoglycemic agents. The secondary outcomes were to determine (5) physician perception of the glycemic control of their patient and (6) the healthcare utilization and costs associated with workup and treatment. We used chi-squared tests for categorical/binary outcome data, and we performed linear and logistic regression analyses to determine whether there were any provider or patient characteristics that were associated with better overall score, diagnostic treatment accuracy, and identification of the underlying etiology for changes in glycemic control. All analyses were conducted in Stata 14.2.

Results

Physician-Practice Survey

A total of 156 board-certified PCPs met the study criteria and were enrolled into the GLUCAR study (Table 1). Over three-quarters of the participants were male and nearly 60% were between the ages of 40 and 55 years; 46.1% were board certified in family medicine and 52.6% were board-certified in internal medicine, with the remainder being double boarded. On average, participants had 21.3±6.3 years of practice experience and currently care for over 3000 patients annually (3034±1389). Three-quarters of participants worked in either a suburban or rural locale, while one-third (38.5%) worked in a multispecialty practice setting, and nearly two-thirds (65.4%) of providers worked in a group practice of any type. Most providers (73.7%) were employed by their practice, and nearly half (48.7%) received a quality bonus.

Table 1.

Baseline Provider Characteristics.

N	156
Male	78.9%
Age
<40	5.1%
40-55	59.6%
>55	35.3%
Board certification
Family medicine	46.1%
Internal medicine	52.6%
Both	1.3%
Years in practice	21.3±6.3
Region
Midwest	23.7%
Northeast	21.8%
South	34.0%
West	20.5%
Locale
Urban	25.0%
Suburban	55.1%
Rural	19.9%
Employed by practice	73.7%
Multispecialty practice	38.5%
Medical practice setting (can choose more than one)
Accountable care organization	16.0%
Solo practice	25.6%
Group practice	65.4%
Hospital based	4.5%
Integrated delivery system	8.3%
HMO (network/staff model)	3.2%
Other	0.0%
Number of active patients	3034±1389
Patients with diabetes	31.3%±16.1%
Receive quality bonus	48.7%
Payer type
Medicare	34.9%
Medicaid	8.5%
Commercial	51.7%
Self	3.8%
Other	1.0%
Tests routinely used to measure glycemic control
HbA1c	100.0%
Random blood sugar	80.8%
Serum fructosamine	14.7%
Serum glycated albumin	8.3%
1,5-Anhydroglucitol	2.6%
Other	3.9%

Variability of Provider Practice

Each participant cared for 3 randomized CPV patients, for a total of 468 simulated patient cases for analysis. The cases were scored on the doctor’s ability to identify, work up, diagnose, and treat these patients presenting with a medical complaint and signs and symptoms of a change in their glycemic control. The overall quality of care score varied widely, from 19.4% to 79.4%, indicating a high degree of practice variation among the participants (see Figure 1). The mean overall score for all participants was 51.3% and had a standard error of 10.6% (range 19.4%-79.4%). Women (+2.4%; P=0.045), providers in the southern region (+2.5%; P=0.015), and those working in a hospital-based practice (+5.3%; P=0.026) performed slightly better than their counterparts. Across care domains, we saw variation in every practice domain, with history taking (65.6%±15.6%) and the physical examination (75.8%±20.5%) the highest, and the composite diagnosis+treatment score the lowest (30.6%±13.1%; range 0.0%-76.9%) (Table 2).

Figure 1.

Histogram of CPV scores.

Table 2.

CPV Results.

	Results
CPV domain
Overall	51.3±10.6
History	65.6±15.6
Physical	75.8±20.5
Workup	56.9±18.6
Diagnosis-treatment	30.6±13.1
Low-value tests (#)	0.9±1.2
Low-value tests (US $a)	$65±$129
Specific items
Primary diagnosis of diabetes	81.8%
Etiology of apparent changes in glycemic control
Postprandial excursions	26.9%
Change in pharmaceutical management	30.8%
Other comorbidities	51.3%
Secondary diagnosis	22.4%
Metformin
Add or continue	62.6%
Increase	20.8%
Add or continue statin	22.3%
Add or continue ACE inhibitor	28.8%
Schedule a follow-up visit	26.5%
Provide routine monitoring of DM progression	26.5%
Offer preventive care services	16.5%

^aBased on Medicare’s 2018 Clinical Lab Fee Schedule and Physician Fee Schedule.

Diagnostic Accuracy

Across all cases we found physicians correctly diagnosed their patients’ primary medical condition (diabetes) 81.8% of the time (Table 2). However, they were only able to identify the underlying etiology of the poor glycemic control 36.3% of the time with only slight differences by case type. In 3 cases of postprandial excursions, physicians identified this as the primary cause 26.9% of the time, and when poor glycemic control was due to a change in medications, they correctly identified steroid use leading to poor control in 30.8% of cases. In the cases where other comorbidities affected glycemic control, physicians identified the clinical condition just over half the time (51.3%). Specifically, in the case where changes in glycemic control were due to a steroid injection in a patient with sciatica, not a single provider who cared for this patient determined the underlying reason for sudden deterioration in the patient’s glycemic control.

We next examined the combined diagnosis+treatment score and found that family medicine physicians (+3.4%; P=0.005), those that practiced in urban settings (+2.8% vs those in suburban or rural settings; P=0.050), and those in hospital practice (+6.5%, P=0.025) were more effective at reaching the right diagnosis and beginning the appropriate therapy. By case type, physicians were marginally more likely to diagnose and provide better treatment (3.2% higher) when another comorbidity was present, compared to the cases of postprandial hyperglycemia (P=0.011) (see Table 3).

Table 3.

Multivariate Linear Regression Results for Diagnosis Treatment Quality-of-Care Score.

	Coefficient	P-value
Female	2.2	0.135
Family medicine	3.4	0.005
Midwest	2.7	0.057
Urban practice	2.8	0.050
Hospital practice	6.5	0.025
Comorbidity case type	3.2	0.011
Constant	25.9	0.000

The Efficacy of HbA1c and Other Testing for Glycemic Control

Providers ordered an HbA1c test 86% of the time (402 out of the 468 cases), indicating their clinical concern for glycemic control. In some CPV cases, an HbA1c test would provide new, helpful information, and in these cases it was ordered in 91.3% of cases. However, in cases when a recent HbA1c was already available or did not provide additional useful information, HbA1c was still ordered 59.5% of the time. Providers working in the northeast region were significantly more likely to order an unnecessary HbA1c than in other regions (OR 4.13, 95% CI, 1.05-16.24). Otherwise, there were no other provider characteristics that indicated ordering an unnecessary HbA1c.

In addition to HbA1c, we looked at other tests to measure glycemic control. Overall, these were not used frequently despite some apparent clinical advantage to detect more recent changes in glycemic control: Serum fructosamine was ordered 7.3% of the time, serum glycated albumin 1.1% of the time, and 1,5-anhydroglucitol was not ordered at all. That few providers ordered these tests which directly measure short-term excursions was notable. Random blood sugar levels were only used in 8.6% of these cases, which was also interesting since 80.8% physicians reported in the survey that they used random glucose levels in their routine practice. In aggregate, when we combined all the instances these tests were used, we found that providers used one of these other tests just 15.0% of the time but that they were 60% more likely to use one or more of these tests in the postprandial cases compared to the other case types (OR 1.61, 95% CI, 0.96-2.72).

Interestingly, ordering any test of glycemic control (HbA1c and others) was not consistently linked to making the primary diagnosis or identifying the cause of hyperglycemia. While ordering an HbA1c was associated with a significantly greater likelihood of making the diagnosis of diabetes (OR 1.97, 95% CI, 1.07-3.64), the HbA1c test was also associated with a significantly lesser likelihood of determining the underlying etiology (OR 0.44, 95% CI, 0.26-0.76). Conversely, ordering other glycemic control tests was associated with a significant increase in recognizing the etiology (OR 1.73, 95% CI 1.02-2.92).

Optimizing Glycemic Therapy

Overall, we discovered that in 55.1% of all cases, providers correctly changed their CPV patients’ glycemic therapy to bring their glucose under control. We looked at what factors influenced correctly changing glycemic therapy, and, interestingly, regardless of the glycemic control test ordering that was ordered, testing was not associated with modifying hypoglycemic treatment (Table 4). This was true whether it was for ordering the HbA1c test (53.7% correct) vs those who did not order the HbA1c test (63.6% correct) (P=0.144), and for those who ordered any of the other glycemic control tests (64.3% correct) vs those who did not order any other test (53.5% correct) (P=0.118). Curiously, we found making or confirming the diagnosis of diabetes was linked with a lower likelihood of adjusting hypoglycemic agent compared to those who did not document their patient’s diabetes (52.0% vs 69.4%, P=0.004). This held true whether the hypoglycemics needed to be increased (50.9% vs 68.6%, P=0.023) or decreased (59.6% vs 70.6%, P=0.354). However, providers who could identify the underlying etiology of a patient’s new hyperglycemia were consistently much more likely to adjust hypoglycemic therapy correctly regardless of the case compared to those who did not (65.9% vs 49.0%, P<0.001). This held true whether the medications needed to be increased (57.4% vs 51.2%, P=0.280) or decreased (92.7% vs 35.0%, P<0.001).

Table 4.

Univariate Factors Associated With Correctly Changing Hypoglycemics Therapy.

	Correctly changed hypoglycemics	P-value
Ordered HbA1c
Yes	53.7%	0.144
No	63.6%
Ordered other test (fructosamine, glycated albumin, and/or random blood sugar)
Yes	64.3%	0.118
No	53.5%
Confirmed diabetes diagnosis
Yes	52.0%	0.004
No	69.4%
Identified underlying etiology
Yes	65.9%	<0.001
No	49.0%

Physicians’ Perception of the Effectiveness of Their Treatment

We asked the providers if they felt their CPV cases had achieved good glycemic control. Only 45.1% providers felt that their simulated patients had done so. By etiology, only 18.6% of the time after caring for the postprandial excursion case did providers feel their patient had good glycemic control. This climbed to 49.4% in their case with medication changes and to 67.3% in their case with other comorbidities. Interestingly, ordering (or not ordering) an HbA1c and seeing the results did not significantly affect whether providers felt these cases had good glycemic control across all cases (46.8% vs 34.9%, P=0.083) mirroring the CPV results above. By case type, only change in medication showed a significant difference, with those ordering an HbA1c more likely to believe their patient was in good glycemic control (55.1% vs 31.6%, P=0.015).

Costs Associated With Low-Value Tests and Improper Treatment

Finally, we looked at excess costs due to waste and poor diabetes management. From the CPV cases, providers ordered on average 0.9±1.2 tests per patient (range 0-6) that did not change the patient management, at a cost of US $65 ± US $129 per case (Table 2, CMMS rates). From the providers’ survey responses (Table 1), we know that these physicians have an active patient panel with, on average, 3034 patients, of which 31.3% have diabetes. If each of these physicians provide US $65 of unneeded workup for their patients, this translates into US $9 629 370 (=156 physicians × 3034 patients/physician × 31.3% patients with diabetes × US $65/patient) in unnecessary diabetes-related workup for the participants alone.

Similarly, we know from previous studies^6,16 that there is a healthcare cost associated with poor vs good glycemic control. Using a difference of US $389⁶ in healthcare costs for patients exhibiting poor control from another study, we know from our data that 44.9% of our cases (=100% − 55.1%) did not correctly change glycemic therapy. This translates into US $25 875 006 (=156 physicians × 3034 patients/physician × 31.3% patients with diabetes × 44.9% in poor control × US $389/patient) of extra preventable cost added into the system, again, for the study participants alone.

Discussion

Adequate glycemic control is challenging for the busy PCP. HbA1c, which has long been the mainstay for diagnosing and altering hypoglycemic therapy, is not suitable in patients with short-term changes in the glucose control. We report on the preliminary results of the GLUCAR study that focused on these patients with diabetes and poor glycemic control. More than 150 providers cared for the same simulated patients. We found that overall care was highly variable in this subset of patients with diabetes who present with common clinical presentations: undetected postprandial hyperglycemia, recently placed on steroids, anemia due to trauma, and gestational diabetes. Of particular note is that HbA1c testing was ubiquitous albeit unhelpful.

We were able to get a particularly close look at the challenges providers face by using simulated patients. With everyone taking care of the same patients, we focused on diagnosis, treatment, and the impact of the workup on the subsequent care. Naturally, we were particularly interested as to whether the hypoglycemic regimen was appropriately altered.

Not only was care varied, but for nearly two-thirds of the cases, providers did not recognize that the patient’s symptoms were due to poor short-term glucose control. Not surprisingly, HbA1c testing was unhelpful in these cases. Nevertheless, it was ordered (and in some instances unnecessarily reordered) in the great majority of cases. Other, possibly more helpful, testing strategies using alternative markers of glycemic control—including serum fructosamine, glycated albumin, and 1,5-anhydroglucitol—were used only 15% of the time. When these testing strategies were used, however, it was much more likely that the etiology of the poor control was uncovered.

The potential significance of better testing, diagnosis, and treatment extends beyond the clinical benefits to decreasing care costs for diabetes by more than US $25 million. Overall, our findings reveal a consistent amount of unnecessary testing and missed opportunities to provide more efficient, less costly care. When the costs of unnecessary testing and overprovision of care estimates from other recent studies are added to our findings, there appears to be the potential for tremendous cost reductions.

When the etiology was correctly identified in our study, this did indeed lead to better treatment and a significantly higher likelihood of having the patients’ hypoglycemic therapy adjusted. While the cases we looked at proved challenging, in general the cases with postprandial hyperglycemic excursions were the most difficult for participants to diagnose and treat compared to patients placed on steroids, those with anemia from recent trauma and status posttransfusion, and a woman with gestational diabetes.

There are a handful of important limitations we want to point out. First, although we attempted to have a nationally representative sample, we were unable to recruit a fully representative sample of women and physicians under the age of 40. The approach we took made it possible to eliminate all of the case-mix variation, helping us identify the specific challenge providers face in recognizing poor glycemic control in order to initiate a corrective change in therapy. Future studies will want to confirm that better workup and diagnosis resulted in better glucose control, although there already is substantial evidence that this is the case in other settings.^17,18 Another limitation is that this study did not look at specialist referrals. We recognize that an alternative to the PCP changing the therapy would be to send the patient to a diabetes specialist. This study, however, was designed specifically to look at PCP practice, where most patients with diabetes are routinely managed.

Notwithstanding, physicians in the GLUCAR study appear to be heavily reliant on the HbA1c test even when a patient presents with new onset symptoms that can only be uncovered through different testing strategies. It is well recognized that the 3-month capture of average glycemic control from a HbA1c test means that shorter term changes in glycemic control and glycemic variability will be missed. In this study, we found that this overreliance led to high variation in clinical practices that led to missed clinical and economic opportunities. An improved diagnostic approach is needed in these types of patients.

Conclusion

In patients with recent changes in their glycemic control, the GLUCAR study found that clinicians rarely recognized the etiology, appeared to be overly reliant on HbA1c testing, and often missed opportunities to improve care and lower costs.