Metformin Use in Practice: Compliance With Guidelines for Patients With Diabetes and Preserved Renal Function

This was an observational, cross-sectional, epidemiological study that included patients ≥18 years of age with type 2 diabetes in primary care settings. This project assembled data from health record sources as outlined below. Patient identities were masked throughout the study in a limited dataset format in accordance with the Health Insurance Portability and Accountability Act.

Data for this study were derived from the Building Infrastructure for Comparative Effectiveness Protocols (BICEP) study, which included 19,570 type 2 diabetes patients obtained from electronic health records (EHRs) of 307 primary care practitioners. The records included practice management system demographics, administrative claims, EHR clinical problem lists, office measures, notes, laboratory results, and medication histories. Type 2 diabetes was defined using a method similar to that defined by Nichols et al. (16) and DeSai t al. (17), triangulating type 2 diabetes from EHR problem lists, type 2 diabetes claims diagnoses, A1C >6.5%, and diabetes medications. Patients with type 1 diabetes, gestational diabetes, or polycystic disease and children <18 years of age were excluded.

A patient-specific study baseline was defined as the first type 2 diabetes encounter for each patient fully recorded in the EHR with a complete medication history after September 2009 and before the last data pull on 12 March 2012.

The primary endpoint was the use of metformin in those with a normal eGFR.

Independent variables included demographics (i.e., age, sex, race, ethnicity, marital status, history of smoking, and primary language spoken), health system variables (i.e., primary insurance, provider specialty, provider age, and number of diabetes cases for provider), office measures (i.e., height, weight, BMI, and blood pressure), laboratory metrics (i.e., eGFR, A1C, total cholesterol, HDL cholesterol, LDL cholesterol, triglycerides, anion gap, alanine aminotransferase, aspartate aminotransferase, microalbumin, serum albumin, and serum creatinine), major diagnoses (i.e., hypertension, heart failure, renal failure, liver disease, history of drug or alcohol abuse, asthma or chronic obstructive lung disease, and comorbidity count), and selected medications (including 11 classes of diabetes medications, 6 classes of cholesterol-lowering agents, antiarrhythmics, 5 classes of diuretics, ACE inhibitors, angiotensin receptor blockers [ARBs], and beta blockers).

The major diagnoses and the comorbidity count were defined using the Hierarchical Condition Category (HCC) model, which provides a disease-specific, validated methodology for risk stratification (18). We used the CMS CY 2011 HCC Risk Adjustment Model to create 184 “Condition Category” variables (19).

Age was aggregated into eight age-groups. Medications were aggregated into groups using Generic Product Identifier six-digit codes. BMI was calculated using height and weight and then aggregated into underweight, normal weight, overweight, obese, and unknown. Office measures and laboratory metrics were also aggregated into logical groups. Baseline laboratory values were transformed into binary variables, distinguishing an abnormal result from a normal or missing result.

Descriptive analysis was done to compare those with a normal eGFR on and off metformin (Table 1) and those with a normal eGFR on and off metformin and not receiving insulin (Table 2). Due to multiple comparisons, P values were defined as significant if <0.01. In both tables, χ² tests were used to compare the proportions of metformin usage in each group, and those with significant associations were used in the statistical model. Among those with a normal eGFR, we used logistic regression to model the binary outcome of metformin use (yes vs. no) to estimate odds ratios (ORs) and 95% CIs for age, comorbidity count, prescriber’s age (dichotomized as ≤40 or >40 years), A1C status, heart failure status, and use of antihypertensive and diabetes drugs (Table 3). All analyses were conducted using SAS® (SAS Institute, Inc.; Cary, NC).

When comparing demographic data for patients with an eGFR >60 mL/min for their metformin status, several variables appear to be associated (Table 1).Women were significantly less likely than men to be on metformin. Worsening A1C was associated with an increased probability of metformin use (Figure 1); a significant increment of metformin use was observed when A1C increased (χ² test P <0.0001) Those <70 years of age had similar rates of metformin use; those ≥70 years of age appeared to have lower rates of use (Figure 2). As number of comorbid conditions increased, patients were less likely to be on metformin (Figure 3). Providers ≤40 years of age were less likely to prescribe their patients metformin; however, this relationship was not maintained when providers were stratified by age 55 years. There was no difference between prescription rates among provider specialties.

Among specific comorbid conditions, only patients with CHF had a significant difference in metformin prescription rate; those with CHF were much less likely to receive a prescription for metformin. No differences were observed between patients with hypertension, hepatic disease, or significant drug or alcohol use.

There was no significant difference in prescription rates for patients who were on insulin. Patients who were also prescribed an ACE inhibitor, an ARB, a statin, or a nonstatin cholesterol agent were all significantly more likely to also be prescribed metformin. Patients who were prescribed any type of beta blocker, any potassium-sparing diuretic, or a loop diuretic were all less likely to receive a prescription for metformin.

Multivariate logistic regression (Table 3 and Figure 3) provided the ORs of a specific variable predicting a patient’s metformin status when simultaneously considering other potent variables. In the fitted model, increasing patient age, having a provider ≤40 years of age, having CHF, and being treated with insulin were strong predictors that a patient would not be on metformin. In contrast, having ≤11 comorbidities, having a higher A1C, and being on an ACE or ARB, statin, nonstatin cholesterol medication, or other diabetes medication were all strong predictors that a patient would be treated with metformin.

Diabetes is a complex, chronic illness requiring continuous medical care with multifactorial risk-reduction strategies beyond glycemic control. Because of the role lifestyle plays in the acquisition of insulin resistance and the chronic and progressive nature of glucose intolerance, type 2 diabetes is rarely seen in isolation; it is often accompanied by a significant burden of comorbid disease such as hypertension, hyperlipidemia, heart disease, vascular disease, and neuropathy, all of which have independent chronic therapy requirements.

It is unfortunate happenstance that many of the conditions caused by diabetes are themselves direct contraindications to its treatment. Providers must carefully balance a patient’s diabetes care needs with those of their other conditions. Thus, there are three general scenarios in which metformin would not be an optimal therapy choice for patients with diabetes: 1) there is an existent contraindication to metformin’s use, 2) patient behavior or preference may play a role in alternate agent selection, or 3) providers may have a preference or personal bias from experience or other factors. Indeed, these three scenarios may be common in clinical practice; of the eligible patients in this study who did not meet exclusionary criteria for metformin use, 33.9% were, nonetheless, not on metformin.

The most prominent and frequently encountered contraindication to metformin use in the outpatient setting is renal failure, and rates of its inappropriate use in this setting are well documented in the literature (8,9). Patients with renal failure were excluded from our analysis. Of the remaining contraindications examined in our study, only CHF was associated with differences in prescription patterns. Patients with CHF were significantly less likely to be prescribed metformin, in proper accordance with guidelines. This trend was also reflected in prescription of other medications; patients on beta blockers, loop diuretics, and potassium-sparing diuretics—agents that are largely used in the treatment of CHF—were significantly less likely to receive metformin. Providers may correctly recognize that CHF is a contraindication to metformin, or some other factor may be at work (15,20). However, there appeared to be no difference in metformin use in patients with either hepatic failure or alcohol abuse. These conditions may be overlooked by providers as a risk for metformin-induced lactic acidosis, or other factors may be weighed in the decision to continue therapy against guidelines.

Patient preference may also play a role in the selection of therapies other than metformin. It is well documented that lifestyle modification is an effective component in both the prevention and management of type 2 diabetes (21,22). Extrapolating information from the 3,153 subjects presented in Table 1, the majority did not take insulin or other diabetes medications, and 31 and 44%, respectively, also did not take metformin. We do not have direct evidence about patient preferences, but our data suggest a few potential explanations. It may be that a large segment of the population may choose to manage diabetes through lifestyle modification alone. Those who had poor glycemic control but were on no medical therapy may represent a patient population that is averse to treatment, given that noncompliance rates for diabetes medical therapy have been shown to be very high (23).

Metformin has several known adverse effects that may make it less palatable for patients. The most common is gastrointestinal upset, occurring in 10.4–19.3% of patients, usually in the first few weeks of therapy. Although it appears that few patients discontinue therapy early in the course of treatment, a significant portion of patients continue to experience these effects at 6 months (24). Thus, the group on other diabetes oral agents may have switched to an alternative therapy to avoid intolerable side effects.

Finally, a segment of the population with poor glycemic control was on metformin and other oral agents, but not insulin. In fact, being on another oral antihyperglycemic agent was a strong predictor of also having a prescription for metformin. This finding likely represents the population who wish to remain on the therapy even in the setting of poor control. Such prescription patterns often stem from the common patient preference to avoid insulin (25).

Guidelines must always be applied in the context of the individual patient. Clinical experience or personal bias may lead to a provider to offer an alternative agent when deemed necessary. Our study demonstrated that decreased metformin use was associated with older age of patients and with those who had a higher burden of disease, as evidenced by their number of comorbidities. The majority of clinical trials for diabetes control have excluded elderly and frail patients with multiple comorbidities. The elderly represent a heterogeneous population, and with less defined guidelines for optimal therapy, providers are granted more leeway in treatment decisions for these patients (26).

Worsening glycemic control was also associated with higher rates of metformin use. For patients with an A1C >8.0%, it as equally likely that they would or would not be prescribed metformin. This result indicates that providers are following the established practice of including metformin in combination with oral therapy and in combination with insulin therapy. Insulin use, however, was associated with a decreased frequency of metformin use.

The American Diabetes Association recommends that patients with diabetes who are >40 years of age and have ≥1 cardiovascular risk factor (i.e., family history of coronary artery disease, hypertension, smoking, dyslipidemia, or albuminuria) receive a statin medication (27). Multiple trials have demonstrated improved cardiovascular outcomes in patients with diabetes who are treated with statins (28). Our study demonstrates that providers prescribing and patients receiving metformin for glycemic control when their eGFR is >60 mL/min are also more likely to be prescribing/receiving medications to control cardiovascular risk. We do not have information about whether this is the result of provider or patient preferences, but it does represent an area in need of further inquiry. Modifications to behavior of both providers and patients can be further explored to improve health outcomes. Evolving continuing medical education strategies have found practices such as office detailing (a practice in which a representative is sent to an office to provide one-on-one educational discussion with the physician) to improve provider behavior (29). More extensive analysis of EHRs may guide us in an approach to improve provider and patient behaviors to promote better health.

Treatment disparities are commonly observed across all fields of treatment. In our study, there appeared to be no differences by race in performance indicators, although the minority populations observed were quite small compared to national averages (30). There was, however, a significant difference in the prescription rate of women with preserved renal function. At this time, the reasons for this difference are unknown, and there appear to be no reported specific diabetes management disparities between sexes reported in the literature.

This study highlights several issues regarding provider compliance with guidelines for the use of metformin. However, several questions are left unresolved and would benefit from further examination. This study was limited by its observational nature, so confounding factors affecting its outcomes cannot be excluded. We were unable to directly assess patient preferences for treatment, so we relied on indirect measures to suggest preferences. Our sampled population also had a very low minority prevalence, limiting its generalization to the population at large.

The original BICEP project was supported by grant number R24HS019474 from the Agency for Healthcare Research and Quality. The continuing research reported here received no additional funding. The authors thank Lisa Godin for her expert assistance with manuscript preparation.

No potential conflicts of interest relevant to this article were reported.