Sumesh Kachroo, PhD

 To identify factors associated with all-cause discontinuation (patient discontinued on their own or physician discontinuation) of oral anticoagulants (OACs) among nonvalvular atrial fibrillation (NVAF) patients.

 We analyzed the MarketScan claims database from October 2009 to July 2012. Adult patients were eligible if they newly initiated an OAC in the study period, had an atrial fibrillation diagnosis (

International Classification of Diseases, Ninth Revision, Clinical Modification

 code 427.31 or 472.32), and had at least 6 months of continuous enrollment after OAC initiation. Multivariable Cox proportional hazards regression was used to assess factors associated with discontinuation. Adjusted hazard ratios (HRs) and 95% CIs were reported.

 Among 12,129 eligible patients, 8143 (67.1%) initiated warfarin and 3986 (32.9%) initiated direct oral anticoagulants (DOACs). Overall, 47.3% of patients independently discontinued during follow-up (mean number of days of follow-up = 416.6 [SD ± 141.7]) with mean time to discontinuation of 120 days (SD ± 114.7). Patients significantly less likely to discontinue included those taking DOACs versus warfarin (HR, 0.91; 95% CI, 0.86-0.97), older patients (≥65 years vs 18 to 34 years) (HR, 0.32; 95% CI, 0.24-0.43), those with diabetes (HR, 0.84; 95% CI, 0.77-0.90), those with prior stroke/transient ischemic attack (HR, 0.65; 95% CI, 0.56-0.75), those with prior pulmonary embolism (HR, 0.71; 95% CI, 0.58-0.88), and those with congestive heart failure (HR, 0.80; 95% CI, 0.74-0.87). Patients with prior bleeding events were significantly more likely to independently discontinue (HR, 1.20; 95% CI, 1.08-1.34).

 The risk of independent discontinuation of OAC treatment among NVAF patients was high. Patients on DOACs compared with warfarin and those with several comorbid conditions had significantly lower risk of discontinuation, while those with prior bleeding were more likely to discontinue.

	As market experience with direct oral anticoagulant therapies (DOACs) grows, it is important to understand the real-world treatment patterns for stroke prevention in patients with atrial fibrillation (AF).

		Our study shows that the risk of discontinuation from any cause of OAC therapy among patients with nonvalvular AF was high.

		Newer medications—DOACs&mdash;may provide better patient outcomes; patients taking DOAC therapy were less likely to discontinue therapy.

Atrial fibrillation (AF) is the most common cardiac arrhythmia seen in clinical practice and is becoming a growing public health concern. Although the magnitude of future trends cannot be definitively predicted, current research indicates that the disease’s incidence and prevalence will gradually increase. Go et al estimated in 2001 that approximately 2.3 million adults in the United States currently had AF, and projected an increase to more than 5.6 million by the year 2050 with more than 50% of affected individuals 80 years or older.

 Colilla et al reported projections in 2013 that AF incidence will increase from 1.2 million in 2010 to 2.6 million in 2030

with the prevalence estimates increasing from 5.2 million in 2010 to 12.1 million in 2030.

 In 2006, Miyaska et al reported the estimated prevalence to be more than 10 million by 2050.

AF results in a 4- to 5-fold increase in the risk of stroke and thromboembolic events, accounting for approximately 15% of all strokes reported in the United States.

 In addition, AF has been associated with increased risk of heart failure, cognitive dysfunction, left ventricular dysfunction, and death.

 AF has enormous socioeconomic implications because, in addition to increasing medical costs and economic burden on the healthcare system, it can lead to a reduction in patients’ quality of life (QOL) and functional status.

 Coyne et al reported that the total annual medical costs for treatment of AF were approximately $6.65 billion in 2005 dollars.

 Kim et al reported that the national incremental AF cost was between $6 and $26 billion in 2008 dollars.

 The economic burden of stroke in the United States for 2008 was estimated at $65.5 billion, whereas the annual estimates for the 27 European Union countries and the United Kingdom were €27 billion and £8.9 billion, respectively, in 2008.

 The QOL of patients with AF has been reported to be significantly poorer compared with the general population, as well as with patients with coronary heart disease.

Historically, stroke-prevention guidelines for AF patients recommended aspirin or no therapy for those with an otherwise low stroke risk, and an oral anticoagulant (OAC) for moderate- to high-risk patients.

-VASc score prediction rule, the number of patients recommended for OAC therapy has dramatically increased compared with those previously recommended under the CHADS

 system. Many patients, especially women 75 years or older (now + 2 from + 1 in CHADS

), have been reclassified as high- rather than low-risk, and OAC use has therefore significantly increased. It has increased even in those still classified as low risk.

For several decades, warfarin has been the primary OAC used for stroke prevention in AF. While highly effective for preventing stroke in AF patients, its significant drawbacks include variable dose requirements and numerous dietary and medication interactions, as well as a commitment to lifelong, regular, frequent monitoring of the patient’s international normalized ratio (INR) to ensure it is within the recommended range.

 As a result, warfarin is associated with relatively frequent bleeding complications, resulting in a high proportion of patients discontinuing therapy (all-cause, initiated by either the patient or their doctor) or receiving suboptimal therapy in usual clinical practice.

In recent years, 3 direct oral anticoagulants (DOACs)—dabigatran, rivaroxaban, and apixaban&mdash;have been approved in the United States for stroke prevention in nonvalvular AF (NVAF). In clinical trials, all 3 were shown to be safe and effective compared with warfarin.

 These new agents have advantages over warfarin because their use doesn’t require regular INR monitoring and they have fewer drug and food interactions. Recently published guidelines have also recommended the use of DOACs as alternatives to the conventional therapy (vitamin K antagonists or antiplatelet agents) in most NVAF patients requiring stroke prevention.

Few studies have shown treatment discontinuation over time in the warfarin-treated patients.

 However, it is absolutely essential to evaluate in detail the treatment patterns and the risk of discontinuation among NVAF patients in the real world, as well as to identify predictors of discontinuation. Although evidence exists for some rationales for discontinuation in warfarin-treated patients, such as bleeding complications,

 little comparable information exists for DOAC-treated patients. Compared with warfarin, DOACs have a shorter half-life; therefore, if the drug is discontinued, noncompliance can result in the anticoagulation effect coming to a halt and thrombosis ensuing.

 It is thus critical to understand real-world discontinuation patterns, including the reasons for discontinuation, for warfarin and DOACs.

Currently, however, little literature exists on recent real-world treatment patterns among AF patients on DOACs. A recent observational study in 70 patients starting on dabigatran (median treatment duration was 140.5 days) reported a discontinuation rate of 10%; 77% of these patients reported treatment satisfaction with dabigatran, although 79% of those previously treated with warfarin preferred dabigatran.

 Another early real-world study on dabigatran adherence in 103 patients reported that only 12% had inadequate adherence. Rivaroxaban and apixaban were approved after dabigatran, and we identified no studies that reported real-world treatment patterns with these drugs.

 As market experience with DOACs grows, it is important to understand the real-world treatment patterns for stroke prevention in patients with AF. The goals of this study were to assess the risk of discontinuation among AF patients on OACs (including DOACs) and to identify factors associated with such discontinuation.

A retrospective cohort study was conducted using the Truven Health MarketScan Research Databases from October 2009 to July 2012. Patients were eligible if: they newly initiated an OAC (warfarin, dabigatran, or rivaroxaban) in the study period, were 18 years or older at the time of treatment initiation, had at least 1 diagnosis of AF in the 12 months prior to treatment initiation (identified by any medical claim associated with an 

International Classification of Diseases, Ninth Revision, Clinical Modification 

code of 427.31 [AF] or 427.32 [atrial flutter]), and had at least 6 months of continuous enrollment (ie, continuous health insurance coverage) after OAC initiation and continuous enrollment in the plan for 12 months prior to study entry. Patients were excluded if they had valvular heart disease, cardiac surgery during the 12-month pre-index period, or used any OAC treatment (warfarin or DOAC) in the 12 months prior to study entry. The index date was defined as date of first OAC prescription after an AF diagnosis during the study eligibility period, and the baseline period was the 12 months prior to index date. The index prescription was defined as the first OAC prescribed in the study period (warfarin, dabigatran, rivaroxaban).

The DOACs included in this study were dabigatran and rivaroxaban, because apixaban was approved in December 2012, after the study’s end date. 

 show the study schematics and the flow chart for patient selection, respectively. Patients were followed until 1 of the following events, whichever earliest, occurred: an interruption in continuous enrollment, a switch or a discontinuation of index treatment, or the end of the study. Discontinuation was defined as having no claims for the index OAC prescription within 60 days of the final day’s supply of the last filled prescription for the index OAC. Patients who received a prescription for an OAC other than the index OAC prescription during the follow-up period were considered switchers, and were thus excluded.

The absolute rate of discontinuation of OAC therapy was determined; additionally, time to discontinuation (in days) and the length of follow-up (in days) were calculated. Patients were censored on the date of discontinuation, end of study period, or interruption in continuous enrollment—whichever occurred earliest. Multivariable Cox proportional hazards regression analysis was conducted to assess the clinical and demographic factors associated with discontinuation in warfarin initiators and DOAC initiators. Hazard ratios (HRs) and the corresponding 95% CIs were calculated. Demographic factors such as insurance type and geographic region, as well as clinical characteristics from the 12-month baseline period&mdash;such as bleeds, stroke, deep vein thrombosis, pulmonary embolism (PE), diabetes, hypertension, myocardial infarction, dyspepsia, renal disease, and congestive heart failure (CHF)&mdash;were included. All analyses were done using SAS version 9.3 (SAS Institute, Cary, North Carolina) and STATA version 13 (STATACorp, College Station, Texas).

A total of 11,825 patients with diagnosis from October 2009 to July 2012 were available for the analyses. 

presents the baseline demographics and baseline comorbidities. A total of 5598 (47.34%) NVAF patients discontinued treatment with OACs. The majority of the total population of patients was 65 years or older (55.07%) and male (59.5%). The most common comorbidities were hypertension (42.62%), diabetes (15.29%), CHF (14.83%), and coronary artery disease (14.67%).

 presents the overall discontinuation rate and the days to discontinuation for both warfarin users and DOAC users. 

 depicts the Kaplan-Meier (K-M) curves that show risk of discontinuation for warfarin and DOAC therapy with time, and 

 shows the factors associated with the risk of discontinuation. Patients taking DOAC therapy were significantly less likely to discontinue compared with patients taking warfarin (reference = warfarin) (HR, 0.91; 95% CI, 0.86-0.97; 

 = .002). As age increased (reference group = aged 18-34 years), the magnitude of risk of discontinuation decreased, with significant results for patients in the age groups 45 to 54 years (HR, 0.56; 95% CI, 0.44-0.71; 

 = .000), 55 to 64 years (HR, 0.42; 95% CI, 0.33-0.53; 

 = .000), and 65 years or older (HR, 0.32; 95% CI, 0.24-0.43; 

 = .000). The patients from the western region of the United States were significantly more likely to discontinue (reference = northeast region) (HR, 1.14; 95% CI, 1.04-1.24; 

 = .003). Patients with prior bleeding events were also more likely to discontinue (HR, 1.20; 95% CI, 1.08-1.34; 

 = .001). The baseline comorbidities that significantly decreased the risk of discontinuation include prior stroke/transient ischemic attack (TIA) (HR, 0.65; 95% CI, 0.56-0.75; 

 = .000), prior PE (HR, 0.71; 95% CI,  0.58-0.88; 

 = .001), a diagnosis of diabetes (HR, 0.84; 95% CI,  0.77-0.90; 

 = .000), and a diagnosis of CHF (HR, 0.80; 95% CI,  0.74-0.87; 

This study highlights the high frequency of all-cause discontinuation of OAC therapy among NVAF patients. Close to half (47%) discontinued during follow-up with a mean time to discontinuation of 120 days. This is consistent with the current literature. Patel et al reported that about 37% of warfarin initiators had discontinued warfarin within 90 days after the initiation of therapy, with 65% discontinuing warfarin therapy within 1 year.

 Deitelzweig et al reported that 51.4% of warfarin initiators with NVAF discontinued warfarin therapy at least once during follow-up.

 Several other studies suggest that approximately 25% of patients on warfarin discontinue therapy within the first year of use.

 Casciano et al have also reported that the underutilization of warfarin is prevalent in NVAF patients and, in addition, nonadherence to warfarin therapy presents additional economic burden on the healthcare system.

 Adherence to therapy is absolutely essential for NVAF patients, as stroke risk is significantly higher during the warfarin discontinuation periods in comparison with periods when the patients are on therapy.

 Ewen et al have also reported that patients who experience 2 or more interruptions in warfarin use have more than twice the risk for stroke (mean follow-up period = 3.4 years) compared with patients with no interruptions.

In addition, a key highlight of the present study is that patients taking DOAC therapy were less likely than patients taking warfarin to discontinue. This result, however, needs to be interpreted with caution. Although the overall risk of discontinuation via the Cox Hazard Model demonstrates the lower risk for DOAC patients compared with warfarin patients (Table 3), it is unclear whether this translates into a meaningful clinical benefit. The K-M curves for discontinuation (Figure 3) show little difference for those patients continuing OAC therapy between the warfarin and DOAC cohorts. Further, the mean time to discontinuation (Table 2) indicates that among patients who discontinue, warfarin patients remain on therapy slightly longer than DOAC patients.

The existing literature has well documented the challenges that plague warfarin therapy, including variable dose requirements and numerous dietary and medication interactions, as well as a lifelong, regular, frequent monitoring of INR to ensure it is within the recommended range.

 On the other hand, DOACs offer patients flexibility, with the absence of required frequent anticoagulation monitoring. Further, DOACs have a more favorable dietary and medication interaction profile compared with warfarin.

A major reason for discontinuing warfarin therapy is its associated bleeding complications and the resulting costs. In their study assessing warfarin-associated bleeding in AF patients, Ghate et al reported the mean adjusted all-cause annual costs as $42,574, $36,571, $22,824, and $22,507 for subjects with intracranial hemorrhage, major gastrointestinal (GI) bleeding, minor GI bleeding, and no bleeding, respectively.

 Kim et al reported a mean cost of $10,819 (SD = $11,536) for hospitalizations due to warfarin-associated bleeding.

 On the other hand, clinical trials and published meta-analyses of the DOACs have shown lower bleeding rates compared with warfarin.

The findings of this study suggest that certain clinical, demographic, and/or healthcare-related characteristics were significantly associated with discontinuation of OACs. We observed that patients with prior bleeding events were more likely to discontinue. Suh et al reported that patients with recent bleeding were more likely to discontinue warfarin compared with patients without recent bleeding (relative risk,â€…1.35; 95% CI, 1.16-1.58).

 The baseline comorbidities that significantly decreased the risk of discontinuation included increasing age, prior stroke/TIA, prior PE, a diagnosis of diabetes, and a diagnosis of CHF. Fang et al reported that the risk of discontinuation was higher in patients 65 years or older compared with patients 85 years or older (HR, 1.33; 95% CI, 1.03-1.72).

 On the contrary, Suh et al reported older age as a risk factor for discontinuation.

 It would be difficult to draw definitive conclusions on the impact of clinical, demographic, and/or healthcare-related characteristics on the discontinuation of OACs. Additional research is needed to build a more detailed understanding, as well as to conduct a comprehensive evaluation on treatment patterns with DOACs using real-world data.

Understanding the real-world usage of warfarin and DOACs is needed to improve existing treatment algorithms. The consequences of discontinuation are immense, often leading to negative health outcomes and increased healthcare costs.

 Given that existing research indicates that the bleeding complications associated with warfarin are a major reason for the drug’s discontinuation,

 it is necessary to thoroughly evaluate how DOACs—which reduce bleeding and/or have a comparatively better bleeding profile

—may encourage improved compliance and thereby provide continuous, uninterrupted stroke protection.

One key strength of our study is its analysis of real-world data on DOACs. The MarketScan claims database, which provides data on patients across the United States, incorporates all medical and pharmacy claims of patients and allows for longitudinal analysis. However, our analysis also has several limitations. The database does not uniformly capture the use of over-the-counter medications, such as aspirin, which are used to help prevent stroke in AF patients and could have an impact on the treatment patterns of the anticoagulants being studied. Also, while the discontinuation of therapy was based on a 60-day gap in pharmacy claims, a lack of INR monitoring means we can’t assess whether the gap was clinically meaningful. In this study, too, we haven’t separately accounted for AF patients who are on short courses of anticoagulation for specific reasons (ie, scheduled for imminent cardioversion or AF ablation procedures). Also, we were not able to identify patients with AF postoperatively, who may have had a short course of anticoagulation therapy. In addition, our analysis was based on early utilization of DOAC, which may change over time. Also, generalizability may be limited because the portion of the study population insured by Medicare only includes that subset of Medicare recipients who are Medicare-eligible retirees with employer-sponsored Medicare Supplemental plans. This patient population may not be representative of all AF patients 65 years or older. In addition, the data do not include uninsured patients.

This study highlights the high risk of discontinuation of OAC therapy among NVAF patients. Despite the Cox Hazard Model indicating a statistically significant reduction in risk of discontinuation for DOAC versus warfarin, the K-M curves are very similar and it is unlikely the small difference seen in discontinuation is clinically relevant. More research utilizing real-world data needs to be conducted to foster detailed understanding of DOAC treatment patterns.

The authors would like to thank Sameer R. Ghate, PhD, for his work in the initial stages of the study. The results of this manuscript were presented at the American Heart Association Scientific Sessions, Dallas, Texas, November 16-20, 2013.

 Bristol-Myers Squibb Global Health Economics and Outcomes Research (SK, MH, XP), Princeton, NJ; Pfizer Global Health Economics and Outcomes Research (XL), New York City, NY; University of Utah Program in Personalized Health Care (DB), Salt Lake City, UT; University of Utah College of Pharmacy (DB, JB), Salt Lake City, UT; University of Utah Comprehensive Arrhythmia Research & Management Center (NM), Salt Lake City, UT.

 This study was sponsored by Bristol-Myers Squibb and Pfizer.     

 Dr Kachroo, Ms Hamilton, and Ms. Pan are employees and stockholders of Bristol-Myers Squibb (BMS). Dr Liu is a Pfizer employee and stockholder. Drs Brixner and Biskupiak received financial support from BMS and Pfizer to conduct this research. Dr Marrouche reports no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (JB, DB, XL, MH, XP); acquisition of data (JB); analysis and interpretation of data (JB, DB, SK, XL, MH, XP); drafting of the manuscript (JB, SK, XL, MH, XP); critical revision of the manuscript for important intellectual content (JB, SK, XL, MH, XP); statistical analysis (JB, SK, DB); obtaining funding (JB, DB); administrative, technical, or logistic support (JB); and supervision (JB).

 Joseph Biskupiak, PhD, MBA, Department of Pharmacotherapy, University of Utah College of Pharmacy, L.S. Skaggs Pharmacy Institute, 30 South 2000 East, Rm 4962, Salt Lake City, UT 84112. E-mail: joe.biskupiak@utah.edu.

1. Go AS, Hylek EM, Phillips KA, et al. Prevalence of diagnosed atrial fibrillation in adults: national implications for rhythm management and stroke prevention: the AnTicoagulation and Risk Factors in Atrial Fibrillation (ATRIA) Study. 

2. Colilla S, Crow A, Petkun W, Singer DE, Simon T, Liu X. Estimates of current and future incidence and prevalence of atrial fibrillation in the U.S. adult population. 

3. Miyasaka Y, Barnes ME, Gersh BJ, et al. Secular trends in incidence of atrial fibrillation in Olmsted County, Minnesota, 1980 to 2000, and implications on the projections for future prevalence. 

4. Dorian P, Kongnakorn T, Phatak H, et al. Cost-effectiveness of apixaban vs. current standard of care for stroke prevention in patients with atrial fibrillation. 

Articles

The risk of discontinuation of oral anticoagulant therapy (both warfarin and direct oral anticoagulant therapies [DOACs]) among nonvalvular atrial fibrillation patients was high. Although the hazard ratio for discontinuation favors DOACs, it is unlikely that the small difference in discontinuation relative to warfarin is clinically meaningful.

Oral Anticoagulant Discontinuation in Patients With Nonvalvular Atrial Fibrillation

Patients with diabetes often exceed desired glycated hemoglobin (A1C) levels for months prior to medication adjustments. To determine if provider and patient characteristics predict glycemic control and treatment intensification.

Observational retrospective cohort study using electronic medical record data. 

 We studied 149 Kaiser Permanente Northwest primary care providers of 14,430 patients with diabetes, of whom 5823 (40.4%) were in optimal control (all A1Cs <7%) and 2446 (17%) were in poor control (at least 1 A1C >9%) in 2011. We also identified a subset of 107 primary care providers of 912 patients with diabetes who were initially in optimal control (A1C <7%) but had a subsequent A1C >7.5% from 2010 to 2011. We used hierarchical linear modeling to assess both patient and provider characteristics as predictors of glycemic control and treatment intensification after incident hyperglycemia.

 Patient characteristics associated with optimal control included older age, lower baseline A1C, shorter diabetes duration, and not using insulin (P <.001 for all). The inverse of these variables predicted poor control. No provider characteristics were associated with glycemic control or intensification. Older patients with a greater change in A1C were more likely to have therapy intensified after loss of glycemic control.

Patient, but not provider characteristics, predicted glycemic control and therapy intensification. Improving systems of care such as disease management services may be a better use of resources than focusing on individual providers.

		In a single integrated health maintenance organization, individual provider characteristics did not predict glycemic control or likelihood of treatment intensification; however, there was little variability between providers. 

		These data suggest that systems of care may contribute more to glycemic control among patient populations compared to individual provider characteristics. 

		Improving systems of care, such as disease management services, may be a better use of resources than focusing on individual providers.

Despite recommendations from the American Diabetes Association,

 only 53% of persons with diabetes in the United States between 2007 and 2010 had glycated hemoglobin (A1C) levels of less than 7%.

 As a progressive disease, type 2 diabetes requires ongoing treatment intensification to achieve and maintain glycemic control.

 Although there are several therapeutic guidelines for managing hyperglycemia,

 previous studies suggest that approximately 50% to 60% of patients with A1Cs above goal are not being prescribed a change in therapy or are actually having their therapy decreased.

 One factor driving this problem is “clinical inertia”—defined as a clinician recognizing a problem but failing to act

—which occurs often in the management of diabetes.

Identifying patient and provider characteristics associated with poor glycemic control and/or delayed therapeutic intensification is essential to improving the health of persons with diabetes. Delays in treatment intensification not only lead to long periods of hyperglycemia, ranging from 1 to 7 years, but also to less likelihood of therapeutic success once therapy is intensified.

 The causes of these long delays in therapy intensification during diabetes management are not clear. Current guidelines suggest that patient characteristics such as older age, frailty, or multiple comorbidities may be legitimate reasons for delaying therapy intensification,

 but previous models showing the influence of such patient characteristics on glycemic control have had limited predictive power.

 Moreover, although some studies have found that diabetes specialists may opt to intensify patients’ treatment sooner than primary care providers (PCPs),

 there has been little examination of specific provider characteristics that may contribute to delays in diabetes therapy intensification.

To further understand reasons for inaction in the context of elevated A1Cs, we examined the extent to which clinician and patient characteristics were associated with glycemic control and treatment intensification after loss of glycemic control.

The study was reviewed and approved by the institutional review board of Kaiser Permanente Northwest.

KPNW is a nonprofit, group-model health maintenance organization that provides integrated comprehensive medical care to about 490,000 individuals in a 75-mile radius around Portland, Oregon. PCPs include physicians, nurse practitioners, and physician assistants of Internal Medicine and Family Practice departments. Patients are assigned a personal PCP, though they also visit other primary care clinicians for short-notice appointments or when their personal clinician is not available. A group of patients assigned to a provider is called a “panel.” To assist providers in caring for their patients with diabetes, KPNW provides evidence-based guidelines based on American Diabetes Association recommendations.

Identification and Follow-up of Patients With Diabetes

KPNW uses healthcare utilization data to track and facilitate operations. An electronic medical record—in use since 1996&mdash;allows providers to record 

) diagnosis codes at each patient contact and to update an electronic problem list (which also utilizes 

 codes). A single regional laboratory processes most KPNW outpatient laboratory tests, and the results are stored in a searchable database. A pharmacy is located in each medical office, and most members have a pharmacy benefit, ensuring almost complete capture of data describing pharmaceutical dispenses.

Subjects were members of the KPNW electronic diabetes registry, established in 1989.

 Patients are automatically added to the registry at the first indication of diabetes. Criteria include physician-entered inpatient or outpatient diagnoses of diabetes (

 code 250.xx), dispensing of an antihyperglycemia agent, an A1C ≥7% (lowered to 6.5% in 2010), or a fasting plasma glucose >125 mg/dL. Clinicians may remove patients from the registry if their inclusion is subsequently determined to be erroneous.

We identified 178 PCPs who were continuously employed during 2011 (

). Of these, 149 (83.7%) had at least 10 patients with diabetes in their panel of patients, and were thus caring for a total of 14,430 diabetics. The outcomes of interest were the probability that patients were in optimal glycemic control (outcome 1A: all A1C values <7%; N = 5823 [40.4%]) or poor control (outcome 1B: at least 1 A1C ≥9%; N = 2446 [17%]) in 2011. We chose a value of <7% based on the American Diabetes Association’s position statement that for many patients with diabetes, a reasonable glycemic goal is A1C <7%.

We identified a subset of the above population consisting of 107 PCPs with at least 4 patients with diabetes on their panel who were initially in optimal control (<7%) but had a subsequent A1C ≥7.5% from 2010 to 2011 (912 patients). Patients were required to have 1 year of follow-up available after the elevated A1C. We assessed the probability of treatment intensification, defined as the addition of a new antihyperglycemic medication (oral agent or insulin) or an increase in dosage of a current oral medication, within 90 days of the elevated A1C (outcome 2). We were unable to capture increases in insulin dosage among those already on insulin.

We examined the demographics of both patients and PCPs. For patients, we also examined time in the KPNW diabetes registry (which has been shown in previous analyses to be a proxy for duration of diabetes), the most recent A1C prior to the observation period, and insulin use. We determined whether patients had a visit to any PCP (not just their own PCP), an endocrinologist, or a specialist during the observation period, and we determined the Deyo-adaptation of the Charlson comorbidity index.

 PCP characteristics included duration of employment at KPNW, primary care training (internal medicine vs family practice), and degree (MD vs nurse practitioner/physician assistant). We also examined the total number of patients and percentage of patients with diabetes on their panels.

We report summary statistics to describe the characteristics of patients and PCPs. Continuous variables were summarized by medians and interquartile ranges and categorical variables by numbers and percentages. We used 2-level hierarchical linear models to assess patient and provider characteristics as predictors of glycemic control and treatment intensification after incident hyperglycemia. The first and second levels of the hierarchical linear models included patient and provider characteristics, respectively. We performed the following process for each of the 3 outcomes: first, we assessed bivariate associations within each of the 2 levels of the model. Secondly, we put together the level-1 patient variables that were statistically significant at a 

 <.10 level and that had enough variability to explain the outcome across almost all (99%) providers in the model. Third, we added all the level-2 provider variables as predictors of the intercept coefficient and evaluated if the provider variables explained additional variance in the outcomes above and beyond the patient characteristics. This last assessment was based on a χ

 test of the deviance difference of the models with and without the provider variables. Laplace estimations were used in all models due to the binary outcomes. Analyses were conducted using SAS version 9.2 (SAS Institute, Cary, North Carolina) and HLM version 7 for Windows (Scientific Software International, Inc, Lincolnwood, IL).

Patients had a median age of 63 years for outcome 1 (prevalent optimal and poor glycemic control) and 62 years for outcome 2 (probability of treatment intensification), and slightly more than half were men (

). Patients had been in the KPNW diabetes registry for an average of 5 years. Glycemic control was <7% for more than half of all patients; close to one-fourth were using insulin. Most saw a PCP, but very few (<2%) saw an endocrinologist.

About half of the PCPs for these patients with diabetes were men, with about an even split between internal medicine and family practice. The majority were MDs and they had worked at KPNW for a median of 6 to 8 years. They had a median of 1200 to 1300 patients on their panels, 8% to 9% of whom had diabetes.

Patient and PCP Characteristics Associated With Optimal Glycemic Control (all A1Cs <7%)

In bivariate analyses, patients who were older and who had at least 1 visit to a specialist other than an endocrinologist were more likely to be in optimal glycemic control (

). On the other hand, male patients with higher A1Cs at baseline, longer duration of diabetes, and insulin use were less likely to have optimal control. Patients with at least 1 visit to an endocrinologist were less likely to be in control; however, only 101 of the 149 PCPs had a mixture of patients who did and did not see an endocrinologist. Having a visit to a specialist other than an endocrinologist was associated with increased odds of being in optimal control. No PCP characteristics were associated with optimal control in the bivariate models. When we examined the patient characteristics in a multivariable model, older patients were still more likely to be in optimal control while patients with higher A1Cs at baseline, longer duration of diabetes, and insulin use were less likely to be in optimal control. For example, each additional year of age increased the odds of being in optimal control by 2% (odds ratio [OR], 1.023; 95% CI, 1.017-1.03; 

 <.001). Having a baseline A1C ≥7% decreased the odds of being in optimal glycemic control by 93% (OR, 0.07; 95% CI, 0.06-0.08; 

 <.001), and for each additional year with diabetes, it decreased by 7% (OR, 0.93; 95% CI, 0.92-0.95; 

 <.001). Users of insulin were almost half as likely to be in optimal control as nonusers of insulin (OR, 0.49; 95% CI, 0.42-0.58; 

 <.001). Adding provider characteristics to the multivariable model did not improve the predictive power of the models.

Patient and PCP Characteristics Associated With Poor Glycemic Control (at least 1 A1C ≥9%)

Patients who were older had a decreased risk of poor control while those who were male, had higher baseline A1Cs, had a longer duration of diabetes, and used insulin had a higher risk of poor control in bivariate models (

). Patients who had at least 1 visit with an endocrinologist were more likely to be in poor control, while having a visit to a specialist other than an endocrinologist was associated with decreased odds of being in poor control. No PCP characteristics were associated with poor control in the bivariate models. Based on the patient characteristic multivariable model, for each additional year of age, the odds of poor control decreased by about 4% (OR, 0.956; 95% CI, 0.949-0.962; 

 <.001). Having a baseline A1C ≥7% increased the odds of poor control by a factor of 12 (OR, 12.4; 95% CI, 9.6-15.9; 

 <.001), and for each year in the diabetes registry, odds of poor control increased by 3% (OR, 1.03; 95% CI, 1.01-1.05; 

 <.001). Users of insulin were more than twice as likely to have poor control than nonusers (OR, 2.1; 95% CI, 1.8-2.5; 

 <.001). Those who had at least 1 visit to a specialist other than an endocrinologist had 20% lower odds of having poor control (OR, 0.80; 95% CI, 0.68-0.95; 

 = .011). When we added PCP characteristics to the multivariable models, the PCP variables did not contribute additional information to the model that just included the significant patient characteristics.

Patient and PCP Characteristics Associated With Treatment Intensification Within 90 Days of Incident Hyperglycemia (elevated A1C >7.5%)

The distribution of the rate of therapy intensification approximated normal (

). Of the 912 patients who were in optimal glycemic control (<7%) followed by an elevated A1C (>7.5%), more than half (58.9%, n = 537) had their therapy intensified within 90 days of the elevated A1C. On average, 58.6% of a PCP’s patients who had been in optimal glycemic control had their therapy intensified within 90 days of an elevated A1C.

Results of the bivariate and multivariable hierarchical linear models of therapy intensification are displayed in 

. A greater change between the A1C below 7% and the subsequent A1C over 7.5% was most strongly associated with the probability of intensification (OR, 1.54; 95% CI, 1.25-1.91; 

 <.001). Older age also increased the probability of intensification (OR, 1.02; 95% CI, 1.00-1.03; 

 = .027) in the multivariable model. PCP characteristics were not associated with the probability of treatment intensification in bivariate models and did not improve the multivariable model.

Studies of poor glycemic control that seek to explain clinical inaction among individuals with diabetes have typically focused on patient characteristics because clinician data are rarely available. Those that focus on clinicians rely on surveys that do not link actual clinical data to their patients’ levels of glycemia.

 Our study may be the first to simultaneously evaluate the association of both patient and provider characteristics, including provider demographics and attributes of provider patient panels, with glycemic control and treatment intensification. We found that patient, rather than provider, characteristics were associated with the probability of optimal or poor glycemic control and treatment intensification when glycemic control is initially lost. These data suggest that systems of care may contribute more to glycemic control among patient populations than individual provider characteristics.

Patients who were older, started out with better glycemic control, had a shorter duration of diabetes, and did not use insulin were more likely to have better glycemic control. A greater increase in A1C coupled with older age were more likely to lead to treatment intensification. In addition, in bivariate models, having at least 1 visit to an endocrinologist during the observation period was associated with poor glycemic control while a visit to a specialist was associated with optimal glycemic control. Patients having difficulty with diabetes management may be more likely to visit an endocrinologist, which would explain the observed association of endocrinology visits with poor control.

It might seem counterintuitive that patients seeing nonendocrinology specialists had improved glycemia; we had expected that visiting other specialists would be associated with worse glycemic control as such patients would be complex with multiple comorbidities. We hypothesize several possible explanations for this unexpected finding. First, PCPs might have been more able to focus on a patient’s diabetes management because other medical problems were being addressed by specialist providers. Another possibility is that having multiple providers encourage lifestyle changes and medication compliance on various encounters may be more effective than just a single provider’s advice. Patients under specialist care may also be more willing to modify their diet or exercise or improve medication compliance because they want to improve a more clinically apparent condition (eg, chest pain, painful knee osteoarthritis) than their “silent” diabetes. Finally, patients who have been able to navigate the health system to order to see a specialist might be more concerned about their health.

In our data, nearly 60% of patients who lost glycemic control after experiencing optimal control had treatment intensified within 3 months. This relatively rapid intensification is more aggressive than has been reported in previous research, including studies conducted in the current setting. For example, in a study conducted when A1C >8% was the trigger for intensification, patients waited an average of 15 to 26 months before their medication was changed, even though 2 to 4 A1C measures were elevated.

 Treatment delays become even longer when the target A1C was <7%. In a recent study from the United Kingdom, it took a median of nearly 3 years for a second medication to be added when A1C was >7% and over 7 years for a third medication to be added.

 However, these prior studies considered only changes from one therapy to another, while our analyses included dosage increases as intensification efforts. This assessment of dosage changes of oral hyperglycemic therapies was a unique strength of this study. Because dosage changes are often not included in studies of treatment intensification, it is not surprising that we found a higher and likely more accurate rate of treatment intensification.

Our study design also required that patients be in optimal control at the start of the observation period and to subsequently experience a minimum of a 0.5 percentage point change in A1C to be eligible for intensification. Thus, we were observing an “incident” loss of glycemic control that was relatively large. Such a change in A1C may be an important trigger for intensifying therapy. Indeed, we found that a change in A1C exhibited the strongest association with treatment intensification. In any event, it seems that chronic hyperglycemia, rather than “incident” loss of control that we studied, may account for much of the long delay in treatment intensification that has been reported elsewhere.

By studying patients with an apparent incident loss of glycemic control, we may have inadvertently included patients who were relatively early in their progression of dysglycemia and, thus, had less comorbidity. This is an important consideration for 2 reasons: first, previous research has shown that among patients with A1C >7%, each additional medical concern raised by the patient reduced the probability of a medication change by 49%.

 Second, among patients on 1 to 3 oral medications who are in a more advanced stage of their disease, insulin may not be initiated for an average of 6 to 7 years after an elevated A1C,

 with an average A1C at insulin initiation of 9.8%.

 Once on insulin, patients still experience low rates of therapy intensification even though insulin can be titrated up to achieve the desired A1C; 70% and 40% of patients on insulin have been shown to have A1C levels above 7% and 8%, respectively.

A unique strength of our study was the ability to assess comprehensive characteristics of clinicians as well as patients and to link those characteristics to glycemic control. To our knowledge, ours is the first such study. We did not find associations between provider characteristics and glycemic level of treatment intensification. This could have been the result of the organizational structure of the study site. Although we are unable to measure it, we know that there is considerable “sharing” of patients across providers. Providers often see other providers’ patients for urgent visits or cover urgent issues when a provider is not available. In the treatment intensification model, the variability in the outcome 

 PCPs was greater or equal to the variability 

 providers, illustrating that variability in the outcome is attributable mostly to patients.

Previous studies suggest that such a lack of variability among providers also exists in other health systems. A study of the Veteran Affairs system found there was almost no provider-level variation in indicators of glycemic control

; rather, most of the variation was attributable to the facility level (patient factors were not examined). This has implications for quality measurement. If little of the variation in diabetes quality measures is attributable to providers, provider quality scores will be inaccurate unless the number of patients with diabetes on a provider’s panel is large. The focus of quality improvement should be on facility-level performance variations.

The lack of provider variability and the lack of an association between provider characteristics and glycemic outcomes also suggest that system level factors may have more influence over the glycemic level of the population than individual provider characteristics. Improving glycemia may be best done by developing indicators or programs that promote systemwide clinical actions such as institutional glycemic targets or protocols for support staff outreach for dysglycemia. For example, KPNW supplements the electronic medical record with a panel support tool that includes an alert to clinicians when individual patients exceed a prespecified A1C goal. Clinicians also have online point-of-care access to internal diabetes guidelines to assist their treatment decision making. More health systems should consider implementing such systemwide programs and future research should focus on how these and other system level factors influence glycemic outcomes.

Many more patient than provider characteristics are associated with optimal or poor glycemic control and treatment intensification when glycemic control is initially lost.

Sumesh Kachroo, PhD

Articles

Oral Anticoagulant Discontinuation in Patients With Nonvalvular Atrial Fibrillation

Do Patient or Provider Characteristics Impact Management of Diabetes?