Outcomes associated with warfarin time in therapeutic range among US veterans with nonvalvular atrial fibrillation

Abstract

Background: Poor quality of warfarin control (time in therapeutic range [TTR] < 65%) can lead to increased risk of adverse events. The objective of this study was to examine the overall quality of international normalized ratio (INR) control and the association of TTR with clinical outcomes including stroke, major bleeding, and all-cause mortality among US warfarin users.

Methods and results: This retrospective observational cohort study utilized the US Veterans Affairs electronic medical records database (VA EMR). Patients with NVAF who newly initiated warfarin from 1 January 2005 to 31 December 2015 were grouped into two cohorts based on TTR <65% or ≥65%. TTR was computed from INR test results. Clinical outcomes assessed were stroke/systemic embolism (SE), hemorrhagic stroke, ischemic stroke, and major bleeding, defined based on hospitalization with those conditions as primary diagnosis, as well as all-cause mortality. Patients were followed from warfarin initiation to the first occurrence of an outcome or censoring. Propensity score weighted time-varying Cox regression was used to evaluate the risk of the clinical events. A total of 127,385 NVAF patients with mean TTR of 51% were included. TTR <65% was observed in 65% of patients. Mean CHA₂DS₂-VAS_C score (SD) was 2.9 (1.5) in the low TTR cohort and 2.7 (1.4) in the high TTR cohort. Patients with TTR <65% had a higher risk for any stroke/SE (HR: 1.57; 95% CI: 1.41–1.75), major bleeding (HR: 2.78; 95% CI: 2.55–3.03) and all-cause mortality (HR: 1.73; 95% CI: 1.67–1.79).

Conclusions: The observed quality of warfarin control in VA EMR suggests room for improvement given the association with elevated risk of adverse clinical outcomes.

Keywords:

• Atrial fibrillation
• hemorrhage
• international normalized ratio
• stroke
• veterans

Introduction

Atrial fibrillation (AF), the most common type of cardiac arrhythmia, poses a significant public health burden globally, with a prevalence of 596.2 per 100,000 males and 373.1 per 100,000 females¹. As the aging population increases, AF is predicted to affect as many as 12.1 million Americans by 2030².

Considering the public health and clinical burden of stroke, bleeding, and other thromboembolic events, anticoagulants have remained the mainstay of therapy to balance the effects of thromboembolism with the risk of hemorrhage. For decades, warfarin has proven efficacious in the treatment of thromboembolism in non-valvular atrial fibrillation (NVAF)^3,4. Dose-adjusted warfarin has been shown to reduce stroke by 64%⁵; however, there are several therapeutic challenges associated with warfarin use. Variability of warfarin dosing has been reported to lead to under- and over-anticoagulation^6–8. Patients who use warfarin should maintain the international normalized ratio (INR) within a relatively narrow therapeutic window between 2.0 and 3.0⁶; the failure to maintain INR within this threshold has been associated in routine medical practice with excess morbidity and higher incidences of stroke, bleeding, and mortality compared to those who succeeded⁹. Therefore, frequent INR monitoring becomes imperative for the optimal management of AF and thromboembolism but represents a burden for the patient among warfarin users. With the availability of non-vitamin K oral anticoagulants (NOACs), it is critical to identify patients with poor warfarin control early to allow physicians to make appropriate anticoagulation treatment decisions before the development of therapeutic consequences.

Time in therapeutic range (TTR), the proportional time that INR values of a patient are between 2 and 3, has been used in observational studies to examine the adequacy of warfarin treatment¹⁰. Good anticoagulation control (TTR ≥65% or 70%) is an important determinant of protection against ischemic stroke and major hemorrhage events¹¹. Unfortunately, TTR has been reported as 55–65% in most clinical trials and approximately 50% in community settings^12–15. In light of these results, improving the quality of INR control may be necessary to achieve high TTRs over a patient’s lifetime¹⁶. Moreover, there is a dearth of real-world evidence on the quality of INR control based on TTR and its impact on clinical outcomes such as major bleeding, stroke, and mortality.

Using a retrospective cohort study, we examined the overall quality of INR control and the association of TTR with clinical outcomes including major bleeding, stroke, and all-cause mortality among US veterans.

Methods

Study design

This study was a retrospective observational cohort study based on National Veteran Affairs (VA) electronic medical record (EMR) data including patients’ records of pharmacy, inpatient, outpatient, and lab results from 1 January 2004 to 31 December 2015. The Veterans Health Administration is the largest integrated health care system in the United States, providing care for veterans across the country at affordable prices. According to US Census Bureau, there were approximately 19 million living US veterans in 2014¹⁷.

This observational study was conducted under the provisions of Privacy Rule 45 CFR 164.514(e), and was exempt from Investigational Review Board review and approval since there was no collection or use of personally identifiable information in the conduct of this study¹⁸.

Patient selection

Patients were included if they had ≥1 inpatient claim or 2 outpatient claims with the diagnosis code for AF (ICD-9-CM code 427.31), were aged ≥18 years, and had ≥1 outpatient warfarin medication fill after the first AF diagnosis. The index date was designated as the first warfarin prescription claim between 1 January 2005 and 30 November 2015.

Patients were required to have ≥13 months of continuous enrollment (12 months pre-index date and ≥1 month post-index date) between 1 January 2005 and 30 November 2015 after the first AF diagnosis, preceded by no anticoagulant use during the 1 year pre-index (baseline) period.

Patients were excluded from the study if they were diagnosed with chronic rheumatic heart disease or valvular heart disease; had a heart valvular replacement, only had an inpatient warfarin prescription fill during study period, or had outpatient prescription fill for any anticoagulant during the baseline period; or had diagnosis of venous thromboembolism (VTE) at any time prior to or on the index date (Supplemental Table 1). Also, patients were excluded if they had temporary AF or contraindications for long-term oral anticoagulant (OAC) therapy (e.g. transient AF, hyperthyroidism, or thyrotoxicosis).

Patient data was assessed from the warfarin initiation date to the therapy discontinuation date (i.e. the last warfarin prescription fill date during the study period + days of supply of that fill +45 days), initiation of non-warfarin anticoagulant (NOACs and low molecular weight heparin [LMWH], unfractionated heparin [UFH], fondaparinux), death, disenrollment, or end of study period, whichever occurred first.

Study measures

Baseline socio-demographic and clinical characteristics during the 12 months prior to index warfarin initiation were measured. Baseline characteristics included demographics, clinical setting, calendar year of warfarin initiation, CHADS₂ and CHA₂DS₂-VASc scores, and comorbidities.

TTR was calculated using the Rosendaal method, and was computed using INR values that were comprehensively recorded in the VA database¹⁹. If INR values were more than 45 days apart within a period of warfarin exposure, the INR value was considered unknown during the time-period from 45 days after the preceding INR value until the next INR value. Overall INR control quality was based on the percentage of time a patient remained in therapeutic range, measured over the entire follow-up period, and was classified as follows for the main comparison cohorts: good INR control, TTR ≥65%; and poor INR control, TTR <65%²⁰. The outcome measures were mortality, stroke/systemic embolism (SE), and major bleeding, defined using hospital claims, which had a bleeding or stroke/SE diagnosis, respectively, as the first listed ICD-9 diagnosis code (Supplemental Table 2). Stroke subcategories were evaluated for hemorrhagic and ischemic stroke. The diagnosis codes used for stroke and major bleeding were based on a validated administrative-claim-based algorithm as well as clinical trial definitions of major bleeding and stroke^14,21,22.

Statistical analysis

Baseline and outcome variables were analyzed descriptively. Counts, percentages, means, medians, and standard deviations were provided for appropriate variables.

The crude incidence rates, and univariate hazard ratios (HRs) for the risk of outcomes of interest were calculated by warfarin control quality (TTR <65% vs. TTR ≥65%). The incidence rates of stroke/SE and major bleeding were calculated as the number of first stroke/SE and major bleeding events, respectively, divided by the total time at risk within the study period and described as the number of stroke/SE and major bleeding events per 100 person-years. Kaplan–Meier survival curves and log-rank testing was used to compare the differences in time to event between the groups.

To estimate the effect of TTR on outcomes, we used propensity score methods to reduce confounding of measured covariates between patients with poor and good warfarin control. Propensity scores were estimated using a logistic regression model with the variables: age, gender, race, US geographic region, calendar year of index date, clinical setting of warfarin initiation, comorbidities and CHA₂DS₂-VASc score²³. We computed the inverse probability of being in the poor warfarin control group weights from the propensity scores²³. All variables were balanced after the propensity score weighting (Supplemental Table 3).

Propensity score weighted time-varying Cox models were used for all outcomes controlling for baseline demographic and clinical covariates²⁴. The overall quality of warfarin control was considered as a categorical predictor (i.e. good vs. poor warfarin control) for the clinical outcomes. Clinical covariates included hypertension (ICD-9-CM: 401-405, 362.11 or on antihypertensive medication), diabetes (ICD-9-CM: 250.xx or on antidiabetic medicine), prior stroke or TIA (ICD-9-CM: 430–436), heart failure (ICD-9-CM: 428.xx, 402.01, 402.11, 402.91, 404.01, 404.03, 404.13, 404.91, 404.93), coronary artery disease (ICD-9-CM: 410.x–414.xx, 429.2), peripheral vascular disease (ICD-9-CM: 440.2x, 440.3x, 440.4, 440.8, 440.9, 443.0, 443.1, 443.8x, 443.9, 444.2x, 444.9, 445.x, 362.30–362.36), prior thromboembolism (ICD-9-CM: 325, 362.3x, 410.xx, 411.1, 411.81), chronic kidney disease (stage I–IV, ICD-9-CM: 585.1–585.4), renal failure (CKD stage V or dialysis or ESRD, ICD-9-CM: 585.5, 585.6, 285.21, 403.01, 404.02, 404.03, 584.5–584.9, 753.13), obesity (ICD-9-CM: 278), dyslipidemia (ICD-9-CM: 272.0, 272.2, 272.4), atherosclerosis (ICD-9-CM: 440.xx), mental disorder (ICD-9-CM: 290–319), cirrhosis/hepatitis (ICD-9-CM: 070.2–070.9, 570, 571.x, 572.2, 572.3, 572.4, 572.8, 573.x), prior GI/GU hemorrhage (ICD-9-CM: 578.x, 531.0, 531.2, 531.4, 531.6, 532.0, 532.2, 532.4, 532.6, 533.0, 533.2, 533.4, 533.6, 534.0, 534.2, 534.4, 534.6, 535.x1, 530.82, 456.0, 456.2, 569.3, 596.7), prostate cancer (ICD-9: 185.xx), lung cancer (ICD-9: 162.xx), predisposition to falls (ICD-9-CM: 290.x–294.x, 331.0, 331.1, 333.4, 345.x, 347, 458.0, 780.2, 780.3, E880–888), and hypertrophic cardiomyopathy (HCM: ICD-9-CM: 425.1). Time-varying co-medication covariates were anti-platelet agents, angiotensin receptor blockers, angiotensin converting enzyme inhibitors, beta-blockers, antiarrhythmic drugs, statins, proton pump inhibitors, and H2-receptor antagonists. Co-medication use was examined during the 12 month baseline and updated every 6 months over follow-up.

Sensitivity analyses were also performed with the TTR cut-point of 60% and 70% as alternatives of poor and good warfarin control. The corresponding propensity score weights were also estimated and applied to time-varying Cox models.

All analyses were carried out using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). Values of p <.05 were considered, a priori, to be statistically significant.

Results

Of the 1,240,405 AF patients in the national VA database between 1 January 2005 and 30 November 2015, 127,385 NVAF patients were selected; 44,347 (34.8%) warfarin patients with TTR ≥65% and 83,038 (65.2%) warfarin patients with TTR <65% were identified after applying the appropriate inclusion and exclusion criteria (Figure 1). Overall, patients were followed for a median of 1.4 years (interquartile range: 3.3 years).

Figure 1. Sample attrition in the VA population: AF patients prescribed warfarin with TTR ≥65% vs. TTR <65%. Abbreviations. AF, atrial fibrillation; ICD-9-CM, International Classification of Disease, 9th revision, Clinical Modification code; OAC, oral anticoagulant; TTR, time in therapeutic range.

Patients had average ages of 71 ± 10 years for TTR <65% and 72 ± 9 years for TTR ≥65%, with the majority of patients in each cohort being ≥75 years (Table 1). Patients with TTR <65% had a significantly higher mean CHA₂DS₂-VAS_C score compared to those with TTR ≥65% (2.1 ± 1.3 vs. 1.9 ± 1.2, p < .001). Patients with poor warfarin control were more likely to have comorbidities, including heart failure (30.8% vs. 21.0%, p < .001), coronary artery disease (48.7% vs. 41.8%, p < .001), peripheral vascular disease (16.3% vs. 11.9%, p < .001) and renal failure (6.8% vs. 3.0%, p < .001).

Table 1. Baseline characteristics of warfarin patients comparing TTR <65% and TTR ≥65% among US veterans.

	TTR <65% N = 83,038		TTR ≥65% N = 44,347
	N/Mean	%/SD	N/Mean	%/SD	p-value
Age (years)	71.0	10.1	71.9	9.2	<.001
Age categories					<.001
<65	24,110	29.0%	10,854	24.5%	–
65–74	27,325	32.9%	15,761	35.5%	–
≥75	31,603	38.1%	17,732	40.0%	–
Gender					<.001
Female	1266	1.5%	550	1.2%	–
Male	81,722	98.5%	43,797	98.8%	–
Geographic region					<.001
Northeast	11,639	14.0%	6651	15.0%	–
Midwest	21,092	25.4%	11,280	25.4%	–
South	32,812	39.5%	16,571	37.4%	–
West	16,723	20.1%	9487	21.4%	–
Other/unknown	772	0.9%	358	0.8%	–
Clinic settings					<.001
Cardiac	4392	5.3%	2122	4.8%	–
Internal medicine	27,257	32.8%	17,643	39.8%	–
Ward	23,122	27.9%	6205	14.0%	–
Other	27,103	32.6%	17,879	40.3%	–
Unknown	1164	1.4%	498	1.1%	–
CHADS2 score	2.1	1.3	1.9	1.2	<.001
CHA2DS2-VASc score	2.9	1.5	2.7	1.4	<.001
CHA2DS2-VASc score ≥2	68,602	82.6%	36,558	82.4%	.420
Medical history at baseline					–
Hypertension	64,750	78.0%	34,534	77.9%	.690
Diabetes	32,938	39.7%	15,301	34.5%	<.001
Prior stroke or TIA	9404	11.3%	4244	9.6%	<.001
Heart failure	25,600	30.8%	9296	21.0%	<.001
Coronary artery disease	40,455	48.7%	18,554	41.8%	<.001
Peripheral vascular disease	13,559	16.3%	5287	11.9%	<.001
Prior thromboembolism	12,450	15.0%	4639	10.5%	<.001
Chronic kidney disease (stage I–IV)	3710	4.5%	1366	3.1%	<.001
Renal failure (CKD stage V or dialysis or ESRD)	5634	6.8%	1350	3.0%	<.001
Obesity	21,831	26.3%	11,561	26.1%	.390
Dyslipidemia	50,819	61.2%	28,847	65.1%	<.001
Atherosclerosis	4832	5.8%	1694	3.8%	<.001
Mental disorder	40,912	49.3%	17,545	39.6%	<.001
Cirrhosis/hepatitis	4178	5.0%	1460	3.3%	<.001
Prior GI/GU hemorrhage	4378	5.3%	1213	2.7%	<.001
Prostate cancer	6333	7.6%	3283	7.4%	.140
Lung cancer	1692	2.0%	415	0.9%	<.001
Predisposition to falls	16,679	20.1%	5802	13.1%	<.001
Hypertrophic cardiomyopathy	321	0.4%	141	0.3%	.059

Abbreviations. CKD: chronic kidney disease; ESRD: end-stage renal disease; GI/GU: gastrointestinal and urogenital; SD: standard deviation; TTR: time in therapeutic range.

Overall, patients had a mean TTR of 51% with a median of 56%. The mean TTR for patients with poor and good warfarin control was 38% and 77%, respectively. For the sensitivity analysis, 44% and 26% of patients had TTR ≥60% and ≥70%, respectively. During the follow-up, on average 34% of the time patients had INR values >3 and 14% of the time warfarin patients had INR values <2. In addition, 7.6% of patients never achieved the therapeutic INR range of 2–3. The TTR distribution is shown in Supplemental Figure 1.

The incidence rate of any stroke/SE (0.77 vs. 0.27 per 100 person years [PY]) and major bleeding (1.92 per 100 PY vs. 0.38 per 100 PY) was higher among patients with poor warfarin control compared to patients with good warfarin control (Table 2). Lastly, all-cause mortality was 6.59 per 100 PY in TTR <65% vs. 3.04 per 100 PY in TTR ≥65%. The Kaplan–Meier curves are shown in Supplemental Figures 2–6.

Table 2. Incidence rates and adjusted hazard ratios for risk of clinical outcomes with TTR <65% vs. TTR ≥65%.

Outcomes	Crude (Unadjusted) Incidence Rates per 100 Person-years		Hazard Ratio (95% CI)^a
	TTR <65% N = 83,038	TTR ≥65% N = 44,347	Crude	p-value	Adjusted^b	p-value
Any stroke/systemic embolism	0.77	0.27	2.18 (1.95–2.43)	<.001	1.57 (1.41–1.75)	<.001
Hemorrhagic stroke	0.07	0.03	1.98 (1.42–2.75)	<.001	1.60 (1.15–2.23)	.006
Ischemic stroke	0.61	0.23	2.07 (1.83–2.34)	<.001	1.51 (1.34–1.70)	<.001
Major bleeding	1.92	0.38	4.06 (3.71–4.43)	<.001	2.78 (2.55–3.03)	<.001
Death	6.59	3.04	2.11 (2.03–2.18)	<.001	1.73 (1.67–1.79)	<.001

^a ≥65% served as the reference group.

^b Adjusting for age, sex, race, region, clinic, index year, medical history (hypertension, diabetes, prior stroke or TIA, heart failure, coronary artery disease, peripheral vascular disease, prior thromboembolism, chronic kidney disease [stage I–IV], renal failure [CKD stage V or dialysis or ESRD], obesity, dyslipidemia, atherosclerosis, mental disorder, cirrhosis/hepatitis, prior GI/GU hemorrhage, prostate cancer, predisposition to falls, hypertrophic cardiomyopathy) and time-varying co-medication use (anti-platelet agents, angiotensin receptor blockers, angiotensin converting enzyme inhibitors, beta-blockers, antiarrhythmic drug, statins, proton pump inhibitors, and H2-receptor antagonists).

Abbreviations. CI: confidence interval; TTR: time in therapeutic range.

Patients with TTR <65% had a significantly higher unadjusted risk of stroke/SE (HR: 2.18; 95% CI: 1.95–2.43, p < .001), major bleeding (HR: 4.06; 95%CI: 3.71-4.43, p < .001) and all-cause mortality (HR: 2.11; 95% CI: 2.03–2.18, p < .001) than patients with TTR ≥65%.

After propensity score weighting, the baseline demographic and clinical variables were well balanced between poor and good warfarin control groups (Supplemental Table 3). During the follow-up, patients with TTR <65% were associated with a 57% increased risk for stroke/SE compared to patients with TTR ≥65% (HR: 1.57; 95% CI: 1.41–1.75, p < .001). In addition, patients with TTR <65% had a higher risk of ischemic (HR: 1.51; 95% CI: 1.34–1.70, p < .001) and hemorrhagic stroke (HR: 1.60; 95% CI: 1.15–2.23, p = .006). Patients with TTR <65% were associated with an increased risk for major bleeding compared to patients with TTR ≥65% (HR: 2.78; 95% CI: 2.55–3.03, p < .001). In addition, patients with poor warfarin control were associated with a 73% increased risk for all-cause mortality compared to patients with good warfarin control (HR: 1.73; 95% CI: 1.67–1.79, p < .001).

As a sensitivity analysis, TTR ≥60% and TTR ≥70% were used to evaluate warfarin control; the outcomes remained consistent with the main analysis. Patients with poor warfarin control had significantly higher risk of major bleeding, stroke/SE, and all-cause mortality (Table 3).

Table 3. Sensitivity analysis: adjusted hazard ratios for risk of clinical outcomes with TTR <60% and TTR <70%.

Outcomes	Adjusted Hazard Ratio (95% CI)^a
	TTR <60%	p-value	TTR <70%	p-value
Any stroke/systemic embolism	1.56 (1.41–1.72)	<.001	1.46 (1.29–1.64)	<.001
Hemorrhagic stroke	1.90 (1.39–2.62)	<.001	1.52 (1.05–2.21)	.026
Ischemic stroke	1.47 (1.31–1.64)	<.001	1.39 (1.22–1.59)	<.001
Major bleeding	2.76 (2.55–2.97)	<.001	3.20 (2.88–3.56)	<.001
Death	1.86 (1.81–1.93)	<.001	1.61 (1.55–1.67)	<.001

^aAdjusting for age, sex, race, region, clinic, index year, medical history (hypertension, diabetes, prior stroke or TIA, heart failure, coronary artery disease, peripheral vascular disease, prior thromboembolism, chronic kidney disease [stage I–IV], renal failure [CKD stage V or dialysis or ESRD], obesity, dyslipidemia, atherosclerosis, mental disorder, cirrhosis/hepatitis, prior GI/GU hemorrhage, prostate cancer, predisposition to falls, hypertrophic cardiomyopathy) and time-varying co-medication use (anti-platelet agents, angiotensin receptor blockers, angiotensin converting enzyme inhibitors, beta-blockers, antiarrhythmic drug, statins, proton pump inhibitors, and H2-receptor antagonists).

Abbreviation. TTR: time in therapeutic range.

Discussion

The current study was a real-world, retrospective observational analysis among US veterans diagnosed with NVAF in order to examine the relative impact of TTR among warfarin users on clinical outcomes risks, including stroke/SE, major bleeding, and all-cause mortality. In the VA population, poor warfarin control was associated with elevated risk of adverse clinical outcomes.

To date, very few real-world observational studies have explored and addressed how TTR predicts the risk of clinical outcomes relevant to warfarin use in the VA population. A recent study in general outpatient clinics showed that AF patients on warfarin had a mean TTR value of 56%, and less than half had TTR >60% with a mean age of 57 years²⁵. In our study, we found a mean TTR of 51% among veterans with a mean age of 70 years. A more recent registry study demonstrated a higher mean TTR of 65 ± 20% with a mean age of 75 years²⁶, which was a closer demographic compared to our sample. However, these results may not necessarily be reflective of the broader US population. Past research has also shown a strong association between low TTR <65% and increased risk of clinical events. Senoo et al. demonstrated that TTR ≥65% was associated with a 78% decreased risk for a combined endpoint of cardiovascular death/stroke and SE events²⁷. A study using the Kaiser Permanente Southern California database showed that patients with TTR ≥55% were associated with a 73% and 60% lower risk of stroke/SE and major bleeding, respectively, which supports the findings in our analysis²⁸. A prior study has shown that warfarin users with a ≥70% TTR had a 79% lower risk of stroke compared to those with TTR ≤30%²⁹. After adjusting for clinical and demographic variables, our multivariate results showed that patients with TTR <65% had a 57% increased risk for major bleeding compared to TTR ≥65% (HR: 1.57; 95% CI: 1.41–1.75, p < .001). A study published in 2013 demonstrated that as TTR decreases from 70% to 30%, the risk for mortality was shown to increase from 80% to as much as 280% among warfarin users²⁹. In our study, we found that TTR ≤65% was associated with a 73% increased risk of death compared to TTR ≥65% (HR: 1.73, 95% CI: 1.67–1.79, p < .001). Our results align with other literature in the past and provide evidence of similar poor outcomes associated with low TTR in the VA database.

Our analysis was conducted from a large, comprehensive administrative national VA EMR database that offers a large sample of elderly veterans to evaluate the association between the quality of INR control and the risk of stroke/SE, major bleeding, and all-cause mortality. In this database, we were able to comprehensively evaluate the clinical relevance of maintaining INR within the desired range. Mortality was evaluated since the VA has death information on all veterans. While the Veterans Affairs Study to Improve Anticoagulation (VARIA) study used the same data sources to identify patient-level predictors of oral anticoagulation control defined as TTR in the VA outpatient settings³⁰, our study examined the association of TTR with clinical outcomes including stroke, major bleeding, and all-cause mortality.

There are several important limitations of this study. First, this was a retrospective cohort study, so we cannot control the exposure or outcome assessment – some ICD-9-CM codes may be incorrectly recorded, misused, or omitted. We also acknowledged the limitation of this study’s retrospective design due to its inability to fully capture defined events (e.g. events that occurred outside the VA system). Because of the study design’s retrospective nature, we cannot randomly assign treatment groups. However, the use of propensity score methods might reduce the confounding of measured covariates between the two treatment groups.

Second, the variables’ true validity may be uncertain using claims and there is the possibility of residual confounding in the modeling results. The administrative databases may lack information on some critical covariates such as over-the-counter drug use and disease duration. Furthermore, it was not possible to capture INR testing outside the VA system in this analysis. Lastly, since this analysis was among VA patients, study findings cannot be generalized to other populations. VA patients are predominantly male and elderly; hence, they are not representative of the entire US population. Furthermore, other clinical information such as medication adherence is limited in the VA EMRs. The presence of a claim for a filled prescription does not indicate that the medication was consumed or that it was taken as prescribed.

The results show that the TTR plays a vital role in predicting cardiovascular outcomes and mortality among elderly veteran NVAF patients treated with warfarin. On average, patients were in the required INR threshold for only half of the total time and close to 10% of the population were never in the required range of 2–3. The low quality of INR control in this population suggests an opportunity for improvement. After adjusting for several relevant demographic and clinical covariates, our analysis showed that poor warfarin control in the VA population was associated with an increased risk for stroke/SE, major bleeding, and all-cause mortality compared to good warfarin control. A sensitivity analysis (data not shown) excluding those who had never reached target range (i.e. TTR = 0, N = 9647) still showed robust results on these adverse clinical outcomes.

Our analysis emphasizes the need for appropriate and frequent INR monitoring among the elderly in the early phases of clinical practice among new warfarin users. Poorly controlled patients may benefit from a switch to a more convenient and at-least-as-effective NOAC regimen, particularly those AF patients with a high-risk profile for stroke or major bleeding. It is important to communicate the benefits of continued (or regular) warfarin monitoring which can optimize the TTR of patients receiving warfarin. As TTR has several disadvantages in assessing anticoagulation management³¹ and is not readily available to physicians in real-world practice, further research is needed to find a better measure in predicting warfarin control so clinicians can decide whether to switch patients to NOACs or make efforts to maintain appropriate INR control.

Transparency

Declaration of funding

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

Declaration of financial/other relationships

S.L., Q.S. and L.S. have disclosed that they are employees of Tulane University School of Public Health and Tropical Medicine, and received a research grant in connection with conducting this study. X.L., M.H., K.F. and S.H. have disclosed that they are employees of Bristol-Myers Squibb Company with ownership of stocks in Bristol-Myers Squibb Company. Y.Z. has disclosed that he is an employee of Xavier University of Louisiana. R.H. has disclosed that he is a former employee of Pfizer Inc., with ownership of stocks in Pfizer Inc.

CMRO peer reviewers on this manuscript have received an honorarium from CMRO for their review work, but have no relevant financial or other relationships to disclose.

Acknowledgements

Allison Keshishian, Neel Vaidya, and Michael Moriarty of STATinMED research provided medical writing and editorial support which was funded by Bristol-Myers Squibb and Pfizer.