The Impact of Body Mass Index and Genetics on Warfarin Major Bleeding Outcomes in a Community Setting

Abstract

Purpose

Several studies have demonstrated an association between body mass index (BMI) and warfarin therapeutic dose, but none evaluated the association of BMI with the clinically important outcome of major bleeding in a community setting. To address this evidence gap, we conducted a case-control study to evaluate the association between BMI and major bleeding risk among patients receiving warfarin.

Methods

We used a case-control study design to evaluate the association between obesity (BMI >30.0) and major bleeding risk among 265 cases and 305 controls receiving warfarin at Group Health, an integrated healthcare system in Washington State. Multivariate logistic regression was used to adjust for potential confounders derived from health plan records and a self-report survey. In exploratory analyses, we evaluated the interaction between genetic variants potentially associated with warfarin bleeding (CYP2C9, VKORC1, and CYP4F2) and obesity on the risk of major bleeding.

Results

Overall, the sample was 55% male, 94% Caucasian, and mean age was 70 years. Cases and controls had an average of 3.4 and 3.7 years of warfarin use, respectively. Obese patients had significantly lower major bleeding risk relative to non-obese patients (Odds Ratio (OR): 0.60, 95% Confidence Interval (CI): (0.39–0.92). The OR was 0.56 (95% CI: 0.35–0.90) in patients with ≥1 year of warfarin use, and 0.78 (95% CI: 0.40–1.54) in patients with <1 year of warfarin use. An exploratory analysis indicated a statistically significant interaction between CYP4F2*3 genetic status and obesity (p=0.049), suggesting a protective effect of obesity on the risk of major bleeding among those wild type for CYP4F2*3, but not among variants.

Conclusions

Our findings suggest that BMI is an important clinical factor in assessing and managing warfarin therapy. Future studies should confirm the major bleeding associations, including the interaction between obesity and CYP4F2*3 status, identified in this study and evaluate potential mechanisms.

INTRODUCTION

Warfarin (Coumadin®) is a commonly prescribed anticoagulant that is highly effective in reducing blood clotting events in patients with medical conditions that confer increased risk (e.g. atrial fibrillation, myocardial infarction, and heart valve replacement).^1–6 However, warfarin management is difficult because giving too high of a dose can result in increased risk of major bleeding (e.g. cerebral hemorrhage), giving too low of a dose can increase risk of serious clots (e.g. stroke), and dose requirements are highly variable. For example, dose requirements are impacted by concomitant drug use, diet, alcohol use, and genetic status.^{4, 7, 8} Collectively, these factors create a significant barrier to appropriate use of this otherwise effective and inexpensive drug that prevents potentially serious clotting events. This is important at a population level because an estimated 2 million Americans currently use warfarin, despite the availability of newer anticoagulant medications.^{2, 4, 9} Accordingly, it is critical to continue to identify warfarin major bleeding risk factors to improve warfarin outcomes, enhance patient and provider confidence, and provide high value anticoagulation care.

Studies have shown that body mass index (BMI) impacts warfarin volume of distribution and clearance, and is positively correlated with coagulation factor levels (VII, VIIIc, fibrinogen)^{10, 11} —suggesting potential mechanisms for BMI to impact clinical outcomes. For example, Mueller and colleagues demonstrated that for each 1-point increase in BMI, the average weekly therapeutic dose of warfarin increased by 0.69 milligrams.¹² Because a typical weekly warfarin dose is 30–40mg, not accounting for BMI can result in clinically important deviations in dose. No studies to date have evaluated if this relationship extends to the clinically important warfarin outcome of major bleeding risk.^10–14

Genetics also play an important role in warfarin outcomes. Recent studies have identified three genes that impact warfarin dose requirements and major bleeding risk–CYP2C9, VKORC1, and CYP4F2.^{4, 7, 15–18} CYP2C9 is primarily responsible for the metabolism of warfarin, VKORC1 is the warfarin drug target, and CYP4F2 is associated with enzymatic activity in the vitamin K/warfarin pathway and may also be associated with vitamin K metabolism.^{4, 13, 16, 17} Variants of CYP2C9 (CYP2C9*2, rs17998523 or *3, rs1057910) and VKORC1 (1173G>A, rs9934438) have been associated with increased major bleeding risk^{19, 20}, while a variant of CYP4F2 (*3, rs2108622) has been associated with decreased risk.^{6, 18} However, no studies have evaluated gene-obesity interactions and major bleeding risk.

The primary objective of this analysis is to evaluate the association between BMI and warfarin major bleeding risk. We also conducted exploratory analyses to evaluate interactions between CYP2C9, VKORC1, and CYP4F2 variants, obesity, and major bleeding risk. The results have the potential to enhance understanding about the clinical significance of BMI in warfarin therapy and inform clinical strategies to mitigate major bleeding risk.

METHODS

Study Design

This retrospective case-control study was conducted using data from an existing case–control study among patients receiving warfarin therapy. The parent study (Warfarin Investigation for Safety and Health, or WISH) was designed to evaluate the association between genetic variants that influence warfarin dose requirements and major bleeding risk.⁶ Study participants were recruited from Group Health (GH), a nonprofit integrated health-care system that insures and provides medical care to approximately 600,000 patients in Washington State.⁶ The sample included 265 cases that experienced a major bleeding event while receiving warfarin and 305 similar controls that did not have major bleeding. For this analysis, the primary exposure was obesity and the outcome of interest was major bleeding. Analyses were adjusted for a variety of factors derived from GH automated databases and a 44-item mailed survey that collected supplemental clinical, behavioral, and demographic information.⁶ Additional details regarding cases, controls, warfarin use, genetic testing, and covariate information was included in previous publications.^{6, 21}

Cases

Cases were GH patients 18 years of age or older who had an inpatient diagnosis of a major bleeding event (reference date) between January 1, 2005 and April 1, 2011. A validated ICD-9 algorithm was used to identify major bleeding events.²² Events were classified as “major bleeding” if they were clinically apparent and resulted in hospitalization, hemoglobin decreased >2mg/dl, and/or more than 2 units of packed red blood cells were transfused.^{22, 23} We required cases to be continuously enrolled in GH for at least 1 year prior to index with no evidence of major bleeding events in the year prior to the index date. The date of the bleeding event was set as the index date. Using GH pharmacy records, we required cases to be using warfarin within three days of the bleeding event.

Controls

Controls were selected using a risk set sampling approach in which patients that met all eligibility criteria above (except major bleeding) were assigned a random index date (date of bleeding event for matched case). The index date was used for collection of information that allowed for the comparison of case and control data from the same time frame of case major bleeding events.

Warfarin Use

Warfarin use was determined using GH automated pharmacy records during the year prior to the index date. The assessment of warfarin use included drug name, dispensing date, days of supply, and strength. Treatment duration was calculated based on the current episode of warfarin use, characterized as a continuous supply of warfarin and not exceeding 90 gap days. Gap days were calculated using the days’ supply field from dispensing data and multiplying by 1.25 to allow for variable adherence.⁶

Body Mass Index

BMI was calculated from height and weight data documented in GH electronic medical records (EMR). Height and weight are routinely collected during patient visits and measures most proximal to the index date were used. After calculating BMI in kg/m², WHO BMI classifications were applied: underweight (<18.5), normal weight (18.5–24.9), overweight (25.0–29.9), and obese (>30.0).²⁴ These classifications were dichotomized into obese and non-obese for the analyses.

Clinical, Demographic, and Behavioral Covariates

Clinical and demographic data displayed in Table 1 were obtained through the GH automated databases that capture all clinical encounters via EMRs and claims data.

Table 1.

Clinical and demographic characteristics by case-control status of warfarin users in a community setting.

Variable	Cases (n=265)	Controls (n=305)	p-value*
Automated Data Variables
Age in Years at Index Date, Mean (SD)	71.1 (12.7)	69.5 (11.2)	0.92
Duration of Plan Enrollment at Index Date (Years),
Mean (SD)	15.8 (6.0)	14.8 (6.3)	0.91
	n (column %)
Male	134 (50.6%)	176 (57.7%)	0.09
Body Mass Index (BMI)			0.01
Not obese (<30.0 kg/m2)	140 (53%)	128 (42%)
Obese (≥30.0 kg/m2)	125 (47%)	177 (58%)
Duration of Warfarin Therapy at Index Date			<0.01
<6 Months	69 (26.0%)	54 (17.7%)
6 Months to 1 Year	26 (9.8%)	24 (7.9%)
>1 Year	170 (64.2%)	227 (74.4%)
Comorbidities¥
Cancer	8 (3.0%)	14 (4.6%)	0.32
Diabetes	58 (21.9%)	81 (26.6%)	0.19
Hypertension	197 (74.3%)	195 (63.9%)	<0.01
Congestive Heart Failure	101 (38.1%)	68 (22.3%)	<0.01
Charlson Comorbidity Index, Mean (SD)			0.07
Score=0	84 (31.8%)	124 (40.8%)
Score=1	70 (26.4%)	76 (25.0%)
Score=2+	110 (41.7%)	104 (34.2%)
Missing	1	1
Self – Report Survey Variables
	n (column %)
Diagnoses associated with warfarin use
Atrial Fibrillation	89 (33.6%)	136 (44.6%)	<0.01
DVT	31 (11.7%)	42 (13.8%)	0.45
PE	27 (10.2%)	26 (8.5%)	0.49
Stroke	27 (10.2%)	20 (6.6%)	0.12
Heart Valve Replacement	54 (20.4%)	32 (10.5%)	0.01
Myocardial Infarction	13 (4.9%)	17 (5.6%)	0.71
Joint Replacement	12 (4.5%)	12 (3.9%)	0.72
CABG	6 (2.3%)	11 (3.6%)	0.36
Other	17 (6.4%)	23 (7.5%)	0.61
Race			0.45
American Indian or Alaska Native	3 (1.2%)	1 (0.3%)
Asian	3 (1.2%)	6 (2.0%)
Black/African American	7 (2.7%)	5 (1.6%)
White/Caucasian	243 (93.5%)	287 (94.4%)
Other	4 (1.5%)	2 (0.7%)
Missing	5	4
Care Setting			0.12
Anticoagulation Clinic	115 (46.4%)	166 (55.1%)
Primary Care	86 (34.7%)	91 (30.2%)
Cardiologist	20 (8.1%)	24 (8.0%)
Other	27 (10.9%)	20 (6.6%)
Missing	17	4
Annual Household Income			0.1
<$25,000	40 (17.1%)	46 (16.1%)
$25,000–$49,999	102 (43.6%)	96 (33.7%)
$50,000–$74,999	45 (18.4%)	79 (27.7%)
$75,000–$99,999	23 (9.8%)	33 (11.6%)
≥$100,000	24 (10.3%)	31 (10.9%)
Missing	31	20

^¥Diagnoses occurred within two years prior to the index date

^*Based on x² tests, Fisher’s exact test, Student’s t-test

Note: Bold p-values indicate statistically significant differences between cases and controls at α=0.05

Supplemental information not available from the databases was collected via a 44-item self-report survey completed by all study participants. Of the 702 cases and 702 controls approached for the survey and genetic testing (see below), 265 (38.0%) and 305 (43.9%) participated. Details on the survey were previously described by Roth et al.⁶

Genetic Testing

A buccal swab was collected from all study participants at the time of the survey via a catch-all mailer (Epicentre, Madison, WI) for analysis of CYP2C9*2 and *3, VKORC1 1173G>A, and CYP4F2*3 polymorphisms. The polymorphism detection assays were designed and performed in the Functional Genomics Laboratory, Center for Ecogenetics and Environmental Health at the University of Washington, Seattle, WA. Genotype and allele frequencies for this study population can be found in a previous publication.⁶

Statistical analyses

Case and control characteristics were compared using χ² tests, Fisher’s exact test for categorical variables and the Student’s t-test for continuous variables following established methods in the literature.²⁵ In the primary analysis, we used unadjusted and multivariate logistic regression to estimate the major bleeding odds ratio (OR) and 95% confidence interval (95% CI) for obese patients relative to non-obese patients. In multivariate analyses, we adjusted for race (white/other), duration of therapy (continuous), age (continuous), gender, diagnoses related to warfarin use (atrial fibrillation, venous thromboembolism, heart valve replacement, other), regular aspirin use, regular use of NSAIDs, congestive heart failure, and hypertension. These covariates were selected based on patient characteristics previously adjusted for in earlier research studying outcomes in warfarin users.^{6, 13, 16}

We also conducted exploratory analyses using multivariate logistic regression models to evaluate the main effects of genetic variants on major bleeding risk and interactions between genotype and obesity. Variants in CYP2C9 (*2, *3) and VKORC1 (1173) were previously associated with an increased bleeding risk^{4, 8, 26}, and CYP4F2*3 was associated with a decreased bleeding risk.¹⁶ To reflect these differences and maximize statistical power, interaction analyses used a variable that combined the presence of any CYP2C9 or VKORC1 risk variant into a binary variable (CYP2C9 or VKORC1 variant), and a separate variable that reflected CYP4F2*3 status. Similar to previous studies, our analyses grouped heterozygous and homozygous variants in a single “variant” category.⁶ The multivariate logistic regression models adjusted for age at index date (continuous), gender (binary), and race (white/other, binary). We did not use Bonferroni correction for testing significance of multiple comparisons.²⁷

All statistical analyses were conducted using the STATA Statistical Package, Version 13.0 (STATA, Austin, TX). This study was approved by the GH institutional review board.

RESULTS

Study Sample Characteristics

Among cases and controls, average age was 71 and 70 years, and 51% and 58% were male, respectively. As shown in Table 1, the only statistically significant differences between cases and controls were in proportion with a hypertension or congestive heart failure diagnosis in their medical records, and self-reported heart valve replacement and atrial fibrillation indication for warfarin.

Primary Analysis of Obesity and Major Bleeding Risk

In unadjusted analyses, obesity; history of deep vein thrombosis, congestive heart failure, or hypertension diagnosis; regular aspirin use or NSAID use; and warfarin treatment duration <1 year each had statistically significant associations with major bleeding (Table 2). In multivariate analysis, only obesity, gender, and history of congestive heart failure diagnosis remained significantly associated with major bleeding risk. Obesity was associated with decreased major bleeding risk in unadjusted (OR=0.61, 95% CI=0.42–0.88) and multivariate (OR=0.60; 95% CI=0.39–0.92) logistic regression models (Table 2).

Table 2.

Unadjusted analysis for association of risk factors on the risk of major bleeding in warfarin therapy patients in a community setting

Comparison	Odds Ratio (95% Confidence Interval)
Caucasian vs. Other Race	0.69 (0.36–1.32)
<1 Year Warfarin Treatment Duration vs. ≥1 Year Warfarin Treatment Duration	1.84 (1.20–2.81)
Age at index date	1.01 (0.99–1.03)
Male vs. Female	0.75 (0.54–1.04)
Body Mass Index ≥30 vs. Body Mass Index <30	0.61 (0.42–0.88)
Warfarin Indication
Atrial fibrillation vs. Other Indications	0.97 (0.68–1.37)
Deep vein thrombosis vs. Other Indications	1.76 (1.15–2.68)
Heart valve replacement vs. Other Indications	3.13 (0.82–11.9)
Regular Aspirin Use vs. No Regular Aspirin Use	1.81 (1.21–2.69)
Regular NSAID Use vs. No Regular NSAID Use	2.51 (1.24–5.12)
Congestive Heart Failure Diagnosis vs. No Congestive Heart Failure Diagnosis	2.15 (1.49–3.10)
Hypertension Diagnosis vs. No Hypertension Diagnosis	1.63 (1.14–2.35)

Note: Bold p-values indicate statistically significant differences between cases and controls at α=0.05

In multivariate models stratified by duration of warfarin use, the major bleeding odds ratio for obese vs. non-obese patients was 0.56 (95% CI: 0.35–0.90) in those with more than 1-year of warfarin use, and 0.78 (95% CI: 0.40–1.54) in those with less than 1-year of warfarin use. A formal test of interaction between obesity and duration of warfarin use was non-significant (p=0.72).

Exploratory Analyses of Genetic Status, Obesity, and Major Bleeding Risk

Odds ratios and their respective confidence intervals are presented for the association between genotype and major bleeding risk in strata of obesity and no obesity. Similarly, we present results for the association between obesity and major bleeding risk in strata of genotype status (Table 4). The interaction between CYP2C9/VKORC1 variant status and obesity on the risk of major bleeding was not statistically significant (p=0.59). Point estimates suggest that the joint effect of obesity is more protective in those patients who were wild type for CYP2C9 or VKORC1, but the result was not statistically significant. However, the interaction between CYP4F2 variant status and obesity on the risk of major bleeding was statistically significant (p=0.049). The OR for major bleeding among obese persons compared to non-obese was 0.98 (95% CI: 0.59–1.62) for those with the CYP4F2*3 variant and 0.49 (95% CI: 0.30–0.79) for those who were wild type CYP4F2*3. The OR for major bleeding among patients with the CYP4F2*3 variant compared to patients who were wild type was 0.98 (95% CI: 0.61–1.56) and 0.49 (95% CI: 0.30–0.81) for obese and non-obese, respectively. The significant gene-environment interaction estimated between CYP4F2*3 genotype and obesity on major bleeding risk would not hold its statistical significance to Bonferroni correction for multiple comparisons indicating the null hypothesis of no interaction might really be true, and the statistically significant result might be due to chance.

Table 4.

Interactions Between Gene and Obesity on Risk of Major Bleeding in Warfarin Users

	Multivariate model* Odds Ratio (95% CI)
Risk factors	Obese	Non-obese	OR for obese vs. non-obese within strata of genotype
CYP2C9 or VKORC1 variant present	0.71 (0.41–1.24)	1.01 (0.58–1.78)	0.70 (0.46–1.06)
CYP2C9 or VKORC1 wild type	0.56 (0.29–1.10)	Reference	0.56 (0.29–1.10)
OR for CYP2C9/VKORC1 variant present vs. wild type within strata of obesity	1.26 (0.74–2.14)	1.01 (0.58–1.78)
CYP4F2*3 variant present	0.48 (0.30–0.77)	0.49 (0.30–0.81)	0.98 (0.59–1.62)
CYP4F2*3 wild type	0.49 (0.30–0.79)	Reference	0.49 (0.30–0.79)
OR for CYP4F2*3 variant present vs. wild type within strata of obesity	0.98 (0.61–1.56)	0.49 (0.30–0.81)

Significant p-values (P<0.05) are in bold.

^*Adjusted for gender, race, age at index date

DISCUSSION

We conducted a case-control study to investigate the association between body mass index and the risk of major bleeding in patients receiving warfarin in a community setting. We found that obese (BMI ≥ 30.0 kg/m²) warfarin users had a statistically significant 40% reduction in major bleeding risk compared to non-obese warfarin users.

This study is the first to show an inverse obesity-bleeding association among warfarin users in a community setting. The results are corroborated by previous evidence demonstrating reduced warfarin benefit (evidenced by altered pharmacokinetic properties associated with warfarin clearance and volume of distribution) and an increased time to achieve therapeutic anticoagulation levels in obese patients indicating sub-therapeutic dosing that may be associated with major bleeding risk.^{10, 11} These studies were relatively small and limited by not adjusting for medications that may interact with warfarin, both of which are addressed in our study. One recent study found an increased bleeding risk in obese individuals taking warfarin, which contrasts our findings; however, that study was conducted at an anticoagulation clinic with a retrospective cohort design, analyzing a smaller number of bleeds, greatly diminishing the statistical power, and, unlike in our analysis, they did not control for key co-morbidity risk factors in their analysis.²⁸ Further, they included any type of bleed (minor or major) compared to only major bleeds used in our study.

Our findings suggest that BMI is an important clinical factor in assessing and managing anticoagulation therapy. This could be attributable to obese patients receiving warfarin doses that are not personalized to their weight, or it could be due to not adjusting the dose to account for long-term weight gain. Future studies should explore these potential mechanisms, and clinicians should carefully consider BMI when determining warfarin doses.¹¹

While results are exploratory, this was the first study to look at the interaction between genetic variants and obesity on bleeding risk. While no interaction was found between CYP2C9/VKORC1 genotype and obesity on major bleeding risk, obesity appeared protective among patients with CYP4F2 wild type but not patients with the variant. We also observed the CYP4F2 variant to be protective among those without obesity but not among those with obesity. These findings suggest the protective association of the CYP4F2*3 variant⁶ and obesity on major bleeding are lessened when both are present. Future studies are needed to validate these results, and additional functional studies are needed to better understand the potential biological rationale for this interaction in order to reconcile the differences of the bleeding risk associations from each factor separately. The findings may also provide evidence for understanding the variability in published findings for associations between pharmacogenomic markers and the risk of major bleeding events in obese warfarin patients. Additional exploration into protective associations against major bleeding risk that may result from being obese and known pharmacogenetic variants that increase risk should follow, since individuals can carry multiple pharmacogenetic variants that confer conflicting major bleeding risks when considered independently.

The study has several limitations. First, we dichotomized BMI into obese and non-obese groups, rather than analyzing finer BMI groupings. This approach was necessary due to small numbers of patients in underweight and normal weight categories, which would have limited statistical power. If heterogeneous effects actually exist across the finer groupings, this could impact the confidence intervals by being too narrow. Being underweight, overweight, and obese are associated with additional individual health risks, but delineating finer BMI groupings is not necessary to inform major bleeding risk. The results on the interaction between obesity and CYP4F2*3 genotype were exploratory and any associations may be due to chance, as statistical significance would not hold under Bonferroni correction for multiple comparisons.²⁷ The size and demographic of the study’s sample creates generalizability limitations. Though this was the largest study of warfarin pharmacogenomics and major bleeding conducted to date, the sample size of 570 patients still limits the power to be able to detect genetic interactions. Furthermore, most participants were Caucasian (overall 94%) and received care in the Seattle metropolitan region, therefore potentially limiting generalizability. Research including diverse groups of warfarin users is needed to determine whether these associations hold in alternative populations and settings.

Conclusion

Obese patients receiving warfarin in a community setting had 40% lower risk of major bleeding relative to non-obese patients. The results are consistent with prior studies demonstrating increased therapeutic dose needs in obese anticoagulation patients. Our findings provide an initial evidence base to support more nuanced investigations of the impact of BMI on major bleeding risk in the future. Although the results indicated reduced bleeding in obese warfarin users, the study did not measure thromboembolic events. Evidence suggests that obesity may predispose to venous thromboembolism, which can be of concern to obese warfarin users.¹⁰ The harm in increased risk of clotting events should be explored further. Future research should investigate obesity and additional genetic variations on major bleeding risk and explore the mechanisms for potential gene-environment interaction effects.

Table 3.

Association between obesity and the risk of major bleeding among warfarin therapy patients in a community setting.

			Unadjusted Model		Adjusted Model*
Exposure	Cases N=265	Controls N=305	OR	95% CI	OR	95% CI
	n (%)
Non-obese (BMI<30kg/m2)	140 (52.8)	128 (42.0)	REF		REF
Obese (BMI≥30 kg/m2)	125 (47.2)	177 (58.0)	0.61	0.42–0.88	0.60	0.39 – 0.92

^*Multivariate model adjusted for: race (white/other), duration of therapy (continuous), age (continuous), gender, diagnoses related to warfarin use (categorized as: atrial fibrillation, VTE, heart valve replacement, and other), regular aspirin use, regular other NSAID use, congestive heart failure diagnosis, hypertension diagnosis

Highlights.

1. The influence of body mass index on bleeding risk among patients receiving the anticoagulant warfarin is poorly understood.

2. Obese patients receiving warfarin had a 40% lower risk of major bleeding compared to non-obese patients.

3. The protective effect of obesity was observed in wild type CYP4F2*3 patients but not patients with variant type CYP4F2*3, which itself has been shown to be protective.

We believe this is the first study to investigate the association between BMI and major bleeding risk on warfarin in a community setting (i.e., not conducted at an academic anticoagulation clinic), as well as the first to evaluate BMI interactions with key genetic risk factors. Our results will be of particular interest to your readership because we demonstrate that obese patients on warfarin are at decreased major bleeding risk—a finding that suggests BMI is an important clinical factor in the assessment and management of anticoagulation therapy. Additionally, our study demonstrated a potentially important gene-environment interaction between CYP4F2 variants and BMI on risk of major bleeding that warrants additional evaluation in future studies.