Abstract
Background
Atrial fibrillation affects more than two million Americans and results in a fivefold increased rate of embolic strokes. The efficacy of adjusted dose warfarin is well documented, yet many patients are not receiving treatment consistent with guidelines. The use of a patient-specific computerized decision support tool may aid in closing the knowledge gap regarding the best treatment for a patient.
Methods
This retrospective, observational cohort analysis of 6,123 Ohio Medicaid patients used a patient-specific computerized decision support tool that automated the complex risk–benefit analysis for anticoagulation. Adverse outcomes included acute stroke, major gastrointestinal bleeding, and intracranial hemorrhage. Cox proportional hazards models were developed to compare the group of patients who received warfarin treatment with those who did not receive warfarin treatment, stratified by the decision support tool’s recommendation.
Results
Our decision support tool recommended warfarin for 3,008 patients (49%); however, only 9.9% received warfarin. In patients for whom anticoagulation was recommended by the decision support tool, there was a trend towards a decreased hazard for stroke with actual warfarin treatment (hazard ratio 0.90) without significant increase in gastrointestinal hemorrhage (0.87). In contrast, in patients for whom the tool recommended no anticoagulation, receipt of warfarin was associated with statistically significant increased hazard of gastrointestinal bleeding (1.54, p = 0.03).
Conclusions
We have shown that our atrial fibrillation decision support tool is a useful predictor of those at risk of major bleeding for whom anticoagulation may not necessarily be beneficial. It may aid in weighing the benefits versus risks of anticoagulation treatment.
KEY WORDS: atrial fibrillation, decision support, anticoagulation, decision aid
INTRODUCTION
Atrial fibrillation is the most prevalent serious cardiac arrhythmia and is a significant risk factor for stroke.1,2 If left untreated, these patients face a significant fivefold increased rate of embolic stroke, and the risk is greatest in the elderly.3 Numerous clinical trials have demonstrated the efficacy of anticoagulation therapy to significantly reduce this risk of thromboembolism and the devastating outcome of ischemic stroke in patients with atrial fibrillation.4–11 Two studies have shown that more than 50% of patients without contraindications to anticoagulation therapy are receiving warfarin, but other studies have documented substantially fewer patients receiving treatment consistent with guidelines.12–20 The challenge is to identify those patients for whom the benefit from treatment outweighs the risk.
Patient-specific characteristics known at the time of decision making substantially alter the patient’s risk of both ischemic stroke and hemorrhage with anticoagulation.1,21–24 Published anticoagulation guidelines are limited in their ability to consider the individual patient’s balance of risk and benefit, especially when the risk of hemorrhage is increased.25,26 Other patient- and physician-related factors might contribute to a decision to withhold warfarin. For example, physicians are less likely to use anticoagulation in older patients, but it is this population that has the greatest risk of stroke, and thus, the greatest opportunity to benefit from intervention.27–29
Given the efficacy of adjusted dose warfarin in reducing the risk of ischemic stroke by 68%, patient-specific computerized decision support may aid in closing the known gap between knowledge and optimal treatment for a patient.1,30,31 Our goal was to validate the ability of an Atrial Fibrillation Decision Support Tool to identify those who would benefit or be harmed from anticoagulation therapy to prevent thromboembolic events in a cohort of Ohio Medicaid patients with non-valvular atrial fibrillation. Short of a randomized trial, this retrospective study was felt to be the best way to predict the performance of the tool. The methodology might be of use to test other tools. We hypothesized that patients receiving anticoagulation treatment concordant with the decision support tool recommendation would have fewer adverse events compared with discordant care.
RESEARCH DESIGN AND METHODS
Basic Design and Data Sources
This study was a retrospective, observational cohort analysis of Ohio Medicaid patients from January 1, 1997 through May 31, 2002. Data were collected from the Ohio Medicaid administrative claims database, which has been well described elsewhere,19,32,33 and the Ohio Mortality Public Use Statistical file. Ohio Medicaid provides coverage for certain low-income and medically vulnerable residents; aged, blind, or disabled and covered families and children. It includes fee-for-service data from all institutions, providers, and pharmacies that provide services to Ohio Medicaid enrollees.
The identified Ohio Medicaid patients were then cross-matched with the Ohio Mortality Public Use Statistical file from January 1, 1998 through May 31, 2002. This file contains data from death certificates for any person who dies in Ohio and Ohio residents who die out of state.
Patient Selection
We identified all patients with two or more claims containing an International Classification of Diseases, Ninth Revision, Clinical Modification code (ICD-9-CM) for atrial fibrillation (427.31) during the study period. Two claims were required for inclusion to increase the likelihood of accurate atrial fibrillation diagnosis. We excluded all patients with lone atrial fibrillation, a history of valvular heart disease (two or more claims with ICD-9-CM codes for mitral valve disease, aortic valve disease, mitral and aortic valve disease, heart valve transplant, or heart valve replacement, or a procedure code for mitral or aortic valve repair or replacement). We included only those patients with 12 consecutive months of enrollment before the first atrial fibrillation diagnosis, which was considered incident atrial fibrillation for the purpose of this study. We followed patients for adverse events until the first month not enrolled in Medicaid; thus, patients were censored at disenrollment.
We used pharmacy claims data to exclude patients who filled prescriptions for warfarin before the initial atrial fibrillation diagnosis. Using the same pharmacy claims data, patients were considered to be treated with warfarin if they filled prescriptions for warfarin within 30 days of the atrial fibrillation diagnosis. Few patients were started on treatment or stopped treatment beyond this 30-day period.
Risk Factors
In the 12-month period before the incident atrial fibrillation diagnosis, we identified patient-specific factors known to influence the risk for stroke and the risk for hemorrhage, and we identified other factors that potentially influence the decision to prescribe warfarin. We used ICD-9-CM for inpatient and outpatient claims, and medication therapeutic class codes were used for pharmacy claims.
Demographic data were used to derive the age, gender, and race for each patient. We identified covariates known to influence the risk of stroke, which include age, hypertension, diabetes mellitus, congestive heart failure, prior stroke, and prior myocardial infarction.1 We identified covariates known to influence the risk of hemorrhage, which include prior gastrointestinal hemorrhage, prior intracranial hemorrhage, anemia, and renal insufficiency.34,35 Any stroke or myocardial infarction that occurred within the prior 90 days to incident atrial fibrillation diagnosis was considered a “recent” event.
We identified, a priori, other covariates that we believe to potentially influence warfarin prescribing but whose effects on stroke and bleeding risk are not quantified in the literature. Psychiatric illness includes schizophrenia, affective psychosis, paranoia, or other non-organic psychosis. Substance abuse includes alcohol dependence, drug dependence, or nondependent alcohol abuse (excluding tobacco use disorder). Social risk factors includes lack of housing, inadequate housing, inadequate material resources, persons living alone, no other household member able to render care, or noncompliance with medical treatment.
Concurrent medication use also may influence warfarin prescribing and risk for hemorrhage. Utilizing medication therapeutic class codes, we defined the categories of: gastrointestinal protection (antacids, anti-ulcer preparations, hemorrhoidal agents/preparations, rectal preparations, H2 inhibitors), analgesics (non-narcotic analgesics, salicylate analgesics, anti-inflammatory agents, non-steroidal anti-inflammatory drugs, miscellaneous analgesics), steroids/immunosuppressants (systemic glucocorticoids, mineralocorticoids, immunosuppressives), and other bleed risk (anti-hemophilic factors, heparin preparations, anti-neoplastics).
Decision Support Tool for Anticoagulation Recommendation
We have described previously a decision analytic tool that incorporates patient-specific risks for ischemic stroke and major bleeding events and calculates expected outcomes for patients with atrial fibrillation with and without warfarin treatment.36–40 This tool is consistent with ACC/AHA/ESC 2006 guidelines. However, while guidelines explicitly address risk stratification for stroke, they provide little guidance on bleeding risk assessment.26 Our tool explicitly accounts for the risk of bleeding and formally addresses the balance of risk of bleeding with the benefit of stroke prevention. This decision support tool is designed to individualize treatment recommendations based upon a patient’s age, gender, and different degrees of risk for thromboembolism and hemorrhage by predicting quality-adjusted life years (QALYs).38–40 Other patient-specific variables used to determine a patient’s risk included a history of diabetes mellitus, hypertension, congestive heart failure, myocardial infarction, prior stroke/TIA, gastrointestinal bleed, anemia, and renal insufficiency. The embedded risk prediction models derived from the medical literature are used to estimate the annual rate of ischemic stroke1 and major bleeding41 from the covariates above (Table 1). We used Decision Maker® for Windows to automate the decision analysis calculations for each patient.
Table 1.
Clinical factor | Risk weights for | |
---|---|---|
Ischemic stroke | Major bleed | |
History of diabetes mellitus | 0.57 | |
History of hypertension | 0.49 | |
History of congestive heart failure | 0.36 | |
History of myocardial infarction | 0.2 | |
Prior stroke/TIA | ||
Either past or present | 0.99 | 0.84 |
Both past and present | 1.69 | |
Age | ||
<60 | 0 | 0 |
60–64 | 0.34† | 0 |
65–69 | 0.34 | 1.03‡ |
70–79 | 0.68 | 1.03 |
80–89 | 1.02 | 1.03 |
History of gastrointestinal bleeding | 1.12 | |
Serious comorbid condition | 1.04 |
Reprinted with permission from MH Eckman et al. Chest 1998 (114)
*This table should not be used for patients with “lone atrial fibrillation” younger than 65 years of age, as this is the referent group against which the weights for clinical risk factors were calculated. These patients have an annual stroke rate of 1%
†Increased risk per decade over age 60 years
‡Age ≥65 years
We automated the calculations using a SAS® Version 9.1 (SAS Institute Inc., Cary, NC, USA) script to develop batch input files for Decision Maker®. Parameter values for the covariates that predict stroke and bleeding risk identified in the 12-month period before the first atrial fibrillation diagnosis for each patient were input to Decision Maker. The output of the decision analysis for each patient was his or her predicted QALYs with and without anticoagulation.
Definition of Study Groups
The expected gain or loss conferred by anticoagulant therapy was determined by calculating the difference in expected utility (in QALYs) between the two strategies. If the calculated gain (ΔQALYs) was zero or greater, we considered this to be a positive recommendation for anticoagulation with warfarin. If the gain (ΔQALYs) was less than zero, anticoagulation was not recommended. These two groups were further stratified by whether they actually received treatment with warfarin or not, forming the four groups. Our overall goal was to determine whether there were fewer adverse events when the decision support tool and actual treatment were concordant. We had no control over who was prescribed warfarin.
Outcomes Assessment
Adverse events included acute stroke, acute gastrointestinal hemorrhage, and acute intracranial hemorrhage based on ICD-9-CM codes recorded on inpatient hospitalization claims. We only used first events from inpatient hospitalization codes to improve the reliability of diagnosis. We also inferred adverse events from the underlying cause of death revealed by ICD-9 codes for 1998 and ICD-10 codes for subsequent years in the Ohio death registry files.
Time at Risk
The date of a patient’s initial atrial fibrillation diagnosis claim was used to define the start of the patient’s period at-risk. Patients were censored at their date of disenrollment from Medicaid or on their date of death. For each outcome of interest, patients were included in the analysis only until the date of the first event, e.g., a patient with an acute stroke was removed from further analysis after the occurrence of the first stroke. As analysis was performed separately for each outcome, occurrence of gastrointestinal hemorrhage, for example, would not be counted as an adverse event or affect time at risk for acute stroke analysis.
Propensity Score for Receiving Warfarin
As this was an observational study in which clinicians and patients were free to make treatment decisions, it is likely that patients prescribed warfarin differed from those who did not receive warfarin. To correct for confounding by indication for warfarin treatment, we developed a propensity score to predict each patient’s likelihood of receiving warfarin.42 We used logistic regression to select the variables that were significant predictors at p < 0.10 to be included in a multivariable model to predict the propensity score. Separate models were developed for the group of patients for whom the decision support tool recommended anticoagulation and for the group the tool recommended withholding anticoagulant therapy. We included this propensity score along with other covariates that might confound adverse outcomes in several Cox proportional hazard analyses to calculate adjusted adverse event rates in these groups.
Statistical Analysis
We used descriptive statistics, including the Student’s t test and chi-square test, to characterize the study population. Event rates for each outcome (stroke, intracranial hemorrhage, and gastrointestinal bleeding) were calculated for each group and for the cohort as a whole.
We performed comparisons within the groups defined by the decision tool recommendations for or against warfarin. Subgroups for comparison were defined by actual treatment with warfarin or not. Cox proportional hazards models (SAS PROC PHREG) were used to determine the unadjusted and adjusted hazard ratios for the outcomes for each type of adverse outcome within each group. The proportional hazards assumptions were met, utilizing time-dependent covariates. The warfarin propensity score was forced into each adjusted model. Incorporating covariates for medications filled before the diagnosis and during the time at risk (including those at the p < 0.10 significance) to the Cox proportional hazards models did not significantly alter the results.
RESULTS
Patient Characteristics
Two or more claims with ICD-9-CM code 427.31 were found among 25,200 patients. After the inclusion and exclusion criteria were applied, 6,123 patients remained in the cohort. These patients were followed for a mean of 581 days.
The mean (SD) age of the study population was 76.2 (13.4) years (Table 2). The majority of patients were women and were white. The population had numerous comorbidities, particularly hypertension, congestive heart failure, diabetes mellitus, and prior myocardial infarction. Many were prescribed analgesics and gastrointestinal protective medications. The decision support tool recommended warfarin for 3,008 patients (49%); however, only 298 (9.9%) of these were prescribed warfarin. Those who actually received warfarin tended to be younger, white, or on analgesics. Across the four study groups, covariates not statistically different included prior myocardial infarction, prior ICH, prior other bleed, and social factors.
Table 2.
Anticoagulation per DST | No anticoagulation per DST | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
Warfarin treatment | No Warfarin treatment | Warfarin treatment | No Warfarin treatment | |||||||
n | % | n | % | p | n | % | n | % | p | |
Observations | 298 | 9.9 | 2,710 | 90.1 | 203 | 6.5 | 2,912 | 93.5 | ||
Age, mean (SD) | 68.6 | (13.7) | 74.6 | (14.3) | <0.01 | 76.1 | (10.1) | 78.6 | (12.1) | <0.01 |
White | 241 | 80.9 | 2,100 | 77.5 | 0.18 | 175 | 86.2 | 2,354 | 80.8 | 0.06 |
Female | 209 | 70.1 | 2,069 | 76.3 | 0.02 | 155 | 76.4 | 2,154 | 74 | 0.45 |
Hypertension | 217 | 72.8 | 1,791 | 66.1 | 0.02 | 59 | 29.1 | 773 | 26.5 | 0.43 |
DM | 132 | 44.3 | 1,161 | 42.8 | 0.63 | 22 | 10.8 | 335 | 11.5 | 0.77 |
CHF | 167 | 56 | 1,505 | 55.5 | 0.87 | 39 | 19.2 | 630 | 21.6 | 0.42 |
Prior MI | 55 | 18.5 | 478 | 17.6 | 0.73 | 22 | 10.8 | 336 | 11.5 | 0.76 |
Prior stroke | 21 | 7 | 236 | 8.7 | 0.33 | 20 | 9.9 | 250 | 8.6 | 0.54 |
Recent stroke | 5 | 1.7 | 42 | 1.5 | 0.87 | 15 | 7.4 | 168 | 5.8 | 0.34 |
Prior GI bleed | 3 | 1 | 19 | 0.7 | 0.47* | 16 | 7.9 | 402 | 13.8 | 0.02 |
Prior ICH | 0 | 0 | 33 | 1.2 | 0.07* | 0 | 0 | 26 | 0.9 | 0.41* |
Prior other bleed | 13 | 4.4 | 147 | 5.4 | 0.44 | 6 | 3 | 94 | 3.2 | 0.83 |
Comorbidity (any of the 3) | 25 | 8.4 | 301 | 11.1 | 0.15 | 60 | 29.6 | 1,193 | 41 | <0.01 |
Anemia | 14 | 4.7 | 178 | 6.6 | 0.21 | 47 | 23.2 | 837 | 28.7 | 0.09 |
Renal disease | 9 | 3 | 162 | 6 | 0.04 | 12 | 5.9 | 413 | 14.2 | <0.01 |
Recent MI | 6 | 2 | 68 | 2.5 | 0.6 | 9 | 4.4 | 182 | 6.3 | 0.3 |
Substance abuse | 17 | 5.7 | 94 | 3.5 | 0.05 | 6 | 3 | 95 | 3.3 | 0.81 |
Psychiatric Dx | 45 | 15.1 | 559 | 20.6 | 0.02 | 23 | 11.3 | 530 | 18.2 | 0.01 |
Social factors | 50 | 16.8 | 655 | 24.2 | <0.01 | 47 | 23.2 | 598 | 20.5 | 0.37 |
Non-compliance | 97 | 32.6 | 1,151 | 42.5 | <0.01 | 72 | 35.5 | 1,079 | 37.1 | 0.65 |
GI Med | 152 | 51 | 1,528 | 56.4 | 0.08 | 93 | 45.8 | 1,537 | 52.8 | 0.05 |
Analgesics | 228 | 76.5 | 1,984 | 73.2 | 0.22 | 157 | 77.3 | 1,897 | 65.1 | <0.01 |
Steroids + | 84 | 28.2 | 756 | 27.9 | 0.92 | 47 | 23.2 | 689 | 23.7 | 0.87 |
Other GI/anemia risk Rx | 28 | 9.4 | 277 | 10.2 | 0.65 | 18 | 8.9 | 365 | 12.5 | 0.12 |
All cause mortality | 91 | 30.5 | 1,323 | 48.8 | <0.01 | 74 | 36.5 | 1,691 | 58.1 | <0.01 |
p: Chi square
*Fisher’s exact
Event rates for each outcome are presented for the cohort and by subgroup in Tables 3, 4 and 5. The stroke rate documented by Medicaid claims (3.37 per 100 patient years) is less than that reported in the literature (4.5% annually untreated with warfarin from pooled analysis).1 We found a large number of strokes among the Ohio mortality files were not documented in the Medicaid claims. Stroke event rates were not statistically different across the groups. Patients had a higher rate of gastrointestinal bleeding (5.76 per 100 patient years) compared with literature reported rates (1.3% per year in warfarin-treated patients).1
Table 3.
N | % | Per 100py | |
---|---|---|---|
Hospital Dx Ischemic Stroke* | 316 | 5.2 | 3.37 |
Die from ischemic stroke† | 242 | 4 | 2.58 |
All ischemic strokes‡ | 523 | 8.5 | 5.57 |
Hospital Dx GI bleed | 516 | 8.4 | 5.6 |
Die from GI bleed | 16 | 0.3 | 0.17 |
All GI bleed | 530 | 8.7 | 5.76 |
Hospital Dx ICH | 63 | 1 | 0.65 |
Die from ICH | 10 | 0.2 | 0.1 |
All ICH | 71 | 1.2 | 0.73 |
Hospital Dx other bleed | 344 | 5.6 | 3.67 |
Die other bleed | 0 | 0 | 0 |
All other bleed | 344 | 5.6 | 3.67 |
*Hospital Dx are adverse events as documented by Ohio Medicaid inpatient claims, ICD-9 CM diagnoses
†Ohio death registry adverse events, ICD-9 CM or ICD-10
‡Combined adverse event from both sources, a patient could either be counted as Hospital Dx or Die from, but not both for “All” outcome
Table 4.
Warfarin N = 298 | No warfarin N = 2,710 | |||||
---|---|---|---|---|---|---|
n | % | Per 100py | n | % | Per 100py | |
Hospital Dx ischemic stroke* | 17 | 5.7 | 3.44 | 141 | 5.2 | 3.44 |
Die from ischemic stroke† | 7 | 2.35 | 1.42 | 91 | 3.36 | 2.22 |
All ischemic strokes‡ | 22 | 7.38 | 4.45 | 220 | 8.12 | 5.37 |
Hospital Dx GI bleed | 23 | 7.72 | 4.56 | 206 | 7.6 | 5.09 |
Die from GI bleed | 0 | 0 | 0 | 7 | 0.26 | 0.17 |
All GI bleed | 23 | 7.72 | 4.56 | 213 | 7.86 | 5.27 |
Hospital Dx ICH | 7 | 2.35 | 1.36 | 29 | 1.07 | 0.69 |
Die from ICH | 0 | 0 | 0 | 5 | 0.18 | 0.12 |
All ICH | 7 | 2.35 | 1.36 | 33 | 1.22 | 0.78 |
Hospital Dx other bleed | 22 | 7.38 | 4.36 | 162 | 5.98 | 3.99 |
Die other bleed | 0 | 0 | 0 | 0 | 0 | 0 |
All other bleed | 22 | 7.38 | 4.36 | 162 | 5.98 | 3.99 |
*Hospital Dx are adverse events as documented by Ohio Medicaid inpatient claims, ICD-9 CM diagnoses
†Ohio death registry adverse events, ICD-9 CM or ICD-10
‡Combined adverse event from both sources, a patient could either be counted as Hospital Dx or Die from, but not both for “All” outcome
Table 5.
Warfarin N = 203 | No warfarin N = 2,912 | |||||
---|---|---|---|---|---|---|
N | % | Per 100py | N | % | Per 100py | |
Hospital Dx ischemic stroke* | 11 | 5.42 | 2.79 | 147 | 5.05 | 3.34 |
Die from ischemic stroke† | 9 | 4.43 | 2.28 | 135 | 4.64 | 3.07 |
All ischemic strokes‡ | 18 | 8.87 | 4.56 | 263 | 9.03 | 5.98 |
Hospital Dx GI bleed | 28 | 13.79 | 7.53 | 259 | 8.89 | 6.04 |
Die from GI bleed | 1 | 0.49 | 0.27 | 8 | 0.27 | 0.19 |
All GI bleed | 29 | 14.29 | 7.79 | 265 | 9.1 | 6.18 |
Hospital Dx ICH | 2 | 0.99 | 0.49 | 25 | 0.86 | 0.55 |
Die from ICH | 0 | 0 | 0 | 5 | 0.17 | 0.11 |
All ICH | 2 | 0.99 | 0.49 | 29 | 1 | 0.64 |
Hospital Dx other bleed | 13 | 6.4 | 3.33 | 147 | 5.05 | 3.33 |
Die other bleed | 0 | 0 | 0 | 0 | 0 | 0 |
All other bleed | 13 | 6.4 | 3.33 | 147 | 5.05 | 3.33 |
*Hospital Dx are adverse events as documented by Ohio Medicaid inpatient claims, ICD-9 CM diagnoses
†Ohio death registry adverse events, ICD-9 CM or ICD-10
‡Combined adverse event from both sources, a patient could either be counted as Hospital Dx or Die from, but not both for “All” outcome
Hazard Ratios
In patients recommended for anticoagulation by the decision support tool, there was a trend towards a decreased hazard for stroke with actual warfarin treatment (Table 6). This difference did not become significant even after adjusting for the propensity of receiving warfarin and prescribed analgesics. Gastrointestinal bleeds, intracranial hemorrhage, and other bleeds were not significantly different between the two groups even after adjusting for the propensity of receiving warfarin. Adjusting for medications and other covariates not included in the decision support tool did not alter the findings.
Table 6.
Anticoagulation recommended By DST | ||||
---|---|---|---|---|
Referent group is those recommended for anticoagulation, but not actually receiving warfarin | ||||
Hazard ratio | Confidence interval | P | ||
All strokes | ||||
Unadjusted | 0.835 | 0.538 | 1.294 | 0.419 |
Adjusted* | 0.904 | 0.580 | 1.407 | 0.654 |
All GIB | ||||
Unadjusted | 0.867 | 0.564 | 1.333 | 0.516 |
Adjusted* | 0.869 | 0.562 | 1.342 | 0.525 |
All ICH | ||||
Unadjusted | 1.743 | 0.771 | 3.940 | 0.182 |
Adjusted* | 1.843 | 0.802 | 4.233 | 0.150 |
All other bleeds | ||||
Unadjusted | 1.093 | 0.700 | 1.706 | 0.697 |
Adjusted* | 1.091 | 0.695 | 1.714 | 0.704 |
* All adjusted models included the propensity score for receiving warfarin
In patients for whom withholding anticoagulation was recommended by the decision support tool (Table 7), there was a similar trend towards a decreased hazard of stroke in those who actually received warfarin. These patients had a statistically significant increased hazard of gastrointestinal bleeding. In the final adjusted Cox proportional hazards model using the covariate of propensity for warfarin prescribing, the relative hazard for gastrointestinal bleeding was 1.54 (p = 0.031). Hazard ratios for intracranial hemorrhage and other bleeds were not significant even after adjustment; however, there were few such outcomes in both groups. Adjusting for medications and other covariates not included in the decision support tool did not alter these findings.
Table 7.
Anticoagulation NOT recommended by DST | ||||
---|---|---|---|---|
Referent group is those not recommended for anticoagulation, and not receiving warfarin | ||||
Hazard ratio | Confidence interval | p | ||
All strokes | ||||
Unadjusted | 0.782 | 0.485 | 1.260 | 0.312 |
Adjusted* | 0.816 | 0.504 | 1.324 | 0.411 |
All GIB | ||||
Unadjusted | 1.293 | 0.881 | 1.897 | 0.189 |
Adjusted* | 1.539 | 1.040 | 2.271 | 0.031 |
All ICH | ||||
Unadjusted | 0.779 | 0.186 | 3.267 | 0.733 |
Adjusted* | 0.823 | 0.195 | 3.545 | 0.804 |
All other bleeds | ||||
Unadjusted | 1.013 | 0.574 | 1.787 | 0.964 |
Adjusted* | 1.008 | 0.567 | 1.793 | 0.978 |
*All adjusted models included the propensity score for receiving warfarin
DISCUSSION
We have shown that there is a high risk of major bleeding in patients for whom our atrial fibrillation decision support tool recommends withholding anticoagulant therapy but who actually receive such treatment and that anti-coagulating these patients may actually result in more harm than benefit. Patients who received anticoagulation, but the decision tool indicated they should not, had a 54% increase in the hazard of gastrointestinal bleeding. We were unable to demonstrate an increase in intracranial hemorrhage and other bleeding events, likely caused by the small number of such events. Strokes were decreased with anticoagulation as expected, but not significantly. These results indicate that gastrointestinal hemorrhage occurred at a greater rate compared to prevention of major stroke. Thus, the risk of anticoagulation likely outweighs its benefit in this group of patients for whom the decision support tool recommended withholding anticoagulant therapy.
In the group of patients for whom the decision support tool recommended anticoagulation therapy, there was a non-statistically significant trend towards a decreased hazard of stroke in those receiving anticoagulant therapy with no increased hazard of gastrointestinal bleeding. The lack of a statistically significant difference in stroke hazard may be secondary to the low overall use of warfarin in this cohort.12–20 As expected, intracranial hemorrhage and other bleeding were increased with warfarin use. While intracranial hemorrhage is devastating, the absolute risk is small. Furthermore, data are lacking to accurately predict future intracranial hemorrhage risk with resumed anticoagulation.43–45
Our study has several limitations. First, we did not have laboratory information documenting the intensity of anticoagulation therapy (i.e., INR values). Consequently, we were not able to adjust our analyses for the intensity of anticoagulation. Second, we were unable to account for the effect of aspirin use on outcomes. Aspirin reduces the risk of stroke, but it is not as effective as warfarin.46–48 Third, we could not reliably identify new subsequent events following the first adverse event (e.g. second stroke). Additionally, major bleeding occurred more frequently than expected. These reflect some of the limitations of using administrative data.49–52 Chart review would enhance the accuracy of diagnoses and warfarin and aspirin use. Lastly, this study can address neither individual patient preferences for receiving warfarin or for different health states nor physician barriers to warfarin use.
This study’s approach might be helpful for preclinical testing other decision support tools, especially with an appropriate dataset. The sole use of administrative data for this preclinical testing may not be sufficient because of issues of completeness and accuracy of the information contained in claims data. This concern is further highlighted by our finding a significant additional number of strokes through the examination of death registry information.
These preliminary results suggest that use of the atrial fibrillation decision support tool might result in more appropriate prescribing of warfarin particularly in patients for whom the balance of risk and benefit favors not treating. The low use of warfarin in this Ohio Medicaid cohort makes it difficult to tell whether the trend towards a decreased risk of stroke in those treated with warfarin concordant with the decision tool’s recommendation is significant. However, a clinical trial of the decision support tool would further clarify the impact and could incorporate physician decision making and patient preferences into the actual treatment decision. We envision the tool better informing such a shared decision making approach, not a substitute for patient-provider discussion.
In conclusion, our study of Ohio Medicaid patients with non-valvular atrial fibrillation demonstrated that a decision support tool for anticoagulation recommendations could identify patients at significant risk for gastrointestinal hemorrhage in whom the decision to anti-coagulate should be weighed more cautiously. The decision support tool suggested that 49% of patients should be considered for anticoagulation, whereas only 9.9% actually received such therapy in this group. Administrative claims data and death files may be insufficient to adequately test decision support tools for all outcomes before clinical trials or use. Further testing of the decision support tool in a clinical setting is desirable to determine if its use can significantly reduce acute strokes while only modestly increasing hemorrhagic events.
Acknowledgments
The authors appreciate the Ohio Department of Jobs and Family Services collaboration for supplying the Ohio Medicaid data. They wish to thank Ronnie D. Horner, PhD, Director and Professor, UC Institute for the Study of Health for his review of this study. They appreciate the valuable assistance from Anthony Leonard, PhD, UC Institute for the Study of Health for his guidance on analysis. Portions of this study have been previously presented as a poster at the Society of Medical Decision Making conference, October 2006.
Funding This study did not receive any internal or external funding.
Conflict of Interest None disclosed.
References
- 1.Laupacis A, Boysen G, Connolly S, et al. Risk factors for stroke and efficacy of antithrombotic therapy in atrial fibrillation. Analysis of pooled data from five randomized controlled trials. Arch Intern Med. 1994;154(13):1449–57. [DOI] [PubMed]
- 2.Ruigomez A, Johansson S, Wallander MA, Rodriguez LA. Incidence of chronic atrial fibrillation in general practice and its treatment pattern. J Clin Epidemiol. 2002;55(4):358–63. [DOI] [PubMed]
- 3.Wolf PA, Abbott RD, Kannel WB. Atrial fibrillation as an independent risk factor for stroke: the Framingham Study. Stroke. 1991;22(8):983–8. [DOI] [PubMed]
- 4.Singer DE. Overview of the randomized trials to prevent stroke in atrial fibrillation. Ann Epidemiol. 1993;3(5):563–7. [DOI] [PubMed]
- 5.The Boston Area Anticoagulation Trial for Atrial Fibrillation Investigators. The effect of low-dose warfarin on the risk of stroke in patients with nonrheumatic atrial fibrillation. N Engl J Med. 1990;323(22):1505–11. [DOI] [PubMed]
- 6.EAFT (European Atrial Fibrillation Trial) Study Group. Secondary prevention in non-rheumatic atrial fibrillation after transient ischaemic attack or minor stroke. Lancet. 1993;342(8882):1255–62. [PubMed]
- 7.Petersen P, Boysen G, Godtfredsen J, Andersen ED, Andersen B. Placebo-controlled, randomised trial of warfarin and aspirin for prevention of thromboembolic complications in chronic atrial fibrillation. The Copenhagen AFASAK study. Lancet. 1989;1(8631):175–9. [DOI] [PubMed]
- 8.Stroke Prevention in Atrial Fibrillation Study Investigators. Stroke Prevention in Atrial Fibrillation Study: final results. Circulation. 1991;84(2):527–39. [DOI] [PubMed]
- 9.Connolly SJ, Laupacis A, Gent M, Roberts RS, Cairns JA, Joyner C. Canadian Atrial Fibrillation Anticoagulation (CAFA) Study. J Am Coll Cardiol. 1991;18(2):349–55. [DOI] [PubMed]
- 10.Ezekowitz MD, Bridgers SL, James KE, Carliner NH, Colling CL, Gornick CC, et al. Warfarin in the prevention of stroke associated with nonrheumatic atrial fibrillation. Veterans Affairs Stroke Prevention in Nonrheumatic Atrial Fibrillation Investigators. N Engl J Med. 1992;327(20):1406–12. [DOI] [PubMed]
- 11.Singer DE, Hughes RA, Gress D, Sheehan MA, Oertel LB, Maraventano SW, et al. The effect of aspirin on the risk of stroke in patients with nonrheumatic atrial fibrillation: The BAATAF Study. Am Heart J. 1992;124(6):1567–73. [DOI] [PubMed]
- 12.Go AS, Hylek EM, Borowsky LH, Phillips KA, Selby JV, Singer DE. Warfarin use among ambulatory patients with nonvalvular atrial fibrillation: the anticoagulation and risk factors in atrial fibrillation (ATRIA) study. Ann Intern Med. 1999;131(12):927–34. [DOI] [PubMed]
- 13.Gottlieb LK, Salem-Schatz S. Anticoagulation in atrial fibrillation. Does efficacy in clinical trials translate into effectiveness in practice. Arch Intern Med. 1994;154(17):1945–53. [DOI] [PubMed]
- 14.Stafford RS, Singer DE. Recent national patterns of warfarin use in atrial fibrillation. Circulation. 1998;97(13):1231–3. [DOI] [PubMed]
- 15.Bungard TJ, Ghali WA, Teo KK, McAlister FA, Tsuyuki RT. Why do patients with atrial fibrillation not receive warfarin. Arch Intern Med. 2000;160(1):41–6. [DOI] [PubMed]
- 16.Samsa GP, Matchar DB, Goldstein LB, Bonito AJ, Lux LJ, Witter DM, et al. Quality of anticoagulation management among patients with atrial fibrillation: results of a review of medical records from 2 communities. Arch Intern Med. 2000;160(7):967–73. [DOI] [PubMed]
- 17.Device may reduce stroke risk from chronic atrial fibrillation. Mayo Clin Health Lett. 2006;24(1):4. [PubMed]
- 18.Fang MC, Stafford RS, Ruskin JN, Singer DE. National trends in antiarrhythmic and antithrombotic medication use in atrial fibrillation. Arch Intern Med. 2004;164(1):55–60. [DOI] [PubMed]
- 19.Johnston JA, Cluxton RJ Jr, Heaton PC, Guo JJ, Moomaw CJ, Eckman MH. Predictors of warfarin use among Ohio Medicaid patients with new-onset nonvalvular atrial fibrillation. Arch Intern Med. 2003;163(14):1705–10. [DOI] [PubMed]
- 20.Reynolds M, Shah J, Essebag V, Olshansky B, Friedman PA, Hadjis T, et al. Patterns and predictors of warfarin use in patients with new-onset atrial fibrillation from the FRACTAL Registry. Am J Cardiol. 2006;97(4):538–43. [DOI] [PubMed]
- 21.Hylek EM, Singer DE. Risk factors for intracranial hemorrhage in outpatients taking warfarin. Ann Intern Med. 1994;120(11):897–902. [DOI] [PubMed]
- 22.Fang MC, Chang Y, Hylek EM, Rosand J, Greenberg SM, Go AS, et al. Advanced age, anticoagulation intensity, and risk for intracranial hemorrhage among patients taking warfarin for atrial fibrillation. Ann Intern Med. 2004;141(10):745–52. [DOI] [PubMed]
- 23.Levine MN, Raskob G, Beyth RJ, Kearon C, Schulman S. Hemorrhagic complications of anticoagulant treatment: the Seventh ACCP Conference on Antithrombotic and Thrombolytic Therapy. Chest. 2004;126(3 Suppl):287S–310S. [DOI] [PubMed]
- 24.Singer DE, Albers GW, Dalen JE, Go AS, Halperin JL, Manning WJ. Antithrombotic therapy in atrial fibrillation: the Seventh ACCP Conference on Antithrombotic and Thrombolytic Therapy. Chest. 2004;126(3 Suppl):429S–56S. [DOI] [PubMed]
- 25.Albers GW, Dalen JE, Laupacis A, Manning WJ, Petersen P, Singer DE. Antithrombotic therapy in atrial fibrillation. Chest. 2001;119(1 Suppl):194S–206S. [DOI] [PubMed]
- 26.Fuster V, Ryden LE, Cannom DS, Crijns HJ, Curtis AB, Ellenbogen KA, et al. ACC/AHA/ESC 2006 guidelines for the management of patients with atrial fibrillation-executive summary: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines and the European Society of Cardiology Committee for Practice Guidelines (Writing Committee to Revise the 2001 Guidelines for the Management of Patients With Atrial Fibrillation). J Am Coll Cardiol. 2006;48(4):854–906. [DOI] [PubMed]
- 27.Cohen N, Almoznino-Sarafian D, Alon I, Gorelik O, Koopfer M, Chachashvily S, et al. Warfarin for stroke prevention still underused in atrial fibrillation: patterns of omission. Stroke. 2000;31(6):1217–22. [DOI] [PubMed]
- 28.Beyth RJ, Antani M, Covinsky KE, Miller DG, Chren MM, Quinn LM, et al. Why isn’t warfarin prescribed to patients with nonrheumatic atrial fibrillation. J Gen Intern Med. 1996;11(12):721–8. [DOI] [PubMed]
- 29.Gage BF, Boechler M, Doggette AL, Fortune G, Flaker GC, Rich MW, et al. Adverse outcomes and predictors of underuse of antithrombotic therapy in Medicare beneficiaries with chronic atrial fibrillation. Stroke. 2000;31(4):822–7. [DOI] [PubMed]
- 30.Singer DE, Go AS. Antithrombotic therapy in atrial fibrillation. Clin Geriatr Med. 2001;17(1):131–47. [DOI] [PubMed]
- 31.Hart RG, Benavente O, McBride R, Pearce LA. Antithrombotic therapy to prevent stroke in patients with atrial fibrillation: a meta-analysis. Ann Intern Med. 1999;131(7):492–501. [DOI] [PubMed]
- 32.Koroukian SM, Cooper GS, Rimm AA. Ability of Medicaid claims data to identify incident cases of breast cancer in the Ohio Medicaid population. Health Serv Res. 2003;38(3):947–60. [DOI] [PMC free article] [PubMed]
- 33.Shireman TI, Heaton PC, Gay WE, Cluxton RJ, Moomaw CJ. Relationship between asthma drug therapy patterns and healthcare utilization. Ann Pharmacother. 2002;36(4):557–64. [DOI] [PubMed]
- 34.Landefeld CS, Anderson PA, Goodnough LT, Moir TW, Hom DL, Rosenblatt MW, et al. The bleeding severity index: validation and comparison to other methods for classifying bleeding complications of medical therapy. J Clin Epidemiol. 1989;42(8):711–8. [DOI] [PubMed]
- 35.Beyth RJ, Quinn LM, Landefeld CS. Prospective evaluation of an index for predicting the risk of major bleeding in outpatients treated with warfarin. Am J Med. 1998;105(2):91–9. [DOI] [PubMed]
- 36.Eckman MH. Patient-centered decision making: a view of the past and a look toward the future. Med Decis Mak. 2001;21(3):241–7. [DOI] [PubMed]
- 37.Johnston JA, Eckman MH. Use of regression modeling to simulate patient-specific decision analysis for patients with nonvalvular atrial fibrillation. Med Decis Mak. 2003;23(5):361–8. [DOI] [PubMed]
- 38.Eckman MH, Pauker SG, eds. Decision Analytical Issues in the Management of Atrial Fibrillation. Philadelphia, PA: Lippincott-Raven Publishers; 1997.
- 39.Eckman MH, Levine HJ, Salem DN, Pauker SG. Making decisions about antithrombotic therapy in heart disease: decision analytic and cost-effectiveness issues. Chest. 1998;114(5 Suppl):699S–714S. [DOI] [PubMed]
- 40.Eckman MH, Falk RH, Pauker SG. Cost-effectiveness of therapies for patients with nonvalvular atrial fibrillation. Arch Intern Med. 1998;158(15):1669–77. [DOI] [PubMed]
- 41.Landefeld CS, Goldman L. Major bleeding in outpatients treated with warfarin: incidence and prediction by factors known at the start of outpatient therapy. Am J Med. 1989;87(2):144–52. [DOI] [PubMed]
- 42.Rosenbaum PR. Observational Studies. 2nd ed. New York, NY: Spinger; 2002.
- 43.Eckman MH, Rosand J, Knudsen KA, Singer DE, Greenberg SM. Can patients be anticoagulated after intracerebral hemorrhage? A decision analysis. Stroke. 2003;34(7):1710–6. [DOI] [PubMed]
- 44.Fang MC, Go AS, Hylek EM, Chang Y, Henault LE, Jensvold NG, et al. Age and the risk of warfarin-associated hemorrhage: the anticoagulation and risk factors in atrial fibrillation study. J Am Geriatr Soc. 2006;54(8):1231–6. [DOI] [PMC free article] [PubMed]
- 45.Rosand J, Eckman MH, Knudsen KA, Singer DE, Greenberg SM. The effect of warfarin and intensity of anticoagulation on outcome of intracerebral hemorrhage. Arch Intern Med. 2004;164(8):880–4. [DOI] [PubMed]
- 46.Lip GY, Edwards SJ. Stroke prevention with aspirin, warfarin and ximelagatran in patients with non-valvular atrial fibrillation: a systematic review and meta-analysis. Thromb Res. 2006;118:321–33. [DOI] [PubMed]
- 47.Sam C, Massaro JM, D’Agostino RB Sr, Levy D, Lambert JW, Wolf PA, et al. Warfarin and aspirin use and the predictors of major bleeding complications in atrial fibrillation (the Framingham Heart Study). Am J Cardiol. 2004;94(7):947–51. [DOI] [PubMed]
- 48.Gage BF, van Walraven C, Pearce L, Hart RG, Koudstaal PJ, Boode BS, et al. Selecting patients with atrial fibrillation for anticoagulation: stroke risk stratification in patients taking aspirin. Circulation. 2004;110(16):2287–92. [DOI] [PubMed]
- 49.Walkup JT, Boyer CA, Kellermann SL. Reliability of Medicaid claims files for use in psychiatric diagnoses and service delivery. Adm Policy Ment Health. 2000;27(3):129–39. [DOI] [PubMed]
- 50.Hennessy S, Bilker WB, Weber A, Strom BL. Descriptive analyses of the integrity of a US Medicaid claims database. Pharmacoepidemiol Drug Saf. 2003;12(2):103–11. [DOI] [PubMed]
- 51.Twiggs JE, Fifield J, Apter AJ, Jackson EA, Cushman RA. Stratifying medical and pharmaceutical administrative claims as a method to identify pediatric asthma patients in a Medicaid managed care organization. J Clin Epidemiol. 2002;55(9):938–44. [DOI] [PubMed]
- 52.Glynn RJ, Monane M, Gurwitz JH, Choodnovskiy I, Avorn J. Agreement between drug treatment data and a discharge diagnosis of diabetes mellitus in the elderly. Am J Epidemiol. 1999;149(6):541–9. [DOI] [PubMed]