The Unrecognized Psychosocial Factors Contributing to Bleeding Risk in Warfarin Therapy
Background and Purpose—Warfarin is an effective drug for the prevention of thromboembolism in the elderly. The major risk for patients taking warfarin is bleeding. We aimed to assess the impact of psychosocial factors, including mood, cognition, social isolation, and health literacy on warfarin instability among community-based elderly patients.
Methods—A case–control study was conducted between March 2008 and June 2009 in a community-based setting. Cases were patients previously stabilized on warfarin who recorded an international normalized ratio ≥6.0. Control subjects were patients whose international normalized ratio measurement was maintained within the therapeutic range. Patient interviews investigated potential predisposing factors to elevated International Normalized Ratio levels.
Results—A total of 486 patients were interviewed: 157 cases and 329 control subjects, with an approximate mean age of 75 years. Atrial fibrillation was the most common primary indication. Adjusted multivariate logistic regression revealed impaired cognition (OR, 1.9; 95% CI, 1.0 to 3.6), depressed mood (OR, 2.2; 95% CI, 1.2 to 3.9), and inadequate health literacy (OR, 4.0;95% CI, 2.1 to 7.4) were associated with increased risk of an elevated International Normalized Ratio.
Conclusions—This study identified impaired cognition, depressed mood, and inadequate health literacy as risk factors for warfarin instability. These had a similar impact to well-recognized demographic, clinical, and medication-related factors and are prevalent among the elderly. These findings suggest that elderly patients prescribed warfarin should be reviewed regularly for psychosocial deficits.
Warfarin is used widely for the treatment and prophylaxis of thromboembolic disorders. It is routinely prescribed for patients with atrial fibrillation to prevent stroke.1 Warfarin's narrow therapeutic window coupled with a variable dose response necessitates periodic monitoring of the International Normalized Ratio (INR), which acts as a marker of warfarin stability as well as a surrogate measure of bleeding risk.1 In Australia, the United Kingdom, and Canada, a shared model of care for monitoring and management exists for the management of warfarin therapy involving specialists, general practitioners (GP), and laboratory services; with GPs retaining clinical responsibility.2,3 Despite routine monitoring, patients taking warfarin are at risk of hemorrhage. Although warfarin-related hemorrhages are usually minor (eg, nose bleeds and bruising), major intracranial, gastrointestinal, and genitourinary bleeding occurs at a rate of 1.2% to 8.1% per annum and are an important deterrent to the use of warfarin.1
To maintain the balance between efficacy and bleeding, warfarin should be maintained within an INR range of 2.0 to 3.0 for patients with atrial fibrillation and 2.5 to 3.5 for patients after mechanical valve replacement.1 The factors that influence the dose needed to maintain the INR within these ranges have been investigated extensively. Demographic factors, including advancing age,4,5 gender,6 race,6 comorbidities (eg, malignancy, diabetes, and previous cerebrovascular events)7,8, and specific concomitant medications, including antibiotics, have been reported to increase the risk of bleeding at the time of being on warfarin therapy.1
Despite this, little is known about how the risk of bleeding is additionally influenced by psychosocial factors such as cognition, social isolation, depressed mood, and health literacy.9 Thirty percent of patients starting warfarin therapy are aged >70 years,2 and it is this population who are at highest risk of developing these characteristics.
The purpose of this study was to determine whether psychosocial factors such as these contributed to elevated INRs (INR ≥6.0) in a community-based population.
Patient Identification and Eligibility
This case–control study recruited patients from a large pathology provider in Melbourne, Australia. Inclusion criteria included aged ≥18 years, residing within a 50-km radius of the central business district, and stabilized on warfarin for a minimum of 3 months. Patients were excluded if they were hospitalized at the time of their INR, living in a care facility, or unable to be interviewed within a 30-day period after registering the INR level of interest.
Cases were identified by an elevated INR ≥6.0 measured during routine monitoring and verified by an immediate duplicate test. An INR ≥6.0 was selected because such levels are unlikely to be caused by individual fluctuations and patients with INRs at and above this level have been shown to be at a significantly higher risk of hemorrhage.1,10 In the system of care in Melbourne, an INR that falls out of the therapeutic range triggers an immediate response by the pathology provider, and patients are reviewed within a minimum of 6 weeks.
For every case, 2 control subjects were randomly selected from the pathology provider's list of patients with INR levels taken on the day the case was identified. INRs of all control subjects were within the therapeutic range for their clinical indication with a small allowance (±0.3 INR units) for acceptable variability.
Participation was initially sought by the pathology provider, who identified and mailed patients an explanatory statement outlining the nature and requirements of the study. After a 7-day period, providing patients the opportunity to decline participation, eligibility was confirmed; consent gained and an interview time was arranged either at their place of residence or at a designated research center. A structured interview to assess potential risk factors took approximately 1 hour to complete. In addition to interviewing the patient, a medication history was obtained and a number of quantitative measures were administered. Medical histories, inclusive of demographic and physiological details and current medication, were confirmed through consultation with the patient's GP.
The structured patient interview was piloted on a separate study of 40 patients in 2007.2 We collected the following information: (1) demographic details: age, socioeconomic status, race/ethnicity (as defined by the patient), primary indication for taking warfarin; (2) current health, comorbidities, visual acuity11; (3) medication review (prescription and nonprescription), self-reported medication adherence12; and (4) psychosocial factors: cognitive function,13 possible signs of depressed mood,14 social connectedness,15 functional health literacy,16 independence with activities of daily living.17,18 Each of these tools has been validated for use in the elderly population. We applied previously adopted cutoffs for each tool.
A total of 150 cases and 300 control subjects were required to provide 80% power to detect a true OR of 2.0 for risk factors having a prevalence of 20% using a probability value of 0.05 as the threshold for statistical significance. To allow for adjustment in a multiple logistic regression model, the sample size was inflated by 10% for each of 10 potential confounders.
Univariate comparison of risk factors was made between cases and control subjects. ORs with 95% CIs were calculated by standard methods. Logistic regression models were used to assess the independent effects of specific characteristics on the risk of having an elevated INR and the significance of interaction terms. Multivariate analysis was undertaken to explore the effect of psychosocial characteristics after controlling for potentially important confounders, including age, sex, race, English as a second language, working/retirement, and primary indication and duration of warfarin therapy. Logistic regression was also used to investigate different definitions of psychosocial deficit using the relevant measured variables in combination as a single index. Presence of mild cognitive impairment was considered as the first definition of psychosocial impairment because it identified the greatest number of patients as “impaired.” Social isolation was included last because it identified the fewest patients and had the weakest association with elevated INR in multivariate analysis. Possible signs of depression were considered before poor health literacy because there is some evidence of an association between health literacy and cognition,19 which was the starting point for defining impairment. Statistical analyses were performed using Stata 10.0 (Stata Corp, College Station, TX).
A total of 157 cases and 329 control subjects were recruited between March 1, 2008, and July 30, 2009, as outlined in the Figure. There were only small differences between enrolled and unenrolled cases and control subjects in relation to all variables collected on unenrolled patients, that is, age (mean and range), gender, INR level, and warfarin dose.
Table 1 summarizes univariate comparisons of the characteristics of cases and control subjects. Cases were more likely than control subjects to be maintained at the higher INR target range according to the primary indication, to have multiple comorbidities, and have poor visual acuity. Cognitive impairment, social isolation, depressed mood, poor health literacy, and functional impairment were all more common among cases in the univariate analysis.
Cases were more likely to be prescribed prednisolone or antibiotics with 20% taking contraindicated macrolides or metronidazole. Cases were less likely to use a tablet dispenser/dosette for medication management and more likely to self-report poor medication adherence. A high proportion of cases reported taking >10 medications.
In the multivariate analysis, cognitive impairment, social isolation, depressed mood, poor health literacy, and functional impairment remained strongly related to a higher bleeding risk (Table 2). Poor health literacy displayed the strongest relationship (adjusted OR, 4.8; 95% CI, 2.9 to 7.8) followed by a high score in the Geriatric Depression Scale-5 (adjusted OR, 3.1; 95% CI, 2.9 to 7.8). Additional adjustment for medications and comorbidities made little difference to these results.Table 2 shows that the strength of the associations diminished when all of the psychosocial factors were present together in the model.
Table 3 shows that the presence of psychosocial factors in combination was associated with increased risk of elevated INR level and that the magnitude of this effect increased with the number of factors considered. A definition of psychosocial deficit based on the presence of all of cognitive impairment, depressed mood, and poor health literacy appeared optimal (adjusted OR, 4.7; 95% CI, 2.7 to 8.2) with little gain from the additional consideration of social isolation.
This large case–control study demonstrated that mild cognitive impairment, social isolation, depressed mood, and poor health literacy are strongly related to warfarin instability. The impact of these factors remained high after accounting for potential confounders, including to age, sex, race, English as a second language, current employment status, duration of therapy, and primary indication for warfarin. This indicated that they had an independent influence on the bleeding risk associated with warfarin therapy. To our knowledge, this is the first study to conduct a comprehensive evaluation of potential risk factors including psychosocial issues as risk factors for elevated INRs. The results are particularly relevant for an aging, community-based population typical of where warfarin treatment is commonly used.
Among similar previous studies, Fang et al20 used the Shortened test of Functional Health Literacy in Adults (s-TOFHLA) to investigate the association among health literacy, warfarin knowledge, adherence, and anticoagulation control in patients presenting to an anticoagulation clinic. Like in our study, they identified a high prevalence of poor health literacy in their cohort. However, they did not identify an impact on adherence or warfarin control.20 Notably they controlled for cognitive impairment as a confounder of health literacy, whereas we found it was an independent risk factor.20
The disparate findings may in part be a consequence of their smaller sample size, demographically different population (a higher proportion of black and Hispanic patients), and differences in the systems of care. In their study, patients were recruited from an anticoagulation clinic where warfarin education was provided and a more satisfactory system of care may have been in place to manage patients with poor health literacy.
Our finding that cognitive impairment was independently associated with an increased bleeding risk is supported by Lackie et al21 who hypothesized that cognitive impairment and functional decline may be surrogate markers of physiological decline and changes in warfarin metabolism.21
The role of depressed mood on INR control has not previously been documented. Our study demonstrates an association between depressed mood and elevated INR and suggests a need for ongoing assessment of possible depression in this high-risk patient group.
Systems of managing warfarin, and other complex medications, should include an assessment of a patient's social network, cognition, and mood because these might identify a subset of patients at high risk. When ≥1 of these factors are compromised, patient suitability for warfarin should be evaluated. If warfarin is prescribed, these patients should have close and proactive monitoring. Self-reported medication adherence is an inadequate measure of risk.
The long-term efficacy and safety of newer anticoagulants remains to be determined. Warfarin remains an effective and widely used anticoagulant and the preferred drug in many circumstances, including atrial fibrillation.22,23 Risks may be mitigated by additional support and supervision, for example, by involving relatives or district nurses in the supervision of medication.21
Like with all case–control studies, our results may be influenced by selection, recall, and interviewer bias. We minimized selection bias by choosing a random sample of control subjects from a large metropolitan pathology provider managing a representative population. Because our results were necessarily based on patient self-report, recall bias may have impacted on our findings. Cognitive impairment may also have impacted on the patient's ability to accurately recall information. We aimed to reduce recall bias and increase accuracy of information obtained by interviewing patients within 30 days of the reported INR level. We also verified medication history and comorbid conditions through the patients' GP medical records. Neither interviewers nor patients could be blinded to the elevated INR, but this was compensated for by using a scripted interview, having a senior researcher periodically audit interviews to ensure consistency, and by using validated tools to assess psychosocial status.
We found cognition, mood, and health literacy strongly influenced the stability of INR levels in patients on warfarin. The presence of multiple psychosocial factors produced a 3.4-fold increase in bleeding risk as measured by the surrogate of an elevated INR. This is comparable to the risk conveyed by other well-established demographic, clinical, and medication-related factors.
We therefore recommend that doctors prescribing warfarin regularly assess the identified psychosocial factors, particularly in elderly patients, and take account of their presence when determining how warfarin therapy should best be managed. When the responsibility for care is fragmented among the specialist, GP, and anticoagulant service, it must be clear who has primary responsibility for monitoring the patient's psychosocial circumstances. When significant impairment is found, patients should be re-evaluated for their suitability for warfarin and steps put in place to ensure adequate supervision of their therapy. These, and other, well-known risk factors should be flagged by computerized systems as a part of long-term management.
Sources of Funding
This work was supported by grants from the National Health and Medical Research Committee (436763) and approved by Human Research Ethics Committees of Monash University and Cabrini Health.
We acknowledge Melbourne Pathology and Prof Leon Piterman for assistance with recruitment and patients and GPs for participating in the research.
- Received February 11, 2011.
- Revision received May 4, 2011.
- Accepted May 5, 2011.
- © 2011 American Heart Association, Inc.
- Lowthian J,
- Joyce C,
- Diug BO,
- Dooley M
- Hughes M,
- Lip GYH
- DiMarco JP,
- Flaker G,
- Waldo AL,
- Corley SD,
- Greene HL,
- Safford RE,
- et al
- Mant J,
- Hobbs FD,
- Fletcher K,
- Roalfe A,
- Fitzmaurice D,
- Lip GY,
- et al
- Uyttenboogart M,
- Stewart RE,
- Vroomen PC,
- De Keyser J,
- Luijcxkx G
- Lackie C,
- Garbarino K,
- Pruetz J