Search tips
Search criteria 


Logo of jgimedspringer.comThis journalToc AlertsSubmit OnlineOpen Choice
J Gen Intern Med. 2008 October; 23(10): 1589–1594.
Published online 2008 July 10. doi:  10.1007/s11606-008-0708-8
PMCID: PMC2533388

Patient Reported Receipt of Medication Instructions for Warfarin is Associated with Reduced Risk of Serious Bleeding Events

Joshua P. Metlay, MD, PhD,corresponding author1,2,3,4,5,6 Sean Hennessy, PharmD, PhD,1,2,3 A. Russell Localio, JD, PhD,1,3 Xiaoyan Han, MS,1,2 Wei Yang, MS,1,2 Abigail Cohen, PhD,1,2,3 Charles E. Leonard, PharmD,1,2,3 Kevin Haynes, PharmD,1,3 Stephen E. Kimmel, MD, MSCE,1,2,3,4 Harold I. Feldman, MD, MSCE,1,2,3,4 and Brian L. Strom, MD, MPH1,2,3,4



Adverse drug events are an important cause of preventable hospitalizations.


To identify whether patient report of receipt of medication instructions and markers of complex care (multiple physicians, recent hospitalization) predict the risk of serious bleeding for older adults on warfarin.


Prospective cohort study of older adults.


Subjects filled new or refill prescriptions for warfarin at the time of enrollment.


Hospitalizations were identified through a state-wide registry. Discharge summaries of hospitalizations for possible warfarin related bleeding events were reviewed by trained abstractors and clinical experts. Incidence rate ratios (IRR) were estimated based on person-months of exposure using Poisson regression models.


From March 2002 through May 2003, we enrolled a total of 2346 adults on warfarin. Over a two-year follow-up period, there were 126 hospitalizations due to warfarin-related bleeding (4.6 hospitalizations per 100 person-years of exposure). Patients who reported receiving medication instructions from either a physician or nurse plus a pharmacist had a 60% reduced rate of subsequently experiencing a serious bleeding event over the next 2 years (adjusted IRR 0.40, 95% CI 0.24–0.68). Having ≥4 physicians providing medication prescriptions over the last 3 months and filling prescriptions at >1 pharmacy over the last 3 months were independently associated with increased bleeding rates (adjusted IRRs 2.37, 95% CI 1.22–4.57 and 1.61, 95% CI 0.97–2.67, respectively).


The rate of warfarin-related hospitalization for bleeding is substantially lower for patients who report receiving medication instructions from a physician or nurse and a pharmacist.

KEY WORDS: anticoagulation, patient communication, medication safety


Warfarin is frequently cited as a leading drug involved in preventable serious adverse drug events.1 Management of warfarin is improved in monitored settings, such as clinical trials and anticoagulation clinics, where close attention is given to levels of anticoagulation and warfarin dose.2 We hypothesized that barriers to optimal communication between patients, pharmacists and physicians increases the risk of adverse events due to warfarin.

A prior study has demonstrated that disagreement between patient and provider reports of warfarin dosing regimens is associated with excessive levels of anticoagulation.3 However, no studies have linked the adequacy of medication information communication to patients with subsequent clinical outcomes. Thus, this study sought to identify patient-reported characteristics of information exchange with physicians, nurses, and pharmacists that subsequently predicted serious bleeding events for older adults on outpatient warfarin therapy.


We conducted a prospective cohort study of serious bleeding events among older adults receiving warfarin. The study protocol was approved by the institutional review board at the University of Pennsylvania. All study participants provided written informed consent.

Study Site

The study sample was drawn from beneficiaries of the Pennsylvania Pharmaceutical Assistance Contract for the Elderly (PACE). This state-run program offers comprehensive prescription coverage to Pennsylvanians over the age of 65 who are not eligible for Medicaid and meet income eligibility requirements.

Subject Enrollment

Details regarding the process of subject enrollment have been reported previously.4 Briefly, subjects were identified through electronic claims transmitted to the PACE program by participating pharmacies. We used a sampling algorithm intended to recruit 2000 new or continuing users of warfarin over a 13-month sampling period (May 1, 2002 through May 31, 2003). Exclusion criteria for the analyses presented in this paper included absence of any claims for warfarin in the PACE database during the follow-up time period and residence in a skilled nursing facility.

Data Collection

Telephone Interviews

Participants completed an interview at baseline and 12 months. For participants who were unable to complete the interview (typically due to cognitive impairment), a proxy interview was sought with a next of kin or caregiver. The baseline interview was conducted within two to four weeks of the pharmacy visit to fill warfarin. Cognitive function was assessed with the six-item cognitive impairment test (6 CIT, score range 0–28),5,6 which at a cutoff of scores ≥8 has a sensitivity and specificity for mild to severe dementia of 90% and 100%, respectively.6 Additional questions regarding receipt of medication instructions were derived through focused interviews conducted by a geriatric nurse with older adults visiting community centers prior to the initiation of this study.4 Specifically, we asked patients to indicate receipt of “any instructions that you may have been given for your warfarin. These instructions might include things you shouldn’t do when taking your medication and things you should do. They could have been written instructions or spoken instructions. You could have received the instructions from the doctor, nurse, or pharmacist. These are not related to the instructions written on your pill bottle. Were you given any instructions of this sort?”

Hospitalization Data

All subjects were followed for 24 months from the month of enrollment. In order to identify hospitalizations during the follow-up period, data were obtained from the Pennsylvania Healthcare Cost Containment Council (PHC4) for within-state hospitalizations. These data are collected quarterly under a mandatory reporting law. We developed a broad list of ICD-9 CM codes suggestive of possible warfarin-related bleeding events. For any hospitalization with a discharge diagnosis (whether principal or not) matching any code on the list, we requested a copy of the discharge summary from the hospital. The codes were purposely selected to maximize sensitivity for case finding.7

Hospital discharge summaries were abstracted to collect information on four criteria relevant to each hospitalization’s likelihood of being due to a warfarin-related bleeding event: a) evidence of clinical signs and symptoms of bleeding, b) evidence of elevated international normalized ratio (INR) upon admission, c) a physician’s attribution in the discharge summary of the hospitalization as due to warfarin toxicity, and d) evidence of administration of vitamin K or fresh frozen plasma. Any hospitalization with clinical signs and symptoms of bleeding with at least one of the other criteria was considered a probable warfarin-related bleeding event. All hospitalizations deemed a probable event were then independently reviewed by two authors. The reviewers independently determined whether a hospitalization was definitely due to warfarin-related bleeding. When the two reviewers disagreed, a third expert independently reviewed the discharge summary and the majority opinion ruled. The rate of initial independent agreement on the presence or absence of a definite warfarin-related bleeding even was 78% (95% CI, 74%–82%).7

Our primary measure of serious bleeding event was any hospitalization due to warfarin-related bleeding.8 However, we separately collected information on whether the bleeding event included intracranial hemorrhage and/or resulted in death.

Data Analysis

Defining Exposure Time

Time at risk was divided for purposes of analysis into 24 consecutive 30-day periods based on the date of enrollment. In keeping with the PACE program’s convention of allowing prescriptions only for 30-day periods, we assumed that each prescription recorded was a 1-month supply.

Defining Risk Factors

For each month of exposure, we counted the number of non-toxicity-related hospitalizations for that month and the prior month. From the PACE database, we calculated the number of unique prescriptions filled per month as a measure of comorbidity and medication complexity.9 We also estimated the total number of different prescribing physicians and unique pharmacies for 3-month rolling windows of exposure to assess the frequency of patient contacts with different physicians. Study subjects who reported receipt of medication instructions at baseline were categorized as receiving instructions throughout the follow-up period. However, for those patients reported not receiving instructions at the baseline interview, their status was updated at the 12-month follow-up interview.

Defining Hospitalization Outcomes

Outcome events were counted if they occurred during a month with a prescription for warfarin or within 30 days of such a prescription month. We disregarded bleeding events that occurred more than 1 month after a drug exposure month, whether at the end of a sequence of exposure months or during a “gap” of known drug prescriptions, because we did not have concurrent evidence of exposure to warfarin during those periods in order to calculate unbiased estimates of bleeding risk.

Statistical Analysis

We calculated the overall rate of observed hospitalizations due to warfarin-related bleeding. We calculated incidence rate ratios (IRRs) using multivariable Poisson regression analysis on person-month level data with the number of warfarin related hospitalizations as the dependent variable, and with robust variance estimates (generalized estimating equations) to allow for differing rates of toxicity across subjects that could not be explained by the observed covariates.10 Variables were then considered for a complete-case-based multivariable model if they were statistically significant in bivariable analysis ( < 0.1) or were pre-specified of clinical interest based on risk factors identified in prior studies. (i.e., use of anticoagulation clinic, cognitive function, advanced age, multiple comorbid conditions).1114

Because of the patterns of missing data, use of only subject months with complete data reduced the sample size for multivariable regression (93.3% of all subject months had complete data). We tested the potential impact on bias of using complete case analysis by using multiple imputation to generate a full dataset for comparison purposes.15 The results from multiple imputation did not differ from those of the complete case analysis and are not presented.

All analyses were conducted with SAS Version 9.1 (Cary, NC).


Of 3956 eligible patients filling warfarin prescriptions and contacted by the PACE program, 2370 (60%) consented to participate and completed the baseline telephone interview. Of these, 95.7% ( = 2268) completed the 12 month follow-up interview and 85.0% ( = 2015) completed 24 months of follow-up. The major reason for failure to complete the follow-up interviews was death ( = 325, 13.7% of the cohort).

When compared to data from PACE, 23 members of the enrolled cohort (1.0% of total) did not have any claims for warfarin (i.e., were erroneously identified initially) and were dropped. The most common indications for warfarin prescriptions included atrial fibrillation (39%), deep venous thrombosis (21%), stroke (18%), heart valve replacement (11%), and pulmonary embolism (9%).

Characteristics of the Cohort

The majority of study subjects were female (77%), white non-Hispanic (95%), and >75 years of age (65%). Only 8% of the participants required a proxy to complete the baseline telephone interview. However, an additional 9% of participants had CIT scores in the moderate and severe cognitive impairment range (scores 8–10 and ≥11, respectively) (Table 1).6

Table 1
Characteristics of Study Subjects

Only 55% of participants reported that they received any medication instructions from a physician or nurse at the time they received their warfarin prescription and 31% reported receiving instructions from both a physician or nurse and a pharmacist (Table 2).

Table 2
Self-reported Information on Medication Utilization Practices

Warfarin-related Bleeding Events

The median length of exposure on warfarin was 11 months (range 1 to 24 months). We identified a total of 126 hospitalizations that were determined to be warfarin-related bleeding events. While the majority of events were gastrointestinal bleeds (67% of total), 11% ( = 14) were intracranial bleeding events, 3% ( = 4) were trauma related and 4% ( = 5) resulted in patient death. Overall, 56 out of the 126 hospitalizations (44%) had a documented international normalized ratio (INR) elevation (>3.5) at the time of admission.

The overall risk of hospitalization due to warfarin-related bleeding event was 4.6 events per 100 person-years of exposure to warfarin (95% CI 3.8, 5.5). The risk of intracranial bleeding was 0.5 events per 100 person-years and the risk of bleeding resulting in death was 0.2 events per 100 person-years. For new users of warfarin, the risk of any serious bleeding event was 4.5 events per 100 person-years of exposure to warfarin (95% CI 2.9, 6.8); and for chronic users of warfarin the risk was 4.7 events per 100 person-years of exposure (95% CI 3.7, 5.8). In addition, the risk of bleeding was not significantly different during the first month of follow-up in the cohort compared to all other months (unadjusted IRR = 0.9, 95% CI 0.4, 1.8).

Risk Factors

In unadjusted analyses, the following factors were associated with a reduced rate of subsequent hospitalization due to warfarin-related bleeding: reporting that medication instructions were received from either a physician or nurse plus a pharmacist (unadjusted IRR = 0.48 95% CI 0.28, 0.75), reporting that it was clear what to do if a dose of warfarin was missed (unadjusted IRR = 0.65, 95% CI 0.43, 0.98), and reporting that it was clear what other drugs to avoid while on warfarin (unadjusted IRR = 0.55, 95% CI 0.34, 0.88).

In analyses adjusting for age, cognitive function, NSAID and/or aspirin use, total number of medications filled, and indication for warfarin, a patient report that he/she received medication instructions for warfarin from either a physician or nurse plus a pharmacist was associated with a reduced risk of subsequent hospitalization due to warfarin-related bleeding compared to reporting that no instructions were received (adjusted IRR 0.40, 95% CI 0.24, 0.68) (Table 3). In absolute terms, subjects reporting no receipt of medication instructions experienced 8.0 bleeding events per 100 person-years of warfarin exposure compared to 3.3 events per 100 person-years of exposure for patients reporting receipt of instructions from either a physician or nurse plus a pharmacist. In addition, filling prescriptions written by ≥4 physicians in the last 3 months compared to only 1 physician writing prescriptions (adjusted IRR 2.37, 95% CI 1.22, 4.57) and filling prescriptions at >1 pharmacy in the last 3 months (adjusted IRR 1.61, 95% CI 0.97, 2.67) were independently associated with the risk of serious bleeding. Other patient factors, such as education level, and visual and auditory functions were neither independent predictors of serious bleeding events nor modifiers of the association between report of receipt of medication instruction and risk of bleeding.

Table 3
Unadjusted and Adjusted Incidence Rate Ratios for Hospitalizations due to Warfarin-related Bleeding*

In a separate analysis adjusting for all the same factors, substituting the type of instructions (verbal vs. written) for the source of instruction, we observed that compared to receiving no instructions, patient reports of receiving written instructions alone or written plus verbal instructions were associated with a reduced risk of bleeding events (adjusted IRR 0.38, 95% CI 0.18, 0.80 and adjusted IRR 0.47, 95% CI 0.29, 0.76, respectively). Patient report of receiving verbal instructions alone was not associated with a reduced risk of bleeding events compared to no instructions (adjusted IRR 0.95, 95% CI 0.51, 1.76).

These specific risk factor estimates remained unchanged after excluding all proxy interviews. In addition, there was no evidence for an interaction between the effect of patient report of receipt of medication instructions and whether the patient was a new vs. chronic user of warfarin. Finally, restriction of the outcome to those bleeding events with a documented INR elevation (>3.5) at the time of admission yielded even stronger measures of association. For example, patient report that he/she received medication instructions from either a physician or nurse plus a pharmacist was associated with an 80% reduction in the subsequent risk of hospitalization due to warfarin-related bleeding (adjusted IRR 0.21, 95% CI 0.09, 0.50).

In models that examined the risk of hospitalization due to any cause during the follow-up period, exposure to multiple physicians and multiple pharmacies remained significant risk factors for any type of hospitalization. However, reporting that medication instructions for warfarin were received from either a physician or nurse plus a pharmacist compared to no instructions was not significantly associated with the risk of any hospitalization (adjusted IRR 0.87, 95% CI 0.71, 1.05).


Patients who report receiving medication instructions from either a physician or nurse plus a pharmacist after they fill prescriptions for warfarin were at substantially reduced risk of experiencing hospitalization due to warfarin-related bleeding over the next two years. Moreover, this reduced risk was independent of patient age, cognitive function, home living arrangement, and number of co-administered medications. The reduced risk was also specific for hospitalization due to warfarin-related bleeding as reporting receipt of medication instructions was not associated with a reduced risk of hospitalization overall.

The overall rate of serious bleeding events in this study (4.6 events per 100 patient-years) is comparable to prior studies which reported rates of serious or major bleeding ranging from 4 to 7 events per 100 patient-years of warfarin exposure.12,16 Prior studies of risk for warfarin-related bleeding events have largely emphasized patient clinical and demographic factors,12,13,17 along with co-administration of potential interacting medications.18 Our study adds to this literature by demonstrating that if patients do not recall receiving instructions on the proper way to use warfarin, they are at increased risk of suffering a serious warfarin-related bleeding event. It is important to recognize that for community-dwelling patients, few systems currently exist to monitor administration and prevent medication errors.19 In this regard, it is notable that only 6% of patients in this study reported receipt of care in a clinic setting dedicated to monitoring anticoagulation levels.

Importantly, we measured patient report of receipt of medication instruction, not the actual receipt of medication instruction. It is likely that in many cases, medication instructions were actually provided by physicians or nurses and pharmacies, but were not recalled by patients when interviewed. Nevertheless, the fact that patients do not recall receiving these instructions is a key predictor of future adverse events.

It is possible that patients’ reports that they received medication instructions from their health care providers might be a surrogate for other elements, such as trust in providers, that improve clinical outcomes by increasing medication adherence and reducing adverse drug events. Physician offices that are more effective at providing medication instruction may also more closely monitor interacting drugs and more appropriately monitor INR levels. In addition, patients who report receiving medication instructions may be more actively involved in managing their healthcare and thus more likely to prompt physicians and nurses for such instructions. Moreover, such patients may be more likely to act to reduce their risk of bleeding, such as reporting to clinic appointments for INR monitoring, and reporting mild symptoms that may identify bleeds at an earlier stage. Thus, this study does not establish that failure to report receipt of medication instructions is causally related to subsequent warfarin-related bleeding events.

A key strength of this study was the prospective design which allowed us to assess patient reported behaviors and knowledge prior to the occurrence of any adverse events, avoiding ascertainment bias. However, one limitation of this study is the potential non-representativeness introduced by non-participants. Our limited comparisons of characteristics indicate that non-participants were older, more likely to be living alone and in the lower income categories.4 We also could not measure all factors that might have confounded the observed associations in this study. Previously reported studies predicting bleeding events for patients on warfarin have included specific comorbid conditions (e.g., diabetes mellitus, GI bleeding, alcoholism, hepatic disease), drug interactions (e.g., antiplatelet agents) and laboratory abnormalities (renal insufficiency),16, 20 which we could not control for in this study.

Prior work has established that as patients become more active and informed participants in health conversations with providers, outcomes are likely to improve.21 Indeed, since the time of this study, new regulation now requires the distribution of a medication guide with all warfarin prescriptions,22 and the Centers for Medical Services now provides a reimbursement mechanism for medication counseling for Medicare recipients,23 but the effectiveness of these interventions is unknown. Regardless of whether such interventions lead to more consistent and better medication counseling activities, a key finding of this study is that patients who do not recall such instructions are at increased risk for bleeding events. Future studies should examine the effectiveness and cost-effectiveness of office-based or home-based programs to identify and more carefully manage these high risk patients on warfarin.

In summary, patients and their caregivers remain key participants in assuring the safe use of medications in outpatient settings. Developing interventions to screen for and manage high risk patients and situations is a critical next step in assuring safer use of warfarin and possibly other high risk medications.


This study was supported by grant P01-HS11530 from the Agency for Healthcare Research and Quality. The funding agency had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript. Portions of this research were presented, in abstract form, at the 22nd International Conference on Pharmacoepidemiology & Therapeutic Risk Management, Lisbon, Portugal (August, 2006), and the Annual Meeting of the Society of General Internal Medicine, Toronto, CA (April, 2007).

We gratefully acknowledge the support of Tom Snedden and Terry Brown from the PACE program as well as the staff at FIRST Health, which provides administrative support for the PACE program. In addition, we thank the research staff at the University of Pennsylvania directed by Research Coordinator Jennifer Holmes.

Dr. Hennessy has received research funding from Pfizer, Inc and Shire. Dr. Kimmel has received research funding from multiple pharmaceutical manufacturers and has served as a consultant to Bayer, Pfizer, and GlaxoSmithKline, and has received honorarium from AstraZeneca. Dr. Feldman has received research funding from Amgen and Roche. Dr. Strom has received research funding from multiple pharmaceutical manufacturers and has served as a consultant to Bristol Myers Squibb. None of the other authors have any conflicts of interest or financial disclosures to declare and none of the authors has received funding from pharmaceutical manufacturers related to warfarin.


This study was supported by grant P01-HS11530 from the Agency for Healthcare Research and Quality. Dr. Metlay was supported by an Advanced Research Career Development Award from the Health Services Research and Development Service of the Department of Veterans Affairs. Dr. Hennessy is supported by NIH grants R01HL076697 and R01AG025152. Dr. Haynes is supported by a National Institutes of Health National Research Service Award (5-F32-AG-026180). Dr. Kimmel is supported by a National Institutes of Health Mid-Career Investigator Award in Patient Oriented Research (K24HL070936) and by NIH grant R01HL066176. Dr. Feldman is supported by a National Institutes of Health Mid-Career Investigator Award in Patient Oriented Research (K24-DK-002651). The funding agencies had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.

Portions of this research were presented, in abstract form, at the 22nd International Conference on Pharmacoepidemiology & Therapeutic Risk Management, Lisbon, Portugal (August, 2006), and the Annual Meeting of the Society of General Internal Medicine, Toronto, CA (April, 2007).


1. Gurwitz JH, Field TS, Harrold LR, et al. Incidence and preventability of adverse drug events among older persons in the ambulatory setting. JAMA. 2003;289:1107–16. [PubMed]
2. van Walraven C, Jennings A, Oake N, Fergusson D, Forster AJ. Effect of study setting on anticoagulation control: a systematic review and metaregression. Chest. 2006;129:1155–66. [PubMed]
3. Schillinger D, Wang F, Rodriguez M, Bindman A, Machtinger EL. The importance of establishing regimen concordance in preventing medication errors in anticoagulant care. J Health Commun. 2006;11:555–67. [PubMed]
4. Metlay JP, Cohen A, Polsky D, Kimmel SE, Koppel R, Hennessy S. Medication safety in older adults: home-based practice patterns. J Am Geriatr Soc. 2005;53:976–82. [PubMed]
5. Blessed G, Tomlinson B, Roth M. The association between quantitative measures of dementia and of senile change in the cerebral grey matter of elderly subjects. Br J Psychiatry. 1968;114:797–811. [PubMed]
6. Brooke P, Bullock R. Validation of a 6 item cognitive impairment test with a view to primary care usage. Int J Geriatr Psychiatry. 1999;14:936–40. [PubMed]
7. Leonard CE, Haynes K, Localio AR, et al. Diagnostic E-codes for commonly used, narrow therapeutic index medications poorly predict adverse drug events. J Clin Epidemiol. 2008;61:561–71. [PubMed]
8. Linkins LA, Choi PT, Douketis JD. Clinical impact of bleeding in patients taking oral anticoagulant therapy for venous thromboembolism: a meta-analysis. Ann Intern Med. 2003;139:893–900. [PubMed]
9. Schneeweiss S, Seeger JD, Maclure M, Wang PS, Avorn J, Glynn RJ. Performance of comorbidity scores to control for confounding in epidemiologic studies using claims data. Am J Epidemiol. 2001;154:854–64. [PubMed]
10. Diggle PJ, Heagerty PJ, Liang K-Y, Zeger SL. Analysis of longitudinal data. Second ed. New York: Oxford University Press; 2002.
11. Chiquette E, Amato MG, Bussey HI. Comparison of an anticoagulation clinic with usual medical care: anticoagulation control, patient outcomes, and health care costs. Arch Intern Med. 1998;158:1641–7. [PubMed]
12. Fihn SD, Callahan CM, Martin DC, McDonell MB, Henikoff JG, White RH. The risk for and severity of bleeding complications in elderly patients treated with warfarin. The National Consortium of Anticoagulation Clinics. Ann Intern Med. 1996;124:970–9. [PubMed]
13. Fang MC, Chang Y, Hylek EM, et al. Advanced age, anticoagulation intensity, and risk for intracranial hemorrhage among patients taking warfarin for atrial fibrillation. Ann Intern Med. 2004;141:745–52. [PubMed]
14. Landefeld CS, McGuire Ed, Rosenblatt MW. A bleeding risk index for estimating the probability of major bleeding in hospitalized patients starting anticoagulant therapy. Am J Med. 1990;89:569–78. [PubMed]
15. Schafer JL. Analysis of incomplete multivariate data. Boca Raton, Fl: Chapman & Hall; 1997.
16. 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;105:91–9. [PubMed]
17. Fihn SD, McDonell M, Martin D, et al. Risk factors for complications of chronic anticoagulation. A multicenter study. Warfarin Optimized Outpatient Follow-up Study Group. Ann Intern Med. 1993;118:511–20. [PubMed]
18. Hylek EM, Heiman H, Skates SJ, Sheehan MA, Singer DE. Acetaminophen and other risk factors for excessive warfarin anticoagulation. JAMA. 1998;279:657–62. [PubMed]
19. Budnitz DS, Layde PM. Outpatient drug safety: new steps in an old direction. Pharmacoepidemiol Drug Saf. 2007;16:160–5. [PubMed]
20. Shireman TI, Mahnken JD, Howard PA, Kresowik TF, Hou Q, Ellerbeck EF. Development of a contemporary bleeding risk model for elderly warfarin recipients. Chest. 2006;130:1390–6. [PubMed]
21. Murray E, Lo B, Pollack L, et al. The impact of health information on the internet on the physician-patient relationship: patient perceptions. Arch Intern Med. 2003;163:1727–34. [PubMed]
22. U.S. Food and Drug Administration. 2006 Safety Alerts for Drugs, Biologics, Medical Devices and Dietary Supplements. Available at: Accessed June 1, 2008.
23. Centers for Medicare and Medicaid Services. Medicare Part D Manual. Available at: Accessed June 1, 2008.

Articles from Journal of General Internal Medicine are provided here courtesy of Society of General Internal Medicine