Warfarin Quality Metrics for Hospitalized Older Adults

Background  Warfarin's adverse drug events are dangerous, common, and costly. While outpatient warfarin management tools exist, there is a dearth of guidance for inpatients. Objectives  We sought to describe a health system's chronic warfarin quality metrics in older inpatients, defined by international normalized ratio (INR) control, explore associations between INR overshoots and clinical outcomes, and identify factors associated with overshoots. Patients/Methods  Data on patients 65 years and older who were prescribed chronic warfarin and admitted during January 1, 2014, to June 30, 2016, were extracted through retrospective chart review. We defined overshoots as INRs 5 or greater after 48 hours of hospitalization. Logistic regression modeling was used to determine risks for overshoots and multivariate analysis for overshoots' association with length of stay (LOS), bleeding, and mortality. Results  Of the 12,107 older inpatients on chronic warfarin, most were 75 years or older (75.7%), female (51.2%), and white (70.0%). While 1,333 (11.0%) of patients had overshoots during the admission, 449 (33.7%) of these reached overshoots after 48 hours. When stratified by overshoots versus no overshoots, LOS more than doubled (15.6 vs. 6.8 days) and the bleed rate was significantly higher (27.4 vs. 8.3%) in the overshoot group. While overall mortality was small (0.4%), the overshoot group's mortality was significantly higher (3.12 vs. 0.28%). Black race and weight were protective against overshoots; history of heart failure and antibiotic/amiodarone exposure were predictive of overshoots. Conclusion  This is the largest study examining warfarin quality metrics for hospitalized adults, specifically older inpatients. Our model may serve as the basis for identifying high-risk warfarin patients to target interventions to reduce adverse drug events.


Introduction
The Agency for Healthcare Research and Quality (AHRQ) labels anticoagulants such as warfarin a high-risk drug. Prior studies have shown that the majority of warfarin-associated bleeds result in serious outcomes, with fatal outcomes reported in up to 10%. 1 Warfarin in particular accounts for the largest number of serious adverse event reports in the Food and Drug Administration's (FDA's) Adverse Event Reporting System for the 1990 and 2000 decades, especially in older patients. 1 In a Medicare-specific population, 8.8% of adverse drug events during hospitalizations were attributed to warfarin. 2 In teaching hospitals, one-third of preventable adverse drug events were related to warfarin. 3 These warfarin-associated adverse drug events have a significant economic burden as well; a review of medical and pharmacy claims for patients with atrial fibrillation on warfarin found that annual all-cause health care costs in patients with intracranial or gastrointestinal bleeds amount to $41,903 per patient and $40,586 per patient, respectively, compared with $24,129 per patient on warfarin without bleeding. 4 Given that warfarin-associated adverse drug events are dangerous, common, and costly, the Department of Health and Human Services' National Action Plan for Adverse Drug Event Prevention has identified the safe use of anticoagulation as a national priority. 5 The Joint Commission's 2017 Hospital National Patient Safety Goals specifically recommend in Aim NPSG.03.05.01 to "Take extra care with patients who take medicines to thin their blood." 6 Achieving anticoagulant safety involves minimizing avoidable adverse drug events, reducing variability in provider care, improving system efficiency, and supporting documentation. 7 Multiple tools exist for warfarin management in the outpatient setting, [8][9][10][11] and dosing algorithms improve time in therapeutic range. 12 Such outpatient protocols include patient factors that may affect sensitivity to warfarin initiation, and subsequently dose adjusting by weekly percentages. 13,14 They are not practical for inpatient use, as they do not account for the nuances of inpatient care, such as frequent use of antibiotics or declining kidney function, and the need to adjust doses on a more frequent basis than every week. Unfortunately, there is a dearth of guidance with regard to inpatient warfarin management. 15 There are very few studies that have explored warfarin management for hospitalized patients; the focus of these studies was on warfarin initiation 16 or these studies did not account for clinical factors affecting maintenance dose. 15 Many studies have supported pharmacy-driven inpatient warfarin management as a method for reducing warfarin-associated adverse drug events. 15,17,18 While this may be effective, it can be impractical, depending on the setting and available resources for dedicating pharmacists to inpatient warfarin management.
We now seek to bridge this gap in warfarin management specifically in the inpatient setting. This study aims to (1) describe a large health system's warfarin quality metrics in older inpatients, defined by the international normalized ratio (INR) control, (2) explore the association between inpatient INR overshoots and clinical outcomes, and (3) identify intrinsic and extrinsic patient factors associated with INR overshoots. We hypothesize that poor warfarin control is common in the inpatient setting and is associated with poor clinical outcomes.

Methods
We conducted a retrospective chart review at a large health system operating in the New York metropolitan area, encompassing seven hospitals (three tertiary and four community hospitals). Data were extracted from electronic health records of patients 65 years and older who were admitted and treated with chronic warfarin between January 1, 2014, and June 30, 2016. For this study, we defined chronic warfarin as documentation of warfarin use as a home medication prior to admission (i.e., admission medication reconciliation). Our local institutional review board approved the study (IRB #16-642).
We defined INR overshoots as supratherapeutic INRs of greater than or equal to 5; it has previously been shown that the incidence of adverse events, specifically bleeding events, rises steeply with these INR values. 19 To identify the quality of inpatient warfarin as a result of dosing during the acute hospitalization (rather than doses taken prior to admission), we limited our analysis to INRs after the initial 48 hours of the hospital stay. , hemiplegia/paraplegia, malignancies, HIV, and liver disease). Additional variables included medications administered during hospitalization (i.e., antibiotics, amiodarone, and statins) and organizational factors (i.e., tertiary vs. community hospital).
Outcomes included hospital LOS, mortality, and clinically significant bleeding. To capture clinically relevant bleeding, patients needed to meet at least two of the following three criteria: (1) an ICD9 code for bleeding (as a hospital diagnosis), (2) RBCs transfused during admission, and/or (3) receipt of a reversal agent during the admission (including any vitamin K, fresh frozen plasma, or prothrombin complex concentrates). The ICD9 bleeding codes used for analysis were derived from members of the New York State Anticoagulation Coalition and from Leonard et al (2008) and are listed in ►Appendix A. 20  analysis was employed to associate INR overshoots with LOS, bleeding, and mortality. Variability across the health system was evaluated with INR overshoots by type of hospital, tertiary care versus community facility. Additional analysis of the impact of patient weight (kg) on INR overshoots was done through chi square testing at 10-kg intervals to establish safety thresholds.

Results
There were 17,494 unique admissions across seven acute care facilities for patients 65 years and over on warfarin. Of these, 12,107 were on chronic warfarin with INR data available and 5,387 were initiated on warfarin during the hospitalization (not included in our target population). Of those on chronic warfarin, 1,020 (8.4%) discontinued warfarin on discharge from the hospital. Patients with INR overshoots were more likely to be discharged without warfarin than those without overshoots (24.3 vs. 7.8%, p < 0.0001). ►Table 1 describes the characteristics of patients on chronic warfarin during their acute hospitalization. The majority of patients (75.7%) were older than 75 years, female (51.2%), and white (70%). One-third had a history of smoking, with the most common comorbid conditions being CHF (46%), DM without chronic complications (31.4%), CVD (24%), COPD ( In evaluating variability in the quality of warfarin management across the seven facilities, the rates of INRs over 5 after the first 48 hours ranged from 3.0 to 5.9%. While there was no significant difference between hospital types (community vs, tertiary) with regard to INR overshoots, admission to a tertiary hospital was found to be protective against bleeding (OR: 0.862, p < 0.0016).
►Table 4 presents demographic, clinical, and organizational variables used in the prediction model for INR overshoots after the initial 48 hours of hospitalization. Using logistic regression, black race and weight were found to be protective against INR overshoots; conversely, history of CHF and antibiotic or amiodarone exposure was predictive of INR overshoots. Moderate or severe CKD trended toward predicting INR overshoots, but did not reach statistical significance (p < 0.068). When adding the variable for INR overshoots to the logistic regression model, we found that (in addition to controlling for age, gender, race, smoking status, ICU stay, heart failure, COPD, DM, CKD, malignancy, liver disease, weight, and antibiotic and amiodarone exposure) an INR greater than or equal to 5 was independently predictive of a longer LOS (p < 0.0001), higher bleed rate (p < 0.0001), and higher mortality (p < 0.0001).
Chi-square testing at 10-kg intervals for body weight found significant thresholds at both 50 and 90 kg: 7.7% of patients less than or equal to 50 kg experienced INR overshoots compared with 3.5% of those over 50 kg (p < 0.0001); 4.1% of patients less than or equal to 90 kg experienced INR overshoots compared with 2.5% of patients over 90 kg (p < 0.0001). At weights over 120 kg, there is no significant difference in incidence of INR overshoots between weight classes. ►Table 5 illustrates an overall downward trend of INR overshoots as weight classes increase, with 7.69% of those 50 kg and under experiencing overshoots, 3.78% of those between 50 and 90 kg, 2.35% between 90 and 120 kg, and 2.98% of those over 120 kg.

Discussion
This is the largest study to date examining warfarin quality metrics for older adults in the inpatient setting. More specifically, we sought to describe chronic warfarin quality metrics across multiple inpatient facilities, explore the association between inpatient INR overshoots and clinical outcomes, and identify intrinsic patient-related factors and extrinsic factors associated with INR overshoots. Warfarin has previously been described as a high-risk medication, and poor control defined by INR has been associated with negative outcomes; our research is novel in that it specifically investigates the clinical outcomes of INR overshoots in a large hospital-based health system and focuses on risk factors for such overshoots.
Approximately half of the patients on chronic warfarin reached supratherapeutic levels of INR greater than 3 during the admission, and approximately one-half of these occurred after the initial 48 hours of hospitalization. This time frame was used to focus on effects of inpatient, rather than outpatient, events including provider dosing and clinical conditions. Roughly 11% of patients reached supratherapeutic INRs at the clinically important threshold of INR ! 5.0, and one-third of these reached this level after the initial 48 hours, with inpatient dosing again likely responsible for these INR overshoots. While the overall bleed rate of 9% is comparable  to rates reported in Medicare patients on anticoagulation, 21 the group with INR overshoots had significantly increased LOS, bleeding, and mortality. Given INR overshoots' association with these adverse events, they may serve as potential surrogate markers for identifying such negative outcomes that health systems seek to avoid. By hospital policy, INRs are checked daily for patients presently treated with warfarin. Despite close monitoring with the rate of INR days being 96% in both the overshoot and no overshoot group, there was still poor control in the overshoot group suggesting that frequency of INR checking did not contribute to differences between adverse outcomes between the groups. Future studies should evaluate interventions to improve the safety of inpatient warfarin dosing and may use INR overshoots in addition to clinical outcomes to evaluate the effectiveness of such interventions. We found that for older patients on chronic warfarin therapy during acute hospitalization, low weight, exposure to antibiotics/amiodarone, and heart failure were independently predictive of INR overshoots, while black race was protective. In additional weight analysis, the largest thresholds for correlation with INR overshoots were found at 50 and 90 kg. At weights above 120 kg, there were no longer associations with INR overshoots. While most of these factors have been identified as markers, or sensitivity classes for outpatient initiation algorithms, this is the first study to evaluate them in an older inpatient population. As examples, the initiation algorithms presented by the University of North Carolina 9 and the University of Wisconsin 10 both include heart failure, low body weight, antibiotics, and amiodarone as intrinsic and extrinsic factors making patients prone to higher warfarin sensitivity. In Kimmel's   pharmacogenetics studies, the clinical dose-revision algorithm that was used as a comparison to pharmacogenetics dosing, black race and body surface area were included as factors increasing the recommended warfarin dose, and amiodarone use as a factor lowering the dose. 22 There were several limitations to our study. This was a retrospective chart review, and predictors of INR overshoots were limited by documentation available from the electronic health record with incomplete information, and difficulty establishing cause and effect. Comorbid conditions were identified through an electronic data pull of ICD9 codes alone, and thus we were unable to differentiate between acute, chronic, and past conditions or to establish temporal associations between such comorbid conditions and INR overshoots. Bleeding events during hospitalization could not be time correlated with INR overshoots, again highlighting the lack of ability to establish cause and effect. We do hope to have improved accuracy of bleeding events by requiring a minimum of two bleed-related orders or coding but were unable to verify this through individual chart analysis. Furthermore, this definition of bleeding may have impacted the finding that admission to a tertiary hospital was associated with a lower risk of bleeding due to differences in transfusion management such as evidence-based, higher thresholds for transfusions. Additionally, while ICU admission was more prevalent in the INR overshoot group, we were unable to assess whether such admissions were due to a concurrent illness causing the overshoot, severe bleeding caused by the overshoot, or poor dosing within the ICU. We also did not assess for scenarios when warfarin was intentionally held (i.e., for procedures) or the need for reversal due to bleeding. To overcome this barrier, our focus for poor control was on supratherapeutic INRs known to be high risk for acute bleeding episodes rather than on subtherapeutic INRs. On the inpatient setting, the risk of subtherapeutic INRs can be mitigated by using heparin-bridging therapies when appropriate. We did not separate surgical and medical patients who may have different risk factors for adverse events. A final limitation was that the use of antibiotics or amiodarone was not correlated in its timing with the administration of warfarin.

Summary and Conclusion
Our study findings indicated that (1) INR overshoots are prevalent in the inpatient setting in an older population across our health system and associated with poor outcomes and (2) INR overshoots are independently associated with low weight, heart failure, non-black race, and antibiotic or amiodarone exposure. In the outpatient setting, both intrinsic patientrelated and extrinsic factors are integrated into warfarin dosing algorithms. We found that similar factors are also associated with inpatient INR metrics. Yet, no such tool exists in the inpatient setting to determine appropriate warfarin dosing, especially for more frequent (than weekly) assessments. Our model may serve as the basis for identifying high-risk patients and developing interventions for inpatient warfarin dosing strategies. Future studies should focus on the impact of the rate of change of the INR (the delta INR) on predicting INR overshoots when combined with clinical factors identified by our prediction model. 17 This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.