Abstract
Available classification tools and risk factors predicting bleeding events in elderly patients after mechanical valve replacement may not be suitable in Asian populations. Thus, we aimed to identify an accurate model for predicting bleeding in elderly patients receiving warfarin after mechanical valve replacement in a Korean population. In this retrospective cohort study, a random forest model was used to determine factors predicting bleeding events among 598 participants. Twenty-two descriptors were selected as predictors for bleeding. Steroid use was the most important predictor of bleeding events, followed by labile international normalized ratio, history of stroke, history of myocardial infarction, and cancer. The random forest model was sensitive (80.77%), specific (87.67%), and accurate (85.86%), with an area under the curve of 0.87, suggesting fair prediction. In the elderly, drug interactions with steroids and overall physical condition had a significant effect on bleeding. Elderly patients taking warfarin for life require lifelong management.
Keywords: aged, medication therapy management, nursing assessment, risk factors, warfarin
1. Introduction
Degenerative diseases are becoming more prevalent in aging populations. The incidence of prosthetic valve replacement continues to increase by more than 300,000 surgeries per year worldwide.[1,2] In Korea, more than 2500 mechanical valve replacements are performed annually, accounting for 25% to 30% of adult cardiac surgeries.[3,4] Mechanical valve replacement is preferred because it is permanent and associated with better survival rates and low risk of reoperation. Mechanical heart valves are more durable than bioprostheses but are more thrombogenic. Therefore, patients with mechanical heart valves require lifelong anticoagulation with an international normalized ratio (INR) target range of 2.0 to 3.5, depending on the valve insertion position; the target INR for such patients is higher than that for patients with atrial fibrillation because of a substantially high risk of thrombosis and systemic embolism.[1,5,6] Anticoagulant therapy is associated with bleeding and thromboembolic risks. Maintaining accurate control and therapeutic levels of INR is important. Patients with mechanical valves have a higher risk of bleeding, which poses significant risks of bleeding complications ranging from 10% to 16%, with a 10% to 12% mortality rate, because they have higher target INRs than patients with atrial fibrillation.[7,8] The incidence of hemorrhagic complications among Asians is higher than that of thrombotic complications, more than twice that among African Americans and Latin Americans and more than 4 times that among Caucasians.[9,10] Moreover, prolonged prothrombin time (PT) in elderly patients taking warfarin is associated with a 2-fold increase in the incidence of bleeding complications, affecting mortality and morbidity rates and quality of life.[11,12] Therefore, it is important to identify risk factors or predictors of bleeding complications in patients taking oral anticoagulant therapy with a vitamin K antagonist after mechanical valve replacement. Continuous monitoring to meet the target range of INR is essential to prevent hemorrhagic complications.
Previous studies identified various bleeding risk factors and proposed assessment tools to determine bleeding risks.[3,13] The stability of PT in patients taking warfarin is affected by several factors including age, race, diet, comorbidity, concomitant medications, alcohol, vitamin K, and CYP2C9 or VKORC1 genes involved in metabolism.[7,13,14] In particular, food is associated with green vegetables, and drug interactions increase the risk of bleeding with acetaminophen, amiodarone, ibuprofen, cephalosporin, steroids, estrogen, and rifampin.[14,15]
Classification tools to predict bleeding events include HEMORR2HAGES (Hepatic or renal disease, Ethanol abuse, Malignancy, Older, Reduced platelet count or function, Rebleeding risk, Hypertension, Anemia, Genetic factors, Excessive fall risk, Stroke), HAS-BLED (Hypertension, Abnormal renal-liver function, Stroke, Bleeding history, Labile INR, Elderly, Drug or alcohol usage), and ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation).[16–18] Various risk factors for bleeding have been published through predictive models, but no match was found except for age. Most tools are characterized by low target PT criteria for patients with atrial fibrillation. Furthermore, these risk factors and tools were developed based on findings from the US and European populations. They may not be suitable for use in Asian populations where ongoing debates exist about the necessity of lower-dose anticoagulant therapy due to population-specific coagulation properties, lifestyle (including nutritional habits), and higher risk of thrombotic complications.[3,10]
The purpose of this study was to determine risk factors associated with the occurrence of bleeding events in patients older than 65 years with mechanical valve replacement in an Asian cohort. We aimed to develop an accurate model for the prediction of bleeding risk that is useful for elderly patients receiving warfarin therapy.
2. Methods
2.1. Design, sample, and setting
In this retrospective cohort study, we followed the STROBE Statement—checklist of items. This study was approved by an appropriate Institutional Review Board and the investigation conformed with the principles outlined in the Declaration of Helsinki.
The study targeted adult cardiac surgery patients receiving warfarin after mechanical valve replacement, with at least 3 years of outpatient follow-up at Asan Medical Center in Seoul. Data collection was performed at a hospital, which is a referral center for cardiac surgery in Seoul, Korea, serving more than 1800 patients a year.
2.2. Data collection and measures
Patients were included in this study if they were Koreans, had oral anticoagulant between 2004 and 2013 and had undergone at least one mechanical valve replacement, continuously presented for health check-up, were aged 65 years or older, and elective surgery. Of the 645 patients who underwent mechanical valve replacement, 47 were excluded because they either died of surgery-related complications or their follow-up records were incomplete.
Based on a literature review,[7,15–18] a checklist including individual, disease-related, and medication-related factors was developed for data collection. A total of 32 independent variables were selected to be tested as bleeding risk factors. The content validity of these variables was assessed based on an extensive literature review, and on the opinion of a group of experts, comprising 2 cardiac surgeons and 5 cardiovascular clinical nurse specialists. The content validity index values for the 32 independent variables were all measured at 0.80 or more.
2.2.1. Outcome variables
Bleeding events (yes/no) during the follow-up period were identified using at least 1 diagnosis code (ICD-9-CM codes and ICD-10-CM codes) from the inpatient or outpatient departments. Major bleeding is fatal bleeding, bleeding from critical areas or organs (intracranial, intrathecal, intraocular, retroperitoneal, intra-articular or pericardial, intramuscular), and causing a fall in hemoglobin level of 20 g/L (1.24 mmol/L) or more, or leading to transfusion of 2 more units of whole blood or red cells.[19]
2.2.2. Independent variables
Individual factors included age, sex, body mass index (BMI), and alcohol abuse (≥20 units/week).
Disease-related factors included operation name; history of stroke, gastrointestinal (GI) bleeding, myocardial infarction (MI), or major bleeding; coronary artery disease (CAD), hypertension, heart failure (left ventricular ejection fraction <45%), diabetes, cancer, hepatic disease, pulmonary embolism, and atrial fibrillation in underlying diseases; and anemia (hemoglobin <13.0 g/dL for men, hemoglobin <12.0 g/dL for women), platelets (<100,000/μL), creatinine (≥1.5 mg/dL), labile INR (<60% time in therapeutic range in patients), and INR on discharge.
Medication-related factors included the use of acetaminophen, amiodarone, antibiotics, antihistamines, antiplatelet agents, digoxin, herbs (ginseng, red ginseng, Chinese medicine, etc.), nonsteroidal anti-inflammatory drugs (NSAIDs), statins, and steroids.
2.3. Data analysis
Statistical analysis was performed using SPSS 23.0 (SPSS Inc, Chicago, IL), and random Forest package of R software (http://www.r-project.org). Continuous variables are expressed as mean and standard deviation, and t-test was performed to identify differences among individual-, disease-, and medication-related factors according to bleeding event. Categorical variables are expressed as frequencies and percentage, and χ2 test and Fisher exact test were used to identify differences among individual-, disease-, and medication-related factors according to bleeding event.
The random forest model was used to determine factors that are important in predicting bleeding risk. The random forest model is an ensemble method using decision tress that is tree-based classification and prediction model.[20] To confirm the rank-related factors with bleeding events in the process of construction of each individual tree, variable importance was calculated with the mean decreased accuracy method. Before model construction, the importance of variables was considered by verifying multicollinearity through correlation analysis between independent variables. After calculating the classification accuracy using out-of-bag samples from each tree in the random forest, we calculated the classification accuracy again, excluding the specific variable of interest, and calculated the difference between the 2 accuracy. The mean difference was averaged to obtain the mean reduction accuracy. This process was repeated for all variables to determine the importance of all variables used in the model. After the model was reconstructed, variable importance and out-of-bag sample were used to confirm the classification accuracy of the model.[21–23]
3. Results
3.1. Individual-, disease-, and medication-related descriptors
The study participants were 69.01 ± 3.42 years on average, and 52.5% were women. In our study, 50.4%, 32.9%, 12.9%, and 3.8% of the patients underwent aortic valve replacement, mitral valve replacement, double valves replacements (aortic valve replacement and mitral valve replacement with mechanical valves), and other procedures (including pulmonary valve replacement and tricuspid valve replacement). Mechanical valve replacement and the Maze procedure were combined in 21.2%. The length of stay was 13.26 ± 12.45 days on average. Peak INR was 4.11 ± 1.89. Smoking was reported by 3.2% of the subjects and alcohol consumption by 4.0%.
Individual-, disease-, and medication-related factors are presented Table 1. All 598 subjects were categorized into 2 groups based on bleeding events: a total of 142 patients had bleeding events and 456 did not. Significant differences were found in bleeding events in terms of individual-, disease-, and medication-related factors. Among individual-related factors, bleeding event of subjects had significantly different BMIs (t = 2.735, P = .006). BMI was lower in the bleeding group than in the non-bleeding group. Among disease-related factors, bleeding event was significantly dependent on history of stroke (χ2 = 49.823, P < .001), history of GI bleeding (χ2 = 33.695, P < .001), history of MI (χ2 = 31.196, P = .026), history of major bleeding (χ2 = 16.212, P < .001), CAD (χ2 = 18.154, P < .001), hypertension (χ2 = 10.088, P = .001), heart failure (χ2 = 8.327, P = .004), diabetes (χ2 = 8.652, P = .003), cancer (χ2 = 38.866, P < .001), hepatic disease (χ2 = 37.467, P < .001), atrial fibrillation (χ2 = 21.904, P < .001), anemia (χ2 = 57.138, P < .001), lower platelet level (χ2 = 13.636, P < .001), upper creatinine level (χ2 = 38.983, P < .001), labile INR (χ2 = 36.147, P < .001), and INR on discharge (t = 2.044, P = .041). Bleeding event of subjects was significantly different for antibiotics (χ2 = 8.695, P = .003), antiplatelet (χ2 = 28.772, P < .001), herb (χ2 = 5.147, P = .023), NSAIDs (χ2 = 5.619, P = .018), and steroids (χ2 = 21.441, P < .001) among medication-related factors. In particular, those who took antibiotics, antiplatelet, herbs, NSAIDs, and steroids were more likely to experience bleeding events (Table 1).
Table 1.
Individual-, disease-, and medication-related factors according to bleeding events (n = 598).
| Bleeding events | ||||||
| Variables | Classification | n (%) or M ± SD (n = 598) | No (n = 456) n (%) or M ± SD | Yes (n = 142) n (%) or M ± SD | χ2 or t | P |
| Individual factor | ||||||
| Age | 69.01 ± 3.42 | 68.88 ± 3.36 | 69.44 ± 3.62 | −1.687 | .092 | |
| Sex | Male | 284 (47.5) | 224 (78.9) | 60 (21.1) | 2.049 | .152 |
| Female | 314 (52.5) | 232 (73.9) | 82 (26.1) | |||
| BMI (kg/m2) | 23.70 | 23.91 ± 0.16 | 23.01 ± 0.25 | 2.735 | .006 | |
| Alcohol abuse (≥20 units/week) | No | 580 (97.0) | 444 (76.6) | 136 (23.4) | 0.942 | .397 |
| Yes | 18 (3.0) | 12 (66.7) | 6 (33.3) | |||
| Disease-related factor | ||||||
| Operation name | AVR | 301 (50.4) | 239 (52.4) | 62 (43.7) | 4.106 | .250 |
| MVR | 197 (32.9) | 141 (30.9) | 56 (39.4) | |||
| DVR (AVR and MVR) | 77 (12.9) | 59 (12.9) | 18 (12.7) | |||
| Others (PVR, TVR, etc) | 23 (3.8) | 17 (3.8) | 6 (4.2) | |||
| Maze operation | No | 471 (78.8) | 360 (78.9) | 110 (77.5) | 1.021 | .847 |
| Yes | 127 (21.2) | 96 (21.1) | 32 (22.5) | |||
| Length of stay | 13.26 ± 12.45 | 12.58 ± 11.61 | 15.49 ± 14.67 | 9.568 | .002 | |
| Previous history of stroke | No | 518 (86.6) | 420 (81.1) | 98 (18.9) | 49.823 | <.001 |
| Yes | 80 (13.4) | 36 (45.0) | 44 (55.0) | |||
| Previous history of GI bleeding | No | 557 (93.1) | 440 (79.0) | 117 (21.0) | 33.695 | <.001 |
| Yes | 41 (6.9) | 16 (39.0) | 25 (61.0) | |||
| Previous history of MI | No | 586 (98.0) | 455 (77.6) | 131 (22.4) | 31.199 | .026∗ |
| Yes | 12 (2.0) | 1 (8.3) | 11 (91.7) | |||
| Previous history of major bleeding | No | 581 (97.2) | 450 (77.5) | 131 (22.5) | 16.212 | <.001 |
| Yes | 17 (2.8) | 6 (35.3) | 11 (64.7) | |||
| CAD | No | 489 (81.8) | 390 (79.8) | 99 (20.2) | 18.154 | <.001 |
| Yes | 109 (18.2) | 66 (60.6) | 43 (39.4) | |||
| Hypertension | No | 297 (49.7) | 243 (81.8) | 54 (18.2) | 10.088 | .001 |
| Yes | 301 (50.3) | 213 (70.8) | 88 (29.2) | |||
| Heart failure (LV EF <45%) | No | 558 (93.3) | 433 (77.6) | 125 (22.4) | 8.327 | .004 |
| Yes | 40 (6.7) | 23 (57.5) | 17 (42.5) | |||
| Diabetes | No | 516 (86.3) | 404 (78.3) | 112 (21.7) | 8.652 | .003 |
| Yes | 82 (13.7) | 52 (63.4) | 30 (36.6) | |||
| Cancer | No | 532 (85.1) | 426 (80.1) | 106 (19.9) | 38.866 | <.001 |
| Yes | 89 (14.9) | 30 (45.5) | 36 (54.5) | |||
| Hepatic disease | No | 531 (88.8) | 425 (80.0) | 106 (20.0) | 37.467 | <.001 |
| Yes | 67 (11.2) | 31 (46.3) | 36 (53.7) | |||
| Pulmonary embolism | No | 597 (99.8) | 455 (76.2) | 142 (23.8) | 0.312 | 1.000∗ |
| Yes | 1 (0.2) | 1 (100.0) | 0 (0.0) | |||
| Atrial fibrillation | No | 396 (66.2) | 325 (82.1) | 71 (17.9) | 21.904 | <.001 |
| Yes | 202 (33.8) | 131 (64.9) | 71 (35.1) | |||
| Anemia (Hb: <13.0 g/dL for men, <12.0 g/dL for women) | No | 333 (55.7) | 293 (88.0) | 40 (12.0) | 57.138 | <.001 |
| Yes | 265 (44.3) | 163 (61.5) | 102 (38.5) | |||
| Lower platelet (<100,000/μL) | No | 582 (97.3) | 450 (77.3) | 132 (22.7) | 13.636 | .001 |
| Yes | 16 (2.7) | 6 (37.5) | 10 (62.5) | |||
| Upper creatinine (≥1.5 mg/dL) | No | 540 (90.3) | 431 (79.8) | 109 (20.2) | 38.983 | <.001 |
| Yes | 58 (9.7) | 25 (43.1) | 33 (56.9) | |||
| Labile INR (<TTR 60%) | No | 576 (96.3) | 451 (78.3) | 125 (21.7) | 36.141 | <.001 |
| Yes | 22 (3.7) | 5 (22.7) | 17 (77.3) | |||
| INR on discharge | 2.11 | 2.09 ± 0.02 | 2.18 ± 0.03 | 2.044 | .041 | |
| Medication-related factors | ||||||
| Acetaminophen | No | 593 (99.2) | 453 (76.4) | 140 (23.6) | 0.736 | .341∗ |
| Yes | 5 (0.8) | 3 (60.0) | 2 (40.0) | |||
| Amiodarone | No | 579 (96.8) | 442 (76.3) | 137 (23.7) | 0.072 | .786 |
| Yes | 19 (3.2) | 14 (73.7) | 5 (26.3) | |||
| Antibiotics | No | 576 (96.3) | 445 (77.3) | 131 (22.7) | 8.695 | .003 |
| Yes | 22 (3.7) | 11 (50.0) | 11 (50.0) | |||
| Antihistamine | No | 582 (97.3) | 446 (76.6) | 136 (23.4) | 1.781 | .230 |
| Yes | 16 (2.7) | 10 (62.5) | 6 (37.5) | |||
| Antiplatelet | No | 509 (85.1) | 408 (80.2) | 101 (19.8) | 28.772 | <.001 |
| Yes | 89 (14.9) | 48 (53.9) | 41 (46.1) | |||
| Digoxin | No | 488 (81.6) | 379 (77.7) | 109 (22.3) | 2.912 | .088 |
| Yes | 110 (18.4) | 77 (70.0) | 33 (30.0) | |||
| Herbs (ginseng and others) | No | 546 (91.3) | 423 (77.5) | 123 (22.5) | 5.147 | .023 |
| Yes | 52 (8.7) | 33 (63.5) | 19 (36.5) | |||
| NSAIDs | No | 547 (91.5) | 424 (77.5) | 123 (22.5) | 5.619 | .018 |
| Yes | 51 (8.5) | 32 (62.7) | 19 (22.5) | |||
| Statin | No | 430 (71.9) | 335 (77.9) | 95 (22.1) | 2.309 | 0.129 |
| Yes | 168 (28.1) | 121 (72.0) | 47 (28.0) | |||
| Steroids | No | 572 (95.7) | 446 (78.0) | 126 (22.0) | 21.441 | <.001 |
| Yes | 26 (4.3) | 10 (38.5) | 16 (61.5) | |||
AVR = aortic valve replacement, BMI = body mass index, CAD = coronary artery disease, DVR = double valve replacement, GI = gastrointestinal, Hb = hemoglobin, INR = international normalized ratio, LV EF = left ventricle ejection fraction, MI = myocardial infarction, MVR = mitral valve replacement, NSAID = nonsteroidal anti-inflammatory drug, PVR = pulmonary valve replacement, TTR = time in therapeutic range, TVR = tricuspid valve replacement.
Fisher exact test.
3.2. Characteristics of the participants of bleeding events
Among the 598 participants, 142 (23.7%) patients with major bleeding events were confirmed. The characteristics of subjects with major bleeding are presented in Table 2. Peak INR of the participants with bleeding events was 5.08 ± 2.38. The most frequent bleeding occurred during 1 to 3 years of taking warfarin after mechanical valve replacement surgery (40.8%). The location of bleeding was in order of Head, Ears, Eyes, Nose, and Throat (28.2%); GI tract (22.5%); and soft tissue (17.6%).
Table 2.
Characteristics of major bleeding events (n = 142).
| Characteristics | Categories | Frequency (%) or mean ± SD |
| Peak INR | 5.08 ± 2.38 | |
| Location of bleeding | HEENT | 40 (28.2) |
| GI | 32 (22.5) | |
| Soft tissue | 25 (17.6) | |
| GU | 22 (15.5) | |
| CNS | 20 (14.1) | |
| Unknown | 3 (2.1) | |
| Duration of warfarin therapy prior to bleed | <1 mo | 8 (5.6) |
| ≥1 mo to <1 yr | 34 (23.9) | |
| ≥1 yr to <3 yr | 58 (40.8) | |
| ≥3 yr | 42 (29.7) | |
| Outcomes | Bleed resolved warfarin continued | 113 (79.6) |
| Bleed resolved warfarin discontinued | 25 (17.6) | |
| Death | 4 (2.8) |
CNS = central nervous system, GI = gastrointestinal, GU = genitourinary, HEENT = Head, Ears, Eyes, Nose, and Throat.
3.3. Random forest prediction model for bleeding events
Twenty-two potential factors were associated with bleeding events based on t-test or χ2-test (Table 3), which were entered into the random forest model. These factors included labile INR less than time in therapeutic range 60%, stroke in previous history, steroid use, cancer, MI in previous history, hepatic disease, GI bleeding in previous history, upper creatinine of more than 1.5 mg/dL, antiplatelet, anemia, major bleeding in previous history, CAD, antibiotics, lower platelet (<1000,000/μL), diabetes, herb, NSAIDs, hypertension, BMI, atrial fibrillation, heart failure, and INR on discharge. The variable importance measures of the 22 potential factors were found to be negative for hypertension, BMI, atrial fibrillation, heart failure, and INR on discharge, and those for NSAIDs were close to 0 (Table 3). To increase the classification accuracy of the model, it was recalculated using 16 variables, excluding these factors. As shown in Figure 1, steroid use was the most important predictor of bleeding events, followed by labile INR, previous history of stroke, previous history of MI, and cancer, whereas diabetes, anemia, herb use, and antibiotic use were not significant predictors of bleeding events in elderly patients after mechanical valve replacement. Using the random forest model, these predictors had a sensitivity of 80.77%, a specificity of 87.67%, and an accuracy of 85.86%. The area under the curve for the model was 0.87, suggesting fair predictive ability.
Table 3.
Rank for factors associated with bleeding events based on MDA.
| Factors associated with bleeding events | Mean deceased accuracy (MDA) |
| Labile INR (<TTR 60%) | 0.01270 |
| Previous history of stroke | 0.01170 |
| Steroids | 0.01012 |
| Cancer | 0.01017 |
| Previous history of MI | 0.00898 |
| Hepatic disease | 0.00686 |
| Previous history of GI bleeding | 0.00643 |
| Upper creatinine (≥1.5 mg/dL) | 0.00586 |
| Antiplatelet | 0.00575 |
| Anemia (Hb: <13.0 g/dL for men, <12.0 g/dL for women) | 0.00493 |
| Previous history of major bleeding | 0.00257 |
| CAD | 0.00286 |
| Antibiotics | 0.00240 |
| Lower platelet (<100,000/μL) | 0.00241 |
| Diabetes | 0.00191 |
| Herb (ginseng and others) | 0.00100 |
| NSAIDs | 0.00044 |
| Hypertension | −0.00035 |
| Body mass index | −0.00002 |
| Atrial fibrillation | −0.00070 |
| Heart failure (LV EF<45%) | −0.00100 |
| INR on discharge | −0.00152 |
CAD = coronary artery disease, GI = gastrointestinal, Hb = hemoglobin, INR = international normalized ratio, LV EF = left ventricle ejection fraction, MDA = mean decreased accuracy, MI = myocardial infarction, NSAID = nonsteroidal anti-inflammatory drug, TTR = time in therapeutic range.
Figure 1.
Random forest prediction model for bleeding events in elderly patients after mechanical valve replacement. CAD = coronary artery disease, DM = diabetes mellitus, INR = international normalized ratio, MI = myocardial infarction.
4. Discussion
In this study, we aimed to identify predictors of bleeding events in elderly patients taking warfarin after mechanical valve replacement and found that drug interactions with steroids and overall physical condition had a significant effect on bleeding in the elderly. Warfarin, a vitamin K antagonist, interacts with numerous drugs and foods, and the list of interactions continues to expand.[24] Controlling anticoagulants is important for patients after mechanical valve replacement because maintenance of the correct target PT INR is essential to maintain valve function. Prescription of non-vitamin K antagonist oral anticoagulants has increased dramatically since their introduction in 2010. Nevertheless, warfarin is generally preferred for elderly patients, patients with higher stroke or bleeding risk, and patients after mechanical valve replacement.[25]
Previous studies on warfarin-related bleeding risk factors identified no common risk factors, except for age.[15–17] The major bleeding incidence in patients older than 65 years of age with mechanical valve replacement was found to be higher than those of previous studies including patients younger than 65 years of age with mechanical valve and patients with nonvalvular atrial fibrillation.[3,25,26] The higher therapeutic PT range in patients after mechanical valve replacement than in those with atrial fibrillation is directly related to bleeding risk. In addition, aging increases the risk of bleeding due to various chronic diseases, drug interactions with warfarin, and fixed dietary habits.
Most descriptors overlapped with those used in HEMORR2HAGES, HAS-BLED, and ATRIA. However, the use of antibiotics, antiplatelet, herbs, NSAIDs, steroids, and atrial fibrillation as descriptors is novel.[15–17] Patients after mechanical valve replacement often use painkillers after surgery and also commonly use other drugs or herbs that interact with warfarin, depending on the Korean or Asian social and cultural backgrounds. Herbal products such as Chinese medicine, garlic, ginseng, green tea, and chamomile, which interact with warfarin, are familiar to Asians and Koreans.[14,26,27] Regarding degenerative changes in the elderly population, aging is associated with the incidence of cancer, rheumatoid disease, and chronic obstructive pulmonary disease, as well as increased use of steroids.
In some studies, women were reported as an important risk factor for major bleeding. However, these findings are not relevant: risk factors for bleeding in Koreans of all ages taking warfarin after mechanical valve replacement showed that woman was an important variable; nevertheless, in the present study, patients older than 65 years of age did not show any differences.[3] Sex is a controversial risk factor for bleeding, and further research is needed. When age was not considered, sex was a very high variable, suggesting that patients older than 65 years had a greater influence on bleeding from underlying diseases, their medications, and eating habits.[27] Overall, our findings suggest that comorbid conditions and potential drug-drug interactions are important predictors of major bleeding. The results accord with those of the previous studies.
Based on our findings, predictors of bleeding events in elderly patients after mechanical valve replacement using random forest model had high sensitivity, specificity, and accuracy. The area under the curve for the model was good, indicating fair prediction. The use of the random forest model to identify both comorbid conditions and potential drug-drug interactions as major predictors of bleeding in patient taking warfarin is appropriate. We found that steroid use was the most important predictor for bleeding events, followed by labile INR, previous history of stroke, previous history of MI, and cancer, whereas diabetes, anemia, use of herbs, and use of antibiotics were not significant predictors of bleeding events in elderly patients after mechanical valve replacement. Most bleeding risk screening tools identified only personal characteristics and diseases related to characteristic such as age, sex, and clinical status and excluded dietary or drug-drug interactions.[16–18] However, these results would be important to establish precautions that should be taken while taking warfarin, particularly changes in PT due to warfarin's potential drug interactions.[24,28] In addition to the common use of warfarin in Korea, lack of awareness regarding its characteristics, function, and precautions should also be considered. In addition to the use of steroids, strict management of PT INR is important to prevent bleeding. The therapeutic PT INR range of an individual depends on the location of the mechanical valve and presence of atrial fibrillation. INR adjustment is performed by taking warfarin; however, individual differences are large. The most notable factor is the individual's strict numerical check and attention to meet the target INR. Drug administration and treatment monitoring of oral anticoagulants taken for life after mechanical valve replacement were selected as a result of decision-making with important ownership of the patient and therefore can be a strength of systematic strategy building.[29,30] At-home monitoring of PT confirmation is also very important and the formation of professional staff available for consult is required. As a result, elderly patients who need to take warfarin for a lifetime should check PT, and should pay attention to personalized influencers and nutritional education. The focus of education should vary by age, and dietary counseling and medication adherence education are important for older patients.
This study has some limitations despite its various strengths. First, because this study collected data from a single tertiary hospital, generalization of its results would not be appropriate. Therefore, conducting a multicenter replication study is necessary. Second, the drug-drug interactions for anticoagulant control were significant risk factors; however, only representative relationship with medication use was confirmed. Any mediating effects of the underlying medications were not verified; thus, examining various drug-drug interactions in the future is necessary. Third, unplanned bleeding is caused by various factors; however, we analyzed risk factors from limited information presented in the electronic medical records. Therefore, more diverse studies investigating short- and long-term risk factors for bleeding events in individuals taking warfarin as well as multicenter and follow-up studies based on Asian and worldwide populations are warranted.
5. Conclusions
In summary, we evaluated risk factors for bleeding in elderly patients who received warfarin after mechanical valve replacement. Control of anticoagulant effects in clinical situations is not easy, and complications of bleeding and embolism caused by taking warfarin have a profound effect on a patient's healthy life and quality of life. Previous studies reported that female is an important factor for bleeding. However, we found that drug interactions with steroids and overall comorbid conditions, such as cancer, heart disease, and stroke, had a significant effect on bleeding in the elderly. Elderly patients who need to take warfarin for a lifetime should check PT, and should pay attention to personalized influencers about medication and eating habits. Therefore, identifying factors influencing warfarin in the Korean sociocultural situation and considering individual medications, underlying diseases, and dietary habits are important to provide individualized education and counseling. A holistic assessment of the patient and a multidisciplinary approach should be provided. Nursing education and interventions are important to control anticoagulant effects and to improve quality of life.
Acknowledgments
The authors would like to thank the reviewers for their critical and helpful comments.
Author contributions
Conceptualization: Jisu Kim, InSil Jang.
Data curation: InSil Jang.
Formal analysis: Jisu Kim.
Funding acquisition: InSil Jang.
Writing – original draft: Jisu Kim.
Writing – review & editing: InSil Jang.
Footnotes
Abbreviations: BMI = body mass index, CAD = coronary artery disease, GI = gastrointestinal, INR = international normalized ratio, MI = myocardial infarction, NSAID = nonsteroidal anti-inflammatory drug, PT = prothrombin time.
How to cite this article: Kim J, Jang I. Predictors of bleeding event among elderly patients with mechanical valve replacement using random forest model: A retrospective study. Medicine. 2021;100:19(e25875).
This research was supported by NRF (National Research Foundation of Korea), Grant No. NRF-2019R1G1A1100166 and the Chung-Ang University Research Grants in 2020. (Recipient: Insil Jang).
The authors have no conflicts of interest to disclose.
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
References
- [1].Poli D, Antonucci E, Pengo V, et al. Mechanical prosthetic heart valves: quality of anticoagulation and thromboembolic risk. The observational multicenter PLECTRUM study. Int J Cardiol 2018;267:68–73. [DOI] [PubMed] [Google Scholar]
- [2].Sun JC, Davidson MJ, Lamy A, et al. Antithrombotic management of patients with prosthetic heart valves: current evidence and future trends. Lancet 2009;374:565–76. [DOI] [PubMed] [Google Scholar]
- [3].Jang I, Choo S, Kim K. Warfarin therapy after mechanical valve replacement: a tool for predicting bleeding events. J Heart Valve Dis 2017;26:215–33. [PubMed] [Google Scholar]
- [4].Jeon YB, Park KY. Selection and complications of prosthetic heart valves. Hanyang Med Rev 2007;27:09–17. [Google Scholar]
- [5].Roberts G, Razooqi R, Quinn S. Comparing usual care with a warfarin initiation protocol after mechanical heart valve replacement. Ann Pharmacother 2017;51:219–25. [DOI] [PubMed] [Google Scholar]
- [6].Matiasz R, Rigolin VH. 2017 Focused update for management of patients with valvular heart disease: summary of new recommendations. J Am Heart Assoc 2018;7:e007596. [Google Scholar]
- [7].Hoffmann M, Zimmermann M, Meyer R, et al. Spontaneous and non-spontaneous bleeding complications in patients with oral vitamin K antagonist therapy. Langenbecks Arch Surg 2014;399:99–107. [DOI] [PubMed] [Google Scholar]
- [8].Wysowski DK, Nourjah P, Swartz L. Bleeding complications with warfarin use: a prevalent adverse effect resulting in regulatory action. Arch Intern Med 2007;167:1414–9. [DOI] [PubMed] [Google Scholar]
- [9].Olesen JB, Lip GYH, Lindhardsen J, et al. Risks of thromboembolism and bleeding with thromboprophylaxis in patients with atrial fibrillation: a net clinical benefit analysis using a ‘real world’ nationwide cohort study. Thromb Haemost 2011;106:739–49. [DOI] [PubMed] [Google Scholar]
- [10].Shen AYJ, Yao JF, Brar SS, et al. Racial/ethnic differences in the risk of intracranial hemorrhage among patients with atrial fibrillation. J Am Coll Cardiol 2007;50:309–15. [DOI] [PubMed] [Google Scholar]
- [11].Hylelt EM, Evans-Molina C, Shea C, et al. Major hemorrhage and tolerability of warfarin in the first year of therapy among elderly patients with atrial fibrillation. Circulation 2007;115:2689–96. [DOI] [PubMed] [Google Scholar]
- [12].Beyth RJ, Quinn L, Landefeld CS. A multicomponent intervention to prevent major bleeding complications in older patients receiving warfarin: a randomized, controlled trial. Ann Intern Med 2000;133:687–95. [DOI] [PubMed] [Google Scholar]
- [13].Shoeb M, Fang MC. Assessing bleeding risk in patients taking anticoagulants. J Thromb Thrombolysis 2013;35:312–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Johnson MP. Transitions of care in patients receiving oral anticoagulants: general principles, procedures, and impact of new oral anticoagulants. J Cardiovasc Nurs 2013;28:54–65. [DOI] [PubMed] [Google Scholar]
- [15].Martín-Pérez M, Gaist D, de Abajo FJ, et al. Population impact of drug interactions with warfarin: a real-world data approach. Thromb Haemost 2018;118:461–70. [DOI] [PubMed] [Google Scholar]
- [16].Gage BF, Yan Y, Milligan PE, et al. Clinical classification schemes for predicting hemorrhage: results from the National Registry of Atrial Fibrillation (NRAF). Am Heart J 2006;151:713–9. [DOI] [PubMed] [Google Scholar]
- [17].Pisters R, Lane DA, Nieuwlaat R, et al. A novel user-friendly score (HAS-BLED) to assess 1-year risk of major bleeding in patients with atrial fibrillation: the Euro Heart Survey. Chest 2010;138:1093–100. [DOI] [PubMed] [Google Scholar]
- [18].Fang MC, Go AS, Chang Y, et al. A new risk scheme to predict warfarin-associated hemorrhage: the ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation) study. J Am Coll Cardiol 2011;58:398–401. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [19].Schulman S, Kearon C. Subcommittee on Control of Anticoagulation of the Scientific and Standardization Committee of the International Society on Thrombosis and Haemostasis. Definition of major bleeding in clinical investigations of antihemostatic medicinal products in non-surgical patients. J Thromb Haemost 2005;3:692–4. [DOI] [PubMed] [Google Scholar]
- [20].Breiman L. Random forests. Machine Learning 2001;45:05–32. [Google Scholar]
- [21].Liaw A, Wiener M. Classification and regression by random Forest. R News 2002;2:18–22. [Google Scholar]
- [22].Strobl C, Boulesteix AL, Kneib T, et al. Conditional variable importance for random forests. BMC Bioinformatics 2008;9:307. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Strobl C, Boulesteix AL, Zeileis A, et al. Bias in random forest variable importance measures: illustrations, sources and a solution. BMC Bioinformatics 2007;8:25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [24].Feng X, Sambamoorthi U, Innes K, et al. Predictors of major bleeding among working-age adults with atrial fibrillation: evaluating the effects of potential drug-drug interactions and switching from warfarin to non-vitamin K oral anticoagulants. Cardiovasc Drugs Ther 2018;32:591–600. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [25].Zhu J, Alexander GC, Nazarian S, et al. Trends and variation in oral anticoagulant choice in patients with atrial fibrillation, 2010–2017. Pharmacotherapy 2018;38:907–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].Donzé J, Rodondi N, Waeber G, et al. Scores to predict major bleeding risk during oral anticoagulation therapy: a prospective validation study. Am J Med 2012;125:1095–102. [DOI] [PubMed] [Google Scholar]
- [27].Di Minno A, Frigerio B, Spadarella G, et al. Old and new oral anticoagulants: food, herbal medicines and drug interactions. Blood Rev 2017;31:193–203. [DOI] [PubMed] [Google Scholar]
- [28].De Queiroz Oliveira JA, Ribeiro ALP, Ribeiro DD, et al. Predictive ability of scores for bleeding risk in heart disease outpatients on warfarin in Brazil. PLoS One 2018;13:e0205970. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [29].Tan CSY, Fong AYY, Jong YH, et al. INR control of patients with mechanical heart valve on long-term warfarin therapy. Global Heart 2018;13:241–4. [DOI] [PubMed] [Google Scholar]
- [30].Seaburg L, Hess EP, Coylewright M, et al. Shared decision making in atrial fibrillation: where we are and where we should be going. Circulation 2014;129:704–10. [DOI] [PubMed] [Google Scholar]