Abstract
Background
Schemas to identify bleeding‐related hospitalizations in claims data differ in billing codes used and coding positions allowed. We assessed agreement across bleeding‐related hospitalization coding schemas for claims analyses of nonvalvular atrial fibrillation (NVAF) patients on oral anticoagulation (OAC).
Hypothesis
We hypothesized that prior coding schemas used to identify bleeding‐related hospitalizations in claim database studies would provide varying levels of agreement in incidence rates.
Methods
Within MarketScan data, we identified adults, newly started on OAC for NVAF from January 2012 to June 2015. Billing code schemas developed by Cunningham et al., the US Food and Drug Administration (FDA) Mini‐Sentinel program, and Yao et al. were used to identify bleeding‐related hospitalizations as a surrogate for major bleeding. Bleeds were subcategorized as intracranial hemorrhage (ICH), gastrointestinal (GI), or other. Schema agreement was assessed by comparing incidence, rates of events/100 person‐years (PYs), and Cohen's kappa statistic.
Results
We identified 151 738 new‐users of OAC with NVAF (CHA2DS2‐VASc score = 3, [interquartile range = 2–4] and median HAS‐BLED score = 3 [interquartile range = 2–3]). The Cunningham, FDA Mini‐Sentinel, and Yao schemas identified any bleeding‐related hospitalizations in 1.87% (95% confidence interval [CI]: 1.81‐1.94), 2.65% (95% CI: 2.57‐2.74), and 4.66% (95% CI: 4.55‐4.76) of patients (corresponding rates = 3.45, 4.90, and 8.65 events/100 PYs). Kappa agreement across schemas was weak‐to‐moderate (κ = 0.47–0.66) for any bleeding hospitalization. Near‐perfect agreement (κ = 0.99) was observed with the FDA Mini‐Sentinel and Yao schemas for ICH‐related hospitalizations, but agreement was weak when comparing Cunningham to FDA Mini‐Sentinel or Yao (κ = 0.52–0.53). FDA Mini‐Sentinel and Yao agreement was moderate (κ = 0.62) for GI bleeding, but agreement was weak when comparing Cunningham to FDA Mini‐Sentinel or Yao (κ = 0.44–0.56). For other bleeds, agreement across schemas was minimal (κ = 0.14–0.38).
Conclusions
We observed varying levels of agreement among 3 bleeding‐related hospitalizations schemas in NVAF patients.
Keywords: Agreement, Anticoagulants, Atrial Fibrillation, Clinical Coding, Hemorrhage
1. INTRODUCTION
Nonvalvular atrial fibrillation (NVAF) is a common cardiac arrhythmia, with 1 in 4 middle‐aged adults in the United States and Europe likely to be diagnosed. NVAF substantially increases patients' risk of stroke by ~5‐fold and mortality by ~2‐fold.1, 2 In patients with a CHA2DS2‐VASc score ≥ 2, oral anticoagulation (OAC) with either a vitamin K antagonist (VKA) or non‐VKA oral anticoagulant (NOAC) is essential for the prevention of ischemic stroke.1, 2 OAC therapy can prevent most ischemic strokes in NVAF patients, and as a result reduce subsequent disability and prolong patients' life spans.
Often prescribers question the real‐world validity of clinical trials, and therefore use real‐world evidence (including claims database studies) to help weigh the benefits (stroke reduction) and risks (bleeding) of OAC use in routine practice settings.3, 4 There have been a substantial number of published claims database studies of OAC users with NVAF that have evaluated major bleeding risk,5, 6, 7, 8, 9 with significant heterogeneity (I 2 = 93.4%)5 in reported rates observed across studies. This variation in major bleeding rates across claims database studies may be due to differences in data sources/populations evaluated, but also the methodologies employed for identifying bleeding events. Retrospective administrative claims analyses rely on schemas (or algorithms)9, 10, 11 that utilize site of service and billing codes to identify whether a bleeding‐related hospitalization (a surrogate for clinically diagnosed major bleed) occurred. Studies of OAC in NVAF patients have used different bleeding‐related hospitalization billing code schemas that vary in regard to the specific codes used and the position the code could appear in (as most healthcare encounters are associated with >1 billing code).
Prior studies have reported prescribers' perception of high‐risk of bleeding on OAC as a frequent reason for withholding or discontinuing patients' OAC therapy.12 It is therefore imperative we gain a better understanding of the impact of using different billing code schemas for detecting bleeding‐related hospitalizations in claims database studies. Here, we sought to assess agreement by comparing incidence, rates of events/100 person‐years (PYs), and Cohen's kappa (κ) across bleeding‐related hospitalization billing code schemas previously used in claims analyses of NVAF patients on OAC.
2. METHODS
We performed a retrospective administrative claims database study using US Truven MarketScan data from January 2012 to June 2015. MarketScan combines 2 separate databases, a commercial and Medicare supplemental database, to cover all age groups, and contains claims from 100 employers, health plans, and government and public organizations representing about 170 million covered lives in the United States, making it representative of the US population as a whole.13 MarketScan captures healthplan enrollment records, participant demographics, International Classification of Diseases, 9th Revision, Clinical Modification (ICD‐9‐CM) diagnosis codes (up to 15 codes for inpatient hospital admissions), Current Procedural Terminology, 4th Revision, and other procedure code types, hospital revenue codes, admission and discharge dates, inpatient mortality data, outpatient medical services data, and prescription dispensing records. All data included in the MarketScan databases are deidentified and are in compliance with the Health Insurance Portability and Accountability Act of 1996 to preserve participant anonymity and confidentiality. This study was deemed exempt from institutional review board oversight.
To be included in the present study, adult patients had to have ≥180 days of continuous medical and prescription coverage prior to initiation of OAC (the baseline period); be OAC naïve during the 180 days prior to the day of the first qualifying OAC dispensing (index date); be newly initiated on apixaban, dabigatran, edoxaban, rivaroxaban, or warfarin; and have ≥2 inpatient or outpatient claims for atrial fibrillation (ICD‐9‐CM code 427.31) without codes suggesting valvular heart disease.
Our primary endpoint was bleeding‐related hospitalizations (a surrogate for major bleeding) as determined by applying 3 different claims database billing code schemas to an identical NVAF cohort as described above.9, 10, 11 Coding schemas to identify bleeding‐related hospitalizations included the automated database case definition for serious bleeding related to oral anticoagulants developed by Cunningham and colleagues,10 suggested coding per the US Food and Drug Administration (FDA) Mini‐Sentinel initiative protocol for assessment of dabigatran (version 3, updated March 27, 2015),11 and the coding schema utilized by Yao and colleagues in recent evaluations of OAC in NVAF patients.9, 14 These billing code schemas all required the bleeding event to be associated with a hospitalization as the site of service, but varied in both the billing codes used to identify bleeding events as well as the coding position (primary or nonprimary) allowed (see Supporting Tables 1–4 in the online version of this article). The primary discharge diagnosis code for a hospitalization is generally considered to reflect the main reason for admission.15
Table 1.
Characteristics of included patients (N = 151 738)
| Characteristic | Value |
|---|---|
| Age, y, mean ± SD | 69.0 ± 12.7 |
| <65 years | 59 730 (39.4) |
| ≥65 years to ≤75 years | 39 469 (26.0) |
| >75 years | 52 539 (34.6) |
| Female | 61 664 (40.6) |
| Alcohol abuse | 2651 (1.7) |
| Anemia | 26 971 (17.8) |
| Diabetes | 43 485 (28.7) |
| Renal failure | 19 328 (12.7) |
| History of gastric or peptic ulcer disease | 152 (0.1) |
| History of major bleeding | 7695 (5.1) |
| Hypertension | 107 868 (71.1) |
| Heart failure | 39 575 (26.1) |
| Liver disease | 10 932 (7.2) |
| Obesity | 18 771 (12.4) |
| Pulmonary disease | 27 734 (18.3) |
| Stroke | 24 564 (16.2) |
| Tobacco abuse | 12 612 (8.3) |
| Vascular disease | 24 122 (15.9) |
| CHADS2 score, median (IQR) | 2 (1–3) |
| 0 | 22 761 (15.0) |
| 1 | 45 052 (29.7) |
| 2 | 41 716 (27.5) |
| 3 | 21 491 (14.2) |
| 4 | 12 516 (8.2) |
| 5 | 5654 (3.7) |
| 6 | 1264 (0.8) |
| CHA2DS2VASC score, median (IQR) | 3 (2–4) |
| 0 | 10 673 (7.0) |
| 1 | 22 138 (14.6) |
| 2 | 27 785 (18.3) |
| 3 | 29 183 (19.2) |
| 4 | 25 630 (16.9) |
| 5 | 17 029 (11.2) |
| 6‐9 | 18 016 (11.9) |
| Modified HAS‐BLED score, median (IQR) | 3 (2–3) |
| 1 | 19 083 (12.6) |
| 2 | 47 356 (31.2) |
| 3 | 48 412 (31.9) |
| 4 | 16 550 (10.9) |
| 5 | 13 969 (9.2) |
| 6–8 | 5084 (3.4) |
| Oral anticoagulant utilized | |
| Apixaban | 19 974 (13.2) |
| Dabigatran | 16 718 (11.0) |
| Edoxaban | 40 (0.03) |
| Rivaroxaban | 46 678 (30.8) |
| Warfarin | 68 319 (45.0) |
| Year of oral anticoagulant initiation | |
| 2012 | 39 983 (26.3) |
| 2013 | 51 153 (33.7) |
| 2014 | 46 073 (30.4) |
| 2015 | 14 589 (9.6) |
Abbreviations: IQR, interquartile range; SD, standard deviation.
Data are presented as no. (%) unless otherwise indicated.
Baseline patient characteristics of the NVAF cohort were reported using descriptive statistics. Schema agreement was assessed by comparing incidence (%), rates of events/100 PYs, and Cohen's κ statistic with accompanying P values. The incidence (proportion) of experiencing any type of bleeding‐related hospitalization during follow‐up along with Clopper‐Pearson 95% confidence interval (CI) and rate/100 PYs, along with Poisson 95% CI were calculated according to each of the 3 billing code schemas. Eligible bleeding events were subcategorized as intracranial hemorrhage (ICH), gastrointestinal (GI) bleeding, or other bleeding. Patients were followed until the occurrence of a bleeding‐related hospitalization, discontinuation/switch of index OAC therapy, disenrollment from insurance plan or end‐of‐study follow‐up. OAC discontinuation was defined as a gap ≥14 days between the most recent OAC fill date and the date when there were no days of OAC supply anticipated to be remaining. Cohen's κ statistic was used to further quantify the magnitude of agreement across schema by estimating the difference between the amount of agreement present and the amount expected due to chance alone (not to compare the accuracy of schemas) and was interpreted as follows: 0 to 0.20 = no agreement, 0.21 to 0.39 = minimal agreement, 0.40 to 0.59 = weak agreement, 0.60–0.79 = moderate agreement, 0.80 to 0.90 = strong agreement, and >0.90 = near perfect agreement.16 We made an a priori decision not to compare bleeding across specific OAC drugs, as such analyses did not address our hypothesis of interest. A P < 0.05 was considered significant in all instances. SAS version 9.3 (SAS Institute Inc., Cary, NC) and IBM SPSS version 22.0 (IBM Corp., Armonk, NY) were used to perform all database management and statistical analysis.
3. RESULTS
In total, 151 738 new users of OAC with NVAF were identified. Baseline characteristics of the study cohort are depicted in Table 1. The mean ± standard deviation (SD) age of the cohort was 69.0 ± 12.7 years; median CHA2DS2‐VASc score was 3 (interquartile range [IQR] = 2–4) and HAS‐BLED score was 3 (IQR = 2–3). Warfarin was the most frequently prescribed OAC (45.0%). Patients were followed for a mean ± SD of 198 ± 210 days.
Across the 3 billing code schemas, any bleeding‐related hospitalization was identified in 1.87% to 4.66% of patients (Figure 1); ICH‐related hospitalizations in 0.32% to 0.89% of patients, GI bleeding‐related hospitalizations in 1.18% to 3.37% of patients, and other bleeding‐related hospitalizations in 0.25% to 0.70% of patients (Table 2). The corresponding rates of these bleeding events were 3.45 to 8.65, 0.59 to 1.64, 2.17 to 6.25, and 0.47 to 1.29/100 PYs, respectively. The percentage of patients identified with any bleeding‐related hospitalization was 2.48‐ and 1.75‐fold higher when using the schema developed by Yao and colleagues compared to either the Cunningham and colleagues or the FDA Mini‐Sentinel schemas, respectively. Similar increases in the relative proportion of patients experiencing an ICH (1.02‐to 2.78‐fold), GI bleed (2.02‐ to 2.86‐fold), and other bleeding‐related hospitalization (1.66‐ to 2.76‐fold) was observed when the Yao and colleagues schema was implemented compared to the others.
Figure 1.
Proportions (A) and rates (B) of total bleeding‐related hospitalizations. Abbreviations: CI, confidence interval; CUN, Cunningham; FDA, Food and Drug Administration; PYs, person‐years
Table 2.
Bleeding‐related hospitalizations in 151 738 nonvalvular atrial fibrillation patients according to different coding schema
| Cunningham10 | FDA | Yao9, 14 | −Fold Change With Yao | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| No. of Events | Proportion, % (95% CI) | Rate, Events/100 PYs (95% CI) | No. of Events | Proportion, % (95% CI) | Rate, Events/100 PYs (95% CI) | No. of Events | Proportion, % (95% CI) | Rate, Events/100 PYs (95% CI) | vs Cunningham | vs FDA | |
| ICH | 488 | 0.32 (0.29‐0.35) | 0.59 (0.54‐0.65) | 1335 | 0.88 (0.83‐0.93) | 1.62 (1.53‐1.71) | 1358 | 0.89 (0.85‐0.94) | 1.64 (1.56‐1.73) | 2.78 | 1.02 |
| GI bleeds | 1791 | 1.18 (1.13‐1.24) | 2.17 (2.07‐2.27) | 2532 | 1.67 (1.60‐1.73) | 3.08 (2.96‐3.20) | 5120 | 3.37 (3.28‐3.47) | 6.25 (6.08‐6.42) | 2.86 | 2.02 |
| Other bleedsa | 642 | 0.42 (0.39‐0.46) | 0.78 (0.72‐0.84) | 385 | 0.25 (0.23‐0.28) | 0.47 (0.42‐0.52) | 1063 | 0.70 (0.66‐0.74) | 1.29 (1.21‐1.37) | 1.66 | 2.76 |
Abbreviations: CI, confidence interval; FDA, Food and Drug Administration; GI, gastrointestinal; ICH, intracranial hemorrhage; PYs, person‐years.
The major extracranial non‐GI bleeding events as defined in the FDA Mini‐Sentinel schema.
The sum of the genitourinary and other bleeding‐related hospitalizations as defined by Cunningham.
Agreement across all schemas was weak to moderate (κ = 0.47 – 0.66, P < 0.001 for all) for total bleeding‐related hospitalizations (Figure 2). Near perfect agreement (κ = 0.99) was observed with the FDA Mini‐Sentinel and Yao schemas for ICH‐related hospitalizations, but agreement was weak when comparing Cunningham to the Yao or FDA Mini‐Sentinel schemas (κ = 0.52–0.53, P < 0.001 for all). FDA Mini‐Sentinel and Yao schema agreement was moderate (κ = 0.62) for GI bleeding, but weak upon comparing Cunningham to the Yao or FDA Mini‐Sentinel schemas (κ = 0.44–0.56, P < 0.001 for all). For other bleeds, agreement across all schemas was minimal at best (κ = 0.14–0.38, P < 0.001 for all).
Figure 2.
Kappa values for agreement among bleeding‐related hospitalization coding schemas. P values for all κ statistics were statistically significant (P < 0.001). Abbreviations: CUN, Cunningham; FDA, Food and Drug Administration; GI, gastrointestinal; ICH, intracranial hemorrhage. κ statistic of 0–0.20 = no agreement, 0.21–0.39 = minimal agreement, 0.40–0.59 = weak agreement, 0.60–0.79 = moderate agreement, and 0.80–0.90 = strong agreement
4. DISCUSSION
We found substantial differences in the incidence, rate, and level of agreement for overall and most subtypes of bleeding‐related hospitalizations (often referred to as major bleeding in schema source studies) when implementing the 3 different claims‐based billing code schemas in an identical NVAF cohort. The differences in the incidences and rates of identified bleeding events across schemas are likely explained by differences in the specific billing codes used by each schema and the positions the corresponding codes were allowed to be identified in. Because the schema by Cunningham and colleagues requires (at minimum) a primary discharge diagnosis code for a bleeding‐related hospitalization event, additional confirmatory evidence for some bleeding codes (including transfusion or processing of blood products for transfusion), and excludes trauma‐related bleeds, it is not surprising it identified the lowest incidences and rates of bleeding events across most bleeding subtypes. The one exception was other bleeds, where Cunningham includes a more comprehensive set of ICD‐9‐CM codes compared to the other schemas studied. Yao and colleagues' schema was found to more frequently identify patients as having a bleeding‐related hospitalization, likely because it allows eligible codes to be located in primary or nonprimary position (up to 15 coding positions in MarketScan hospitalization data). The increase in bleeding‐related hospitalizations with Yao was most pronounced for GI bleeds, with this schema identifying ~2 to 3 times more bleeding events than either the Cunningham or FDA Mini‐Sentinel schemas (both restrict GI bleeding codes to the primary position for event identification). The FDA Mini‐Sentinel schema's allowance of primary and nonprimary position for ICH codes explains its near‐perfect agreement with Yao (coding between these 2 schemas differs only in Yao's exclusion of patients with a primary V57 code for rehabilitation and a few additional trauma codes), but weak agreement with Cunningham. Whereas other site bleeding‐related hospitalizations were relatively infrequent in this study and contributed to only a small fraction of the total bleeding‐related hospitalization events identified, there was substantial disagreement across all 3 billing code schema for this subtype of bleeds.
Of the 3 schemas evaluated in our study, only Cunningham's has undergone prior validation through medical record review.10 In their validation study performed within Tennessee Medicaid enrollees receiving warfarin, Cunningham and colleagues reviewed the medical records of 186 (charts available) of 236 identified bleeding‐related hospitalizations per their algorithm. Of these 186, 89% (95% CI: 83%‐92%) were adjudicated as true bleeding‐related hospitalizations based upon medical record review. Nearly 72% (95% CI: 64.6%‐77.5%) of these bleeding events were classified as definite, whereas 17.2% (95% CI: 12.5%‐23.3%) were classified as probable. The Cunningham algorithm demonstrated positive predictive values of at least 86% across all bleeding subtypes (ICH = 94%, GI = 86%, genitourinary = 89%, and other = 96%).
An additional strength of the Cunningham algorithm was that it had previously shown estimates of bleeding‐related hospitalization rates in OAC patients that were relatively consistent with clinically diagnosed major bleeding rates from prospective registry studies that have implemented centralized adjudication of events. In a preplanned pooled analysis of the international, prospective, observational XANTUS (Xarelto for Prevention of Stroke in Patients With Atrial Fibrillation), XANAP (Xarelto for Prevention of Stroke in Patients With Atrial Fibrillation in Asia‐Pacific), and XANTUS‐EL (Xarelto for Prevention of Stroke in Patients With Nonvalvular Atrial Fibrillation, Eastern Europe, Middle East, Africa, and Latin America) studies examining the rate of major bleeds (defined as clinically overt bleeding at a critical site or associated with a ≥ 2 g/dL hemoglobin drop, transfusion or death) in 11 121 rivaroxaban patients with NVAF, major bleeding occurred at a rate of 1.7 events/100 PYs (95% CI: 1.5‐2.0) among a population with a mean CHADS2 score of 2.0 ± 1.3.17 In a population with a similar degree of comorbidity (CHADS2 = 2.2), Tamayo and colleagues identified a rate of bleeding‐related hospitalizations (called major bleeding in their publication) of 2.86 events/100 PYs (95% CI: 2.61‐3.13) after applying the Cunningham algorithm to claims data for 27 467 NVAF patients receiving rivaroxaban captured in the US Department of Defense database.7 Although the bleeding‐related hospitalization rate observed with the Cunningham algorithm in our study (3.45 events/100 PYs, median CHADS2 = 2) was somewhat higher than seen in the pooled XANTUS or the study by Tamayo and colleagues, this increase might at least partially be explained by the inclusion of patients receiving warfarin (45.0%) in our analysis instead of NOAC drugs only. NOAC drugs as a class have been shown in a meta‐analysis of phase III clinical trials to reduce patients' risk of major bleeding (relative risk = 0.86, 95% CI: 0.73‐1.00) and ICH (relative risk = 0.48 (95% CI: 0.39‐0.59) vs warfarin.18 The shorter duration of patient follow‐up in our study (mean = 6.6 months) may also explain the higher rate of bleeding‐related hospitalizations, given that multinational GARFIELD‐AF (Global Anticoagulant Registry in the FIELD‐Atrial Fibrillation) data suggest patients' risk of a major bleed is highest (56% higher vs the overall rate) in the first 4 months after treatment initiation.19 Whereas this may have impacted our incidence and event rate calculations, the κ statistics (which are intended to show consistency between coding algorithms, not relative accuracy) should not have been affected, because we applied all 3 schema to the same exact patient cohort and duration of follow‐up.
If a coding schema used in a claims database comparison of ≥2 OAC drugs overestimates bleeding‐related hospitalization rates, it is possible the overestimation would have a smaller impact on the hazard ratio compared to the absolute risk difference (Δ, the inverse of which is the number needed to harm [NNH]). For example, assuming two sets of event rates for major bleeding of 2.5/100 PYs (OAC A1) vs 2.0/100 PYs (OAC B1) or 6.0/100 PYs (OAC A2) vs 4.8/100 PYs (OAC B2), the resulting hazard ratios would be identical (at 1.25), but absolute differences and interpretation of the clinical relevance would vary substantially (Δ1 = 0.5%/100 PYs, NNH = 200; and Δ2 = 1.2%/100 PYs, NNH = 84). Most claims analyses of OAC in NVAF patients report both hazard ratios and absolute risk rates,7, 20, 21, 22 with bleeding event estimates of the latter ranging from 2.5 to 7.5 events/100 PYs. Given the potential impact that overestimation (or underestimation) of bleeding risk may have on the interpretation of claims database study results, it is important that steps are taken whenever possible to assure the most accurate estimation of event rates to allow for optimal risk–benefit assessment and to not scare clinicians away from proper OAC use (either initiation, continued use, or appropriate dosing) due to undue concern over bleeding. In clinical practice, the prescription and dosing of OAC is frequently inconsistent with guideline recommendations.12, 23 Analysis of GARFIELD‐AF found that 36.9% of patients with a CHA2DS2‐VASc ≥2 were not treated with OAC.19 Furthermore, a higher than anticipated use of reduced‐dose NOAC therapy has been observed in routine clinical practice,23 with experts suggesting this prescribing pattern is being at least in part driven by concerns about bleeding.24
Our study has several limitations. The deidentified nature of the claims data made it impossible for us to adjudicate bleeding events (in even a subset of patients) according to a standard, accepted clinical definition. Rather, claims database analyses have used billing code schemas that identify bleeding‐related hospitalizations as a surrogate for major bleeding. Due to our inability to adjudicate events, we cannot concretely say which algorithm was most sensitive and specific. However, the greater consistency in bleeding rates seen between the validated Cunningham algorithm in the present analysis and prior claims database analyses,7 in comparison to prospective studies that adjudicated events,17, 19 indirectly suggests it may be the most accurate, and that other algorithms overestimate bleeding rates. Due to common limitations of claims data, including missing data, misclassification bias, and the potential for upcoding,15 it is unlikely a perfect (100% sensitive and specific) claims‐based coding schema for detecting major bleeding can be created. Moreover, this would require a consensus on which billing codes are most representative of accepted clinical definitions of major bleeding. This being said, the previously mentioned limitations of claims data are unlikely to explain our study's findings, as all 3 schema were applied to the same exact NVAF patient cohort.
5. CONCLUSION
Centralized adjudication is the gold‐standard method for accurately verifying major bleeding events. However, this is not possible in retrospective analyses of administrative claims databases that rely instead on billing code schemas to identify bleeding‐related hospitalizations and to evaluate severity of bleeds. Although our study could not determine the most accurate coding schema for identifying bleeding‐related hospitalizations in NVAF patients, we were able to show that there is substantial variation in levels of agreement among the 3 frequently used billing code schemas. Clinicians should carefully consider the schema used and its impact on rates of bleeding‐related hospitalizations when interpreting claims analyses of OAC in NVAF patients.
Author contributions
All authors contributed to the design, analysis, interpretation of data, drafting the article, or revising it critically for important intellectual content and approved the final version to be published. Craig I. Coleman is the senior and corresponding author and guarantor, and affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.
This study was supported by Bayer AG, Berlin, Germany.
Conflicts of interest
Craig I. Coleman has received research grants from Bayer AG and Janssen Scientific Affairs LLC. He has served as a consultant for Boehringer‐Ingelheim Pharmaceuticals, Janssen Scientific Affairs LLC, Bayer AG, and Portola Pharmaceuticals. Tatsiana Vaitsiakhovich, Bernhard Schaefer, Anna‐Katharina Meinecke, and Daniel Eriksson are employees of Bayer AG. Elaine Nguyen, Erin R. Weeda, Thomas J. Bunz, and Nitesh A. Sood have no disclosures germane to this article.
Supporting information
Appendix S1.
Coleman CI, Vaitsiakhovich T, Nguyen E, et al. Agreement between coding schemas used to identify bleeding‐related hospitalizations in claims analyses of nonvalvular atrial fibrillation patients. Clin Cardiol. 2018;41:119–125. 10.1002/clc.22861
Funding information Bayer AG, Grant/Award number: NA; Food and Drug Administration
REFERENCES
- 1. Kirchhof P, Benussi S, Kotecha D, et al. 2016 ESC Guidelines for the management of atrial fibrillation developed in collaboration with EACTS. Eur Heart J. 2016;37:2893–2962. [DOI] [PubMed] [Google Scholar]
- 2. January CT, Wann LS, Alpert JS, et al. 2014 AHA/ACC/HRS guideline for the management of patients with atrial fibrillation: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines and the Heart Rhythm Society. J Am Coll Cardiol. 2014;64:e1–e76. [DOI] [PubMed] [Google Scholar]
- 3. Oyinlola JO, Campbell J, Kousoulis AA. Is real world evidence influencing practice? A systematic review of CPRD research in NICE guidances. BMC Health Serv Res. 2016;16:299. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. MHP Health . The tricky second album. From RCTs to RWE, does industry have the right mix for clinicians' tastes? https://issuu.com/MHPCommunications/docs/mhp_health_tricky_second_album. Accessed December 21, 2017.
- 5. Weeda ER, White CM, Peacock WF, Coleman CI. Rates of major bleeding with rivaroxaban in real‐world studies of nonvalvular atrial fibrillation patients: a meta‐analysis. Curr Med Res Opin. 2016;32:1117–1120. [DOI] [PubMed] [Google Scholar]
- 6. Coleman CI, Antz M, Bowrin K, et al. Real‐world evidence of stroke prevention in patients with nonvalvular atrial fibrillation in the United States: the REVISIT‐US study. Curr Med Res Opin. 2016;32:2047–2053. [DOI] [PubMed] [Google Scholar]
- 7. Tamayo S, Peacock F, Patel M, et al. Characterizing major bleeding in patients with nonvalvular atrial fibrillation: a pharmacovigilance study of 27 467 patients taking rivaroxaban. Clin Cardiol. 2015;38:63–68. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Abraham NS, Singh S, Alexander GC, et al. Comparative risk of gastrointestinal bleeding with dabigatran, rivaroxaban, and warfarin: population based cohort study. BMJ. 2015;350:H1857. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Yao X, Abraham NS, Sangaralingham LR, et al. Effectiveness and safety of dabigatran, rivaroxaban, and apixaban versus warfarin in nonvalvular atrial fibrillation. J Am Heart Assoc. 2016;5:e003725. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Cunningham A, Stein CM, Chung CP, Daugherty JR, Smalley WE, Ray WA. An automated database case definition for serious bleeding related to oral anticoagulant use. Pharmacoepidemiol Drug Saf. 2011;20:560–566. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Go AS, Singer D, Cheetham TC, et al. Mini‐Sentinel medical product assessment. A protocol for assessment of dabigatran. Version 3.https://www.sentinelinitiative.org/sites/default/files/Drugs/Assessments/Mini‐Sentinel_Protocol‐for‐Assessment‐of‐Dabigatran_0.pdf. Published March 27, 2015. Accessed April 20, 2017. [Google Scholar]
- 12. Kakkar AJ, Mueller I, Bassand JP, et al. Risk profiles and antithrombotic treatment of patients newly diagnosed with atrial fibrillation at risk of stroke: perspectives from the international, observational, prospective GARFIELD registry. PLoS One. 2013;8:e63479. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. MarketScan research databases . Data for healthcare research. Truven Health website. http://truvenhealth.com/Portals/0/assets/ACRS_11223_0912_MarketScanResearch_SS_Web.pdf. Accessed May 17, 2017.
- 14. Noseworthy PA, Yao X, Abraham NS, Sangaralingham LR, McBane RD, Shah ND. Direct comparison of dabigatran, rivaroxaban, and apixaban for effectiveness and safety in nonvalvular atrial fibrillation. Chest. 2016;150:1302–1312. [DOI] [PubMed] [Google Scholar]
- 15. Gandhi SK, Salmon W, Kong SX, Zhao SZ. Administrative databases and outcomes assessment: an overview of issues and potential utility. J Manag Care Spec Pharm. 1999;5:215–222. [Google Scholar]
- 16. Viera AJ, Garrett JM. Understanding interobserver agreement: the kappa statistic. Fam Med. 2005;37:360–363. [PubMed] [Google Scholar]
- 17. Camm AJ, Amarenco P, Haas S, et al. Safety analysis of rivaroxaban: a pooled analysis of the global XANTUS programme (real‐world, prospective, observational studies for stroke prevention in patients with atrial fibrillation). Eur Heart J. 2017;38(suppl 1):ehx504.P3592. [Google Scholar]
- 18. Ruff CT, Giugliano RP, Braunwald E, et al. Comparison of the efficacy and safety of new oral anticoagulants with warfarin in patients with atrial fibrillation: a meta‐analysis of randomised trials. Lancet. 2014;383:955–962. [DOI] [PubMed] [Google Scholar]
- 19. Bassand JP, Accetta G, Camm AJ, et al. Two‐year outcomes of patients with newly diagnosed atrial fibrillation: results from GARFIELD‐AF. Eur Heart J. 2016;37:2882–2889. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Graham DJ, Recihman ME, Wernecke M, et al. Stroke, bleeding, and mortality risks in elderly Medicare beneficiaries treated with dabigatran or rivaroxaban for nonvalvular atrial fibrillation. JAMA Intern Med. 2016;176:1662–1671. [DOI] [PubMed] [Google Scholar]
- 21. Li X, Deitelzweig S, Keshishian A, et al. Effectiveness and safety of apixaban versus warfarin in non‐valvular atrial fibrillation patients in “real‐world” clinical practice. A propensity‐matched analysis of 76,940 patients. Thromb Haemost. 2017;117:1072–1082. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Maura G, Blotiere PO, Bouillon K, et al. Comparison of the short‐term risk of bleeding and arterial thromboembolic events in nonvalvular atrial fibrillation patients newly treated with dabigatran or rivaroxaban versus vitamin K antagonists: a French nationwide propensity‐matched cohort study. Circulation. 2015;132:1252–1260. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Nguyen E, White CM, Patel MR, et al. Doses of apixaban and rivaroxaban prescribed in real‐world Untied States cardiology practices compared to registration trials. Curr Med Res Opin. 2016;32:1277–1279. [DOI] [PubMed] [Google Scholar]
- 24. Weitz JI, Eikelboom JW. Appropriate apixaban dosing: prescribers take note. JAMA Cardiol. 2016;1:635–636. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Appendix S1.