Effect of Anticoagulation on Hospitalization Costs After Intracranial Hemorrhage in Atrial Fibrillation
A Registry Study
Background and Purpose—Intracranial hemorrhage (ICH) is the most feared adverse event with oral anticoagulant therapy in patients with atrial fibrillation. The health economic aspects of resuming oral anticoagulant therapy after ICH are unknown. The aim was to estimate hospitalization costs of thromboembolism and hemorrhage subsequent to ICH in 2 patient groups with atrial fibrillation surviving the first 90 days post ICH: (1) patients resuming warfarin therapy within 90 days post ICH and (2) patients discontinuing therapy.
Methods—Retrospective data from Danish national registries were linked to identify patients with atrial fibrillation who suffered an ICH between January 1, 1997, and April 1, 2011. Study start was 90 days after incident ICH. Mortality was evaluated using the Kaplan–Meier estimate. Occurrence of hospitalization-requiring thromboembolism and hemorrhage was used to estimate hospitalization costs by linkage of International Classification of Diseases, Tenth Revision, codes to Danish Diagnosis–Related Group tariffs. The effect of resuming warfarin therapy on average 3-year hospitalization costs was estimated by regression analysis adjusted for between-group differences in baseline characteristics.
Results—In the inclusion period, 2162 patients had an ICH; 1098 survived the first 90 days and were included for analysis, and of those, 267 resumed warfarin therapy. Therapy resumption reduced the mean 3-year hospitalization cost of hospitalized patients significantly by US$ 1588 (95% confidence interval, −2925 to −251) and was significantly correlated with fewer hospitalization days per hospitalized patient (−4.6 [95% confidence interval, −7.6 to −1.6]). The marginal effect of therapy resumption on hospitalization costs per patient was US$ −407 (95% confidence interval, −815 to 2).
Conclusions—Resuming warfarin therapy within 90 days after ICH in patients with atrial fibrillation is associated with a decrease in average hospitalization costs.
- atrial fibrillation
- cost and cost analysis
- intracranial hemorrhages
Nonvalvular atrial fibrillation (AF) is associated with a 5-fold increased risk of embolic strokes,1 which is significantly reduced by oral anticoagulant therapy (OAC), either with vitamin K antagonists (eg, warfarin) or with non-vitamin K antagonist oral anticoagulants (NOACs).2 OAC is associated with an increased risk of bleeding where the most serious is intracranial hemorrhage (ICH), which can be devastating to patients.3 Nonetheless, the net clinical benefit of anticoagulating patients with AF has been estimated to outweigh the inherent risk of hemorrhages.4
OAC is recommended after ICH for the majority of patients.2 Understandably, the iatrogenic nature of hemorrhagic events during OAC may, however, cause reluctance among clinicians to prescribe them to patients at an increased risk of hemorrhage.5 This may especially be valid if patients have had previous, life-threatening hemorrhage, such as ICH.
In this study, we investigate the health economic effect of resuming warfarin therapy within 90 days after an incident ICH in patients with AF. This includes an analysis of survival and hospitalization costs related to thromboembolism and hemorrhage for a 3-year period in 2 groups: (1) patients who resume warfarin therapy within 90 days after ICH and (2) patients who do not.
Information was retrieved from 3 well-validated Danish registries: the Danish Civil Registration System, the National Patient Registry, and the Danish National Prescription Registry.6–8 In Denmark, all citizens are provided with a unique personal identification number at birth or immigration, allowing linkage of information from the different data sets at the individual level. Information on patients’ sex, birthday, migration, and vital status was obtained from the Danish Civil Registration System. Data on diagnoses and hospital admission and discharge dates were retrieved from the Danish National Patient Registry. Diagnoses are coded according to the 10th revision of the Danish version of the International Classification of Diseases (ICD). The Danish National Prescription Registry was used to identify patients’ purchase of prescription medication. Prescriptions are coded according to the Anatomic Therapeutics Chemical Classification.
By linkage of the registries, a cohort was established comprising patients with AF on warfarin therapy who had been hospitalized with a diagnosis of ICH (ICD-10 diagnoses: I60–I62, S063C, S064, S065, or S066). Nonvalvular AF was defined as an ICD-10 diagnosis I48 with the baseline absence of mitral stenosis and mechanical heart valves (ICD-10 diagnoses: I05 and Z952–Z954). Diagnosis of AF should be registered at a minimum of 90 days before ICH, and warfarin purchase should be registered no more than 180 days before ICH. Patients who redeemed prescription of heparins or phenprocoumon within 180 days before ICH were excluded as were patients with erroneous birthday or migration registration or who immigrated less than 1 year before ICH.
The index date, that is, study start date, was defined as 90 days after the date of the incident ICH. A blanking period of 90 days was considered appropriate to exclude resource consumption and mortality referable to the ICH rather than to complications related to warfarin therapy or lack thereof. ICHs should be registered between January 1, 1997, and April 1, 2011, ensuring a potential, full 3-year follow-up period for all patients (data censored at June 30, 2014). The group resuming therapy included patients who had a prescription of warfarin after the ICH date, and before the index date, whereas the discontinuation group had no further prescriptions of warfarin up to the index date (Figure 1).
Patients’ comorbidity status at index date was evaluated using the CHA2DS2-VASc and a modified HAS-BLED scores2,9 (Table I in the online-only Data Supplement). The modified HAS-BLED score was calculated with the omission of the point for labile international normalized ratio values as this information is not available in Danish registries.10 Baseline medication status was determined from registered medication prescriptions within 90 days before the index date.
Outcomes included hemorrhagic and thromboembolic complications that occurred after the index date and could be related to warfarin therapy or lack thereof. Occurrence of complications was estimated from primary diagnoses in the Danish National Patient Registry using ICD-10 codes (Table II in the online-only Data Supplement). Hemorrhagic complications included further nontraumatic ICH, gastrointestinal hemorrhage, and major extracranial hemorrhage. Major extracranial hemorrhage was defined as acute anemia, conjunctival, retinal, and vitreous hemorrhage; hemothorax, recurrent, persistent, and unspecified hematuria, hemorrhage from the respiratory passages; and unclassified hemorrhage (Table II in the online-only Data Supplement). Thromboembolic complications included pulmonary and systemic embolism, deep venous thrombosis, myocardial infarction, transient ischemic attack, and ischemic stroke. Only hospitalization-requiring complications were included. Death from any cause was registered to estimate patient survival.
A Danish hospital sector perspective was applied as no records of resource consumption in the primary sector or outside the health sector were available. All costs are given in US dollar based on a purchasing power parity for gross domestic product of Danish kroner (DKK) 762.03 per US$ 100.11,12 Costs accruing in future years were discounted to represent 2015 values using a discount rate of 3%.13
The Danish Diagnosis–Related Group (dkDRG) classification system from 201514 was used to estimate hospitalization costs. The dkDRG classification system is used for activity-based settlement with the hospitals in Denmark where the hospitals are reimbursed by exogenously defined tariffs associated with each dkDRG group.15 On the basis of registered hospital admission (ICD-10 codes), patients were allocated to dkDRG groups, and corresponding dkDRG tariffs were identified on the basis of primary diagnosis assuming that secondary diagnoses would not influence resource consumption substantially.14,16 To accommodate more resource heavy hospitalizations, a reimbursement of US$ 256 per extra day of hospitalization was applied if the length of stay exceeded the trim point of the allocated dkDRG group.14 When admission and discharge date of a hospitalization were the same, the length of stay was registered to be one day.
All analyses were performed according to the intention-to-treat principle based on patients’ anticoagulation status at index date. Incidence of therapy resumption was evaluated by the cumulative incidence function with death as competing risk.
Baseline characteristics of the treatment groups at index date were given as mean and SDs for quantitative information and percentages for qualitative information. Statistically significant differences were identified by Student t test for continuous variables and χ2 test for dichotomous variables. Survival after index date was summarized using the Kaplan–Meier estimate, and group differences were estimated using the Cox proportional hazard ratio. Incidence rates per 100 person-years of all outcomes and crude and adjusted hazard ratios of composite outcomes for the resumption group versus the discontinuation group are given in Table III in the online-only Data Supplement.
The effect of therapy resumption on mean discounted, 3-year hospitalization cost per patient (ci) was estimated by regression analysis. As a large proportion of patients incurred no relevant hospitalization costs and the costs conditional to (ci>0) displayed a skewed distribution, a 2-part model was used to model separately (1) the probability of incurring any hospitalization costs and (2) the level of costs of those with positive costs.17,18 The conditional expectation of the cost was divided into 2 parts:
In the first part, the effect of therapy resumption on the probability of having any hospitalizations was evaluated for all patients using a logistic regression model. The second part was used to analyze the level of total costs for patients who were hospitalized and was constructed via a generalized linear model with a γ-distribution and identity link function.18 The 2 parts of the model were combined to estimate the marginal effect of therapy resumption on the predicted, mean, discounted, 3-year hospitalization cost per patient as given by the partial derivatives of the marginal expectation17,18:
The same 2-part model was used to estimate the effect of therapy resumption on the predicted, total number of hospitalization days per patient by linking the probability of hospitalization with the number of hospitalization days of hospitalized patients.
The effect of varying the index date to day 42, 60, 120, and 180 post ICH on marginal costs and hospitalization days of therapy resumption was investigated as sensitivity analyses.
All models included the indicator variable for therapy resumption and were adjusted for the individual components of the CHA2DS2-VASc and modified HAS-BLED scores at index date. These were considered a priori determined covariates that might affect the risk of complications. Bleeding history was omitted because of the qualifying event. The presence of abnormal renal and hepatic function was included as a composite covariate because of a low prevalence of abnormal hepatic function. Age groups were created for ages <65, 65≥ and <75, and ≥75 at index date (Table 1). All explanatory variables were dichotomous. Cost analyses were performed with robust SEs to account for possible overdispersion.18
Significance tests were 2 tailed. Statistical significance was assumed for P values <0.05 for all tests. Stata/MP 13 (StataCorp LP, College Station, TX) was used for statistical analysis. No ethical approval is needed for anonymous register-based studies in Denmark.
From January 1, 1997, to April 1, 2011, 2162 patients were hospitalized because of ICH while on warfarin therapy. A total of 1064 patients died before index date, leaving 1098 patients for inclusion (Figure 1). Two hundred sixty-seven patients (24.3%) had a prescription of warfarin therapy before the index date and were included in the resumption group. As illustrated in Figure 2, the cumulative incidence of therapy resumption remained linear up to the index date after which it seemed to stagnate. One hundred seventy patients (20.5%) in the discontinuation group resumed therapy after the index date.
Patients in the resumption group tended to be younger (mean age, 74.4 versus 76.7 years; Figure I in the online-only Data Supplement), and a larger proportion of patients were men (female sex; 34.8% versus 39.2%; Table 1). All patients had a high risk of thromboembolism and hemorrhage although the resumption group had a lower risk of both compared with the discontinuation group (CHA2DS2-VASc score: 3.5 versus 3.8 and modified HAS-BLED score: 3.7 versus 3.9; Figure II and III in the online-only Data Supplement, distributions into dichotomous score groups). Also, mortality in the 3-year follow-up period was lower for the resumption group with a crude hazard ratio of 0.73 (95% confidence interval [CI], 0.37 to 0.93; Figure 3). When adjusting for baseline differences, the adjusted hazard ratio for mortality was 0.79 (95% CI, 0.61 to 1.01; Table III in the online-only Data Supplement). On the basis of a total survival time of 680 years, mean survival in the resumption group was 2.6±0.8 years. The total survival time in the discontinuation group was 1943 years, giving a mean survival of 2.3±1.0 years, consistent with the lower mortality in the resumption group versus the discontinuation group.
A total of 339 hospitalizations were observed during the 3-year follow-up period with causes summarized in Table 2. During the follow-up period, 19.5% (n=52) of patients in the resumption group and 21.7% (n=180) in the discontinuation group experienced ≥1 hospitalizations. On the basis of multivariable regression analysis, the predicted probability of hospitalization was 21.1% (95% CI, 20.7 to 21.5) for the entire cohort. The probability of hospitalization in the 2 groups was comparable (odds ratio, 0.92 [95% CI, 0.65 to 1.31]; Table 3).
The total, discounted, 3-year hospitalization cost per hospitalized patient was right skewed with a median cost of US$ 4300 (mean, US$ 5088±4198) in the resumption group and US$ 4545 (mean, US$ 6524±5189) in the discontinuation group. On the basis of regression analysis, the predicted average hospitalization costs of hospitalized patients for the entire cohort was US$ 6231 (95% CI, 6154 to 6308). Therapy resumption reduced the discounted, 3-year hospitalization cost of hospitalized patients significantly by US$ −1588 (95% CI, −2925 to −251; Table 3). Patients in the resumption group had a median of 4 hospitalization days per hospitalized patient (mean, 6.9±9.5), whereas patients in the discontinuation group had a median of 7 hospitalization days per hospitalized patient (mean, 11.6±16.6). For the entire cohort, the predicted mean number of hospitalization days for hospitalized patients was 10.8 days (95% CI, 10.5 to 11.0). Therapy resumption reduced the number of hospitalization days per hospitalized patient significantly (−4.6 [95% CI, −7.6 to −1.6]).
The 2 parts of the model were joined to estimate the incremental 3-year, discounted hospitalization costs and days of hospitalization per patient at marginal variation of therapy resumption, that is, the presence versus the absence of resumption of warfarin therapy at index date. The estimated marginal effect of therapy resumption was adjusted for the effect of the other covariates included in the model.
Therapy resumption seemed to reduce the mean discounted, 3-year hospitalization costs per patient borderline significantly (marginal effect, US$ −407 [95% CI, −815 to 2]). When combining the logistic and the generalized linear model in the 2-part model, the uncertainty surrounding the probability of hospitalization (odds ratio, 0.92 [95% CI, 0.65 to 1.31]) affects the joint model uncertainty, causing insignificance for the resulting marginal cost of therapy resumption. However, therapy resumption still reduced the mean number of hospitalization days significantly (marginal effect, −1.1 [95% CI, −1.9 to −0.3]) because of a larger effect of therapy resumption on hospitalization days of hospitalized patients (Table 3).
When shortening the blanking period, the total patient population alive at index date and the absolute number of patients who were hospitalized increased. However, fewer patients were included in the therapy resumption group and fewer hospitalizations were observed in this group (Table IV in the online-only Data Supplement). The effect of therapy resumption on marginal costs and days of hospitalization increased and became significant when shortening the blanking period. In accordance with this, the marginal effect decreased when extending the blanking period although it remained negative in all analyses (Table 3).
To our knowledge, this is the first study to investigate the health economic effect of resuming warfarin therapy after ICH in patients with AF. The principal finding of this study is that resumption of warfarin therapy reduced the total, discounted, 3-year hospitalization cost of hospitalized patients significantly. Second, resumption of warfarin therapy was significantly correlated with fewer hospitalization days per patient. Consequently, resumption of warfarin therapy seemed to reduce the mean, discounted, 3-year hospitalization cost per patient by US$ −407 (95% CI, −815 to 2).
A 3-year follow-up was expected to sufficiently capture between-group differences in resource consumption within the hospital sector referable to the initial decision whether to resume therapy or not. Application of a longer time horizon was not substantiated by the size of the patient cohort, and the diminution of the cohort at the end of the applied follow-up period did not encourage the extension of the time horizon. Furthermore, it is likely that resource consumption for a longer follow-up would be influenced by other factors than the initial treatment strategy, rendering the correlation between it and belatedly occurring complications questionable.
Studies suggest that patients’ risk of thromboembolism and death tend to be the highest in the period right after an ICH, and evidence suggests that early OAC resumption positively affects both.10,19 The effect of changing the index date was investigated in sensitivity analyses, which concordantly indicated that therapy resumption was associated with lower per-patient hospitalization costs and fewer hospitalization days. Application of the index date of 90 days post ICH in the main analysis was considered appropriate to allow for a sufficient size of the resumption group for analysis while still capturing any possible effect of early resumption.
Survival is likely to affect resource consumption, thereby possibly introducing survivorship bias in estimations of hospitalization costs. Considering that the discontinuation group had a higher mortality, the probability of hospitalization in this group may have been falsely lowered because of less time at risk and, hence, also hospitalization costs. Assuming a survivorship bias, a larger marginal effect of therapy resumption might have been observed if no between-group differences in survival were observed.
A large proportion of patients in both groups had prescriptions of antiplatelets (Table 1), which may have affected the survival and probability of hospitalization. It could be relevant to stratify patients according to their antiplatelet status to investigate the effect of antiplatelet monotherapy and any cumulative effect of antiplatelet and warfarin therapies combined could have on cost accumulation. However, because of the size of the data set used, subgroup stratifications were not relevant.
The inclusion period in the present analysis effectively excluded patients receiving NOACs. Indeed, NOACs might offer a more acceptable choice of OAC after ICH compared with warfarin as NOACs have presented an improved safety profile compared with warfarin.20 However, as previous ICH was an exclusion criterion in all large randomized controlled trials on NOACs, this is only a hypothesis.
The annual costs related to medication and monitoring of warfarin therapy amount to US$ 950 per patient when assuming complete adherence to a defined daily dose of warfarin of 7.5 mg (US$ 17.93/100 tablets, 2.5 mg21,22) and monitoring performed every third week in general practice (consultation [US$ 24.03] and laboratory tests [US$ 20.22]15,23). Hence, the costs of monitoring and medication are not counteracted by the estimated marginal effect of resuming warfarin therapy on hospitalization costs. However, as post-ICH OAC resumption offers lower mortality, fewer thromboembolisms, and a similar risk of hemorrhage compared with no therapy,10,19 the summarized effect on patients’ health-related quality of life may render it cost effective, despite higher costs.24
Because of the strict inclusion criteria applied in the analysis, the patient population was not large and the results should be interpreted with caution. Furthermore, the practice pattern in the Danish secondary sector may be dissimilar to the performance in other healthcare systems, and results may, therefore, not be generalizable.
As this was a register-based study, cerebral imaging identifying the subtype of the qualifying event was not available. It is highly probable that ICH severity may have affected the decision whether to resume therapy. Furthermore, patient-specific circumstances that may have caused patients and clinicians to refrain from therapy resumption, such as ICH severity and patients’ subsequent functional status, were unknown and could not be included for analysis, and hence, it is likely that results are the subject of influence of unobserved confounding. The substantial differences in between-group baseline characteristics, such as age and comorbidity at index date, are likely to have affected the choice whether to resume warfarin therapy. The analyses presented here were adjusted for differences in clinically relevant variables as identified in the registries. Alternatively, analyses could have been performed on propensity score–matched groups if the cohort had been larger.
This study does not inform on causality between continuous warfarin therapy and occurrence of further complications. With the intention-to-treat design, redemption of warfarin prescription before index date was used as an indicator of therapy resumption, but whether patients subsequently discontinued therapy was not included for analysis. Furthermore, some patients possibly resumed therapy in hospital or had spare medication from previous prescriptions before the ICH, which could postpone their need of prescriptions and incorrectly allocate them to the discontinuation group. In both cases, this would bias toward the null. The quality of warfarin therapy also affects patients’ prognosis, but this information was not available for analysis.
The applied perspective was that of the Danish hospital sector and, hence, did not include resource consumption in the primary sector. However, substantial costs also accrue outside the hospital sector after hospital-handled thromboembolic and hemorrhagic complications.25 In all probability, focusing only on the hospital sector has led to an underestimation of actual costs, and consequently, the effect of resuming warfarin therapy after ICH on true costs may be underestimated.
This study evaluated the health economic effect of resuming warfarin therapy within 90 days after an ICH in patients with AF from the perspective of the Danish hospital sector. Hospitalization costs because of thromboembolic and hemorrhagic complications of hospitalized patients were decreased significantly if patients had resumed warfarin therapy within 90 days after the qualifying event. When adjusting for differences in patient characteristics, therapy resumption was associated with a decrease of the mean, discounted, 3-year hospitalization costs of patients with AF.
This study adds to the evidence base supporting the resumption of OAC for the majority of patients with AF who have had severe hemorrhage.
We thank professor Lars Holger Ehlers at the Danish Center for Healthcare Improvements, Aalborg University, for his helpful feedback in designing the study.
Dr Larsen has served as an investigator for Janssen Scientific Affairs, LLC, and Boehringer Ingelheim and has been on the speaker bureaus for Bayer, BMS/Pfizer, Roche Diagnostics, Boehringer Ingelheim, and Takeda Pharma. Dr Lip has served as a consultant for Bayer/Janssen, Astellas, Merck, Sanofi, BMS/Pfizer, Biotronik, Medtronic, Portola, Boehringer Ingelheim, Microlife, and Daiichi-Sankyo and has been on the speaker bureaus for Bayer, BMS/Pfizer, Medtronic, Boehringer Ingelheim, Microlife, Roche, and Daiichi-Sankyo. The other authors report no conflicts.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.115.012338/-/DC1.
- Received December 4, 2015.
- Revision received January 12, 2016.
- Accepted January 19, 2016.
- © 2016 American Heart Association, Inc.
- Wolf PA,
- Abbott RD,
- Kannel WB.
- Camm AJ,
- Lip GY,
- De Caterina R,
- Savelieva I,
- Atar D,
- Hohnloser SH,
- et al.,
- ESC Committee for Practice Guidelines (CPG)
- Hart RG,
- Diener HC,
- Yang S,
- Connolly SJ,
- Wallentin L,
- Reilly PA,
- et al
- Vest-Hansen B,
- Riis AH,
- Christiansen CF.
- Kildemoes HW,
- Sørensen HT,
- Hallas J.
- Lip GY,
- Nieuwlaat R,
- Pisters R,
- Lane DA,
- Crijns HJ.
- 11.↵PPP and exchange rates. Organisation for Economic Co-operation and Development (OECD.stat). http://stats.oecd.org/Index.aspx?datasetcode=SNA_TABLE4. Accessed May 28, 2015.
- Kristensen FB,
- Sigmund H.
- 14.↵Sundhedsdokumentation, Sektor for National Sundhedsdokumentation og Forskning. Takstsystem 2015 – vejledning. Online: Statens Serum Institut; February 17, 2015. http://www.ssi.dk/~/media/Indhold/DK%20-%20dansk/Sundhedsdata%20og%20it/NSF/Sundhedsokonomi%20og%20finansiering/Sundhedokonomi/Takster/Takster%202015/Takstvejledning2015.ashx.
- Pedersen KM.
- 16.↵InteraktivDRG. Statens Serum Institut. http://drgservice.ssi.dk/grouper. Accessed May 21, 2015.
- Gregori D,
- Petrinco M,
- Bo S,
- Desideri A,
- Merletti F,
- Pagano E.
- Jones AM,
- Rice N,
- D’Uva TB,
- Balia S.
- Nielsen PB,
- Larsen TB,
- Skjøth F,
- Gorst-Rasmussen A,
- Rasmussen LH,
- Lip GY.
- Ruff CT,
- Giugliano RP,
- Braunwald E,
- Hoffman EB,
- Deenadayalu N,
- Ezekowitz MD,
- et al
- 21.↵ATC/DDD index – B01AA Vitamin K antagonists. WHO Collaborating Center for Drug Statistics Methodology. http://www.whocc.no/atc_ddd_index/?code=B01AA03&showdescription=yes. Accessed May 21, 2015.
- 22.↵medicinpriser.dk (active substance: warfarin). Danish Medicines Agency. http://www.medicinpriser.dk. Accessed May 27, 2015.
- 23.↵Honorartabel 01-04-2015 DAGTID; §50 Grundydelser, §60 Laboratorieundersøgelser. Danish Medical Association. http://www.laeger.dk/portal/page/portal/LAEGERDK/Laegerdk/P_L_O/Overenskomster/Honorartabel/Honorartabel%201-4-15/Honorartabel%2001-04-2015%20-%20DAGTID. Accessed May 27, 2015.
- 24.↵Process and Methods Guides: Guide to the methods of technology appraisal 2013. Online: National Institute for Health and Clinical Excellence; April 2013. http://www.nice.org.uk/article/pmg9/resources/non-guidance-guide-to-the-methods-of-technology-appraisal-2013-pdf.
- Jakobsen M,
- Fredslund EK,
- Kolodziejczyk C.