Thrombosis Research 133 (2014) 1061–1067
Contents lists available at ScienceDirect
Thrombosis Research journal homepage: www.elsevier.com/locate/thromres
Regular Article
Impact of short periods with worsened or improved INR control on life expectancy and QALYs in patients with atrial fibrillation Eva Lesén a,⁎, Ingela Björholt a, Åse Björstad a, Martin Fahlén b, Anders Odén c a b c
Nordic Health Economics AB (NHE), Gothenburg, Sweden Department of Medicine, Kungälv Hospital, Kungälv, Sweden Department of Mathematical Sciences, Chalmers University of Technology, Gothenburg, Sweden
a r t i c l e
i n f o
Article history: Received 24 June 2013 Received in revised form 12 February 2014 Accepted 31 March 2014 Available online 4 April 2014 Keywords: Warfarin Atrial fibrillation International normalized ratio Quality-adjusted life years
a b s t r a c t Introduction: Warfarin-treated patients with poor international normalized ratio (INR) control, measured with time in therapeutic range (TTR) or the standard deviation of transformed INR (SDTINR), have an increased risk for clinical events. To what extent only a short period with an altered INR control may influence outcomes remains unknown. This study assessed the impact of transient periods of worsened or improved INR control on life expectancy and quality-adjusted life years (QALYs) among warfarin-treated patients with atrial fibrillation (AF) using both metrics. Materials and methods: Warfarin-treated patients with AF, registered in the patient record system Journalia during years 1985–2000, were included. Information on all-cause mortality was collected from the Cause of Death Register. Hypothetical scenarios where patients were assumed to have a transiently altered INR control during 30 days were modeled statistically using hazard functions, and the impact on remaining life expectancy and QALYs was assessed. Results: When using SDTINR, a 70-year old man within the 20th best INR control percentile was estimated to lose 7.4 days of life or 0.0100 QALYs from a 30-day long worsened INR control to that of an average 70-year old male patient. Correspondingly, 4.0 days of life or 0.0059 QALYs would be gained if a 70-year old man within the 20th worst INR control percentile would have an average INR control during 30 days. The magnitudes were smaller when TTR was used to determine INR control. Conclusions: Even short periods of an altered INR control is expected to have impact on life expectancy and QALYs among patients with AF. © 2014 Elsevier Ltd. All rights reserved.
Introduction Atrial fibrillation (AF) has been estimated to affect about 1-3% of the general population [1–4]. The prevalence increases steeply with age from approximately 0.1% among individuals younger than 55 years to 9% among those aged 80 years and older [5]. AF is a major risk factor for stroke; the condition increases the risk 5-fold [6]. The oral anticoagulant agent warfarin has been estimated to reduce the risk of stroke by 64% among patients with AF, but also increases the risk for bleeding [5]. Warfarin has a narrow therapeutic index, i.e. the margin between a beneficial treatment intensity and an intensity which increases the risk for adverse events is small [7]. Furthermore, warfarin interacts with several medicines as well as food and the sensitivity to its effect varies both
Abbreviations: AF, Atrial Fibrillation; INR, International Normalized Ratio; QALY, Quality-Adjusted Life Years; SD, Standard Deviation; SDTINR, Standard Deviation of Transformed International Normalized Ratio; TTR, Time in Therapeutic Range. ⁎ Corresponding author at: Nordic Health Economics, Medicinaregatan 8b, SE-413 90 Gothenburg, Sweden. Tel.: +46 707 44 77 85; fax: +46 31 741 17 01. E-mail address: [email protected] (E. Lesén).
http://dx.doi.org/10.1016/j.thromres.2014.03.052 0049-3848/© 2014 Elsevier Ltd. All rights reserved.
between and within patients. Regular control of treatment intensity is therefore required. The complexity associated with warfarin use and the fear of adverse events might hinder the use of warfarin among patients who could otherwise benefit from this treatment. Previous research does suggest under-use among patients at a high risk for stroke [8–10], and approximately only half of all of eligible AF patients have been estimated to be treated with warfarin [11,12]. A reliable measure of treatment intensity is thus of great clinical importance. For most AF patients treated with warfarin, the standard therapeutic target is to have an international normalized ratio (INR) within 2–3, although values in the lower range have been suggested as optimal [13]. The proportion of time the patient spends within INR 2–3 is used to measure treatment intensity and quality, and a large value for the time in therapeutic range (TTR) is associated with a lower risk for stroke, bleeding and death [14,15]. The current paradigm in warfarin treatment is that a high TTR indicates successful treatment quality. However, Lind et al. (2012) showed that this perception could be questioned [16]. The authors introduced a metric based on the standard deviation (SD) of transformed INR values (SDTINR), i.e. the variability in INR values over time. This new metric was found to be superior to TTR in
1062
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
the ability to predict the risk for clinical events among warfarin-treated patients with AF. Thus, it has previously been shown that patients with a poor INR control, measured with TTR or with SDTINR, have an increased risk for clinical events, but to what extent only a short period with an altered INR control has an impact on outcomes remains unknown. The objective of this study was to assess the impact of a transient period of a worsened (or improved) INR control on life expectancy and quality-adjusted life years (QALYs) lost (or gained) due to the altered probability of death during these periods, among warfarin-treated patients with AF. The INR control was determined both by TTR and by SDTINR. Materials and Methods Study Population Journalia is a patient record system used to monitor warfarin therapy among outpatients in approximately 100 hospital-based and centralized anticoagulation clinics in Sweden. The system includes patient-level data on warfarin monitoring, such as the patients’ age, sex, INR-values and date of monitoring. A previously extracted dataset from Journalia was used for this study [16]. The study population encompassed patients with AF and warfarin treatment, registered in Journalia during the years 1985–2000. Patients with fewer than five INR measurements were excluded. No further inclusion or exclusion criteria were applied. This nationally representative dataset included 19,177 patients and encompassed approximately 36,000 patient-years. Data from the more recent years were overrepresented. Information on all-cause mortality was collected from the Cause of Death Register, held by the Swedish National Board of Health and Welfare. The register linkage was performed with the unique person identification number as the key. The study was approved by the Local Ethical Review Board in Gothenburg (registration number Ö 211–01). Study Design A schematic illustration of the study design is presented in Fig. 1. Information on demographics, INR control and mortality for the study population was based on data from Journalia and the Cause of Death Register. These data were used to estimate the short-term risk of death in relation to INR control as well as the average remaining life expectancy and QALYs. Hypothetical scenarios were created. In one scenario, patients of a specific age and sex were assumed to continuously have a relatively “good” INR control. In another scenario, patients
were assumed to have a transiently worsened INR control during a short time period, and thereafter the same “good” INR control as in the first scenario. The level of INR control during this 30-day period corresponded to that of an average patient of the same age and sex. By statistical modeling using hazard functions, the differences between the two scenarios in the total estimated remaining life expectancy and QALYs in a life-long perspective were estimated, i.e. the number of days of life or QALYs that could be lost due to a worsened INR control during 30 days. The analyses concerning an improved INR control were the same as described above, but performed in the opposite direction (i.e. patients with a relatively “poor” INR control were assumed to have a transiently improved INR control during 30 days). How these alterations in INR control may be achieved was beyond the scope of this study, but the scenario in which a worsened INR control was modeled could e.g. reflect a change in diet or a less rigid therapy management during a vacation period when usual everyday routines may be changed. Correspondingly, the scenario concerning an improved INR control could correspond to a situation where the dose adjustments or INR monitoring frequencies would be improved, or if patients would be treated with a medication less sensitive to drug and food interactions. The order in which these two scenarios are presented in this paper is arbitrary and is not intended to reflect a ranking of their relative assumed importance in clinical practice. Rationale for the Study Design The focus of this study was thus to assess the impact of a short and transient alteration in INR control on life expectancy and QALYs in a life-long perspective. The rationale for the choice of study design was as follows. The calculation of SDTINR and TTR require data on INR values over a period of time, during which factors other than INR control may change. These time-covarying factors may be more or less feasible to adjust for. Thus, analyses based on actually observed alterations in INR control would unavoidably demand a long observation time, and the possibility to isolate the effect of an altered INR control would be hindered. Instead, hypothetical scenarios of abrupt alterations in INR control were created and modeled statistically. Nevertheless, observed patient data were used for the calculation of the short-term risk of death in relation to the patients’ INR control, as well as the average remaining life expectancy and QALYs. Statistical Analyses Calculation of INR Control INR has a skewed distribution. For the calculation of SDTINR, INR was transformed using the inverse of the standardized normal distribution function to achieve a normally distributed variable, as described in
Fig. 1. Schematic illustration of the study design. Information on demographics, INR control and mortality for the study population was based on data from Journalia and the Cause of Death Register. In the upper scenario, patients were assumed to continuously have a relatively “good” INR control. In the lower scenario, patients were assumed to have a transiently worsened INR control corresponding to that of an average patient of the same age and sex during a short time period, and thereafter the same “good” INR control as in the upper scenario. By statistical modeling, the differences between the two scenarios in the total estimated remaining life expectancy and QALYs in a life-long perspective were estimated. Analyses concerning the scenario of an improved INR control were also performed, but in the opposite direction (i.e. patients with a relatively “poor” INR control were assumed to have a transiently improved INR control during 30 days). Abbreviations: INR, international normalized ratio; QALYs, quality-adjusted life years.
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
1063
detail previously [16]. TTR was calculated via linear interpolation of INR measurements and the time between these, using the Rosendaal method [17]. For every new measurement of INR, SDTINR and TTR were updated. If the total number of INR measurements exceeded 19 (corresponding to approximately one year), the first ones were deleted from the calculation so that no more than the 19 most recent INR measurements were used. Each patient was followed from one INR measurement to the next or to death, if either occurred within 7 weeks (a common maximum interval between monitoring visits in the study population). After these 7 weeks, the patient was censored until a new INR measurement occurred, which was then used as a starting point for a new period. For SDTINR, there is currently no established cut-off for a good or poor INR control. To avoid using a cut-off with unknown validity and also to increase the comparability of the results between TTR and SDTINR, patients were instead categorized into percentiles based on their level of INR control. The association between SDTINR and TTR is inverse, i.e. a high quality of INR control corresponds to low SDTINR-values and high TTR-values. The values for TTR were inverted so that the lower percentiles correspond to a high quality of INR control for both SDTINR and TTR.
This study estimated the short-term risk of death, and thereby assumed that the INR level only influenced the risk of death during the following 30-day period. However, in practice, the level of INR control may influence the risk of death after the 30-day period as well. Thus, any residual benefit or disadvantage before or after the alteration in INR control was not captured in this study. Still, the consequences of the altered risk of death during this short period on the remaining days of life and QALYs were based on the patients' entire remaining life expectancy. Thus, the number of life days lost (or gained) could be substantial, even if the probability of dying was low (within 30 days). A more detailed description of these calculations, including the hazard functions, is available in the online supplement.
Modeling the Impact of a Transiently Altered INR Control on Remaining Life Expectancy and QALYs Based on data from Journalia and the Cause of Death Register, the risk (hazard) of death within 30 days was calculated according to the patients’ age, sex and INR control. The hazard function was in the form exp(β0 + β1 * current age + β2 * sex + β3 * Z), i.e. the same as in Lind et al. [16]. The variable Z was equal to either TTR or SDTINR, in both cases a time-varying covariate. These hazard functions thus yielded estimates of the 30-day risk of death for each value of TTR and SDTINR. The remaining life expectancy among patients with AF and warfarin treatment was then estimated based on integrals of hazard functions, using data from Journalia and the Cause of Death Register. These calculations are described in more detail in the online supplement. Briefly, the functions consider the risk of death at each point in time and the probability of having survived until that point in time, summed over the patients’ entire life span. These functions thus enabled the estimation of the remaining life expectancy according to age, sex and INR control. The impact of a short alteration in the level of INR control on the estimated remaining life expectancy in a life-long perspective was then modeled as follows: In one hypothetical scenario, patients of a specific age and sex were assumed to continuously have a “good” INR control, and their remaining life expectancy was estimated. In a hypothetical scenario of a transiently worsened INR control, the INR control during the first 30 days was assumed to correspond to that of an average patient of the same age and sex (i.e. a poorer INR control). Thereafter, the INR control was assumed to be at the same level as in the scenario of a continuously “good” INR control. The remaining life expectancy in this scenario was estimated. The difference between these two scenarios in the estimated remaining life expectancy was then assessed, and this difference thus corresponded to the number of days of life that could be lost due to the 30-day long period with a worsened INR control. To estimate the remaining QALYs instead of life expectancy, age- and sex adjusted utilities were applied. The utility data were based on a previous Swedish survey of 11,698 randomly selected individuals from the general population [18]. Due to the lack of adequate data, the applied utility values were not adjusted for the presence of AF and warfarin treatment. An annual discount rate of 3.5% was applied in the calculation of QALYs. The analyses concerning an improved INR control were the same as described above, but performed in the opposite direction, i.e. patients with a relatively poor INR control were assumed to have an INR control level corresponding to an average warfarin-treated AF patient of the same age and sex during 30 days, and the impact on the remaining life expectancy and QALYs were estimated.
Characteristics of the Study Population
Statistical Software Statistical analyses were performed using software written in Basic. The software is based on the same structure and has the same constructor (co-author Professor Anders Odén) as the statistical application in FRAX® (www.shef.ac.uk/FRAX), developed by the World Health Organization. Results
As described in the Methods, the study population included 19,177 individuals. The mean (SD) age was 73.4 (6.1) years, and 59% were men. The average 70-year old man had an SDTINR of 0.86 and a TTR of 62%. The included patients were followed for a mean of 1.9 years. Table 1 shows the estimated remaining life expectancy and QALYs in the study population according to age and sex. This was estimated for all ages, but is presented for a selection of ages. Categorization of Patients According to their INR Control Patients were categorized into percentiles based on their INR control, as determined by SDTINR and TTR, respectively. Fig. 2 shows the values for SDTINR and TTR according to each percentile. The 99th percentile thus corresponds to the 1% of patients with the worst INR control according to SDTINR and TTR, respectively. Impact of a Transiently Altered INR Control During 30 Days Fig. 3 shows the expected impact on days of life among 70-year old men in each percentile due to a hypothetical 30-day long alteration in INR control corresponding to that of an average 70-year old man. This average patient is thus found in the percentile where the difference in days of life equals zero (by definition). Patients below this percentile would lose days of life if their INR control would transiently worsen to the INR control level of an average patient. Correspondingly, patients above that percentile would gain days of life. The figure also shows that the magnitude of the lost (or gained) days of life would be larger Table 1 Estimated remaining life years and QALYs in the study population. Remaining life years
Remaining QALYs
Age (years)
Men
Women
Men
Women
20 30 40 50 60 70 80
53.0 43.5 34.3 25.8 18.2 11.8 7.1
59.4 49.7 40.3 31.3 23.0 15.8 10.0
20.1 17.8 15.1 12.2 9.2 6.2 4.1
19.8 17.7 15.2 12.3 9.2 6.5 4.6
The remaining life expectancy and QALYs were estimated for all ages, but are presented here only for a selection of ages. Abbreviation: QALYs, quality-adjusted life years
1064
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
control would correspond to that among an average warfarin-treated patient with AF of the same age and sex.
Fig. 2. Level of INR control for patients categorized into percentiles 1 to 99 for SDTINR and TTR, respectively. As shown in the figure, higher percentiles correspond to poorer INR control. Abbreviations: INR, international normalized ratio; SDTINR, standard deviation of transformed international normalized ratio; TTR, time in therapeutic range.
when SDTINR was used to determine INR control, as compared to TTR. The corresponding curves for QALYs would have a similar shape as the curves concerning life expectancy, and are therefore not presented. Worsened INR Control As stated above, the results regarding the magnitudes of the expected loss in days of life due to a transiently worsened INR control during 30 days are illustrated for all percentiles of 70-year old men in Fig. 3. Results concerning two examples of patient percentiles are described here, namely those with the 20th and the 2.5th best INR control, respectively. When SDTINR was used to determine INR control, a 70-year old man with the 20th best INR control (i.e. on the 20th percentile) would lose 7.4 days of life or 0.0100 QALYs due to the 30-day long period with a worsened INR control to that of an average 70-year old man with AF and warfarin. When TTR was used to determine INR control, the corresponding loss would be 3.7 days or 0.0051 QALYs. The 70-year old men with the 2.5th best INR control according to SDTINR would lose 15.5 days of life or 0.0196 QALYs. The corresponding loss according to TTR would be 6.6 days or 0.0089 QALYs. To illustrate how the results vary according to age, Fig. 4 shows the expected loss in days of life and QALYs if patients with the 2.5th best INR control according to SDTINR and TTR, respectively, were assumed to have a worsened INR control during 30 days. The worsened INR
Fig. 3. Expected gains and losses in days of life due to a 30-day long alteration in INR control. Patients were categorized into percentiles based on their INR level; higher percentiles correspond to poorer INR control. The figure shows the results for 70-year old men in each percentile. The level of INR control during the 30-day period corresponded to that of an average 70-year old warfarin-treated man with AF. By definition, this average patient is found in the percentile where the difference in days of life equals zero. Abbreviations: AF, atrial fibrillation; INR, international normalized ratio; SDTINR, standard deviation of transformed international normalized ratio; TTR, time in therapeutic range.
Improved INR Control The results regarding the magnitudes of the expected gains in days of life due to a transiently improved INR control during 30 days are illustrated for all percentiles of 70-year old men in Fig. 3. When SDTINR was used to determine INR control, a 70-year old man with the 20th worst INR control (i.e. on the 80th percentile) would gain 4.0 days of life or 0.0059 QALYs due to the 30-day long period with an improved INR control to that of an average 70-year old man with AF and warfarin. When TTR was used to determine INR control, the corresponding gain would be 0.6 days or 0.0009 QALYs. The 70-year old men with the 2.5th worst INR control according to SDTINR would gain 10.8 days of life or 0.0168 QALYs. The corresponding gain according to TTR would be 3.4 days or 0.0049 QALYs. Fig. 5 illustrates how the results vary according to age, and shows the expected gain in days of life and QALYs if patients with the 2.5th worst INR control according to SDTINR and TTR, respectively, were assumed to have an improved INR control during 30 days. The improved INR control would correspond to that among an average warfarin-treated patient with AF of the same age and sex. Discussion This study has illustrated that a worsened INR control during 30 days only can be expected to lead to lost days of life and QALYs in a life-long perspective among warfarin-treated patients with AF. This scenario could reflect a change in diet or a less rigid therapy management during a vacation period. A 70-year old man with the 20th best INR control according to SDTINR could be expected to lose 7.4 days of life or 0.0100 QALYs if his INR control would worsen during 30 days to that of an average 70-year old warfarin-treated man with AF. Considering that the INR control was assumed to be worsened during 30 days only, and that the worsened control was to that of an average patient, as opposed to an extremely poorly controlled patient, the magnitudes of the expected loss in days of life and QALYs are quite substantial. This study further illustrated that even a short and transient period of an improved INR control may have an impact on remaining days of life and QALYs. This scenario could e.g. reflect a situation where the dose adjustments or INR monitoring frequencies would be improved, or if patients would be treated with a medication less sensitive to drug and food interactions. For 70-year old men with the 20th worst INR control according to SDTINR, a 30-day improvement to that of an average 70-year old man was estimated to yield an additional 4.0 days of life or 0.0059 QALYs. Again, given that the INR control was assumed to be improved only during 30 days and only to the level of an average patient, the magnitudes are relatively large. No previous study has estimated the impact on life expectancy or QALYs in relation to a worsened or improved INR control as determined by SDTINR, to our knowledge. There are a few previous studies analyzing the impact of an improved INR control which apply a similar approach using TTR. Rose et al. (2011) simulated the number of adverse events that could be avoided and the QALYs that could be gained if the TTR would be improved in a population of 67,077 AF patients on anticoagulation [19]. The adverse events included ischemic stroke, major hemorrhage and all-cause mortality. An increase in TTR by 5% gave rise to a gain of 863 QALYs in their total population over a 2 year horizon, which could be converted to an average gain of 0.0129 QALYs per patient. Regier et al. (2006) compared self-managed to physicianmanaged anticoagulation control and modeled the impact on QALYs [20]. TTR was assumed to be 72% in the self-managed group and 63% in the physician-managed group. The adverse events included thrombosis, minor and major hemorrhage, death following major thrombotic event and death following major hemorrhagic event. Over a 5-year period, the average gain in QALYs in the self-managed group compared to
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
1065
Fig. 4. Expected differences in days of life and QALYs among patients with the 2.5th best INR control due to a 30-day long worsened INR control to the INR control corresponding to that of an average warfarin-treated AF patient of the same age and sex. Abbreviations: AF, atrial fibrillation; INR, international normalized ratio; QALYs, quality-adjusted life years; SDTINR, standard deviation of transformed international normalized ratio; TTR, time in therapeutic range.
the physician-managed group was 0.07. Sorensen et al. (2009) modeled the cost-effectiveness of warfarin treatment among 70-year old AF patients using previously published rates of stroke, bleeding, adherence
etc. [21]. When a TTR of 100% was assumed for these patients, the remaining life years and QALYs in a lifetime perspective were estimated to 9.7 and 7.2, respectively. Under conditions assumed to represent
Fig. 5. Expected differences in days of life and QALYs among patients with the 2.5th worst INR control due to a 30-day long improved INR control to the INR control corresponding to that of an average warfarin-treated AF patient of the same age and sex. Abbreviations: AF, atrial fibrillation; INR, international normalized ratio; QALYs, quality-adjusted life years; SDTINR, standard deviation of transformed international normalized ratio; TTR, time in therapeutic range.
1066
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
real-world warfarin control (TTR 48%), the corresponding figures were 9.3 life years and 6.8 QALYs. The designs differ between these previous studies and the current study, and direct comparisons of the absolute impact on life expectancy and QALYs are therefore not appropriate. Still, all of these previous studies suggest that an improved quality of INR control, albeit measured with TTR, may lead to an increase in life expectancy and QALYs. This is in accordance with the findings reported in this study. The current study also showed that the impact of a transiently altered INR control on life expectancy and QALYs was more pronounced when SDTINR was used to categorize patients according to their quality of INR control, than when TTR was used.
Acknowledgments
Strengths and Limitations
Appendix A. Supplementary Description
This study was based on nationally representative data including detailed information on warfarin treatment and monitoring frequency in a large patient population. These data were used to calculate the shortterm risk of death in relation to the patients’ INR control, as well as the average remaining life expectancy and QALYs. By creating hypothetical scenarios, the impact of an altered INR control on the remaining life expectancy and QALYs was modeled statistically. The estimation of future QALYs did not incorporate the effects of events other than death, e.g. stroke or bleeding, since such analyses would be associated with major uncertainties due to the lack of reliable information on the severity of these events. Also, assumptions regarding the patients’ future INR control as well as the impact of the events on the patients’ entire remaining life course would need to be made. These uncertainties are evidently not present in the case of death, since this event occurs only once for each patient and has the same consequences in terms of future QALYs for all affected patients (i.e. 0). Furthermore, the applied utility values were not adjusted for the presence of AF and warfarin treatment, due to the lack of applicable data on such ageand sex stratified utilities. Warfarin patients in Sweden are generally considered to be wellcontrolled (according to TTR) compared to warfarin patients in other countries, which could possibly limit the generalizability to other settings. However, the findings in Lind et al. [16] show that SDTINR has a greater ability than TTR to predict the risk for clinical events. Also, Lind et al. showed that the correlation between TTR and SDTINR was weak (r = −0.27) [16]. The perception of the well-controlled patient population in Sweden might therefore be questioned. Furthermore, the data used in this study were from 1985–2000. The use of preventive cardiovascular drugs has increased and the mortality due to cardiovascular disease has decreased in Sweden during and after this period [22–26]. Also, the life expectancy in Sweden has increased somewhat in later years, and is currently higher than in many other countries [27]. The dataset encompassing years 1985–2000 may thus have produced results that are more comparable to a later period for many other countries. However, since this study assessed the impact of transient periods with an altered INR control, and since percentiles were used instead of absolute cut-off values (e.g. a TTR ≥70%) to categorize patients according to their level of INR control, the age of the dataset and the proportion of well-controlled patients in Sweden compared to other countries are of minor importance.
A supplementary description of statistical calculations in this article can be found online at http://dx.doi.org/10.1016/j.thromres.2014.03. 052.
Conclusions Even short and transient periods with an altered INR control can be expected to have an impact on remaining life expectancy and QALYs among patients with AF. The findings suggest that it is important to continuously maintain a good anticoagulation control, since even a short period with a reduced quality of control may be harmful. Furthermore, even short term efforts to improve anticoagulation control may be beneficial.
Conflict of Interest Statement This study was sponsored by Bristol-Myers Squibb and Pfizer. The authors have no other conflicts of interest to report.
This study was sponsored by Bristol-Myers Squibb and Pfizer. The sponsors had no role in the collection, analysis or interpretation of data, but provided comments on the study design and on the manuscript. The sponsors supported the decision to submit the manuscript for publication.
References [1] Ericson L, Bergfeldt L, Bjorholt I. Atrial fibrillation: the cost of illness in Sweden. Eur J Health Econ 2011;12:479–87. [2] Go AS, Hylek EM, Phillips KA, Chang Y, Henault LE, Selby JV, et al. Prevalence of diagnosed atrial fibrillation in adults: national implications for rhythm management and stroke prevention: the AnTicoagulation and Risk Factors in Atrial Fibrillation (ATRIA) Study. JAMA 2001;285:2370–5. [3] SBU – Statens beredning för medicinsk utvärdering (Swedish Council on Health Technology Assessment). Förmaksflimmer - förekomst och risk för stroke (Atrial fibrillation - prevalence and risk for stroke) [Report in Swedish]. http://www.sbu. se/upload/Publikationer/Content0/2/Formaksflimmer_forekomst_risk_for_stroke. pdf; 2013. [4] Bjorck S, Palaszewski B, Friberg L, Bergfeldt L. Atrial fibrillation, stroke risk, and warfarin therapy revisited: a population-based study. Stroke 2013;44:3103–8. [5] Hart RG, Pearce LA, Aguilar MI. Meta-analysis: antithrombotic therapy to prevent stroke in patients who have nonvalvular atrial fibrillation. Ann Intern Med 2007;146:857–67. [6] Wolf PA, Abbott RD, Kannel WB. Atrial fibrillation as an independent risk factor for stroke: the Framingham Study. Stroke 1991;22:983–8. [7] Summary of product characteristics: Warfarin. Available from the Swedish Medical Products Agency http://www.lakemedelsverket.se. [8] Frykman V, Beerman B, Ryden L, Rosenqvist M. Management of atrial fibrillation: discrepancy between guideline recommendations and actual practice exposes patients to risk for complications. Eur Heart J 2001;22:1954–9. [9] Bradley BC, Perdue KS, Tisdel KA, Gilligan DM. Frequency of anticoagulation for atrial fibrillation and reasons for its non-use at a Veterans Affairs medical center. Am J Cardiol 2000;85:568–72. [10] Ogilvie IM, Newton N, Welner SA, Cowell W, Lip GY. Underuse of oral anticoagulants in atrial fibrillation: a systematic review. Am J Med 2010;123:638–45. [11] Friberg L, Hammar N, Ringh M, Pettersson H, Rosenqvist M. Stroke prophylaxis in atrial fibrillation: who gets it and who does not? Report from the Stockholm Cohort-study on Atrial Fibrillation (SCAF-study). Eur Heart J 2006;27:1954–64. [12] Baczek VL, Chen WT, Kluger J, Coleman CI. Predictors of warfarin use in atrial fibrillation in the United States: a systematic review and meta-analysis. BMC Fam Pract 2012;13:5. [13] Oden A, Fahlen M. Oral anticoagulation and risk of death: a medical record linkage study. BMJ 2002;325:1073–5. [14] Gallagher AM, Setakis E, Plumb JM, Clemens A, van Staa TP. Risks of stroke and mortality associated with suboptimal anticoagulation in atrial fibrillation patients. Thromb Haemost 2011;106:968–77. [15] Wan Y, Heneghan C, Perera R, Roberts N, Hollowell J, Glasziou P, et al. Anticoagulation control and prediction of adverse events in patients with atrial fibrillation: a systematic review. Circ Cardiovasc Qual Outcomes 2008;1:84–91. [16] Lind M, Fahlen M, Kosiborod M, Eliasson B, Oden A. Variability of INR and its relationship with mortality, stroke, bleeding and hospitalisations in patients with atrial fibrillation. Thromb Res 2012;129:32–5. [17] Rosendaal FR, Cannegieter SC, van der Meer FJ, Briet E. A method to determine the optimal intensity of oral anticoagulant therapy. Thromb Haemost 1993;69:236–9. [18] Burstrom K, Johannesson M, Diderichsen F. Health-related quality of life by disease and socio-economic group in the general population in Sweden. Health Policy 2001;55:51–69. [19] Rose AJ, Berlowitz DR, Ash AS, Ozonoff A, Hylek EM, Goldhaber-Fiebert JD. The business case for quality improvement: oral anticoagulation for atrial fibrillation. Circ Cardiovasc Qual Outcomes 2011;4:416–24. [20] Regier DA, Sunderji R, Lynd LD, Gin K, Marra CA. Cost-effectiveness of self-managed versus physician-managed oral anticoagulation therapy. CMAJ 2006;174:1847–52. [21] Sorensen SV, Dewilde S, Singer DE, Goldhaber SZ, Monz BU, Plumb JM. Costeffectiveness of warfarin: trial versus "real-world" stroke prevention in atrial fibrillation. Am Heart J 2009;157:1064–73.
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067 [22] Wandell PE. Drug use in patients with atrial fibrillation in Swedish primary health care: a comparison 5 years apart. Eur J Clin Pharmacol 1999;55:333–7. [23] Walley T, Folino-Gallo P, Stephens P, Van Ganse E. Trends in prescribing and utilization of statins and other lipid lowering drugs across Europe 1997–2003. Br J Clin Pharmacol 2005;60:543–51. [24] Carlsson AC, Wandell P, Sundquist K, Johansson SE, Sundquist J. Differences and time trends in drug treatment of atrial fibrillation in men and women and doctors'
1067
adherence to warfarin therapy recommendations: A Swedish study of prescribed drugs in primary care in 2002 and 2007. Eur J Clin Pharmacol 2013;69:245–53. [25] Silwer L, Lundborg CS. Patterns of drug use during a 15 year period: data from a Swedish county, 1988–2002. Pharmacoepidemiol Drug Saf 2005;14:813–20. [26] National Board of Health and Welfare. Official statistics of Sweden. Statistics – Health and Medical Care. Causes of Death 2010; 2011. [27] World Health Organization. World health statistics; 2012.
Contents lists available at ScienceDirect
Thrombosis Research journal homepage: www.elsevier.com/locate/thromres
Regular Article
Impact of short periods with worsened or improved INR control on life expectancy and QALYs in patients with atrial fibrillation Eva Lesén a,⁎, Ingela Björholt a, Åse Björstad a, Martin Fahlén b, Anders Odén c a b c
Nordic Health Economics AB (NHE), Gothenburg, Sweden Department of Medicine, Kungälv Hospital, Kungälv, Sweden Department of Mathematical Sciences, Chalmers University of Technology, Gothenburg, Sweden
a r t i c l e
i n f o
Article history: Received 24 June 2013 Received in revised form 12 February 2014 Accepted 31 March 2014 Available online 4 April 2014 Keywords: Warfarin Atrial fibrillation International normalized ratio Quality-adjusted life years
a b s t r a c t Introduction: Warfarin-treated patients with poor international normalized ratio (INR) control, measured with time in therapeutic range (TTR) or the standard deviation of transformed INR (SDTINR), have an increased risk for clinical events. To what extent only a short period with an altered INR control may influence outcomes remains unknown. This study assessed the impact of transient periods of worsened or improved INR control on life expectancy and quality-adjusted life years (QALYs) among warfarin-treated patients with atrial fibrillation (AF) using both metrics. Materials and methods: Warfarin-treated patients with AF, registered in the patient record system Journalia during years 1985–2000, were included. Information on all-cause mortality was collected from the Cause of Death Register. Hypothetical scenarios where patients were assumed to have a transiently altered INR control during 30 days were modeled statistically using hazard functions, and the impact on remaining life expectancy and QALYs was assessed. Results: When using SDTINR, a 70-year old man within the 20th best INR control percentile was estimated to lose 7.4 days of life or 0.0100 QALYs from a 30-day long worsened INR control to that of an average 70-year old male patient. Correspondingly, 4.0 days of life or 0.0059 QALYs would be gained if a 70-year old man within the 20th worst INR control percentile would have an average INR control during 30 days. The magnitudes were smaller when TTR was used to determine INR control. Conclusions: Even short periods of an altered INR control is expected to have impact on life expectancy and QALYs among patients with AF. © 2014 Elsevier Ltd. All rights reserved.
Introduction Atrial fibrillation (AF) has been estimated to affect about 1-3% of the general population [1–4]. The prevalence increases steeply with age from approximately 0.1% among individuals younger than 55 years to 9% among those aged 80 years and older [5]. AF is a major risk factor for stroke; the condition increases the risk 5-fold [6]. The oral anticoagulant agent warfarin has been estimated to reduce the risk of stroke by 64% among patients with AF, but also increases the risk for bleeding [5]. Warfarin has a narrow therapeutic index, i.e. the margin between a beneficial treatment intensity and an intensity which increases the risk for adverse events is small [7]. Furthermore, warfarin interacts with several medicines as well as food and the sensitivity to its effect varies both
Abbreviations: AF, Atrial Fibrillation; INR, International Normalized Ratio; QALY, Quality-Adjusted Life Years; SD, Standard Deviation; SDTINR, Standard Deviation of Transformed International Normalized Ratio; TTR, Time in Therapeutic Range. ⁎ Corresponding author at: Nordic Health Economics, Medicinaregatan 8b, SE-413 90 Gothenburg, Sweden. Tel.: +46 707 44 77 85; fax: +46 31 741 17 01. E-mail address: [email protected] (E. Lesén).
http://dx.doi.org/10.1016/j.thromres.2014.03.052 0049-3848/© 2014 Elsevier Ltd. All rights reserved.
between and within patients. Regular control of treatment intensity is therefore required. The complexity associated with warfarin use and the fear of adverse events might hinder the use of warfarin among patients who could otherwise benefit from this treatment. Previous research does suggest under-use among patients at a high risk for stroke [8–10], and approximately only half of all of eligible AF patients have been estimated to be treated with warfarin [11,12]. A reliable measure of treatment intensity is thus of great clinical importance. For most AF patients treated with warfarin, the standard therapeutic target is to have an international normalized ratio (INR) within 2–3, although values in the lower range have been suggested as optimal [13]. The proportion of time the patient spends within INR 2–3 is used to measure treatment intensity and quality, and a large value for the time in therapeutic range (TTR) is associated with a lower risk for stroke, bleeding and death [14,15]. The current paradigm in warfarin treatment is that a high TTR indicates successful treatment quality. However, Lind et al. (2012) showed that this perception could be questioned [16]. The authors introduced a metric based on the standard deviation (SD) of transformed INR values (SDTINR), i.e. the variability in INR values over time. This new metric was found to be superior to TTR in
1062
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
the ability to predict the risk for clinical events among warfarin-treated patients with AF. Thus, it has previously been shown that patients with a poor INR control, measured with TTR or with SDTINR, have an increased risk for clinical events, but to what extent only a short period with an altered INR control has an impact on outcomes remains unknown. The objective of this study was to assess the impact of a transient period of a worsened (or improved) INR control on life expectancy and quality-adjusted life years (QALYs) lost (or gained) due to the altered probability of death during these periods, among warfarin-treated patients with AF. The INR control was determined both by TTR and by SDTINR. Materials and Methods Study Population Journalia is a patient record system used to monitor warfarin therapy among outpatients in approximately 100 hospital-based and centralized anticoagulation clinics in Sweden. The system includes patient-level data on warfarin monitoring, such as the patients’ age, sex, INR-values and date of monitoring. A previously extracted dataset from Journalia was used for this study [16]. The study population encompassed patients with AF and warfarin treatment, registered in Journalia during the years 1985–2000. Patients with fewer than five INR measurements were excluded. No further inclusion or exclusion criteria were applied. This nationally representative dataset included 19,177 patients and encompassed approximately 36,000 patient-years. Data from the more recent years were overrepresented. Information on all-cause mortality was collected from the Cause of Death Register, held by the Swedish National Board of Health and Welfare. The register linkage was performed with the unique person identification number as the key. The study was approved by the Local Ethical Review Board in Gothenburg (registration number Ö 211–01). Study Design A schematic illustration of the study design is presented in Fig. 1. Information on demographics, INR control and mortality for the study population was based on data from Journalia and the Cause of Death Register. These data were used to estimate the short-term risk of death in relation to INR control as well as the average remaining life expectancy and QALYs. Hypothetical scenarios were created. In one scenario, patients of a specific age and sex were assumed to continuously have a relatively “good” INR control. In another scenario, patients
were assumed to have a transiently worsened INR control during a short time period, and thereafter the same “good” INR control as in the first scenario. The level of INR control during this 30-day period corresponded to that of an average patient of the same age and sex. By statistical modeling using hazard functions, the differences between the two scenarios in the total estimated remaining life expectancy and QALYs in a life-long perspective were estimated, i.e. the number of days of life or QALYs that could be lost due to a worsened INR control during 30 days. The analyses concerning an improved INR control were the same as described above, but performed in the opposite direction (i.e. patients with a relatively “poor” INR control were assumed to have a transiently improved INR control during 30 days). How these alterations in INR control may be achieved was beyond the scope of this study, but the scenario in which a worsened INR control was modeled could e.g. reflect a change in diet or a less rigid therapy management during a vacation period when usual everyday routines may be changed. Correspondingly, the scenario concerning an improved INR control could correspond to a situation where the dose adjustments or INR monitoring frequencies would be improved, or if patients would be treated with a medication less sensitive to drug and food interactions. The order in which these two scenarios are presented in this paper is arbitrary and is not intended to reflect a ranking of their relative assumed importance in clinical practice. Rationale for the Study Design The focus of this study was thus to assess the impact of a short and transient alteration in INR control on life expectancy and QALYs in a life-long perspective. The rationale for the choice of study design was as follows. The calculation of SDTINR and TTR require data on INR values over a period of time, during which factors other than INR control may change. These time-covarying factors may be more or less feasible to adjust for. Thus, analyses based on actually observed alterations in INR control would unavoidably demand a long observation time, and the possibility to isolate the effect of an altered INR control would be hindered. Instead, hypothetical scenarios of abrupt alterations in INR control were created and modeled statistically. Nevertheless, observed patient data were used for the calculation of the short-term risk of death in relation to the patients’ INR control, as well as the average remaining life expectancy and QALYs. Statistical Analyses Calculation of INR Control INR has a skewed distribution. For the calculation of SDTINR, INR was transformed using the inverse of the standardized normal distribution function to achieve a normally distributed variable, as described in
Fig. 1. Schematic illustration of the study design. Information on demographics, INR control and mortality for the study population was based on data from Journalia and the Cause of Death Register. In the upper scenario, patients were assumed to continuously have a relatively “good” INR control. In the lower scenario, patients were assumed to have a transiently worsened INR control corresponding to that of an average patient of the same age and sex during a short time period, and thereafter the same “good” INR control as in the upper scenario. By statistical modeling, the differences between the two scenarios in the total estimated remaining life expectancy and QALYs in a life-long perspective were estimated. Analyses concerning the scenario of an improved INR control were also performed, but in the opposite direction (i.e. patients with a relatively “poor” INR control were assumed to have a transiently improved INR control during 30 days). Abbreviations: INR, international normalized ratio; QALYs, quality-adjusted life years.
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
1063
detail previously [16]. TTR was calculated via linear interpolation of INR measurements and the time between these, using the Rosendaal method [17]. For every new measurement of INR, SDTINR and TTR were updated. If the total number of INR measurements exceeded 19 (corresponding to approximately one year), the first ones were deleted from the calculation so that no more than the 19 most recent INR measurements were used. Each patient was followed from one INR measurement to the next or to death, if either occurred within 7 weeks (a common maximum interval between monitoring visits in the study population). After these 7 weeks, the patient was censored until a new INR measurement occurred, which was then used as a starting point for a new period. For SDTINR, there is currently no established cut-off for a good or poor INR control. To avoid using a cut-off with unknown validity and also to increase the comparability of the results between TTR and SDTINR, patients were instead categorized into percentiles based on their level of INR control. The association between SDTINR and TTR is inverse, i.e. a high quality of INR control corresponds to low SDTINR-values and high TTR-values. The values for TTR were inverted so that the lower percentiles correspond to a high quality of INR control for both SDTINR and TTR.
This study estimated the short-term risk of death, and thereby assumed that the INR level only influenced the risk of death during the following 30-day period. However, in practice, the level of INR control may influence the risk of death after the 30-day period as well. Thus, any residual benefit or disadvantage before or after the alteration in INR control was not captured in this study. Still, the consequences of the altered risk of death during this short period on the remaining days of life and QALYs were based on the patients' entire remaining life expectancy. Thus, the number of life days lost (or gained) could be substantial, even if the probability of dying was low (within 30 days). A more detailed description of these calculations, including the hazard functions, is available in the online supplement.
Modeling the Impact of a Transiently Altered INR Control on Remaining Life Expectancy and QALYs Based on data from Journalia and the Cause of Death Register, the risk (hazard) of death within 30 days was calculated according to the patients’ age, sex and INR control. The hazard function was in the form exp(β0 + β1 * current age + β2 * sex + β3 * Z), i.e. the same as in Lind et al. [16]. The variable Z was equal to either TTR or SDTINR, in both cases a time-varying covariate. These hazard functions thus yielded estimates of the 30-day risk of death for each value of TTR and SDTINR. The remaining life expectancy among patients with AF and warfarin treatment was then estimated based on integrals of hazard functions, using data from Journalia and the Cause of Death Register. These calculations are described in more detail in the online supplement. Briefly, the functions consider the risk of death at each point in time and the probability of having survived until that point in time, summed over the patients’ entire life span. These functions thus enabled the estimation of the remaining life expectancy according to age, sex and INR control. The impact of a short alteration in the level of INR control on the estimated remaining life expectancy in a life-long perspective was then modeled as follows: In one hypothetical scenario, patients of a specific age and sex were assumed to continuously have a “good” INR control, and their remaining life expectancy was estimated. In a hypothetical scenario of a transiently worsened INR control, the INR control during the first 30 days was assumed to correspond to that of an average patient of the same age and sex (i.e. a poorer INR control). Thereafter, the INR control was assumed to be at the same level as in the scenario of a continuously “good” INR control. The remaining life expectancy in this scenario was estimated. The difference between these two scenarios in the estimated remaining life expectancy was then assessed, and this difference thus corresponded to the number of days of life that could be lost due to the 30-day long period with a worsened INR control. To estimate the remaining QALYs instead of life expectancy, age- and sex adjusted utilities were applied. The utility data were based on a previous Swedish survey of 11,698 randomly selected individuals from the general population [18]. Due to the lack of adequate data, the applied utility values were not adjusted for the presence of AF and warfarin treatment. An annual discount rate of 3.5% was applied in the calculation of QALYs. The analyses concerning an improved INR control were the same as described above, but performed in the opposite direction, i.e. patients with a relatively poor INR control were assumed to have an INR control level corresponding to an average warfarin-treated AF patient of the same age and sex during 30 days, and the impact on the remaining life expectancy and QALYs were estimated.
Characteristics of the Study Population
Statistical Software Statistical analyses were performed using software written in Basic. The software is based on the same structure and has the same constructor (co-author Professor Anders Odén) as the statistical application in FRAX® (www.shef.ac.uk/FRAX), developed by the World Health Organization. Results
As described in the Methods, the study population included 19,177 individuals. The mean (SD) age was 73.4 (6.1) years, and 59% were men. The average 70-year old man had an SDTINR of 0.86 and a TTR of 62%. The included patients were followed for a mean of 1.9 years. Table 1 shows the estimated remaining life expectancy and QALYs in the study population according to age and sex. This was estimated for all ages, but is presented for a selection of ages. Categorization of Patients According to their INR Control Patients were categorized into percentiles based on their INR control, as determined by SDTINR and TTR, respectively. Fig. 2 shows the values for SDTINR and TTR according to each percentile. The 99th percentile thus corresponds to the 1% of patients with the worst INR control according to SDTINR and TTR, respectively. Impact of a Transiently Altered INR Control During 30 Days Fig. 3 shows the expected impact on days of life among 70-year old men in each percentile due to a hypothetical 30-day long alteration in INR control corresponding to that of an average 70-year old man. This average patient is thus found in the percentile where the difference in days of life equals zero (by definition). Patients below this percentile would lose days of life if their INR control would transiently worsen to the INR control level of an average patient. Correspondingly, patients above that percentile would gain days of life. The figure also shows that the magnitude of the lost (or gained) days of life would be larger Table 1 Estimated remaining life years and QALYs in the study population. Remaining life years
Remaining QALYs
Age (years)
Men
Women
Men
Women
20 30 40 50 60 70 80
53.0 43.5 34.3 25.8 18.2 11.8 7.1
59.4 49.7 40.3 31.3 23.0 15.8 10.0
20.1 17.8 15.1 12.2 9.2 6.2 4.1
19.8 17.7 15.2 12.3 9.2 6.5 4.6
The remaining life expectancy and QALYs were estimated for all ages, but are presented here only for a selection of ages. Abbreviation: QALYs, quality-adjusted life years
1064
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
control would correspond to that among an average warfarin-treated patient with AF of the same age and sex.
Fig. 2. Level of INR control for patients categorized into percentiles 1 to 99 for SDTINR and TTR, respectively. As shown in the figure, higher percentiles correspond to poorer INR control. Abbreviations: INR, international normalized ratio; SDTINR, standard deviation of transformed international normalized ratio; TTR, time in therapeutic range.
when SDTINR was used to determine INR control, as compared to TTR. The corresponding curves for QALYs would have a similar shape as the curves concerning life expectancy, and are therefore not presented. Worsened INR Control As stated above, the results regarding the magnitudes of the expected loss in days of life due to a transiently worsened INR control during 30 days are illustrated for all percentiles of 70-year old men in Fig. 3. Results concerning two examples of patient percentiles are described here, namely those with the 20th and the 2.5th best INR control, respectively. When SDTINR was used to determine INR control, a 70-year old man with the 20th best INR control (i.e. on the 20th percentile) would lose 7.4 days of life or 0.0100 QALYs due to the 30-day long period with a worsened INR control to that of an average 70-year old man with AF and warfarin. When TTR was used to determine INR control, the corresponding loss would be 3.7 days or 0.0051 QALYs. The 70-year old men with the 2.5th best INR control according to SDTINR would lose 15.5 days of life or 0.0196 QALYs. The corresponding loss according to TTR would be 6.6 days or 0.0089 QALYs. To illustrate how the results vary according to age, Fig. 4 shows the expected loss in days of life and QALYs if patients with the 2.5th best INR control according to SDTINR and TTR, respectively, were assumed to have a worsened INR control during 30 days. The worsened INR
Fig. 3. Expected gains and losses in days of life due to a 30-day long alteration in INR control. Patients were categorized into percentiles based on their INR level; higher percentiles correspond to poorer INR control. The figure shows the results for 70-year old men in each percentile. The level of INR control during the 30-day period corresponded to that of an average 70-year old warfarin-treated man with AF. By definition, this average patient is found in the percentile where the difference in days of life equals zero. Abbreviations: AF, atrial fibrillation; INR, international normalized ratio; SDTINR, standard deviation of transformed international normalized ratio; TTR, time in therapeutic range.
Improved INR Control The results regarding the magnitudes of the expected gains in days of life due to a transiently improved INR control during 30 days are illustrated for all percentiles of 70-year old men in Fig. 3. When SDTINR was used to determine INR control, a 70-year old man with the 20th worst INR control (i.e. on the 80th percentile) would gain 4.0 days of life or 0.0059 QALYs due to the 30-day long period with an improved INR control to that of an average 70-year old man with AF and warfarin. When TTR was used to determine INR control, the corresponding gain would be 0.6 days or 0.0009 QALYs. The 70-year old men with the 2.5th worst INR control according to SDTINR would gain 10.8 days of life or 0.0168 QALYs. The corresponding gain according to TTR would be 3.4 days or 0.0049 QALYs. Fig. 5 illustrates how the results vary according to age, and shows the expected gain in days of life and QALYs if patients with the 2.5th worst INR control according to SDTINR and TTR, respectively, were assumed to have an improved INR control during 30 days. The improved INR control would correspond to that among an average warfarin-treated patient with AF of the same age and sex. Discussion This study has illustrated that a worsened INR control during 30 days only can be expected to lead to lost days of life and QALYs in a life-long perspective among warfarin-treated patients with AF. This scenario could reflect a change in diet or a less rigid therapy management during a vacation period. A 70-year old man with the 20th best INR control according to SDTINR could be expected to lose 7.4 days of life or 0.0100 QALYs if his INR control would worsen during 30 days to that of an average 70-year old warfarin-treated man with AF. Considering that the INR control was assumed to be worsened during 30 days only, and that the worsened control was to that of an average patient, as opposed to an extremely poorly controlled patient, the magnitudes of the expected loss in days of life and QALYs are quite substantial. This study further illustrated that even a short and transient period of an improved INR control may have an impact on remaining days of life and QALYs. This scenario could e.g. reflect a situation where the dose adjustments or INR monitoring frequencies would be improved, or if patients would be treated with a medication less sensitive to drug and food interactions. For 70-year old men with the 20th worst INR control according to SDTINR, a 30-day improvement to that of an average 70-year old man was estimated to yield an additional 4.0 days of life or 0.0059 QALYs. Again, given that the INR control was assumed to be improved only during 30 days and only to the level of an average patient, the magnitudes are relatively large. No previous study has estimated the impact on life expectancy or QALYs in relation to a worsened or improved INR control as determined by SDTINR, to our knowledge. There are a few previous studies analyzing the impact of an improved INR control which apply a similar approach using TTR. Rose et al. (2011) simulated the number of adverse events that could be avoided and the QALYs that could be gained if the TTR would be improved in a population of 67,077 AF patients on anticoagulation [19]. The adverse events included ischemic stroke, major hemorrhage and all-cause mortality. An increase in TTR by 5% gave rise to a gain of 863 QALYs in their total population over a 2 year horizon, which could be converted to an average gain of 0.0129 QALYs per patient. Regier et al. (2006) compared self-managed to physicianmanaged anticoagulation control and modeled the impact on QALYs [20]. TTR was assumed to be 72% in the self-managed group and 63% in the physician-managed group. The adverse events included thrombosis, minor and major hemorrhage, death following major thrombotic event and death following major hemorrhagic event. Over a 5-year period, the average gain in QALYs in the self-managed group compared to
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
1065
Fig. 4. Expected differences in days of life and QALYs among patients with the 2.5th best INR control due to a 30-day long worsened INR control to the INR control corresponding to that of an average warfarin-treated AF patient of the same age and sex. Abbreviations: AF, atrial fibrillation; INR, international normalized ratio; QALYs, quality-adjusted life years; SDTINR, standard deviation of transformed international normalized ratio; TTR, time in therapeutic range.
the physician-managed group was 0.07. Sorensen et al. (2009) modeled the cost-effectiveness of warfarin treatment among 70-year old AF patients using previously published rates of stroke, bleeding, adherence
etc. [21]. When a TTR of 100% was assumed for these patients, the remaining life years and QALYs in a lifetime perspective were estimated to 9.7 and 7.2, respectively. Under conditions assumed to represent
Fig. 5. Expected differences in days of life and QALYs among patients with the 2.5th worst INR control due to a 30-day long improved INR control to the INR control corresponding to that of an average warfarin-treated AF patient of the same age and sex. Abbreviations: AF, atrial fibrillation; INR, international normalized ratio; QALYs, quality-adjusted life years; SDTINR, standard deviation of transformed international normalized ratio; TTR, time in therapeutic range.
1066
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067
real-world warfarin control (TTR 48%), the corresponding figures were 9.3 life years and 6.8 QALYs. The designs differ between these previous studies and the current study, and direct comparisons of the absolute impact on life expectancy and QALYs are therefore not appropriate. Still, all of these previous studies suggest that an improved quality of INR control, albeit measured with TTR, may lead to an increase in life expectancy and QALYs. This is in accordance with the findings reported in this study. The current study also showed that the impact of a transiently altered INR control on life expectancy and QALYs was more pronounced when SDTINR was used to categorize patients according to their quality of INR control, than when TTR was used.
Acknowledgments
Strengths and Limitations
Appendix A. Supplementary Description
This study was based on nationally representative data including detailed information on warfarin treatment and monitoring frequency in a large patient population. These data were used to calculate the shortterm risk of death in relation to the patients’ INR control, as well as the average remaining life expectancy and QALYs. By creating hypothetical scenarios, the impact of an altered INR control on the remaining life expectancy and QALYs was modeled statistically. The estimation of future QALYs did not incorporate the effects of events other than death, e.g. stroke or bleeding, since such analyses would be associated with major uncertainties due to the lack of reliable information on the severity of these events. Also, assumptions regarding the patients’ future INR control as well as the impact of the events on the patients’ entire remaining life course would need to be made. These uncertainties are evidently not present in the case of death, since this event occurs only once for each patient and has the same consequences in terms of future QALYs for all affected patients (i.e. 0). Furthermore, the applied utility values were not adjusted for the presence of AF and warfarin treatment, due to the lack of applicable data on such ageand sex stratified utilities. Warfarin patients in Sweden are generally considered to be wellcontrolled (according to TTR) compared to warfarin patients in other countries, which could possibly limit the generalizability to other settings. However, the findings in Lind et al. [16] show that SDTINR has a greater ability than TTR to predict the risk for clinical events. Also, Lind et al. showed that the correlation between TTR and SDTINR was weak (r = −0.27) [16]. The perception of the well-controlled patient population in Sweden might therefore be questioned. Furthermore, the data used in this study were from 1985–2000. The use of preventive cardiovascular drugs has increased and the mortality due to cardiovascular disease has decreased in Sweden during and after this period [22–26]. Also, the life expectancy in Sweden has increased somewhat in later years, and is currently higher than in many other countries [27]. The dataset encompassing years 1985–2000 may thus have produced results that are more comparable to a later period for many other countries. However, since this study assessed the impact of transient periods with an altered INR control, and since percentiles were used instead of absolute cut-off values (e.g. a TTR ≥70%) to categorize patients according to their level of INR control, the age of the dataset and the proportion of well-controlled patients in Sweden compared to other countries are of minor importance.
A supplementary description of statistical calculations in this article can be found online at http://dx.doi.org/10.1016/j.thromres.2014.03. 052.
Conclusions Even short and transient periods with an altered INR control can be expected to have an impact on remaining life expectancy and QALYs among patients with AF. The findings suggest that it is important to continuously maintain a good anticoagulation control, since even a short period with a reduced quality of control may be harmful. Furthermore, even short term efforts to improve anticoagulation control may be beneficial.
Conflict of Interest Statement This study was sponsored by Bristol-Myers Squibb and Pfizer. The authors have no other conflicts of interest to report.
This study was sponsored by Bristol-Myers Squibb and Pfizer. The sponsors had no role in the collection, analysis or interpretation of data, but provided comments on the study design and on the manuscript. The sponsors supported the decision to submit the manuscript for publication.
References [1] Ericson L, Bergfeldt L, Bjorholt I. Atrial fibrillation: the cost of illness in Sweden. Eur J Health Econ 2011;12:479–87. [2] Go AS, Hylek EM, Phillips KA, Chang Y, Henault LE, Selby JV, et al. Prevalence of diagnosed atrial fibrillation in adults: national implications for rhythm management and stroke prevention: the AnTicoagulation and Risk Factors in Atrial Fibrillation (ATRIA) Study. JAMA 2001;285:2370–5. [3] SBU – Statens beredning för medicinsk utvärdering (Swedish Council on Health Technology Assessment). Förmaksflimmer - förekomst och risk för stroke (Atrial fibrillation - prevalence and risk for stroke) [Report in Swedish]. http://www.sbu. se/upload/Publikationer/Content0/2/Formaksflimmer_forekomst_risk_for_stroke. pdf; 2013. [4] Bjorck S, Palaszewski B, Friberg L, Bergfeldt L. Atrial fibrillation, stroke risk, and warfarin therapy revisited: a population-based study. Stroke 2013;44:3103–8. [5] Hart RG, Pearce LA, Aguilar MI. Meta-analysis: antithrombotic therapy to prevent stroke in patients who have nonvalvular atrial fibrillation. Ann Intern Med 2007;146:857–67. [6] Wolf PA, Abbott RD, Kannel WB. Atrial fibrillation as an independent risk factor for stroke: the Framingham Study. Stroke 1991;22:983–8. [7] Summary of product characteristics: Warfarin. Available from the Swedish Medical Products Agency http://www.lakemedelsverket.se. [8] Frykman V, Beerman B, Ryden L, Rosenqvist M. Management of atrial fibrillation: discrepancy between guideline recommendations and actual practice exposes patients to risk for complications. Eur Heart J 2001;22:1954–9. [9] Bradley BC, Perdue KS, Tisdel KA, Gilligan DM. Frequency of anticoagulation for atrial fibrillation and reasons for its non-use at a Veterans Affairs medical center. Am J Cardiol 2000;85:568–72. [10] Ogilvie IM, Newton N, Welner SA, Cowell W, Lip GY. Underuse of oral anticoagulants in atrial fibrillation: a systematic review. Am J Med 2010;123:638–45. [11] Friberg L, Hammar N, Ringh M, Pettersson H, Rosenqvist M. Stroke prophylaxis in atrial fibrillation: who gets it and who does not? Report from the Stockholm Cohort-study on Atrial Fibrillation (SCAF-study). Eur Heart J 2006;27:1954–64. [12] Baczek VL, Chen WT, Kluger J, Coleman CI. Predictors of warfarin use in atrial fibrillation in the United States: a systematic review and meta-analysis. BMC Fam Pract 2012;13:5. [13] Oden A, Fahlen M. Oral anticoagulation and risk of death: a medical record linkage study. BMJ 2002;325:1073–5. [14] Gallagher AM, Setakis E, Plumb JM, Clemens A, van Staa TP. Risks of stroke and mortality associated with suboptimal anticoagulation in atrial fibrillation patients. Thromb Haemost 2011;106:968–77. [15] Wan Y, Heneghan C, Perera R, Roberts N, Hollowell J, Glasziou P, et al. Anticoagulation control and prediction of adverse events in patients with atrial fibrillation: a systematic review. Circ Cardiovasc Qual Outcomes 2008;1:84–91. [16] Lind M, Fahlen M, Kosiborod M, Eliasson B, Oden A. Variability of INR and its relationship with mortality, stroke, bleeding and hospitalisations in patients with atrial fibrillation. Thromb Res 2012;129:32–5. [17] Rosendaal FR, Cannegieter SC, van der Meer FJ, Briet E. A method to determine the optimal intensity of oral anticoagulant therapy. Thromb Haemost 1993;69:236–9. [18] Burstrom K, Johannesson M, Diderichsen F. Health-related quality of life by disease and socio-economic group in the general population in Sweden. Health Policy 2001;55:51–69. [19] Rose AJ, Berlowitz DR, Ash AS, Ozonoff A, Hylek EM, Goldhaber-Fiebert JD. The business case for quality improvement: oral anticoagulation for atrial fibrillation. Circ Cardiovasc Qual Outcomes 2011;4:416–24. [20] Regier DA, Sunderji R, Lynd LD, Gin K, Marra CA. Cost-effectiveness of self-managed versus physician-managed oral anticoagulation therapy. CMAJ 2006;174:1847–52. [21] Sorensen SV, Dewilde S, Singer DE, Goldhaber SZ, Monz BU, Plumb JM. Costeffectiveness of warfarin: trial versus "real-world" stroke prevention in atrial fibrillation. Am Heart J 2009;157:1064–73.
E. Lesén et al. / Thrombosis Research 133 (2014) 1061–1067 [22] Wandell PE. Drug use in patients with atrial fibrillation in Swedish primary health care: a comparison 5 years apart. Eur J Clin Pharmacol 1999;55:333–7. [23] Walley T, Folino-Gallo P, Stephens P, Van Ganse E. Trends in prescribing and utilization of statins and other lipid lowering drugs across Europe 1997–2003. Br J Clin Pharmacol 2005;60:543–51. [24] Carlsson AC, Wandell P, Sundquist K, Johansson SE, Sundquist J. Differences and time trends in drug treatment of atrial fibrillation in men and women and doctors'
1067
adherence to warfarin therapy recommendations: A Swedish study of prescribed drugs in primary care in 2002 and 2007. Eur J Clin Pharmacol 2013;69:245–53. [25] Silwer L, Lundborg CS. Patterns of drug use during a 15 year period: data from a Swedish county, 1988–2002. Pharmacoepidemiol Drug Saf 2005;14:813–20. [26] National Board of Health and Welfare. Official statistics of Sweden. Statistics – Health and Medical Care. Causes of Death 2010; 2011. [27] World Health Organization. World health statistics; 2012.
