552819
research-article2014
AOPXXX10.1177/1060028014552819Annals of PharmacotherapyDesai et al
Research Report
Comparative Persistence on β-Blockers Versus Calcium Channel Blockers for Ventricular Rate Control in Nonelderly Patients With Atrial Fibrillation
Annals of Pharmacotherapy 1–10 © The Author(s) 2014 Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/1060028014552819 aop.sagepub.com
Vibha C. A. Desai1, Christina M. L. Kelton, PhD2, Anne H. Metzger, PharmD1, Teresa M. Cavanaugh, PharmD1, Jeff J. Guo, PhD1, and Pamela C. Heaton, PhD1
Abstract Background: For patients with atrial fibrillation (AF), early treatment is essential to prevent serious complications such as stroke. Several randomized clinical trials have shown that rate-control may be as effective as rhythm-control medications, whereas the latter have serious side effects. Little evidence exists, however, about which class of rate-control medication— β-blockers (BBs) or calcium channel blockers (CCBs)—may be superior. Objective: The objective was to compare the long-term persistence on BBs versus CCBs in nonelderly adult patients with AF. Methods: A longitudinal retrospective cohort study for patients 40 to 60 years old with newly diagnosed AF (identified by ICD-9 code 427.31) was performed using data from Ohio Medicaid physician, institutional, and pharmacy claims from January 2006 through June 2011. A Cox proportional hazard regression, with time to change out of rate-control therapy as the dependent variable, was estimated to compare persistence on (proxy for effectiveness of) rate-control medication across drug classes. A propensity-score analysis was used to control for selection bias. Additional covariates included age, development of heart failure, and medication adherence. Results: Out of 1239 patients included in the cohort, 1016 received a BB; 223 received a CCB. Over time, patients on CCBs were significantly more likely to switch out of rate-control therapy (hazard ratio = 1.89; 95% CI = 1.14-3.09) than patients on BBs. Conclusions: Evidence suggests that nonelderly AF patients, when prescribed ratecontrol therapy, persist longer on BBs than CCBs. Because this is the first long-term study comparing the 2 drug classes in the nonelderly population, further research is suggested. Keywords atrial fibrillation, comparative effectiveness, persistence, rate-control drugs, β-blockers, calcium channel blockers, rhythmcontrol medications, Medicaid claims data
Introduction Approximately 600 000 adults younger than 65 years, in the United States, suffer from atrial fibrillation (AF),1-4 the most common form of cardiac arrhythmia, characterized by rapid and disorganized atrial activation.4 Because nonelderly adults potentially face a number of decades living with this chronic disease, early treatment for AF is essential to discourage serious complications that develop over time from an uncontrolled heart rate. It is known that 15% of all strokes in the United States are caused by AF.5 A patient with AF is 4 times more likely to have a stroke, and he or she is twice as likely to die.6-8 An uncontrolled heart rate in AF results in symptoms such as shortness of breath, dizziness, light-headedness, weakness, and chest pain and can lead to cardiomyopathy and heart failure (HF), both of which are severely disabling conditions.9 With the
symptoms that may accompany AF, patients are likely to experience comorbidities like anxiety and depression, which may significantly reduce quality of life and lead to lost productivity at work.10 Treatment for AF depends on a number of factors, including how long the patient has had the disease, how severe the symptoms are, and the underlying cause of AF. For some patients, such as those with lone AF or those who are severely symptomatic, it may be recommended that the 1
University of Cincinnati Academic Health Center, Cincinnati, OH, USA University of Cincinnati, Cincinnati, OH, USA
2
Corresponding Author: Christina M. L. Kelton, University of Cincinnati, Carl H. Lindner College of Business, 414 Lindner Hall, 2925 Campus Green Drive, Cincinnati, OH 45221, USA. Email: [email protected]
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on October 20, 2014
2
Annals of Pharmacotherapy
heart’s rhythm be restored either through electrical cardioversion or pharmacologically by the blocking of sodium or potassium channels.11 However, rhythm-control drugs are associated with serious side effects such as organ toxicities; patients taking these drugs require frequent clinical monitoring.12 Electrical cardioversion carries a risk for thromboembolism, ventricular proarrhythmia, and sinus node arrest.13,14 Other patients are treated using a rate-control therapy, the purpose of which is to slow the rapid heart rate associated with AF. Pharmacological agents used for rate control include digoxin, β-blockers (BBs), and calcium channel blockers (CCBs). Although generally better tolerated than antiarrhythmic drugs, BBs and CCBs are certainly not free of potential side effects, nor do they promise a long-term solution to AF.10 A number of randomized clinical trials (RCTs), such as PIAF, RACE, AFFIRM, STAF, and HOT-CAFÉ, in the early 2000s,15-19 showed no survival advantage of rhythmcontrol drugs or electrical cardioversion over rate-control drugs. They also showed no difference between the 2 drug strategies in stroke rate and noninferiority of rate-control relative to rhythm-control drugs for prevention of morbidity. Evidence from these RCTs led to a more prominent role for rate-control therapy in the 2006 Guidelines for the Management of Patients With Atrial Fibrillation.20 Clinical practice changed over the decade 2001-2010 to favor rate-control medications. In a study based on 2 national outpatient visit survey databases, the percentage of AF patient visits mentioning only a rate-control medication was found to trend upward (P = 0.074) from 41.9% (95% CI = 34.1%-49.7%) in 2001 to 47.3% (95% CI = 38.3%56.2%) in 2010, whereas the percentage of visits mentioning both rhythm- and rate-control drugs had a statistically significant upward trend (P = 0.015) as well from 3.1% (95% CI = 1.2%-5.0%) to 12.5% (95% CI = 7.9%-17.1%).21 Regardless of the rhythm-versus-rate-control decision, there is very limited clinical evidence on the choice between a BB and CCB if rate control is the selected first-line treatment. A crossover open-label study of 5 drug regimens consisting of atenolol monotherapy, diltiazem monotherapy, their combinations with digoxin, and digoxin alone found that the atenolol plus digoxin combination produced the most effective 24-hour ventricular rate control.22 A randomized crossover study comparing low-dose diltiazem plus digoxin with low-dose betaxolol plus digoxin showed that the latter was superior in controlling ventricular rate.23 The only long-term study, which used the AFFIRM trial data, found that adequate rate control was achieved over time in a significantly higher percentage of patients treated with a BB alone than with a CCB alone.24 Because only elderly patients were enrolled in the AFFIRM trial, the results from the study are suggestive only of what might be expected for younger patients. Thus, a gap in the literature and in clinical care on the long-term comparative
effectiveness of BBs versus CCBs, especially for nonelderly patients, exists. Medicaid is the largest public health care payer for nonelderly adults in the United States. Medicaid data provide a detailed, longitudinal record of patients’ health care use, diagnoses, procedures, and drug prescriptions from health care claims. Because we cannot measure effectiveness directly and clinically using claims data, the objective of this study was to evaluate the comparative persistence (approximating effectiveness) on BBs versus CCBs in nonelderly adults with AF. Persistence was defined as the length of time a patient remained on rate-control medication before switching to a rhythm-control medication, undergoing electrical cardioversion, catheter ablation, or pacemaker implantation. This study is the first to compare the longterm persistence on BBs versus CCBs in nonelderly adults with AF.
Methods Data Data were taken from inpatient, physician, and prescription claims for Ohio Medicaid managed-care and fee-for-service beneficiaries from January 1, 2006, through June 30, 2011.
Cohort Selection Patients were included if they met the following 4 criteria: (1) they were between the ages of 40 and 60 years (inclusive) at the time of first diagnosis of (first claim for) AF (ICD-9 code 427.31) anytime between July 1, 2006, and June 30, 2010 (allowing at least 1 year of follow-up until June 30, 2011, for each patient); we refer to the first AF claim date as the AF incident date; (2) they had at least 1 claim for a BB or CCB (identified by generic drug codes given in Appendix A) between the incident date and June 30, 2010 (again allowing 1 year of follow up), with the first prescription claim date called the drug assignment date (assignment to BB or CCB); (3) they were continuously enrolled in Medicaid for 6 months prior to their incident date and during the time period between their incident and assignment dates; and (4) they had no diagnosis of AF anytime in the 6 months prior to their incident date. These criteria used in patient selection allowed the formation of BB and CCB subcohorts (study arms) of younger patients newly diagnosed with AF. Patients were excluded if they had had at least 1 claim for a rhythm-control medication (medications listed in Appendix A) or at least 1 claim for electrical cardioversion (ICD-9 procedure code 99.61, 99.62, or 99.69) in the 6 months prior to their assignment date. Patients with claims for both a BB and a CCB within a window of 15 days before and after their assignment date were excluded because they could not be clearly assigned to one study arm or the other.
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on October 20, 2014
3
Desai et al To exclude patients with paroxysmal AF who may not require continuous medication, all patients who had only a single claim (the one from their assignment date) for an AF drug during the first 12 months following their assignment date were excluded. Patients who had been diagnosed with HF (ICD-9 428.xx), cardiomyopathy (ICD-9 425.xx), coronary artery disease (ICD-9 414.0), or myocardial infarction (ICD-9 410.xx or 412.xx) or who had undergone coronary artery bypass surgery or a revascularization procedure (ICD-9 procedure code 36.1x, 36.2x, or 36.3x) in the 6 months prior to their assignment date were excluded as well to avoid selection bias uncontrollable by propensity-score analysis.
Treatment Persistence We measured treatment persistence as the length of time that the patient remained on rate-control medication. Specifically, to measure persistence, we measured time from assignment date to the date of the first claim for one of the following events: a prescription for a rhythm-control medication, electrical cardioversion, catheter ablation (ICD-9 procedure code 37.34), or pacemaker implantation (ICD-9 procedure codes 37.80 - 37.83). We did not consider a switch from BB to CCB or CCB to BB as an event—that is, the patient was still considered persistent following such a switch between rate-control drugs. Censoring events included death and disenrollment from Medicaid.
Covariates Baseline characteristics included age at assignment date; gender; diagnosis of a pulmonary disease, including acute or chronic pulmonary heart disease, diseases of pulmonary circulation, obstructive chronic bronchitis, emphysema, asthma, chronic airway obstruction not elsewhere classified (ICD-9 415.xx-417.xx, 491.2, 492.xx, 493.xx, or 496.xx), hepatic disease (ICD-9 570.xx-576.xx), renal disease (ICD-9 580.xx or 598.xx), hypertension (ICD-9 401. xx-405.xx), diabetes (ICD-9 249.xx or 250.xx), or angina pectoris (ICD-9 413.xx) in the 6 months before the assignment date; a claim for a cardiovascular drug (a statin, antihypertensive, or diuretic identified by therapeutic class codes given in Appendix A) within 6 months preassignment; a claim for a rate-control drug (BB, CCB, or digoxin) within 6 months preassignment; and Medicaid patient status (managed care or fee for service). A majority of these variables were considered in the AFFIRM-based study.24 Other covariates were (1) diagnosis of HF postassignment, (2) switching to or adding another rate-control drug (BB, CCB, or digoxin) postassignment (up to 15 days prior to any study event to rule out drug initiation concurrent with that event), and (3) degree of adherence to therapy. Adherence was assessed by a medication possession ratio (MPR) calculated as the total number of days of supply
between the first claim and last claim for the medication minus the number of days of supply for the last claim, divided by the time elapsed between the first claim and last claim.25,26 An MPR ≤0.8 was considered low adherence.27-29 Although other medication adherence measures, such as the proportion of days covered (PDC), have also been advocated in the literature,30 an article published earlier in this journal found that the MPR and PDC produced almost identical adherence estimates for the situation the authors considered.31
Correcting for Selection Bias A logistic regression model, with treatment assignment (BB or CCB) as the dependent variable and the baseline patient characteristics as independent variables, was estimated to obtain propensity scores for the patients in the cohort. Patients were then stratified into propensity-score quintiles. Pairwise tests for equality between the distribution of baseline characteristics within quintiles between the 2 study arms were run to ensure that the quintile grouping adequately controlled for selection bias.32
Time-to-Event Analysis Kaplan-Meier curves were generated from the time-toevent data for each of the study arms. A log-rank test was used to determine statistical difference between the curves.33 Using a Cox proportional hazard model, time to event (persistence on rate control) was regressed on study-arm assignment (CCB or BB), a subset of the baseline patient characteristics (all but drug claims for statins, antihypertensives, and diuretics for parsimony, and gender and pulmonary disease because of detected multicollinearity with propensity-score quintiles), the postassignment factors discussed above, and the propensity quintiles. Additional statistical tests were run to ensure correct model specification.34 All analyses were conducted using SAS version 9.2 and R version 2.14.2. The institutional review board at the University of Cincinnati approved the study protocol.
Results Cohort Selection A total of 3436 patients met all inclusion criteria. After the exclusion of 2197 patients for the reasons delineated above, the study cohort consisted of 1239 patients.
Events As Table 1 shows, the median follow-up period was 19.4 months. Over time, 91 (7.34%) patients did not persist on rate-control medication, with 63 patients switching to or
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on October 20, 2014
4
Annals of Pharmacotherapy
Table 1. Patient Events. Study Cohort
Total a
Event Number of patients Switch to or add rhythm-control drug Electrical cardioversion Catheter ablation and/or pacemaker Death Medicaid disenrollment End of study period without event Median time to event, months Median follow-up period, months
β-Blocker Arm a
Calcium-Channel-Blocker Arm
n (%)
n (%)
n (%)a
1239 (100) 63 (5.08) 15 (1.21) 13 (1.05) 0 (0.00) 604 (48.75) 544 (43.91) 7.37 19.37
1016 (100) 44 (4.33) 14 (1.38) 8 (0.79) 0 (0.00) 486 (47.84) 464 (45.67) 7.42 19.35
223 (100) 19 (8.52) 1 (0.45) 5 (2.24) 0 (0.00) 118 (52.92) 80 (35.87) 6.68 19.37
a
Percentage of column total.
adding a rhythm-control drug, 15 patients experiencing electrical cardioversion, and 13 patients undergoing catheter ablation and/or pacemaker implantation. No patient in the study died during the follow-up period. There were 1016 (82.0%) patients who were initially prescribed a BB, and the remaining 223 (18.0%) started with a CCB medication. Whereas 66 (6.5%) patients on a BB had an event (did not persist on rate-control medication), 25 (11.2%) patients on a CCB did not persist. Moreover, the median time to event for patients in the BB group versus the CCB group was longer (7.4 months vs 6.7 months; see Table 1.)
Patient Characteristics Descriptive statistics for the study cohort, stratified by study arm, are given in Table 2. Patients in the 2 arms were similar with respect to baseline characteristic percentages, with 3 exceptions: (1) patients prescribed a CCB were more likely to have pulmonary disease (54.7% vs 38.7%; P < 0.0001); (2) patients prescribed a CCB were more likely to be female (72.2% vs 57.3%; P < 0.0001); and (3) patients prescribed a CCB were less likely to have had prior ratecontrol drug use (63.2% vs 72.0%; P = 0.010). However, within propensity-score quintiles, proportions were not significantly different between study arms (nor were the proportions of any of the other baseline characteristics). (See Appendix B for propensity-score quintiles.) In Table 2, we also provide the postassignment patient characteristics stratified by study arm. During their followup period, 244 (24.0%) patients in the BB arm and 54 (24.2%) patients in the CCB arm (P = 0.949) had a new diagnosis of HF. Similarly, there was no statistically significant difference (P = 0.524) in the proportion of patients with low medication adherence between the 2 study arms (500 or 49.2% of patients in the BB arm vs 115 or 51.6% of
patients in the CCB arm). However, the proportion of CCB patients switching to or adding a BB (27.8%) was significantly (P < 0.0001) higher than the proportion of BB patients switching to or adding a CCB (7.6%). After the assignment date, 21 (33.9%) of the patients who switched to or added a BB in the CCB arm were diagnosed with HF. Similarly, 13.5% of patients in the CCB arm versus 9.7% of the patients in the BB arm (P = 0.091) had a switch to or addition of digoxin during the follow-up.
Persistence Patterns Figure 1 depicts the Kaplan-Meier curves for the 2 study arms. Consistent with the greater percentage of events and shorter median time to event for the CCB arm versus BB arm patients, the CCB persistence curve lies consistently below (P = 0.022) the BB curve. Whereas, at 60 months postassignment, 87.5% of BB patients remained on a rate-control drug regimen, that percentage for the CCB patients was 78.4%. To ensure that the results in Figure 1 were not affected by intra– drug-class switching, another set of Kaplan-Meier curves was derived for the 1100 patients with no switching. Exactly the same pattern was seen as in Figure 1; percentages at 60 months were the same as in Figure 1, and again, curve equality was rejected (P = 0.018). (See Appendix C.) Results from the Cox proportional hazard regression in Table 3 show that patients who received a CCB initially were 1.89 (95% CI = 1.14-3.08) times more likely to experience one of the study events (ie, less likely to persist on rate-control therapy) compared with patients who initially received a BB. The only other statistically significant coefficient was that for renal disease. Patients with renal disease were significantly less likely to have an event (hazard ratio = 0.44; CI = 0.22-0.89)—that is, more likely to persist with rate control. All statistical tests showed good model specification.
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on October 20, 2014
5
Desai et al Table 2. Patient Characteristics. Study Cohort
Total
β-Blocker Arm
Calcium-Channel-Blocker Arm
Characteristic
n (%)a
n (%)a
n (%)a
P Valueb
Number of patients Preassignment characteristics Age, years ± standard deviation Female Pulmonary disease Hepatic disease Renal disease Hypertension Diabetes Angina pectoris Statin use Antihypertensive use Diuretic use Prior use of rate-control drug Medicaid managed-care plan Postassignment characteristics Diagnosis with heart failure Low medication adherence Rate-control class switch or addition Switch to or addition of digoxin
1239 (100)
1016 (100)
223 (100)
51.6 ± 5.8 743 (59.97) 515 (41.57) 116 (9.36) 249 (20.10) 924 (74.58) 504 (40.68) 88 (7.10) 537 (43.34) 549 (44.31) 464 (37.45) 872 (70.38) 500 (40.36)
51.6 ± 5.9 582 (57.28) 393 (38.68) 100 (9.84) 211 (20.77) 761 (74.90) 419 (41.24) 73 (7.19) 453 (44.59) 455 (44.78) 371 (36.52) 731 (71.95) 413 (40.65)
51.4 ± 5.7 161(72.20) 122 (54.71) 16 (7.17) 38 (17.04) 163 (73.09) 85 (38.12) 15 (6.73) 84 (37.67) 94 (42.15) 93 (41.70) 141 (63.23) 87 (39.01)
0.737
research-article2014
AOPXXX10.1177/1060028014552819Annals of PharmacotherapyDesai et al
Research Report
Comparative Persistence on β-Blockers Versus Calcium Channel Blockers for Ventricular Rate Control in Nonelderly Patients With Atrial Fibrillation
Annals of Pharmacotherapy 1–10 © The Author(s) 2014 Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/1060028014552819 aop.sagepub.com
Vibha C. A. Desai1, Christina M. L. Kelton, PhD2, Anne H. Metzger, PharmD1, Teresa M. Cavanaugh, PharmD1, Jeff J. Guo, PhD1, and Pamela C. Heaton, PhD1
Abstract Background: For patients with atrial fibrillation (AF), early treatment is essential to prevent serious complications such as stroke. Several randomized clinical trials have shown that rate-control may be as effective as rhythm-control medications, whereas the latter have serious side effects. Little evidence exists, however, about which class of rate-control medication— β-blockers (BBs) or calcium channel blockers (CCBs)—may be superior. Objective: The objective was to compare the long-term persistence on BBs versus CCBs in nonelderly adult patients with AF. Methods: A longitudinal retrospective cohort study for patients 40 to 60 years old with newly diagnosed AF (identified by ICD-9 code 427.31) was performed using data from Ohio Medicaid physician, institutional, and pharmacy claims from January 2006 through June 2011. A Cox proportional hazard regression, with time to change out of rate-control therapy as the dependent variable, was estimated to compare persistence on (proxy for effectiveness of) rate-control medication across drug classes. A propensity-score analysis was used to control for selection bias. Additional covariates included age, development of heart failure, and medication adherence. Results: Out of 1239 patients included in the cohort, 1016 received a BB; 223 received a CCB. Over time, patients on CCBs were significantly more likely to switch out of rate-control therapy (hazard ratio = 1.89; 95% CI = 1.14-3.09) than patients on BBs. Conclusions: Evidence suggests that nonelderly AF patients, when prescribed ratecontrol therapy, persist longer on BBs than CCBs. Because this is the first long-term study comparing the 2 drug classes in the nonelderly population, further research is suggested. Keywords atrial fibrillation, comparative effectiveness, persistence, rate-control drugs, β-blockers, calcium channel blockers, rhythmcontrol medications, Medicaid claims data
Introduction Approximately 600 000 adults younger than 65 years, in the United States, suffer from atrial fibrillation (AF),1-4 the most common form of cardiac arrhythmia, characterized by rapid and disorganized atrial activation.4 Because nonelderly adults potentially face a number of decades living with this chronic disease, early treatment for AF is essential to discourage serious complications that develop over time from an uncontrolled heart rate. It is known that 15% of all strokes in the United States are caused by AF.5 A patient with AF is 4 times more likely to have a stroke, and he or she is twice as likely to die.6-8 An uncontrolled heart rate in AF results in symptoms such as shortness of breath, dizziness, light-headedness, weakness, and chest pain and can lead to cardiomyopathy and heart failure (HF), both of which are severely disabling conditions.9 With the
symptoms that may accompany AF, patients are likely to experience comorbidities like anxiety and depression, which may significantly reduce quality of life and lead to lost productivity at work.10 Treatment for AF depends on a number of factors, including how long the patient has had the disease, how severe the symptoms are, and the underlying cause of AF. For some patients, such as those with lone AF or those who are severely symptomatic, it may be recommended that the 1
University of Cincinnati Academic Health Center, Cincinnati, OH, USA University of Cincinnati, Cincinnati, OH, USA
2
Corresponding Author: Christina M. L. Kelton, University of Cincinnati, Carl H. Lindner College of Business, 414 Lindner Hall, 2925 Campus Green Drive, Cincinnati, OH 45221, USA. Email: [email protected]
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on October 20, 2014
2
Annals of Pharmacotherapy
heart’s rhythm be restored either through electrical cardioversion or pharmacologically by the blocking of sodium or potassium channels.11 However, rhythm-control drugs are associated with serious side effects such as organ toxicities; patients taking these drugs require frequent clinical monitoring.12 Electrical cardioversion carries a risk for thromboembolism, ventricular proarrhythmia, and sinus node arrest.13,14 Other patients are treated using a rate-control therapy, the purpose of which is to slow the rapid heart rate associated with AF. Pharmacological agents used for rate control include digoxin, β-blockers (BBs), and calcium channel blockers (CCBs). Although generally better tolerated than antiarrhythmic drugs, BBs and CCBs are certainly not free of potential side effects, nor do they promise a long-term solution to AF.10 A number of randomized clinical trials (RCTs), such as PIAF, RACE, AFFIRM, STAF, and HOT-CAFÉ, in the early 2000s,15-19 showed no survival advantage of rhythmcontrol drugs or electrical cardioversion over rate-control drugs. They also showed no difference between the 2 drug strategies in stroke rate and noninferiority of rate-control relative to rhythm-control drugs for prevention of morbidity. Evidence from these RCTs led to a more prominent role for rate-control therapy in the 2006 Guidelines for the Management of Patients With Atrial Fibrillation.20 Clinical practice changed over the decade 2001-2010 to favor rate-control medications. In a study based on 2 national outpatient visit survey databases, the percentage of AF patient visits mentioning only a rate-control medication was found to trend upward (P = 0.074) from 41.9% (95% CI = 34.1%-49.7%) in 2001 to 47.3% (95% CI = 38.3%56.2%) in 2010, whereas the percentage of visits mentioning both rhythm- and rate-control drugs had a statistically significant upward trend (P = 0.015) as well from 3.1% (95% CI = 1.2%-5.0%) to 12.5% (95% CI = 7.9%-17.1%).21 Regardless of the rhythm-versus-rate-control decision, there is very limited clinical evidence on the choice between a BB and CCB if rate control is the selected first-line treatment. A crossover open-label study of 5 drug regimens consisting of atenolol monotherapy, diltiazem monotherapy, their combinations with digoxin, and digoxin alone found that the atenolol plus digoxin combination produced the most effective 24-hour ventricular rate control.22 A randomized crossover study comparing low-dose diltiazem plus digoxin with low-dose betaxolol plus digoxin showed that the latter was superior in controlling ventricular rate.23 The only long-term study, which used the AFFIRM trial data, found that adequate rate control was achieved over time in a significantly higher percentage of patients treated with a BB alone than with a CCB alone.24 Because only elderly patients were enrolled in the AFFIRM trial, the results from the study are suggestive only of what might be expected for younger patients. Thus, a gap in the literature and in clinical care on the long-term comparative
effectiveness of BBs versus CCBs, especially for nonelderly patients, exists. Medicaid is the largest public health care payer for nonelderly adults in the United States. Medicaid data provide a detailed, longitudinal record of patients’ health care use, diagnoses, procedures, and drug prescriptions from health care claims. Because we cannot measure effectiveness directly and clinically using claims data, the objective of this study was to evaluate the comparative persistence (approximating effectiveness) on BBs versus CCBs in nonelderly adults with AF. Persistence was defined as the length of time a patient remained on rate-control medication before switching to a rhythm-control medication, undergoing electrical cardioversion, catheter ablation, or pacemaker implantation. This study is the first to compare the longterm persistence on BBs versus CCBs in nonelderly adults with AF.
Methods Data Data were taken from inpatient, physician, and prescription claims for Ohio Medicaid managed-care and fee-for-service beneficiaries from January 1, 2006, through June 30, 2011.
Cohort Selection Patients were included if they met the following 4 criteria: (1) they were between the ages of 40 and 60 years (inclusive) at the time of first diagnosis of (first claim for) AF (ICD-9 code 427.31) anytime between July 1, 2006, and June 30, 2010 (allowing at least 1 year of follow-up until June 30, 2011, for each patient); we refer to the first AF claim date as the AF incident date; (2) they had at least 1 claim for a BB or CCB (identified by generic drug codes given in Appendix A) between the incident date and June 30, 2010 (again allowing 1 year of follow up), with the first prescription claim date called the drug assignment date (assignment to BB or CCB); (3) they were continuously enrolled in Medicaid for 6 months prior to their incident date and during the time period between their incident and assignment dates; and (4) they had no diagnosis of AF anytime in the 6 months prior to their incident date. These criteria used in patient selection allowed the formation of BB and CCB subcohorts (study arms) of younger patients newly diagnosed with AF. Patients were excluded if they had had at least 1 claim for a rhythm-control medication (medications listed in Appendix A) or at least 1 claim for electrical cardioversion (ICD-9 procedure code 99.61, 99.62, or 99.69) in the 6 months prior to their assignment date. Patients with claims for both a BB and a CCB within a window of 15 days before and after their assignment date were excluded because they could not be clearly assigned to one study arm or the other.
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on October 20, 2014
3
Desai et al To exclude patients with paroxysmal AF who may not require continuous medication, all patients who had only a single claim (the one from their assignment date) for an AF drug during the first 12 months following their assignment date were excluded. Patients who had been diagnosed with HF (ICD-9 428.xx), cardiomyopathy (ICD-9 425.xx), coronary artery disease (ICD-9 414.0), or myocardial infarction (ICD-9 410.xx or 412.xx) or who had undergone coronary artery bypass surgery or a revascularization procedure (ICD-9 procedure code 36.1x, 36.2x, or 36.3x) in the 6 months prior to their assignment date were excluded as well to avoid selection bias uncontrollable by propensity-score analysis.
Treatment Persistence We measured treatment persistence as the length of time that the patient remained on rate-control medication. Specifically, to measure persistence, we measured time from assignment date to the date of the first claim for one of the following events: a prescription for a rhythm-control medication, electrical cardioversion, catheter ablation (ICD-9 procedure code 37.34), or pacemaker implantation (ICD-9 procedure codes 37.80 - 37.83). We did not consider a switch from BB to CCB or CCB to BB as an event—that is, the patient was still considered persistent following such a switch between rate-control drugs. Censoring events included death and disenrollment from Medicaid.
Covariates Baseline characteristics included age at assignment date; gender; diagnosis of a pulmonary disease, including acute or chronic pulmonary heart disease, diseases of pulmonary circulation, obstructive chronic bronchitis, emphysema, asthma, chronic airway obstruction not elsewhere classified (ICD-9 415.xx-417.xx, 491.2, 492.xx, 493.xx, or 496.xx), hepatic disease (ICD-9 570.xx-576.xx), renal disease (ICD-9 580.xx or 598.xx), hypertension (ICD-9 401. xx-405.xx), diabetes (ICD-9 249.xx or 250.xx), or angina pectoris (ICD-9 413.xx) in the 6 months before the assignment date; a claim for a cardiovascular drug (a statin, antihypertensive, or diuretic identified by therapeutic class codes given in Appendix A) within 6 months preassignment; a claim for a rate-control drug (BB, CCB, or digoxin) within 6 months preassignment; and Medicaid patient status (managed care or fee for service). A majority of these variables were considered in the AFFIRM-based study.24 Other covariates were (1) diagnosis of HF postassignment, (2) switching to or adding another rate-control drug (BB, CCB, or digoxin) postassignment (up to 15 days prior to any study event to rule out drug initiation concurrent with that event), and (3) degree of adherence to therapy. Adherence was assessed by a medication possession ratio (MPR) calculated as the total number of days of supply
between the first claim and last claim for the medication minus the number of days of supply for the last claim, divided by the time elapsed between the first claim and last claim.25,26 An MPR ≤0.8 was considered low adherence.27-29 Although other medication adherence measures, such as the proportion of days covered (PDC), have also been advocated in the literature,30 an article published earlier in this journal found that the MPR and PDC produced almost identical adherence estimates for the situation the authors considered.31
Correcting for Selection Bias A logistic regression model, with treatment assignment (BB or CCB) as the dependent variable and the baseline patient characteristics as independent variables, was estimated to obtain propensity scores for the patients in the cohort. Patients were then stratified into propensity-score quintiles. Pairwise tests for equality between the distribution of baseline characteristics within quintiles between the 2 study arms were run to ensure that the quintile grouping adequately controlled for selection bias.32
Time-to-Event Analysis Kaplan-Meier curves were generated from the time-toevent data for each of the study arms. A log-rank test was used to determine statistical difference between the curves.33 Using a Cox proportional hazard model, time to event (persistence on rate control) was regressed on study-arm assignment (CCB or BB), a subset of the baseline patient characteristics (all but drug claims for statins, antihypertensives, and diuretics for parsimony, and gender and pulmonary disease because of detected multicollinearity with propensity-score quintiles), the postassignment factors discussed above, and the propensity quintiles. Additional statistical tests were run to ensure correct model specification.34 All analyses were conducted using SAS version 9.2 and R version 2.14.2. The institutional review board at the University of Cincinnati approved the study protocol.
Results Cohort Selection A total of 3436 patients met all inclusion criteria. After the exclusion of 2197 patients for the reasons delineated above, the study cohort consisted of 1239 patients.
Events As Table 1 shows, the median follow-up period was 19.4 months. Over time, 91 (7.34%) patients did not persist on rate-control medication, with 63 patients switching to or
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on October 20, 2014
4
Annals of Pharmacotherapy
Table 1. Patient Events. Study Cohort
Total a
Event Number of patients Switch to or add rhythm-control drug Electrical cardioversion Catheter ablation and/or pacemaker Death Medicaid disenrollment End of study period without event Median time to event, months Median follow-up period, months
β-Blocker Arm a
Calcium-Channel-Blocker Arm
n (%)
n (%)
n (%)a
1239 (100) 63 (5.08) 15 (1.21) 13 (1.05) 0 (0.00) 604 (48.75) 544 (43.91) 7.37 19.37
1016 (100) 44 (4.33) 14 (1.38) 8 (0.79) 0 (0.00) 486 (47.84) 464 (45.67) 7.42 19.35
223 (100) 19 (8.52) 1 (0.45) 5 (2.24) 0 (0.00) 118 (52.92) 80 (35.87) 6.68 19.37
a
Percentage of column total.
adding a rhythm-control drug, 15 patients experiencing electrical cardioversion, and 13 patients undergoing catheter ablation and/or pacemaker implantation. No patient in the study died during the follow-up period. There were 1016 (82.0%) patients who were initially prescribed a BB, and the remaining 223 (18.0%) started with a CCB medication. Whereas 66 (6.5%) patients on a BB had an event (did not persist on rate-control medication), 25 (11.2%) patients on a CCB did not persist. Moreover, the median time to event for patients in the BB group versus the CCB group was longer (7.4 months vs 6.7 months; see Table 1.)
Patient Characteristics Descriptive statistics for the study cohort, stratified by study arm, are given in Table 2. Patients in the 2 arms were similar with respect to baseline characteristic percentages, with 3 exceptions: (1) patients prescribed a CCB were more likely to have pulmonary disease (54.7% vs 38.7%; P < 0.0001); (2) patients prescribed a CCB were more likely to be female (72.2% vs 57.3%; P < 0.0001); and (3) patients prescribed a CCB were less likely to have had prior ratecontrol drug use (63.2% vs 72.0%; P = 0.010). However, within propensity-score quintiles, proportions were not significantly different between study arms (nor were the proportions of any of the other baseline characteristics). (See Appendix B for propensity-score quintiles.) In Table 2, we also provide the postassignment patient characteristics stratified by study arm. During their followup period, 244 (24.0%) patients in the BB arm and 54 (24.2%) patients in the CCB arm (P = 0.949) had a new diagnosis of HF. Similarly, there was no statistically significant difference (P = 0.524) in the proportion of patients with low medication adherence between the 2 study arms (500 or 49.2% of patients in the BB arm vs 115 or 51.6% of
patients in the CCB arm). However, the proportion of CCB patients switching to or adding a BB (27.8%) was significantly (P < 0.0001) higher than the proportion of BB patients switching to or adding a CCB (7.6%). After the assignment date, 21 (33.9%) of the patients who switched to or added a BB in the CCB arm were diagnosed with HF. Similarly, 13.5% of patients in the CCB arm versus 9.7% of the patients in the BB arm (P = 0.091) had a switch to or addition of digoxin during the follow-up.
Persistence Patterns Figure 1 depicts the Kaplan-Meier curves for the 2 study arms. Consistent with the greater percentage of events and shorter median time to event for the CCB arm versus BB arm patients, the CCB persistence curve lies consistently below (P = 0.022) the BB curve. Whereas, at 60 months postassignment, 87.5% of BB patients remained on a rate-control drug regimen, that percentage for the CCB patients was 78.4%. To ensure that the results in Figure 1 were not affected by intra– drug-class switching, another set of Kaplan-Meier curves was derived for the 1100 patients with no switching. Exactly the same pattern was seen as in Figure 1; percentages at 60 months were the same as in Figure 1, and again, curve equality was rejected (P = 0.018). (See Appendix C.) Results from the Cox proportional hazard regression in Table 3 show that patients who received a CCB initially were 1.89 (95% CI = 1.14-3.08) times more likely to experience one of the study events (ie, less likely to persist on rate-control therapy) compared with patients who initially received a BB. The only other statistically significant coefficient was that for renal disease. Patients with renal disease were significantly less likely to have an event (hazard ratio = 0.44; CI = 0.22-0.89)—that is, more likely to persist with rate control. All statistical tests showed good model specification.
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on October 20, 2014
5
Desai et al Table 2. Patient Characteristics. Study Cohort
Total
β-Blocker Arm
Calcium-Channel-Blocker Arm
Characteristic
n (%)a
n (%)a
n (%)a
P Valueb
Number of patients Preassignment characteristics Age, years ± standard deviation Female Pulmonary disease Hepatic disease Renal disease Hypertension Diabetes Angina pectoris Statin use Antihypertensive use Diuretic use Prior use of rate-control drug Medicaid managed-care plan Postassignment characteristics Diagnosis with heart failure Low medication adherence Rate-control class switch or addition Switch to or addition of digoxin
1239 (100)
1016 (100)
223 (100)
51.6 ± 5.8 743 (59.97) 515 (41.57) 116 (9.36) 249 (20.10) 924 (74.58) 504 (40.68) 88 (7.10) 537 (43.34) 549 (44.31) 464 (37.45) 872 (70.38) 500 (40.36)
51.6 ± 5.9 582 (57.28) 393 (38.68) 100 (9.84) 211 (20.77) 761 (74.90) 419 (41.24) 73 (7.19) 453 (44.59) 455 (44.78) 371 (36.52) 731 (71.95) 413 (40.65)
51.4 ± 5.7 161(72.20) 122 (54.71) 16 (7.17) 38 (17.04) 163 (73.09) 85 (38.12) 15 (6.73) 84 (37.67) 94 (42.15) 93 (41.70) 141 (63.23) 87 (39.01)
0.737
