Effects of Nonpersistence With Medication on CV Disease
Effects of Nonpersistence With Medication on CV Disease
In ONTARGET, 25,620 patients were randomized. Persistence to study medication over their complete observation period was seen in 20,991 patients, whereas 4,629 patients (18.1%) stopped study medications before their final visit or death (nonpersistent patients). Among the 4,629 nonpersistent patients, 18.8% had interrupted taking their study medication before. In persistent patients, the percentage of temporal interruptions was 11.5%.
The number of patients not taking study medication steadily increased over time (online Appendix Supplementary Figure 1). Most discontinuations of study medication occurred within the first year of the study (33.7% of all noncompleters) (online Appendix Supplementary Table I).
(Enlarge Image)
Supplementary Figure 1.
Supplementary Proportion of patients noncompliant to study medication.
Table I summarizes the baseline characteristics by persistence with study medications and by the time of stopping (before or after the first primary event) together with results of univariate χ tests. Results of the Cox regression analysis of the time to permanent stopping of study medications including all baseline characteristics simultaneously are depicted in Table II. An increased risk of nonpersistence with study medications (after adjustment for all other factors) was associated with older age, female gender, black ethnicity, smoking, history of diabetes, previous stroke/transient ischemic attack (TIA), and history of depression (all P < .01). On the other hand, the persistence rate was higher in Asians, in individuals reporting regular physical activities, and in previous users of ACE-Is. The numbers of drugs, additional to study medication, in nonpersistent and persistent patients were 4.51 (1.8) and 4.35 (1.8), respectively (P < .01). In patients who stopped medication, there was no apparent difference when patients became nonpersistent, before or after an event, for most of the patient characteristics. However, in patients stopping after an event, the percentage of patients with previous stroke/TIA, history of diabetes or hypertension, and previous use of ACE-Is was higher. Some differences were present between patients stopping for an adverse event (54%) or for other reasons (46%) (see online Appendix Supplementary Table II).
When comparing persistent and nonpersistent patient groups, nonpersistent patients had higher event rates and significantly increased HRs for all nonfatal CV end points (Table III). However, this "fixed" analysis also included events of nonpersistent patients before they stopped taking study medications. Thus, the Cox model including the patients' status (being on or off medication) as a time-dependent covariate better assesses the relationship between persistence and outcome; in this analysis, we see a clearly increased risk of nonpersistence for CV death and chronic heart failure (CHF) hospitalization but not for MI and stroke (Table III). Figure 1A-F shows Kaplan-Meier-curves displaying the proportion of patients with an event separately for persistent patients over the complete study period and for nonpersistent individuals after cessation of study medication. These descriptive curves (which are not adjusted for any confounders) visualize the increased risk of CV death and CHF hospitalization and the combined end points after cessation of study medication. Non-CV events (injury, poisoning, and procedural complications) were more frequent in nonpersistent (6.5%) than in persistent patients (4.0%, P < .0001), but this was not statistically different when analyzed in a time-dependent manner and adjusted for confounders (HR 1.03, 99% CI 0.81–1.31, P = .75).
(Enlarge Image)
Figure 1.
Kaplan-Meier curves for the primary composite outcome and the individual components thereof in patients being on and off study medication. Three-fold end point corresponds to the primary outcome in the HOPE trial.
The association of nonpersistence with CV end points was further explored by looking at the Kaplan-Meier curves stratified by the year of discontinuation. Figure 2 shows cumulative CV events in patients with continuous persistence and patients who stopped study medications within the first, second, third, and forth year and later. For all end points, there was a sharp increase of events in the year of discontinuation of study medications. For CV death, it has to be considered that the presentation is biased because patients stopping later than after 1 year cannot have died before (immortal time bias). In patients who did not stop the study medications, event rates for nonfatal were comparable with the nonpersistent patients before and later than 1 year after discontinuation, but in the year of discontinuation, a rapid increase was observed. This increase was more or less independent of the year in which the discontinuation occurred, and this applied to all individual component outcomes and consequently to the composite end points (online Appendix Supplementary Figure 2, and Supplementary Table III).
(Enlarge Image)
Figure 2.
Kaplan-Meier curves showing time-dependent outcomes in subgroups discontinuing study medication at different time points. Outcomes in patients with no discontinuation are incorporated.
(Enlarge Image)
Supplementary Figure 2.
Supplementary Kaplan-Meier curves summarizing the occurrence of the 4-fold end point: 3-fold end point (A), CV death (B), MI (C), stroke (D), and heart failure hospitalization (E) after an event (F). For comparison, the event rates of patients without discontinuation are shown.
Figure 3 shows for each end point that the average persistence was highest without events, whereas in patients who experienced an event, the average persistence was lower when the event occurred early and higher when the event occurred late. This finding suggests (as does the observation of the sharp increase of events in the year of discontinuation) that events may not only occur as a result of nonpersistence but that nonpersistence may also be resulting from a previous nonfatal event.
(Enlarge Image)
Figure 3.
Rate of persistence patients who had events at different time points during study follow-up. The 4-fold end point comprised CV death, MI, stroke, and heart failure hospitalization. The 3-fold end point is CV death, MI, and stroke, which was used a primary outcome in the HOPE trial.
To study the association of nonfatal CV events on persistence more thoroughly, the proportion of patients stopping study medications after first MI, stroke, and hospitalization for heart failure was compared with the same proportion in patients without an event of the same type (Figure 4). For MI and stroke, the rate of nonpersistence sharply increased within the first year after the event (HR > 3) and afterward leveled off to increase with a similar slope as patients without an event. Only after CHF hospitalizations, there was a significant increase of discontinuations over the complete observation period (more than 3-fold in the first year and doubled afterward). Similar observations have been made for other CV events, which were not components of the composite end points, such as new onset of CHF, end-stage renal disease, revascularization, angina pectoris, TIAs, and new atrial fibrillation as well as for patients with malignancy (online Appendix Supplementary Figure 3). However, the opposite was observed with improved persistence to study medications after new onset of diabetes mellitus. In the online Appendix Supplementary Figure 4, for each end point, the nonpersistent patients were separated into those who have stopped medication independent of an event, and those who had stopped after an event, and the incidence of stopping was calculated per patient-years (PYs). The data show that all adjudicated primary and secondary end points, except the new onset of diabetes mellitus, were significantly associated with increased rates of stopping study medications.
(Enlarge Image)
Figure 4.
Effect of components of the composite end point MI (A), stroke (B), and hospitalization for heart failure (C) on nonpersistence.
(Enlarge Image)
Supplementary Figure 3.
Supplementary Effect of nonfatal events on the persistence rate of patients. These adjudicated outcomes were not components of the individual end points. AF, atrial fibrillation; ESRD end-stage renal disease.
(Enlarge Image)
Supplementary Figure 4.
Supplementary Hazard ratio of nonpersistence occurrence after the first event or before an event in the trial.
Figure 5 shows that the rate of discontinuation was clearly dependent on the total number of events with increasing rates of nonpersistence in patients with multiple events. Therefore, we explored whether discontinuation after an event is predictive for subsequent outcomes. Online Appendix Supplementary Figure 5 shows the Kaplan-Meier curve for the primary end point in persistent patients and in patients who became nonpersistent before or after an event. All 3 groups were significantly different. Patients who were nonpersistent after a primary event experienced more subsequent events (6.95 per 100 PYs, adjusted HR vs persistent patients 1.54 [1.13–2.11, P = .0004]) than patients who became nonpersistent before a primary event (4.75 per 100 PYs, adjusted HR vs persistent patients 1.31 [1.16–1.48, P < .0001]); persistent patients had 3.43 primary events per 100 PYs. The adjusted and unadjusted HRs for stopping after versus before an event for major individual end points are given in the online Appendix Supplementary Figure 6. Among the nonpersistent patients, 54% of study medication discontinuation could be attributed to adverse events, whereas in 46%, the stopping was related to other reasons. In patients who stopped before a primary outcome event, there was no difference whether the stop was related to an adverse event or not (HR, 1.06, 95 CI, 0.84–1.34, P = .49). Some of the nonpersistent patients were taking open-label nonstudy ACE-Is (40.0%) and ARBs (18.8%), the rate being higher than in persistent patients (ACE-Is 6.9%, ARBs 2.1%). Open-label therapy with ACE-I or ARBs was related to a reduced hazard of CV and non-CV death, but an increased hazard of MI or heart failure hospitalization (online Appendix Supplementary Figure 7).
(Enlarge Image)
Figure 5.
Number of nonfatal events in population and persistence (%, ordinate).
(Enlarge Image)
Supplementary Figure 5.
Supplementary Kaplan-Meier curves showing time-dependent outcomes (4-fold end point) in patients who did not stop medications or who stopped before an event or after an event.
(Enlarge Image)
Supplementary Figure 6.
Supplementary Hazard ratio of having stopped study medication after or before an event.
(Enlarge Image)
Supplementary Figure 7.
Supplementary Hazard ratio of receiving open-label ACE-I/ARB therapy or none of such therapies in patients nonadherent to study medication.
Results
In ONTARGET, 25,620 patients were randomized. Persistence to study medication over their complete observation period was seen in 20,991 patients, whereas 4,629 patients (18.1%) stopped study medications before their final visit or death (nonpersistent patients). Among the 4,629 nonpersistent patients, 18.8% had interrupted taking their study medication before. In persistent patients, the percentage of temporal interruptions was 11.5%.
The number of patients not taking study medication steadily increased over time (online Appendix Supplementary Figure 1). Most discontinuations of study medication occurred within the first year of the study (33.7% of all noncompleters) (online Appendix Supplementary Table I).
(Enlarge Image)
Supplementary Figure 1.
Supplementary Proportion of patients noncompliant to study medication.
Table I summarizes the baseline characteristics by persistence with study medications and by the time of stopping (before or after the first primary event) together with results of univariate χ tests. Results of the Cox regression analysis of the time to permanent stopping of study medications including all baseline characteristics simultaneously are depicted in Table II. An increased risk of nonpersistence with study medications (after adjustment for all other factors) was associated with older age, female gender, black ethnicity, smoking, history of diabetes, previous stroke/transient ischemic attack (TIA), and history of depression (all P < .01). On the other hand, the persistence rate was higher in Asians, in individuals reporting regular physical activities, and in previous users of ACE-Is. The numbers of drugs, additional to study medication, in nonpersistent and persistent patients were 4.51 (1.8) and 4.35 (1.8), respectively (P < .01). In patients who stopped medication, there was no apparent difference when patients became nonpersistent, before or after an event, for most of the patient characteristics. However, in patients stopping after an event, the percentage of patients with previous stroke/TIA, history of diabetes or hypertension, and previous use of ACE-Is was higher. Some differences were present between patients stopping for an adverse event (54%) or for other reasons (46%) (see online Appendix Supplementary Table II).
Impact of Persistence on CV Events
When comparing persistent and nonpersistent patient groups, nonpersistent patients had higher event rates and significantly increased HRs for all nonfatal CV end points (Table III). However, this "fixed" analysis also included events of nonpersistent patients before they stopped taking study medications. Thus, the Cox model including the patients' status (being on or off medication) as a time-dependent covariate better assesses the relationship between persistence and outcome; in this analysis, we see a clearly increased risk of nonpersistence for CV death and chronic heart failure (CHF) hospitalization but not for MI and stroke (Table III). Figure 1A-F shows Kaplan-Meier-curves displaying the proportion of patients with an event separately for persistent patients over the complete study period and for nonpersistent individuals after cessation of study medication. These descriptive curves (which are not adjusted for any confounders) visualize the increased risk of CV death and CHF hospitalization and the combined end points after cessation of study medication. Non-CV events (injury, poisoning, and procedural complications) were more frequent in nonpersistent (6.5%) than in persistent patients (4.0%, P < .0001), but this was not statistically different when analyzed in a time-dependent manner and adjusted for confounders (HR 1.03, 99% CI 0.81–1.31, P = .75).
(Enlarge Image)
Figure 1.
Kaplan-Meier curves for the primary composite outcome and the individual components thereof in patients being on and off study medication. Three-fold end point corresponds to the primary outcome in the HOPE trial.
The association of nonpersistence with CV end points was further explored by looking at the Kaplan-Meier curves stratified by the year of discontinuation. Figure 2 shows cumulative CV events in patients with continuous persistence and patients who stopped study medications within the first, second, third, and forth year and later. For all end points, there was a sharp increase of events in the year of discontinuation of study medications. For CV death, it has to be considered that the presentation is biased because patients stopping later than after 1 year cannot have died before (immortal time bias). In patients who did not stop the study medications, event rates for nonfatal were comparable with the nonpersistent patients before and later than 1 year after discontinuation, but in the year of discontinuation, a rapid increase was observed. This increase was more or less independent of the year in which the discontinuation occurred, and this applied to all individual component outcomes and consequently to the composite end points (online Appendix Supplementary Figure 2, and Supplementary Table III).
(Enlarge Image)
Figure 2.
Kaplan-Meier curves showing time-dependent outcomes in subgroups discontinuing study medication at different time points. Outcomes in patients with no discontinuation are incorporated.
(Enlarge Image)
Supplementary Figure 2.
Supplementary Kaplan-Meier curves summarizing the occurrence of the 4-fold end point: 3-fold end point (A), CV death (B), MI (C), stroke (D), and heart failure hospitalization (E) after an event (F). For comparison, the event rates of patients without discontinuation are shown.
Association Between Events and Subsequent Persistence
Figure 3 shows for each end point that the average persistence was highest without events, whereas in patients who experienced an event, the average persistence was lower when the event occurred early and higher when the event occurred late. This finding suggests (as does the observation of the sharp increase of events in the year of discontinuation) that events may not only occur as a result of nonpersistence but that nonpersistence may also be resulting from a previous nonfatal event.
(Enlarge Image)
Figure 3.
Rate of persistence patients who had events at different time points during study follow-up. The 4-fold end point comprised CV death, MI, stroke, and heart failure hospitalization. The 3-fold end point is CV death, MI, and stroke, which was used a primary outcome in the HOPE trial.
To study the association of nonfatal CV events on persistence more thoroughly, the proportion of patients stopping study medications after first MI, stroke, and hospitalization for heart failure was compared with the same proportion in patients without an event of the same type (Figure 4). For MI and stroke, the rate of nonpersistence sharply increased within the first year after the event (HR > 3) and afterward leveled off to increase with a similar slope as patients without an event. Only after CHF hospitalizations, there was a significant increase of discontinuations over the complete observation period (more than 3-fold in the first year and doubled afterward). Similar observations have been made for other CV events, which were not components of the composite end points, such as new onset of CHF, end-stage renal disease, revascularization, angina pectoris, TIAs, and new atrial fibrillation as well as for patients with malignancy (online Appendix Supplementary Figure 3). However, the opposite was observed with improved persistence to study medications after new onset of diabetes mellitus. In the online Appendix Supplementary Figure 4, for each end point, the nonpersistent patients were separated into those who have stopped medication independent of an event, and those who had stopped after an event, and the incidence of stopping was calculated per patient-years (PYs). The data show that all adjudicated primary and secondary end points, except the new onset of diabetes mellitus, were significantly associated with increased rates of stopping study medications.
(Enlarge Image)
Figure 4.
Effect of components of the composite end point MI (A), stroke (B), and hospitalization for heart failure (C) on nonpersistence.
(Enlarge Image)
Supplementary Figure 3.
Supplementary Effect of nonfatal events on the persistence rate of patients. These adjudicated outcomes were not components of the individual end points. AF, atrial fibrillation; ESRD end-stage renal disease.
(Enlarge Image)
Supplementary Figure 4.
Supplementary Hazard ratio of nonpersistence occurrence after the first event or before an event in the trial.
Figure 5 shows that the rate of discontinuation was clearly dependent on the total number of events with increasing rates of nonpersistence in patients with multiple events. Therefore, we explored whether discontinuation after an event is predictive for subsequent outcomes. Online Appendix Supplementary Figure 5 shows the Kaplan-Meier curve for the primary end point in persistent patients and in patients who became nonpersistent before or after an event. All 3 groups were significantly different. Patients who were nonpersistent after a primary event experienced more subsequent events (6.95 per 100 PYs, adjusted HR vs persistent patients 1.54 [1.13–2.11, P = .0004]) than patients who became nonpersistent before a primary event (4.75 per 100 PYs, adjusted HR vs persistent patients 1.31 [1.16–1.48, P < .0001]); persistent patients had 3.43 primary events per 100 PYs. The adjusted and unadjusted HRs for stopping after versus before an event for major individual end points are given in the online Appendix Supplementary Figure 6. Among the nonpersistent patients, 54% of study medication discontinuation could be attributed to adverse events, whereas in 46%, the stopping was related to other reasons. In patients who stopped before a primary outcome event, there was no difference whether the stop was related to an adverse event or not (HR, 1.06, 95 CI, 0.84–1.34, P = .49). Some of the nonpersistent patients were taking open-label nonstudy ACE-Is (40.0%) and ARBs (18.8%), the rate being higher than in persistent patients (ACE-Is 6.9%, ARBs 2.1%). Open-label therapy with ACE-I or ARBs was related to a reduced hazard of CV and non-CV death, but an increased hazard of MI or heart failure hospitalization (online Appendix Supplementary Figure 7).
(Enlarge Image)
Figure 5.
Number of nonfatal events in population and persistence (%, ordinate).
(Enlarge Image)
Supplementary Figure 5.
Supplementary Kaplan-Meier curves showing time-dependent outcomes (4-fold end point) in patients who did not stop medications or who stopped before an event or after an event.
(Enlarge Image)
Supplementary Figure 6.
Supplementary Hazard ratio of having stopped study medication after or before an event.
(Enlarge Image)
Supplementary Figure 7.
Supplementary Hazard ratio of receiving open-label ACE-I/ARB therapy or none of such therapies in patients nonadherent to study medication.
Source...