論文ID: CJ-16-0287
To the Editor:
In August 2012, we reported the safety and efficacy of once-daily rivaroxaban for stroke and systemic embolism prevention in Japanese patients with atrial fibrillation (AF) as a result of the phase III J-ROCKET AF study.1 In December 2014, the United States Food and Drug Administration (FDA) issued a medical device recall notice for the Alere™ INRatio® Monitor System (formally known as the HemoSense INRatio device). The FDA reported that this whole-blood point-of-care device may provide an international normalized ratio (INR) result that is lower than an automated, plasma-based laboratory INR in patients with certain specific medical conditions, including abnormal hematocrit levels, conditions associated with raised fibrinogen levels, and bleeding or unusual bruising.2,3 In September 2015, Bayer Yakuhin Ltd (Bayer) identified that the recall notice of the Alere INRatio device was applicable to the HemoSense INRatio devices used in the J-ROCKET AF study. To evaluate the effect of possible device malfunction leading to lower INR values and inappropriately high doses of warfarin and bleeding events, Bayer and Johnson & Johnson performed 3 sensitivity analyses based on data from the J-ROCKET AF study.
The sensitivity analyses were planned to exclude the effect of subsets of the study population listed in the Alere recall notice (relevant medical condition [RMC]) (Table 1),4 in order to provide a robust evaluation of the effect of the recall on the principal safety and primary efficacy endpoints. First, we defined the term “recall-related time of observation” to consider not only the presence of RMCs and adverse event (AE)/serious adverse event (SAE) start and stop dates but also the timing of prothrombin time (PT)-INR measurements performed with the device in association with these intervals. In brief, the mere presence of a RMC was not sufficient to warrant exclusion from observation because it was the performance of a PT-INR in association with this condition that was of most relevance. For example, if a patient developed a RMC but a PT-INR was not measured during that interval, the device recall would not be relevant. Conversely, PT-INRs performed while a patient had a RMC may exert an effect even after the AE/SAE stop date, until a subsequent (ie, unaffected) PT-INR was performed. Next, 2 physicians blinded to treatment assignments independently reviewed baseline medical history and AEs, identifying any conditions that matched those listed in the Alere letter. The 2 reviewers then compared their findings, and any discrepancies were discussed and resolved. Descriptions of the 3 sensitivity analyses are as follows. For sensitivity analysis no. 1, the safety outcome was the principal safety endpoint (composite of major and non-major clinically relevant bleeding), and the efficacy outcome was the primary efficacy endpoint (composite of stroke and non-central nervous system [CNS] systemic embolism). The trial population used was the safety population and the data scope was on-treatment. Patients who had a relevant chronic medical condition reported at baseline (ie, in their relevant medical history on the electronic case report form at the screening visit) were censored after the performance of their first PT-INR of the study. For the remaining patients, the analysis also specified time intervals if they had a RMC, with censoring as defined by recall-related time of observation, as described above. For sensitivity analysis no. 2, the safety outcome was the principal safety endpoint and the efficacy outcome was the primary efficacy endpoint. The trial population used was the safety population, and the data scope was on-treatment. This analysis completely excluded patients who had recall-related time of observation before the endpoint event or censoring. In other words, the only patients included in this analysis were those who did not have any recall-related time of observation before the endpoint event or censoring. For sensitivity analysis no. 3, the safety outcome was the principal safety endpoint and the efficacy outcome was the primary efficacy endpoint. The trial population used was the safety population, and the data scope was on-treatment. As above, this analysis can be considered in terms of the occurrence of an endpoint or a censoring event (ie, reaching the end of treatment with the study drug). For this analysis, if the endpoint occurred or the end of treatment with the study drug was reached outside of a recall-related time of observation, the patient was included. What occurred before the event was not relevant, provided the event occurred outside of the period of recall-related time of observation.
| Anemia of any type with hematocrit <30% (AlereTM INRatio® PT/INR Test Strips) or <25% (AlereTM INRatio®2 PT/INR Test Strips, Heparin Insensitive) |
| Any conditions associated with elevated fibrinogen levels including: |
| • Acute inflammatory conditions (examples may include acute viral or bacterial infections such as pneumonia or influenza) |
| • Chronic inflammatory conditions (examples may include rheumatoid arthritis, Crohn’s disease, ulcerative colitis, infectious liver diseases such as hepatitis, or inflammatory kidney diseases such as diabetic nephropathy and glomerulonephritis) |
| • Severe infection (eg, sepsis) |
| • Chronically elevated fibrinogen for any reason |
| • Hospitalized or advanced stage cancer or end stage renal disease patients requiring hemodialysis |
| Any bleeding or unusual bruising, clinically observed or reported by the patient |
Note: in addition, the letter to HealthCare Professionals asked them to “…adhere to the following precaution[s] in order to obtain the most accurate results: ‘Only use the Alere INRatio® PT/INR Monitor system on patient samples within the hematocrit range of 30% to 55%’.”
In all 3 analyses, rivaroxaban remained non-inferior to warfarin for the principal safety outcome of major and non-major clinically relevant bleeding (Table 2). The largest difference in event rate and hazard ratio was observed in sensitivity analyses 2 and 3, respectively. That is, the event rates per 100 patient-years in sensitivity analysis 2 increased in both treatments compared with the original analysis and in sensitivity analysis 3 the hazard ratio increased modestly compared with the original hazard ratio, but the upper limit of the 2-sided 95% confidence interval for each of the 3 sensitivity analyses remained well below the non-inferiority margin prespecified in the original study analysis, allowing the non-inferiority of rivaroxaban to be confirmed in line with the original analysis. In regard to the primary efficacy outcome of the composite of stroke and non-CNS systemic embolism, evaluation of the event rates and hazard ratios in the analyses showed little change from the original analysis for rivaroxaban vs. warfarin (Table 3). The results of these 3 sensitivity analyses, which took into consideration that the INRatio device was used to assess PT-INR values in the study, showed that the effect of potentially inaccurate PT-INR readings did not alter the original conclusion of the safety of rivaroxaban in the J-ROCKET AF study.
| Analysis | Rivaroxaban | Warfarin | Rivaroxaban vs. Warfarin | |||
|---|---|---|---|---|---|---|
| n=639 n/Jc |
Event rate (100 Pt-yr)b |
n=639 n/J |
Event rate (100 Pt-yr) |
Hazard ratio (95% CI) (a) |
P value (b) |
|
| Original trial result | 138/639 | 18.04 | 124/639 | 16.42 | 1.11 (0.87, 1.42) | <0.001 |
| Sensitivity analysis 1 | 122/690* | 17.68 | 105/661* | 15.88 | 1.13 (0.87, 1.47) | <0.001 |
| Sensitivity analysis 2 | 104/401 | 24.42 | 92/389 | 22.51 | 1.11 (0.84, 1.46) | <0.001 |
| Sensitivity analysis 3 | 122/556 | 18.48 | 105/544 | 16.26 | 1.15 (0.88, 1.49) | <0.001 |
aPrincipal Safety Endpoint is a composite of major and non-major clinically relevant bleeding events. bEvent rate 100 Pt-yr: number of events per 100 patient-years of follow-up. cn=number of subjects with events, and J=number of subjects in each subgroup except for sensitivity analysis 1 (marked with*) where J is time of observation not related to the recall (in patient-years). (a) For all analyses except sensitivity analysis 1, hazard ratio (95% confidence intervals (CI)) from the Cox proportional hazard model with treatment as a covariate. For sensitivity analysis 1, hazard ratio (95% CI) from the Cox proportional hazard model with treatment as a covariate, recall-related time of observation as a time-dependent covariate, and their interaction. (b) P value (one-sided) for non-inferiority of rivaroxaban vs. warfarin by a non-inferiority margin of 2 in hazard ratio.
Output of the original J-ROCKET AF safety analysis and the 3 sensitivity analyses performed in the present report. In the safety population, rivaroxaban remained non-inferior to warfarin for the principal safety outcome of major and non-major clinically relevant bleeding. In the 3 sensitivity analyses performed, no meaningful shifts in the respective hazard ratios for rivaroxaban vs. warfarin were observed and the CI did not widen appreciably. The upper limit of the 2-sided 95% CI for each of the 3 sensitivity analyses remained well below the non-inferiority margin of 2 in terms of risk (hazard) ratio prespecified in the original study analysis, so the non-inferiority of the study drug was still secured and declared. Of the 3 sensitivity analyses, the largest difference in event rate and hazard ratio was observed in sensitivity analyses 2 and 3, respectively. The event rates per 100 patient-years in sensitivity analysis 2 increased in both treatments compared with the original analysis (24.42 as opposed to 18.04/100 Pt-yr of rivaroxaban therapy, and 22.51 vs. 16.42/100 Pt-yr of warfarin therapy), but the hazard ratio was unchanged from the original study result. In sensitivity analysis 3, the hazard ratio increased modestly compared with the original hazard ratio, but the upper boundary of the 95% CI remained well below the non-inferiority margin.
| Analysis | Rivaroxaban | Warfarin | Rivaroxaban vs. Warfarin | |||
|---|---|---|---|---|---|---|
| n=639 n/Jc |
Event rate (100 Pt-yr)b |
n=639 n/J |
Event rate (100 Pt-yr) |
Hazard ratio (95% CI) (a) |
P value (b) |
|
| Original trial result | 11/639 | 1.26 | 22/639 | 2.60 | 0.48 (0.23, 1.00) | 0.050 |
| Sensitivity analysis 1 | 9/787* | 1.14 | 18/729* | 2.47 | 0.47 (0.21, 1.04) | 0.064 |
| Sensitivity analysis 2 | 5/368 | 1.05 | 16/353 | 3.77 | 0.29 (0.11, 0.79) | 0.015 |
| Sensitivity analysis 3 | 9/543 | 1.21 | 18/529 | 2.57 | 0.47 (0.21, 1.06) | 0.068 |
aPrimary Efficacy Endpoint is a composite of stroke and non-central nervous system (CNS) systemic embolism. bEvent rate 100 Pt-yr: number of events per 100 patient-years of follow-up. cn=number of subjects with events, and J=number of subjects in each subgroup except for sensitivity analysis 1 (marked with*) where J=time of observation not related to the recall (in patient-years). (a) For all analyses except sensitivity analysis 1, hazard ratio (95% confidence interval (CI)) from the Cox proportional hazard model with treatment as a covariate. For sensitivity analysis 1, hazard ratio (95% CI) from the Cox proportional hazard model with treatment as a covariate, recall-related time of observation as a time-dependent covariate, and their interaction. (b) Unadjusted P value (two-sided) for the exploratory hypothesis (hazard ratio equals to 1) of rivaroxaban vs. warfarin.
Event rates and hazard ratios for the time to the first occurrence of the primary efficacy events (composite of stroke and non-CNS systemic embolism) while on-treatment in the safety population for the original J-ROCKET AF efficacy analysis, as well as for the 3 sensitivity analyses performed. Evaluation in the safety population of the event rates and hazard ratios for the 3 sensitivity analyses showed that the efficacy event rates with rivaroxaban and warfarin for the primary efficacy endpoint remained similar to the original analysis. In sensitivity analysis 2, which included only those patients without any recall-related time of observation ahead of any endpoint event or censoring, the event rate decreased in the rivaroxaban group and the event rate increased in the warfarin group compared with the original analysis (1.05 as opposed to 1.26/100 Pt-yr of rivaroxaban therapy, 3.77 as opposed to 2.60/100 Pt-yr of warfarin therapy) and the hazard ratio was considerably lower, and the upper boundary of the 95% CI in this comparison was below the line of unity.
Y.O., M. Kajikawa, and M. Kato report employment by Bayer Yakuhin Ltd. G.P. reports employment by Johnson & Johnson. M.H. reports receiving consultant fee from Bayer Yakuhin Ltd, and lecture honoraria from Bayer Yakuhin Ltd and Boehringer Ingelheim.
(Released online April 22, 2016)
