Effect of Time after Randomization

As note earlier, in the Kaplan-Meier plot of the primary endpoint, Figure 24, much of the divergence in the curves for the event rates occurs in the 180 day of treatment. After that time, the curves tend to stay about the same distance apart, suggesting a

convergence of the event rates after the first 6 months of treatment.

Accordingly, we asked our colleagues in the Davison of Biometrics I to examine the relationship of time after randomization to the event rates and hazard ratio (rivaroxaban vs. warfarin) for primary endpoint events. Figure 25 is a plot of the annual event rate over time since randomization in each treatment arm. The curve for warfarin (in green) is above the curve for rivaroxaban (in pink) at all time points, but the distance between the curves decreases sharply from randomization until about 1 year of treatment. After that, the curves remain close together as the rate falls over time in each arm.

Figure 25. Estimated Hazard Functions over Time since Randomization

The curves in both Figure 24 and Figure 25 suggest that the event rates in the two treatment arms approach each other over the first six months of treatment and become nearly similar after one year. Using the Cox proportional hazards model, the HR for

event occurring up to Day 180 was calculated as 0.69 (95% CI, 0.432, 0.860). Starting at Day 180, the HR and CI are 0.895 (0.710, 1.13), consistent with the suggestion in the two figures.

We asked the sponsor to confirm these findings. The sponsor determined that the on treatment primary efficacy endpoint event rates in the rivaroxaban and warfarin arms (safety population), in the first 180 days after randomization, were, respectively, 1.66 and 2.66 events per hundred patient years, yielding a hazard ration of 0.62 (95% CI, 0.44, 0.88). For days 181 on, the analogous rates were 1.71 and 1.96 with a hazard ratio of 0.87 (95% CI, 0.70, 1.10), very similar to the rates obtained by FDA. The sponsor’s event rate data for these and other time periods are shown in Table 30. The data show a progressive increase in the hazard ratio over the first year of the study to 0.89, with stabilization after that.

These data confirm that once patients are stabilized on warfarin therapy, event rates with rivaroxaban and warfarin are quite similar.

Table 30: Primary Event Rates In Various Time Periods Safety Population, On Treatment

Time Interval --- Rivaroxaban ---­ --- Warfarin --- Rivaroxaban vs. Warfarin

from Event Rate Event Rate

Random­

ization n/N (100 pt-yr) n/N (100 pt-yr) Hazard Ratio _{(95% CI)} p-value

1-30 13/7061 2.29 20/7082 3.50 0.65 (0.33,1.31) 0.233 31-60 7/6766 1.28 16/6830 2.89 0.44 (0.18,1.08) 0.072 61-90 10/6585 1.87 15/6664 2.77 0.67 (0.30,1.50) 0.334 91-180 24/6439 1.56 35/6518 2.24 0.70 (0.41,1.17) 0.173 181-360 56/6058 1.97 64/6190 2.21 0.89 (0.62,1.27) 0.524 ≥ 361 80/5546 1.57 95/5613 1.84 0.86 (0.64,1.15) 0.304 1-180 53/7061 1.66 86/7082 2.66 0.62 (0.44,0.88) 0.007 ≥ 181 136/6058 1.71 158/6190 1.96 0.87 (0.70,1.10) 0.253

A likely explanation for the rising HR over time is sub-optimal TTR in the warfarin arm in the early weeks of study treatment. Table 31 is a display of the mean and SD of global INR at weekly intervals until week 4, then 4 week intervals until week 56, and then 8 week intervals until week 180, when only 1 subject had INR data. The data indicate that the mean (SD) INR over the course of the study was 2.40 (0.38). During the first week, mean INR was 2.26 (1.09), but by week 2 it was 2.26 (1.06). By week 4, mean INR was 2.38 (0.86) and the mean remained near that value for most of the next three years of treatment. However, the SD fell gradually over this period, suggesting less variance,

which might explain the narrowing difference in the event rates over time, as INR below 2 and above 3 would be associated with increased primary efficacy event risk. Analysis of only mean INR could obscure important temporal trends in the data, and we plan to examine other approaches to evaluating INR control.

Table 31. Mean INR Over The Course Of ROCKET

Time since Random -ization N INR Mean (SD) Time since Random­ ization N INR Mean (SD) Entire Study 7025 2.40 (0.38) WEEK 56 5343 2.40 (0.73)

WEEK 1 6672 2.26 (1.09) WEEK 64 4731 2.42 (0.73) WEEK 2 6351 2.49 (1.06) WEEK 72 4200 2.43 (0.75) WEEK 3 1669 2.43 (1.02) WEEK 80 3803 2.41 (0.73) WEEK 4 6576 2.38 (0.86) WEEK 88 3249 2.44 (0.71) WEEK 8 6630 2.34 (0.81) WEEK 96 2811 2.43 (0.73) WEEK 12 6487 2.38 (0.81) WEEK 104 2355 2.43 (0.71) WEEK 16 6358 2.40 (0.80) WEEK 112 1860 2.41 (0.72) WEEK 20 6246 2.40 (0.77) WEEK 120 1406 2.40 (0.70) WEEK 24 6130 2.41 (0.80) WEEK 128 1037 2.46 (0.68) WEEK 28 6041 2.40 (0.76) WEEK 136 664 2.43 (0.65) WEEK 32 5931 2.40 (0.75) WEEK 144 343 2.38 (0.68) WEEK 36 5813 2.42 (0.75) WEEK 152 163 2.50 (0.60 WEEK 40 5734 2.40 (0.74) WEEK 160 56 2.45 (0.46) WEEK 44 5653 2.42 (0.75) WEEK 168 13 2.46 (0.54) WEEK 48 5577 2.42 (0.70) WEEK 172 8 2.59 (0.63) WEEK 52 5501 2.44 (0.74) WEEK 180 1 1.5

We asked the Sponsor to provide tables of mean time in ranges on INR during specified intervals of treatment in the warfarin arm overall and the subsets of patients were VKA experienced and VKA naïve at baseline. These data are displayed in Table 32, Table 33, and Table 34, respectively.

Table 32. Mean Time In Specified Ranges Of INR During Intervals Of Treatment Safety Population

Table 33. Mean Time In Specified Ranges Of INR During Intervals Of Treatment Safety Population, Patients with VKA use at Baseline

Table 34. Mean Time In Specified Ranges Of INR During Intervals Of Treatment Safety Population, Patients with No VKA use at Baseline

The VKA naïve patients had a time in the INR therapeutic range of 2 to 3 (TTR) of 31% from day 1-30 and did not exceed 50% until the period from day 181-360, despite substantial attrition. Out of range values were mostly on the low side (i.e., <2), but about 21% were > 3 in the first 30 days. The VKA experience patients started with a TTR of 48% in the first 30 days and reached 63% in the period from day 181-360. At all time points until the very last days of the study when one VKA experienced patient remained on treatment, TTR was substantially lower in the VKA naïve patients than in the VKA experienced patients.

Consistent with the TTR data, the primary efficacy event rate data show a substantial difference between the VKA experienced patients and the VKA naïve patients in the pattern of event rates and rivaroxaban vs. warfarin arm hazard ratios over the course of the study. Data for the overall population, VKA experienced, and VKA naïve patients are summarized in for the periods from Day 1 to 180 and Day 181 and beyond for the safety population on treatment in Table 35.

Table 35. Primary Efficacy Endpoint Events By Baseline VKA Status And Time

Period

Safety Population, On Treatment Population and

Time Interval

--- Rivaroxaban ---­ --- Warfarin ---

Rivaroxaban vs. Warfarin Event Rate Event Rate

From Randomization

n/N

(100 pt-yr) n/N (100 pt-yr) Hazard Ratio

(95% CI) p-value All patients 1-180 53/7061 1.66 86/7082 2.66 0.62 (0.44,0.88) 0.007 ≥ 181 136/6058 1.71 158/6190 1.96 0.87 (0.70,1.10) 0.253 VKA Experienced 1-180 25/4401 1.24 47/4437 2.28 0.54 (0.33,0.88) 0.014 ≥ 181 89/3839 1.71 93/3985 1.73 0.99 (0.74,1.32) 0.948 VKA Naïve 1-180 28/2660 2.37 39/2645 3.33 0.71 (0.44,1.16) 0.171 ≥ 181 47/2219 1.72 65/2205 2.43 0.71 (0.49,1.03) 0.072

In both VKA naïve patients VKA experienced patients, the warfarin arm event rate falls by roughly 25% from the 0-180 day period to the ≥ day 181period, but the rates are higher in the VKA naïve patients in both periods. The absolute reduction in rates

between the two periods is also somewhat larger in the VKA naïve patients (a reduction of 0.7 vs. 0.55 events per 100 patient-years).

In the rivaroxaban arm, the event rate falls from the earlier to later period in the VKA naïve patients, but moves in the opposite direction in the VKA experienced patients. Overall there is only a small rise in the event rate in the rivaroxaban arm from the earlier to later period.

Thus, most of the observed increase in the hazard ratio from the early period to the later period in the “all patients” rows of Table 33 results from the decrease in the warfarin arm event rate over time, which was larger in the VKA naïve patients in absolute terms. This suggests that poor warfarin control played in role in the relative poor results for warfarin from day 0 to 180, but that reductions in event rates in both the VKA naïve and

experienced patients contributed to the overall reduction. The differing patters of change in the event rates over time in the VKA naïve and experienced subgroups in the events rates in the rivaroxaban arm are difficult to explain, and may be due to chance.

Reviewer Comment: The fact that patients who were VKA experienced at baseline therapy had similar event rates after 180 days on study regardless of treatment arm suggests that such patients may have little to gain (except perhaps convenience) from switching to rivaroxaban. This is another argument

against suggesting in labeling that rivaroxaban is superior to warfarin in preventing thrombotic events in non-valvular AFib patients.