Electronic Supplementary Material s13

Electronic Supplementary Material

Article Title: A State Transition Model for Health Outcomes Associated with Vorapaxar Treatment as an Add-on to Standard Care Antiplatelet Therapy in the Prevention of Thrombotic Events for Patients with a Recent Myocardial Infarction

Authors: John D. Whalen, BA1
Glenn Davies, DrPH, MS2
Mark Du, MPH3
Mustafa Oguz, PhD1
Lori D. Bash, PhD, MPH4
Ipek Ozer-Stillman, MS5

Institutions: 1 Evidera, Metro Building, 6th Floor, 1 Butterwick, London, W6 8DL, UK
2 Merck & Co., Inc., 351 North Sumneytown Pike, North Wales, PA, 19454, US
3 Evidera, 7101 Wisconsin Avenue, Suite 1400, Bethesda, MD, 20814, US
4 Merck & Co., Inc., 126 East Lincoln Avenue, Rahway, NJ, 07065, US
5 Evidera, 430 Bedford Street, Suite 300, Lexington, MA, 02420, US

Note: Ms. Ozer-Stillman and Mr. Whalen were employees of Evidera at the time this work was conducted and this manuscript was developed.

Corresponding Author:

Dr. Mustafa Oguz
Evidera, Metro Building, 6th Floor, 1 Butterwick, London, W6 8DL, UK
Phone: +44 (0) 208 576 5016 Fax: +44 (0) 20 8576 5195
E-mail:

Appendix A. Description of Risk Equation Development Process

Risk equations were developed from a subset of patients in the qualifying MI cohort of TRA 2°P. Patients were included in this analysis if they did not have a history of stroke or TIA and were receiving either ASA+CLO or ASA alone at baseline. Patients who received clopidogrel monotherapy or other anticoagulant/antithrombotic agents were not included; this exclusion accounted for less than 2% of patients without prior stroke or TIA. Summary characteristics of patients are presented in Table 4.

The events considered in this analysis include cardiovascular (CV) death, ischemic stroke, hemorrhagic stroke, and myocardial infarction (MI). Patients who did not experience one of these events were censored at their last visit in the study. At the end of the analysis, a total of six equations were developed; one equation for each baseline antiplatelet therapy (ASA+CLO and ASA) and event. Tables 5, 6, and 7 below presents the parameters, which include the intercept, coefficients of the covariates, and the shape parameter of the Weibull fit. For each of the six equations, exponential and log-normal distributions were also considered.

The fit statistics provided in Table 8, including the negative log likelihood, Akaike information criterion (AIC), and Bayesian information criterion (BIC), were very similar across all three distributions for all equations. Ultimately, the decision to pursue a Weibull distribution was based on its common implementation in the literature, and to provide consistency in modeling the risk of major CV events. The original Framingham equations were based on Weibull distributions and have been used in many studies subsequently, because of their fit to long-term observational data [41].

An analysis was carried out in which for each of the six equations, the distribution that yielded the most favorable (i.e., lowest) fit statistics. The results in Table 9 show that, under a situation in which the best fit for each equation was chosen, the absolute discounted QALYs increased by approximately 0.10, but the incremental QALYs gained associated with VOR+SC had virtually no change, relative to the base case in which all equations assumed a Weibull distribution.

For a given individual patient profile, these risk equations can be used to derive a survivor function. Under a Weibull distribution, the survivor function is defined as

S(t) = e(-λ(tγ))

where λ is the scale parameter, and t is time expressed in days. The scale parameter is parameterized in the following way:

λ = e(-γ*xβ)

where xβ is the dot product of patient characteristics and the risk equation coefficients.

As an example calculation, consider a hypothetical patient profile with the following characteristics: treated with vorapaxar, male, 62 years old, 80 kg, non-diabetic, no PAD, qualifying NSTEMI (i.e., no STEMI, STEMI type not unknown), GFR > 60, history of PCI, no history of CABG, entry into TRA 2°P six months after suffering an MI, and on aspirin monotherapy at baseline. To calculate the cumulative probability of not having had an MI within the first year of follow-up, we first determine the scale parameter:

λ =e(-0.7621*11.1206)= 2.0861*10-4

Then, this cumulative probability can be calculated:

S(t365 days) = e(-2.0861*10-4*(3650.7621))= 0.981

Table 4. Summary Characteristics of Patients Included in the Risk Equation Development

/ Vorapaxar Arm / Placebo Arm /
Variable / Statistic / ASA+CLO / ASA / ASA+CLO / ASA /
Female / n (%) / 1255 (19.54%) / 428 (22.63%) / 1232 (19.03%) / 399 (21.88%)
Age, years / Mean
SD
Median
25%
75%
N / 58.41
10.58
58
51
66
6424 / 59.78
10.48
60
52
67
1891 / 58.15
10.36
58
51
65
6474 / 59.37
10.61
59
52
67
1824
Weight <60 kg / n (%) / 284 (4.42%) / 142 (7.51%) / 281 (4.34%) / 131 (7.18%)
Diabetes / n (%) / 1291 (20.10%) / 481 (25.44%) / 1327 (20.50%) / 445 (24.40%)
History of PAD / n (%) / 502 (7.81%) / 172 (9.10%) / 524 (8.09%) / 175 (9.59%)
STEMI / n (%) / 3440 (53.55%) / 976 (51.61%) / 3429 (52.97%) / 948 (51.97%)
STEMI type Unknown / n (%) / 318 (4.95%) / 158 (8.36%) / 354 (5.47%) / 134 (7.35%)
GFR (mL/min/1.73 m2) / Mean
SD
Median
25%
75%
N / 83.10
20.67
82.31
69.89
95.52
6352 / 80.91
21.64
80.80
66.77
93.39
1869 / 83.48
20.28
82.47
70.24
95.77
6395 / 81.43
22.02
80.33
67.74
93.75
1802
Prior PCI / n (%) / 5644 (87.86%) / 966 (51.08%) / 5656 (87.36%) / 961 (52.69%)
Prior CABG / n (%) / 520 (8.09%) / 557 (29.46%) / 518 (8.00%) / 548 (30.04%)
Time Since Qualifying MI Event
<=1 month
>1–3 months
>3–6 months
>6–12 months
>12 months
Missing / n (%) / 2020 (31.44%)
1781 (27.72%)
1170 (18.21%)
1419 (22.09%)
19 (0.30%)
4 (0.06%) / 269 (14.23%)
410 (21.68%)
390 (20.62%)
802 (42.41%)
12 (0.63%)
2 (0.11%) / 2148 (33.18%)
1770 (27.34%)
1157 (17.87%)
1362 (21.04%)
15 (0.23%)
10 (0.15%) / 239 (13.10%)
418 (22.92%)
414 (22.70%)
734 (40.24%)
12 (0.66%)
3 (0.16%)

ASA, Dual antiplatelet therapy with aspirin; CLO, Clopidogrel; CABG, Coronary artery bypass graft; GFR, Glomerular filtration rate; MI, Myocardial infarction; PAD, Peripheral arterial disease; PCI, Percutaneous coronary intervention; STEMI, ST segment elevation myocardial infarction.

Table 5. Risk Equation Parameter Values for MI

/ MI /
ASA+CLO (SE) / ASA (SE) /
Vorapaxar (0=yes, 1=no) / −0.258 (0.089) / −0.351 (0.211)
Sex (0=male, 1=female) / −0.095 (0.115) / −0.118 (0.257)
Age (years) / −0.001 (0.005) / −0.025 (0.011)
Weight (kg) / −0.003 (0.003) / −0.016 (0.006)
Diabetes (0=yes, 1=no) / 0.419 (0.101) / 0.544 (0.227)
Peripheral Arterial Disease (0=yes, 1=no) / 0.634 (0.128) / 1.01 (0.279)
STEMI (0=no, 1=yes) / 0.452 (0.179) / 0.475 (0.361)
STEMI type unknown (0=no, 1=yes) / −0.073 (0.094) / −0.073 (0.226)
GFR <= 60 (0=no, 1=yes) / −0.367 (0.119) / −0.606 (0.249)
History of PCI (0=yes, 1=no) / 0.246 (0.135) / 0.215 (0.225)
History of CABG (0=yes, 1=no) / 0.975 (0.12) / −0.373 (0.231)
Time to randomization / 0.048 (0.014) / 0.014 (0.03)
Intercept / 8.559 (0.11) / 12.695 (0.273)
Weibull Shape (γ) / 0.859 (0.015) / 0.762 (0.033)

ASA, Dual antiplatelet therapy with aspirin; CLO, Clopidogrel; CABG, Coronary artery bypass graft; GFR, Glomerular filtration rate; MI, Myocardial infarction; PCI, Percutaneous coronary intervention; SE, Standard error; STEMI, ST segment elevation myocardial infarction.

Table 6. Risk Equation Parameter Values for Stroke

/ Stroke /
ASA+CLO (SE) / ASA (SE) /
Vorapaxar (0=yes, 1=no) / −0.669 (0.225) / −0.712 (0.37)
Sex (0=male, 1=female) / −0.041 (0.267) / 0.536 (0.473)
Age (years) / −0.028 (0.011) / −0.032 (0.018)
Weight (kg) / −0.001 (0.006) / 0.004 (0.011)
Diabetes (0=yes, 1=no) / 0.27 (0.235) / 0.652 (0.37)
Peripheral Arterial Disease (0=yes, 1=no) / 0.348 (0.306) / 0.497 (0.485)
STEMI (0=no, 1=yes) / 0.337 (0.451) / −0.448 (0.649)
STEMI type unknown (0=no, 1=yes) / 0.146 (0.216) / −0.594 (0.366)
GFR <= 60 (0=no, 1=yes) / 0.186 (0.3) / −0.477 (0.412)
History of PCI (0=yes, 1=no) / 0.138 (0.312) / 0.512 (0.383)
History of CABG (0=yes, 1=no) / 0.588 (0.283) / −0.775 (0.481)
Time to randomization / −0.025 (0.03) / 0.087 (0.051)
Intercept / 12.556 (0.416) / 12.818 (0.578)
Weibull Shape (γ) / 1.048 (0.021) / 1.039 (0.038)

Table 7. Risk Equation Parameter Values for CV Death

CV Death
ASA+CLO (SE) / ASA (SE)
Receiving vorapaxar (0=yes, 1=no) / −0.161 (0.145) / −0.196 (0.21)
Sex (0=male, 1=female) / −0.022 (0.178) / 0.108 (0.258)
Age (years) / −0.037 (0.008) / −0.009 (0.011)
Weight (kg) / −0.004 (0.004) / −0.01 (0.006)
Diabetes (0=yes, 1=no) / 0.629 (0.159) / 0.594 (0.226)
Peripheral Arterial Disease (0=yes, 1=no) / 0.823 (0.183) / 0.847 (0.272)
STEMI (0=no, 1=yes) / 0.352 (0.234) / −0.152 (0.388)
STEMI type unknown (0=no, 1=yes) / −0.602 (0.162) / −0.255 (0.219)
GFR <= 60 (0=no, 1=yes) / −0.898 (0.175) / −0.887 (0.245)
History of PCI (0=yes, 1=no) / −0.342 (0.18) / −0.219 (0.223)
History of CABG (0=yes, 1=no) / 0.53 (0.183) / −0.733 (0.262)
Time to randomization / 0.062 (0.023) / 0.113 (0.033)
Intercept / 12.061 (0.182) / 11.172 (0.281)
Weibull Shape (γ) / 1.003 (0.018) / 1.019 (0.029)

Table 8. Fit Statistics for Risk Equation Distributions

Group / Event / Distribution / -2log likelihood / AIC / BIC
ASA+CLO / CV Death / Exponential / 2133.355 / 2159.355 / 2256.174
Weibull / 2133.353 / 2161.353 / 2265.620
Log Normal / 2144.777 / 2172.777 / 2277.044
Stroke / Exponential / 1166.664 / 1192.664 / 1289.483
Weibull / 1166.466 / 1194.466 / 1298.732
Log Normal / 1169.053 / 1197.053 / 1301.319
MI / Exponential / 6765.964 / 6791.964 / 6888.783
Weibull / 6747.761 / 6775.761 / 6880.028
Log Normal / 6743.590 / 6771.590 / 6875.857
ASA Only / CV Death / Exponential / 938.316 / 964.316 / 1044.949
Weibull / 938.282 / 966.282 / 1053.117
Log Normal / 941.436 / 969.436 / 1056.270
Stroke / Exponential / 402.786 / 428.786 / 509.418
Weibull / 402.736 / 430.736 / 517.570
Log Normal / 399.471 / 427.471 / 514.306
MI / Exponential / 1664.679 / 1690.679 / 1771.311
Weibull / 1651.033 / 1679.033 / 1765.868
Log Normal / 1645.571 / 1673.571 / 1760.405

Table 9. Comparison of Discounted QALYs under Base Case versus under Each Equation Modeled with Best Statistical Fit

Under All Weibull (Base Case) / Under Best Fits
VOR+SC / SC / VOR+SC / SC
8.27 / 7.96 / 8.36 / 8.05

Appendix B. Comparison of Model Results with TRA 2°P

For the modeled population of patient profiles associated with the placebo arm of the qualifying MI cohort without a history of stroke or TIA, the model results were compared with the results observed in TRA 2°P among the corresponding population. To ensure that the model was as closely aligned as possible with the trial design, the following model settings were implemented:

· The time horizon of the model was set to 2.6 years, based on the median follow-up reported in Morrow et al. (trial publication) [19].

· For each patient profile, the model start time was set to each profile’s specific time to randomization/time since qualifying MI.

· The treatment duration of clopidogrel after an MI was set to an indefinite length (i.e. patients are on clopidogrel over the full 2.5 year time horizon), as opposed to the base case assumption of a 12-month clopidogrel treatment duration after an MI based on antiplatelet treatment guidelines. The duration was set to an indefinite length for this validation analysis, because discontinuation from clopidogrel was not fully tracked in TRA 2°P and only determined at the discretion of the clinician (i.e. cannot determine exactly when clopidogrel treatment was finished for all patients).

· The TRA 2°P risk equations were used for the full time horizon, and the reference risks derived from PLATO [17] and UK Hospital Episode Statistics [14] are not applied during the first 90 days after an MI.

· The reference bleeding rates were changed to represent the TIMI major, TIMI minor, intracranial hemorrhage, and fatal bleeds among patients in the qualifying MI cohort without a history of stroke or TIA on ASA+CLO at baseline (as shown in Table 10). As such, the clopidogrel-related hazard ratios from CAPRIE and CURE originally leveraged in the base case analysis were instead set to 1, because the treatment effect of clopidogrel on bleeding is not being considered for this validation analysis.

· The model structure for the original Q1 and Q2+ stroke health states was altered slightly. Instead of assigning an elevated risk of CV death during the first 90 days after a stroke based on UK Hospital Episode Statistics [14], and then subsequently adjusting the risk of non-CV death with a standardized mortality ratio [28], this validation analysis assumed that for each 90 day-cycle after a stroke, the risk of death was constant. The risk of death post-stroke was set to 0.72% for each 90-day cycle (based on the 3-year Kaplan-Meier estimate of 8.3% for all-cause mortality in the placebo care arm of the qualifying stroke cohort in TRA 2°P, data on file for the derivation) [19].