15 February 2020 by James West
Baseline risk evaluation predicts likelihood of clinical worsening: A sample size re-analysis from Freedom-EV
Background: Multiparametric risk analyses like the Reveal 2 or French non-invasive (French) risk scores predict outcomes for patients with pulmonary arterial hypertension (PAH). Freedom-EV was a global, event driven study of oral treprostinil in participants who had recently started monotherapy as treatment for PAH. In Freedom-EV, we prospectively planned to analyze the change in the French non-invasive risk score, and, post-hoc, we also analyzed the change in Reveal 2 score. For this post-hoc analysis, we hypothesized that participants who had clinical worsening (CW) events started with a higher baseline risk score than those who did not. We hypothesized that enriching a study selecting only those with higher risk scores would allow a smaller number of participants.
Methods: We calculated the French and Reveal 2 risk scores for four groups: active treatment with CW events, active without events, placebo with events, placebo without events. We used Fisher’s exact analysis to compare categorical French risk scores and a non-parametric ANCOVA to analyze continuous Reveal 2 scores. We re-analyzed sample size enrolling only participants with 0 or 1 French low risk factors with an underlying hazard ratio of 0.625 (observed among higher risk participants in Freedom-EV) and a placebo median event time of 20 months. We assumed accrual during 36 months (total study duration 42 months) with a 19% active dropout rate and 10% placebo dropout.
Results: Both scores demonstrated statistically higher risk among those with CW events as compared to those without. Our analyses suggest that we would have 90% power to demonstrate the same hazard ratio with 191 CW events and 430 higher risk participants (French score of 0 or 1 low risk factors) as compared to the 205 CW events and 690 participants actually enrolled.
Conclusion: For event driven studies, enriching for those with higher baseline risk may be reasonable (For table, see 'CTA - view full pdf')
** Reveal 2 difference between groups, ANCOVA, p < 0.001, ## French difference between groups, Fisher’s exact, p < 0.001.