Project summary
Under the current clinical paradigm, the majority of patients sharing a pathology tend to be treated in a similar
manner according to clinical guidelines that are based on previously conducted clinical trials combined into meta-
analyses. While some situations are amendable to the stratification of care, that is using more or less intensive
therapy based on the presence of specific risk factors, true personalization of care (i.e. therapeutic or
management selected based on a comprehensive review of patient characteristics and possibly including
patient-specific prediction models) remains exceedingly rare despite the potential for improved patient-level
outcomes. One important question in this regard that has not yet been answered is the extent to which a
personalized approach would result in clinical benefits should it be used in a large number of patients. At this
time, given the paucity of examples of large scale implementation of personalized care, it is not possible to
directly provide an answer; however, we could use existing data to generate a reliable approximation through
computer simulations. Thus, we propose to use data from ~130 previously published NHLBI funded randomized
clinical trials to simulate the effect of personalized medicine and compare the group-level outcomes to results
expected in the same patient population without using a personalized approach to treatment choice. Specifically,
for each clinical trial included in this study, we will create arm-specific prediction models for the primary outcome
and apply it to the opposite study group, thus estimating the theoretical, patient-specific probability of achieving
the primary outcome had they been assigned to the opposite trial arm. Simulations will then be performed
separately for all trials where patients are respectively assigned to: 1) the treatment arm of the trial, 2) the control
arm of the trial or 3) to whatever arm carries the lowest probability of adverse outcomes (i.e. predictive allocation).
We will then calculate the net benefit of predictive allocation by comparing the cumulative prevalence of
outcomes in that simulation vs. either the simulation where all patients are assigned to either the treatment arm
(for positive trials) or where all patients are assigned to the control arm (for negative trials). Finally, we will
compile the data from all included trials and identify factors that are associated with changes in the net benefit
of predictive allocation, including both trial-specific risk factors and performance metrics of the prediction model
used for patient allocation. This study will allow us, for the first time, to estimate the potential improvement, at
the population level, that would be associated with the widespread utilization of a personalized approach to
treatment choice. We will also generate crucial information in regards to the clinical scenarios and situations
where such an approach would generate the highest benefits. This information will be essential for the efficient
and targeted implementation of future personalized medicine programs.