Randomized G-Computation Models in Healthcare Systems

Healthcare system quality improvements depend both on the availability of innovative technologies and on proper investments to transfer experimental policies into daily practices that could be easily adopted in all hospitals. Unfortunately, funds are generally not enough to cover all the addressable issues and the policy makers are faced with the difficult problem to decide where to allocate the money to produce the most relevant positive outcomes. To support this decision process, data gathering, and analysis play a key role. In this contribution we propose a simplified pipeline that starting from observational data to achieve statistical conclusions as valid as in designed randomized studies. After detailing the proposed analytic method, its soundness is proved using an important case study: the problem of the reduction of Healthcare-Associated Infections, and especially those acquired in Intensive Care Units. In particular, we show how to estimate the preventable proportion of Intubation-Associated Pneumonia in ICUs. In our study, using G-Computation based approach, we found out that the preventable proportion for IAP is of 44%. Interestingly, when bundle compliance is added in the statistical model, the preventable proportion for IAP is of 40%.