An Observational Study: The Effect of Diuretics Administration on Outcomes of Mortality and Mean Duration of I.C.U. Stay

TL;DR

This observational study investigates the impact of diuretics on ICU patient outcomes, introducing a reusable software approach for treatment outcome analysis and applying machine learning for prediction.

Contribution

It presents a stepwise, software-based methodology for analyzing treatment effects using observational data, and explores machine learning for outcome prediction.

Findings

Diuretics are associated with longer ICU stays in certain patient groups.

In less sick patients, diuretics significantly reduce 30-day mortality.

The study demonstrates a reusable approach for observational treatment effect analysis.

Abstract

This thesis conducts an observational study into whether diuretics should be administered to ICU patients with sepsis when length of stay in the ICU and 30-day post-hospital mortality are considered. The central contribution of the thesis is a stepwise, reusable software-based approach for examining the outcome of treatment vs no-treatment decisions with observational data. The thesis implements, demonstrates and draws findings via three steps: Step 1. Form a study group and prepare modeling variables. Step 2. Model the propensity of the study group with respect to the administration of diuretics with a propensity score function and create groups of patients balanced in this propensity. Step 3. Statistically model each outcome with study variables to decide whether the administration of diuretics has a significant impact. Additionally, the thesis presents a preliminary machine learning based method using Genetic Programming to predict mortality and length of stay in ICU outcomes for the study group. The thesis finds, for its study group, in three of five propensity balanced quintiles, a statistically significant longer length of stay when diuretics are administered. For a less sick subset of patients (SAPS ICU admission score < 17) the administration of diuretics has a significant negative effect on mortality.