AI as an intervention: improving clinical outcomes relies on a causal approach to AI development and validation.

The primary practice of healthcare artificial intelligence (AI) starts with model development, often using state-of-the-art AI, retrospectively evaluated using metrics lifted from the AI literature like AUROC and DICE score. However, good performance on these metrics may not translate to improved clinical outcomes. Instead, we argue for a better development pipeline constructed by working backward from the end goal of positively impacting clinically relevant outcomes using AI, leading to considerations of causality in model development and validation, and subsequently a better development pipeline.

A comparison of phage susceptibility testing with two liquid high-throughput methods.

Phage therapy is a promising antibacterial strategy, especially given that drug-resistant bacterial infections are escalating worldwide. Because phages are not active against all strains of a given species, phages being considered for therapeutic use would ideally be tested against bacterial isolates from individual patients prior to administration. Standardized, clinically validated phage susceptibility testing (PST) methods are needed for assessing phage activity.

Pediatric Critical Care Illness Severity Toolkit: Stata Commands for Calculation of Pediatric Index of Mortality and Pediatric Logistic Organ Dysfunction Scores.

Introduction: Illness severity scoring tools, such as PRISM III/IV, PIM-3, and PELOD-2, are widely used in pediatric critical care research. However, their application is hindered by complex calculation processes, privacy concerns with third-party online calculators, and challenges in accurate implementation within statistical packages.

Methods: We have developed a comprehensive, open-source toolkit for implementing the PIM-3, Simplified PIM-3, and PELOD-2 scores.

Decrypting the molecular basis of cellular drug phenotypes by dose-resolved expression proteomics.

Proteomics is making important contributions to drug discovery, from target deconvolution to mechanism of action (MoA) elucidation and the identification of biomarkers of drug response. Here we introduce decryptE, a proteome-wide approach that measures the full dose-response characteristics of drug-induced protein expression changes that informs cellular drug MoA. Assaying 144 clinical drugs and research compounds against 8,000 proteins resulted in more than 1 million dose-response curves that can be interactively explored online in ProteomicsDB and a custom-built Shiny App.