Research & Resources

Our work is grounded in peer-reviewed research and validated with real hospital data. We publish our findings to advance the field and demonstrate that our AI delivers measurable clinical outcomes.

Published Research

Peer-Reviewed 2024

Development and validation of 'Patient Optimizer' (POP) algorithms for predicting surgical risk with machine learning

Kowadlo G, Mittelberg Y, Ghomlaghi M, Stiglitz DK, Kishore K, Guha R, Nazareth J, Weinberg L

BMC Medical Informatics and Decision Making · Vol. 24, Article 70

Abstract

Developed machine learning algorithms to assess pre-operative risk and predict post-operative complications using data from 11,475 adult surgical admissions (2015–2020). Logistic regression and XGBoost models achieved strong performance for predicting complications (AUROC 0.755) and length-of-stay (AUROC 0.841). The study validated that small datasets with limited features can produce clinically useful predictions, and introduced multivalue length-of-stay prediction rather than binary classification.

Read full paper
Preprint 2026

Trustworthy personalized treatment selection: causal effect-trees and calibration in perioperative medicine

Mittelberg Y, Stiglitz DK, Kowadlo G

medRxiv · Posted March 4, 2026

Abstract

Developed a deployment-readiness framework integrating causal inference, interpretable effect-trees, and calibration assessment to distinguish actionable signal from unreliable variation. Using 130,000+ surgical operations from the INSPIRE perioperative dataset (2011–2020), estimated treatment effects using causal forests with double machine learning. In a prostate procedures case study, neuraxial anesthesia was associated with substantially lower post-operative opioid use (ATE = −1.38 medications). The framework produced clinically interpretable subgroups and identified which were reliable for deployment through calibration analysis.

Read full paper

Tools

Free Tool

MedData Tracker

An automated system for discovering, evaluating, and curating open medical datasets relevant to perioperative medicine and blood test reference ranges.

Automated discovery — scans PhysioNet, Zenodo, PubMed, and the open web daily for new datasets
AI scoring — Claude evaluates each dataset for relevance to perioperative medicine, blood test reference ranges, and related domains
Structured metadata — availability, licensing, sample size, variable categories, acquisition instructions, and more
Human-curated — every dataset is reviewed and approved by a domain expert before publication
Explore MedData Tracker

Guides

Clinical and technical guides coming soon. In the meantime, visit our blog for insights on AI in perioperative care.