± HEALTH In our modern society, digital devices, social media platforms, and artificial intelligence (AI) tools have b…+ CLIMATE The growing complexity of electric vehicle charging station (EVCS) operations—driven by grid constraints, ren…± HEALTH Objectives This study aim to develop, compare and internally validate machine-learning models for predicting…+ HEALTH Diabetes management faces a number of obstacles, such as fragmented healthcare data, privacy concerns, poor e…− SCIENCE Background Craniofacial soft tissue thickness (CFSTT) is a critical parameter in forensic facial reconstructi…+ HEALTH Background Technological innovation in total joint arthroplasty (TJA) has largely focused on intraoperative p…± HEALTH Infective endocarditis (IE) continues to be an often fatal condition despite improvements in cardiac surgical…± HEALTH Purpose of review Systemic sclerosis-associated interstitial lung disease (SSc-ILD) is the leading cause of d…
TruaceTracing the truth around AIFriday, September 11, 2026
Subscribe
Health·G Space·Evidence-backed gain·Published 2026-09-10

Advancing preoperative planning technology in total joint arthroplasty with a real-time machine learning calculator: 1-year mortality risk in a value-based care era

Abstract: Background Technological innovation in total joint arthroplasty (TJA) has largely focused on intraoperative precision through robotics, navigation, and implant design, while preoperative decision-making remains comparatively underdeveloped. Accurate estimation of patient-specific risk is central to surgical indications, yet existing tools provide limited resolution for consequential outcomes such as 1-year mortality. Methods A machine learning model was developed using the TriNetX Research Network. Patients unde…

TRV-2026-1050Peer-reviewedPermanent record — cite & verify
Advancing preoperative planning technology in total joint arthroplasty with a real-time machine learning calculator: 1-year mortality risk in a value-based care era

Flickr - Official U.S. Navy Imagery - Sgt. Joshua Elliott uses a Primus RS machine during physical therapy. by Official Navy Page from United States of America MCSN Joseph A. Boomhower/U.S. Navy. Public domain

The quick read

In a study published September 9, 2026, investigators built a machine learning calculator to estimate 1-year mortality after primary total knee and hip arthroplasty using 43 routinely available preoperative variables from the TriNetX Research Network. On internal validation the model achieved AUROC 0.761 with Brier score 0.006, and stratified risk monotonically from 0.574% overall to 3.57% in the top 5% of predicted risk.

The work matters because preoperative decision-making in joint replacement has lagged behind intraoperative precision tools, and individualized mortality estimation could inform surgical indications and value-based care discussions. Uncertainty remains because results are limited to internal validation with modest overprediction in the highest decile, and the authors note that prospective deployment depends on future external validations.

Main points
  • Study used TriNetX Research Network patients undergoing TKA or THA with verifiable 1-year follow-up; observed mortality was 0.574% (n=1,344).
  • Model trained with nested cross-validation and isotonic calibration; Brier score 0.006 and average precision 0.037 on internal validation.
  • Risk stratification showed monotonic increase: highest quintile 1.488% mortality, highest decile 2.50%, top 5% 3.57%.
  • Top predictors by SHAP were age (mean |SHAP| = 0.407), male sex (0.218), serum creatinine (0.162), hemoglobin (0.153), and BMI (0.145).
Gain

An XGBoost model using 43 routine preoperative variables estimated 1-year mortality after total knee and hip arthroplasty with AUROC 0.761 and stable calibration, stratifying patients so the top 5% had 6.2-fold higher mortality than baseline.

The rundown

Researchers developed an XGBoost model incorporating 43 preoperative variables from the TriNetX Research Network for patients undergoing total knee or hip arthroplasty with verifiable 1-year follow-up. Training used nested cross-validation with isotonic calibration, and feature contributions were assessed using SHapley Additive exPlanations (SHAP).

Performance included a sensitivity-anchored threshold of 0.235% that captured 90.1% of all deaths and a Youden-optimal threshold of 0.767% that achieved 83.9% specificity. Calibration remained stable across risk deciles with mean absolute error below 0.15% for D1-D9, while the web calculator was derived from machine-learning-derived inputs for point-of-care use.

Sources

Reader signal

How should this claim be treated?

The debate