https://www.thelancet.com/journals/landig/article/PIIS2589-7500(24)00044-X/fulltext

Evaluation of a machine-learning model based on laboratory parameters for the prediction of acute leukaemia subtypes: a multicentre model development and validation study in France

Lancet Digit Health. 2024 May;6(5):e323-e333. PMID: 38670741 DOI: 10.1016/S2589-7500(24)00044-X

Vincent Alcazer, Grégoire Le Meur, Marie Roccon, Sabrina Barriere, Baptiste Le Calvez, Bouchra Badaoui, Agathe Spaeth, Olivier Kosmider, Nicolas Freynet, Marion Eveillard5, Carolyne Croizier, Simon Chevalier, Pierre Sujobert

Methods

We conducted a multicentre, model development, and validation study based on 19 routine laboratory parameters collected at disease onset in 1410 acute leukaemia patients from six independent French University Hospitals. Using training (n=679) and external validation (n=731) cohorts, several machine learning models were evaluated with a custom sensitivity analysis for variable selection. An additional prospective cohort of n=66 patients was also used for further validation.

Based on ten routine laboratory parameters, our final eXtreme Gradient Boosting (XGB)-model showed an AUC [95%CI] of 0.97 [0.95-0.99], 0.90 [0.83-0.97], and 0.89 [0.82-0.95] for APL, ALL, and AML diagnoses, respectively. Optimal cutoffs to guide clinical decisions were then set, leading to an accuracy of 99.7/99.5/98.8% for confident predictions and 96.1/87.9/86.3% for overall predictions of APL, ALL, and AML, respectively. These results were confirmed in the prospective cohort. The final model was integrated into a web-app with a user-friendly graphical interface, AI-PAL.

AI-PAL is a free and open-source software package built in R, with a user-friendly interface provided via Shiny, that enables clinical hematologists and biologists to diagnose the three main subtypes of acute leukemia based solely on routine biological parameters.

AI-PAL has a ready-to-use online version available at https://alcazerv.shinyapps.io/AIPAL/.

You can install a local version from GitHub either by cloning the repository or directly by downloading the package in R: You’ll need to have R (>= 4.1.0) and the remotes package installed.

install.packages("remotes")
remotes::install_github("VincentAlcazer/AIPAL")

AIPAL::run_app()

The AI-PAL Shiny app will open in your default web browser.

/! This work is currently not published and is available for personal use or review only. /! 

If you encounter any problem with the software or find a bug, please report it on GitHub:

• Create a new issue on the Github page
• Try to describe the problem/bug with reproductible steps