SickleSense AI: An Ensemble Machine Learning Framework for Predicting Pain Severity in Sickle Cell Disease

Judging how severe a vaso-occlusive crisis (VOC) is in sickle cell disease (SCD) still depends largely on the patient's own pain rating, which is subjective and can differ from one patient or clinician to the next. This paper presents SickleSense AI, a multimodal clinical decision-support framework that sorts VOC severity into Mild, Moderate and Severe classes using only fifteen routinely recorded physiological and hematological measurements. Pain scores are kept out of the input entirely, so no label information leaks into the features. The work uses two cohorts: a real hospital cohort of 23 patients (N = 23) and a clinically constrained synthetic cohort of 5,000 records (N = 5,000) produced with a Gaussian Copula and constrained boundary sampling. The classifier is a two-level stacking ensemble in which Random Forest, Extra Trees and XGBoost act as base learners and a Logistic Regression meta-learner combines their outputs. On the augmented cohort the ensemble reached a five-fold cross-validated accuracy of 96.40% (±0.6%) and a macro F1-score of 0.9635. On an independent holdout of 1,000 records it identified all 333 severe cases, giving a severe-crisis recall of 100% with no false negatives. Feature importance analysis ranked hemoglobin (24.5%), respiratory rate (18.2%) and heart rate (16.4%) as the most influential predictors. Inference takes under 50 ms, which makes the tool a candidate for decision support in remote and tribal Primary Health Centres (PHCs). Because the evaluation relies largely on synthetic data, validation on larger independent real-world datasets is still required.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-14
DOI
https://doi.org/10.5281/zenodo.24217361
Primary Topic
Artificial Intelligence in Healthcare
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

SickleSense AI: An Ensemble Machine Learning Framework for Predicting Pain Severity in Sickle Cell Disease

Tushar Funde, Yatharth Shrirame, Rohan Gupta, Nayan Warhadkar
Zenodo (CERN European Organization for Nuclear Research)
Artificial Intelligence in Healthcare
preprint

SickleSense AI: An Ensemble Machine Learning Framework for Predicting Pain Severity in Sickle Cell Disease

Tushar Funde, Yatharth Shrirame, Rohan Gupta, Nayan Warhadkar
preprint en

Abstract

Judging how severe a vaso-occlusive crisis (VOC) is in sickle cell disease (SCD) still depends largely on the patient's own pain rating, which is subjective and can differ from one patient or clinician to the next. This paper presents SickleSense AI, a multimodal clinical decision-support framework that sorts VOC severity into Mild, Moderate and Severe classes using only fifteen routinely recorded physiological and hematological measurements. Pain scores are kept out of the input entirely, so no label information leaks into the features. The work uses two cohorts: a real hospital cohort of 23 patients (N = 23) and a clinically constrained synthetic cohort of 5,000 records (N = 5,000) produced with a Gaussian Copula and constrained boundary sampling. The classifier is a two-level stacking ensemble in which Random Forest, Extra Trees and XGBoost act as base learners and a Logistic Regression meta-learner combines their outputs. On the augmented cohort the ensemble reached a five-fold cross-validated accuracy of 96.40% (±0.6%) and a macro F1-score of 0.9635. On an independent holdout of 1,000 records it identified all 333 severe cases, giving a severe-crisis recall of 100% with no false negatives. Feature importance analysis ranked hemoglobin (24.5%), respiratory rate (18.2%) and heart rate (16.4%) as the most influential predictors. Inference takes under 50 ms, which makes the tool a candidate for decision support in remote and tribal Primary Health Centres (PHCs). Because the evaluation relies largely on synthetic data, validation on larger independent real-world datasets is still required.

Zenodo (CERN European Organization for Nuclear Research)
Raisoni Group of Institutions (IN)
Artificial Intelligence in Healthcare
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon