🏥 Hospital Readmission Prediction (Logistic Regression)
Author: Isaac Tosin Adisa
📌 Overview
This model predicts 30-day hospital readmission risk using structured clinical features derived from the MIMIC-IV dataset. It serves as the linear baseline in an integrated multi-model comparative framework alongside XGBoost and LightGBM, designed to evaluate the trade-offs between model complexity, predictive performance, calibration quality, explainability, and subgroup fairness.
The model outputs calibrated probabilities suitable for downstream clinical risk stratification workflows. As a logistic regression model, it offers strong interpretability by design — coefficients map directly to feature-level log-odds, making it transparent and auditable without requiring post-hoc explanation tools.
This model is released alongside a fully reproducible pipeline and open-source implementation to facilitate independent validation and reuse.
📊 Dataset
| Property | Value |
|---|
| Source | MIMIC-IV (v2.2) |
| Total admissions | 415,231 |
| 30-day readmission prevalence | ~18% |
| Feature count | 26 structured clinical features |
| Split | Train / Validation / Test (temporal split) |
Features include demographics, admission type, primary diagnosis category, comorbidity burden (Elixhauser), length of stay, lab value summaries, procedure counts, and prior utilization history.
⚙️ Training
| Setting | Value |
|---|
| Framework | scikit-learn |
| Solver | lbfgs |
| Regularization | L2 (tuned via cross-validation) |
| Class imbalance | class_weight="balanced" |
| Feature scaling | StandardScaler (applied pre-fit) |
| Calibration | Platt scaling (post-hoc) |
📈 Performance
| Metric | Value | Notes |
|---|
| AUC-ROC | ~0.67 | Linear baseline discrimination |
📊 Logistic Regression serves as the interpretable linear baseline in this framework. Its performance provides a lower-bound reference for evaluating the marginal gains of tree-based models (XGBoost, LightGBM) against the cost of reduced transparency.
🔍 Explainability
Logistic Regression is inherently interpretable — no post-hoc explanation method is required.
- Feature coefficients directly encode the direction and magnitude of each variable's contribution
- Odds ratios can be derived directly from model weights
- Compatible with standard clinical audit and regulatory review workflows
⚖️ Fairness Evaluation
The model was evaluated across 16 demographic and clinical subgroups, including stratifications by age group, sex, race/ethnicity, insurance type, and admission source.
All subgroups satisfy the following thresholds:
| Metric | Threshold |
|---|
| ΔAUC (vs. overall) | ≤ 0.05 |
| ΔFNR (vs. overall) | ≤ 0.10 |
No subgroup exhibited clinically meaningful performance degradation under these criteria.
🚀 Usage
1import joblib
2import numpy as np
3
4# Load model
5model = joblib.load("logreg.pkl")
6
7# Replace with your 26 clinical features (must be StandardScaler-transformed)
8X = np.array([[...]])
9
10# Returns 30-day readmission probability
11pred = model.predict_proba(X)[0][1]
12
13print(f"Readmission risk: {pred:.3f}")
⚠️ Input features must be scaled using the same
StandardScaler fitted during training before inference. See the
repository for the full feature schema and preprocessing pipeline.
🎯 Intended Use
- Linear baseline benchmarking against tree-based models
- Clinical ML interpretability research
- Demonstration of explainable and fair AI systems
- Reproducibility and model comparison
⚠️ Limitations
- Linear model — logistic regression cannot capture non-linear feature interactions present in complex clinical data; tree-based models may outperform it on discrimination metrics.
- Retrospective validation only — model was trained and evaluated on historical MIMIC-IV data; prospective validation has not been performed.
- Single institution — MIMIC-IV reflects one academic medical center (BIDMC); generalizability to other institutions requires local validation.
- No causal claims — feature associations do not imply clinical causation.
- Requires local validation before any deployment in a clinical decision support context.
- Credentialed dataset — MIMIC-IV requires PhysioNet credentialing; this model card does not distribute the underlying data.
🔗 Links
📜 Citation
1@misc{adisa2025readmission,
2 title={Hospital Readmission Prediction with Explainability and Fairness},
3 author={Adisa, Isaac Tosin},
4 year={2026},
5 eprint={2604.22535},
6 archivePrefix={arXiv},
7 primaryClass={cs.LG}
8}
License
This model is released under the
MIT License. The underlying MIMIC-IV dataset is subject to its own
PhysioNet credentialed access agreement.
