Interpretable Multiple Myeloma Prognosis with Observational Medical Outcomes Partnership Data

Salma Rachidi; Aso Bozorgpanah; Eric Fey; Alexander Jung

cs.LGMar 20, 2026

Interpretable Multiple Myeloma Prognosis with Observational Medical Outcomes Partnership Data

Authors: Salma Rachidi, Aso Bozorgpanah, Eric Fey, Alexander Jung

Organizations: Aalto University, Dept. of Computer Science, Espoo, Finland · Helsinki University Hospital, Helsinki, Finland

Abstract

Machine learning (ML) promises better clinical decision-making, yet opaque model behavior limits the adoption in healthcare. We propose two novel regularization techniques for ensuring the interpretability of ML models trained on real-world data. In particular, we consider the prediction of five-year survival for multiple myeloma patients using clinical data from Helsinki University Hospital. To ensure the interpretability of the trained models, we use two alternative constructions for a penalty term used for regularization. The first one penalizes deviations from the predictions obtained from an interpretable logistic regression method with two manually chosen features. The second construction requires consistency of model predictions with the revised international staging system (R-ISS). We verify the usefulness of the proposed regularization techniques in numerical experiments using data from 812 patients. They achieve an accuracy up to 0.721 on a test set and SHAP values show that the models rely on the selected important features.

Figures & tables

Aspect	Description
Data Points	812 myeloma patients treated at HUS who have known 5-year survival status and available values for albumin, LDH and B2M
Features	Age at diagnosis, 16 blood measurements (albumin, LDH, B2M, hemoglobin, platelets, creatinine, C reactive protein, alkaline phosphatase, free lambda light chains, free kappa light chains, leukocytes, protein, ionized calcium, immunoglobulin A, immunoglobulin G and immunoglobulin M) and percentage of plasma cells in bone marrow
Label	Binary : 1 if the patient died within 5 years from diagnosis, 0 if the patient survived at least 5 years
Data distribution	417 data points with label 1 and 395 with label 0 stage 1: 18/111 (16.2 % ) with label 1 stage 2: 325/595 (54.6 %) with label 1 stage 3: 74/106 (69.8 %) with label 1

TABLE I : Summary of the dataset

	Auxiliary model		Stage-based		$\bm{\alpha=0}$
Dataset	Acc	AUC	Acc	AUC	Acc	AUC
$\mathcal{D}^{(\rm kf)}$	0.646	0.645	0.634	0.627	0.736	0.734
$\mathcal{D}^{(\rm test)}$	0.697	0.696	0.631	0.621	0.689	0.686

TABLE II : Metrics on baselines

Panel of Figure 3 — Fig. 1 : Comparison of scores and loss values with respect to $\alpha$ for models using the auxiliary-alignment regularizer ( 2 ).

Rank	1st	2nd	3rd
$\alpha$
0	lambda	age	LDH
1	age	LDH	kappa
2	age	LDH	albumin
3	age	LDH	hemoglobin
4	age	LDH	hemoglobin

TABLE III : Summary of SHAP value ranks for the auxiliary-alignment regularizer. The columns show which features were ranked 1st, 2nd or 3rd for each value of alpha according to the mean absolute Shapley values on

\mathcal{D}^{(\rm test)}

. ’lambda’ and ’kappa’ correspond to the respective free light chains.

Panel of Figure 5 — Fig. 2 : Comparison of scores and loss values with respect to $\alpha$ for models using the stage-consistency regularizer ( 3 ).

Feature	LDH	Albumin	B2M	Age
$\alpha$
0	3rd	5th	8th	2nd
1	1st	2nd	5th	3rd
2	1st	2nd	3rd	7th
3	1st	2nd	3rd	7th
4	1st	2nd	3rd	11th

TABLE IV : Summary of SHAP value ranks for the stage-consistency regularizer. The columns show the ranks of the features used in Algorithm 1 for each value of alpha according to the mean absolute Shapley values on

\mathcal{D}^{(\rm test)}

Interpretable ML

Interpretable machine learning develops models and explanations that make predictions, evidence, and decision processes understandable to people.

See all 306 papers

Explainable Machine Learning for Sepsis Outcome Prediction Using a Novel Romanian Electronic Health Record Dataset
Apr 6, 2026Andrei-Alexandru Bunea, Ovidiu Ghibea, Dan-Matei Popovici +2Sepsis Prediction Clinical Outcome Prediction
POLITEHNICA Bucharest National University of Science and Technology, Bucharest, Romania · Carol Davila University of Medicine and Pharmacy, Faculty of Medicine, Bucharest, Romania · General Surgery Department, University Emergency Hospital of Bucharest, Bucharest, Romania +1
Retrieval-Augmented Interpretable Learning: Towards Task-Specific Zero-Shot Models in Healthcare
Jul 20, 2026Sazan Mahbub, Caleb N. Ellington, Zhiyuan Li +4Zero-Shot Learning Meta-Learning
Carnegie Mellon University · GenBio AI · University of Wisconsin–Madison +2
Enhancing SHAP Explainability for Diagnostic and Prognostic ML Models in Alzheimer Disease
Mar 6, 2026Pablo Guillén, Enrique Frias-MartinezExplainability Evaluation Explanation Stability
MIU City University, Miami, FL, USA · Research Institute UNIR iTED, Universidad Internacional de La Rioja (UNIR), Spain

Apr 6, 2026J/K move · Enter open · S save

Explainable Machine Learning for Sepsis Outcome Prediction Using a Novel Romanian Electronic Health Record Dataset

Andrei-Alexandru Bunea, Ovidiu Ghibea, Dan-Matei Popovici, Ion Daniel+1

POLITEHNICA Bucharest National University of Science and Technology, Bucharest, Romania · Carol Davila University of Medicine and Pharmacy, Faculty of Medicine, Bucharest, Romania · General Surgery Department, University Emergency Hospital of Bucharest, Bucharest, Romania+1

Sepsis Prediction Clinical Outcome Prediction Interpretable ML

Open paper AskPDF arXiv

We develop and analyze explainable machine learning (ML) models for sepsis outcome prediction using a novel Electronic Health Record (EHR) dataset from 12,286 hospitalizations at a large emergency hospital in Romania. The dataset includes demographics, International Classification of Diseases (ICD-10) diagnostics, and 600 types of laboratory tests. This study aims to identify clinically strong predictors while achieving state-of-the-art results across three classification tasks: (1)deceased vs. discharged, (2)deceased vs. recovered, and (3)recovered vs. ameliorated. We trained five ML models to capture complex distributions while preserving clinical interpretability. Experiments explored the trade-off between feature richness and patient coverage, using subsets of the 10--50 most frequent laboratory tests. Model performance was evaluated using accuracy and area under the curve (AUC), and explainability was assessed using SHapley Additive exPlanations (SHAP). The highest performance was obtained for the deceased vs. recovered case study (AUC=0.983, accuracy=0.93). SHAP analysis identified several strong predictors such as cardiovascular comorbidities, urea levels, aspartate aminotransferase, platelet count, and eosinophil percentage. Eosinopenia emerged as a top predictor, highlighting its value as an underutilized marker that is not included in current assessment standards, while the high performance suggests the applicability of these models in clinical settings.

Abstract

Interpretable ML

Explore similar work

Explainable Machine Learning for Sepsis Outcome Prediction Using a Novel Romanian Electronic Health Record Dataset

Retrieval-Augmented Interpretable Learning: Towards Task-Specific Zero-Shot Models in Healthcare

Enhancing SHAP Explainability for Diagnostic and Prognostic ML Models in Alzheimer Disease