Survival Meets Classification: A Novel Framework for Early Risk Prediction Models of Chronic Diseases

Shaheer Ahmad Khan; Muhammad Usamah Shahid; Muddassar Farooq

cs.LGMar 12, 2026

Survival Meets Classification: A Novel Framework for Early Risk Prediction Models of Chronic Diseases

Authors: Shaheer Ahmad Khan, Muhammad Usamah Shahid, Muddassar Farooq

Organizations: CureMD Research, 80 Pine St 21st Floor, New York, NY 10005, United States

Abstract

Chronic diseases are long-lasting conditions that require lifelong medical attention. Using big EMR data, we have developed early disease risk prediction models for five common chronic diseases: diabetes, hypertension, CKD, COPD, and chronic ischemic heart disease. In this study, we present a novel approach for disease risk models by integrating survival analysis with classification techniques. Traditional models for predicting the risk of chronic diseases predominantly focus on either survival analysis or classification independently. In this paper, we show survival analysis methods can be re-engineered to enable them to do classification efficiently and effectively, thereby making them a comprehensive tool for developing disease risk surveillance models. The results of our experiments on real-world big EMR data show that the performance of survival models in terms of accuracy, F1 score, and AUROC is comparable to or better than that of prior state-of-the-art models like LightGBM and XGBoost. Lastly, the proposed survival models use a novel methodology to generate explanations, which have been clinically validated by a panel of three expert physicians.

Figures & tables

Disease	ICD10 codes	Patient Count
Hypertension (HTN)	I10-I13	54654
Type 2 Diabetes Mellitus (DM)	E11	31865
Chronic Kidney Disease (CKD)	N18, I12, I13, [E08-E13].22	15316
Chronic Ischemic Heart Disease (CHD)	I25	13804
Chronic Obstructive Pulmonary Disease (COPD)	J41-J44	15131

Table 1: Patient counts for each chronic disease that were eligible for model training, and their respective criteria

Panel of Figure 1 — Figure 1: Distribution of observation times for hypertension using the three described approaches. Similar plots for the remaining diseases can be found in Appendix 0.B .

	Classifiers			Random survival forest
Disease	Random Forest	XGB	LGBM	C-Index	F1 (RS)	F1 (LN)	F1 (SP)
Hypertension	0.728	0.751	0.751	0.689	0.762	0.749	0.743
Heart	0.730	0.738	0.753	0.667	0.750	0.750	0.742
CKD	0.726	0.735	0.748	0.658	0.749	0.750	0.742
COPD	0.726	0.737	0.743	0.675	0.737	0.750	0.721
Diabetes	0.741	0.752	0.752	0.688	0.766	0.749	0.735

Table 2: Evaluation of Approach 1: Similar on the validation set.

Panel of Figure 2 — Figure 2: Average survival curves for the hypertension training set. Clusters are made using the true and predicted labels found using survival probability for classification. Similar curves for the remaining diseases can be found in Appendix 0.C .

	Classifiers			Random survival forest
Disease	Random Forest	XGB	LGBM	C-Index	F1 (RS)	F1 (LN)	F1 (SP)
Hypertension	0.728	0.744	0.744	0.708	0.762	0.747	0.750
Heart	0.799	0.808	0.814	0.720	0.813	0.815	0.796
CKD	0.768	0.787	0.787	0.718	0.790	0.792	0.796
COPD	0.788	0.785	0.790	0.731	0.794	0.798	0.787
Diabetes	0.759	0.773	0.767	0.729	0.780	0.768	0.774

Table 3: Evaluation of Approach 2: Overlap on the validation set.

	Classifiers			Random survival forest
Disease	Random Forest	XGB	LGBM	C-Index	F1 (RS)	F1 (LN)	F1 (SP)
Hypertension	0.712	0.741	0.743	0.720	0.747	0.741	0.742
Heart	0.763	0.766	0.784	0.716	0.788	0.786	0.772
CKD	0.757	0.763	0.775	0.718	0.772	0.775	0.777
COPD	0.769	0.772	0.773	0.729	0.775	0.778	0.761
Diabetes	0.757	0.767	0.770	0.745	0.780	0.770	0.775

Table 4: Evaluation of Approach 3: Distinct on the validation set.

Figure 3: Average F1 scores of LGBM and RSF using the three classification techniques

Disease	C Index	Accuracy	Precision	Recall	NPV	Specificity	AUROC	AUPRC	F1 score
Hypertension	0.709	0.742	0.723	0.779	0.764	0.705	0.828	0.819	0.755
Heart	0.741	0.788	0.758	0.838	0.823	0.739	0.869	0.852	0.819
CKD	0.729	0.789	0.767	0.827	0.814	0.751	0.870	0.859	0.796
COPD	0.730	0.784	0.761	0.815	0.809	0.753	0.869	0.871	0.799
Diabetes	0.728	0.784	0.819	0.733	0.756	0.836	0.872	0.896	0.778

Table 5: Performance of the Random Survival Forest on the test set.

Panel of Figure 4 — Figure 4: Feature importances found using SurvSHAP and our custom implementation. Similar plots for the remaining diseases can be found in Appendix 0.D .

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Feature Name	Feature Description
age_group	Categorizes the age of a person into pre-defined interval or bins
gender	Identifies the gender of a person
race_mapping	A numerical representation of an individual’s race
bmi_group	Categorizes the BMI of person into pre-defined interval or bins
map_group	Categorizes mean arterial pressure (MAP) of a person into pre-defined intervals or bins

Appendix

Table 6: Common Features included in all models

Feature Name	Feature Description
E78	Code for Disorders of lipoprotein metabolism and other lipidemias (Present/Absent)
I25	Code for Chronic ischemic heart disease (Present/Absent)
K21	Code for Gastro-esophageal reflux disease (Present/Absent)
K29	Code for Gastritis and duodenitis (Present/Absent)
N17	Code for Acute kidney failure (Present/Absent)
R60	Code for Edema, not elsewhere classified (Present/Absent)

Appendix

Table 7: Diabetes Features - Part 1

Feature Name	Feature Description
dexamethasone	Medications under GPI code 2210 (Prescribed/Not Prescribed)
sleep_disorder	Custom Feature Group for codes G47, F51 and Z72.820 (Present/Absent)
horomonal_antineoplastic	Medications under GPI code 2140 (Prescribed/Not Prescribed)
smoking_history	Custom Feature Group indicating smoking history (Present/Absent)
family_diabetes	Family History of Diabetes (Present/Absent)
metabolic_disorder	Custom Feature Group for codes E88 and Z86.3 (Present/Absent)

Appendix

Table 8: Diabetes Features - Part 2

Feature Name	Feature Description
E11	Code for Type 2 diabetes mellitus (Present/Absent)
E78	Code for Disorders of lipoprotein metabolism and other lipidemias (Present/Absent)
G89	Code for Pain, not elsewhere classified (Present/Absent)
I25	Code for Chronic ischemic heart disease (Present/Absent)
F41	Code for Other anxiety disorders (Present/Absent)
K29	Code for Gastritis and duodenitis (Present/Absent)

Appendix

Table 9: Hypertension Features

Feature Name	Feature Description
E11	Code for Type 2 diabetes mellitus (Present/Absent)
D50	Code for Iron deficiency anemia (Present/Absent)
R53	Code for Malaise and fatigue (Present/Absent)
I25	Code for Chronic ischemic heart disease (Present/Absent)
R06	Code for Abnormalities of breathing (Present/Absent)
R73	Code for Elevated blood glucose level (Present/Absent)

Appendix

Table 10: CKD Features

Feature Name	Feature Description
DM	Elixhauser Comorbidity Group for Diabetes without chronic complications (Present/Absent)
D50	Code for Iron deficiency anemia (Present/Absent)
HTN	Elixhauser Comorbidity Group for Primary hypertension (Present/Absent)
E78	Code for Disorders of lipoprotein metabolism and other lipidemias (Present/Absent)
R51	Code for Headache (Present/Absent)
R06	Code for Abnormalities of breathing (Present/Absent)

Appendix

Table 11: Heart Features

Feature Name	Feature Description
DM	Elixhauser Comorbidity Group for Diabetes without chronic complications (Present/Absent)
R07	Code for Pain in throat and chest (Present/Absent)
R09	Code for Other symptoms and signs involving the circulatory and respiratory system (Present/Absent)
R53	Code for Malaise and fatigue (Present/Absent)
E78	Code for Disorders of lipoprotein metabolism and other lipidemias (Present/Absent)
F41	Code for Other anxiety disorders (Present/Absent)

Appendix

Table 12: COPD Features

Explore similar work

Subtyping patients with chronic disease using longitudinal BMI patterns
Nov 9, 2021Md Mozaharul Mottalib, Jessica C Jones-Smith, Bethany Sheridan +1Clinical Risk Prediction
Automatic Cardiac Risk Management Classification using large-context Electronic Patients Health Records
Mar 10, 2026Jacopo Vitale, David Della Morte, Luca Bacco +5Clinical Risk Prediction Longitudinal EHR Modeling
Research Unit of Intelligent Health Technologies, Universit`a Campus Bio-Medico di Roma, Rome, Italy. · Central Diagnostic Laboratory, University Medical Center Utrecht, Utrecht, The Netherlands.
Scalable Clinical Data Infrastructure and Comparative ML Evaluation for Hospitalisation Risk Prediction in Elderly Patients with Multiple Long-Term Conditions using CPRD
Aug 29, 2026Asra Aslam, Volodymyr Chapman, Maurice M. O'Connell +9Clinical Prediction Clinical Risk Prediction
School of Information, University of Sheffield, Sheffield, United Kingdom · School of Medicine, Faculty of Medicine and Health, University of Leeds, Leeds, United Kingdom · Division of Informatics, University of Manchester, Manchester, United Kingdom +2

Nov 9, 2021J/K move · Enter open · S save

Subtyping patients with chronic disease using longitudinal BMI patterns

Md Mozaharul Mottalib, Jessica C Jones-Smith, Bethany Sheridan, Rahmatollah Beheshti

Clinical Risk Prediction

Open paper AskPDF arXiv

Obesity is a major health problem, increasing the risk of various major chronic diseases, such as diabetes, cancer, and stroke. While the role of obesity identified by cross-sectional BMI recordings has been heavily studied, the role of BMI trajectories is much less explored. In this study, we use a machine-learning approach to subtype individuals' risk of developing 18 major chronic diseases by using their BMI trajectories extracted from a large and geographically diverse EHR dataset capturing the health status of around two million individuals for a period of six years. We define nine new interpretable and evidence-based variables based on the BMI trajectories to cluster the patients into subgroups using the k-means clustering method. We thoroughly review each cluster's characteristics in terms of demographic, socioeconomic, and physiological measurement variables to specify the distinct properties of the patients in the clusters. In our experiments, the direct relationship of obesity with diabetes, hypertension, Alzheimer's, and dementia has been re-established and distinct clusters with specific characteristics for several of the chronic diseases have been found to be conforming or complementary to the existing body of knowledge.

Abstract

Explore similar work

Subtyping patients with chronic disease using longitudinal BMI patterns

Automatic Cardiac Risk Management Classification using large-context Electronic Patients Health Records

Scalable Clinical Data Infrastructure and Comparative ML Evaluation for Hospitalisation Risk Prediction in Elderly Patients with Multiple Long-Term Conditions using CPRD