EUVIMEDSkopri EUVIMED
Open access · CC BY 4.0

Predicting cancer-related INA-CBG classifications using explainable and uncertainty-aware machine learning: A retrospective hospital-based study

Suryanto Nugroho, Raden Venantius Hari Ginardi, I Ketut Eddy Purnama

Array · 2026-12 · 32 · 101414

DOI: 10.1016/j.array.2026.101414 ↗

At a glance

Journal article in Array, 2026-12. Written by Suryanto Nugroho and 2 co-authors. 47 references. The full text is freely available (open access, CC BY 4.0).

AstrattDettalji biblijografiċiCiteRelated studies

Abstract (original language)

Accurate and transparent reconstruction of Indonesia Case-Based Groups (INA-CBG) classifications is relevant to hospital case-mix management and administrative quality assurance. This study evaluated the recoverability of recorded cancer-related INA-CBG classifications using explainable and uncertainty-aware machine learning. A retrospective analysis was conducted using a single-hospital inpatient repository containing 165,391 hospitalization episodes recorded between November 2016 and April 2026. After cohort selection, 20,098 cancer-related episodes across 11 INA-CBG categories were included. Twenty-nine categorical predictors comprising 12 ICD-10 diagnosis fields and 17 ICD-9-CM procedure fields were used to compare logistic regression, random forest, XGBoost, and LightGBM. A predictor-signature-controlled partitioning strategy separated 12,884 development episodes from 7,214 held-out test episodes, preventing identical coding configurations from appearing in both partitions. Logistic regression achieved the highest performance in the primary comparison, with an accuracy of 99.43%, balanced accuracy of 0.8664, and macro-F1 score of 0.8738. Top-2 and top-3 accuracy reached 99.86% and 99.97%, respectively. Incorrect predictions showed lower mean confidence (0.6622 versus 0.9911) and higher normalized predictive entropy (0.2979 versus 0.0166) than correct predictions. Probability calibration yielded an expected calibration error of 0.0054 and a multiclass Brier score of 0.0090. Supplementary analyses demonstrated substantial feature sparsity, variation in minority-class performance, and sensitivity of nonlinear classifiers to hyperparameter selection. These findings indicate that routinely coded clinical information can reproduce recorded cancer-related INA-CBG classifications with high internal accuracy while providing insights into residual prediction uncertainty and classification disagreements. External and prospective validation is required before operational implementation.

Cite

APA

Nugroho, S., Ginardi, R. V. H., & Purnama, I. K. E. (2026). Predicting cancer-related INA-CBG classifications using explainable and uncertainty-aware machine learning: A retrospective hospital-based study. Array, 32, 101414. https://doi.org/10.1016/j.array.2026.101414

Vancouver

Nugroho S, Ginardi RVH, Purnama IKE. Predicting cancer-related INA-CBG classifications using explainable and uncertainty-aware machine learning: A retrospective hospital-based study. Array. 2026;32:101414. doi:10.1016/j.array.2026.101414

BibTeX
@article{Nugroho2026,
  author = {Nugroho, Suryanto and Ginardi, Raden Venantius Hari and Purnama, I Ketut Eddy},
  title = {Predicting cancer-related INA-CBG classifications using explainable and uncertainty-aware machine learning: A retrospective hospital-based study},
  journal = {Array},
  year = {2026},
  volume = {32},
  pages = {101414},
  doi = {10.1016/j.array.2026.101414}
}
Start for free →