# Children Comorbidity Score, a Simple Predictor for In-hospital Mortality: A Nationwide Inpatient Database Study in Japan

## TL;DR

This study developed a new comorbidity score for children in Japan that better predicts in-hospital mortality than existing scores.

## Contribution

A novel Children Comorbidity Score (CCS) was developed using ICD-10 codes and Lasso regression for pediatric mortality prediction.

## Key findings

- The CCS had better discrimination (C-statistic 0.720) than CCC (0.649) and CCI (0.544) in predicting mortality.
- CCS showed good calibration and provided the highest net benefit in decision curve analysis.
- The CCS performed consistently well across age, sex, and BMI categories.

## Abstract

Utilizing a nationwide inpatient database in Japan, we aimed to develop a novel comorbidity score for pediatric patients to predict in-hospital mortality―the Children Comorbidity Score (CCS)―based on the International Classification of Diseases, 10th Revision (ICD-10) codes.

We retrospectively analyzed pediatric patients hospitalized between 2010 and 2017 using the Japanese Diagnosis Procedure Combination database. Eighty percent of the data was used as a training set, where we applied Lasso regression to a model with 56 candidate comorbidity categories to predict in-hospital mortality. We employed the 1-standard-error rule in Lasso regression to derive a parsimonious model and forced the entry of 12 categories of pediatric Complex Chronic Conditions (CCC). Thus, we developed the CCS, an integer-based comorbidity score using the selected variables with nonzero coefficients. The remaining 20% of the data was used as the test set, where we evaluated the CCS’s predictive performance using C-statistics, calibration, and decision curve analysis, comparing it with two other scores: a CCC-based score using ICD-10 codes and the Charlson Comorbidity Index (CCI).

Among 1,968,960 pediatric patients, we observed 6,492 (0.33%) in-hospital mortalities. The developed integer-based CCS, utilizing 10 comorbidity categories via variable selection by Lasso regression, had better discrimination ability (C-statistics, 0.720 [95% confidence intervals (CI), 0.707-0.734]) than the CCC (0.649 [0.636-0.662]) and CCI (0.544 [0.533-0.555]). The superior discrimination of the CCS was consistent across all age categories, sexes, and body mass index categories. The CCS showed good calibration, with a calibration slope of 1.027 (95% CI, 0.981-1.073). Decision curve analysis indicated that the CCS provided the highest net benefit compared to either of the reference models.

The ICD-10-based CCS outperformed conventional comorbidity scores in predicting in-hospital mortality and would be useful in comorbidity assessment among pediatric inpatients.