# A Creatinine–CAR Composite Index (CCAR) Optimized by Machine Learning for Prognosis in Cancer Cachexia

## TL;DR

This study introduces a new machine learning-optimized index, CCAR, that improves survival prediction for cancer cachexia patients and is available via a web-based calculator.

## Contribution

The novel CCAR index combines creatinine and CAR, optimized via machine learning, for improved prognosis in cancer cachexia.

## Key findings

- The CCAR index outperformed traditional markers in predicting survival across three patient cohorts.
- High CCAR scores were strongly associated with worse overall survival in cancer cachexia patients.
- A web-based calculator was developed to enable real-time CCAR computation and survival prediction.

## Abstract

Cancer cachexia is a multifactorial syndrome associated with poor prognosis and impaired quality of life in cancer patients. However, survival prediction in cancer cachexia remains difficult due to the lack of reliable biomarkers.

This retrospective cohort study analysed data from 1,367 patients with cancer cachexia diagnosed according to the 2011 Fearon consensus, using the multicentre INSCOC database. The cohort was divided into a training set (n = 959), an internal validation set (n = 408), and an independent external validation cohort (n = 284). The mean age of the entire cohort was 58.7 ± 10.9 years, and 39.4% were female. LASSO regression identified creatinine (Cr) as a key predictor. The CCAR (Cr + CAR) index was then constructed using Cr and CAR within a random forest model. Prognostic performance was assessed by Harrell's concordance index (C‐index), time‐dependent AUC, Kaplan–Meier analysis and multivariate Cox regression, with overall survival (OS) as the primary endpoint.

The CCAR index consistently outperformed conventional inflammation‐ and nutrition‐related markers across all three cohorts. The C‐index values for CCAR were 0.777 in the training cohort, 0.789 in the internal validation cohort, and 0.765 in the external validation cohort, compared with 0.627–0.660 for CAR ALONE. Patients in the high‐CCAR group had significantly worse OS than those in the low‐CCAR group (log‐rank p < 0.0001 for all cohorts). Multivariate Cox regression confirmed that CCAR was an independent prognostic factor for OS (HR 3.31, 95% CI 2.94–3.72 in the training set; HR 3.51, 95% CI 2.93–4.22 in the validation set; HR 3.21, 95% CI 2.55–4.03 in the external cohort; all p < 0.001). A web‐based calculator (https://heyangzhang.pythonanywhere.com) was developed for real‐time CCAR computation and survival predictions.

The CCAR index provides a robust, easily accessible tool for predicting survival outcomes in cancer cachexia patients. The web‐based CCAR calculator demonstrates significant clinical applicability and improves patient risk stratification, offering potential for guiding early interventions and personalized treatment strategies in clinical settings.