Incorporating inflammatory biomarkers into a prognostic risk score in patients with non-ischemic heart failure: a machine learning approach

Jiayu Feng; Xuemei Zhao; Boping Huang; Liyan Huang; Yihang Wu; Jing Wang; Jingyuan Guan; Xinqing Li; Yuhui Zhang; Jian Zhang

doi:10.3389/fimmu.2023.1228018

Incorporating inflammatory biomarkers into a prognostic risk score in patients with non-ischemic heart failure: a machine learning approach

Front Immunol. 2023 Aug 15:14:1228018. doi: 10.3389/fimmu.2023.1228018. eCollection 2023.

Authors

Jiayu Feng¹, Xuemei Zhao¹, Boping Huang¹, Liyan Huang¹, Yihang Wu¹, Jing Wang¹, Jingyuan Guan¹, Xinqing Li¹, Yuhui Zhang¹, Jian Zhang^{1

2}

Affiliations

¹ State Key Laboratory of Cardiovascular Disease, Heart Failure Center, National Center for Cardiovascular Diseases, Fuwai Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
² Key Laboratory of Clinical Research for Cardiovascular Medications, National Health Committee, Beijing, China.

Abstract

Objectives: Inflammation is involved in the mechanisms of non-ischemic heart failure (NIHF). We aimed to investigate the prognostic value of 21 inflammatory biomarkers and construct a biomarker risk score to improve risk prediction for patients with NIHF.

Methods: Patients diagnosed with NIHF without infection during hospitalization were included. The primary outcome was defined as all-cause mortality and heart transplantations. We used elastic net Cox regression with cross-validation to select inflammatory biomarkers and construct the best biomarker risk score model. Discrimination, calibration, and reclassification were evaluated to assess the predictive value of the biomarker risk score.

Results: Of 1,250 patients included (median age, 53 years, 31.9% women), 436 patients (34.9%) experienced the primary outcome during a median of 2.8 years of follow-up. The final biomarker risk score included high-sensitivity C-reactive protein-to-albumin ratio (CAR) and red blood cell distribution width-standard deviation (RDW-SD), both of which were 100% selected in 1,000 times cross-validation folds. Incorporating the biomarker risk score into the best basic model improved the discrimination (ΔC-index = 0.012, 95% CI 0.003-0.018) and reclassification (IDI, 2.3%, 95% CI 0.7%-4.9%; NRI, 17.3% 95% CI 6.4%-32.3%) in risk identification. In the cross-validation sets, the mean time-dependent AUC ranged from 0.670 to 0.724 for the biomarker risk score and 0.705 to 0.804 for the basic model with a biomarker risk score, from 1 to 8 years. In multivariable Cox regression, the biomarker risk score was independently associated with the outcome in patients with NIHF (HR 1.76, 95% CI 1.49-2.08, p < 0.001, per 1 score increase).

Conclusions: An inflammatory biomarker-derived risk score significantly improved prognosis prediction and risk stratification, providing potential individualized therapeutic targets for NIHF patients.

Keywords: biomarker; inflammation; machine learning; non-ischemic heart failure; predictive model.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Biomarkers
Female
Heart Failure* / diagnosis
Humans
Machine Learning
Male
Middle Aged
Prognosis
Risk Factors

Substances

Biomarkers

Grants and funding

This work was supported by the National Nature Science Foundation of China (grant number 81873472) and the Chinese Academy of Medical Sciences Initiative for Innovative Medicine (grant number 2020-I2M-1-002).