Page 147 - EJMO-9-1

Eurasian Journal of Medicine and
Oncology
Machine learning insights into heart failure outcomes

predictive models can aid clinicians in risk stratification 4.3. Performance metrics of machine learning
and treatment planning for HF patients. Moreover, the models
emphasis on longitudinal data, particularly the “time” The performance metrics presented in Table 3 provide
variable representing follow-up duration, echoes previous insights into the effectiveness of various machine learning
research highlighting the importance of temporal aspects in models in predicting death events among HF patients. The
understanding disease progression and outcomes in HF. 11 logistic regression and random forest models demonstrated
Our study underscores the utility of data-driven relatively higher accuracy and AUC-ROC values compared
approaches in identifying relevant predictors of adverse to the SVM. Furthermore, the GBM model demonstrated
outcomes in HF and highlights the importance of robust performance across multiple metrics, such as
comprehensive risk assessment in clinical practice. In accuracy, precision, recall, F1-score, and AUC-ROC.
addition, our study also emphasizes the value of data- Comparing our findings on machine learning model
driven strategies in risk stratification and treatment performance with previous studies reveals consistent
planning, a sentiment echoed by several researchers trends regarding the effectiveness of certain algorithms in
advocating for the integration of predictive modeling into predicting death events among HF patients. Our results
clinical practice to improve patient care and outcomes. align with previous research indicating that logistic
12
15
However, further validation and exploration of additional regression and random forest models often demonstrate
factors are warranted to enhance the predictive accuracy higher accuracy and AUC-ROC values compared to SVM
16
and generalizability of prognostic models for HF patients, classifiers. This finding suggests that logistic regression
as emphasized in the literature. 12 and random forest algorithms may be more suitable for risk
prediction in HF populations compared to SVM classifiers
4.2. Correlation matrix due to their robust performance metrics. Moreover, our
The correlation matrix offers valuable insights into the observation of the excellent performance of the GBM
relationships between various clinical attributes and their model is consistent with studies highlighting the superior
associations with the occurrence of death events among predictive power of ensemble learning techniques, such
17
patients with HF. The positive and negative correlations as boosting algorithms, in healthcare applications. The
identified within the matrix provide a deeper insight into comprehensive assessment of various performance metrics,
the factors influencing patient outcomes. including accuracy, precision, recall, and F1-score, further
strengthens the validity of our findings and underscores
The correlation matrix analysis corroborates existing the importance of selecting appropriate machine learning
literature on HF prognosis by identifying key predictors algorithms tailored to specific clinical contexts. However,
of mortality among HF patients. Specifically, the positive as noted in previous research, the performance of machine
correlation between age and serum creatinine levels aligns learning models can be influenced by various factors,
with numerous studies that have demonstrated age-related including data quality and sample size, emphasizing the
declines in renal function and their associations with need for rigorous validation and refinement processes
adverse outcomes in HF patients. Similarly, the negative to verify their reliability and applicability in real-world
13
correlation between ejection fraction and death events healthcare settings. 18
highlights the prognostic significance of cardiac function,
consistent with extensive research indicating that reduced 4.4. Prediction analysis of machine learning models
ejection fraction is a strong predictor of mortality in HF The confusion matrix offers valuable insights into the
patients. Our findings underscore the importance of accuracy of the linear regression model in predicting death
14
these clinical parameters in risk assessment and treatment events among HF patients. By categorizing predictions
planning for HF patients, echoing the sentiments of previous into true positives, false positives, true negatives, and
studies advocating for comprehensive risk stratification false negatives, the matrix enables a thorough assessment
approaches in clinical practice. Moreover, our correlation of the model’s accuracy and effectiveness. The model
10
matrix analysis enhances understanding of the complex achieved a relatively high number of true positives,
interplay between various clinical parameters and their correctly identifying cases where death events occurred.
impacts on patient outcomes, aligning with the broader However, it also exhibited a notable number of false
literature emphasizing the multifactorial nature of HF positives, incorrectly predicting positive outcomes where
prognosis. However, further validation and exploration of no death events occurred. Conversely, although the
13
additional factors are warranted to strengthen prognostic model demonstrated a high number of true negatives
models and improve patient outcomes in HF management, by accurately identifying cases where no death events
as suggested by previous research. occurred, it also had some false negatives, failing to

Volume 9 Issue 1 (2025) 139 doi: 10.36922/ejmo.6583