Page 146 - EJMO-9-1

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

Figure 4D illustrates the confusion matrix representing function and is a strong predictor of adverse outcomes
the performance of a GBM model in predicting outcomes in HF. The dataset also includes information on
for HF patients. The GBM model correctly predicted 13 comorbidities, such as diabetes, high blood pressure, and
true positives and made four false positive predictions. It smoking status. These comorbidities are prevalent among
accurately predicted 31 instances where both the actual HF patients and contribute to disease progression and
and predicted outcomes were negative (true negatives), complications. In particular, diabetes and high blood
but misclassified 12 instances as negative when the actual pressure are known risk factors for the development
outcome was positive (false negatives). of HF and are associated with poor prognoses and high
mortality rates. Although smoking is less prevalent among
4. Discussion HF patients compared to other cardiovascular diseases, it

The original dataset provides a comprehensive overview remains a significant modifiable risk factor that warrants
of the clinical and demographic characteristics of HF attention in clinical management strategies. In addition,
patients. It comprises various attributes, including age, the dataset incorporates a temporal aspect represented by
anemia status, serum creatinine levels, ejection fraction, the “time” variable, which indicates the follow-up duration
and comorbidities, such as diabetes and high blood for each patient. This temporal dimension allows for the
pressure. These attributes offer valuable insights into the assessment of disease progression over time and facilitates
heterogeneity of HF patients and their associated risk longitudinal analyses of patient outcomes and treatment
factors for adverse outcomes. The distribution of age responses. The dataset also includes a binary outcome
9
within the dataset reflects the typical demographic profile variable, “DEATH EVENT,” indicating whether a patient
of HF patients, with ages ranging from 50 to 75 years. died during the follow-up period. This outcome variable
Advanced age is a well-established risk factor for HF, as the serves as the target variable for predictive modeling and
occurrence of HF rises with age, primarily due to changes risk stratification analyses.
in cardiac structure and function that occur over time. Overall, the original dataset provides a rich
The dataset’s representation of age diversity underscores source of information for studying HF epidemiology,
the importance of considering age as a key factor in risk prognostication, and therapeutic interventions. By
assessment and treatment planning for HF patients. leveraging the diverse array of clinical and demographic
Several clinical biomarkers and physiological parameters variables available in the dataset, researchers and clinicians
are included in the dataset, such as serum creatinine can gain valuable insights into the complex interplay of
levels, ejection fraction, and platelet counts. Platelets play factors influencing HF outcomes and develop tailored
a significant role in chronic HF outcomes through their approaches to patient care.
involvement in inflammation, thrombosis, and endothelial
dysfunction, which can exacerbate disease progression and 4.1. Feature importance analysis
lead to adverse events. Elevated platelet activation is linked The selected attributes based on the correlation analysis
to vascular inflammation and thrombotic complications, with the target variable “DEATH EVENT” encompass
while changes in platelet counts may reflect underlying various clinical indicators commonly associated with
conditions, such as anemia or systemic inflammation, HF. These attributes include demographic factors, such
both associated with poor prognosis. This study identified as age and sex, as well as physiological parameters, such
platelets as a moderate predictor of mortality, underscoring as serum creatinine levels, ejection fraction, and platelet
their potential as a biomarker for risk stratification. count. The feature importance scores obtained from the
Clinically, these findings highlight the importance of random forest regressor model highlight the significant
incorporating platelet metrics into broader predictive predictive power of certain attributes, particularly “time,”
models for personalized treatment strategies. Future which represents the follow-up period. This underscores
research should validate these results in larger cohorts and the importance of longitudinal data in understanding and
explore platelet function markers to enhance their utility predicting outcomes in HF patients.
in managing HF.
Compared to previous studies, our findings corroborate
Moreover, these markers play crucial roles in assessing existing literature on HF prognosis regarding the
cardiac function, renal function, and overall disease significance of certain clinical indicators in predicting
severity in HF patients. Elevated serum creatinine levels, adverse outcomes. For instance, our identification of age,
for example, are indicative of impaired renal function, serum creatinine levels, and ejection fraction as significant
which is commonly observed in HF patients and is predictors of death events aligns with numerous studies
associated with increased mortality risk. Similarly, a that have highlighted the prognostic value of these factors
reduced ejection fraction reflects compromised cardiac in HF patients. The inclusion of these attributes in
10

Volume 9 Issue 1 (2025) 138 doi: 10.36922/ejmo.6583