Combined learning models for survival analysis of patients with pulmonary hypertension

Abstract

Background Pulmonary hypertension is a disease that manifests itself by excessive and chronic elevation of pressure in the pulmonary arteries. Pulmonary Hypertension (PH) presents non-specific initial symptoms, and this leads to an often-late diagnosis with a prognosis that can compromise the survival of the patient.Methods This paper will propose a combined methodological approach of the survival Support Vector Machine (SVM), Cox Proportional-Hazards Model (Cox-PHM), and Association Rule Mining (ARM) methods to examine the associated factors that are critical to the survival prognosis of patients with pulmonary hypertension (PAH).Results This approach was applied to data from 171 patients with PAH from 9 hospitals in 4 sub-Saharan African countries. For the study, nine PAH-related parameters were recorded for each patient. The SVM and Cox-PHM models showed that ''Associated with Schistosomiasis'' was the most significant parameter on death status. These allowed the ARM model to identify all the rules related to this significant parameter to provide more necessary information, such as detecting any other parameter that may enter the game with more precision.Conclusion The interpretation of results from the combined approach offers clarifications that are useful for a good understanding of the etiological factors and influential variables in the prognosis of survival of pulmonary hypertension. This provides an additional perspective in line with the messages of recent international recommendations which suggest that early diagnosis and treatment contribute to improving the vital prognosis of patients.

Keywords

• Survival analysis
• Learning patterns
• Association rules
• Interpretability