Data-Driven Phenotyping in Heart Failure With Preserved Ejection Fraction

Heart Failure with Preserved Ejection Fraction (HFpEF) is a complex condition, and people with HFpEF can have different symptoms and clinical profiles. Understanding these differences may help improve how the condition is described and studied. This study has two parts: a retrospective analysis and a cross-sectional assessment.

In the retrospective part, the researchers will collect clinical information that was previously recorded in the hospital's clinical records during past hospitalizations for HFpEF at the Local Health Unit of the Leiria Region (ULS RL). The data will be reviewed and prepared for analysis using standard data quality procedures. After the database is complete, the researchers will use data-driven methods to look for patterns among participants, in order to identify groups of people who share similar characteristics ("phenotypes") without setting predefined categories. Methods will include descriptive statistics, correlation analysis and feature selection using algorithmic approaches such as ReliefF. For phenotyping, unsupervised machine-learning techniques including K-means clustering and principal component analysis (PCA) will be applied.

The cross-sectional part will invite a sample of participants selected to represent each phenotype (planned 15 participants for each phenotype) identified in the retrospective analysis. Selected participants will complete a single on-site visit including informed consent verification, a structured clinical review and a standardized cardiopulmonary exercise test (CPET) performed according to local and international guidelines. The CPET procedures will follow the laboratory protocol, namely calibration of equipment, resting measurements, incremental workload protocol, continuous gas exchange, and electrocardiogram (ECG) monitoring. CPET data will be recorded in digital format and transferred securely to the study database.

The study will also evaluate phenotype concordance with widely used HFpEF tools (H2FPEF and HFA-PEFF) and describe differences in physiological responses during CPET across phenotypes, to help clarify how useful they are in describing different forms of HFpEF. Analyses will emphasize exploratory, data-driven evaluation and estimation of effect sizes, consistent with the phenotyping objectives of the study. Where relevant, associations between phenotype membership and CPET variables will be explored descriptively and through correlation-based analyses.

Ethical and data protection procedures are in place. Personal identifiers will be removed and replaced by study ID codes. A linkage file (study ID to personal identifiers) will be stored on an encrypted device with access restricted to the student investigator. Electronic study data will be housed on secure servers with role-based access control. Data will be retained according to institutional policy and relevant legislation. Only de-identified datasets will be used for analysis and sharing. Safety procedures for CPET include pre-test screening for absolute contraindications, continuous ECG and blood pressure monitoring during the test, availability of emergency equipment and immediate clinical oversight by qualified personnel. Adverse events during CPET will be recorded and reported per the Ethics Committee requirements.