Predicting incident heart failure from population-based nationwide electronic health records: protocol for a model...

Area of Science:

• Cardiology
• Health Informatics
• Predictive Analytics

Background:

• Heart failure (HF) presents a growing global health challenge, marked by significant morbidity, mortality, and escalating healthcare expenditures.
• Current diagnostic approaches often lead to delayed detection of HF, typically at advanced symptomatic stages requiring hospitalization.
• Existing statistical models for predicting incident HF face limitations in performance and scalability for widespread clinical application.

Purpose of the Study:

• To develop and validate robust prediction models for new-onset heart failure (HF) utilizing routinely collected primary care electronic health records (EHRs).
• To establish prediction horizons of 1, 5, and 10 years for incident HF risk.
• To enable earlier identification and targeted diagnostics for HF through implementable EHR-based models.

Main Methods:

• Employing logistic regression and supervised machine learning techniques on large-scale primary care EHR datasets (CPRD-GOLD for derivation, CPRD-AURUM for external validation).
• Utilizing patient-level linked data from primary care, secondary care, and mortality records.
• Assessing model performance through discrimination, calibration, and clinical utility metrics, focusing on routinely accessible variables.

Main Results:

• The study will derive and validate prediction models for incident heart failure (HF) risk.
• Performance evaluation will determine the models' accuracy in predicting HF over 1, 5, and 10-year horizons.
• The models will be assessed for their potential clinical utility in identifying at-risk individuals.

Conclusions:

• Validated EHR-based prediction models can enhance the early detection of heart failure (HF).
• This approach offers a scalable solution for identifying individuals at risk of HF in routine clinical practice.
• Improved early identification through predictive modeling can facilitate timely interventions and potentially alter HF disease trajectories.