Approaching a nationwide registry: analyzing big data in patients with heart failure

Randomized controlled trials usually lack generabilizity to real-world context. Real-world data, enabled by the use of big data analysis, serve as a connection between the results of trials and the implementation of findings in clinical practice. Nevertheless, using big data in the healthcare has difficulties such as ensuring data quality and consistency. This article aimed to examine the challenges in accessing and utilizing healthcare big data for heart failure (HF) research, drawing from experiences in creating a nationwide HF registry in Türkiye.

We established a team including cardiologists, HF specialists, biostatistics experts, and data analysts. We searched certain key words related to HF, including heart failure, nationwide study, epidemiology, incidence, prevalence, outcomes, comorbidities, medical therapy, and device therapy. We followed each step of the STROBE guidelines for the preparation of a nationwide study. We obtained big data for the TRends-HF trial from the National Healthcare Data System. For the purpose of obtaining big data, we screened 85,279,553 healthcare records of Turkish citizens between January 1, 2016 and December 31, 2022.

We created a study cohort with the use of ICD-10 codes by cross-checking HF medication (n = 2,722,151). Concurrent comorbid conditions were determined using ICD-10 codes. All medications and procedures were screened according to ATC codes and SUT codes, respectively. Variables were placed in different columns. We employed SPSS 29.0, MedCalc, and E-PICOS statistical programs for statistical analysis. Phyton-based codes were created to analyze data that was unsuitable for interpretation by conventional statistical programs. We have no missing data for categorical variables. There was missing data for certain continuous variables. Propensity score matching analysis was employed to establish similarity among the studied groups, particularly when investigating treatment effects.

To accurately identify patients with HF using ICD-10 codes from big data and provide precise information, it is necessary to establish additional specific criteria for HF and use different statistical programs by experts for correctly analyzing big data.