gms | German Medical Science

Introduction: Cardiovascular diseases are the main causes of morbidity and mortality worldwide. They typically progress silently over many years and manifest themselves acutely in middle/advanced age when outcomes are poor despite optimal treatment. Personalised risk prediction is therefore of utmost importance to healthcare researchers, patients, and professionals alike. The ACRIBiS (Advancing Cardiovascular Risk Identification with Structured Clinical Documentation and Biosignal Derived Phenotypes Synthesis) project within the BMBF-funded German Medical Informatics Initiative strives to improve and standardise clinical documentation of routine data including the calculation of selected cardiovascular risk scores.

At present, clinicians estimate a patient’s individual risk through online calculators that are not well integrated into the clinical workflow or sufficiently quality assured, and, thus, of minimal help when communicating risk to patients. Here, we present an easy-to-use software to calculate three cardiovascular risk scores, whose use is recommended by current best-practice guidelines.

Methods: From the list of scores selected for the ACRIBiS project, we decided on one score for three high-risk patient groups:

1.

The CHA₂DS₂-VASc score to predict the risk of stroke in patients with atrial fibrillation [1]

2.

The SMART (Secondary Manifestations of ARTerial disease) score to predict recurrent ischaemic events in patients with pre-existing vascular diseases [2]

3.

The MAGGIC (Meta-Analysis Global Group in Chronic Heart Failure) score to predict mortality in patients with heart failure [3]

We developed and implemented score calculators in Python and used automated testing with randomly generated data in Selenium to compare scores to those generated by available online calculators. Additionally, we tested our calculators with data extracted from the UK Biobank (https://www.ukbiobank.ac.uk/) (Application Number 155031) and the MIMIC-IV [4] datasets.

Results: We successfully benchmarked our score calculators against online calculators with our open source software made available on Github (https://github.com/IMI-HD/rapid_cv_score). Our results were consistent with the results of the respective online calculators. We also identified relevant patient records in the UK Biobank and MIMIC-IV datasets to compute the CHA₂DS₂-VASc score for 48,703 patients (MIMIC-IV: 27,075, UK Biobank: 21,628) and the SMART score for 104,552 patients (MIMIC-IV: 25,228, UK Biobank: 79,324). Subsequently, we compared the scores’ prediction with follow-up data from the aforementioned resources.

Discussion: Risk prediction tools are an important prerequisite for personalized treatment. In the ACRIBiS project, we explore a more standardised approach as incorrect score calculation may have dangerous consequences. We successfully implemented and validated score calculators with randomly generated data. To test our implementation with real world data, we used the MIMIC-IV and the UK Biobank datasets for the CHA₂DS₂-VASc and the SMART score. Future steps include testing of the MAGGIC score calculation in both datasets and pilot testing at all ACRIBiS partner sites. Moreover, we plan to implement additional relevant scores (e.g., HAS-BLED, ABC-AF, SMART-REACH, Barcelona BioHF, and CHARGE-AF) and update our software package continuously. Taken together, our validated calculators can be used to calculate scores based on clinical routine data.

Conclusion: Integrated and validated score calculation can potentially increase the use of personalized risk assessment tools and allow better risk stratification in patients with cardiovascular diseases.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.