How can life-threatening secondary diseases be prevented in congenital heart disease and when is medical intervention necessary? To answer this question, researchers at the Competence Network for Congenital Heart Defects set their hopes on a targeted application of artificial intelligence (AI).

Data of More than 10,000 Adults with Congenital Heart Disease

An international research team for the first time addressed the question in which way specifically trained computational processes, so-called deep learning algorithms (DL algorithms), are suitable for assessing the prognosis and guiding the therapy in congenital heart disease. The team was led by ACHD researcher Gerhard-Paul Diller. The AI study took place in collaboration with the renowned Royal Brompton Hospital in London. Within the study, the data of over 10,000 adult patients with congenital heart disease or pulmonary hypertension who were followed-up medically between 2000 and 2008, were analyzed algorithmically.

44,000 Medical Recordings Fed the Algorithms

The DL algorithms were trained by means of clinical and demographic data, ECG parameters, cardiopulmonary stress tests and selected laboratory markers. Using these raw data, biometricians taught the algorithms to identify the diagnosis group, as well as the complexity of the disease, and to assign these to the correct degree of severity as classified by the New York Heart Association (NYHA classification). Furthermore, models were developed to detect inconclusive diagnoses and anomalous disease courses. These models enabled the multidisciplinary teams to classify the respective diagnostic findings and to calculate the prognosis.

Accuracy of Over Ninety Percent

The DL algorithms were based on over 44,000 medical recordings. In total, they calculated the diagnosis, as well as the disease complexity and the NYHA classification with an accuracy of over ninety percent in random tests. Likewise, they were able to predict when patients required treatment again with an accuracy of over ninety percent.

Expanding AI Research Activities to Improve the Prognosis

In the follow-up period of eight years, 785 out of the 10,019 included patients deceased. Based on an analysis of the survival time, the degree of severity that had been derived automatically from the clinical information was crucial for the life expectancy, independently of the parameters of demography, exercise capacity, laboratory and ECG. The scientists hope that expanding AI research and the algorithmic computation of further data might help developing individualized treatment strategies. Furthermore, this might promote an increased life expectancy and quality of life also in severe congenital heart defects significantly.