Heart Failure in Norway: Combining Artificial Intelligence and a Database of 15000 Patients to Understand Clinical Characteristics, Mortality and Health Care Resource Use

The prevalence of Heart failure above 70 years of age is 10% and 5 year mortality rate above 60%, higher than for cancer. The readmission rate first after hospitalisation is 44% despite the availability of life prolonging and life quality enhancing treatment. There is a lack of resources for adequate diagnostic workup necessary for implementing evidence-based treatment. This projects aims at assessing the impact of guidelines based diagnostic workup and guidelines based treatment of heart failure on mortality and readmission rates. As the symptoms defining the degree of heart failure and the discharge medication only is available in the electronic patient files, artificial intelligence is used to retrieve this information to assess if treatment is according to guidelines.

The project is using first a rule based text processing approach using IBM Watson, then advancing to a machine learning approach using readmission and mortality as endpoints.

The project has access to digitally stored echocardiographic measurements as well as digital ECG's and lab data on 15 000 patients admitted with a diagnosis of Heart failure. If the retrieval of symptoms and function by artificial intelligence is successful, the next step is to assess if those benefitting the most from echocardiography can be identified using information from the ECG's, lab data or symptoms and functional capacity as described in the Electronic Health Records.