Bayesian Hemodynamics Model for Personalized Monitoring of Congestive Heart Failure Patients - Translating Physician's Reasoning Into Computational Models, Part II

Heart failure (HF) is a world-wide problem. At the moment 26 million people are living with this chronic disease and the prevalence is still increasing. Besides this growing number in prevalence, HF is also responsible for almost 1 million hospitalizations a year in the US and in Europe. Consequently, this has also a major economic impact especially due to recurrent admissions of these patients. Adequate prediction of decompensation could prevent (un)necessary admissions as a result of heart failure. Philips is developing a Bayesian model for chronic heart failure, enabling monitoring of patients with heart failure in the hospital and at home. An important characteristic of such a Bayesian model is that it is a knowledge-based model, in contrast to data-mining based models, and requires only a few patient data to get started (10-20 patients). Another important characteristic is that these 'knowledge-based models' are applicable in any setting, again in contrast to data-mining based models. This makes the proposed model different from conventional data-mining approaches to modelling. During a hospital admission, the model will be "filled in" with personal patient data. Subsequently, during the rest of the hospital stay or after release from the hospital, a number of symptoms and lab measurement variables ("observables"), will be the input for the model. The output of the model (the result) will be a probability of improvement (versus worsening) of the condition of the patient or the status of the heart failure condition on a scale (from 1-10). The model can deal with less input variables than the number it has been "personalized" with. With less input measurements, naturally the reliability of the result will be reduced. This modelling approach basically captures the clinical way of thinking into a model. If interpreted in the right way using smart Bayesian modelling, the GP or geriatrician will be able to monitor and treat the majority of heart failure patients. This fits in current thinking to reduce HC costs by keeping patients at home and out of the hospital.

The clinical investigation is designed to evaluate whether the outcome of the "Bayesian Hemodynamics model" compares with the cardiologist's status assessment. The purpose of this study is to validate the computer model that has been developed to assess the status of a heart failure patient. With the model, the investigators aim to support healthcare professionals with early detection of deterioration of heart failure patients and with providing the right treatment when it is needed. If successful, this could help heart failure patients to stay at home longer and reduce hospital admissions.

The clinical literature review is documented in report, Personalized Heart Failure Monitoring using a Bayesian network, Anja v.d. Stolpe, Wim Verhaegh, Folke Noertemann, PR-TN 2017/00180.

This clinical investigation is needed, because no complete datasets, including ground truth assessments by cardiologists, are available, neither in existing databases, nor in clinical literature.

The clinical investigation needs to be performed on a population that fulfills the inclusion/exclusion criteria described in Chapter 6, because the "Bayesian Hemodynamics model" is only valid for these cases.