Siri Pflüger

Siri Pflüger

Master's Thesis

Analyzing the Health Status of Heart Failure Patients based on Telemonitoring Data using Machine Learning

Katharina Jäger (M.Sc.), Madeleine Flaucher (M.Sc.), Dr.-Ing. Heike Leutheuser, Prof. Dr. Björn Eskofier, Dr. Sebastian Eckl, Patricia Trißler

06 / 2022 – 12 / 2022


Heart failure (HF) is one of the leading causes of death worldwide [1]. A massive deterioration in cardiac output is referred to as decompensated heart failure. The heart can no longer supply the organism with sufficient blood and oxygen to maintain a stable metabolism. Typical symptoms are fluid retention and dyspnea or tachycardia even at rest [2]. According to the German Federal Statistical Office, HF is the most common single diagnosis of fully hospitalized patients in Germany in 2019 [3]. The incidence of disease and the total hospitalization rate of patients with HF have been steadily increasing for years. Analysis of data from 7 million patients with statutory health insurance shows that around 16 % of HF patients died within two years [4]. Mortality increases with age and the severity of the disease [5].

According to guidelines, current care of HF patients recommends regular monitoring to detect changes in clinical and psychosocial situations and adjust treatment if necessary [4]. However, the literature reveals that patients often have difficulties constantly monitoring their health status, recognizing symptoms early, and knowing when to contact the physician [6]. Telemonitoring provides the opportunity to support these patients with an easy-to-use mobile and sensor-based technology for the home environment. Furthermore, telemonitoring has been shown to significantly reduce the rehospitalization rate and mortality of chronic heart failure patients and thus also reduce healthcare costs [7]. The ProHerz-App (ProCarement GmbH, Forchheim, Germany) offers a digital telemonitoring application for the individualized care of HF patients. Until now, recommendations for action and therapy have been based on defined guidelines and the experience of physicians and heart failure nurses [4, 8]. However, literature shows that machine learning can predict HF events with reasonable accuracy and, in some cases, well before clinical diagnosis [9, 10]. The use of modern machine learning techniques in the ProHerz telemonitoring system aims to increase and accelerate the objectivity of physicians’ assessment of health status.

Therefore, this Master’s thesis aims to analyze the progressive health status of HF patients to detect cardiac decompensation using machine learning techniques. This is done in three steps and refers to the objective and subjective evaluation of the disease. In the first step, relevant parameters which assess the health status and make an imminent decompensation recognizable at an early stage are determined from guidelines. Furthermore, expert interviews will be conducted to verify and extend the guideline-defined parameters with evidence-based knowledge. The subjective perception of symptoms will be analyzed and included in the evaluation using patient surveys. In a final step, the gained knowledge will be applied to analyze the ProHerz telemonitoring dataset for decompensation patterns using machine learning techniques. The aim is to analyze whether correlations between the measured vital parameters and the health status questionnaire data exist and whether health status classification can be implemented on the telemonitoring data. The data used for the analysis was acquired in a previous study by ProCarement, which includes daily measurements of vital signs and questionnaire data from 70 patients.

[1] C. W. Tsao u. a., „Heart Disease and Stroke Statistics—2022 Update: A Report From the American Heart Association“, Circulation, Bd. 0, Nr. 0, S. CIR.0000000000001052, doi: 10.1161/CIR.0000000000001052.
[2] U. C. Hoppe und E. Erdmann, „Chronische Herzinsuffizienz“, in Klinische Kardiologie: Krankheiten des Herzens, des Kreislaufs und der herznahen Gefäße, E. Erdmann, Hrsg. Berlin, Heidelberg: Springer, 2011, S. 123–179. doi: 10.1007/978-3-642-16481-1_5.
[3] „Deutscher Herzbericht 2020“, Deutsche Herzstiftung e.V., Frankfurt am Main, 32. Version, Juni 2021.
[4] Bundesärztekammer (BÄK), Kassenärztliche Bundesvereinigung (KBV), Arbeitsgemeinschaft der Wissenschaftlichen Medizinischen Fachgesellschaften (AWMF). Nationale VersorgungsLeitlinie Chronische Herzinsuffizienz – Langfassung, 3. Auflage. Version 3. 2019. DOI: 10.6101/AZQ/000482.
[5] S. Störk, F. Peters-Klimm, J. Bleek, R. Ninic, und A. Klöss, „Sektorübergreifende Versorgung bei Herzinsuffizienz“, in Krankenhaus-Report 2021: Versorgungsketten – Der Patient im Mittelpunkt, J. Klauber, J. Wasem, A. Beivers, und C. Mostert, Hrsg. Berlin, Heidelberg: Springer, 2021, S. 109–130. doi: 10.1007/978-3-662-62708-2_7.
[6] K. A. Sethares, M.-E. Sosa, P. Fisher, und B. Riegel, „Factors Associated With Delay in Seeking Care for Acute Decompensated Heart Failure“, Journal of Cardiovascular Nursing, Bd. 29, Nr. 5, S. 429–438, Okt. 2014, doi: 10.1097/JCN.0b013e3182a37789.
[7] C. Ebner, P. Kastner, und G. Schreier, „Telemonitoring bei Herzschwäche Patienten – Von der Wissenschaft zur Anwendung“. Tagungsband der eHealth2009 und eHealth Benchmarking 2009, 7. Mai 2009. Zugegriffen: 31. Januar 2022.
[8] T. A. McDonagh u. a., „2021 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure“, European Heart Journal, Bd. 42, Nr. 36, S. 3599–3726, Sep. 2021, doi: 10.1093/eurheartj/ehab368.
[9] M. Shah, R. Zimmer, M. Kollefrath, und R. Khandwalla, „Digital Technologies in Heart Failure Management“, Current Cardiovascular Risk Reports, Bd. 14, Nr. 8, 2020, doi: 10.1007/s12170-020-00643-7.
[10] J. Wu, J. Roy, und W. F. Stewart, „Prediction modeling using EHR data: challenges, strategies, and a comparison of machine learning approaches“, Med Care, Bd. 48, Nr. 6 Suppl, S. S106-113, Juni 2010, doi: 10.1097/MLR.0b013e3181de9e17.