Optimized Multi-modality Machine Learning Approach During Cardio-toxic Chemotherapy to Predict Arising Heart Failure (MERMAID)

The present project will develop an automated machine learning approach using multi-modality data (imaging, laboratory, electrocardiography and questionnaire) to increase the understanding and prediction of arising heart failure in patients scheduled for cardio-toxic chemotherapy. This algorithmus will be developed by the technical cooperation partner at Technion, the institut for biomedical engineering in Haifa, Israel.

Study Overview

• Status: Unknown
• Conditions: Toxicity Due to Chemotherapy

Detailed Description

The present project will develop an automated machine learning approach using multi-modality data (imaging, laboratory, electrocardiography and questionnaire) to increase the understanding and prediction of arising heart failure in patients scheduled for cardio-toxic chemotherapy. This algorithmus will be developed by the technical cooperation partner Prof. Adam who leads the Technion, the institut for biomedical engineering.

Specific aims:

1. To collect all achievable data from patients scheduled for cardiotoxic chemotherapy at baseline, up to 6 months after ending therapy - regarding imaging (MRI, echocardiography with conventional and strain parameter), electrocardiography, biomedical markers (to define the function of liver, kidney, heart and hematopoietic bone marrow), clinical parameter and quality of life questionnaire:
2. To optimize and evaluate a robust machine learning approach that integrate and assess all these data to detect early myocardial damage and to identify an optimal parameter (single or in combination) for prediction of subclinical left ventricular (LV) dysfunction (stage 1 of the current study).
3. To perform a clinical study (stage 2 of the current study) of chemotherapy patients, and to identify subclinical LV dysfunction, which will be used to guide cardioprotective therapy using the new machine learning approach in comparison to the actual standard procedure using only echocardiographic left ventricular ejection fraction (LVEF).

The purpose of this study is to evaluate and optimize a machine learning approach to combine and integrate data from different imaging modalities with laboratory, electrocardiography and questionnaire information to define the value of all these parameter in patient management, by identification of subclinical LV dysfunction, which will be used to guide cardioprotective therapy in comparison to a standard approach using only conventional echocardiographic parameters.

MRI, conventional echocardiographic parameters and echocardiographic myocardial deformation imaging are employing different modalities and approaches to obtain insight into myocardial tissue and deformation. We hypothesize that a new and optimized automated algorithm using these modalities and integrating laboratory, electrocardiography and questionnaire information will improve the detection of early LV dysfunctions, and will bring new insight to the potential response of chemo patients to cardiotoxic therapy. We expect that this algorithm leads to the use of adjunctive therapy that will limit the development of LV dysfunction, interruptions of chemotherapy and development of heart failure in follow-up and thus will reduce morbidity and costs.

• Study Type: Observational
• Enrollment (Anticipated): 470

Contacts and Locations

Study Locations

• Germany
  • Aachen, Germany, 52074
    • Department of Cardiology, RWTH Aachen University Hospital

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: 18 years to 100 years (Adult, Older Adult)
• Accepts Healthy Volunteers: No
• Genders Eligible for Study: Female

• Sampling Method: Non-Probability Sample

Study Population

The machine learning based algorithm will be trained on a supervised cohort of 200 chemo-treated patients and an age matched control group of 200 normal subjects will achieve the desired accuracy to detect subtle changes in LV function (stage 1 of the current study).

The stage 2 part of the current study will be performed in patients undergoing cardiotoxic chemotherapy (N=70), randomized for comparing a surveillance strategy using machine learning approach (group A, N=35) from conventional surveillance based on conventional echocardiographic parameter as LVEF (group B, N=35). Patients coming to the echo lab for echo surveillance of LV function will be randomized to optimized automated algorithm or receive standard LVEF alone.

Description

Inclusion Criteria:

1. Patients Patients scheduled for chemotherapy at increased risk of cardiotoxicity (regarding 200 Chemo patients in stage 1 study and 70 Chemo patients in stage 2 study):

   • use of anthracycline with
   • trastuzumab (Herceptin) in breast-cancer with the HER2 mutation OR
   • tyrosine kinase inhibitors (eg sunitinib) OR
   • cumulative anthracycline dose >450g/m2 of doxorubicin, or equivalent other anthracycline cumulative dose (eg for epirubicine >900g/m2) OR
   • -increased risk of heart failure (HF) (age >65y, type 2 diabetes mellitus, hypertension, previous cardiac injury eg. myocardial infarction)
2. Female aged > 18 years
3. Written informed consent prior to study participation
4. The subject is willing and able to follow the procedures outlined in the protocol The department of gynecology at the RWTH University hospital will inform the principal investigator about these patients.

Exclusion Criteria:

1. Valvular stenosis or regurgitation of >moderate severity
2. History of previous heart failure (baseline New York Heart Association - NYHA >2)
3. Inability to acquire interpretable images (identified from baseline echo)
4. Contraindication to perform a MRI
5. Oncologic (or other) life expectancy <12 months
6. Pregnant and lactating females
7. Patient has been committed to an institution by legal or regulatory order
8. Participation in a parallel interventional clinical trial
9. The subject received an investigational drug within 30 days prior to inclusion into this study
10. Relevant renal insufficiency

Study Plan

How is the study designed?

Design Details

• Observational Models: Cohort
• Time Perspectives: Prospective

Number of groups / cohorts

4

Cohorts and Interventions

Group / Cohort
Supervised cohort of 200 chemo-treated patients; cohort of 200 patients undergoing a chemo therapy accordingly to the inclusion criteria patients
Age matched control group of 200 normal subjects; 200 age matched control group of subjects from the outpatient clinic who are not chemo-treated and who fit the inclusion and exclusion criteria
A:machine learning approach (N=35); 70 female patients undergoing cardiotoxic chemotherapy accordingly to the inclusion criteria will be randomized into two arms (group A and B).
B: conventional echocardiographic parameters (N=35); 70 female patients undergoing cardiotoxic chemotherapy accordingly to the inclusion criteria will be randomized into two arms (group A and B).

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Time Frame
Change in LVEF from baseline to one year, as determined by MRI as gold standard according to random study group allocation	one year

Collaborators and Investigators

• Sponsor: RWTH Aachen University
• Collaborators: Technion, Israel Institute of Technology

Study record dates

Study Major Dates

• Study Start: January 1, 2017
• Primary Completion (Anticipated): January 1, 2019
• Study Completion (Anticipated): January 1, 2019

Study Registration Dates

• First Submitted: September 16, 2016
• First Submitted That Met QC Criteria: October 13, 2016
• First Posted (Estimate): October 17, 2016

Study Record Updates

• Last Update Posted (Estimate): October 17, 2016
• Last Update Submitted That Met QC Criteria: October 13, 2016
• Last Verified: October 1, 2016

More Information

Terms related to this study

• Additional Relevant MeSH Terms: Heart Diseases; Cardiovascular Diseases; Heart Failure
• Other Study ID Numbers: 16-079

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No
• product manufactured in and exported from the U.S.: No