Final Report Summary - DOPPLER-CIP (Determining Optimal non-invasive Parameters for the Prediction of Left vEntricular morphologic and functional Remodeling in Chronic Ischemic Patients) Executive Summary:4.1.1 Executive summaryChronic ischemic heart disease is a disease in which an imbalance exists between the delivery of oxygen to the heart muscle with respect to its oxygen consumption. Most often, this imbalance is the result of atheromatous plaque build-up in one or more of the coronary arteries, which compromises normal myocardial perfusion. Although ischemic heart disease does not necessarily cause an immediate threat to the patient, it does give rise to morphological and functional remodelling of the heart. The former implies that the shape of the heart (cavities and walls) changes over time, which leads to enlarged hearts and ultimately heart failure (i.e. a life threatening situation in which the heart cannot pump enough blood in order to provide the whole body with nutrients and oxygen). The latter type of remodelling causes the heart to function less well that in combination with the morphologic changes accelerates the evolution towards heart failure.Restoring blood flow to a region of chronic ischemia can stop the process of remodelling and allow partial or full recovery of function. The current therapeutic approaches are either to restore blood flow by mechanical dilatation of the stenosed coronary vessel (i.e. balloon angioplasty) or surgically by-passing the stenosis with a venous or arterial vessel graft (i.e. coronary artery by-pass grafting). Pre-procedural investigations as well as the therapeutic measures undertaken are costly and carry a considerable procedural risk for the patient. It is therefore current practice to assess possible adverse morphologic remodelling and functional recovery of the heart before the decision for invasive examinations and treatments is taken. To this end, several non-invasive tests can be used such as bicycle stress testing while monitoring changes in the electrocardiogram (ECG); echocardiography (echo); magnetic resonance imaging (MRI); computed tomography (CT) or single positron emission computed tomography (SPECT). Although each of these modalities have shown accurate results under study conditions in selected patient populations, to date, no large scale studies have made a direct comparison of the different methodologies towards predicting adverse morphologic remodelling or functional recovery of the myocardium after medical therapy. The lack of such information results in a sub-optimal use of the methodologies at hand. The goal of DOPPLER-CIP was therefore to set up such a study and directly contrast the different imaging parameters and modalities with respect to their prognostic power towards remodelling of the heart.Over a period of 3 years (i.e. January 1st 2010 till December 31st 2012), 676 patients with suspicion of ongoing ischemic heart disease were recruited in 7 clinical centres across Europe. Upon recruitment, several clinical tests (i.e. blood samples, quality of life questionnaire, etc.) as well as imaging examinations were performed. All data was analysed in a blinded manner in specialized core-labs. Two years after recruitment, 613 of these patients (i.e. ~91%) returned for a follow-up examination including imaging. As such, remodelling of the hearts of these patients over the 2 year follow-up period could be determined and the most prognostic baseline measurement could be determined. Although detailed statistical analysis of the data is still ongoing at the time of writing, preliminary findings indicate that morphologic characteristics of the ventricle at baseline best predict both morphologic and functional remodelling. The findings of this study will not only improve the treatment of an individual patient but will also provide valuable information towards policy makers in order to ensure sustainability of the European health care systems.Project Context and Objectives:4.1.2 Description of the project context and objectives (max 4 pages)4.1.2.a Ischemic heart diseaseCardiovascular disease is currently the leading cause of death in industrialized countries and is expected to develop into this position in the emerging countries by 2020 [Murray97]. Coronary artery disease (CAD) is the most prevalent cause of cardiovascular disease and is associated with high mortality and morbidity. Patients with CAD represent a very large proportion of all acute medical hospitalizations in Europe. The clinical presentations of CAD include stable angina pectoris (i.e. chest pain), unstable angina, acute myocardial infarction, heart failure and sudden death. However, ischemia can also be ‘silent’ in the sense that it is present without clear symptoms.Chronic ischemic heart disease originates from an imbalance between myocardial oxygen supply and myocardial oxygen demand. In CAD, this imbalance is secondary to the presence of flow limiting coronary stenosis. Exercise increases myocardial oxygen demand and will thus disclose ischemia early.Restoration of myocardial blood supply has the potential to improve ventricular contractile function and survival. However, as any intervention carries risk, patients that are unlikely to benefit are better refrained from this therapy. For this reason it is crucial to determine which patients with ischemic cardiomyopathy would benefit from myocardial revascularization. 4.1.2.b Prevalence of ischemic heart disease The prevalence of chronic angina patients in Europe can be estimated as high as 30.000-40.000 per million inhabitants [Julian97]. The prevalence increases with age and is twice as common in men as it is in women. Ischemic heart disease either associated with stable angina or not, is the leading cause of heart failure in developed countries, accounting for over two-thirds of cases [Chapter 14, Crea].4.1.2.c Non-invasive diagnosis of ischemic heart diseaseThe presence of chronic ischemic heart disease can be investigated by a number of non-invasive techniques. The primary aim of these techniques is to provide a regional map of the heart in which the amount of viable myocardium is quantified and wall motion improvement or clinical benefit after coronary revascularization can be predicted:• Electrocardiogram (ECG): The cardiac contraction is associated with the movement of ions (i.e. charged particles) at a molecular level which causes changes in the electric field surrounding the heart at a macroscopic level. These macroscopic changes can be measured in a so-called electrocardiogram (ECG). The ECG is the cornerstone of each cardiac examination with important diagnostic and therapeutic information.• Scintigraphy: By injecting radioactive molecules in the blood stream, the perfusion of organs can be imaged by detecting the location where particles – intrinsically associated with radioactive decay – are emitted. Hereto, the patient needs to be surrounded by an array of detectors so that a computer reconstruction can be made from the source of the detected emissions based on time of flight. As such, this technique is commonly referred to as single positron (i.e. the particle) emission computed tomography (SPECT). By attaching the radioactive material to molecules involved in metabolic processes, also metabolic activity can be visualized.• Echocardiography: In this technique, acoustic waves – not detectable by the human ear - are induced in the human body. As these waves will be reflected at boundaries between different structures in the body (i.e. cause an echo), this information can be used to construct an image. The ultrasound exam of the heart is the most frequently used cardiac imaging technique because of its user-friendliness, its availability and its relatively low cost.• Magnetic Resonance Imaging (MRI): This technique makes use of the fact that certain atoms abundantly present in the body behave like microscopic magnets that can be manipulated by applying external magnetic fields. Hereto, the patient is put in a strong static magnetic field (induced by large superconducting coils) while other magnetic fields and electromagnetic detectors (i.e. antennae) are used to disturb these microscopic magnets and sense their response to this disturbance. As such, 2D or 3D images of the human body, i.e. the heart, can be made. Interestingly, by adding paramagnetic contrast agents, this technique allows to directly visualize scar tissue.All of the methodologies described above have advantages and disadvantages. The methodology of choice will therefore often depend on the preference and experience of the treating physician and the local hospital guidelines or customs. Moreover, technological developments have brought about a multitude of new imaging modes and parameters that have shown accurate results under study conditions in selected patient populations. There was thus a need to make an objective comparison of methodologies in order to predict patient outcome based on these (new) non-invasive imaging techniques.This was the primary objective of DOPPLER-CIP: determining the optimal non-invasive imaging parameter (myocardial function, perfusion, ventricular blood flow, cell integrity) and methodology (ECG, echocardiography, scintigraphy, MRI) that best predicts adverse morphological and function ventricular remodelling of the heart. Moreover, as the different technical options come at a different cost, the secondary objective of DOPPLER-CIP was to interpret the prognostic findings in the context of their cost-effectiveness. Finally, as a third objective of DOPPLER-CIP, the collected data would be used to look into the pathophysiologic mechanisms of ischemic heart disease.To reach the above goals, a multi-centre study was set up in which patients with suspicion of chronic ischemic heart disease were recruited and imaged with the different imaging modalities both at baseline and after 2 years of follow-up.References[Julian97] Julian DG, Bertrand ME, Hjalmarsson A. Management of stable angina pectoris. European Heart Journal, 18:394-413, 1997[Murray97] Murray CJ, Lopez AD. Alternative projections of mortality and disability by cause 1990-2020: Global Burden of Disease Study. Lancet, 24;349(9064):1498-504, 1997Project Results:4.1.3 Main S&T results/foregrounds4.1.3.a Overall study designThe overall strategy of DOPPLER-CIP was to conduct a multi-centre clinical study in which all participating (clinical) partners collected data according to standard operating procedures and in which all partners contributed in the analysis and interpretation of these data. The latter was implemented through the installation of core-lab facilities. In this way, consistent parameter extraction could be assured to avoid study bias. A further measure to avoid bias was blinding of data available to the core-labs.Roughly, DOPPLER-CIP was divided into four phases. Phase 1 (May 2009-December 2009) aimed at setting up the multi-centre study. It included defining a standardized imaging / data analysis protocol; the installation of the different core-labs and a central database; and a quality assurance step in which all sites were validated in terms of appropriate data acquisition and all core-labs were validated in terms of appropriate data analysis.The second phase of DOPPLER-CIP (January 2010-November 2014) was related to data acquisition and analysis. During this phase, patients were recruited at the different clinical sites; imaging data were acquired and distributed to the different core-labs; and all data was analysed. Data analysis was focused around 4 categories of parameters: morphological parameters; parameters related to global left ventricular function; parameters related to regional left ventricular function; and myocardial perfusion parameters. Over a period of two years (i.e. 24 months), adverse cardiac events such as sudden death, myocardial infarction, heart failure and mechanical revascularization were registered. At the end of this follow-up period, all patients were invited to undergo a second imaging study in order to define reference data for effective morphologic and functional changes. These data were analysed in the same manner as the baseline data (i.e. blinded core-lab analysis).The third and final phase of DOPPLER-CIP (December 2014-February 2015) focused on the interpretation and statistical analysis of the data.4.1.3.b Details on the study implementationDOPPLER-CIP aimed at including about 675 patients with suspicion of chronic ischemic heart disease. The size of this study population was estimated based on a statistical power analysis showing that this would be the number of patients required to allow to detect 8-9% difference in accuracy amongst the prognostic parameter tested. Patient recruitmentPatients eligible for participating in the study were patients that presented at the out-patient clinic with typical clinical symptoms of chronic ischemic heart disease such as fatigue, dyspnoea, (un)stable angina. Patients that had a record of acute coronary syndromes within the last 3 months; that already had a pacemaker or Internal Cardiac Defibrillator (ICD); that showed significant arrhythmias; or that had a concomitant disease that made 2-year survival unlikely were not eligible. Moreover, patients that were recruited for the study but that proved to have non-ischemic cardiomyopathy or more than moderate valve lesions were excluded. Once the patient was considered eligible, (s)he was included in the study after informed consent had been obtained. At the start of the project, a study protocol was established in which the detailed inclusion and exclusion criteria were defined and in which it was specified which (clinical) data had to be collected in every patient including the procedures that needed to be followed to obtain that information. In addition, template documents to obtain informed consent were distributed amongst all partners; translated to the local language; and approved by the local ethical committees.Baseline data collection Upon inclusion, baseline imaging data were acquired. Standard diagnostic tests were done for all patients and included: electrocardiogram (ECG) bicycle testing with measurement of VO2max, blood sampling and assessment of quality of life (using the MacNew Heart Disease Health-related Quality of Life Questionnaire (MacNew) - Short Form 36 (SF-36)). In addition, as much as possible of the following data was acquired whenever practically possible and ethically justifiable:• Echocardiography including two-dimensional imaging, three-dimensional imaging, contrast imaging and Myocardial Velocity Imaging (MVI)• Magnetic Resonance Imaging (MRI) including cine-MRI, contrast imaging (first-pass and delayed enhancement), MRI-tagging and phase-contrast MRI• Single Positron Emission Computed Tomography (SPECT) stress (dipyridamole) testing• Stress echo and/or stress MRI (using either dobutamine, adenosine or dipyridamole as a stressor)• Coronary angiography (when clinically required)For all modalities, protocols on data acquisition were defined at the start of the project detailing which modalities could be used in which operational modes; scanner settings to be applied; data sets to be taken; data formats for storage; etc. Obviously, it was not possible to acquire all data in all patients for reasons related to patient comfort and logistics. However, as many of the data sets presented above as logistically possible and ethically justifiable were acquired in every patient. Please note that DOPPLER-CIP did not consider CT acquisitions in its protocol as the aim of the study was not to characterize the coronary anatomy and as cardiac CT can currently not be justified for the mere assessment of cardiac morphology, function and perfusion given the radiation and contrast load for the patient.After baseline acquisitions, each patient received optimal treatment according to the individual physician’s expertise and based on European guidelines. It was not the intension of this study to attempt to optimize different therapeutic strategies.Data transferUpon recruitment of a patient, a case registration by the clinical site was made in the central database via a web interface resulting in the automatic generation of a study specific ID for this patient. Baseline clinical data (e.g. gender, weight, cholesterol levels, etc.) were captured electronically through the web interface of the database using standardized electronic case report forms (eCRFs). The electronic study platform was based on the OpenClinica platform. It supports HIPAA (Health Insurance Portability and Accountability Act) guidelines and is designed as a standards-based extensible, modular and open source platform.At the clinical site, images were anonymized with the generated study ID. For this anonymization process, dedicated software was developed that would automatically retrieve the study ID from the OpenClinica platform to avoid typing errors. In this way, not only the privacy of the study subjects was guaranteed but also bias during the later data analysis was avoided. As such, anonymized data were sent to the central core-lab via a secured internet connection (i.e. a Virtual Private Network – VPN) in order to be stored in a PACS system dedicated to the study. Hereto, an open source PACS system “DCM4CHEE” was used which is based on the same software architecture as OpenClinica. All imaging data was stored in a standard DICOM format.Blood plasma samples (10 vials of 2ml) were labelled using the generated study ID and transferred to the central core lab through specialized commercial couriers.Data analysisImages were divided according to modality by the central core-lab and sent to specialized core-labs for analysis. To guarantee that all core-labs were fully blinded during analysis, the images were re-anonymized and labelled with a new study ID. For the transfer of the (anonimyzed) images again DICOM network operations over secure network connections were used.Core-labs were defined based on the expertise of the different partners in the project and detailed in the table below. The parametric data generated by the core-labs were entered into a central SQL database. Imaging parameters were sent to the central core-lab by regular Email in customized formats as defined either by the processing software (i.e. the data exports provided by these analysis tools) or by the respective core-lab.In order to verify the quality of the measurements done by a given core-lab, a subset of the data sets (i.e. 15 data sets in total) were given a distinct study ID (i.e. re-anonimized) and re-send to the core-lab. As such, the measurement variability of a given core-lab could be quantified.Patient follow-upTwo years after inclusion of the patient in the study, a follow-up study was performed including the same quality of life assessment and the same non-invasive measurements as the ones done at baseline for that particular patient. At follow-up, only parameters relating to morphology or function were assessed. In other words, no follow-up perfusion or stress tests were performed unless required for good clinical practice. The follow-up data were analysed in the different core-labs (as defined above) in order to objectively determine morphologic or functional changes that occurred in a particular patient. This information was used in combination with the registry on cardiac events as study end-point and served as reference during further statistical analysis of the prognostic value of the baseline data. Important to note is that the re-anonimization of the data by the central core-lab was done using distinct study IDs so that the core-lab could not link baseline to follow-up studies of a given patient. In this way, data analysis remained fully blinded also at follow-up.A schematic overview of the study protocol of DOPPLER-CIP is given in the Figure 1 while the work/data flow is illustrated in Figure 2.Statistical analysisThe imaging parameter database of DOPPLER-CIP was statistically analysed in order to address the primary objective of the study, i.e. determining the optimal non-invasive imaging parameter (myocardial function, perfusion, ventricular blood flow, cell integrity) and methodology (ergometry, echocardiography, scintigraphy, MRI) to predict adverse morphological ventricular remodelling and functional recovery. Hereto, patients had to be classified in patients that showed morphological/functional remodelling and patients that didn’t; subsequently the best predicting parameter had to be identified:• Patient classification: Patients were classified according to whether their ventricle remodelled morphologically or functionally at follow-up. Obviously, this required a clear definition of “remodelling”. Hereto, a ventricle was defined to have remodelled in terms of morphology – characterized by its end-diastolic volume (EDV) – or function – characterized by ejection fraction (EF) – in case the follow-up and baseline measurements differed by more than two standard deviations of the intrinsic measurement variability. In other words, if the change in EDV / EF was more than what could reasonably be expected as a result of measurement variability, the ventricle was considered to be remodelled.Given that different imaging modalities could have been available to define remodelling, a ‘waterfall’ system was used in which the modality with both baseline and follow-up data and with the highest measurement reproducibility was used first and the least reproducible measurement was used as last fall-back option in case the other (paired) options were not available for that particular patient.• Determining the best predictor: Given that the imaging parameter database contains about 600 variables, simply running a c-statistic on all parameters would make the type I statistical error unacceptably large implying that it would be very unlikely to find significant differences in prognostic power of individual parameters. Therefore, a two-step procedure was implemented as follows:o All imaging parameters were categorized into ‘physiologic parameter groups’, i.e. left ventricular morphology, left atrial morphology, left ventricular global systolic function, etc. A total of 11 physiologic groups were defined.o Each physiologic group was decomposed in its principle components and the number of components that could explain 80% of the variability in the data of that group was retained.o Missing data was accounted for using multiple data imputations.o Given that an individual c-statistic would overestimate the predictive value of a given parameter, a cross-validation methodology was applied.o The c-statistic of each physiologic group was contrasted and the best physiologic group was selected for further detailed analysis.o This detailed analysis existed out of an analysis on the c-statistic of the parameter within this physiologic group.4.1.3.c Main findings of the studyStudy populationOver the 3-year recruitment period, 676 patients were included in the study out of which 25% were female and 46% had a previous myocardial infarction (MI). The age distribution of the study population is shown in Figure 3 when stratified for gender (left) and previous MI (right). Some other demographic data of the study population such as body mass index (BMI), systolic blood pressure, New York Heart Association (NYHA) class, McNew total score and total cholesterol levels are shown in Figures 4-8.About 20% of the total patient population were smokers with similar proportions for both genders.Two years after recruitment, 613 of the patients (i.e. ~91%) returned for a follow-up examination including imaging.Patient classificationBased on the definition used, 168 out of 613 patients (i.e. 33%) remodelled morphologically while this was 106 (i.e. 21%) for functional remodelling.Best predicting parameter groupFigure 9 (left) shows the predictive power of each of the physiological parameter groups for predicting changes in EDV while Figure 9 (right) gives the accuracies for functional remodelling. Overall, left ventricular morphological parameters best predicted both morphologic and functional remodelling.Best predicting imaging parametersWhen focusing on the best predicting parameter group in order to further identify the best predicting individual parameter, the internal left ventricular diameter measured by M-mode echocardiography as well as the left ventricular end-diastolic volume assessed by cine-MRI came out as the better predictors for both types of remodelling as detailed in Table 1 and 2 below.Interpretation of the dataThe data presented above needs to be interpreted in the context of the limitations of the study conducted to date. Indeed, during statistical analysis, several decisions had to be taken ad hoc. For example, how much of the variability within a parameter group needs to be explained by the principle component analysis? How much should 2 paired measurements differ before a patient is classified in the remodelling group? And others. Obviously, the outcome of the study should be independent of these statistical choices. Hereto, a sensitivity analysis was performed whereby the different parameters in the statistical analysis were varied and the effect on the outcome was mapped. Although this analysis, to date, consistently points towards the LV morphological parameter group as the most predictive one, this sensitivity analysis remains incomplete at the time of writing and is currently being elaborated.Similarly, patient classification was currently performed using a ‘waterfall system’ based on reproducibility of the measurement, which may have induced bias. This is currently being investigated further. In this context, it should be noted that the measurement variability has become critically important towards the classification of the study population. For this reason, the consortium will repeat – in the following weeks – this variability assessment (even though the study formally ended) as it is key.Strong conclusive data on the primary objective of DOPPLER-CIP is currently thus still lacking but is expected in the next few weeks. As such, the secondary objective of DOPPLER-CIP (i.e. interpreting the effectiveness of a given modality / parameter in the context of its cost) could not yet be addressed although all cost-related information as well as the models to predict cost-effectiveness has been prepared. It is thus anticipated that shortly after the primary objectives of DOPPLER-CIP have been reached, the cost-effectiveness analysis will be finalized as well.Potential Impact:4.1.4 Potential impact and mean dissemination activities and exploitation of results4.1.4.a Socio-economic impact and wider societal implicationsWith an estimated 17.5 million deaths in 2012, cardiovascular disease (CVD) remains the leading cause of death in the world and accounts for 31% of all global deaths [WHO15]. Out of these, the majority of deaths (i.e. 7.3 million) were due to coronary heart disease. By 2030, the same report predicts an annual increase to 23.3 million deaths from CVDs. Similarly, coronary heart disease is responsible for 5% of all disability-adjusted life years (DALYs) lost [Murray12] and this number is expected to almost double by 2020. In Europe alone, each year, CVD causes over 4 million deaths, which are 47% of all deaths [Nichols12]. Overall, cardiovascular disease is estimated to cost the European Union’s economy almost €196 billion a year out of which around 54% is due to direct health care costs [Nichols12]. At a global level, the direct and indirect costs of CVD was approximately US$ 863 billion and is estimated to rise to US$ 1,044 billion by 2030 [WEF11]. Based on these hallucinating numbers, the increasing need for efficient techniques for patient stratification towards treatment planning (i.e. the goal of DOPPLER-CIP) is obvious.The results of DOPPLER-CIP will lead to a better definition of potentially adverse conditions in patients with CAD such as an impending left ventricular remodelling. This knowledge cannot only be used to support decision making towards treatment at an early stage but can also provide important information during therapy follow-up. Especially the accurate prediction of an impending left ventricular remodelling would allow supporting the decision for or against particular early therapeutic measures. The study will thus have a positive impact on the quality of health care for the individual cardiovascular patient in addition to having a positive impact on economical aspects related to cardiac imaging. Indeed, to date, no large-scale imaging studies had been conducted to provide evidence-based guidelines towards which parameters to measure and what modalities to use in order to accurately predict adverse cardiac remodelling in an economically viable manner. DOPPLER-CIP is the first large, multi-centre study that will enable cardiologists for the first time to select diagnostic imaging modalities on a wide and robust evidential basis. This will have impact on diagnostic cost-effectiveness by avoiding unnecessary examinations, on patient safety by avoiding over-treatment and, thus, on European health care systems by reducing costs.4.1.4.b Main dissemination activities and exploitation of resultsTo date, the scientific output of DOPPER-CIP remains limited. Given that the last patients received their follow-up examinations in November 2014, this is not unexpected. The major scientific contributions and dissemination is thus expected in the following months.The study protocol and the design of the study was published in the Scandinavian Cardiovascular Journal in 2013 [Rademakers13] and presented at the “European Molecular Imaging Meeting” in Antwerp 2014.Based on the preliminary statistical findings of the study (cf. above) a ‘late breaking science’ abstract on the primary goals of DOPPLER-CIP was submitted to the meeting of the European Society of Cardiology (ESC; London, August 29th-September 2nd) – the leading international conference on cardiology. Moreover, based on the preliminary findings, the chair of the technical program committee of the meeting of the European Association of CardioVascular Imaging (EACVI; Sevilla, December 2nd – December 5th) allocated a full session to the study. The major findings of the study will be submitted for publication at the earliest possible but for sure before the end of 2015.In the meanwhile, the data acquired at the recruitment phase of the study, has been used for several methodological studies. For example, one of the partners recently proposed a new algorithm for fully automatic analysis of cine MRI images [Queiros14]. The MR imaging data available in DOPPLER-CIP as well as the extracted volumetric measurements were used to validate this new algorithm in a large cohort of patients in a realistic, multi-vendor clinical setting. The results of this study were submitted for publication in the European Heart Journal – Cardiovascular Imaging (EHJ-CVI) and is currently under revision [Queiros15]. Similarly, volumetric ultrasound data is being used to validate automatic segmentation algorithms for the left ventricle [Barbosa13] and left atrium [Almeida14] while the tagged MRI data has been used for a preliminary validation of a recent automatic software tool [Morais13]. At the time of writing, multiple other sub-studies have been initiated that will result in further publication output from this project in the following months and years.Finally, next to scientific publication of the main findings of DOPPLER-CIP, the results of this study will form the basis to prepare new guidelines for the use of non-invasive imaging for the prognosis of patients with chronic ischemic heart disease. Similarly, the findings of the cost-effectiveness analysis of DOPPLER-CIP will be fed back to the European health care systems. Moreover, a automated software tool was developed for efficient analysis of left ventricular morphology and function. A snapshot of the graphic user interface of this software is shown in Figure 10.Throughout the duration of the project, the project’s website was kept up-to-date where the latest statistics on patient recruitment and scientific board reports were continuous made available.List of Websites:4.1.5 Project website and contact detailsThe project’s website is http://www.dopplercip.beFor further details, updates or information on the study, please contact jan.dhooge@uzleuven.be Related documents final1-final_report_en_figs_and_tables.pdf
