Computer Aided Clinical Trials
Starting from a database of 100’s of real patient electrogram records, we describe how to develop and use a large
in-silico cohort consisting of 10,000+ heart models to improve the planning and execution of a clinical trial (CT)
for implantable cardioverter defibrillators (ICDs). We illustrate our approach by applying it retrospectively to a
real CT that compares two discrimination algorithms (DA) within ICDs for the detection of potentially fatal
cardiac arrhythmias. The CT posited that one algorithm would be better than the other but the results of the trial
were opposite to this hypothesis. With our in-silico cohort we are able to provide early insight prior to a CT by
generating patient populations across a range of heart conditions and distributions and also feed the same signals
to multiple devices across a range of configurations. This effort is an early step towards using computer modeling
as regulatory-grade evidence for medical device certification.
Introduction: Clinical trials for ICDs rely on historical data to formulate hypotheses and estimate statistics. This
process can be assisted by early and fast large-scale experiments on computer models of the heart. We sought to
estimate, using model-based analysis, the specificity and sensitivity of two different algorithms for discriminating
between Ventricular Tachycardia (VT) and SupraVentricular Tachycardia (SVT), as a first step in planning a trial
for comparing devices running the two algorithms.
Materials and Methods: We begin by generating a population of 10,000+ synthetic heart models and
implementing diagnostic algorithms of two very commonly used ICD platforms, i.e. Boston Scientific and
Medtronic. We performed conformance testing to validate our device models against real ICDs. Now, using the
closed-loop of the device models and in-silico cohort we conducted multiple trials to compare the performance of
the two algorithms to appropriately discriminate between potentially fatal ventricular tachycardias (VT) and nonfatal
SupraVentricular Tachycardias (SVTs).
Results and Discussion: This model-based trial aims to evaluate the specificity and sensitivity of two VT/SVT
discrimination algorithms (DA): Rhythm ID (RHID, Boston Scientific) and PR Logic + Wavelet (PRLW,
Medtronic). We created a computer model of cardiac electrical activity, and used it to simulate 11,400 arrhythmia
episodes [atrial Flutter (AFL), atrial fibrillation, other SVT, sustained and non-sustained VT and VF]. We
implemented the two DA in software using descriptions from publicly available literature. Each episode is run
through both DA. The results of our in-silico pre-clinical trial indicate that Boston Scientific's algorithm was less
able to discriminate between SVT and VT and so may lead to inappropriate therapy. We further demonstrated that
the result continues to hold if we vary the characteristics of the synthetic population and device parameters.
Simulations took less than a day to complete.
Conclusions: While in-silico trials do not seek to replace a CT, they may provide early insight into the factors
that affect the outcome at a fraction of the cost and duration and without the ethical issues. This model-based
analysis can guide CT investigators in the calculation of effect size, sample size, and choice of eligibility criteria.
Figure 1: Overview of a model-based clinical trial. 1) EGM recordings of real patients are adjudicated to create a table of EGM morphologies of various tachycardias. 2) Subsets of these morphologies are combined with a timing model to create a synthetic heart model. 3) Through variation of the parameters of the model, an entire synthetic cohort is generated and simulated to produce synthetic EGM signals. 4) Various device evaluation experiments can be executed with this synthetic cohort.