ProfessorDarrel Francis

Darrel Francis is Professor of Cardiology at the National Heart and Lung Institute. He specialises in using quantitative techniques, derived from mathematics, engineering, statistics and machine learning, to problems that affect patients with heart disease.

He trained in Medical Sciences at Cambridge University and trained as a Junior Doctor in Oxford Radcliffe Trust, and developed a strong interest in understanding mechanisms, diagnosis and treatments in quantitative terms.

He moved to Imperial College London for his thesis funded by a British Heart Foundation Junior Research Fellowship, which developed mathematical techniques to analyse phenomena in cardiac dynamics and cardiorespiratory interaction where purely qualitative approaches may be misleading or confusing. He completed his Clinical Specialty training in Cardiology in London while setting up his research group.

Research Themes
Supported by the British Heart Foundation and the European Union, a 10-year programme of research is underway tackling a series of mechanistic and practical problems in clinical cardiology. The group takes special care to try to use measurement methods with verifiable reliability so that results can be trusted to not have arisen by chance or inadvertent exaggeration.

Cardiac "Resychronization" Therapy or not? You decide.






Image: Yacht Plot: Study results mainly determined by bias-resistance of design

We now implant tens of thousands of biventricular pacemakers into patients in heart failure: they improve symptoms and save lives. But it is far from clear how they achieve this. For speed in a busy life, we call them "resynchronization" devices, but this can be misconstrued to imply that they work by making the different ventricular regions contract at the same time.

This assumption has created an uncritical market for indices of "dyssynchrony" which have been introduced in great multitude by enthusiastic proponents who have often claimed strong prediction of response, but these claims are gradually being recognised to be the result of inadvertent measurement bias.

With colleague Dr Zachary Whinnett and their clinical research fellows Dr Afzal Sohaib and Dr Andreas Kyriacou, and international collaborators, Prof Francis is using detailed physiological experiments to explore by reliable methods the mechanism by which these devices give their benefits, so that the scope and direction of further advances can be more confidently evaluated.

When is an optimization not an optimization?
Guidelines recommend that all biventricular pacing devices undergo optimization of their settings. This is a process of making adjustments while observing heart function and then selecting the setting which gives the greatest function. However, the commonly-recommended techniques for optimization have not undergone basic testing by those recommending them.

First, real-life expert cardiologists confidently contradict each other as to which setting is best. More worryingly, the experts contradict themselves as much as they contradict each other, even when viewing identical data less than 10 minutes later. This means that the method (and not the individual) is at fault, and that experts' confidence in a technique is no guide to its reliability.

Second, guidelines often pay only lip service to the existence of beat-to-beat biological variability that makes it necessary to make multiple measurements at each setting so that the standard error of the mean is sufficiently small to detect the subtle differences between settings. Lacking this precaution, guideline-recommended optimization algorithms can be considered a form of roulette which may in about half of cases be worsening the efficiency of the heart, rather than improving it.

With colleagues Dr Zachary Whinnett and Dr Charlotte Manisty, the group is developing a "basic science of optimization". This involves explicit quantitative recognition of the presence of noise as well as signal. This programme of research will provide tools so that clinicians worldwide can reliably evaluate their own local optimization procedures and then conduct optimization confident in the knowledge that the process is not merely placebo - or worse.

Individualised narrow-error-bar physiological measurements
With Dr Zachary Whinnett, Dr Charlotte Manisty and clinical and engineering team members, Prof Francis is developing reliable methods for pacemaker optimization. So far the group have identified the following steps that make the process of obtaining a reliable (i.e. reproducible) optimum quicker and easier:

Raise the heart rate, because the between-setting differences become more distinct
Fit a parabola, rather than simply picking-the-highest value
Repeatedly conduct transitions of AV delay, and measure the pre-post difference
Measure immediately, before compensatory vasodilatation attenuates the signal
Choose a variable for its good signal-to-noise ratio, rather than arbitrarily
Exercise optima can be calculated from resting measurements
Allow enough time for optimization: not less than 12-minutes, even with these steps
Assess VV optimization methods in atrial fibrillation, where AV delay cannot confound
Don't trust experts' recommendations - check the numbers for yourself
In parallel, the group are also carrying out invasive studies, with two major findings so far. First, biventricular pacing enhances coronary flow by increasing the diastolic suction wave.

The second finding is currently controversial. Biventricular pacing appears to increase myocardial oxygen consumption, rather than decreasing it as is commonly assumed. A strength of the study was the careful prior design to ensure error bars were small enough for such interpretations to be made.

Rapid, reliable quantification of physiology through technology-assisted ultrasound measurements
With research fellow Dr Graham Cole, engineering team Drs Massoud Zolgharni and Niti Dhutia and Keith Willson, the group is working towards better tools to support clinical echocardiography for use in clinical cardiology, research, and general medical and surgical environments.

The group is developing better methods to measure indices of ventricular function, intravascular volume status, mitral regurgitation and aortic stenosis, so that more healthcare workers can learn to use the technology and have greater confidence in the reliability and comparability of measurements made by different observers. This may lead to earlier and more accurate diagnosis, and more efficient longitudinal monitoring of patients.

Reliable Science
With colleague Dr Justin Davies and clinical research fellows Dr Matthew Shun-Shin, Dr Sukh Nijjer, Dr Shai Sen, Dr Ricardo Petraco and Dr Judith Finegold, the group seeks to rectify a weakness in Cardiology that reliability of research conclusions is not greatly valued: it is considered sufficient to publish a claim and move on.

Mistaken findings come from two broad sources: random error (displacing results in one direction or another by chance) and bias (displacing results consistently in one direction).

Random Error (Scatter)






Image: Unmentioned reality of raw clinical data

Most clinical measurements have some degree of random error, although this is rarely discussed, perhaps because scientific comments on sources of measurement variability can be misinterpreted as criticism of clinical skill.

The group is developing ways of making measurements that deliver error levels small enough for the desired purpose. For example, when tracking changes in a chronic disease in a patient over time, almost all measured variables fluctuate randomly far more than the disease progresses, i.e. the apparent changes seen on each visit are almost exclusively error. This needs open discussion so that better measurement methods can be sought, and reliably recognised when found.

Non-random Error (Bias)
Relatively little effort is put in many cardiological studies to eliminate bias, in selection or exclusion of patients, or the making of measurements. The consequences can be very large, especially in studies of association betweeen variables where the observers are unblinded and have a strong prior belief.

Unfortunate combination of random error and bias
When random variability between beats is large, clinicians use "clinical judgement" to select which value to report, unaware of the devastating consequences for subsequent use of such selected data in research, where this can cause gross exaggeration.

The Reliable Science drive encourages open discussion of these problems of random error, bias and - worst of all - both entangled together, so that they can be dealt with while there is still time, i.e. before the study begins.

Consultant Cardiologist
Prof Francis is a Consultant Cardiologist at Imperial College Healthcare NHS Trust. This work keeps the focus of the group on the day-to-day problems faced by cardiology patients and their physicians.

This work includes cardiac emergencies as part of the hospital's 24/7 emergency angioplasty service for the care of heart attack patients, and inpatient and outpatient care, as well as outreach clinics such as those in Harrow. He conducts and analyses tests on Cardiology patients, including echocardiography, stress echocardiography and cardiopulmonary exercise testing, and has special expertise in cardiopulmonary interaction.

In 2009 he was awarded the Fellowship of the Royal College of Physicians.

Academic Cardiology Trainees
Aside from research, teaching, clinical practice as a cardiologist, and dodging administrative activities, Prof Francis' other long-term interest is in guidance of cardiology trainees with an interest in clinical research.

He is the academic counterpart and deputy to Professor Jamil Mayet, Training Programme Director for Cardiology in North West Thames for which Imperial College NHS Trust is the Lead Provider.

The group is committed to enhancing Imperial College's position as the most attractive location in the UK for those wishing to pursue such a dual path. Francis has been recognised by the Rector of Imperial College for his commitment to mentoring.

To help protect clinical trainees (academic or otherwise) from being misused as a "service", the group has argued that this is usually a disservice, as can be seen by analogy, algebra and audit.