Interview with Roger Jelliffe

Can you tell us a little bit about your respective roles? What is a typical day like for you?

Michael Neely runs our LAPKB (Laboratory of Applied Pharmacokinetics and Bioinformatics) lab now, and has done a magnificent job of moving everything onward and upward. I could not be happier! He is much smarter and works much harder than I ever could.

I am retired now and consult for him, and only work about half time. My day may be over at the lab, or perhaps working a bit more from home now. I also go to meetings and give talks at various places.

Is there anything that your laboratory does or that is done at your hospital/centre that you would consider innovative?

Most definitely. We all are indebted to Alain Mallet for introducing the nonparametric (NP) maximum likelihood (NPML) approach to population PK/PD modeling, Alan Schumitzky took this on and developed the nonparametric expectation – maximization (NPEM) method. Then Bob Leary, when working with us, developed the nonparametric adaptive grid (NPAG) method, which made everything much faster.

Further, David Bayard and Mark Milman in our lab developed multiple model (MM) dosage design, which uses NP models and computes dosage regimens that hit target goals with maximum precision (minimum expected weighted squared error). No other dosing method does this. It is a unique innovation of our lab. You need NP models to do this, parametric models cannot even perceive the issue of precision in dosing.

In addition, David Bayard has developed a new multiple model optimal (MMopt) sampling strategy to calculate the most informative times to get samples for optimal TDM protocols. It performs much better than D-optimal and similar designs. MMopt is useful not only for optimal estimation of model parameter distributions, but also for optimizing specific clinical tasks such as sampling optimally to hit a future target serum concentration best, or a desired area under the curve (AUC) best. This is also a totally new, improved, and innovative approach.

Further still, David Bayard has implemented a new method for fitting TDM data from acutely ill and unstable ICU patients to better track drug behavior in them. All the usual methods of fitting TDM data assume that the model parameters are fixed and constant throughout the period of data analysis. He implemented the interacting multiple model (IMM) approach for tracking rapidly maneuvering aircraft so as to detect rapid changes in a patient’s model parameters as each new TDM data point becomes available, for much improved tracking of drug behavior in highly unstable patients with high intra-individual variability.  High intra-individual variability used to be regarded as a real problem in TDM. The IMM method does much to overcome this.

We have also tried hard to introduce to the laboratory community an improved (the correct) way of describing assay credibility and precision by using the reciprocal of the assay variance rather than percent coefficient of variation. We also developed our method of estimating creatinine clearance when renal function is changing from the rate of change between a pair of serum creatinine samples rather than using only a single sample.

What technological innovations have entered into use during your career that have permitted a change, or evolution, in practice?

We used to do our work in 1966 over acoustic couplers where you put the phone in a wood box to connect a typewriter to a time-shared mainframe computer. The advent of the personal computer and the laptop have totally changed all this and moved it to the bedside.

In addition, parallel computing has helped a lot for analysis of large systems of multiple drugs that interact with each other. Michael Neely’s Pmetrics software now permits analysis of such large, interacting, multidrug systems. A real step forward!

How did you become interested in your area of expertise?

I have been an adult cardiologist. I wanted to give digitalis glycosides such as digoxin better to patients. In 1966 I took a course in radioisotopes at Oak Ridge Institute for Nuclear Studies, and learned that the equations of radioactive decay also describe the behavior of many drugs. I got the idea that if you know the half-time of a drug, you can control the total amount of drug you permit a patient to have in the body by giving the right doses at the right dose intervals. Everything else grew from this very simple, but very powerful, idea. I am actually very glad that I escaped the traditional education in PK that most people get today!

I am also so grateful to Dr. Richard Bellman for helping me so much when I knew nothing, and for introducing me to the concept of stochastic adaptive control of systems.

Is there anything that you’ve seen elsewhere or heard about and thought “I’d like to incorporate that idea at my center”?

Absolutely. I was stunned when I heard Alain Mallet first present his method of nonparametric population PK modeling at a conference in Airlee, Virginia, in 1983. I went back to our lab and told the guys about this, and Alan Schumitzky took hold of it and started the process of NP population modeling in our lab. Yes, absolutely!

What sort of research do you have on the horizon that you think might influence clinical practice in the future?

One is the incorporation of process noise into the description of drug behavior. You can have great radar and track a plane precisely, but the wind blows it this way and that. That is process noise. Current error models describe measurement errors, and lump other clinical sources of uncertainty into that category. But many clinical uncertainties are the errors with which doses are prepared and given, and errors in recording of the times at which doses and blood samples are given and taken, and in the misspecification of the model of the drug compared to the greater complexities of physiological reality. That is process noise. Dr. Neely is working to correctly implement process noise now for both linear and nonlinear drug models.

In addition, all our current PK equations still describe changes in model parameters as piecewise constant. However, other equations used to describe the changes in gravitational attraction upon orbiting satellites use parameters that change instantaneously. Use of such equations for PK/PD systems might further improve our understanding and control of drug behavior in unstable ICU patients.

What do you consider is the future for TDM and CT? What are you excited about? What are the challenges we face?

I think the greatest challenge we face is the strong tendency shown over the last 60 years to resist quantitative approaches in medicine. The biomedical community and the medical school curricula have simply failed to respond to the quantitative advances in PK/PD which have been made in the last 60 years. Clinical pharmacology used to be a respected medical subspecialty. Now it is dead. The biomedical community and medical education simply have failed to keep up. There are not enough trained people now to teach clinically meaningful quantitative methods such as PK and decision theory to medical students, where the teaching should begin. We continue instead, decade after decade, to turn out untrained physicians who have no idea of the proper use of drugs for patient care.

The IATDMCT leads this field of education, I believe, because it is the society most focused on the individual patient, and the one most free of domination by the pharmaceutical industry and its continued “one size fits all” approach to drug therapy. TDM can be on the verge of great changes as our new quantitative capabilities become translated into real changes in clinical medicine.

Dr. Neely and I are editing a book, “Individualized Drug Therapy for Patients“, which should come out in November by Elsevier. We are most excited about this.