Predictive performance of the Domino model during low-dose ketamine infusions in volunteers

 

J. Halliday and M. Ogden, P. Corlett, G. Honey, M. Lee, T. de Smet, P. Fletcher, D. Menon, A.Absalom

University Departments of Anaesthesia and Psychiatry, Addenbrookes Hospital, Cambridge, UK

Introduction:

Ketamine is being used to investigate the glutaminergic hypofunction theory of schizophrenia. We have performed 4 studies of the effects of target-controlled infusions of ketamine (Domino pharmacokinetic model1). The aim of our study was to assess the predictive performance of the Domino model, and to assess the likely performance of 3 other models.

Methods:

Following local ethics committee approval, 57 healthy volunteers performed cognitive function tests and/or underwent functional MRI (fMRI) imaging whilst receiving either saline or racemic ketamine administered using a TCI device (Anaetech Ltd, Leeds UK) programmed with the Domino model. In study 1 (n = 14) and study 2 (n = 12), volunteers received ketamine on two occasions, once at a target concentration of 50 ng/ml and once at a target of 100 ng/ml. During study 3 (n = 15) volunteers received TCI ketamine at a target concentration of 100 ng/ml on one occasion, whereas during the 4th study (n = 16) volunteers received ketamine on one occasion but at two different target concentrations (100 ng/ml for 105 mins followed by 200 ng/ml for a further 105 minutes). For each study, at each target concentration, 2 or 3 venous blood samples were withdrawn during infusion, with a further sample after the infusion ended. Samples were centrifuged, and plasma was removed, frozen and later assayed by gas chromatography for ketamine. The total number of samples was 251.

                Prediction performance was assessed using the criteria recommended by Varvel et al. Samples withdrawn after the infusions stopped were not included in the divergence calculation. The slope of the regression curve of performance errors over time was also calculated, again excluding samples withdrawn after infusion. To estimate the predictive performance of other models (Clements 125, Clements 250, and Hijazi), we used Stanpump to calculate the plasma concentrations the 3 other models would have predicted at the times the samples were withdrawn, and subjected these predicted  concentrations to the same Varvel analysis.

Results:

In all studies the predictive accuracy of the Domino model changed over time. Soon after a target concentration increase, the Domino model significantly overpredicted the ketamine concentration, typically by ~100%, samples after ~ 60 min were generally clustered around the predicted concentration. For later samples the model underpredicted the concentration. The performance criteria are summarised in Figure 1.

 

Figure1: Comparison of Domino, Clements 125, Clements 250 and Hijazi model parameters.

                Parameter

	Parameter
Model	MDPE	MDAPE	Wobble	Divergence	Slope of PE's
Domino	-4.8	34.7	23.7	0.0495	0.4276
Clements125	-20.5	27.0	27.2	-0.0618	-0.0098
Clements250	3.9	24.0	29.1	-0.1615	-0.0959
Hijazi	24.2	47.5	45.8	0.1683	0.4448

 

Discussion:

The systematic change in predictive performance of the Domino model over time is sub-optimal for fMRI and cognitive studies (where steady state plasma concentrations are ideal). This problem is not well shown with the Varvel criteria, but is better shown by the slope of the regression curve of the PEs over time. For this reason, we believe that the Clements250 model may have been a better model for control of TCI ketamine. The measured concentrations have been subjected to NONMEM analysis to develop a new model that we will test prospectively in a subsequent study.

References:

1. Domino EF et al. Anesth Analg 1982; 61:87-92

2. Varvel JR et al. J Pharmacokinet Biopharm 1992; 20:63-94

3. Clements JA et al. Br J Anaesth 1981; 53: 27-30

4. Hijazi Y et al. Br J Anaesth 2003; 90: 155-60