Getting the dose right: anaesthetic drug delivery and the posological sweet spot

Posology, a scientific term not in common usage, is the science of drug dosage; it is thus a branch of clinical pharmacology (or perhaps a synonym of sorts). Combining the Greek words 'posos' (how much) and 'logos' (science), posology can be thought of more simply as 'dosology'. In the posology of anaesthesia, the fundamental question anaesthetists must answer each day is: 'What is the right anaesthetic dosing strategy for my next patient?' In this issue of the British Journal of Anaesthesia, van den Berg and colleagues1van den Berg J Eleveld D De Smet T van den Heerik AS van Amsterdam K Influence of Bayesian optimization on the performance of propofol target-controlled infusion.Br J Anaesth. 2017; 119: 918-933Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar report a novel approach to optimizing posology in anaesthesia. Their study was an attempt to personalize target-controlled infusion (TCI) therapy with a single observation from the patient. Taking a Bayesian approach, the authors started with pharmacokinetic (PK) parameters from a population model2Eleveld DJ Proost JH Cortínez LI Absalom AR Struys MM A general purpose pharmacokinetic model for propofol.Anesth Analg. 2014; 118: 1221-1237Crossref PubMed Scopus (96) Google Scholar and then adjusted them based on the difference between the predicted drug concentration and the observed drug concentration measured in real time from a single blood sample from the patient. Bayesian estimations of PK model parameters have a decades-long history since their introduction by Sheiner and colleagues in 1979.3Sheiner LB Beal S Rosenberg B Marathe VV Forecasting individual pharmacokinetics.Clin Pharmacol Ther. 1979; 26: 294-305Crossref PubMed Scopus (383) Google Scholar Bayesian methods are intuitively appealing, in part because the approach is somewhat similar to how humans solve problems: start with information that is available a priori, and adjust based on the difference between the a priori information and the observation, normalized by their variability. This moves the adjusted system from the a priori starting point (e.g. the population-based PK model parameters) towards the specific situation at hand, the individual patient's PK parameters. Unless the individual patient is perfectly represented by the population PK model, Bayesian adjustment should improve PK model performance. On the contrary, if the a priori information already allows good predictions of observations (in this instance, if concentrations predicted by the population PK model are close to observed concentrations), Bayesian adjustment is not expected to improve model performance much. The study by van den Berg and colleagues1van den Berg J Eleveld D De Smet T van den Heerik AS van Amsterdam K Influence of Bayesian optimization on the performance of propofol target-controlled infusion.Br J Anaesth. 2017; 119: 918-933Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar tested this hypothesis in a sophisticated way. Given that the PK model was sufficiently good, there was essentially no improvement in accuracy, although a modest reduction in model bias was achieved. A clear message from the study is that the propofol model of Eleveld and colleagues performs well in the patients and conditions in which it was applied in the study.2Eleveld DJ Proost JH Cortínez LI Absalom AR Struys MM A general purpose pharmacokinetic model for propofol.Anesth Analg. 2014; 118: 1221-1237Crossref PubMed Scopus (96) Google Scholar The Bayesian adjustment, therefore, was not very useful in this instance. However, despite the 'negative' findings, the authors have done something important by demonstrating that real-time, real-world Bayesian adjustment of a pharmacological model in the acute care clinical setting is feasible. Various permutations of their Bayesian adaptation approach can be applied to pharmacokinetic and pharmacodynamic (PD) models that are currently implemented in numerous technologies, including open-loop TCI systems and closed-loop delivery systems, among others. Although the Bayesian adaptation approach was not fruitful in this study, it might be useful for other models, particularly less robust models with poorer overall performance. Why are investigations such as the study by van den Berg and colleagues1van den Berg J Eleveld D De Smet T van den Heerik AS van Amsterdam K Influence of Bayesian optimization on the performance of propofol target-controlled infusion.Br J Anaesth. 2017; 119: 918-933Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar undertaken in the first place? What is the driving force motivating studies like this? The answer is simple; getting the dose right is the fundamental pharmacological task of clinical anaesthesia. And this is complicated. In most therapeutic areas within medicine, the 'decision space' for rational dosing can be conceptualized along axes of effectiveness and safety, and there is, ideally, considerable overlap between the two (i.e. high therapeutic indices as in Fig. 1A). For many anaesthetic drugs, not only is the overlap of 'safe and effective' much smaller (i.e. low therapeutic indices), there is also a third axis, 'efficiency', to be considered when choosing a drug and formulating a rational dosing scheme (see Fig. 1B). In the context of anaesthesia practice, pharmacological efficiency describes how the choice of drug and the dosing schedule impact the ratio of patient care quality and costs in terms of emergence times, restoration of protective reflexes, time to return of spontaneous ventilation, need for postanaesthesia monitoring, etc. Most therapeutic areas in medical practice are not constrained by this efficiency imperative (i.e. no need to turn the therapy on and off with precision).4Egan TD Shafer SL Target-controlled infusions for intravenous anesthetics: surfing USA not!.Anesthesiology. 2003; 99: 1039-1041Crossref PubMed Scopus (52) Google Scholar In contrast, in the operating room, the coma of anaesthesia must be produced and reversed on demand, as though it were a 'light switch'.5Brown EN Lydic R Schiff ND General anesthesia, sleep, and coma.N Engl J Med. 2010; 363: 2638-2650Crossref PubMed Scopus (623) Google Scholar 6Egan TD Is anesthesiology going soft?: trends in fragile pharmacology.Anesthesiology. 2009; 111: 229-230Crossref PubMed Scopus (30) Google Scholar In devising a dosing strategy to achieve these goals, having more axes in the decision space and having less overlap between these axes mean that the dosing 'sweet spot' (i.e. the optimal posological area) is small and must be targeted accurately. The dosing sweet spot exists at the relatively small nexus of safety, effectiveness, and efficiency. Hitting this sweet spot is challenging, because the position and size of the conceptual circles shown in Figure 1B are known only with a considerable degree of uncertainty. Typical dosing schemes are based on population PK and PD models; individual patients are sometimes not well described by these models. Thus, personalizing the models, as with the study by van den Berg and colleagues,1van den Berg J Eleveld D De Smet T van den Heerik AS van Amsterdam K Influence of Bayesian optimization on the performance of propofol target-controlled infusion.Br J Anaesth. 2017; 119: 918-933Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar is an important goal of contemporary anaesthetic pharmacology research. Personalizing models to account for common variables that impact a drug's disposition and effects is a well-established aim. For example, recent work has advanced our understanding of the influence of body weight and age on the clinical pharmacology of propofol and remifentanil, refining the existing models,2Eleveld DJ Proost JH Cortínez LI Absalom AR Struys MM A general purpose pharmacokinetic model for propofol.Anesth Analg. 2014; 118: 1221-1237Crossref PubMed Scopus (96) Google Scholar 7Eleveld DJ Proost JH Vereecke H et al.An allometric model of remifentanil pharmacokinetics and pharmacodynamics.Anesthesiology. 2017; 126: 1005-1018Crossref PubMed Scopus (42) Google Scholar, 8Kim TK Obara S Egan TD et al.Disposition of remifentanil in obesity: a new pharmacokinetic model incorporating the influence of body mass.Anesthesiology. 2017; 126: 1019-1032Crossref PubMed Scopus (17) Google Scholar, 9Minto CF Schnider TW Egan TD et al.Influence of age and gender on the pharmacokinetics and pharmacodynamics of remifentanil. I. Model development.Anesthesiology. 1997; 86: 10-23Crossref PubMed Scopus (897) Google Scholar, 10Egan TD Huizinga B Gupta SK et al.Remifentanil pharmacokinetics in obese versus lean patients.Anesthesiology. 1998; 89: 562-573Crossref PubMed Scopus (216) Google Scholar, 11Schnider TW Minto CF Shafer SL et al.The influence of age on propofol pharmacodynamics.Anesthesiology. 1999; 90: 1502-1516Crossref PubMed Scopus (768) Google Scholar and optimizing anaesthetic drug administration through understanding PK and PD interactions.12van den Berg JP Vereecke HE Proost JH et al.Pharmacokinetic and pharmacodynamic interactions in anaesthesia. A review of current knowledge and how it can be used to optimize anaesthetic drug administration.Br J Anaesth. 2017; 118: 44-57Abstract Full Text Full Text PDF PubMed Scopus (33) Google Scholar The study by van den Berg and colleagues1van den Berg J Eleveld D De Smet T van den Heerik AS van Amsterdam K Influence of Bayesian optimization on the performance of propofol target-controlled infusion.Br J Anaesth. 2017; 119: 918-933Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar extended this approach by incorporating information about the disposition of propofol in individual patients into an existing PK model. There are parallel efforts aimed at increasing the size of the optimal posological area by moving the circles of Figure 1B inward or increasing their size, by reducing the uncertainty about the size and position of the circles, and by relaying the real-time location and trajectory of the individual anaesthetic procedure to the clinician in an actionable format. These include anaesthetic drug development, PK and PD research, more robust and accurate concentration and effect sensors, and making PK and PD information available to the clinician through advanced pharmacological displays at the point of care. With a larger optimal posological area and increased personalized situational awareness, drug delivery decisions are better informed and have larger error margins, minimizing adverse effects and enhancing the likelihood of successful therapy.van den Berg and colleagues1van den Berg J Eleveld D De Smet T van den Heerik AS van Amsterdam K Influence of Bayesian optimization on the performance of propofol target-controlled infusion.Br J Anaesth. 2017; 119: 918-933Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar attempted to personalize anaesthetic drug delivery without relying on closed-loop control. Instead, they personalized the PK model exploiting Bayesian concepts. In doing so, they show a path on how to integrate point-of-care i.v. drug concentration monitoring into drug delivery automation. They have also shown the value of carefully selecting a population-based PK model, which in their study situation already fitted the patients so well that personalizing it did not further improve drug delivery performance. Wrote and edited the manuscript: K.K., T.D.E. K.K. was a salaried employee of Dräger (Lübeck, Germany) until May 2014. T.D.E. is on the Associate Editorial Board of the British Journal of Anaesthesia.