IN this issue of Anesthesiology, Barr et al. 1make a useful contribution to our understanding of the clinical pharmacology of intensive care unit (ICU) sedation. It is estimated that approximately $1 billion is spent each year in the United States alone on drugs used for sedation in the ICU. Misuse of these drugs contributes to morbidity, mortality, and expense. Optimization of ICU sedation will require characterization of the clinical pharmacology of sedative drugs. Unfortunately, there are a paucity of studies that use blinded designs and intention-to-treat analysis, which report pertinent baseline data and which use specific dosing schedules and standardized cointerventions. 2The report by Barr et al. is an example of the type of study needed to redress this deficiency.
Barr et al. characterize and compare the pharmacokinetics and pharmacodynamics of lorazepam and midazolam when used for postoperative sedation of surgical ICU patients. Lorazepam and midazolam are commonly used ICU sedatives, but it has been unclear which is the optimal agent. A previous and influential comparison of midazolam and lorazepam by Pohlman et al. 3showed no difference in efficacy and, more notably, no difference in the duration of effect of the two drugs after discontinuation, despite known differences in their pharmacokinetics. In fact, the mean duration of effect was less for lorazepam than for midazolam, although the difference was not statistically significant. This report, along with economic considerations, led many institutions to adopt lorazepam for routine ICU sedation.
Simplistically, one might assume that the most reliable method of comparing the duration of effect of two drugs is by direct measurement, but the difficulty with this approach is that the conclusion cannot be extrapolated beyond the depth or duration of sedation (because drug half-time varies with the duration of administration) used in that particular study. Pharmacokinetic–pharmacodynamic modeling, as used by Barr et al. , takes us past this restriction.
Any difference in the duration of effect of two drugs must arise from either pharmacokinetics or pharmacodynamics. Drug A may have a shorter duration of effect than drug B either because it is cleared from the effect site more rapidly (a pharmacokinetic difference) or because the difference in the concentrations defining appropriate sedation and “recovery” is smaller for drug A than for drug B (a pharmacodynamic difference). In their study, Barr et al. carefully characterized the pharmacokinetics of midazolam and lorazepam in ICU patients. Although their findings were significantly different from those previously reported for healthy volunteers, they confirm that the context-sensitive decrement times (the time required for a given percentage decrease in plasma concentration) were much smaller for midazolam than for lorazepam.
If the plasma concentrations of midazolam decrease more rapidly than those of lorazepam after discontinuation of drug administration, the only way that the effect of lorazepam could be shorter-lived than that of midazolam would be if the concentration “decrement” (the difference between the concentration associated with adequate sedation and the concentration associated with return of an appropriate level of consciousness) is less for lorazepam than for midazolam. This question can be approached with pharmacodynamic modeling. Barr et al. use the Ramsay scale, subjectively evaluated ordinal scores of 1–6 (with 6 being unresponsive), to assess the level of sedation. The pharmacodynamic model assumed that the probability of a level of sedation greater than or equal to some value ss (where ss ranges from 2–6) is given by
where C50,ssis the plasma benzodiazepine concentration, P (Sedation ≥ ss) is 50%, and γ determines the slope of the concentration–effect curve. Using this model, the authors predict midazolam C50,ssvalues of 68, 101, 208, 304, and 375 ng/ml for sedation scores of 2–6, respectively, and the comparable values for lorazepam were 34, 51, 104, 152, and 188 ng/ml. These estimates indicate that the relative concentration decrements for recovery from midazolam and lorazepam sedation are not different. However, there is a potential flaw in the pharmacodynamic model used by Barr et al. The reader may have noted that the ratio of C50,ssfor lorazepam and midazolam is exactly 2 for each score. This occurs because the authors assume that the ratio of C50,ssvalues for lorazepam and midazolam potency is the same for all levels of sedation. This is a major assumption that could introduce bias. However, this concern is mitigated by the authors’ allowing γ, the parameter determining the shape of the concentration–effect relation, to be different for midazolam and lorazepam. This adjustment in the model should allow differences in the relative concentration decrements to be detected. Using this approach, the authors find only trivial differences in the relative concentration decrements for recovery after midazolam and lorazepam. This result derived by modeling is consistent with the observed midazolam:lorazepam concentration ratio during sedation of 1.8 and the observed midazolam:lorazepam concentration ratio at the time of extubation of 2.2. Furthermore, the observed concentrations have similar ratios at each level of sedation. Therefore, both formal modeling and empirical observations indicate that the relative concentration decrements for midazolam and lorazepam are not markedly different.
Given the well-characterized pharmacokinetic differences between midazolam and lorazepam and the minimal pharmacodynamic differences, one must conclude that the duration of effect of midazolam is less than that of lorazepam. How then do we explain the observations of Pohlman et al. 3? The most likely explanation is that Pohlman et al. did not blind the clinicians involved in the study and did not carefully control cointerventions, such as the use of analgesics. The primary endpoint of their study, time to return to baseline mental status, may be uninterpretable because we cannot be certain of standardization of sedation or use of analgesics. However, we must also note that there were differences in the demographics of the patients studied by Barr et al. and Pohlman et al. Two are of particular interest. First, the duration of sedation was considerably longer in the study by Pohlman et al. Could tolerance be developing? Second, the patients in the study by Pohlman et al. had higher Acute Physiology and Chronic Health Evaluation scores than those enrolled in the Stanford study (18.5 vs. 9.0). This raises the question of whether underlying illness alters sedation requirements, a possible explanation for why the patients studied by Pohlman et al. required relatively high doses of both agents. It is well-known that the movement response to surgical stimulus is mediated by components of the central nervous system more primitive than the cerebral cortex. Is it possible that lung injury also elicits a response at levels below the cerebral cortex? This is certainly plausible from an evolutionary point of view. If this supposition were true, it would suggest that sedating patients with drugs that work primarily on the cerebral cortex might not be an efficient strategy. The supposition that underlying illness alters sedation requirements can also explain the clinical conundrum of patients who are difficult to sedate and require high doses of sedatives during the early phases of mechanical ventilation, when lung injury is at its peak, but who then prove to be oversedated when this injury resolves and weaning from the ventilator is begun. Hopefully, further studies of this important clinical problem will be conducted using the sophisticated modeling techniques described in the study in this issue of Anesthesiology but that are longitudinal, focusing on how pharmacodynamics (and pharmacokinetics) change during the period of mechanical ventilation.