Safer discharge from intensive care to hospital wards

Intensive care in the United Kingdom is certainly underprovided relative to many developed counties. The United States spends over 1% of its gross national product on providing intensive care, while Britain spends around 0.05%—possibly twentyfold less.¹ But intensive care remains largely outside the evidence based paradigm—apparently for ethical reasons. Judging the appropriateness of intensive care provision still depends solely on apparent unmet need and observed associations of prognostic indicators with mortality. This week's BMJ sees another such study.²

It seems too easy to claim benefit in intensive care on the basis of biological plausibility and observational comparison alone. For example, a recent Cochrane overview of 30 trials on the effect of intravenous albumin for acute renal failure,³ which showed significant harm, was dismissed by some enthusiasts.⁴ The writings of respected and dispassionate authors who have held intensive care to be immune from randomisation do not help this apparent impasse.⁵,⁶ In the absence of any rigorous measurement of attributable effect, intensive care will continue to be provided on the basis of evidence that is unacceptable in other areas of health enhancement. With 21st century medical technology this is particularly unfortunate.

The acute physiology score (APACHE) developed by Knaus and others⁷ was designed to provide reliable, physiology based, indices of likely benefit by predicting hospital mortality from measurements made among critically ill adults in hospital. This score has been shown to be a potent measure of casemix that predicts mortality well in the British context⁸ and hence enables comparisons of intensive care performance to be disentangled somewhat from practice variations, be they discretionary or enforced by shortage. Obviously cases that are refused admission to intensive care units cannot be readily compared, since scoring is impossible unless a patient is admitted.⁹ In this week's BMJ another score is proposed, which again predicts hospital mortality based on physiological measurements (p 1274).²

Daly et al looked at some 13 000 intensive care patients in 20 centres and derived a score based on physiological and other measurements on the last day in intensive care to predict risk of death before hospital discharge. The objective of the work is to provide intensivists with an index predicting risk of hospital death associated with discharge, so that patients may be discharged sensibly from scarce intensive care unit beds to make room for severely ill patients. The authors also use it to identify extra capacity needed in intensive care units to avoid the discharge of high risk patients. The main outcome of interest should not, of course, be mortality: it should also include a measure of survival duration and also be quality adjusted.

The measure was derived from a subset of data collected from one of the intensive care units and subsequently tested on a different subset from the same unit and on data from 19 other British centres.² Having derived a score that predicted hospital mortality well on the test data, the authors then used the score to estimate the fate of those discharged while deemed to be at high risk compared with those who were not. Taking patients who were above the risk threshold at some stage during a minimum of a three day stay in the intensive care unit, Daly et al compared the high risk discharged patients with those discharged while at lower risk. The latter had lower actual mortality than the former. The crux of the paper is this comparison, and it is used to predict the consequence of one or two days extra stay in intensive care for high risk patients, when their score would then assign them to the low risk group.

These prognostic indices cannot explain all the intrinsic determinants of mortality in a dynamic system on any absolute scale, whatever the amount of discrimination shown, and sadly the shortfall cannot even be guessed at. Disentangling intrinsic patient risk from effects of care will remain impossible without randomisation. Such evidence about the risk of unnecessary death can be seriously misleading if intensive care is going to have to be assessed in this way. For example providing more intensive care beds in response to refused admissions to intensive care seems to lead, because of consequent changes in the threshold for referral and admission, to a greater total number of refusals¹⁰—and thus, logically, to still more beds.

Similarly here the logic of these extrapolations seems to assume that all clinical decisions in intensive care units are made on the basis of intrinsic need, independently of extrinsic influence, in a static system of patient care. It assumes that experienced clinicians do not assess risk and possible benefit very well in individual patients—which could give rise to some of the observed differences. Assuming that the dominant determinant of actual risk of hospital death is a physiological risk score, however discriminatory, is unwise. Who knows what complex processes led to discharge in each case, and how they might change under different influences on the individual clinical decision. The effect of high dependency beds, to name but one factor, is unassessed by this work.

We need to understand more about the determinants of death in critically ill patients because many lives are at risk and the care is expensive, but observational comparisons, incorporating sophisticated indices of risk, can only raise hypotheses. Daly et al certainly suggest a prospective test of their hypotheses—but secure validation of the score itself (against, for example, current APACHE scoring, which has been validated) would be another prerequisite. In the end, intensive care provision at the margin of possible benefit simply has to be assessed by random allocation like everything else about which there is legitimate doubt. There is currently no substitute—unless we are to end up spending 1% of gross national product on intensive care—whatever its actual effect.

Papers p 1274