ARTICLES
Consequences of discharges from intensive care at night
C Goldfrad, K Rowan
Summary
Background It is generally believed that pressure for beds on
intensive-care units (ICUs) has increased in the UK. This
study used discharge at night as a proxy measure to
investigate pressure.
Methods Night was defined in two ways: “out of office hours’
from 2200 to 0659 h and “the early hours of the morning”
from 0000 to 0459 h. The rate of discharge at night was
compared for 21 295 adult admissions to 62 ICUs covering
the period 1995–98 with 10 806 admissions to 26 ICUs
covering the period 1988–90. With data solely from
1995–98, the consequences of discharge at night and
premature discharge were investigated.
Findings Overall, 2269 (21·0%) admissions did not survive
the ICU in 1988–90 compared with 4487 (21·1%) in
1995–98. Of ICU survivors, 2·7% were discharged at night
(2200–0659 h) in 1988–90 compared with 6·0% in
1995–98. In 1995–98, night discharges (2200–0659 h) had
a higher crude (odds ratio 1·46, 95% CI 1·18–1·80) and
case-mix adjusted (1·33, 1·06–1·65) ultimate hospital
mortality. Higher odds ratios were observed when the
definition of night was 0000–0459 h. Premature discharge
was commoner at night, 42·6% vs 5·0% and its importance
was apparent when incorporated into the logistic-regression
model (premature discharge 1·35, 1·10–1·65; night
discharge 1·17, 0·92–1·49).
Interpretation Night discharges from ICU are increasing in
the UK. This practice is of concern because patients
discharged at night fare significantly worse than those
discharged during the day. Night discharges are more likely
to be “premature” in the view of the clinicians involved. The
implication of these results is that many hospitals have
insufficient intensive-care beds. In deciding whether or not to
invest more resources in intensive care we must, however,
consider the cost-utility of this particular service compared
with other ways that additional resources could be used.
Lancet 2000; 355: 1138–42
Intensive Care National Audit & Research Centre (ICNARC),
Tavistock House, Tavistock Square, London WC1H 9HR, UK
(C Goldfrad MSc, K Rowan PhD)
Correspondence to: K Rowan
(e-mail: kathy@icnarc.demon.co.uk)
1138
Introduction
It is generally believed that pressure for beds on intensive
care units (ICUs) has increased. Although no rigorous
research evidence exists, the increasing number of reports
and correspondence related to this subject does suggest
that the pressure on bed availability in ICUs may be
greater now than before.
Increasing numbers of admissions with a concomitant
decrease in length of stay in the ICU have been
reported.1–3 Occupancy rates have been described as very
high.4 The shortage of available ICU beds in London,
UK, and elsewhere2 has been detailed and a wide variation
in provision of facilities has been recorded.4–6 Cancelled
operations due to the lack of available ICU beds4 and high
rates of refused admissions have been reported, both
regionally7 and nationally.5,6 The transfer of patients over
long distances in search of an ICU bed has also been
reported8–10 and the potential dangers of transferring
critically ill patients has been highlighted.11 Premature
discharge of patients has been described.3,12,13
A national bed register covering England was
introduced in 1996 by the Department of Health to ease
the problem of finding a suitable bed for a critically ill
patient.14 In addition, the Intensive Care Society, the
professional organisation of intensive-care doctors, has
called for both an increase in ICU facilities in areas of low
provision and a formal transport system.15
High dependency units (HDUs) have been proposed
for low-risk, short-stay ICU admissions to ease the
pressure.1,3,12,16,17 However, a shortage of such intermediate
facilities also seems to exist.
With the results of the Intensive Care Society’s UK
APACHE II study from July, 1988 to September, 1990,
which indicated a high death rate on the ward after
discharge from ICU (variation 6–16% across ICUs),18 and
the belief that discharging patients from ICUs at night
does not constitute good quality care and, where possible,
is to be avoided, our study used discharge at night as a
proxy measure to investigate pressure on ICUs. The aim
was to investigate the change, over time, in the rate of
discharge at night from ICUs and to find whether there
were any adverse consequences following discharge at
night.
By the use of two high-quality clinical databases, the
UK APACHE II study database,18,19 and the Intensive
Care National Audit & Research Centre’s Case Mix
Programme Database (CMPD),20 the rate of discharge at
night for 1988–90 was compared with the rate for
December, 1995, to April, 1998. We used data solely
from the CMPD to compare the consequences of
discharge at night with discharge during the day.
Methods
Databases
Data were extracted for the 10 806 admissions to 26 ICUs in the
UK APACHE II study database, covering 1988–90, and for the
22 059 admissions to 62 ICUs in the CMPD, covering 1995–98.
All these data had been collected prospectively. Age exclusion
criteria (age <16 years) used in the UK APACHE II study were
THE LANCET • Vol 355 • April 1, 2000
 ARTICLES

APACHE II CMPD
All ICUs Same ICUs All ICUs Same ICUs
ICUs 26 9 62 9
Adult (肁16 years) admissions 10 806 4131 21 295* 4064†
ICU survivors (% adult admissions) 8528 (78·9%) 3219 (77·9%) 16 789 (78·8%) 3255 (80·1%)
Discharges 2200–0659 h (% ICU survivors) 234 (2·7%) 94 (2·9%) 1009 (6·0%) 182 (5·6%)
Range of discharges at night across ICUs 0·6–9·6% 1·9–4·0% 0·7–17·2% 2·2–15·0%
Discharges 0000–0459 h (% ICU survivors) 91 (1·1%) 43 (1·3%) 433 (2·6%) 81 (2·5%)
Range of discharges at night across ICUs 0–4·0% 0–2·3% 0–9·4% 1·1–7·0%
Admissions before applying age exclusion criteria (admissions <16 years) used in APACHE II study=*22 059 and †4206.
Table 1: Proportion of discharges at night from ICUs participating in UK APACHE II study compared with CMPD

applied to the CMPD resulting in 21 295 adult admissions.
Deaths in ICU were excluded from the analyses.
Data
Data were extracted for “time of discharge from ICU”. We
consulted with ICU colleagues and, before the analysis was done,
defined night as “out of office hours” (2200–0659 h) and second
as “the early hours of the morning” (0000–0459 h). Each
admission surviving the ICU stay was categorised as either a
night or day discharge according to the two definitions.
Solely by the use of the CMPD, data were extracted on the
case mix (age, medical history, acute severity). A severe [past
medical] history was defined by the presence of one or more of 16
severe chronic conditions. Acute severity was measured by the
APACHE II score,19 which encompasses weighting for age,
medical history, acute severity (defined by derangement from the
normal range, in the first 24 h in the ICU, for 12 physiological
variables), and for surgical status. Given that two admissions with
the same APACHE II score but with different reasons for
admission can have very different risks of death,21 the APACHE
II probability of hospital death, defined as death before discharge
from hospital after intensive care, was used to describe case mix
overall. Probabilities of hospital death were estimated by the UK
APACHE II model.22
Subsequent readmissions to ICU were identified by the
variables “postcode”, “date of birth”, and “sex”, and confirmed
by the participating ICUs. For discharges directly transferred to
another ICU, in either the same or another hospital, survival data
(alive/dead) at final discharge from ICU or hospital were
extracted. For deaths at final discharge from hospital, the time to
hospital death, in days, was calculated.
Data were also extracted for the variables “reason for discharge
from ICU” and “destination following discharge from ICU”. The
former was based on a clinician’s subjective assessment of a
20
18
patient’s readiness for discharge in the light of the needs of other
patients for the available ICU beds. No attempt was made to
impose standard explicit criteria for this variable.
Length of stay in the original ICU was calculated, in part days,
by the use of data for the variables “date/time of admission to
ICU” and “date/time of discharge from ICU”. For discharges
directly transferred to another ICU, in either the same or another
hospital, data for the variables “date of discharge from ICU” and
“date of ultimate discharge from ICU” were used to calculate, in
days, length of stay in subsequent ICUs after discharge from the
original ICU. Similarly, for all discharges, we used data for the
variable “date of ultimate discharge from hospital” to calculate
the length of stay in hospital after discharge from the original
ICU. In order to verify whether patients were being discharged at
night due to pressure on ICU beds, the time, in hours, to the next
admission (turnover time for the bed) was calculated from the
“date/time of discharge from ICU” for one admission to the
“date/time of admission” for the next admission to the same
ICU.
Analyses
The overall proportion of discharges at night, and the proportion,
by hour, were compared for the two periods for all ICUs and,
additionally, for the ICUs common to both databases. For
1995–98 solely (CMPD), discharges at night were compared
with discharges during the day for case mix, readmission rates
and crude and case-mix-adjusted, ultimate hospital mortality.
Time to death, reason for discharge and time to next admission,
destination after discharge, length of stay in the original ICU, in
subsequent ICUs and in hospital were also compared.
Case-mix adjustment was done by the APACHE II method.
Admissions staying less than 8 h in the ICU were excluded from
calculation of APACHE II scores. In addition, admissions with
burns, after cardiac surgery, directly transferred from another
1988–1990 (n=8 528)
1995–1998 (n=16 789)

16
Proportion of discharges (%)

14

12

10

0
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5 6
Hour of discharge (07:00–06:59 h)
Proportion of discharges by hour for ICUs in UK APACHE II study compared with CMPD

THE LANCET • Vol 355 • April 1, 2000 1139

ARTICLES

Night discharges Day discharges Night discharges Day discharges

(n=1009) (n=15 747) (n=1009) (n=15 747)
Mean (95% CI) age (years) 57·5 (56·4–58·7) 58·2 (57·9–58·5) Reason for discharge
Admissions 肁85 years 29 (2·9%) 386 (2·5%) Fully ready for discharge 445 (44·1%) 13 593 (86·3%)
Admissions with medical history* 127 (12·6%) 2259 (14·3%) Early discharge due to shortage of ICU bed 430 (42·6%) 780 (5·0%)
Admissons eligible for calculation of 792 (78·5%) 14 792 (93·9%) Current level of care continuing in another unit 94 (9·3%) 632 (4·0%)
APACHE II scores† (% of total) Delayed discharge due to shortage of ward 18 (1·8%) 387 (2·5%)
Mean (95% CI) APACHE II score 15·5 (15·1–16·0) 14·6 (14·5–14·7) beds
Admissions eligible for calculation of 706 (70·0%) 13 293 (84·4%) Discharge for palliative care 17 (1·7%) 322 (2·0%)
APACHE II probabilities‡ (% of total) Self-discharge against medical advice 5 (0·5%) 31 (0·2%)
Median (interquartile range) APACHE II 16·2 (7·6–32·1) 13·7 (6·7–26·7)
Median (interquartile range) time to next 2·3 (0·8–15·3) 5·5 (2·0–20·5)
probability (%) of hospital death
admission (h)
*Presence of one or more of 16 defined conditions/therapies: biopsy proven cirrhosis,
portal hypertension, hepatic encephalopathy, very severe cardiovascular disease, severe Median (interquartile range) time to next 1·2 (0·5–5·4) 2·3 (0·9–10·3)
respiratory disease, AIDS, metastatic disease, acute myelogenous leukaemia or acute admission for premature discharge (h)
lymphocytic leukaemia or multiple myeloma, chronic myelogenous leukaemia or chronic
lymphocytic leukaemia, lymphoma, congenital immunohumoral or cellular immune Table 4: Reason for discharge for night (2200–0659 h) versus
deficiency state/home ventilation, chronic renal replacement therapy, steroid treatment, day discharges
radiotherapy, chemotherapy. †Admissions staying less than 8 h ICU were excluded from
the calculation of APACHE II scores. ‡Admissons with burns or after cardiac surgery and
direct transfers from another ICU were excluded from the calculation of APACHE II
(␹2=21·96, p=0·00) and the ultimate hospital mortality
probabilities. was 1·4-fold greater (␹2=23·05, p=0·00). The time to
Table 2: Case mix for night (2200–0659) versus day death in hospital after discharge from the ICU was similar
discharges for night and day discharges. When comparing outcomes
between groups of intensive-care admissions, it is essential
ICU, readmissions and admissions for which hospital outcome
to adjust for case mix. The population eligible for case-
data were missing were excluded from calculation of APACHE II
probabilities and from case-mix-adjusted ultimate hospital
mix adjustment using the APACHE II method (excluding
mortality. admissions staying less than 8 h in the ICU, admissions
with burns or following cardiac surgery, direct transfers in
Results from another ICU, readmissions and patients for whom
Trend in night discharges the hospital outcome was missing) was n=12 951. The
There were 2269 (21·0%) admissions who did not survive odds of hospital death for discharges at night compared
the ICU during 1988–90 compared with 4487 (21·1%) with discharges during the day were significantly
during 1995–98. Overall, 2·7% of discharges occurred at increased both for crude (odds ratio 1·46 [95% CI
night in 1988–90 compared with 6·0% in 1995–98 (table 1·18–1·80]) and for case-mix-adjusted ultimate hospital
1), a 2·2-fold increase. The proportion of discharges at mortality (1·33 [1·06–1·65]). When the definition of
night varied 16-fold across ICUs in 1988–90 compared discharge at night was restricted to discharges between
with 25-fold in 1995–98. Similar results were seen when 0000 h and 0459 h, higher odds ratios were seen, 1·62
the definition for discharge at night was restricted to (1·19–2·21) and 1·53 (1·11–2·13), respectively (table 3).
discharge from ICU between 0000 h and 0459 h. When The variation in the proportion of night discharges
the ICUs common to both databases, denoted “same (0·7–17·2%) and in ultimate hospital mortality
ICUs”, were compared (n=9), similar results were seen. (5·0–24·8%), across the 62 ICUs, suggested
The proportion of discharges were consistently greater heterogeneity. To address the hypothesis that high ICU
from 1600 h to 0700 h for 1995–98 compared with night discharge was associated with high ICU ultimate
1988–90 (figure 1). hospital mortality, potential clustering was accounted for
by robust estimates of variance.23 After adjusting for a
Case mix for night versus day discharges possible cluster effect of ICUs, night discharge remained
significant (p=0·036). Graphical representation of
For 1995–98, the case mix of discharges at night
mortality ratios (observed hospital deaths divided by
compared with discharges during the day are presented in
APACHE II expected hospital deaths) for each ICU
table 2. Mean age was similar, the proportion with a
ordered by proportion of night discharges showed no
severe past medical history was slightly lower for
relation between excess case-mix adjusted ultimate
discharges at night and the acute severity of illness,
hospital mortality and increasing proportion of night
measured by the mean APACHE II score, was higher, as
discharges.
was the median probability of hospital death.

Outcome for night versus day discharges Reason for discharge

The proportion of admissions discharged at night who Only 44·1% of discharges at night were judged by
were subsequently readmitted for intensive care was clinicians to be fully ready for discharge compared with
similar to the proportion discharged during the day (7·9% 86·3% of discharges during the day (table 4). Premature
vs 6·4%, Pearson ␹2=3·65, p=0·06). Ultimate ICU
Definition of night OR (95% CI)
mortality was 2·5-fold greater for night discharges
2200–0659 h
Night discharges Day discharges Crude* 1·46 (1·18–1·80)
(n=1009) (n=15 747) Case-mix adjusted* 1·33 (1·06–1·65)
After adjustment for premature discharge 1·17 (0·92–1·49)†
Readmitted for intensive care 80 (7·9%) 1008 (6·4%)
Mortality at ultimate discharge from ICU 27 (2·7%) 166 (1·1%) 0000–0459 h
Mortality at ultimate discharge from hospital* 183 (18·1%) 2052 (13·0%) Crude* 1·62 (1·19–2·21)
Median (interquartile range) time to hospital 8 (2–20) 8 (2–21) Case-mix adjusted* 1·53 (1·11–2·13)
death for hospital non-survivors (days) After adjustment for premature discharge 1·33 (0·95–1·87)

*Ultimate discharge status from hospital missing for 3·7% of night discharges and for *Ultimate hospital mortality for night versus day discharges.
2·5% of day discharges. Table 5: Outcome for night versus day discharges—impact of
Table 3: Outcome for night versus day discharges premature discharge

1140 THE LANCET • Vol 355 • April 1, 2000


Night discharges Day discharges
(n=1009)
Destination after discharge from ICU
Ward, same hospital 701 (69·5%)
HDU (same or other hospital) 178 (17·6%)
ICU (same or other hospital) 113 (11·2%)
Normal residence 9 (0·9%)
Other hospital (neither ICU or HDU) 8 (0·8%)
Median (interquartile range) stay in ICU 1·3 (0·4–3·6)
(part days)
Median (interquartile range) length of stay 4 (1–14)
in subsequent ICU after discharge from
original ICU (days)*
Median (interquartile range) length of stay 11 (5–21)
in hospital after discharge from original
ICU (days)
*Discharges directly transferred to another ICU (same or other hospital).
Table 6: Discharge destination and length of stay for night
(2200–0659 h) versus day discharges
discharge (“early discharge due to shortage of ICU beds”)
was judged to be much commoner at night (42·6%) than
during the day (5·0%). In contrast, only a small
proportion of discharges at night were judged to be
“delayed due to shortage of ward beds”. These clinical
judgments were corroborated by the finding that the
median time to the next admission at night was less than
half that during the day (2·3 h vs 5·5 h) and was halved
again for those for whom clinicians judged the discharge
to be premature (1·2 h).
Ultimate hospital mortality was greater for patients who
were discharged prematurely (18·9 vs 13·3%, ␹2=29·4,
p<0·01). The difference remained significant after
adjustment for case mix (odds ratio 1·41 [95% CI
1·17–1·71]). The importance of premature discharge was
apparent when it was incorporated into the logistic-
regression model: premature discharge odds ratio 1·35
(1·10–1·65); night discharge odds ratio 1·17 (0·92–1·49,
table 5). Premature discharge and night discharge were
correlated (Pearson correlation coefficient=0·55, p<0·01).
Discharge destination
Most discharges at night went to the ward (table 6).
However, 28·8% were discharged either to a HDU or
another ICU compared with 16·9% for discharges during
the day. Discharges at night stayed a shorter time in the
ICU compared with those discharged during the day.
Length of stay in a subsequent ICU was shorter for night
discharges than for day discharges. Length of stay in
hospital after discharge from ICU was similar for night
and day discharges (p=0·48, Mann-Whitney U test).
Ultimate hospital mortality was associated with a
patient’s destination after discharge from an ICU. For the
majority of night discharges who went to the ward, crude
ultimate hospital mortality was 17·7% compared with
36·4% for those discharged to another ICU (same or
other hospital) and 13·8% for those discharged to a HDU
(same or other hospital). The impact of night discharge to
another ICU, in terms of case-mix adjusted, ultimate
Night discharges
direct to the
ward (n=701)
Readmitted for intensive care 49 (7·0%)
Mortality at ultimate discharge from hospital 124 (17·7%)
Median (interquartile range) time to death 7 (2–20)
for hospital non-survivors (days)
Table 7: Outcome for night discharges direct to the ward
versus day discharges
THE LANCET • Vol 355 • April 1, 2000
ARTICLES
Definition of night OR (95% CI)
(n=15 747)
2200–0659 h
Crude* 1·42 (1·11–1·82)
12 291 (78·1%) Case-mix adjusted* 1·37 (1·06–1·78)
2005 (12·8%) After adjustment for premature discharge 1·18 (0·90–1·56)
673 (4·2%)
0000–0459 h
244 (1·5%)
Crude* 1·73 (1·21–2·48)
527 (3·3%)
Case-mix adjusted* 1·73 (1·19–2·53)
1·6 (0·8–3·9) After adjustment for premature discharge 1·47 (0·99–2·17)
*Ultimate hospital mortality for discharges direct to the ward versus day discharges.
7 (2–16) Table 8: Outcome for night discharges direct to the ward
versus day discharges—impact of premature discharge
10 (5–20) hospital mortality, could not be investigated due to
sample size (only 39 night discharges were eligible for
case-mix adjustment). The impact of night discharge to a
HDU could not be investigated because of bias
introduced by the type of unit (discharge to HDU not
reported for combined ICU/HDUs).
For those discharged directly to the ward at night, both
readmission for intensive care (7·0%) and ultimate
hospital mortality (17·7%) were similar to all discharges at
night (table 7). The population eligible for case-mix
adjustment using the APACHE II method (excluding
admissions staying less than 8 h in the ICU, admissions
with burns or following cardiac surgery, direct transfers in
from another ICU, readmissions and patients for whom
hospital outcome was missing) was n=12 784. The odds
of hospital death for discharges direct to the ward at night
compared with discharges during the day were
significantly increased both for crude (1·42 [1·11–1·82])
and for case-mix adjusted ultimate hospital mortality
(1·37 [1·06–1·78]). When the definition of discharge at
night was restricted to discharges between 0000 h and
0459 h, higher odds ratios were observed, 1·73 both for
crude (1·21–2·48) and for case-mix adjusted (1·19–2·53)
ultimate hospital mortality (table 8).
Ultimate hospital mortality was greater for patients
deemed to have been discharged prematurely (19·0% vs
13·2%, ␹2=29·7, p<0·01). The difference remained
significant after adjustment for case mix (odds ratio 1·45
[1·20–1·77]). The importance of premature discharge was
apparent when it was incorporated into the logistic-
regression model: premature discharge odds ratio 1·39
(1·13–1·72); night discharge 1·18 (0·90–1·56). Premature
discharge and night discharge were correlated (Pearson
correlation coefficient 0·53, p<0·01).
Discussion
Night discharges from ICU doubled in the UK over the
past decade—a worrying trend because patients
discharged at night fare significantly worse than those
discharged during the day. Before considering possible
explanations for these findings, it is important to
recognise a potential methodological limitation—the
adequacy of the UK APACHE II model for case-mix
adjustment. While we can never be certain that all
potential risk factors have been taken into account, the
model used was developed and extensively validated in
Day discharges the UK.19 There could be unknown confounders, such as
(n=15 747)
will-to-live or genetic predisposition, and this uncertainty
can only be resolved by a randomised trial. However, it is
1008 (6·4%)
2052 (13·0%)
hard to imagine such a trial ever being done.
8 (2–21) The rising proportion of night discharges reflects
increasing demand on intensive-care beds. Given that the
level of need (as measured by the median probability of
hospital mortality which was 19·6% in 1988/90 vs 20·1%
1141
ARTICLES
in 1995/98) has not changed significantly, the increase in
demand seems to have arisen from an increase in the
incidence of severely ill patients identified by clinicians as
suitable for intensive care. One contributor will be the
rising rate of surgical procedures; for example, coronary
artery bypass grafting rose 10% per year in the period in
question.24
The main reason why night discharges do worse than
day discharges is that they are more likely to be premature
in the view of the clinicians involved. Their judgments
were validated by the shorter bed turnover interval seen
after discharge. Indeed, while night was still associated
with an increased risk of subsequent death, the risk ceased
to be significant with premature discharge in the model.
Other factors that might account for a worse outcome for
night discharges include poorer quantity and quality of
care available at night both during transfer and at the
destination. Transfers in the middle of the night may be
traumatic both physically and psychologically for patients.
What are the implications of these findings? From the
point of view of clinical audit, it may seem that the
incidence of premature discharges (whether at night or
during the day) would be a useful measure of unmet need.
We would caution against the use of this, because it
depends on clinical judgment and is, therefore, vulnerable
to manipulation. Instead, given the correlation between
the proportions of premature and of night discharges, we
would recommend the latter be adopted as it is most
unlikely to be manipulated by clinicians. It also has the
practical advantages of being a process rather than an
outcome measure.
The implications of our findings for the health system
are that many hospitals do not have enough intensive-care
beds. In deciding whether or not to invest more resources
in intensive care we must, however, consider the cost-
utility of this particular service compared with other ways
that additional resources could be used. Meanwhile, it is
important that all ICUs ensure that they are making best
use of their facilities.
Contributors
C Goldfrad and K Rowan jointly participated in the study design, analysis
of data, and writing of the paper.
Acknowledgments
We thank everyone in the ICUs participating in the Case Mix Programme
and those responsible for local funding. We are grateful for Richard
Marsh’s encouragement to pursue these analyses. We are grateful to Nick
Black and David Spiegelhalter and the anonymous journal referees, for
comments on an earlier draft of the paper. Finally, we acknowledge the
Department of Health and the Welsh Health Common Services Authority
for the intial, 2 year, pump-priming funds to establish ICNARC; Alasdair
Short for his continual support as our clinical advisor; and our sister
organisation the Intensive Care Society.
1142
References
1 Kilpatrick A, Ridley S, Plenderleith L. A changing role for intensive
therapy: is there a case for high dependency care? Anaesthesia 1994;
49: 666–70.
2 Mitchell I, Grounds M, Bennett D. Intensive care in the ailing UK
health care system. Lancet 1995; 345: 652.
3 Bull AR. Rationing intensive care—crisis reactions to shortages should
be replaced by analysis of strategic services. BMJ 1995; 310: 1010.
4 Smith GB, Taylor BL, McQuillan PJ, Nials E. Rationing intensive
care—intensive care provision varies widely in Britain. BMJ 1995;
310: 1412–13.
5 Metcalfe A, McPherson K. Study of provision of intensive care in
England, 1993. London: Department of Health, 1995.
6 Metcalfe MA, Sloggett A, McPherson K. Mortality among
appropriately referred patients refused admission to intensive-care
units. Lancet 1997; 350: 7–11.
7 Working Party. Review of intensive care services in the West
Midlands. Birmingham: NHS Executive West Midlands, 1994.
8 Anon. ICU patient moved three times a day. Health Serv J 1995;
Jan 26: 6.
9 Chadda D. Intensive care units are overloaded, inquiry finds. Health
Serv J 1995; July 20: 4.
10 Dyer O. Bed register could be set up after patient’s death. BMJ 1995;
310: 688–89.
11 Bion J. Rationing intensive care: preventing critical illness is better,
and cheaper, than cure. BMJ 1995; 310: 682–83.
12 Ryan DW, Bayly PJM, Weldon OGW, Jingree M. A prospective
two-month audit of the lack of a high-dependency unit and its impact
on intensive care. Anaethesia 1997; 52: 265–75.
13 McQuillan P, Pilkington S, Allan A, et al. Confidential inquiry into
quality of care before admission to intensive care. BMJ 1998; 316:
1853–58.
14 Warden J. Emergency bed register announced. BMJ 1996; 313: 575.
15 Wallace PGM, Lawler PG. Bed shortages—regional intensive care unit
transfer teams are needed. BMJ 1997; 314: 369.
16 Ryan DW. Rationing intensive care—high dependency units may be
the answer. BMJ 1995; 310: 1010–11.
17 Peacock JE, Edbrooke DL. Rationing intensive care—data from
one high dependency unit supports their effectiveness. BMJ 1995;
310: 1413.
18 Rowan KM, Kerr JH, Major E, McPherson K, Short A, Vessey MP.
Intensive Care Society’s APACHE II study in Britain and Ireland—I:
variations in case mix of adult admissions to general intensive care
units and impact on outcome. BMJ 1993; 307: 972–77.
19 Rowan KM, Kerr JH, Major E, McPherson K, Short A, Vessey MP.
Intensive Care Society’s APACHE II study in Britain and Ireland—II:
outcome comparisons of intensive care units after adjustment for case
mix by the American APACHE II method. BMJ 1993; 307: 977–81.
20 Rowan K. Intensive Care Society has set up a centre for national
audit. BMJ 1996; 313: 1007–08.
21 Gunning K, Rowan K. ABC of intensive care: outcome data and
scoring systems. BMJ 1999; 319: 241–44.
22 Rowan KM. Outcome comparisons of intensive care units in Great
Britain and Ireland using the APACHE II method. Oxford: University
of Oxford, 1992 (DPhil thesis).
23 Huber PJ. The behaviour of maximum likelihood estimates under
non-standard conditions: proceedings of the Fifth Berkeley
Symposium on Mathematical Statistics and Probability. Berkeley:
University of California Press, 1967: 1: 221–23.
24 Black NA, Langham S, Coshall C, Parker J. Impact of the 1991 NHS
reforms on the availability and use of coronary revascularisation in the
UK (1987–1995). Heart 1996; 76: 1–31.
THE LANCET • Vol 355 • April 1, 2000