help button home button ClinMed NetPrints
HOME HELP FEEDBACK BROWSE ARTICLES BROWSE BY AUTHOR
Warning: This article has not yet been accepted for publication by a peer reviewed journal. It is presented here mainly for the benefit of fellow researchers. Casual readers should not act on its findings, and journalists should be wary of reporting them.

This Article
Right arrow Abstract Freely available
Services
Right arrow Similar articles in this netprints
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Daly, K.
Right arrow Articles by Chang, R. W S
Right arrow Search for Related Content
PubMed
Right arrow Articles by Daly, K.
Right arrow Articles by Chang, R. W S

clinmed/2000060005v1 (July 27, 2000)
Contact author(s) for copyright information

 

Triaging ICU discharges to reduce mortality from inappropriate early discharges

Kathleen Daly, SRN, BSc
St Thomas’ Hospital, London

R Beale, FRCA
Consultant Intensive Care Physician
St Thomas’ Hospital, London

RWS Chang, BSc MS FRCS
Consultant Transplant Surgeon
St George’s Hospital, London

Correspondence to:      R W S Chang


Renal Offices
St George’s Hospital
Blackshaw Road
London SW17 0QT
Phone: 0181 725 3869
Fax: 0181 725 2068
Email: renechang@compuserve.com


Abstract

Objectives:   The aim of this study was to develop a predictive model to triage ICU discharges to reduce post ICU discharge deaths.

Design:  Logistic regression analyses and modelling of data of  patients who were discharged alive from intensive care units. 

Setting:  20 United Kingdom ICUs between 1989 to 1998.  

Participants:  3 data sets of ICU survivors – Guy’s Development (5,475 patients ); Guy’s Validation (1,136); and the Riyadh ICU Program Users Group (RIPUG) Validation (7,313) from 19 other ICUs. 

Main outcome measures:  Post ICU discharge mortality and power of triage model.

Findings:   Patients had a post-ICU discharge mortality of up to 12.4%.

The triage model identified patients at risk from death on the ward with a sensitivity and specificity of 65.5% and 87.6%, respectively and an area under the ROC curve of 0.86. Among patients who stayed in the ICU for more than 3 days, post ICU discharge mortality was reduced by 39% for patients at risk if they remained in the ICU for another 48 hours compared to those who were discharged on the same day.  There were 2,875 at risk among the patients in the Validation data sets with a post discharge mortality of 25%.  An additional 5,750 (16%) ICU bed days would be required if these patients were to remain another 48 hours in the ICU to reduce their post-discharge mortality.

Conclusions:  An intensive care discharge triage model has been developed that may reduce post ICU discharge mortality by up to 39%.  However, implementing such an approach will require a major increase in the provision of fully staffed ICU beds in the United Kingdom.


Introduction

The winter of 1999 has highlighted the acute shortage of intensive care beds in the United Kingdom. One of the consequences of a shortage of ICU beds is that patients are often discharged early to make room for the more severely ill.  The United Kingdom APACHE II study, in 1993, reported a step-up in mortality of between 6.1 to 16.3 %1, 2   following discharge from intensive care.  This increase in mortality has been attributed to premature and inappropriate discharge from intensive care and delayed admission to intensive care 3, 4..  Previous attempts to examine the causes of death following intensive care unit (ICU) discharge have focused on factors occurring after ICU discharge 5 and following hospital discharge 6, 7.  Recently, Goldfrad and Rowan examined factors effecting patient outcome that are present prior to ICU discharge 8. Using discharges at night as a proxy measure of inappropriate early discharge from the ICU they reported a 1.4 fold increase in ultimate hospital mortality for these patients. Other studies have highlighted significantly higher severity of illness scores or TISS scores on the day of ICU discharge among patients who died subsequently on the wards than hospital survivors 9,10.  In the current provision of ICU beds in the UK, there will always be a need to discharge patients to make room for the more severely ill.  It is therefore important to have an objective way to identify patients at risk of dying on the ward if they were discharged too early.

 

This paper reports the development of discharge triage predictive model to identify those patients who died on the ward after discharge from the ICU, who would have benefited from a longer stay in the ICU and explored the implications of its use.  

 


Patients and Methods

All ICU survivors discharged from Guy’s Hospital’s ICU between 1st June 1990 and 31st December 1998 were included in the study as were survivors from 19 other UK ICUs between June 1989 and September 1996 (The Riyadh ICU Program Users Group database).  Daily physiological and treatment data collected prospectively using the Riyadh Intensive Care Program (Medical Associated Software House, London, UK) were analysed to identify variables that had a significant influence on ward death following ICU discharge.   Severity of illness and intensity of treatment were measured using the Acute Physiology and Chronic Health Evaluation II (APACHE II) system 11, the Organ Failure Score (OFS)12 and the Therapeutic Intervention Scoring System (TISS) 13.  This data together with demographic data, including the presence of chronic ill health (as defined using APACHE II criteria), and patients’ hospital outcome, were entered daily onto the computer by a team of specifically trained nurses and doctors. 

 

Model Development Data Set

There were 6,319 patients admitted to the 13 bed general (medical, surgical and cardiothoracic) adult ICU at Guy’s hospital between 30th June 1990 and 31st December 1996.  Eight hundred and forty-four (13.4%) patients who died in the ICU were excluded from the analysis.  Of the remaining 5,475 (87%) ICU survivors, 200 (3.7%) patients died on the wards whilst 5,275 (96.3%) survived to leave hospital. Twenty-five (12.5%) ward deaths and 117 (2.2%) hospital survivors were re-admitted to the ICU during the same period of hospitalisation.  Only data from the patient’s last day in the ICU during their first ICU admission was used to develop the predictive model.  There were 3,133 (57.2%) patients who were admitted to the ICU following cardiothoracic surgery (97% of which was following elective surgery) – a relatively low risk group. For the purposes of analyses, a variable denoting whether or not the patient had cardiothoracic surgery coded 1 and 0, respectively was created.

Bivariate analysis was performed to identify those variables that differentiated the ward deaths from the hospital survivors. Variables found to have a statistically significant influence on survival following ICU discharge were subjected to multivariate logistic modelling to select variables that produced the ‘best’ predictive model of ward death following ICU discharge. 

 

A stepwise forward logistic regression procedure was used to derive the model.  The large disparity in the number of survivors compared to non-survivors resulted in a loss of discriminatory power of the logistic regression function.  To correct for this, a random selection of 250 cases from the 5,275 hospital survivors were merged with the 200 ward deaths to obtain a data set for model development.  This process was repeated twenty times to produce 20 modelling data sets.  The ‘best’ model and cut-off were then selected.  Calibration of the model was assessed by the Hosmer Lemeshow ‘goodness of fit statistic’ 14 for significance (p > 0.05).  Discrimination was assessed using ROC curve analysis 15. 

Model Validation

The model’s ability to triage ICU discharges was then evaluated by applying the triage model on a Validation Data Set derived from ICU survivors from Guy’s Hospital ICU admitted between 1st January 1997 to 31st December 1998 and ICU survivors from 19 other UK ICUs admitted between June 1989 and September 1996.

Use of model to alter outcome

There is not much use for a model to identify patients at risk, if the outcomes of these patients cannot be altered.  To test the ability to alter patient outcomes, we selected patients who stayed in the ICU for more than three days and had a discharge triage score which equalled or exceeded the cut-off (0.6) any time within the 48 hours prior to ICU discharge.  Any ward deaths known to be ‘not for resusitation’ at ICU discharge were excluded from the analysis.  Patients who stayed in the ICU for more than 72 hours were classified into 4 groups depending on when the discharge triage model last predicted them to be at risk of dying on the wards (i.e. the cut-off  was equalled or exceeded) and the timing of their discharge from the ICU.  Group 0 were patients last predicted on the day of ICU discharge. Group 1 were patients last predicted 24 hours prior to discharge; and Group 2 were last predicted 48 hours prior to discharge; Group 3 were patients who were not at risk within the 48 hours prior to discharge.

 

Data analysis was performed using the statistical software package SPSS version 9.0 (Woking, UK).  Categorical data were analysed using Chi-Squared tests.  Non-normally distributed continuous data were evaluated with the Mann-Whitney test.  Logistic Regression Analysis was used to develop the predictive model.  A p value of < 0.05 was considered significant. 

 


Results

The patients’ demographic data, clinical features and severity of illness are given in Table 1.  Higher severity of illness scores (mean APACHE II score 19.2 v 13.5), longer ICU stay (median days 3 v 1), a longer period on mechanical ventilation (median days 3 v 1), and a greater requirement for renal support (21% v 5%), were positively associated with ward death. Cardiac surgery, however, was positively associated with hospital survival (15% v 59%).  Whilst there was an overall reduction in the APACHE II scores during the patients stay in the ICU, the scores for ward deaths remained significantly higher at ICU discharge (mean APACHE II score 15.8 v 11.9).  Analyses of the components of the APACHE II determined that increasing age (median age 67 v 62 years), presence of previous chronic ill health (40% v 22%), higher acute physiology score (mean 10.1 v 8.0), and a greater number of organs in failure for a longer duration (44% v 24%), contributed to the positive association with death of the overall score.  Consequently, these variables were used initially to develop the model.

 

Forward stepwise multivariate analyses were performed on each of the 20 modelling data sets. The ‘best’ model was that which demonstrated the best ‘fit’ and had the largest area under the ROC curve. This model selected the following five variables - patient’s age, chronic health points, acute physiology score (APS) at ICU discharge, ICU length of stay and whether or not the patient had had cardiothoracic surgery, for inclusion in the model.  These same five variables were selected in 15 out of the 20 analyses.  A cut-off of 0.6 had the best sensitivity and specificity – 65.5% and 87.6%, respectively.  Table 2 gives details of the final model and Figure 1 shows its ROC curve.

The model was validated using two other data sets.  As the results of the Guy’s Validation and the RIPUG Validation data sets were similar, the merged results are presented (Table 3).  The sensitivity and specificity were 74.3%% and 71.1%% respectively; and the area under the ROC curve was 0.80 (95% CI 0.79-0.82).  The area under the ROC curve ranged from 0.68 - 0.87 for the 20 individual ICUs. Patients identified as at  risk by the model had a mortality of  25% while the mortality of those not at risk was 4% giving a relative risk of 5.61 (95% CI 4.89-6.44).

 

Evaluation of the ability of the model to alter outcome of patients identified as at risk  demonstrated significant differences in post ICU discharge mortality between Groups 0, 1 and 2 (Table 4).  In the Development Data Set, 14% of patients discharged from the ICU with a risk of ward death >=0.6 died on the ward.  Among the patients who equalled or exceeded the threshold and stayed an additional 48 hours in the ICU, during which time the predictive threshold of the model fell below 0.6, only 4% died on the ward.  This reduction in mortality was statistically significant (p=0.034).  In the Validation Data Set there was a reduction in mortality from 28% to 17% among those who stayed another 48 hours (p=0.011).  The relative risk of patients identified as at high risk of dying was reduced from 6.76 to 3.46 if these patients were to spend another 48 hours in the ICU before their discharge.

Potential impact on the provision of Intensive Care Beds in the United Kingdom

The Validation Data Set was used to estimate the impact on the provision of intensive care resources if the model was used to reduce the post ICU discharge mortality. There were 8,449 patients who stayed in the ICU for a total of 34,588 days.  They had an overall post ICU discharge mortality of 11.3%.  2,875 patients were predicted as likely to die on the ward on the day of discharge (had a triage score >=0.6) and had a post ICU discharge mortality of 25%.  Assuming that our model is valid, the post ICU mortality could be reduced by nearly 40% if these patients stayed another 2 days before discharge.  This would have required a total of 5,750 additional ICU bed days or the provision of fully staffed ICU bed days would have to be increased by 17%.


Discussion

A significant number of patients are dying on the wards following ICU discharge.  An increase in mortality after discharge from ICU of between 9% to 27% have been reported 16,  17.  Using discharges at night as a proxy measure of inappropriate early discharge, Goldfrad and Rowan 8 were able to demonstrate an increase in post ICU discharge mortality among these patients from 11.9% to 15.4%. A recent study in Newcastle 9 analysed patients’ TISS scores on the day of ICU discharge.  They found that patients who were discharged with a TISS score of 20 or greater had a ward mortality rate of 21.4% compared with of only 3.7% for those patients discharged with a TISS score of less than 10.  Increasing age, male sex and admission APS were associated with death post ICU discharge.  TISS reflects the intensity of treatment and may vary between ICUs depending on the treatment philosophy of the units.  Our discharge triage model used objective data (age, presence of end stage disease, the degree of physiology derangement, the length of stay and the presence or absence of cardiac surgery) in a logistic regression equation to identify patients at risk from inappropriate early discharge. We were able to do this as the Riyadh ICU Program database captures daily data throughout a patient’s stay in the ICU.  Among patients in the Model Development data set, patients identified as at risk had a mortality of 14% compared to a mortality of only 1.5% among those not at risk. This is despite the finding that the post ICU discharge mortality at Guy’s is only 3.7% which is relatively low, due probably to the high proportion of ‘low risk’ cardiothoracic patients within our particular patient cohort and relatively better provision of ICU beds.  The post ICU discharge mortality of the patients identified to be a risk in the Validation data set was 25% compared to a mortality of 4% for patients not at risk; a relative risk of 5.61 (4.89 – 6.44 95% C.I.).   

 

Identifying patients at risk is of no use if their outcome cannot be altered.  By modelling a ‘what if’ situation whereby patients at risk and discharged on the same day were compared with patients who stayed for another 24 to 48 hours, it was possible to demonstrate a reduction of relative risk from 6.76 (4.78-9.56 95% C.I.) to 3.46 (2.21-5.41 95% C.I.).  However, the finding of this modelling exercise will need to be confirmed by a prospective study.

. 

If the results reported in this study can be confirmed in a prospective study, it will have a major impact on the provision of ICU beds in the United Kingdom.  We have estimated that there needs to be an increase by 16% in the number of ICU beds in order to reduce the number of post ICU discharge deaths by 39%.  This increase is probably an under estimate of the overall requirement, as our model only accounts for deaths as a result of being discharged too early.  We have not looked into the effects of delayed admission to the ICU because of the lack of beds nor have we estimated the number of extra ICU beds required in order not to have to cancel major surgery.

Development of new technologies and treatments together with an ageing population now mean that we are treating patients who previously would not have survived or in whom ICU admission would not have been deemed appropriate.  The UK already has limited resources allocated for the provision of intensive care facilities compared to many of its European counterparts 18, and regional differences in the number of available ICU beds have been demonstrated 19.  Although the overall number of ICU and high-dependency beds has increased over the past 10 years there has been a concurrent rise in hospital activity 20, 21.   A report by the Audit Commission in 1999 found that up to 25% (with a median value of 5%) of patients were still being discharged prematurely to allow more seriously ill patients to be admitted 22.  Neither our discharge triage model nor discharge guidelines published by the Department of Health 23, which deal with the process of care will have any impact until and unless the major structural problem of a shortfall of at least 16% in intensive care bed provision which we have identified, is corrected. 

 

The implications of this study on intensive care resources are tremendous.  Our model and the results of evaluation are strongly suggestive that post-ICU discharge mortality can be reduced.  However, the results of the validation were not based on use of the model in real time.  It is therefore important that before any strategic decisions are made or the model is used in clinical practice a prospective study to further evaluate the model is carried out.  We propose that in such a study, patients who equal or exceed the threshold (0.6) on the day they are considered for discharge from the ICU are randomised into two groups; a) those who are discharged on the same day and b) those who are retained in the ICU for another 48 hours or until their probability of post ICU discharge death is less than 0.6. Additional resources will be required during the course of such a study. We have estimated that to do the study on a 13 bedded ICU will require funding for an additional 2 ICU beds for a period of just under 3 years.

 


References

1.      Rowan KM., Kerr JH., Major E., McPherson K., Short A., Vessey MP.  Intensive Care Society’s APACHE II study in Britain and Ireland-1: Variations in case mix of adult admissions to general intensive care units and impact on outcome. BMJ 1993; 307: 972-976

2.      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: 977-981

3.      Bion J. Rationing intensive care. BMJ 1995; 310: 682-683

4.      Ryan DW. Providing intensive care. BMJ 1996; 312: 654

5.      Wallis CB., Davies HTO., Shearer AJ. Why do patients die on general wards after discharge from intensive care units?  Anaesthesia 1997; 52: 9-14

6.      Dragstead L., Qvist J. Outcome from intensive care. A five-year study of 1308 patients: underlying causes of death. European Journal of Anaesthesiology 1990; 7: 159-68

7.      Ridley S., Purdie J. Cause of death after critical illness. Anaesthesia 1992; 47: 116-119

8.      Goldfrad C, Rowan K.  Consequences of discharges from intensive care at night. Lancet 2000; 355: 1138-1142.

9.      Smith L., Orts CM., O’Neil I., Batchelor AM., Gascoigne AD., Baudouin SV. TISS and mortality after discharge from intensive care. Intensive Care Medicine 1999; 25: 1061-1065

10.  Daly K., Bihari D. Multiple Organ Failure in the critically ill. Outcomes and costs -a six month follow- up study. Intensive Care Medicine (1995); 21(suppl.1): S81.

11.  Knaus  WA,  Draper EA,  Wagner WP, Zimmerman JE. APACHE II: A severity of   disease classification system. Criti Care Med 1985; 13: 818- 829

12.  Chang RWS, Jacobs S, Lee B. Predicting outcome intensive care unit patients using computerised trend analysis of daily APACHE II scores corrected for organ system failure. Intensive Care Medicine 1988; 14: 558 – 566

13.  Cullen DJ, Keene R, Waternaux C, Kunsman JM, Caldera DL, Peterson H. Results, charges and benefits of intensive care for critically ill patients: Update Crit Care Med 1984; 12 (2): 102-106

14.  Lemeshow S , Hosmer DW. A review of goodness of fit statistics for use in the development of logistic regression models. Am J Epidemiol  1982; 115: 92-106.

15.   Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operator characteristic (ROC) curve. Radiology 1982; 143: 29-36.

16.   Munn J, Willatts SM, Tooley MA. Health and activity after intensive care.         Anaesthesia 1995; 50: 1017-1021

17.   Goldhill DR, Sumner A. Outcome of intensive care patients in a group of British Intensive care units. Crit Care Med 1998; 26: 1337-1345

18.   Vincent JL, Suter P, Bihari D, Bruining H. Organisation of intensive care units in Europe: lessons from the EPIC study. Intensive Care Medicine 1997; 23: 1181- 1184

19.   Metcalf MA, McPherson K. Study of intensive care in England. London: Department of Heath.1995

20.   Ridley SA, Burchett K, Burns A, Gunning K. A comparison of hospital and critical-care activity. Anaesthesia 1999; 54: 521-528

21.   Hensher M, Edwards N, Stokes R. International trends in the provision and   utilisation of hospital care. BMJ 1999; 319:845-848 

22.  Audit Commission. Critical to Success: The place of efficient and effective critical care services within the acute hospital. Audit Commission October 1999

23.   Working Group. Guidelines on admission to and discharge from intensive care and high dependency care units. London: Department of health 1996


Acknowledgement:  This study was funded by a research grant from the Special Trustees of St Thomas’ Hospital.  We would like to thank members of the Riyadh ICU Program Users Group for access to their database.

 

Competing Interests:

René W S Chang is a Director of Medical Associated Software House Limited.


Table 1. Demographic characteristics and clinical features of the ICU survivors for the three data sets

 

A

Development Data Set

B

 Validation Data Set

A versus B

p value

 

Ward Deaths

n = 200

Hospital

Survivors

n = 5275

p

Ward Deaths

n = 958

Hospital Survivors

n = 7491

p

Ward Deaths

Hospital

Survivors

Age (median years)

67 (31-93)

62 (17-101)

0.0001

72 (16-96)

63 (16-96)

0.0001

0.0001

0.0001

Sex (male %)

65%

71%

ns

56%

61%

0.002

0.027

ns

Patients with Chronic ill Health

40%

22%

0.0001

28%

15%

0.0001

0.001

0.0001

Mean Day 1 APACHE II score

19.2

(18.2-20.2)

13.5

(13.4-13.7)

0.0001

17.4

(17.0-17.9)

12.0

(11.9-12.1)

0.0001

0.001

0.0001

Mean Last APACHE II

score

15.8

(14.9-16.6)

11.9

 (11.8-12.0)

0.0001

15.8

(15.4-16.2)

10.1

(10.0-10.2)

0.0001

ns

0.0001

Mean Day 1 ROD

30.4

(27.3-30.0)

11.2

(10.9-11.6)

0.0001

29.7

(28.4-31.0)

14.2

(13.9-14.6)

0.0001

ns

0.0001

Mean Day 1 APS

13.5

(12.6-14.4)

9.6

(9.5-9.7)

0.0001

11.6

(11.2-12.0)

8.1

(8.0-8.2)

0.0001

0.0001

0.0001

Mean Last APS

10.1

(9.3-10.9)

8.0

(7.9-8.1)

0.0001

9.9

(9.5-10.3)

6.2

(6.1-6.3)

0.0001

ns

0.0001

Mean Day 1 OFS

19.6

(18.6-20.7)

13.7

(13.5-13.8)

0.0001

17.9

(17.4-18.3)

12.1

(11.9-12.2)

0.0001

ns

0.0001

Mean Last OFS

16.1

(15.2-17.0)

12.0

(11.9-12.1)

0.0001

16.2

(15.8-16.6)

10.2

(10.1-10.3)

0.0001

ns

0.0001

Patients with >1 organs in failure at ICU discharge

44%

24%

0.0001

54%

35%

0.0001

ns

0.045

Median day 1

TISS points

34 (4-78)

33 (3-89)

ns

33.5 (2-79)

29 (2-93)

0.0001

ns

0.0001

Median Last

TISS points

28 (4-54)

31 (2-79)

0.0001

29 (2-75)

25 (2-79)

0.0001

ns

0.0001

Number of cardiothoracic patients (%)

15%

59%

0.0001

0.8%

7%

0.0001

0.0001

0.0001

Number of  ventilated patients

Ventilated days ( median)

69%

3 (0-52)

81%

1 (1-260)

0.0001

0.0001

57%

3 (1-87)

48%

2 (1-198)

0.0001

0.003

0.002

0.0001

Number of dialysed patients

Dialysis days (median)

21%

4 (1-34)

5%

4 (1-57)

0.0001

ns

6%

5.5 (1-45)

3%

4 (1-75)

0.0001

0.0001

0.0001

0.0001

Length of ICU stay (median days)

3 (1-64)

1 (1-283)

0.0001

3 (2-112)

2 (1-219)

0.0001

ns

0.0001

Length of hospital stay

(median days)

10 (0-303)

7 (0-677)

0.0001

7 (0-256)

9 (0-281)

ns

0.0001

ns

Median values with ranges. Mean values with 95% confidence intervals.

Supplementary Table 1. The Sensitivity specificity (using a 0.5, 0.6 and 0.7 cut-offs), ROC curve analysis and Hosmer- Lemeshow statistic produced from each of the 20 data sub-sets

 

Sensitivity

 (number of deaths correct out of 200 patients)

Specificity

(number of survivors correct out of 250 patients)

roc (%)

(95% CI’s)

Chi SquareD

Significance

 

0.5

0.6

0.7

0.5

0.6

0.7

Model 1

70.5%  (141

57.0%  (114)  

44.5%  (89)

80. 0%  (200)

86.4%   (216)

92.4%  (231)

83.0%

3.8514

0.8703

Model 2

71.0%  (142)

60.5%  (121)

48.0%  (96)

79.2%   (198)

87.2%   (218)

93.6%  (234)

83.7%

4.8551

0.7729

Model 3*

69.0%  (138)

54.0%  (108)

33.5%  (67)

71.2%   (178)

82.8%   (207)

92.0%  (230)

79.5%

13.3017

0.1019

Model 4

76.0%  (152)

62.0%  (124)

51.0%  (102)

80.4%   (201)

87.2%   (218)

92.4%  (231)

84.7%

4.2161

0.8371

Model 5

74.0%  (148)

65.0%  (130)

49.5%  (99)

81.2%   (203)

87.2%   (218)

92.0%   (230)

84.8%

5.6835

0.6826

Model 6*

72.0%  (144)

60.5%  (121)

48.5%  (97)

80.0%   (200)

88.0%   (220)

92.8%   (232)

83.9%

7.2631

0.5085

Model 7

72.5%  (145)

61.0%  (122)

48.5%  (97)

82.0%   (205)

86.8%   (217)

91.6%   (229)

84.0%

7.7146

0.4618

Model 8

73.0%  (146)

61.5%  (123)

50.5%  (101)

83.6%   (209)

88.0%   (220)

92.8%   (232)

85.2%

6.8178

0.5564

Model 9*

73.0%  (146)

61.5%  (123)

42.5%  (85)

78.4%   (196)

84.0%   (210)

91.6%   (229)

82.0%

4.9558

0.7623

Model 10

73.0%  (146)

64.0%  (128)

55.0%  (110)

83.2%   (208)

87.6%   (219)

94.0%   (235)

82.0%

6.5971

0.5807

Model 11

76.5%  (153)

66.0%  (132)

54.0%  (108)

82.4%   (206)

85.6%   (214)

91.2%   (228)

85.2%

11.0842

0.1970

Model 12

70.5%  (141)

57.0%  (114)

44.5%  (89)

80.0%   (200)

86.4%   (216)

92.4%   (231)

82.3%

3.8514

0.8703

Model 13*

73.0%  (146)

62.5%  (125)

44.0%  (88)

78.4%   (196)

85.6%   (214)

92.8%   (232)

82.5%

6.0224

0.6447

Model 14

75.5%  (151)

61.0%  (122)

51.5%  (103)

80.4%   (201)

86.4%   (216)

91.6%   (229)

84.9%

4.2283

0.8360

Model 15

72.0%  (144)

58.5%  (117)

47.0%  (94)

78.4%   (196)

86.8%   (217)

92.4%   (231)

82.7%

4.0832

0.8493

Model 16

72.0%  (144)

57.0%  (114)

46.5%  (93)

80.4%   (201)

86.4%   (216)

92.4%   (231)

82.8%

2.7520

0.9489

Model 17

76.5%  (153)

65.5%  (131)

56.0%  (112)

80.8%   (202)

88.0%   (220)

92.0%   (230)

82.8%

8.9501

0.3465

Model 18

73.0%  (146)

65.5%  (130)

52.0%  (104)

82.4%   (206)

87.6%   (219)

93.6%   (234)

85.6%

5.9186

0.65637

Model 19

70.5%  (141)

58.0%  (116)

46.0%  (92)

78.0%   (195)

84.4%   (211)

92.0%   (230)

82.4%

8.8309

0.3568

Model 20*

76.0%  (152)

65.5%  (131

53.0%  (106)

80.4%   (201)

87.2%   (218)

93.2%   (233)

85.1%

2.5648

0.9586

*Data sub-sets did not include all of the 5 variables selected in each of the other 15 data sub-sets


Table 2.  The five variables selected by forward step-wise multiple logistic regression analysis for the discharge triage predictive model.

Variable

B

R

Significance

Age

0.0532

0.2202

0.0000

Chronic Health Points

0.2501

0.1258

0.0006

Acute Physiology Score

0.1556

0.2005

0.0000

Cardiac Surgery

-2.1084

-0.3074

0.0000

Length of ICU stay

0.0447

0.1032

0.0034

Constant

-4.5821

 

0.0000


Table 3.  Predictive power of Discharge Triage Model

 

Alive

Died

Total

Development Data Set

>=0.6

770 (86%)

130 (14%)

900

<0.6

4505 (98.5%)

70 (1.5%)

4575

Relative risk of dying = 9.44 (7.12 – 12.51 95% CI )

Validation Data Set

>=0.6

2163 (75%)

712 (25%)

2875

<0.6

5328 (96%)

246 (4%)

5574

Relative risk of dying = 5.61 (4.89-6.44 95% CI)


Table 4.  Comparison of post ICU discharge mortality for patients discharged from the ICU on the day of prediction (Group 0), patients who stayed an additional 24 hours (Group 1) and patients who stayed in the ICU for an additional 48 hours (Group 2) and patients not predicted at all (Group 3)  

Alive

Died

Total

Development Data Set

Group 0

326 (86%)

53 (14%)

379 (100%)

Group 1

71 (93%)

5 (7%)

76

Group 2

52 (96%)

2 (4%)

54

Group 0 v Group 1 p=0.077;  Group 0 v Group 2 p=0.034

Relative risk of dying Groups 1 & 2 v Group 0 = 0.385 (0.18-0.826 95% CI)

Validation Data Set

Group 0

581 (72%)

230 (28%)

811

Group 1

126 (87%)

19 (13%)

145

Group 2

86 (83%)

17 (17%)

103

Group 3

776 (96%)

34 (4%)

810

Group 0 v Group 1 p=0.0001; Group 0 v Group 2 p=0.011 

Relative risk of dying Groups 1 & 2 v Group 0 = 0.512 (0.371-0.706  95% CI)

Relative risk of dying Group 0 v Group 3 = 6.76 (95% CI= 4.78 – 9.56)

Relative risk of dying Groups 1 & 2 v Group 3 = 3.46 (95% CI=2.21 – 5.41)

Mortality is reduced by 39.3% if patients stayed another two days


Figure 1. ROC curve of the discharge triage model. 


Figure 2. ROC curve for Validation Data Set






This Article
Right arrow Abstract Freely available
Services
Right arrow Similar articles in this netprints
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Daly, K.
Right arrow Articles by Chang, R. W S
Right arrow Search for Related Content
PubMed
Right arrow Articles by Daly, K.
Right arrow Articles by Chang, R. W S


HOME HELP FEEDBACK BROWSE ARTICLES BROWSE BY AUTHOR