Rates of in-hospital arrests, deaths and intensive care admissions: the effect of a medical emergency team

One approach to providing an early response to at-risk inpatients throughout the hospital is the medical emergency team (MET),^9,10 which replaces the conventional cardiac arrest team. The MET responds to specific clinical criteria (such as bradycardia, tachycardia, hypotension and threatened airway) in order to prevent further deterioration. Others have advocated similar approaches to reduce unexpected hospital deaths and morbidity.^11,12

In our study, selected outcomes in a hospital with a MET were compared with outcomes in two hospitals with conventional cardiac arrest teams. These outcomes were rates of cardiorespiratory arrest, unanticipated admission to the ICU or high dependency unit (HDU), death, and deaths where there was no prexisting "do not resuscitate" (DNR) order.

The study was approved by the ethics committees of each participating hospital and the University of New South Wales.

At Hospital 1, the cardiac arrest team was replaced by a MET, which any staff member could call for immediate assistance. Staff could summon the MET if concerned about a patient's condition or if the patient's vital signs exceeded certain levels (Box 1).¹⁰ An education program explained the MET's role to all new staff. However, calling the MET when criteria were met was not compulsory. The MET consisted of the ICU registrar and senior nurse, and medical registrar.

At Hospitals 2 and 3, the arrest team was paged by nursing or medical staff for cardiorespiratory arrest. The arrest team consisted of the ICU registrar, medical registrar, and ICU or coronary care nurse.

Soon after an event, the patient's medical record was reviewed for demographic information. In addition, for cardiac arrests and deaths, documentation of a DNR order before arrest or death was recorded. Each ICU/HDU admission was classified as to whether the patient was admitted to ICU/HDU for the same reason he or she was admitted to hospital. If not, the ICU/HDU admission was defined as unanticipated. For example, a patient admitted to ICU with respiratory distress after a cholecystectomy would be unanticipated.

Data were collected by three critical care nurses (one at each hospital) trained in the use of a specifically designed form, which was piloted for two weeks before data collection. The nurses were familiar with the medical record at the three hospitals. Where the information contained in the history was unclear, the attending staff were asked for clarification.

Data were entered into a database.¹³ Data cleaning was performed and any anomalies checked by reference to the datasheet or medical record. Data were then exported to SAS¹⁴ for analysis.

For any patient, one event could result in additional events (eg, cardiac arrest followed by unanticipated ICU admission and then death). The "index event" was defined as the event the data collectors considered the first in a series of events. The casemix-adjusted rates of index events were compared between the three hospitals as secondary endpoints.

To ensure that any decline in the rate of unanticipated admissions was not caused by excess anticipated admissions, the casemix-adjusted rate of all ICU/HDU admissions was calculated as a control measure.

Using simple and multiple logistic regression, we modelled the probability of an event occurring during hospitalisation, adjusted for patient demographics and diagnostic characteristics. Models were derived independently for each total event and for each index event. Parameters were added to the model in a stepwise fashion.

To prevent overparameterising the models (where minor, non-significant differences cumulatively hide true differences), when C (equivalent to the area under the receiver operator characteristics curve) reached 0.85 no further parameters were added to the model. This always occurred with fewer than six parameters used. Demographic and casemix independent variables that were tested for use in the models are detailed in Box 2.

The models developed used groups of diagnostic categories based on ICD-9-CM codes using the principal diagnosis and the stay diagnosis only.¹⁵ The ICD-9-CM code groupings used are available from the principal author (PJB). The performance of the models was assessed by Hosmer-Lemeshow goodness-of-fit tests.¹⁶

The risk of an event occurring in a hospital compared with the MET hospital was presented as an adjusted odds ratio with 95% confidence intervals. A level of significance of 5% was used in all statistical tests.

The rates of DNR orders in dying patients were 77% in Hospital 1, and 64% and 70% in Hospitals 2 and 3, respectively (P = 0.006).

There was a significantly reduced rate of unanticipated ICU/HDU admissions at the MET intervention hospital after casemix adjustment (for both the total event rate and the index rate). After adjustment, Hospital 2 had 49 (95% CI, 20-87) more unanticipated ICU/HDU admissions over a six-month period, and Hospital 3 had 92 (95% CI, 47-146) more, compared with Hospital 1. The rate of all ICU/HDU admissions was lower at Hospital 1 than at one control hospital, and trended to lower than at the other.

There was no statistically significant difference in cardiac arrest rate or death rate after casemix adjustment. The casemix-adjusted death rate in patients where there was no documentation found of a DNR order was significantly higher at Hospital 2, translating to 27 (95% CI, 7-53) extra non-DNR deaths.

In the cardiac arrest models, the variables that were adjusted for were emergency admissions, age over 74, heart disease, lung disease and infectious disease as diagnoses.

In the total death models, the terms adjusted for were emergency admissions, age over 74, single day stay emergency admissions, and cancer and infectious disease. The same demographic variables were used in the index death model, although in the total non-DNR death model both age ≥ 75 and age 65-74 were used in the model.

In the unanticipated ICU/HDU models, the variables adjusted for were single day admission, emergency admission, and cancer and gastrointestinal disease diagnoses. The total unanticipated ICU/HDU model also included the variables stroke and infectious disease as diagnoses.

All models satisfied the Hosmer- Lemeshow test.¹⁶

The models we used appear to adequately fit the data, according to the Hosmer-Lemeshow goodness-of-fit tests, and with good model performance measured by C statistics. However, multiple methods of casemix adjustment are possible, and these may give divergent results. This is a limitation of casemix adjustment methodology.²⁰

To avoid concealing real differences by excessive modelling, parameters were added stepwise by multivariate analysis until the models reasonably represented the data. The terms which appeared in the final models were usually those that could be expected to influence the outcomes. Thus, advanced age and emergency admissions were factors in the death and cardiac arrest models. The cardiac arrest models also included the terms for heart disease, lung and infectious disease. Infectious disease was an unexpected variable and was also significant in the total death model.

Other differences (such as levels of hospital funding, ICU/HDU capacity, the number and seniority of medical and nursing staff, and the level of out-of-hours cover) may also have contributed to the results. However, to adjust for these would have been more difficult than for the variables studied, which relate directly to the patients at risk and are easily and reliably obtained.

The reduction in unanticipated ICU/HDU admissions that was seen in the MET intervention hospital could result from many factors. One possible explanation is that the MET was effective and able to intervene on the wards and prevent further deterioration. Another possible reason may relate to differences in referral practices: perhaps the presence of MET backup engendered a feeling that ICU/HDU referral was not needed. Misclassification of ICU/HDU admissions as anticipated rather than unanticipated was excluded as an explanation of the difference by the finding that the rate of all ICU/HDU admissions was lower at the intervention hospital than one control hospital and trended to lower at the other.

The lack of efficacy of the MET to prevent cardiorespiratory arrest and modify death rate may be related to lack of sensitivity of calling criteria, or because pathophysiological processes (eg, shock) become irreversible. Another possible explanation for the lack of effect of the MET on event rates is underutilisation. Based on a previous study,²¹ up to 706 MET calls could have been expected, and yet only 150 were made. Frequent education is probably also required to ensure the appropriate calling of a MET.²² No special efforts regarding staff education in the study period were made. The clinical staff of the hospital were unaware of the study, to negate any possible Hawthorne effect.²³

Finally, organisational changes such as introduction of a MET are difficult to implement in hospitals.^24,25 Our results probably reflect the effectiveness of the implementation of the MET system as much as the concept of early intervention.

Is the benefit observed useful and worth pursuing? If it is possible to reduce unanticipated ICU admissions without increased mortality this may result in cost saving. It has been estimated that the US spends about 1% of its gross national product on intensive care facilities.²⁶ However, any savings in intensive care would be offset by the cost of establishing and maintaining a MET. The MET may also have unexpected costs and benefits on processes such as staff satisfaction with care provided. Again, these need to be quantified.

We believe that the MET concept should be studied further in a larger sample of institutions.

1. Brennan TA, Leape LL, Laird N, et al. Incidence of adverse events and negligence in hospitalised patients: results of the Harvard Medical Practice Study I. N Engl J Med 1991; 324: 370-376.
2. Leape LL, Brennan TA, Laird N, et al. Nature of adverse events in hospitalised patients: results of the Harvard Medical Practice Study II. N Engl J Med 1991; 324: 377-384.
3. Wilson R McL, Runciman WB, Gibberd RW, et al. The Quality in Australian Health Care Study. Med J Aust 1995; 163: 458-471.
4. Schein RMH, Hazday N, Pena M, et al. Clinical antecedents to inhospital cardiopulmonary arrest. Chest 1990; 98: 1388-1392.
5. Franklin C, Mathew J. Developing strategies to prevent inhospital cardiac arrest: analyzing responses of physicians and nurses in the hours before the event. Crit Care Med 1994; 22: 246-247.
6. Lundberg JS, Perl TM, Wiblen T, et al. Septic shock: an analysis of outcomes for patients with onset on hospital wards versus intensive care units. Crit Care Med 1998; 26: 1020-1024.
7. Goldhill DR, Sumner A. Outcome of intensive care patients in a group of British intensive care units. Crit Care Med 1998; 26: 1337-1345.
8. McQuillan P, Pilkington S, Allan A, et al. Confidential inquiry into quality of care before admission to intensive care. BMJ 1998; 316: 1853-1858.
9. Lee A, Bishop G, Hillman KM, Daffurn K. The medical emergency team. Anaesth Intensive Care 1995; 23: 183-186.
10. Hourihan F, Bishop G, Hillman KM, et al. The medical emergency team: a new strategy to identify and intervene in high risk patients. Clin Intensive Care 1995; 6: 269-272.
11. Frank ED. A shock team in a general hospital. Anesth Analg 1967; 46: 740-745.
12. Goldhill DR. Introducing the postoperative care team [editorial]. BMJ 1997; 314: 389.
13. Microsoft Access [computer program]. Version 2.0. Redmond, Wa: Microsoft, 1994.
14. SAS for Windows [computer program]. Version 6.12. Cary, NC: SAS Institute Inc, 1997.
15. Stremple JF, Bross DS, Davis CL, McDonald GO. Comparison of postoperative mortality and morbidity in VA and nonfederal hospitals. J Surg Res 1994; 56: 405-416.
16. Hosmer DW, Lemeshow S. Applied logistic regression. New York: John Wiley and Sons, 1989.
17. Iezzoni LI. The risks of risk adjustment. JAMA 1997; 278: 1600-1607.
18. Dubois RW, Rogers WH, Moxley JH, et al. Hospital inpatient mortality. Is it a predictor of quality? N Engl J Med 1987; 317: 1674-1680.
19. Green J, Passman LJ, Wintfield N. Analyzing hospital mortality. The consequences of diversity in patient mix. JAMA 1991; 265: 1849-1853.
20. Iezzoni LI, Shwartz M, Ash A, et al. Severity measurement methods and judging hospital death rates for pneumonia. Med Care 1996; 34: 11-28.
21. Hillman KM, Bishop G, Lee A, et al. Identifying the general ward patient at high risk of cardiac arrest. Clin Int Care 1996; 7: 242-243.
22. Daffurn KD, Lee A, Hillman KM, et al. Do nurses know when to summon emergency assistance? Intensive Crit Care Nurs 1994; 10: 115-120.
23. Grufferman S. Complexity and the Hawthorne effect in community trials [editorial]. Epidemiology 1999; 10: 209-210.
24. Garside P. Organisational context for quality: lessons from the fields of organisational development and change management. Qual Health Care 1998; 7 Suppl: S8-15.
25. Koeck C. Time for organisational development in healthcare organisations [editorial]. BMJ 1998; 317: 1267-1268.
26. Cerra FB. Healthcare reform: the role of coordinated critical care. Crit Care Med 1993; 21: 457-464.

Department of Intensive Care, St George Hospital, Sydney, NSW.
Theresa C Jacques, MB BS, FFICANZCA, Director.

Department of Intensive Care, Illawarra Regional Hospital, Wollongong, NSW.
E Grant Simmons, MB BS, FFICANZCA, Director.

Reprints will not be available from the authors.
Correspondence: Dr P J Bristow, Intensive Care Offices, Alfred Hospital, Commercial Road, Prahran, VIC 3181.
p.bristowATalfred.org.au

Make a comment