Physiological abnormalities in early warning scores are related to mortality in adult inpatients{dagger}

D. R. Goldhill* and A. F. McNarry

The Anaesthetics Unit, The Royal London Hospital, London E1 1BB, UK

*Corresponding author. Anaesthesia and Critical Care, The Royal National Orthopaedic Hospital Trust, Brockely Hill, Stanmore, Middlesex, HA7 4LP, UK. E-mail: david.goldhill@rnoh.nhs.uk
{dagger}This article is accompanied by Editorial II. Presented to the Anaesthetic Research Society in Glasgow, April 2003.

Accepted for publication: December 17, 2003


    Abstract
 Top
 Abstract
 Introduction
 Methods and results
 Comment
 References
 
Background. Early warning scores using physiological measurements may help identify ward patients who are, or who may become, critically ill. We studied the value of abnormal physiology scores to identify high-risk hospital patients.

Methods. On a single day we recorded the following data from 433 adult non-obstetric inpatients: respiratory rate, heart rate, systolic pressure, temperature, oxygen saturation, level of consciousness, urine output for catheterized patients, age and inspired oxygen. We also noted the care required and given.

Results. Twenty-six patients (6%) died within 30 days. They were significantly older than survivors (P<0.001). Their median hospital stay was 26 days (interquartile range 16–39). Mortality increased with the number of physiological abnormalities (P<0.001), being 0.7% with no abnormalities, 4.4% with one, 9.2% with two and 21.3% with three or more. Patients receiving a lower level of care than desirable also had an increased mortality (P<0.01). Logistic regression modelling identified level of consciousness, heart rate, age, systolic pressure and respiratory rate as important variables in predicting outcome.

Conclusions. Simple physiological observations identify high-risk hospital inpatients. Those who die are often inpatients for days or weeks before death, allowing time for clinicians to intervene and potentially change outcome. Access to critical care beds could decrease mortality.

Br J Anaesth 2004; 92: 882–4

Keywords: complications, abnormal physiology; intensive care, audit, scoring systems; intensive care, outcome; intensive care outreach; mortality, inpatient


    Introduction
 Top
 Abstract
 Introduction
 Methods and results
 Comment
 References
 
Recording a patient’s physiological observations is part of the daily ward routine. Ashworth1 found that abnormal values identify patients with greater mortality risk. Patients with abnormal values had a 90-day mortality of 20%, compared with 1.6% overall. Early warning scores are used to identify patients who are, or who may become, critically ill. They award an increasing number of points to worsening physiological values until a trigger score is reached. Although these are being introduced throughout the UK and their use has been encouraged by several bodies,24 the scores have yet to be scientifically validated. We studied the relevance of abnormal physiological values for identifying high-risk hospital inpatients.


    Methods and results
 Top
 Abstract
 Introduction
 Methods and results
 Comment
 References
 
Ethics committee approval was obtained. The week before the study, senior nursing staff on all wards were briefed and encouraged to ensure that comprehensive charting took place on the study day. Between 08:30 and 18:00 on December 17th, 2002 every adult non-obstetric bed area in the Royal London Hospital was visited. If a bed was unoccupied, one return visit was made before the bed area was excluded from the investigation. Physiological variables recorded were: respiratory rate, heart rate (HR), systolic pressure, temperature, oxygen saturation (SpO2), level of consciousness (LOC; alert/confused/responds to voice/responds to pain/unresponsive), supplemental oxygen (FIO2) (air/<=40%/>40%), weight, and urine output for catheterized patients. The level of the patient’s critical care needs and location were classified using definitions provided by the Intensive Care Society5 (level 0: normal ward care in an acute hospital; level 1: care in an acute ward with additional support; level 2: care for a single organ failure or step down from a higher level of care; level 3: advanced respiratory support or support of two organ systems and including all complex patients with multi-organ failure).

If recorded within 8 h of our visit, we used the most recent values on the patient’s observation chart. If not, we made new observations ourselves. Outcome at 30 days (discharged alive/died in hospital/inpatient) was retrieved from the hospital records system. We used the normal ranges defined by our intensive care outreach service (called the Patient At Risk Team in our hospital): respiratory rate 10–19 bpm; HR 50–99 beats min–1; systolic pressure 100–179 mm Hg; temperature 36.0–37.4°C; SpO2 >=95%; LOC alert; urine output (catheterized patients) 0.5–3 ml kg–1 min–1.

Statistical analysis was with SPSS v10. For the backward stepwise logistic regression (likelihood ratio) model, all variables except age were treated as normal or abnormal (binary variables); age was treated as a continuous variable. The odds ratios thus provided relate to the increased risk of mortality if the variable is abnormal (using the definitions listed) except for age where the odds ratio is the risk of mortality caused by being 1 yr older. At each step in the regression, for a variable to be removed from the model it had to have a P value greater than 0.1.

We surveyed 548 beds of which 98 were unoccupied on two visits. We excluded 13 intensive care unit (ICU) patients, three patients known to be ‘not for resuscitation’ and one duplicate observation caused by a transfer, leaving 433 data sets for analysis. Values are given as mean (SD) or median (interquartile range).

The 26 patients who died were older than the survivors (P<0.001, unpaired t-test), mean age 73 (range 38–91) yr vs 60 (18–97) yr. Their median hospital stay was 26 (16–39) days, with death at a median of 10.5 (4–21) days after the study. The only patient with no abnormalities who died did so 21 days after the study. Mortality increased significantly with the number of abnormalities (P<0.001, logistic regression; explanatory variable: total number of abnormalities) (Table 1). Respiratory rate and HR were the variables most frequently abnormal (abnormal in 54% and 13%, respectively, of cases with one abnormality and in 96% and 63%, respectively, of cases with at least three abnormalities).


View this table:
[in this window]
[in a new window]
 
Table 1 Thirty-day outcome, location and days in hospital before and after the study according to the number of abnormalities recorded. CI, confidence interval; IQR, interquartile range; *P=0.087; {dagger}P=0.014, {ddagger}P=0.001 (by logistic regression); apatients recorded as died in hospital within 30 days of study; bincreased risk of mortality when compared with no abnormalities; clocation capable of delivering levels of care (patients already in level 3 were excluded); dbefore study = days in hospital before study; after study = days in hospital until death, discharge or end of study period (30 days)
 
In a separate analysis, the seven initial variables (respiratory rate, HR, systolic pressure, temperature, SpO2, LOC, age) were analysed by backward stepwise logistic regression. The odds ratio for mortality for the five most significant variables were, in decreasing level of significance: LOC 4.63 (95% confidence intervals 1.79–12.00; P<0.05), HR 3.86 (1.49–9.97; P<0.05), age 1.04 (1.01–1.07; P<0.05), systolic pressure 3.16 (1.12–8.90; P<0.05) and rerspiratory rate 2.49 (0.92–6.72).

This model had a sensitivity of 7.7%, specificity of 99.8% and a positive predictive value of 66.7%. The sensitivity and specificity of the model was calculated using a classification cut-off of 0.5; that is, if the probability of death obtained from the model was greater than 0.5 the subject was classified as having died, conversely if the probability of death obtained from the model was less than 0.5 the subject was classified as a survivor.

Urine output was only recorded for patients with catheters (n=52, 12%) and was not included in the logistic regression model. Oxygen therapy was given to 39 patients (9%). In these two groups (catheterized or receiving oxygen) 30-day mortality was 19.2% and 30%, respectively.

Estimated levels of care were available for 384 (88.7%) of all patients, which included 23 (88.5%) of the 26 who died. The 34 patients thought by data collectors to be receiving a lower level care than desirable had a greater mortality (20.6%, P<0.01, Fisher’s exact test) compared with the 349 patients (mortality 5.4%) judged to be receiving appropriate care.

Forty-three (9.9%) of the patients in the study were seen at some point during their hospital stay by the intensive care outreach team. The team reviewed six of the patients who died. In three of these patients, increased treatment was considered inappropriate, one was admitted to ICU and two were discharged from outreach follow-up before the study day. Only two of the other 20 patients who died were transferred to the ICU.


    Comment
 Top
 Abstract
 Introduction
 Methods and results
 Comment
 References
 
The results are striking, given the limitations of data accuracy and completeness. Patients were in different phases of their illness, some recovering, some still to deteriorate. Some beds were empty, waiting for new admissions. We do not have data on patients absent from the wards, some of whom would have been in the operating theatre, the endoscopy suite or the radiology department. The physiological values were single unvalidated measurements, usually taken by ward staff at some time during the study day. These values may have been subject to measurement and recording errors and were not always collected at a single time. The proximity of the study to Christmas may have affected admission and discharge decisions. The results depend directly on our definition of physiological normality. Different definitions could give a different set of results. Our definitions were taken from an early warning score that is used to identify patients who may develop critical illness. As such, it is not intended to predict mortality, and this may be shown by the low sensitivity and high specificity we report. These measures would also be affected by different definitions of normality. This may also explain the difference in the magnitude of our results when compared with those of Ashworth.1 Our definition of 30-day mortality may also have influenced the results as we assumed those discharged alive from hospital did not die later within the 30-day study period.

Experience suggests that respiratory rate is an important indicator of an at-risk patient.6 Although ventilatory frequency was the most common abnormality we found, it did not make an independent statistically significant contribution to the logistic regression model. There may be several explanations for this. These include the limitations of the study, with outcome sometimes days or weeks after data collection, or that an abnormal respiratory rate may co-exist with other abnormalities that make a greater contribution to the model. Large-scale prospective studies are necessary to determine the physiological variables and values that identify high-risk patients.

The data collectors were not confident enough to assess the appropriate ‘level of care’ requirement for some of the patients. Considering the 89% of patients for whom this information is available, those cared for at a lower level than ideal had an increased mortality. We have found that the longer patients are in hospital before they are admitted to ICU, the greater their mortality.7 About 25% of admissions to ICU from the ward occur after the patient has deteriorated to the point of cardiorespiratory arrest.8 Patients at high risk are present on the wards and their condition may deteriorate during their hospital admission. Early intervention may be beneficial and this should include assessment for critical care.

We excluded three patients known, on the study day, to be ‘not for further resuscitation’. We may not have known of other patients for whom treatment was limited. Three patients who died were assessed by our outreach service at some time during their hospital stay and an increase in treatment was considered inappropriate. The majority (87%) of patients with three or more abnormalities were in level-0 beds. If only half of these patients could have benefited from critical care, we would need twice the number of critical care beds in the hospital to accommodate them.

We found an association between easily recordable physiological derangements and mortality. Most patients with physiological abnormalities who died were in hospital for many days. This suggests that an early warning score could identify some patients early enough to allow interventions to take place in an appropriate location. Therefore, the opportunity exists to intervene and improve outcome for high-risk ward patients.


    Acknowledgements
 
We are grateful to the following colleagues who assisted in data collection and entry: Fultara Begum, Caroline Cooper, Louise Crosby, Guy Jackson, Ann McGinley, Gerlinde Mandersloot, Simon Nourse, Huw Owen Reece, Penny Robinson, and Stuart Withington. Statistical advice was provided by the Statistical Services Unit, University of Sheffield.


    References
 Top
 Abstract
 Introduction
 Methods and results
 Comment
 References
 
1 Ashworth S. A prelude to outreach: prevalence & mortality of ward patients with abnormal vital signs. In: Proceedings of the 15th Annual Congress of the ESICM. Intensive Care Med 2002; 28 Suppl 1: S21

2 Royal College of Physicians of London.The Interface Between Acute General Medicine and Critical Care. Report of a working party of the Royal College of Physicians. London: Royal College of Physicians of London, 2002

3 Intensive Care Society. Guidelines for the Introduction of Outreach Services. Intensive Care Society Standards. London: Intensive Care Society, 2002

4 Department of Health. Comprehensive Critical Care: a Review of Adult Critical Care Services. London: DoH, 2000

5 Intensive Care Society. Levels of Critical Care for Adult Patients. Intensive Care Society Standards. London: The Intensive Care Society, 2002

6 Goldhill DR, Worthington L, Mulcahy A, Tarling M, Sumner A. The patient-at-risk team: identifying and managing seriously ill ward patients. Anaesthesia 1999; 54: 853–60[CrossRef][ISI][Medline]

7 Goldhill DR, McNarry A. The longer the patient is in hospital before ICU admission the higher the mortality. Br J Anaesth 2002; 89: 356–7P

8 Goldhill DR, Sumner A. Outcome of intensive care patients in a group of British intensive care units. Crit Care Med 1998; 26: 1337–45[ISI][Medline]