A physiologically-based early warning score for ward patients: the association between score and outcome*
Preliminary analysis of this data was presented at the Intensive Care Society, London, December 2002.
Summary
We analysed the physiological values and early warning score obtained from 1047 ward patients assessed by an intensive care outreach service. Patients were either referred directly from the wards (n = 245, 23.4%) or were routine critical care follow-ups. Decisions were made to admit 135 patients (12.9%) to a critical care area and limit treatment in another 78 (7.4%). An increasing number of physiological abnormalities was associated with higher hospital mortality (p < 0.0001) ranging from 4.0% with no abnormalities to 51.9% with five or more. An increasing early warning score was associated with more intervention (p < 0.0001) and higher hospital mortality (p < 0.0001). For patients with scores above one (n = 660), decisions to admit to a critical care area or limit treatment were taken in 200 (30.3%). Scores of all physiological variables except temperature contributed to the need for intervention and all variables except temperature and heart rate were associated with hospital mortality.
Patients admitted from the wards to an intensive care unit (ICU) have a much higher mortality than those admitted from the emergency department or the operating theatres [1]. Patients may be on the wards for days or even weeks before admission to an ICU. The longer they are in hospital before ICU admission, the worse their prognosis [2]. Approximately 25% of deaths in patients admitted to an ICU occur after discharge back to a ward and many of these deaths are in relatively low-risk patients [1]. After hospital admission, but before ICU referral, management may be suboptimal and this is associated with an increased mortality [3, 4]. In addition, critical incidents on the wards that result in serious adverse outcomes are common [5]. Early identification of the sickest patients (or those who may become so) may allow earlier intervention, including admission to a critical care unit, thus potentially improving their outcome.
The majority of patients who suffer a cardiorespiratory arrest in hospital have gross physiological abnormalities recorded in the hours preceding the arrest [6–10]. Abnormal physiological measurements, along with a patient's history, examination and investigations, are central to objectively identifying at-risk ward patients [10]. One way of identifying potentially critically ill patients on the wards is through physiologically based early warning scores [11, 12]. These assign an increasing number of points to increasingly deranged physiological values. The early warning scores presently in use have both physiological parameters and scores selected by clinicians based on their experience. For an early warning score to be applicable to a wider group of hospital patients, information is needed as to which physiological variables are most important and at what values. At The Royal London Hospital the intensive care outreach service uses an early warning score, the Patient-At-Risk (PAR) score, to help identify and trigger a response for such patients.
This study had two main objectives. These were to explore the raw physiological data in order to learn more about the importance and contribution of each physiological parameter, and to determine the value of the subjectively derived Patient-At-Risk score as a means of identifying ward patients who could benefit from critical care support.
Methods
Ethics committee approval was obtained for the data analysis presented. Members of the intensive care outreach service routinely reviewed two groups of patients: primary referrals from the wards of any patient causing concern or who triggered an early warning score response, and patients discharged to a ward from a critical care area (ICU, surgical High Dependency Unit (HDU) and the Coronary Care Unit (CCU)).
Patients were reviewed until they were discharged from the outreach service, were admitted to a critical care area, died on the ward or a decision was made that critical care admission would be inappropriate. This was considered an outreach service patient episode and could consist of one or more assessments. Each time a patient was assessed, a Patient-At-Risk score was calculated from seven physiological variables (Table 1).
| Points scored | |||||||
|---|---|---|---|---|---|---|---|
| 3 | 2 | 1 | 0 | 1 | 2 | 3 | |
| Temperature; °C | <35.0 | 35.0–35.9 | 36.0–37.4 | 37.5–38.4 | ≥38.5 | ||
| Heart rate; beats.min−1 | <40 | 40–49 | 50–99 | 100–114 | 115–129 | ≥130 | |
| Systolic blood pressure; mmHg | <70 | 70–79 | 80–99 | 100–179 | ≥180 | ||
| Respiratory rate; breaths.min−1 | <10 | 10–19 | 20–29 | 30–39 | ≥40 | ||
| S pO2; % | <85% | 85–89% | 90–94% | ≥95% | |||
| Level of consciousness | Alert | Confused | Responds to voice | Responds to pain or unresponsive | |||
| Urine output; ml.kg−1.h−1 | nil | <0.5 | dialysis* | 0.5–3 | >3 | ||
- * Dialysis: normally dialysis dependent.
In common with other presently available early warning scores, points are awarded for physiological derangement. Raw physiological values were recorded for temperature, heart rate, arterial systolic blood pressure, respiratory rate, and oxygen saturation (SpO2). Level of consciousness was classified as alert, confused, responds to voice, responds to pain and unresponsive. Urinary output was classified by volume of urine in ml.kg−1.h−1 and whether patients were usually dialysis dependent. Values in the predefined normal range scored zero and a maximum of up to three points was awarded for physiological derangement. The individual points were summed for the total Patient-At-Risk score.
Other information recorded included the patient's sex, age, and any decisions following outreach service involvement. Hospital outcome was retrieved from the hospital's patient administration system. All this information was recorded onto a form and entered into a Microsoft Access (Microsoft Corporation, One Microsoft Way, Richmond, WA) database. Validation of the data consisted of a visual check of the forms and error checking within the database. Data was extracted for all patients seen by the outreach service. If patients had more than one outreach service episode during their hospital stay, only the data from the first episode were analysed.
Each outreach service episode had three possible outcomes: alive remaining on the wards (WARD), transferred to a critical care area (CRITICAL CARE), and either treatment limitation decision that critical care was not appropriate or death while under review by the outreach service (THERAPY LIMIT). The patients' status at hospital discharge was also recorded. Data was grouped by Patient-At-Risk score with scores of seven and eight being amalgamated, as were scores above eight.
In order to understand the association between physiological values and outcome, the values of all parameters, except level of consciousness and urinary output, were ordered from lowest to highest and then divided into groups of 50 patients. The highest and lowest values for this range were noted, as was the hospital mortality. A range of values was identified for each parameter associated with a hospital mortality of < 15%, 15% to < 25%, 25% to < 35% and ≥ 35%. Hospital mortality was also determined for groups of level of consciousness and urinary output defined in the Patient-At-Risk score.
We explored the relationship between hospital mortality and physiological abnormality, defined for each variable as a Patient-At-Risk score of greater than zero. The relationship of the Patient-At-Risk score to decisions made following the outreach episode and hospital outcome was also examined.
Statistical analyses were performed with Statistical Package for the Social Sciences version 11.5 for Windows (SPSS, Chicago, IL) and GraphPad Prism version 4.00 for Windows (GraphPad Software, San Diego, CA) using logistic regression analysis, t-test, Mann–Whitney U-test, Kruskal–Wallis test, Chi-Squared tests and Chi-Squared for trend as appropriate.
Results
The outreach service database contained 1552 outreach service episodes with 2933 Patient-At-Risk scores between 17 August 2001 and 27 January 2003. We excluded 350 episodes in patients who had previous outreach service assessments during the same hospital admission. A further 36 episodes were excluded where the patient was in a critical care area when first seen. Scores, not the physiological values, were recorded in the earliest days of our outreach service and so, after excluding patients with missing values, there were 1047 episodes with complete data for analysis.
Table 2 contains details of the 1047 patients grouped by outcome following completion of the outreach service episode. Primary referrals accounted for 245 (23.4%) patients with the others being follow-up assessments of patients who had been discharged from a critical care area. The primary referrals were in hospital for a median of 4 days (interquartile range 2–11 days) before assessment. A decision was made to admit 81 (33.1%) of the primary referrals to a critical care unit and a treatment limitation decision was made in a further 37 (15.1%). In the critical care discharge group, 54 (6.7%) were re-admitted to a critical care area while under outreach service surveillance and treatment limitation decisions were made in a further 41 (5.1%). Table 3 shows similar data grouped by hospital outcome.
| WARD | CRITICAL CARE | THERAPY LIMIT | ||
|---|---|---|---|---|
| n (% of total no. of patients) | 834 (79.7) | 135 (12.9) | 78 (7.4) | |
| Age; years (SD) | 53.6 (20.0) | 61.1 (19.3) | 72.4 (15.4) | p < 0.001 |
| Male; (%) | 509 (61.0) | 73 (54.1) | 41 (52.6) | NS |
| Receiving oxygen; (%) | 421 (50.5) | 111 (82.2) | 67 (85.9) | p < 0.001 |
| Stay before assessment; days | 6 [3–12] | 5 [2–18] | 8 [3–18] | NS |
| Hospital stay; days | 21 [11–38] | 24 [11–47] | 13 [6–29] | p < 0.001 |
| Primary referral; (%) | 127 (15.2) | 81 (60.0) | 37 (47.4) | p < 0.001 |
| Hospital mortality; (%) | 56 (6.7) | 38 (28.1) | 59 (75.6) | p < 0.001 |
- WARD: remained on ward receiving ongoing care. CRITICAL CARE: patient transferred to a critical care area within the hospital. THERAPY LIMIT: treatment limitation decision that critical care was not appropriate or death while under review by outreach service; Primary referral: ward patient referred to the outreach service because of a trigger score or general clinical concern. The other patients were routine follow-ups after discharge from a critical care area.
- NS: not statistically significant (p > 0.05). The percentage in row one is of the total number of patients (n = 1047). All other percentage are of the number in each column.
| Hospital outcome | |||
|---|---|---|---|
| Alive | Dead | ||
| n (% of total no. of patients) | 894 (85.4) | 153 (14.6) | |
| Age; years (SD) | 53.4 (20.1) | 70.4 (14.4) | p < 0.001 |
| Male; (%) | 540 (60.4) | 83 (54.2) | NS |
| Receiving oxygen; (%) | 479 (53.6) | 120 (78.4) | p < 0.001 |
| Stay before assessment; days | 6 [3–12] | 9 [4–21] | p = 0.001 |
| Hospital stay; days | 21 [11–38] | 20 [9–41] | NS |
| Primary referral; (%) | 172 (19.2) | 73 (47.7) | p < 0.001 |
| Outreach service outcome; n (%) | p < 0.001 | ||
| WARD | 778 (87.0) | 56 (36.6) | |
| CRITICAL CARE | 97 (10.9) | 38 (24.8) | |
| THERAPY LIMIT | 19 (2.1) | 59 (38.6) | |
- Outcome at the time of hospital discharge; Primary referral: ward patient referred to the outreach service because of a trigger score or general clinical concern. The other patients were routine follow-ups after discharge from a critical care area; Outreach service outcome – WARD: remained on ward receiving ongoing care, CRITICAL CARE: patient transferred to a critical care area within the hospital, THERAPY LIMIT: treatment limitation decision that critical care was not appropriate or death while under review by the outreach service; NS: not statistically significant (p > 0.05). The percentage in row one is of the total number of patients (n = 1047). All other percentages are of the number in each column.
The physiological data have been presented as a range of values associated with a range of hospital mortality (Table 4). Table 4 also gives the incidence for each range of the physiological values.
| Hospital mortality range | |||||||
|---|---|---|---|---|---|---|---|
| ≥35% | 25 – <35% | 15 – <25% | <15% | 15 – <25% | 25 – <35% | ≥35% | |
| Temperature; °C | <35.5 | 35.5–36.4 | 36.5–37.5 | 37.6–38.4 | ≥38.5 | ||
| n (%) | 26 (2.5) | 262 (25.0) | 641 (61.2) | 84 (8.0) | 34 (3.2) | ||
| Mortality (%) | 38 | 17 | 12 | 17 | 26 | ||
| Heart rate; beats.min−1 | <60 | 60–99 | 100–119 | ≥120 | |||
| n (%) | 35 (3.3) | 687 (65.6) | 234 (22.3) | 91 (8.7) | |||
| Mortality; % | 23 | 12 | 16 | 31 | |||
| Systolic blood pressure; mmHg | <90 | 90–99 | 100–109 | 110–159 | ≥160 | ||
| n (%) | 34 (3.2) | 50 (4.8) | 110 (10.5) | 722 (69.0) | 131 (12.5) | ||
| Mortality (%) | 50 | 30 | 17 | 11 | 16 | ||
| Respiratory Rate; breaths.min−1 | <6 | 6–24 | 25–29 | 30–34 | ≥35 | ||
| n (%) | 1 (0.1) | 769 (73.7) | 130 (12.5) | 64 (6.1) | 80 (7.7) | ||
| Mortality (%) | 100 | 9 | 21 | 28 | 41 | ||
| S pO2; % | <85 | 85–89 | 90–94 | ≥95 | |||
| n (%) | 24 (2.3) | 37 (3.5) | 146 (14.0) | 837 (80.2) | |||
| Mortality (%) | 46 | 35 | 17 | 12 | |||
| Level of consciousness | Alert | Confused or responds to voice or responds to pain | Unresponsive | ||||
| n (%) | 724 (69.6) | 288 (27.7) | 28 (2.7) | ||||
| Mortality (%) | 9 | 25 | 50 | ||||
| Urine output; ml.kg−1.h−1 | 0 – <0.5 | ≥0.5 or on dialysis* | |||||
| n (%) | 116 (12.4) | 819 (87.6) | |||||
| Mortality (%) | 40 | 11 | |||||
- * Dialysis: normally dialysis dependent.
There was a highly statistically significant relationship between the number of physiological abnormalities, defined as a Patient-At-Risk score of greater than 0, and hospital mortality (Chi-Squared for trend p < 0.0001) (Table 5).
| No. of physiological abnormalities | ||||||
|---|---|---|---|---|---|---|
| 0 | 1 | 2 | 3 | 4 | ≥5 | |
| Patients; (%) | 176 (16.8) | 261 (24.9) | 262 (25.0) | 196 (18.7) | 98 (9.4) | 54 (5.2) |
| Mortality (%) | 4.0 | 4.6 | 10.7 | 23.0 | 33.7 | 51.9 |
| Odds ratio | 2.12 (1.08–4.18) | 3.61 (1.93–6.78) | 9.00 (4.64–17.45) | 22.35 (10.17–49.13) | 26.0 (10.31–65.60) | |
- Physiological abnormality for each parameter was defined as the range associated with a Patient-At-Risk early warning score of more than zero (Table 1).
- Logistic regression explanatory variable = number of physiological abnormalities. Odds ratio compared to zero abnormalities; significance for one abnormality p = 0.03, for other number of abnormalities p < 0.0001.
- An increasing number of physiological abnormalities was associated with a higher hospital mortality (Chi-squared for trend, p < 0.0001).
Table 6 gives details of the outcome of the outreach service episode and hospital outcome by Patient-At-Risk score. When looking at the ability of the scores to discriminate between patients who needed intervention (CRITICAL CARE or THERAPY LIMIT combined) from those who did not, the area under the receiver operating characteristic (ROC) curve of the Patient-At-Risk score was 0.822. Increasing Patient-At-Risk scores were associated with increased intervention and hospital mortality (Chi-Squared for trend p < 0.0001). Examining the Patient-At-Risk score, binary logistic regression using backward conditional data entry at a value of p < 0.05 showed that all physiological components except temperature contributed to the model predicting the need for intervention and all components except temperature and heart rate contributed to the model predicting hospital outcome.
| Patient-At-Risk score | Outreach outcome (%) | Hospital outcome (%) | ||||
|---|---|---|---|---|---|---|
| Score | n (%) | WARD | CRITICAL CARE | THERAPY LIMIT | Alive | Dead |
| 0 | 176 (16.8) | 98.3 | 1.7 | 0.0 | 96.0 | 4.0 |
| 1 | 211 (20.2) | 95.3 | 4.7 | 0.0 | 96.2 | 3.8 |
| 2 | 173 (16.5) | 88.4 | 7.5 | 4.0 | 88.4 | 11.6 |
| 3 | 142 (13.6) | 81.7 | 14.1 | 4.2 | 90.8 | 9.2 |
| 4 | 112 (10.7) | 68.8 | 23.2 | 8.0 | 83.0 | 17.0 |
| 5 | 78 (7.4) | 61.5 | 20.5 | 17.9 | 76.9 | 23.1 |
| 6 | 55 (5.3) | 65.5 | 20.0 | 14.5 | 72.7 | 27.3 |
| 7 & 8 | 57 (5.4) | 42.1 | 29.8 | 28.1 | 47.4 | 52.6 |
| ≥9 | 43 (4.1) | 14.0 | 44.2 | 41.9 | 46.5 | 53.5 |
- Outreach outcome: WARD: remained on ward receiving ongoing care; CRITICAL CARE: patient transferred to a critical care area within the hospital; THERAPY LIMIT: treatment limitation decision that critical care was not appropriate or death while under review by outreach.
- Hospital outcome: Alive: alive at discharge or transfer from hospital; Dead: died in hospital.
Selecting a suitable trigger score will determine the outreach service workload. There were 387 patients with Patient-At-Risk scores of one or less, of whom 15 (3.9%) died in hospital and interventions were made in 13 (3.4%). There were 660 patients scoring two or more points, of whom 138 (20.9%) died in hospital and interventions were made in 200 (30.3%). If the threshold were to be increased to four or more points, 345 patients would be identified, of whom 105 (30.4%) died in hospital and interventions were made in 154 (44.6%).
Discussion
Intensive care outreach services were introduced throughout much of the United Kingdom in the latter half of 2000 and early 2001. The Audit Commission's 1999 report, ‘Critical to Success’, first used the term ‘outreach’ in this context [13]. The ‘highest priority recommendations’ included agreeing ‘danger signs’ to help identify patients at risk of deteriorating. Comprehensive Critical Care [14], a Department of Health Report published in 2000, further developed the concept of outreach.
Professor Hillman and colleagues pioneered medical emergency teams (METs) in Australia [15] with call-out criteria based upon deranged physiological values. The introduction of medical emergency teams has been shown to reduce cardiorespiratory arrests on the wards and decrease mortality [16, 17]. In our hospital we piloted our outreach service in 1997 [18]. This showed that there were large numbers of critically ill patients on the wards. When our outreach service was aware of the seriously ill ward patients, cardiorespiratory arrests were prevented.
Compared to hospital survivors, patients seen by our outreach service who died in hospital were older and were more likely to be primary referrals than patients recently discharged from a critical care area. The high percentage of primary referrals who were admitted to a critical care area, died or had decisions made to limit treatment, demonstrates that this group of patients was appropriately selected for involvement with the outreach service.
Our study included only those patients already selected to receive outreach care. They are therefore likely to be among the sickest patients in the hospital. We were interested in the ability of the Patient-At-Risk score to identify, from within this group, patients with poor outcome. However, the Patient-At-Risk score is not used to predict outcome. In common with other physiologically based early warning scores it is a screening tool designed to alert clinical staff to a potential problem with a patient. Our data confirm the strong association between abnormal physiology and decisions made. It is interesting that a single array of measurements taken on the wards when the patient is seen for the first time by the outreach service also provides valuable information about hospital outcome. Critical care scoring systems such as APACHE II [19] predict outcome on the basis of physiological abnormality measured in the ICU but there is little information about the relationship between physiological abnormality and outcome in ward patients.
The physiological variables in the Patient-At-Risk score are routinely measured on a ward. Our analysis depends upon the range of values used to assign points in the Patient-At-Risk score. Although these values are in broad agreement with other groups, our conclusions might have been different if other thresholds had been selected. A review of the physiological data (Table 4) suggests that gross abnormalities of all the physiological parameters are associated with adverse outcomes. More data are necessary to permit formal statistical analysis to derive critical thresholds.
A single score has the benefit of being simple and widely applicable. It is likely, however, that different groups of patients will influence the score in different ways. For example, different triggers may be appropriate for medical or surgical patients, or for primary ward referrals as opposed to patients stepping down to the wards from critical care areas. A logical next step would be to introduce more physiological monitoring onto hospital wards. An early warning score could be used to trigger a graded response where lower scores mandate a minimum frequency of monitoring. As scores increase, the frequency of monitoring could increase and patients could be placed in higher intensity care areas. At certain thresholds a prompt medical response should be initiated and standards could be laid down with respect to the time to attend, and the seniority and expertise of personnel to be contacted.
Our data are from a single institution in one health care system. More information from a wider group of hospitals is required to validate the conclusions. Compared to other health care systems, the United Kingdom is short of critical care facilities, which may, in part, be responsible for the higher mortality among high-risk patients [20]. Because the outreach service only saw selected patients we do not know what percentage of the hospital population could potentially benefit from assessment and intervention. A point prevalence study in our hospital in 2002 found that 10.9% of the patients not on the ICU had three or more physiological abnormalities and a 30-day mortality of 21.3%[21]. A further 20.1% had two abnormalities and a 9.2% 30-day mortality. If only those patients with a Patient-At-Risk score of two or greater are assessed, a high-risk group is identified in whom decisions on critical care admission or treatment limitation are made in about 30%. However, even the patients who the outreach service concluded were well enough to remain on the ward received outreach care. These patients were assessed by the outreach service and suggestions were often made regarding the most appropriate treatment interventions. Follow-up continued, sometimes for several days, until it was felt that the outreach service had nothing more to contribute directly. This follow-up role may save lives, decrease the number of critical care re-admissions and provide valuable continuity between the critical care units and the wards [22]. Changes in the delivery of healthcare and the training of doctors [23] and nurses may have resulted in less ward based knowledge and skills to be able to recognise, assess and manage seriously ill patients. The outreach service is also involved in educating ward staff to look after seriously ill patients.
These findings are relevant to all staff involved in the delivery of critical care. Patients are not concerned where their care is delivered and the ICU does not exist in isolation from the rest of the hospital. Evidence suggests that there are considerable numbers of critically ill patients on acute wards in the UK. Many are in hospital for days before their eventual demise or admission to a critical care area. Physiologically based early warning scores are one way in which patients could be identified and tracked. There is now some evidence to suggest that early intervention may improve outcome. The implementation of a system to ensure regular, accurate measurement and recording of physiological values at the bedside should be possible. We would suggest that this is an essential part of any hospital-wide system involving high-risk patients.
