OUP user menu

The Simple Clinical Score predicts mortality for 30 days after admission to an acute medical unit

J. Kellett , B. Deane
DOI: http://dx.doi.org/10.1093/qjmed/hcl112 771-781 First published online: 17 October 2006


Background: Predictive scores such as APACHE II and SAPS II have been used to assess patients in intensive care units, but only the modified early warning (MEW) score has been used to assess acutely ill general medical patients.

Design: Observational study of predictors of mortality.

Setting: Small Irish rural hospital.

Methods: From 17 February 2000 to 29 January 2004, 9964 consecutive patients admitted as acute medical emergencies were divided into a derivation cohort of 6736 patients and a validation cohort of 3228 patients.

Results: In the derivation cohort, 316 patients (4.7%) died within 30 days of hospital admission. Under univariate analysis, age, vital signs and 18 categorical variables were associated with increased risk of death, and nine with reduced risk. Logistic regression identified 16 independent predictors of 30-day mortality, from which the Simple Clinical Score was derived, stratifying patients into five risk classes. In each class, mortality was not significantly different between the derivation and validation cohorts: 0–0.1% for very low risk, 1.5–1.6% for low risk, 3.8–3.9% for average risk, 9.0–10.3% for high risk, and 29.2–34.4% for very high risk.

Discussion: The Simple Clinical Score quickly and accurately identifies patients at both a low and high risk of death from the first to the 30th day after admission, enabling prompt triage and placement within a health-care facility.


Although the measurement of vital signs has been standard practice for over a century, there have been few attempts to quantify their clinical performance. The predictive scores APACHE II1 and SAPS II2 were derived from patients requiring intensive care unit (ICU) treatment, while early warning scores and criteria for calling medical emergency teams were developed empirically by expert committees based on audit of only 100 or so seriously ill patients.3 Since these instruments were designed to only deal with ICU patients, they lack sensitivity and cannot be used to rule out severe illness. Only the modified early warning (MEW) score, validated on 709 patients, has been used to assess acutely ill general medical patients.4 A MEW score >4 is associated with an increased risk of death (OR 5.4, 95%CI 2.8–10.7) and indicates an urgent need for intensive care. The huge advantage of the MEW score is that it simply collates the results of the classic vital signs and can, therefore, be used by anyone at the bedside. However, while a high MEW score is a specific test for serious illness, the score lacks sensitivity. Since the vast majority of patients have scores of 2 or less, the score cannot be used to rule out the possibility of life-threatening illness.

The purpose of the study was to determine the important, immediately available clinical findings that can identify how sick medical patients are when they present to hospital. The study was not confined only to those patients that subsequently required intensive care but prospectively studied, at the time of admission, all patients admitted to an acute medical unit. Thirty-day mortality was used as a surrogate marker for severity of illness. We used only information easily obtained at the bedside.


Nenagh Hospital is a small general hospital in rural Ireland serving a population of 60 000. It has a 36-bed acute medical unit, with 2800 admissions per year, almost all of which are unplanned emergencies. It is served by three consultant physicians, each assisted by a team of three physicians in training; each team is on-call every third day. The hospital has a five-bed ICU capable of cardiac monitoring, external and temporary transvenous pacing, non-invasive and invasive ventilation, etc. Renal dialysis, haematology and oncology units are available at Limerick Regional Hospital, 25 miles away. More than half of all acute medical admissions in Ireland are to small medical units similar to ours.5 Our overall in-hospital mortality rate for acute medical patients is 3.7%, and not significantly different from the rate of 3.3% reported by the Limerick Regional Hospital, the nearest teaching hospital and tertiary referral centre. From 17 February 2000 to 29 January 2004, all medical patients admitted to Nenagh Hospital had their history and physical data entered into a computerized database. Those aged <14 years, and those who died within 15 min of arrival to the hospital, were excluded from the study.

The primary purpose of this study was to determine how much prognostic information could be gained from the vital signs and clinical predicament at the time of admission. All variables collected had to be immediately available at the time of admission, and not require laboratory investigations, clinical expertise or equipment beyond that normally immediately available. Diagnoses such as chronic obstructive airway disease, chronic renal failure, malignancy, etc. were not included, because their immediate and accurate determination would have required either considerable diagnostic skill or expert interpretation of instantly available and reliable medical records. On the other hand, stroke was included, as this common yet serious diagnosis is usually obvious.

In all, six continuous variables (age, blood pressure, pulse rate, respiratory rate, temperature, oxygen saturation) and 23 categorical variables were collected. Two categorical variables captured functional status (Zubrod score,6 ability to stand), eight clinical presentation (mental status, new stroke, intoxication, breathlessness, syncope, gastrointestinal symptoms, chest pain, headache), seven demographic or social features (sex, marital status, tobacco and alcohol consumption, health insurance, self referral, nursing home residency), five prior health status (diabetes, prior stroke, prior myocardial infarction, prior heart failure, prior medication), and one ECG findings. The doctor on-call filled out a standard data collection form when the patient first presented to the hospital. A full-time data collection officer (BD) then, after discussion with the admitting physician and usually within 24 h of admission, corrected any errors or omissions before entering the data into an Epi-Info version 6.0 database.

All deaths registered within the hospital's catchment area from 17 February 2000 to 6 March 2004 were examined, to identify patients who died after discharge but within 30 days of admission. Of the 11 124 patients admitted during the study period, 89 who remained in hospital for >30 days, 785 with incomplete data, and 286 who lived outside the hospital's catchment area (and hence were unavailable for follow-up) were excluded from the study. The final study population comprised 9964 patients, all with complete data. The study population was divided into two cohorts: a derivation cohort comprising all patients admitted before October 28, 2002 and a validation cohort comprising all patients admitted after that date (Table 1). There were some statistically significant differences between the derivation and validation cohorts, including blood pressure, temperature, gender, nursing home residency, self-referral, private health insurance, etc. The derivation cohort was examined for both univariate and independent predictors of mortality within 30 days of admission. The Simple Clinical Score was developed from the independent predictors of mortality that were identified. Receiver operator characteristic (ROC) curves then compared the performances of this score in both the derivation and validation cohorts.

View this table:
Table 1

Comparison between derivation cohort and validation cohort

Neither the medical nor nursing staff received any special training on how to measure vital signs during the study. Respiratory rate was measured manually. Blood pressure, pulse rate and oxygen saturation were measured electronically using Vital Care 506 DXN (Criticare Systems Inc.) and checked manually if needed. Body temperature was measured by the Genius First Temp tympanic thermometer (Tyco Healthcare AG). ECGs were classified as normal or abnormal according to the interpretation of the Marquette MAC-PC computerized ECG interpretation program. An ECG was considered to be normal if the only abnormality recorded was tachycardia or bradycardia. Patients were considered unwell prior to their current illness if they usually needed to spend some part of normal daytime in bed (i.e. a Zubrod score of 2 or more).6 If patients had given up smoking or alcohol for >1 year they were deemed to be ex-smokers or ex-drinkers, respectively. Altered mental status was divided into those patients with and without coma: coma was defined as either being unresponsive or responsive only to pain at the time of admission and, hence, unable to stand unaided. Altered mental status without coma included confusion, agitation or obtundation. The presence or absence of abnormal brain stem responses or Glasgow Coma Scale was not recorded.

Statistical significance was tested using Student's t-test and χ2 analysis. Logistic regression using Logistic (Gerard E. Dallal, Tufts University) software identified independent predictors of mortality.7 ROC curves were constructed and the area under the curves (AUC) compared according to the method of Hanley and McNeil8 using ROC Analyzer (R.M. Centor and J. Keightley, Richmond VA).


In the derivation cohort, 316 patients (4.7%) died within 30 days of hospital admission: 40 (12.7%) died within 24 h of admission, 135 (42.7%) within 1 week and 224 (70.9%) within 2 weeks of admission. Seventy-one patients died after discharge from hospital; 22 of these after discharge to a nursing home. There were significant differences in the mean values of age and all the classic vital signs, except temperature, between those patients who died and those who survived. Patients who died also had significantly lower blood oxygen saturations (Table 2). There was a clear relationship between age and 30-day mortality that showed a slight gender difference. Of the 1016 women aged <55 years, only two (0.2%) died, while three (0.3%) of the 909 men aged <50 years died. No significant gender differences could be demonstrated within any other variable examined.

View this table:
Table 2

Comparison of continuous variables for derivation cohort

Mean values of vital signs expressed as continuous variables are of little clinical value, since excess mortality is only observed either above and/or below their normal ranges. For example, although there were no differences in the mean body temperatures between those who died and those who survived, both a low and high temperature were associated with an increased chance of death. Therefore, continuous variables were converted into categorical variables by placing them into groups that were then bar-graphed against 30-day mortality. By a process of visual inspection these bar graphs revealed the practical convenient cut-offs. In all cases, the cut-off for patients with a low mortality (i.e. normal ranges) was obvious (Table 3). Further incremental cut-offs were chosen when there was a further substantial increase in 30-day mortality. For example, oxygen saturation graphed into 5% decrements showed a dramatic increase in 30-day mortality between 90% and 95%, which approximately doubled below an oxygen saturation of 90%.

View this table:
Table 3

Comparison of categorical variables derived from continuous variables for derivation cohort

Further univariate analysis of the derivation cohort revealed that neither syncope nor gastrointestinal symptoms nor prior history of myocardial infarction were associated with an increased risk of 30-day mortality. However, 18 categorical variables were associated with an increased risk of mortality, and nine were associated with a significantly reduced risk of death (Table 4). Coma and altered mental status had the highest association with mortality, particularly if the influence of intoxication (which conferred a greatly reduced chance of death) was excluded.

View this table:
Table 4

Comparison of other categorical variables for derivation cohort

Independent predictors of 30-day mortality were determined by logistic regression. In addition to all the continuous variables, all 18 categorical variables associated with an increased mortality and the nine variables associated with a reduced mortality were tested by logistic regression: only 16 were independent predictors of mortality (Table 5). These were then converted into categorical variables that were assigned points, adjusted so that the logistic regression model's odds ratio for each variable became approximately equal. While the initial guesses of the points for each independent predictor were based on the χ2 and p values from the univariate analysis, the final judgements were made by trial and error to produce similar odds ratios for each variable. In the final logistic regression model, the odds ratios ranged from 1.4 for age groups to 1.8 for coma without intoxication. The combined total of all these points yielded the Simple Clinical Score for the prediction of mortality (Table 6).

View this table:
Table 5

Odds ratios for the 16 independent predictors of 30-day mortality in the derivation cohort logistic regression model

View this table:
Table 6

Total points awarded for the 16 independent predictors of 30-day mortality

ROC curves were constructed using the Simple Clinical Score to predict mortality from the first to the 30th day after admission. The ROC curves of the derivation and validation cohorts had an identical area under the curve (AUC) from the first to the 30th day after admission (Figure 1). The Simple Clinical Score's points naturally fell into five risk classes according to the mortality rate observed within them: very low risk (0–3 points), low risk (4–5 points), average risk (6–7 points), high risk (8–11 points) and very high (⩾12 points). Very high and very low risk patients are easily identified at either end of the ROC curve (Figure 1b). The remaining patients with between 4–11 points divide into three approximately equal groups. While the theoretical maximum score is 31 points, the mean points score for very high risk patients was 13.3 ± 1.7 (median 13.0) and ranged from 12 to 21 points. Approximately one third of patients admitted were classed as very low risk, and only two of these died, compared to the 5% of patients classed as high risk, about a third of whom died. The remaining patients were more or less equally divided into those with low, average and high risk. There were no significant differences in observed mortality from the time of admission and the next 30 days between the derivation or validation cohorts in any of the five risk classes: the higher the class the greater the mortality (Figure 2). Moreover, the higher the risk class. the greater the age, length of hospital stay, readmission rate and the need for nursing home care after discharge (Table 7).

Figure 1.

a Comparison of the Simple Clinical Score ROC curves for mortality within 24 h of hospital admission for the derivation and validation cohorts. The area under the curve (AUC) for mortality on the first 24 h after admission was 90.2% (standard error (SE) 0.019) for the derivation cohort and 90.9% (SE 0.027) for the validation cohort (p 0.42). b Comparison of the Simple Clinical Score ROC curves for mortality within 30 days of hospital admission for the derivation and validation cohorts. The area under the curve (AUC) for mortality up to 30 days after admission was 85.8% (SE 0.009) for the derivation cohort and 85.6% (SE 0.013) for the validation cohort (p 0.43). The Simple Clinical Score points fell naturally into five risk classes: very high (⩾12 points) and very low risk (0–3 points) patients are easily identified at either end of the ROC curve. The remaining patients, with 4–11 points, divide into three approximately equal groups.

Figure 2.

Mortality at 30 days after admission in very low risk, low risk, average risk, high risk and very high risk patients.

View this table:
Table 7

Outcomes by Simple Clinical Score risk class

Only eight of the 16 independent predictors of 30-day mortality were statistically significant independent predictors of mortality within 24 h of hospital admission. These were: an abnormal ECG, temperature, breathlessness, blood pressure, coma without intoxication, oxygen saturation, a history of prior illness, and inability to stand unaided. The AUC for a logistic regression model that used only these eight variables was 88.9% for the derivation cohort and 88.4% for the validation cohort (cf. 90.2% and 90.9% for the full model). The three most important variables in the full model were age, ECG findings and inability to stand unaided. If any of these three variables were removed from the model, the AUC for 30-day mortality fell from 85.8% to 84.6%. The three least important variables were the relationship between pulse and systolic blood pressure, systolic blood pressure and respiratory rate. If any of these were removed, the AUC for 30-day mortality fell from 85.8% to 85.7%. A model that only included the seven variables of age, ECG findings, history of prior illness, oxygen saturation, inability to stand unaided or nursing home residence, respiratory rate and stroke still had an AUC of 84.2% for the derivation cohort and 84.7% for the validation cohort. However, the AUC for predicting mortality within 24 h for this model only increased to 85.6% for the derivation cohort and 85.4% for the validation cohort.

Out of the entire patient population (i.e. derivation and validation cohorts combined) 59 patients died within 24 h of admission: 21 (36%) from between 30 min and 6 h after admission. Only one low risk patient died during this time: a 67-year old man, from acute myocardial infarction. Of the seven average risk patients who died within 24 h of admission, four died from acute myocardial infarction, two from arrhythmias associated with chronic heart disease and one 27-year-old man with cerebral palsy and mental retardation died from septicaemia. The remaining 51 (86%) patients who died within 24 h were either high or very high risk patients.


Ideally, prediction rules should be easy, quick and cheap to use, and predict something important and clinically relevant. They should also provide accurate predictions over a wide range of clinical situations.9 The Simple Clinical Score satisfies all these criteria. Inspection of the Simple Clinical Score showed that it fell naturally into five risk classes. There was a small group of very high risk patients (⩾12 points), and the remainder could be divided into four roughly equal groups with near zero (0–3 points), half average (4–5 points), average (6–7 points) and twice average mortality (8–11 points). All patients in this study were admitted as acute emergencies; the need for full resuscitation was therefore assumed. The Simple Clinical Score may be useful in the subsequent review of resuscitation status. Clearly it would be inappropriate to re-consider the resuscitation status of very low or low risk patients. On the other hand, palliative care rather than resuscitation may often be the appropriate management for frail elderly high risk patients.

The variables that were tested to devise the Simple Clinical Score were chosen to reflect the symptoms, signs and historical data that were likely to be instantly available when a patient presents as an emergency. Although there were several statistically significant differences between the derivation and validation cohorts, these differences were of doubtful clinical significance. There were substantially more self-referred patients in the validation cohort and slightly more with private health insurance. We have already reported significant differences between self-referred and physician-referred patients, and have also examined the impact of private insurance.10 However, neither of these factors was an independent predictor of 30-day mortality. No degree of hypertension predicted mortality, and a temperature >39°C only just reached statistical significance as an independent predictor of mortality. These findings may reflect the relatively low prevalence of high fever, and the fact that, thanks to modern drugs, few patients now suffer from malignant hypertension. Although numerous studies report the benefits of smoking cessation both ex-smokers and ex-drinkers carried an excess risk of death. This is probably explained by the ‘ill quitter effect’.11

It is possible that over time the treatment in our hospital might have changed and, therefore, it might have been wiser to select our derivation and validation cohorts by a means other than their date of admission. However, the validation cohort yielded results identical to those of the derivation cohort. It may well be that some of the Simple Clinical Score's independent variables are not truly independent, as a result of some ‘double counting’. It is hard to believe, for example, that respiratory rate, the symptom of breathlessness and oxygen saturation are all independent of each other. Nevertheless, at least in our validation cohort, the Simple Clinical Score works in practice. Only time will tell if it works equally well in other patient populations and clinical circumstances.

The Simple Clinical Score ROC curve for 30-day mortality has an AUC of between 85% and 90% and, hence, has considerably more predictive power than the MEW score (reported to have AUC of 67%, or 72% if age was included).4 Since the responses to pain and verbal stimuli of patients with an altered mental status without coma were not recorded, the full MEW score for our patient cohorts could not be retrospectively calculated. However, a modified MEW score, which assigned one point for patients with altered mental status without coma and two points for patients with coma, had an AUC of 64.7% for the derivation cohort and 64.9% for the validation cohort. Moreover, using this modified MEW score, 8% of very low risk and only 34% of very high risk patients had a MEW score >4, suggesting that the MEW score occasionally identifies serious illness when it is absent, and misses it when present.

Other scoring systems, such as APACHE II and SAPS II, require several time-consuming laboratory investigations, and have not been validated in general medical patients. Unlike the MEW and other scores derived from patients requiring intensive care,3 the Simple Clinical Score accurately identified patients at extremely low risk of death. One of the surprising features of the score is its ability, from a single determination, to predict the risk of mortality from the first to the 30th day after admission. Even though the in-hospital mortality rate of our patients was comparable to that of our neighbouring hospital, one of the main concerns with this study is whether or not our patients are representative of general medical patients elsewhere, and thus whether our findings can be applied universally. HIV status was not routinely tested for and, as far as we know, none of the patients in this study were HIV-positive. The fact that surgical admissions were excluded may explain why gastrointestinal symptoms were not associated with mortality. Furthermore, other hospitals may not admit intoxicated patients so readily.

Unlike most triage systems, which require extensive staff training before they can be implemented,12 the determination of the Simple Clinical Score should require only minimal skill and training before any health care worker can estimate it quickly and accurately. In addition to age and the classic vital signs of pulse, blood pressure, temperature and respiratory rate, the score requires oxygen saturation and an ECG. These investigations should be standard care in all emergency departments. Combined with a cursory assessment of mental status and prior health, they yield almost all the information required to determine the immediate risk of death. Although more parsimonious models may be statistically valid, there seems little clinical advantage in abbreviating the Simple Clinical Score. The variables identified by the score resonate with clinical intuition, and the collection of all 16 variables is unlikely to require more work or resources than collecting just a few of them. Although obtaining a more detailed medical history, laboratory tests and imaging may assist in establishing exactly what is wrong with the patient and how best to treat it, the score instantly flags the intensity, skill level and speed of treatment required.

Despite lack of evidence of their sensitivity, specificity and usefulness, a variety of early warning scores and outreach programs criteria have been adopted in many centres.13 For example, although the MEW score has gained considerable popularity for the identification of severely ill patients and monitoring their progress, it has not been shown to improve outcomes of adverse events in medical emergency admissions.14 Unlike other predictive scores,3 the Simple Clinical Score can quickly identify patients with either a very low or a very high risk of death. It is, therefore, both a sensitive and specific identifier of illness severity. It remains to be seen whether its use will improve patient outcomes.

However, the Simple Clinical Score should not be used to decide the need for hospital admission. All patients in this study were admitted to hospital and investigated and treated to the best of our ability. Only two very low risk patients died: both after discharge, one from metastatic throat cancer and one from an overdose of alcohol. However, the fact that a patient has a very low risk of death should not imply that hospital admission is unnecessary. Most of the average or low risk patients who died within 24 h of admission died from acute myocardial infarction. Although patients with chest pain may have a reduced risk of death, they still need to be admitted and treated to rule out active ischaemic heart disease. Hospitals offering a higher or lower quality of care may find that the different risk classes are associated with mortality rates that are correspondingly lower or higher. Indeed, the predictive score may prove to be a valuable tool for assessing and comparing the quality of clinical care between different units.

The Simple Clinical Score quickly and accurately identifies patients at both a low and high risk of immediate death, and their likely outcome at 30 days. Appropriate triage and placement within a health care facility can thus be promptly initiated, without waiting for the results of further investigations. A prospective multi-centre study will be required to validate this instrument further.


The authors would like to acknowledge the help of all the medical, nursing and administrative staff of Nenagh Hospital in the collection of data that made this paper possible. In particular, we would like to thank Mrs Marie Kennedy for her meticulous help.


View Abstract