QJM Advance Access originally published online on July 3, 2007
QJM 2007 100(8):501-507; doi:10.1093/qjmed/hcm055
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Laboratory risk factors for hospital mortality in acutely admitted patients
From the Liverpool School of Tropical Medicine, Liverpool, UK
Address correspondence to Professor G.V. Gill, Liverpool School of Tropical Medicine, Pembroke Place, Liverpool L3 5QA. email: g.gill{at}liv.ac.uk
Received 7 September 2006 and in revised form 5 February 2007
 |
Summary |
|---|
 Top Summary IntroductionMethodsResultsDiscussionReferences |
|---|
Background: Many factors affecting hospital mortality in acutely admitted patients are poorly understood. Although scoring systems exist for critically ill patients, usually in intensive care units (ICUs), there are no specific mortality prediction systems for general acute admissions.
Aim: To assess the relationship between simple admission laboratory variables on the risk of in-patient mortality.
Design: Retrospective analysis of hospital admissions and laboratory databases.
Methods: Where possible, all deceased patients in the 12-month period of study were matched with two surviving controls. The laboratory database was then analysed for admission investigations, including serum sodium, plasma glucose, and white blood cell (WCC) count. Abnormalities of these variables were then compared between cases (those who subsequently died), and controls (those who survived).
Results: There were 16 219 admissions, with an overall mortality of 7.6%. We investigated 602 cases and 1073 controls. Hyperglycaemia (glucose >11.0 mmol/l) (OR 2.0, p < 0.0001); severe hyponatraemia (sodium <125 mmol/l) (OR 4.0, p < 0.0001); and leukocytosis (WCC >10 x 109/l) (OR 2.0, p < 0.001) were significantly associated with mortality. The respective associations on logistic regression analysis were: glucose, OR 1.7, p = 0.02; sodium, OR 4.4, p < 0.0001; WCC, OR 1.5, p = 0.006. Low glucose levels, high sodium levels, and low WCC levels were also associated with increased mortality, leading to ‘U-shaped’ mortality associations. The effect of more than one laboratory abnormality being present was cumulative, in a linear fashion.
Discussion: Plasma glucose, serum sodium and WCC are measured in most acutely admitted patients, and abnormalities of these variables have associations with in-hospital mortality. This may provide the basis for the development of a mortality risk scoring system.
|
Introduction |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
The ability to predict likely outcome in acutely admitted hospital patients can be beneficial in several ways. High-risk patients can receive especially intensive management from health workers, or in cases of extreme adverse prognosis, intensive management may be curtailed in favour of a more palliative approach. For patients and their relatives, an accurate assessment of likely outcome may be helpful on humanitarian grounds.
There are two approaches to such outcome assessment measures: one essentially clinical, the other investigative. Clinical risk scores have been used for many years, particularly in intensive care unit (ICU) situations. The APACHE II system is a well-known example of this, and uses mainly various clinical parameters to predict outcome on admission to an ICU.1 Outside ICU situations, the MEWS score similarly assesses risk stratification, though this is designed mainly to identify patients who may benefit from higher-dependency care.2
Risk assessment on the basis of laboratory investigations is also commonly used, but is usually applied in specific disease situations, and generally gives arbitrary (rather than numerical) assessments of risk. Examples include the degree of elevation of serum troponin T and abnormalities of the electrocardiogram (ECG) in patients with myocardial infarction,3 or the degree of plasma hyperosmolality in diabetic patients with hyperglycaemic emergencies.4
There are, however, common laboratory measurements which appear to be associated with general mortality risk in hospital patients. For example, hyponatraemia (particularly of severe degrees) has been known for many years to be associated with high mortality in general hospital patients.5–8 More recently, leukocytosis (in non-infective situations) has been shown to be associated with increased coronary and general mortality.9–11 Finally, admission plasma or blood glucose (in non-diabetic situations) strongly predicts mortality in patients admitted with myocardial infarction,12–14 stroke,15 and other general medical problems.16 Of particular interest and importance, reduction of glucose levels in critically ill patients by insulin treatment improves outcome,17 suggesting that this laboratory abnormality at least is a modifiable variable.
Fortuitously, the vast majority of acutely ill patients admitted to Western hospitals have all three of these variables measured routinely. Serum or plasma sodium (Na) is measured as part of ‘U&E’ (urea and electrolytes), or a ‘renal profile’. The white blood cell count (WCC) is included in a ‘full blood count’ or ‘haematogram'; and plasma or blood glucose is also usually routinely measured. We have therefore undertaken a large database exploration of these measurements and their effect on in-patient mortality, in a large British teaching hospital. In particular, we have case-controlled deceased patients, and explored causes as well as occurrence of death, and the cumulative effects when more than one of the abnormal laboratory variables is present.
|
Methods |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
A computerized hospital database analysis was carried out in a single calendar year (January to December 2002) of all acute admissions to a British urban teaching hospital (Royal Liverpool University Hospital). An acute admission was defined as a patient admitted after presenting to the accident and emergency department, or referred by their general practitioner with an acute medical or surgical event. There were 16 219 such acute admissions of whom 1227 (7.6%) died during hospitalization. All the deceased patients were electronically selected as cases, and for each case, two controls were sought from the admission lists, matched for sex, age band (10-year intervals), hospital specialty, and the nearest sequential date of admission. For some patients—notably those admitted to the intensive care unit (ICU)—it was not possible to select a second control; and for other patients, no appropriate matching control could be found.
Of the 602 cases, 123 were matched to two controls, 405 to a single control, and 74 were unmatched. Of the controls, 422 were not matched to cases. Some patients had missing laboratory variables as follows: sodium 1, glucose 271, chloride 2, potassium 97, bicarbonate 1, urea 1, creatinine 1, anion gap 103, platelets 7, neutrophils 27 and lymphocytes 27. The final group for analysis was thus 602 cases (deceased) and 1073 controls (survivors)
For the purposes of analysis, we defined hyperglycaemia as a random plasma glucose >7.0 mmol/l, hyponatraemia as a plasma sodium (Na) <135 mmol/l, and leukocytosis as white blood cell (WBC) count >10.0 x 109/l. As well as our major test variables of serum sodium, plasma glucose, and white blood cell count, we also analysed the effect of other measurements available with these: serum potassium and urea (with the renal profile), and haemoglobin, platelets, neutrophils and lymphocytes (with the blood count). For each variable, we designed ‘bandings’ of ranges. For example, 2.5–7.0 mmol/l was the normal band for serum urea, and the high and low bands were >7.0 and <2.5 mmol/l, respectively. For sodium and glucose (which we wished to investigate in more detail), more bandings were chosen. Mortality odd ratios (OR) were determined for these bandings, against the mortality for these in the normal ranges.
Data analysis used the Statistical Package for Social Sciences (SPSS), version 14. Univariate analysis of 2 x 2 tables was by Fischer's exact test or 
2 tests (depending on number of observations). Logistic regression analysis was performed to simultaneously investigate the impact of the measured variables on mortality. Because for some cases no matching control could be found, an unconditional logistic regression (i.e. ignoring matching) was carried out. This allowed the whole dataset to be analysed and, more importantly, meant that unmatched cases were not dropped, which could have introduced selection bias.
Ethical approval was obtained from the Ethical Committees of the Royal Liverpool University Hospital and the Liverpool School of Tropical Medicine.
|
Results |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
Group characteristics
The 602 deceased patients were of mean (±SD) age 75 ± 13 years; 79% were acute medical admissions and 21% surgical, and the duration of hospital stay was 20 ± 26 days (mean±SD). The 1073 controls (survivors) had a mean age of 76 ± 12 years, 85% were medical, and the mean length of hospitalization was 18 ± 23 days. None of these parameters were significantly different between patients and controls.
Mortality risk factors
The risk of mortality for each of the measured laboratory variables, together with appropriate confidence intervals and levels of significance, is shown in Table 1, both for univariate analysis and logistic regression. Because many variables were correlated with each other (Table 2), we relied on logistic regression to indicate true mortality links. The latter analysis showed particularly strong statistical associations of mortality with hyponatraemia, hypernatraemia, leukopenia, thrombocytopenia and raised creatinine.
|
|
Mortality pattern for major laboratory variables
Figures 1–3
show the pattern of mortality risk for the three main laboratory variables: plasma glucose (Figure 1), serum sodium (Figure 2), and white blood cell count (Figure 3). It can be seen that using the bandings from Table 1, in each case there was a ‘U’-shaped curve, with higher mortality risk at both abnormally high and low levels. The effect was particularly marked for sodium.
|
|
|
Cumulative effect of laboratory variables
Figure 4 shows the mortality risk of abnormalities of the three major variables (hyperglycaemia, hyponatraemia, and leukocytosis) individually, and also in combination. It can be seen that combinations of these abnormalities led to a graded and approximately linear increase in mortality risk.
|
Mortality causes
Using standard coded causes of death (ICD-10), and the major significant risk factors of hyponatraemia, leukocytosis, and hyperglycaemia; mortality causes were compared between those with and without these laboratory variables for each diagnostic category. Comparing hyponatraemia and normonatraemic patients, only pneumonia was over-represented in those with hyponatraemia (mortality 11.8% vs. 7.8%, p = 0.04). Similarly with leukocytosis, again only pneumonia showed a significant excess (13.3% vs. 4.3%, p < 0.0001). For hyperglycaemia, there was a significant excess of myocardial infarction (6.9% vs. 2.7%, p < 0.001), as well as pneumonia (11.9% vs. 7.1%, p < 0.01).
|
Discussion |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
Despite the considerable importance of hospital mortality, there has been little robust research on early predictive factors. Our study involved a large cohort, with a carefully selected group of case-controls, and we also studied acute general admissions. Previous research in the area has tended to concentrate on specific diagnostic categories (e.g. myocardial infarction12–14 or stroke15) or especially high-risk groups of patients (e.g. in intensive care units1). Also, we investigated the effect of laboratory variables which are almost universally measured in all patients admitted to Western hospitals: electrolytes, renal function tests, plasma glucose, haemoglobin and white blood cell count.
Univariate analysis showed statistically significant associations with mortality for abnormal levels of most of the variables tested (Table 1), but the more critical logistic regression analysis did not always confirm these associations. In particular, the relationship between hyperglycaemia and mortality was greatly reduced in significance. It may be that at least some of the findings significant on univariate analysis are age-dependent, a hypothesis that could be tested using a different control group that is not age-matched. Hypernatraemia (>145 mmol/l) and severe hyponatraemia (<125 mmol/l) maintained highly significant mortality associations with logistic regression; as did uraemia, thrombocytopenia, leukocytosis, and lymphopenia.
An especially interesting finding was that for the common variables of plasma glucose, serum sodium, and white blood cell count, there was a ‘U-shaped’ relationship between their levels and mortality (Figures 1–3
). Thus both low and high levels were associated with excess death risk, compared to levels within the normal range. This common effect has not been previously described to our knowledge.
Perhaps not surprisingly, when hyponatraemia, hyperglycaemia and leukocytosis existed together (either two of the three, or all three), there was a cumulative effect on mortality. Figure 4 suggests that the cumulative effect is linear. The odds ratio (OR) for mortality ranged from 1.6 to 2.3 for one abnormality, 2.7 to 3.9 for two, and 4.0 for all three.
Our data on the cause of mortality should be interpreted with caution. Mortality codings tend to be restrictive, and coding inaccuracies are well known. Nevertheless, we did show an association between hyperglycaemia and myocardial infarction (MI) as a cause of death, as previously reported.12–14 We did not, however, find an association between leukocytosis and MI, which has been described previously.18
In conclusion, abnormalities of a number of commonly measured laboratory variables on admission to hospital were strongly associated with subsequent in-patient mortality in this patient group. Particular associations were seen with severe hyponatraemia, hyperkalaemia, uraemia, leucocytosis and lymphopoenia. In many cases, the causal link with mortality remains to be determined. Our findings raise the possibility of the development of a laboratory-measurement-based predictive scoring system, to assess potential outcome during hospital admission.
|
References |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
1. Markgraf R, Deutchinoff G, Pientka L, Scholtent T. Comparison of Acute Physiology and Chronic Health Evaluation II and III and Simplified Acute Physiology Score II: a prospective cohort study evaluating these methods to predict outcome in a German interdisciplinary intensive care unit. Crit Care Med (2000) 28:26–33.[CrossRef][ISI][Medline]
2. Subbe CP, Kruger M, Rutherford P, Gemme L. Validation of a Modified Early Warning Score in medical admissions. Q J Med (2001) 94:521–6.
3. Heidenrich PA, Alloggiamento T, Melsop K, McDonald KM, Go AS, Hlatky MA. The prognostic value of troponin in patients with non-ST elevation acute coronary syndromes: a meta analysis. J Am Coll Cardiol (2001) 38:478–85.
4. MacIsaac RJ, Lee LY, McNeil KJ, Tsalamandris C, Jerums G. Influence of age on the presentation and outcome of acidotic and hyperosmolar diabetic emergencies. Intern Med J (2002) 32:379–85.[CrossRef][ISI][Medline]
5. Baron D, Hutchinson TA. The outcome of hyponatraemia in a general hospital population. Clin Nephrol (1984) 22:72–6.[ISI][Medline]
6. Tierney WM, Martin DK, Greenlee MC, Zerbe RL, McDonald CJ. The prognosis of hyponatremia at hospital admission. J Gen Intern Med (1986) 1:380–5.[ISI][Medline]
7. Natkunam A, Shek CC, Swaminathan R. Hypoanatraemia in a hospital population. J Med (1992) 22:83–96.[ISI]
8. Nzerue CM, Baffoe-Bonnie H, Falana B, Dai S. Predictors for outcome in hospitalized patients with severe hyponatraemia. J Nat Med Assoc (2003) 95:335–43.[Medline]
9. Cannon CP, McCabe CH, Wilcox RG, Bentley JH, Braunwald E. Association of white blood cell count with increased mortality in acute myocardial infarction and unstable angina pectoris. Am J Cardiol (2001) 8:636–9.
10. Labry LO, Campion EW, Glynn RJ, Vokonas PS. White blood cell count as a predictor of mortality: results of 18 years from the Normative Aging Study. J Clin Epidemiol (1990) 43:153–7.[CrossRef][ISI][Medline]
11. Reding MT, Hibbs JR, Morisson VA, Swaim WR, Filice GA. Diagnosis and outcome of 100 consecutive patients with extreme granulocytic leukocytosis. Am J Med (1998) 104:12–16.[CrossRef][ISI][Medline]
12. Capes SE, Hunt D, Malmberg K, Gerestein HC. Stress hyperglycemia and increased risk of death after myocardial infarction in patients with and without diabetes: a systematic overview. Lancet (2000) 355:773–8.[CrossRef][ISI][Medline]
13. Norhammar AM, Ryden L, Malmberg K. Admission plasma glucose: independent risk factor for long-term prognosis after myocardial infarction. Diabetes Care (1999) 22:1827–31.
14. Ainla T, Baburin A, Teesalu R, Rahu M. The association between hyperglycaemia on admission and 180-day mortality in acute myocardial infarction patients with and without diabetes. Diabetic Medicine (2005) 22:1321–5.[CrossRef][ISI][Medline]
15. Williams LS, Rotich J, Qi R, et al. Effects of admission hyperglycaemia on mortality and costs in acute ischemic stroke. Neurology (2002) 59:67–71.
16. Umpierrez GE, Isaac SD, Bazargan N, You X, Thaler LM, Kitabchi AK. Hyperglycaemia: an independent marker of in-hospital mortality in patients with undiagnosed diabetes. J Clin Endocrinol Metab (2002) 87:978–82.
17. van den Berghe G, Wouters P, Weekers F, et al. Intensive insulin therapy in critically ill patients. N Engl J Med (2001) 345:1359–67.
18. Barron HV, Harr SD, Radford MJ, Wong Y, Krumholz HM. The association between white blood cell count and acute myocardial infarction mortality in patients <65 years of age: findings from the cooperative cardiovascular project. J Am Coll Cardiol (2001) 38:1654–61.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||