Identifying the sick: can biochemical measurements be used to aid decision making on presentation to the accident and emergency department

T. R. Hucker1, G. P. Mitchell1, L. D. Blake1, E. Cheek2, V. Bewick2, M. Grocutt3, L. G. Forni1 and R. M. Venn1,*

1 Department of Critical Care and 3 Accident and Emergency Department, Worthing Hospital, Lyndhurst Road, Worthing, West Sussex BN11 2DH, UK. 2 School of Computing, Mathematical and Information Sciences, University of Brighton, West Sussex, UK

* Corresponding author. E-mail: richard.venn{at}wash.nhs.uk

Accepted for publication February 11, 2005.


    Abstract
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Clinical implications
 Appendix
 References
 
Background. Early and accurate identification of patients who may benefit from aggressive optimal medical intervention is essential if improved outcomes in terms of survival are to be achieved. We studied the usefulness of routine clinical measurements and/or markers of metabolic abnormality in the early identification of those patients at greatest risk of deterioration on presentation to the accident and emergency department.

Methods. We conducted a prospective observational study in the accident and emergency department of a 602-bed district general hospital. Routine clinical measurements (heart rate, systolic blood pressure, temperature, oxygen saturation in room air, level of consciousness and ventilatory frequency) and venous blood analysis for metabolic markers (pH, bicarbonate, standard base excess, lactate, anion gap, strong ion difference, and strong ion gap) and biochemical markers (Na+, K+, Ca+, Cl+, albumin, urea and creatinine) were recorded from unselected consecutive hospital admissions over two 3-month periods (September–November 2002 and February–April 2003).

Results. Logistic regression analysis showed that neither conventional clinical measurements upon presentation to the accident and emergency department nor venous biochemical and metabolic indices have good discriminatory ability when used as single predictors of either hospital mortality or length of hospital stay. Selecting variables from all the clinical and venous blood measurements gave a parsimonious model containing only age, heart rate, phosphate and albumin (area under the receiver operating characteristic curve, 0.82 [95% CI 0.76, 0.87]).

Conclusions. A combination of clinical and venous biochemical measurements in the accident and emergency department proved the best predictors of hospital mortality. Consequently, they may be helpful as a triage tool in the accident and emergency department to help identify patients at risk of deterioration.

Keywords: audit, accident and emergency ; clinical observations ; complications, mortality ; hospital stay ; outcome ; physiology, abnormal


    Introduction
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Clinical implications
 Appendix
 References
 
Early and accurate identification of patients who may benefit from aggressive optimal medical intervention is essential if improved outcomes in terms of survival are to be achieved, as shown in the study by Rivers and colleagues.1 This study showed that early goal-directed therapy in the emergency department, before admission to the intensive care unit (ICU), significantly improved survival. In this instance, optimal implies detection of that period whereby pathophysiological processes may be reversed by appropriate therapies such as fluid resuscitation, correction of acidaemia or inotropic support. Traditionally, conventional markers of deranged physiology have included measurement of the heart rate, arterial blood pressure, ventilatory frequency, and urine output and assessment of these in terms of ‘normal values’. However, in certain groups derangements of such variables may be poor as well as late indicators of underlying physiological disturbance.2 In the ICU a variety of tools are utilized in order to predict which patients are likely to deteriorate. Of these, markers of underlying acidosis such as pH, standard base excess (SBE) and lactate, appear to be among the simplest and most useful for the identification of such physiological disturbance.36 Although there are a variety of causes of acidaemia, the early identification of patients with acidosis due to tissue hypoxia may enable improved triage and management decisions to be made.

Recently, there has been a resurgence of interest in physicochemical approaches to further defining acid–base physiology employing variables other than the base excess and plasma bicarbonate concentration as measures of metabolic acidosis.3 This approach describes the acid–base status through three variables that are assumed to be independent: , strong ion difference (SID) and the total concentration of the non-volatile weak acids, principally albumin and phosphate.3 79 SID is calculated as the difference between the sum of the major fully ionized cations (Na+, K+, Ca2+ and Mg2+) and that of the fully ionized anions (principally Cl). The pH, SBE and are presumed to be dependent on these and hence cannot be primarily or individually altered. This approach has been used in the clinical setting,3 10 11 but to date has not proved to be a particularly useful tool in the critical care setting when compared with other more conventional metabolic markers.4 5 12 This may reflect the inevitable delay in admission from the accident and emergency (A&E) department to the critical care setting, which may influence outcome. Also, initial management such as the type of fluid used for resuscitation has been shown to influence SID.13 Therefore application of this approach may identify the sick patient in the A&E department if applied initially in their admission before instituting fluid resuscitation.

The aim of this study is to investigate whether routine clinical measurements (heart rate, blood pressure, percentage oxygen saturations, level of consciousness and ventilatory frequency) and/or markers of metabolic abnormality (e.g. lactate, SBE, anion gap [AG], strong ion gap [SIG]) are useful in the early identification of those patients at greatest risk of deterioration following presentation to the A&E department.


    Methods
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Clinical implications
 Appendix
 References
 
This prospective observational study was conducted in the Accident and Emergency Department of Worthing General Hospital over a 6-month period. Worthing General Hospital is a 602-bed district general hospital situated on the south coast of England, with an annual A&E attendance of >50 000. The majority of general practitioner referrals are admitted via an emergency assessment unit or the relevant speciality wards, thereby bypassing the A&E department.

Following local research ethics committee approval, consecutive hospital admissions from the A&E department over two 3-month periods (September–November 2002 and February–April 2003) were considered. These periods were selected not only for convenience but also as they are representative of the more typical referral patterns rather than the somewhat busier mid-winter period. Patients were excluded if they were referred by a general practitioner, transferred from a secondary or tertiary medical facility, <18 yr, not requiring diagnostic venepuncture, admitted under the specialities of ear, nose and throat, obstetrics and gynaecology, or maxillofacial surgery, died shortly after admission to the A&E department or had suffered a fatal out-of-hospital event, or were palliative care admissions.

In accordance with the Data Protection Act, posters were displayed in the A&E department and permission was obtained during venepuncture for blood analysis data to be used anonymously for research purposes at a later date. Venous blood was drawn and then analysed using an ABL 700 blood gas analyser (Radiometer, Copenhagen, Denmark) to measure venous pH and . Bicarbonate () and SBE were calculated using the Henderson–Hasselbach equation and the Siggaard–Andersen formula, respectively. Simultaneously, sera were analysed using the Roche modular biochemistry analyser (Roche Diagnostics, Lewes, UK). The analysers underwent daily calibration and quality control checks.

Sera were analysed for Na+, K+, Ca2+, Cl, , albumin, urea, creatinine and lactate. AG, SID and SIG were calculated using the formulae given in the Appendix. Since it is not usual practice to measure magnesium in the A&E department, it was assumed constant and was not measured for the calculation of SIDapparent.

During the second period of study, in addition to the venous blood analysis, the following clinical variables were recorded by the admitting nurse: heart rate, systolic blood pressure, temperature, oxygen saturation in room air (), level of consciousness (defined as Alert, responsive to Verbal command, responsive to Pain, or Unresponsive [AVPU score]) and ventilatory frequency. Patients requiring oxygen did not have their measured on air if considered medically inappropriate and these data were not included in subsequent analysis. Furthermore, the case notes for all study patients who died were reviewed, thus ensuring accuracy of reporting.

All patients were followed up to determine hospital survival and length of hospital stay. All therapeutic management was at the discretion of the attending physician.

Statistical analysis
The summary statistics indicated that a number of the variables did not conform to a symmetrical distribution and needed to be transformed for further statistical analysis. The logarithms of lactate (mmol litre–1) and urea (mmol litre–1) and the reciprocal of creatinine (µmol litre–1) were used.

Two sample t-tests were used to assess which of the quantitative variables were related to survival. Fisher's exact test was used to assess the relationship between AVPU score and survival. P-values <0.05 were considered significant. For each variable that showed a significant difference between survivors and non-survivors in the t-test, discrimination was assessed using the area under the receiver operating characteristic curve (AUROC). Age is clearly an important factor in survival, and further analysis was needed to assess whether each variable had prognostic ability even when the age of the patient was taken into consideration. Logistic regression was used to model survival on each prognostic variable in combination with age. Additionally, urea and creatinine are age dependent and therefore were adjusted for age by replacing them with their residuals after fitting quadratic curves.

A model including all the clinical variables and age was produced using multiple logistic regression. Stepwise procedures were used in order to identify which, if any, of the venous blood measurements added significantly to the discrimination provided by the clinical variables. A parsimonious model was then derived using stepwise logistic regression on all the measured variables, both clinical and venous blood. The criterion for the inclusion of a variable in the model was that the P-value for the likelihood ratio test was <0.05. The Hosmer–Lemeshow test was used to assess the calibration for each model, in addition to using AUROC to evaluate the discrimination. Furthermore, to investigate the difference between the surgical and medicine specialities, a categorical variable for speciality, in addition to its interactions with all of the measured variables, was included.

After an initial investigation, length of stay for survivors was grouped into 1 day and >1 day and two sample t-tests were used to assess which of the quantitative variables were related to this categorical variable. Patients who did not survive were omitted from this analysis. Where possible, analyses were carried out on the data collected over both study periods. All analyses were performed using the SPSS statistical package (Chicago, IL, USA).


    Results
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Clinical implications
 Appendix
 References
 
Patients
During the observation period 2221 patients satisfied the study inclusion and exclusion criteria (1044 in first 3-month period, and 1177 in second 3-month period). Venous blood analysis data were obtained for 1424 of these patients (58% of the total sample; 681 in the first period and 743 in the second). Data were not available for 797 patients. Of the patients from whom data were collected, 1296 (91%) survived and 128 (9%) died. In the group where no data were available there were 715 survivors and 82 non-survivors.

For the patients included in the study, the mean length of stay was 8.0 days for the survivors and 9.5 days for the non-survivors. Clinical measurements and age were recorded for 672 of the patients during the second 3-month period. Of these 672 patients, 599 (90%) survived and 73 (10%) died. The mean ages were 66.2 yr for the survivors and 79.3 yr for the non-survivors. The percentage of patients admitted by medical specialty is shown in Table 1.


View this table:
[in this window]
[in a new window]
 
Table 1 Percentage of patients admitted by medical specialty.

 
In-hospital mortality
All variables showed statistically significant differences between survivors and non-survivors with the exception of Na+, K+, Cl, Ca2+ and the calculated variable SIDapparent (Table 2). Fisher's exact test indicated that a significantly (P=0.007) greater proportion of patients classified as alert survived compared with the proportion of those at other levels of consciousness (Table 3).


View this table:
[in this window]
[in a new window]
 
Table 2 Means (95% CI) and t-tests of survivors and non-survivors for measured and calculated variables. The only variables that did not show statistically significant differences between survivors and non-survivors were the measured variables Na+, K+, Cl and Ca2+, and the calculated variable SID apparent

 

View this table:
[in this window]
[in a new window]
 
Table 3 Cross-classification of level of consciousness with survival. Data are shown as absolute number (percentage). Fisher's exact test indicated that a significantly greater (P=0.007) proportion of patients classified as alert survived compared with the proportion of those at other levels of consciousness

 
AUROC is a measure of the ability to discriminate between survivors and non-survivors. Individually none of the variables appear to be good predictors of mortality. The logistic regression analyses show that the prognostic ability of each variable is significantly improved by including age (the P-value for the change in the model when age was included was >0.05). Table 4 also indicates that all the variables except temperature add significantly to the discriminatory ability of age. The highest AUROC value is 0.78 for SIDeffective combined with age (Table 4).


View this table:
[in this window]
[in a new window]
 
Table 4 Logistic regression table showing the odds ratios and AUROC values for each variable allowing for age. Data are from second 3-month period: n=672; number of deaths=73; number of survivors=599. Without allowing for age, none of the variables appeared to be good predictors of mortality; the best were urea, albumin and SIDeffective with AUROC values of 0.73, 0.72 and 0.71, respectively. The logistic regression analysis shows that the prognostic ability of each variable is significantly improved by including age. The table indicates that, with the exception of temperature, all of the variables add significantly to the discriminatory ability of age.

 
The multiple logistic regression model including all the clinical variables gave an AUROC of 0.67, but the addition of age gave an improved AUROC of 0.76 (Table 5). The addition of the venous blood measurements one at a time showed that phosphate and albumin improved the model significantly. The AUROC for this model was 0.84. The parsimonious model obtained by using all the clinical and venous blood measurements included only the variables age, pulse, phosphate and albumin (AUROC=0.82). With AUROC values >0.8, the final two models showed good discrimination. These models were also well calibrated, as indicated by the Hosmer–Lemeshow goodness-of-fit tests (P=0.686 and P=0.768, respectively). The coefficients for calculating the probabilities of death using the parsimonious model are given in Table 6. The inclusion of speciality (medicine/surgical) and its interactions with the other variables did not significantly improve the model.


View this table:
[in this window]
[in a new window]
 
Table 5 Results from the multiple logistic regression analyses (n=672)

 

View this table:
[in this window]
[in a new window]
 
Table 6 Coefficients for calculating predicted probabilities of death from the parsimonious logistic regression model. The predicted probabilities P are calculated using the equation ln (P/[1–P])=–4.80+0.05xage+0.02xpulse+1.97xphosphate–0.12xalbumin

 
Length of hospital stay
The variables that showed a significant difference for length of stay (grouped into 1 day and >1 day) were Na+, Ca2+, Cl, albumin, urea, creatinine, AGadjusted, SIDeffective, SIG, age, temperature, , ventilatory frequency and heart rate (Table 7). With the exception of age, the variables that showed a significant difference did not change when omitting the data for the elderly and orthopaedic specialities (predominantly the more elderly patients).


View this table:
[in this window]
[in a new window]
 
Table 7 Means (95% CI), and t-tests for all measured and calculated variables for length of hospital stay of 1 day and >1 day for survivors

 
Small positive correlations were found between length of hospital stay, urea and creatinine, but linear regression produced no useful model (R2=0.05).


    Discussion
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Clinical implications
 Appendix
 References
 
The logistic regression analyses show that conventional clinical measurements upon presentation to the A&E department did not have good discriminatory ability to predict either hospital mortality or length of hospital stay. These findings concur with previous work performed on a critically ill patient cohort, which demonstrated that heart rate, systolic blood pressure and urine output did not predict outcome.2 This has implications when designing early warning scoring systems in order to detect a patient population at risk of deterioration. Recent comprehensive critical care guidelines14 recommend the use of an early warning system such as MEWS15, which employs a score including systolic blood pressure and heart rate. To date most scoring systems have not undergone rigorous validation. Our results suggest that such variables in isolation may have little or no practical usefulness in detecting patient deterioration.

None of the venous blood measurement data had an AUROC ≥0.8, which is usually taken as good discriminatory ability.16 A combination of all the clinical variables and age gave an AUROC of 0.76. Phosphate and albumin significantly improved this model, giving good discrimination with an AUROC of 0.84. Selecting variables from all the clinical and venous measurements gave a parsimonious model containing only age, pulse, phosphate and albumin (AUROC=0.82).

Although urea and creatinine showed significant differences between survivors and non-survivors, they had poor discriminatory ability for mortality when adjusted for age. However, the mean values for non-survivors were only 10 mmol litre–1 and 115 µmol litre–1, respectively, reflecting the seriousness of small increases in our traditional measurements of renal function. It is already recognized that minor elevations in both urea and creatinine equate to adverse mortality and morbidity in patients with community-acquired pneumonia.17 Serum phosphate appears to be a promising predictor of outcome allowing for age, and this may be explained by phosphate being a surrogate marker of renal insufficiency. However, phosphate alone offers no clinical usefulness for predicting outcome, since the mean values and 95% confidence intervals for survivors and non-survivors were within the normal laboratory range.

Of course, the venous metabolic and biochemical markers investigated in this study are not diagnostic tools, and it is clearly important to diagnose the underlying cause of the metabolic abnormality. However, they appear to have a role in identification of the sick patient and as such may be usefully employed as a triage tool. The clinician should be aware that even minor abnormalities in these metabolic and biochemical markers signify the potential for deterioration and mortality and thus may warrant urgent action. Perhaps of greater importance are the trends in these variables, in that continued deterioration, particularly in the face of treatment, should herald more aggressive intervention. Neither clinical nor venous blood measurements were useful in predicting length of hospital stay. However, length of hospital stay is often a poor indicator of outcome because of its dependence on so many factors,18 and this may partially explain why this study failed to identify a useful discriminator. Although it is well recognized that members of certain groups, such as the elderly, may have a prolonged hospital stay, we still failed to find a hospital stay discriminator even when this population was excluded from the analysis.

Methods and limitations of this study
Assessing the ‘metabolic state’ of a patient can be difficult given the often complex nature of the problems encountered. Traditionally, this has utilized arterial blood gas analysis rather than venous analysis as presented here. This may explain some of the differences between this study and previous work addressing the critically ill.4 12 However, recent work involving 246 patients with acute illness found that a significant correlation exists between arterial and venous pH for patients in the A&E department.19 The authors concluded that the venous pH estimation is an acceptable substitute for an arterial estimation in the A&E department. Correlations have also been found to exist between arterial and venous metabolic indices in acutely ill patients20 and in those with uraemia or diabetic ketoacidosis.2123 Moreover, arterial blood gas sampling has significant associated morbidity and consequently is not routinely used in all hospital admissions. Venous puncture offers minimum morbidity and, as shown here, is performed on the majority of patients admitted via the A&E department. Thus assessment of metabolic derangement on a venous blood sample is a practicable and acceptable additional investigation.

Invariably, resources limited capture of data from all patients admitted during the busiest periods of the trial. However, it is very unlikely that this would have resulted in systematic bias. It is also difficult to assess the accuracy of the data at the extremes of the measured ranges owing to the relatively small number of patients presenting with such values. However, it could be argued that those presenting with metabolic and biochemical data only marginally outside the normal range may benefit most from interventional therapies before the onset of organ dysfunction, given that hitherto they may not have been identified as at significant risk. A further source of inaccuracy is that patients may have received medical intervention (e.g. intravenous fluids from paramedical staff) before assessment of observational and venous blood measurements. As discussed previously, this can affect several of the variables studied.13 However, this reflects the real-life clinical situation and therefore does not detract from the pragmatic conclusions of the study. A difference in mode and urgency of treatment of the admitted patient following transfer out of the A&E department may also influence the outcome variables, mortality and length of hospital stay. Of course, we could not control for this, but given the large number in our study population we would not expect systematic bias as a consequence.

A final limitation of an observational study of this kind is that although we have shown that these venous blood values may assist us in the A&E department as a triage tool, we do not know whether intervention aimed specifically at normalizing these values will improve mortality. It would seem unlikely that correcting these markers will universally ‘cure all ills’. At the very least, we would expect that further stratification would be required according to the cause of the metabolic derangement.


    Clinical implications
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Clinical implications
 Appendix
 References
 
We have shown that the use of venous blood markers in the A&E department may help to identify patients at risk of deterioration, although validation is required in an independent population. This is well established in the critically ill arena, although differences exist in the predictive abilities of the various metabolic and biochemical markers between these two populations. No conclusions can be drawn about the potential benefits that may be accrued from early intervention. However, others have addressed this question in individuals with circulatory collapse secondary to severe sepsis. Early and aggressive medical intervention demonstrated an impressive survival advantage when metabolic markers were used to assist in the identification of these patients. Indeed, the markers were used to guide the resuscitation process itself.1 It is hoped that future work will show that a similar strategy employed in a heterogeneous group of patients, such as those investigated in this study, will reduce mortality. The current provision of Level 1 and Level 2 facilities, which would be required to manage these patients, remains woefully inadequate in the UK and represents major resource implications for the National Health Service. However, based on current evidence,1 the provision of ‘metabolic resuscitation beds’ (Level 1 care) may well reduce mortality and also prove cost-effective.


    Appendix
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Clinical implications
 Appendix
 References
 
A complete description of the derivations of formulae described below is available in the literature.3

Anion gap observed:

Anion gap adjusted for abnormal albumin concentrations:

Normal albumin defined as 45 g litre–1.

Apparent strong ion difference (SIDapparent ):

where . Nanormal is defined as 142 mmol litre–1. Mg2+ was not measured in this study since its measurement is not usual clinical practice in the A&E department (assumed constant for calculation). Effective strong ion difference (SIDeffective):

where albumin=albuminmeasured(0.123xpH–0.631) and .

Strong ion gap (SIG):


    Acknowledgments
 
We gratefully acknowledge the cooperation of the nursing and medical staff of the Accident and Emergency Department and the Biochemistry Department at Worthing Hospital. We are grateful to Mr Derek Paterson-Moody of Morpheus I.T. for technical support, and Mrs Kathleen Durick and Mrs Pat Kerry for administrative support, in performing this study.


    References
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Clinical implications
 Appendix
 References
 
1 Rivers E, Nguyen B, Havstad S, et al. Early goal-directed therapy in the treatment of severe sepsis and septic shock. N Engl J Med 2001; 345: 1368–77[Abstract/Free Full Text]

2 Bland RD, Shoemaker WC, Abraham E, Cobo JC. Hemodynamic and oxygen transport patterns in surviving and nonsurviving postoperative patients. Crit Care Med 1985; 13: 85–90[ISI][Medline]

3 Fencl V, Jabor A, Kazda A, Figge J. Diagnosis of metabolic acid–base disturbances in critically ill patients. Am J Respir Crit Care Med 2000; 162: 2246–51[Abstract/Free Full Text]

4 Smith I, Kumar P, Molloy S, et al. Base excess and lactate as prognostic indicators for patients admitted to intensive care. Intensive Care Med 2001; 27: 74–83[CrossRef][ISI][Medline]

5 Bernardin G, Pradier C, Tiger F, Deloffre P, Mattei M. Blood pressure and arterial lactate level are early indicators of short-term survival in septic shock. Intensive Care Med 1996; 22: 17–25[ISI][Medline]

6 Waldau T, Larsen VH, Bonde J, Fogh-Andersen N. Lactate, pH, and blood gas analysis in critically ill patients. Acta Anaesthesiol Scand 1995; 39: 267–71[ISI]

7 Stewart PA. Modern quantitative acid–base chemistry. Can J Physiol Pharmacol 1983; 61: 1444–6[ISI][Medline]

8 Fencl V, Leith DA. Stewart's quantitative acid–base chemistry: applications in biology and medicine. Respir Physiol 1993; 91: 1–16[CrossRef][ISI][Medline]

9 Sirker AA, Rhodes A, Grounds RM, Bennett ED. Acid–base physiology: the ‘traditional’ and the ‘modern’ approaches. Anaesthesia 2002; 57: 348–56[CrossRef][ISI][Medline]

10 Gilfix BM, Bique M, Magder S. A physical chemical approach to the analysis of acid–base balance in the clinical setting. J Crit Care 1993; 8: 187–97[CrossRef][ISI][Medline]

11 Kellum AJ, Kramer DJ, Pinsky MR. Strong ion gap: a methodology for exploring unexplained anions. J Crit Care 1995; 10: 51–5[CrossRef][ISI][Medline]

12 Cusack RJ, Rhodes A, Lochhead P, et al. The strong ion gap does not have prognostic value in critically ill patients in a mixed medical/surgical adult ICU. Intensive Care Med 2002; 28: 864–9[CrossRef][ISI][Medline]

13 Morgan TJ, Venkatesh B, Hall J. Crystalloid strong ion difference determines metabolic acid–base change during in vitro hemodilution. Crit Care Med 2002; 30: 259–61[CrossRef][ISI][Medline]

14 UK Department of Health. Comprehensive Critical Care: a Review of Adult Critical Care Services 2000. Available online at www.doh.gov.uk/pdfs/criticalcare.pdf

15 Stenhouse C, Coates S, Tivey M, Allsop P, Parker T. Prospective evaluation of modified early warning score to aid earlier detection of patients developing critical illness on a general surgical ward. Br J Anaesth 2000; 84: 663P.

16 Ridley S. Severity of illness scoring systems and performance appraisal. Anaesthesia 1998; 53: 1185–94[CrossRef][ISI][Medline]

17 Anonymous. Guidelines for the management of adults with community-acquired pneumonia. Am J Respir Crit Care Med 2001; 163: 1730–54[Free Full Text]

18 Chumbley GM, Hall GM. Recovery after major surgery: does the anaesthetic make any difference? Br J Anaesth 1997; 78: 347–8.[Free Full Text]

19 Kelly A-M, McAlpine R, Kyle E. Venous pH can safely replace arterial pH in the initial evaluation of patients in the emergency department. Emerg Med 2001; 18: 340–2[CrossRef]

20 Gennis PR, Skovron ML, Aronson ST, Gallagher EJ. The usefulness of peripheral venous blood in estimating acid–base status in acutely ill patients. Ann Emerg Med 1985; 14: 845–9[ISI][Medline]

21 Gokel Y, Paydas S, Koseoglu Z, Alparslan N, Seydaoglu G. Comparison of blood gas and acid–base measurements in arterial and venous blood samples in patients with uremic acidosis and diabetic ketoacidosis in the emergency room. Am J Nephrol 2000; 20: 319–23[CrossRef][ISI][Medline]

22 Ma OJ, Rush MD, Godfrey MM, Gaddis G. Arterial blood gas results rarely influence emergency physician management of patients with suspected diabetic ketoacidosis. Acad Emerg Med 2003; 10: 836–41[Abstract/Free Full Text]

23 Brandenburg MA, Dire DJ. Comparison of arterial and venous blood gas values in the initial emergency department evaluation of patients with diabetic ketoacidosis. Ann Emerg Med 1998; 31; 459–65.[ISI][Medline]