Introduction

St James’ Hospital (SJH) is a tertiary referral centre for various specialties, but is on continuous call for emergency medical admissions. In 2002 there was a major reorganisation of acute care with the establishment of an Acute Medical Admissions Unit (AMAU). Following this system change, key quality markers including hospital length of stay; hospital emergency department ‘wait-times’ ¹, in-hospital mortality and readmissions were improved^2,3. Further improvement in the outcomes for unselected emergency medical admissions may be achieved by closer monitoring and early detection of deteriorating patient. Previous studies have shown that deterioration in the clinical status of patients can be anticipated by changes in physiological parameters⁴ and laboratory variables^5,6. Much work had been completed in this field, focusing on both physiological- and laboratory-derived variables, alone or in combination^6,7. This has led to the development of Early Warning Systems tools for ward based patients to identify patients at risk and prevent further deterioration^7,8.

Emergency medical patients have frequent blood and biochemistry analyses; on our electronic patient record, there is an automated calculation of the GFR value following each creatinine determination. A more robust estimate of the glomerular function can be made usingthe Modification of Diet in Renal Disease (MDRD) formula, which was devised in 1999.⁹ It uses six variables—age, race, gender, serum creatinine, urea and albumin levels. The MDRD formula to estimate GFR has not been validated in acute kidney injury⁹, however, standing alone; some of the variables used in these equations have been independently shown to predict mortality in an acute setting^10,11. We have previously demonstrated that admission eGFR can predict the in-hospital mortality in emergency medical admissions¹². Used in the latter context, the eGFR represents an aggregate predictive score with significant evidence for these^8,13,14. The aim of this study therefore, was to explore how the eGFR value might change during an acute hospital admission, and whether it could be used to predict in-hospital mortality. The setting was a study of all unselected acute medical emergencies admitted to a large university teaching hospital over a 10-year period between January 2002 and December 2011.

Methods

An anonymous dedicated patient database was created in 2002 to characterise emergency admitted patients, using components from the administration system (PAS), the emergency department database, and the hospital biochemical, haematology and microbiological databases; some discharge and procedural codes were derived from the hospital in-patient enquiry (HIPE) scheme (Table 1). The admission and discharge eGFR values were used to calculate the delta GFR value, the change during the hospital episode. The extended MDRD equation (MDRD Value = 170 x Sr. Creatinine–0.999 x Age–0.176 x [0.742 if Female] x [1.210 if Black] x BUN–0.176 x Albumin+0.318) was utilized. We categorised patients by any improvement in delta GFR (Group I) or of this variable over the clinical episode: deteriorations were of <= 5 (Group II), > 5 <=10 (Group III), 10 <=20 (Group IV), 20 <=40 (Group V) and > 40 ml/min-1/m² (Group VI). Mortality was any in-hospital death at a designated cut-offs; we examined the association of change in eGFR from admission, at designated cut-offs of 24 hr, 72 hr, 7 day, 14 and 30 days on mortality by that point. Descriptive statistics for baseline demographic data included, as appropriate, means / standard deviations (SD), medians / inter-quartile ranges (IQR) or percentages. Comparisons between categorical variables and mortality were with Chi-square tests. We examined the association between mortality at cut-offs (24 hr, 72 hr, 7 day, 14 day and 30 day) and the following predictor variables: age, change in eGFR over the hospital episode, the O2 saturation and troponin status (all at time of presentation in the Emergency Department).

Results

Renal function over time

The median GFR at admission was 75.3 ml/min/1.73m2 (IQR: 53.8, 97.7); the grouped data suggested an increase following admission being at 24 hr – 87.4 ml/min/1.73m2 (IQR: 67.2, 107.6), at 72 hr – 85.7 ml/min/1.73m2 (IQR: 64.7, 106.8), at 7 day – 80.8 ml/min/1.73m2 (IQR: 59.9, 101.8) and at 30 days – 72.5 ml/min/1.73m² (IQR: 52.9, 93.2). However, as patients had different lengths of hospital stay; direct comparisons must be done to establish whether on a paired basis patients dropped their MDRD value over time. Most patients showed small improvements in MDRD value; the change from baseline to discharge or death value were respectively +1.3, +2.6, +2.8 and +2.2 ml/min/1.73m² at the sample points of 24, 72 hr, 7 day and 30 days. Patients who died had a much lower GFR at admission 48.9 ml/min/1.73m2 (IQR: 30.4, 69.7) but showed a crucial difference with a substantial fall by day 2 – 33.5 ml/min/1.73m2 (IQR: 19.9, 59.4) with little further change thereafter – 3 day 36.3 ml/min/1.73m2 (IQR: 20.6, 64.7) - 7 day 38.4 ml/min/1.73m² (IQR: 20.1, 64.5) and 30 days – 40.2 ml/min/1.73m2 (IQR: 21.1, 64.1). For patients who died, the change from baseline to discharge or death value were respectively -3.9, -3.7, -4.2 and -7.5 ml/min/1.73m2 at the sample points of 24, 72 hr, 7 day and 30 days.

Changes in GFR by group and mortality risk (Figure 1)

We analysed the GFR value by the change over the clinical episode. We categorised patients based on the variation of GFR from admission in 6 groups: as any improvement in delta GFR (Group I) or a deterioration of one of five categories: <= 5 (Group II), > 5 <=10 (Group III), 10 <=20 (Group IV), 2 <=40 (Group V) and > 40 ml/min-1/m2 (Group VI). These groups respectively contained 53.6%, 23.7%, 8.2%, 7.7%, 5.0% and 1.8% of the patients with mortalities of 4.4%, 4.2%, 6.6%, 9.4%, 15.7 and 24.5%. The unadjusted OR for a 30-day death for the five groups with declines of <5, >=5 < 10, >=10 < 20, >=20 < 40 and >=40 ml/min/1.73m2, compared with the baseline of an improved GFR, were 0.96 (0.87, 1.06), 1.54 (1.35, 1.74), 2.74 (2.0, 3.52), 4.06 (3.64, 4.54) and 4.06 (3.64, 4.54) respectively.

Risk prediction based on lab data and change in GFR (Figure 2)

We identified an efficient model using only four predictors of 30-day outcome; these were age, change in GFR between admission and end of the hospital episode, troponin status (negative or positive) and O2 saturation level at time of admission. The fully adjusted Odds Ratios for each of the classes of GFR deterioration showed a stepwise increase in risk of 30-day death, compared with those showing no change or an improvement. The prediction was of a high order at each of the time points assessed of 24 hr, 72 hr, 7 days, 14 days and 30 days. The Area under the Receiver Operator Curve at each of these prediction time points were 24 hr 0.94 (0.92, 0.96), 72 hr 0.93 (0.91, 0.94), 7 days 0.91 (0.90, 0.92), 14 days 0.90 (0.89, 0.90) and 30 days 0.88 (0.87, 0.89). The fully adjusted for each deteriorating group vs. improvers at 30 days were OR’s of (Group II) 1.42 (95% CI: 1.20, 1.68), (Group III) 1.59 (1.27, 1.99), (Group IV) 2.71 (2.24, 3.27), (Group V) 5.56 (4.54, 6.81), and (Group V) 11.9 (9.0, 15.6) The model fit across each of the deciles assessed was good, as evidenced by the Hosmer-Lemeshow chi² statistic = 8.5 and a non-significant p value of 0.58 – indicating no significant deviation between the predicted and observed observations within each of the deciles (Figure 2).

Discussion

Acute kidney injury is not uncommon in emergency medical admission; what our data emphasises is that, for emergency medical admissions, homeostasis of the internal environment is the norm. Blood tests are routinely collected (often daily) in emergency hospitalised medical patients; therefore there is an opportunity to deploy these as a monitoring and an early warning system. A combination of markers, when formalized, then constitute an aggregate score system. Hucker et al suggested that a combination of clinical and biochemical measurements in the accident and emergency department proved the best predictors of hospital mortality based on the logistic regression analyses rather than each of them taken individually¹⁵. These results were supported by Prytherch et al and Asadollahi et al, both confirming the relationship between admission abnormal laboratory data and a predicted worse outcome in general medical patients^13,14. The difficulty is that these studies could require specific knowledge of different scoring system, complex calculation or additional adjustments for age of clinical variables in order to predict the outcome.

Others agree that there are major ramifications for changes in renal function in specific contexts and that even small alteration can have important consequences. Grigorian Shamagian et al demonstrated that renal dysfunction was a powerful predictor of mortality and morbidity in patients hospitalized for CHF, irrespective of left ventricular systolic function¹⁶. Similarly, Gottlieb et al and Smith et al demonstrated increase in mortality and length of stay with even very small increases in creatinine in patients admitted with congestive heart failure^17,18. The limitation of these studies is that they were performed on selected groups rather than with unselected medical emergencies. In contrast, this study was performed on an unselected group of patients, representing all emergency medical admissions over a 10 yr period. Therefore the findings can be applied to all medical patient groups. Nearly 80% of patients, the eGFR will either show no change or improve over the course of the admission. For the remaining 20% of patients, the drop in renal function is exponentially related to the risk of an in-hospital death. It is important to note that we are not considering patients with advanced or deteriorating renal function; as the average eGFR value on admission was 74 ml/min/1.73m² (IQR 54 – 98), a threshold cut-off of 5 ml/min/1.73m² (Group I) represents a change of 6.7% from baseline. This variation would be at the level that might be considered important in a clinical context.

There has been debate about whether absolute decline in renal function or relative decrease (for example 25%) would provide a better discriminator¹⁹. This in our view would be a high threshold as only 5.8% of our patients encountered such a deficit, but with a mortality of 50.7% and an OR of 15.5 (95% CI: 13.9 – 17.3) for a 30-day in-hospital death. Thus any significant deterioration in GFR, following an emergency medical hospital admission, must be considered a high-risk event and mandate closer monitoring.

We have previously shown that admission eGFR is a reliable tool to predict mortality following acute medical admission. This study expands on this work by demonstrating any deterioration in the admission eGFR value is an indicator of increased risk of death during that admission. The hope is that a move to standardise of National Early Warning Systems will provide a basis to detect early deterioration in perturbed physiology, initiate corrective action and further improve clinical outcomes²⁰. As patients frequently have biochemistry determinations during such an admission, algorithms could be devised to ‘track and trigger’ with such deterioration. The impact of an automatic laboratory based alerting system, providing feedback to the clinical care process, would be of great interest.

Correspondence: E Berzan
Division of Internal Medicine, St James’s Hospital, James’s St, Dublin 8
Email: [email protected]

Acknowledgements

Our consultant medical colleagues and the non-consultant members of the ‘on-call’ teams without which the AMAU initiative could not have been progressed; and the contribution of S Donnelly, her Clinical Nurse Managers and the ancillary professions related to medicine (SCOPE).

References

1. Moloney E, O'Riordan D, Silke B, Bennett K, Smith D: Impact of an acute medical admission unit on length of hospital stay, and emergency department 'wait times', Quarterly Journal of Medicine, 2005. 98: 283-9.

2. Rooney T, O'Riordan D, Silke B, Moloney E, Bennett K: Impact of an acute medical admission unit on hospital mortality: a 5-year prospective study, Quarterly Journal of Medicine, 2008. 101:457-65.

3. Scott I, Vaughan L, Bell D: Effectiveness of acute medical units in hospitals: a systematic review, International Journal for Quality in Health Care, 2009. 21:397-407.

4. Kause J, Parr M, Flabouris A, Hillman K, Smith G, Prytherch D: A comparison of antecedents to cardiac arrests, deaths and emergency intensive care admissions in Australia and New Zealand, and the United Kingdom--the ACADEMIA study, Resuscitation, 2004. 62: 275-82.

5. Froom P, Shimoni Z: Prediction of hospital mortality rates by admission laboratory tests, Clinical Chemistry, 2006. 52: 325-8.

6. Asadollahi K, Gill G, Hastings I, Beeching N: Laboratory risk factors for hospital mortality in acutely admitted patients, Quarterly Journal of Medicine, 100: 501-507.

7. Physicians RCo. National Early Warning Score (NEWS): Standardising the assessment of acute- illness severity in the NHS. Report of a working party. 2012 [cited; Available from: www.rcplondon.ac.uk/ national-early-warning-score

8. Smith G, Prytherch D, Schmidt P, Featherstone P. Review and performance evaluation of aggregate weighted 'track and trigger' systems, Resuscitation, 2008. 77:170-9.

9. Levey A, Bosch J, Lewis J, Greene T, Rogers N, Roth D. A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group, Annals of Internal Medicine, 1999. 130:461-70.

10. Lyons O, Whelan B, Bennett K, O'Riordan D, Silke B. Serum albumin as an outcome predictor in hospital emergency medical admissions. European Journal of Internal Medicine 2010. 21:17-20.

11. Whelan B, Bennett K, O'Riordan D, Silke B. Serum sodium as a risk factor for in-hospital mortality in acute unselected general medical patients, Quarterly Journal of Medicine, 2009. 102: 175-82.

12. Chin J, O'Dowd S, Wan Md Adnan W, Bennett K, O'Riordan D, Mellotte J. Using the MDRD value as an outcome predictor in emergency medical admissions. Nephrology Dialysis Transplantation, 2011. 26:3155-9.

13. Prytherch D, Sirl J, Schmidt P, Featherstone P, Weaver P, Smith G. The use of routine laboratory data to predict in-hospital death in medical admissions, Resuscitation, 2005. 66: 203-7.

14. Asadollahi K, Hastings I, Gill G, Beeching N. Prediction of hospital mortality from admission laboratory data and patient age: a simple model, Emergency Medicine Australasia 2011. 23:354-63.

15. Hucker T, Mitchell G, Blake L, Cheek E, Bewick V, Grocutt M. Identifying the sick: can biochemical measurements be used to aid decision making on presentation to the accident and emergency department, British Journal of Anaesthesia, 2005. 94: 735-41.

16. Grigorian Shamagian L, Varela Roman A, Pedreira Perez M, Gomez Otero I, Virgos Lamela A, Gonzalez-Juanatey J. [Renal failure is an independent predictor of mortality in hospitalized heart failure patients and is associated with a worse cardiovascular risk profile]. Revista Española de Cardiología, 2006. 59: 99-108.

17. Gottlieb S, Abraham W, Butler J, Forman D, Loh E, Massie B. The prognostic importance of different definitions of worsening renal function in congestive heart failure, Journal of Cardiac Failure, 2002. 8: 136-141.

18. Smith G, Vaccarino V, Kosiborod M, Lichtman J, Cheng S, Watnick S. Worsening renal function: what is a clinically meaningful change in creatinine during hospitalization with heart failure? Journal of Cardiac Failure, 2003. 9: 13-25.

19. Metr M, Nodari S, Parrinello G, Bordonali T, Bugatti S, Danesi R. Worsening renal function in patients hospitalised for acute heart failure: clinical implications and prognostic significance, European Journal of Heart Failure, 2008. 10: 188-195.

20. M. Jones, NEWSDIG: The National Early Warning Score Development and Implementation Group, Clinical Medicine, 2012. 12: 501-503.