a b s t r a c t

Background and objective: Quality management of Acute kidney injury is dependent on early detection, which is currently deemed to be suboptimal. The aim of this study was to identify combinations of variables as- sociated with AKI and to derive a prediction tool for detecting patients attending the emergency department (ED) or hospital with AKI (ED-AKI).

Design, setting, participants and measurements: This retrospective observational study was conducted in the ED of a tertiary university hospital in Wales. Between April and August 2016 20,421 adult patients attended the ED of a University Hospital in Wales and had a serum creatinine measurement. Using an electronic AKI reporting system, 548 incident adult ED-AKI patients were identified and compared to a randomly selected cohort of adult non-AKI ED patients (n = 571). A prediction model for AKI was derived and subsequently internally validated using bootstrapping. The primary outcome measure was the number of patients with ED-AKI. Results: In 1119 subjects, 27 variables were evaluated. Four ED-AKI models were generated with C-statistics rang- ing from 0.800 to 0.765. The simplest and most practical multivariate model (model 3) included eight variables that could all be assessed at ED arrival. A 31-point score was derived where 0 is minimal risk of ED-AKI. The model discrimination was adequate (C-statistic 0.793) and calibration was good (Hosmer & Lomeshow test 27.4). ED-AKI could be ruled out with a score of <2.5 (sensitivity 95%). internal validation using bootstrapping yielded an optimal Youden index of 0.49 with sensitivity of 80% and specificity of 68%.

Conclusion: A risk-stratification model for ED-AKI has been derived and internally validated. The discrimination of

this model is objective and adequate. It requires refinement and external validation in more generalisable settings.

(C) 2020

1. Background

Acute Kidney Injury is a global health issue complicating 4-20% of all hospital admissions with an incidence as high as 70% in crit- ically ill patients [1,2]. It is characterised by an abrupt loss of kidney function that is strongly associated with high mortality and morbidity [3,4]. In patients attending the University Hospital of Wales as emergen- cies with AKI, the 30-day mortality is 26% [5]. The majority of these pa- tients are admitted through the hospital front door either through the emergency department (ED) or the acute medical admissions unit.

In 2009 the National Confidential Enquiry into Patient Outcomes and Death (NCEPOP) in the United Kingdom reported that the recognition, management and documentation of AKI was gravely suboptimal espe- cially in those patients who died with an AKI [6]. NCEPOP made the

* Corresponding author at: Emergency Medicine Unit, The University of Hong Kong, China.

E-mail address: [email protected] (T.H. Rainer).

case that management was poor due to a poverty of accurate early diag- nosis and early intervention. The literature is replete with calls for early diagnosis and early intervention of AKI. The National Institute for Health and Care Excellence (NICE) recommends that risk assessment is made at the hospital front door especially therefore in emergency depart- ments [7]. Yet there are no validated bespoke risk tools for detecting AKI let alone high-risk AKI in the context of the emergency department. It is rare that AKI is immediately life-threatening. As a result it receives little attention from emergency physicians. Yet delays in detection and treatment result in an in-hospital mortality rate of 20-40% that is higher than acute stroke and acute myocardial infarction combined [7,8]. We have previously shown that more cases of index hospital AKI e-alerts occur in the emergency department (53%) than the rest of the hospital combined and that this is true of all three stages of AKI [8]. Also, 24% cases of AKI deteriorate significantly to a higher stage relative to the inci- dent creatinine measurement and the mortality in this group is 38.8% [5]. We have also shown that ED-AKI is an independent risk factor for mortal- ity and that in those patients who die with associated AKI, 58% occur

https://doi.org/10.1016/j.ajem.2020.12.017

0735-6757/(C) 2020

within 7 days of admission to hospital. There is little other, if any, specific data on the recognition, detection and treatment of AKI in the ED. How- ever, the NCEPOD report on hospital mortality primarily due to acute renal failure includes EDs [NCEPOD]. It judges that in pre-admission AKI, good clinical practice occurred in only 55% cases, assessment of risk factors was poor, subsequent delays in recognition occurred in 43% cases, that a fifth of post-admission AKI was predictable and avoidable and that complications were missed in over 50% cases.

The introduction of an electronic alert in Wales was an attempt to fa- cilitate early detection of AKI [8-10]. Once an elevated creatinine is iden- tified in the laboratory an e-alert is triggered. However, the e-alert is dependent on an appropriate urea and creatinine blood tests being re- quested. Further, the evidence is that the introduction of the e-alert makes little difference to outcomes [11].

We hypothesise that information that is easily obtained from adult patients at a triage assessment in an emergency department (ED) may be used to determine the risk of ED-AKI in patients and to guide the re- quest for ED Screening tests and early monitoring of bladder volume and urinary output and quality. The purpose of this retrospective, single- centre, observational study in an ED and acute medical unit (AMU) was to identify combinations of add-on tests to estimate the probability of ED-AKI and to produce a model for risk-stratifying such patients. Spe- cifically, we aimed to estimate the accuracy of individual variables at ED clinical presentation for predicting ED-AKI. Secondly, we aimed to de- rive and internally validate a prediction score for ED-AKI.

1. Methods
   1. Ethics and study design

Ethical approval has been obtained from Cardiff University and Car- diff and Vale UHB to conduct a retrospective, single-centre, study in the ED and AMU of the University Hospital of Wales, Wales, UK. Written consent was not required from all participants as the study was retro- spective, anonymized, complied fully with the Declaration of Helsinki and Good Clinical Practice Guidelines and was part of an ongoing Quality improvement project.

• 1. Setting and electronic reporting system

The study was conducted in the ED and AMU of the University Hos- pital of Wales which receive 300 and 20 new patients respectively each day and which are associated with the Cardiff and Vale Nephrology and Transplant directorate. This centre is a tertiary referral centre providing all aspects of renal services for patients in South East Wales.

• 1. Participants

The study cohort including basic characteristics, data collection, ep- idemiology, demographics and outcomes have been previously re- ported [5]. Out of 17,693 patients attending the ED, 548 incident adult ED-AKI patients were identified. A non-AKI control cohort of 600 pa- tients was generated from the 17,145 incident patients attending the ED with no AKI alert by random selection using the random number generating function.

• 1. Measurements

Twenty-seven variables collected by Research staff are shown in Table 1. They include patient characteristics, medical history, conven- tional risk factors including diabetes, hypertension, lipid profile, smoking history, family history and investigations which may be avail- able within an emergency department. We did not collect point of care venous blood gas, current medication, laboratory tests, or physiological observations, as these were not available in the administrative repository.

• 1. Definitions

The definition of AKI accord with the current ‘Kidney Disease: Im- proving Global Outcomes’ (KDIGO) guidelines [8]. Patients for whom the first e-alert was generated from a creatinine value measured in pri- mary care were classified as primary care AKI.

Pre-existing chronic kidney disease (PeCKD) was defined as an eGFR (calculated by modified MDRD formula) <60 ml/min/1.73m² derived from the baseline SCr [12]. ‘PeCKD and worsening eGFR’ was defined as a decline in eGFR >15% or a decrease in eGFR >5 ml/min/1.73m² (with a baseline eGFR <60 ml/min/1.73m²) [13].

• 1. Outcomes

The primary outcome measure was the number of cases with ED- AKI. ED-AKI is defined as occurring in the ED prior to discharge or admission.

• 1. Statistical analysis

The unit of analysis was the patient. Statistical analysis was carried out using MedCalc 18.5-64-bit for Windows XP/Vista/7/8/10 (MedCalc Software, Ostend, Belgium). Descriptive statistics are presented as me- dians and interquartile ranges, means and standard deviations or 95% confidence interval (95% CI) and numbers and percentages.

Initially univariate analysis was used to identify variables that may be used in the model. Univariate logistic regression was performed on all 27 variables in order to determine the unadjusted odds ratios. P values <=0.05 were considered statistically significant.

Models were then devised according to the available data along spe- cific steps of the patient pathway. Models 1 to 3 include data only avail- able at ED triage. Model 4 includes data available at triage and early in the clinical pathway.

Multivariate Cox proportional hazard modelling was used to determine the significance of variables for ED-AKI. Stepwise backward variable selection with significance level < 0.05 were considered statis- tically significant. Several combinations were tested depending on the patient’s place in the ED pathway and the availability of data.

For the most practical score a risk score was devised by rounding the raw regression coefficients to the nearest integer. The risk score was then estimated for each patient according to the results of the two diag- nostic tests. The coefficients of each of the variables were summed and rounded in order to develop a 31-point risk-score where 0 is lowest risk and 31 is highest. After performing logistic regression where the depen- dent variable was ED-AKI and the independent variable ‘risk score’, we re-assessed the calibration of the model.

The performance of the model to predict the risk of ED-AKI was es- timated by calibration and discrimination. Calibration was assessed using the Hosmer-Lemeshow goodness-of-Fit test. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC, c-statistic). The model accuracy was thus reported as sensi- tivity and specificity, goodness of fit, and AUC.

The model was validated internally using bootstrapping with 1000 itinerations each with 978 randomly selected samples. This generated an optimal sensitivity and specificity wit 95% confidence intervals. As this was a secondary analysis, an a priori sample size calculation was not performed.

1. Results

Descriptive results from the cohort have been previously reported [5]. Table 1 shows the univariate regression analysis with unadjusted odds ratios for all variables. The highest odds ratios were evident in older age groups with Multiple comorbidities and prior CKD.

Table 2 shows the accuracy of various models along the patient path- way to predict ED-AKI. Models vary from 71 to 73% overall accuracy and

Table 1

Univariate Cox regression analysis of prognostic factors for ED-AKI

No		Number of cases	Unadjusted Odds Ratio	95% CI	AUROC	P Value
	Demographic
1	Age, per year	1119	1.038	1.031-1.045	0.706	<0.0001*
2	Age group, per decade
	Reference: 18-30 years	125	–	–	0.708	–
	>30-40 years	101	1.105	0.574-2.127		0.7658
	>40-50 years	106	1.438	0.771-2.684		0.253
	>50-60 years	151	3.096	1.787-5.365		<0.0001*
	>60-70 years	174	6.404	3.736-10.977		<0.0001*
	>70-80 years	168	8.195	4.739-14.173		<0.0001*
	>80-90 years	226	8.143	4.826-13.742		<0.0001*
	>90 years	68	9.419	4.770-18.596		<0.0001*
3	Male Gender	1119	1.417	1.120-1.793	0.543	0.0037*
4	Reference: Female NH/RH Resident	1119	3.502	2.360-5.195	0.565	<0.0001*
5	Reference: Not NH/RH Resident Previous hospital admission in 30 days	1119	1.490	1.078-2.058	0.527	0.016*
6	Reference: No admission Pre-admission CKD 3a-5	1119	4.446	3.110-6.356	0.598	<0.0001*
	Reference: No CKD
7	Previous recent creatinine Reference: No recent creatinine	1119	2.422	1.893-3.099	0.604	<0.0001*
	Non-CKD Comorbidities
8	Liver Disease	1119	6.325	2.430-16.463	0.522	0.0002
9	Peripheral Vascular Disease	1119	3.932	1.782-8.680	0.519	0.0007*
10	Diabetes	1119	3.457	2.487-4.805	0.590	<0.0001
11	Dementia	1119	2.875	1.749-4.725	0.534	<0.0001*
12	Lung Disease	1119	2.716	1.829-4.034	0.549	<0.0001*
13	Hypertension	1119	2.570	2.011-3.283	0.612	<0.0001
14	Hyperlipidaemia	1119	2.257	1.707-2.985	0.575	<0.0001
15	Active Malignancy	1119	2.504	1.680-3.723	0.543	<0.0001
16	Cardiovascular Disease	1119	1.889	1.453-2.455	0.565	<0.0001*
17	Vasculitis	1119	1.996	1.054-3.779	0.512	0.0297
18	Number of comorbidities	1119			0.712
	6		17.701	2.041-153.523		0.0091
	5		10.621	5.156-21.878		<0.0001
	4		6.777	4.223-10.876		<0.0001
	3		6.212	4.111-9.387		<0.0001
	2		7.028	4.816-10.255		<0.0001
	1 Reference: 0		4.794	3.361-6.839		<0.0001

with areas under the ROC curve from 0.765 to 0.800. Calibrations range from 14.7 to 27.4.

Table 3 presents adjusted odds ratios, raw coefficients and a simpli-

fied score for the eight variables used in model 3. The highest adjusted

Table 2

Effect of add-on tests to C-statistics for six models for ED-AKI

C-statistic

? C-statistic P Value Hosmer

P Value R² Overall

	AUROC (95% CI)			& Lemeshow Test			Correct Classification
Model 1	0.765	–	<0.0001	19.827	<0.0110	20%	71.05%
	(0.740-0.790)
Model 2	0.773	0.008	<0.0001	14.707	0.0651	21%	71.13%
	(0.747-0.797)
Model 3	0.793	0.028	<0.0001	27.427	0.0006	24%	72.21%
	(0.774-0.822)
Model 4	0.800	0.015	<0.0001	25.249	0.0014	25%	72.74%
	(0.776-0.824)

Model 1 includes data available at ED triage (N = 1119). The six variables are age, gender, nursing/residential home status, previous hospital admission, previously known CKD, and cre- atinine test performed in the previous 30 days.

Model 2 includes data available at ED triage (N = 1119). The six variables are age group, gender, nursing/residential home status, previous hospital admission, previously known CKD, and creatinine test performed in the previous 30 days.

Model 3 includes data available at ED triage (N = 1119). This model includes eight variables which are age group, gender, nursing/residential home status, previous hospital admission, previously known CKD, and creatinine test performed in the previous 30 days, and two individual comorbidities – liver disease and diabetes.

Model 4 includes data available early in the ED patient pathway (N = 1119). The 10 variables in model 4 include nine from model 3 with the further addition of ED serum sodium available as point of care. The addition of baseline creatinine or baseline eGFR added no improvement in discrimination to the model.

R² – Cox & Snell.

Table 3

Adjusted odds ratios, raw coefficients and the score chart for Model 3 for predicting ED- AKI (N = 1119)

Variable	Adjusted Odds ratio	Raw Coefficient	Score?
Liver disease	7.43	2.01	/8
Age Group			/7
Age Group > 90 years	5.95	1.78	7
Age Group > 80–90 years	5.03	1.62	6
Age Group > 70–80 years	4.95	1.60	6
Age Group > 60–70 years	3.86	1.35	5
Age Group > 50–60 years	2.10	0.74	3
Diabetes	2.51	0.92	/4
Nursing/Residential Home	2.14	0.76	/3
Known CKD	2.05	0.72	/3
Creatinine within 30 days	2.01	0.70	/3
Previous hospital admission	1.67	0.51	/2
Male Gender	1.44	0.36	/1
Maximum		7.76	31

* Score developed from raw coefficient rounded to nearest 0.25.

100

80

60

Sensitivity

40

20

Score_Total

odds ratios are for patients with known liver disease, older age groups and diabetes. Model 3 is the preferred model because it is the simplest to use, generates the highest C-statistic, utilises information that should be available on first arrival at triage in the ED, does not depend on blood tests and has 95%CIs that overlap with model 4. Model 4 generates mar- ginally better overall accuracy but is dependent on a point of care so- dium blood test in the ED. Thus, there may be a time delay to identifying high-risk AKI patients.

Table 4 shows the predictive analyses for model 3. The most bal- anced probability for ED-AKI is >6 out of 31. A sensitivity of 97% can be achieved with scores <2, and a specificity of 95% can be achieved with score > 13. Fig. 1 shows the AUROC curve for model 3.

Table 5 (Appendix) shows the summary table for sensitivity and specificity after bootstrapping. After bootstrapping, the Youden index was 0.4878 (95%CI 0.4369 to 0.5376) at an optimal cut off at >6 (95% CI >6 to >6) giving a sensitivity of 80.5% and specificity of 68.3%.

Fig. 2 shows the probability of ED-AKI based on the 31-point scoring method. At a score of 0 the probability of ED-AKI is <5%. At score < 2 the probability of ED-AKI is 12%. The prediction strength of the model levels off at a score > 15, where the probability of ED-AKI is 80%.

There is a positive correlation between the model 3 score and peak creatinine (r = 0.415, 95%CI 0.365-0.462, p < 0.0001). There was no correlation between the score and AKI stage I to III.

1. Discussion

In this study we have identified variables that are associated with ED-AKI and that are easily obtained in an ED. We have derived four

Table 4

Predictive analyses for Model 3 for ED-AKI (N = 1119)

0

0 20 40 60 80 100

AUC = 0.793

P < 0.001

100-Specificity

Fig. 1. AUROC Curve for Model 3 Score.

models and identified one (Model 3) that is most practical and appro- priate for the ED-triage setting. We have validated the model internally using bootstrapping and found it to be adequate for discrim- ination and to have good calibration. This is an objective model and a first step towards the development of a pragmatic test to guide early assessment.

This study evaluated 27 variables that are readily available ei- ther at ED triage or soon after a point of care blood gas assessment. It involved a complete group of consecutive AKI cases determined from a high-quality e-alert system. The control group from the same time period was too large for comparable evaluation so randomised selection has been used to select a similar and man- ageable cohort for assessment. The randomised selection provides a fair representation of non-AKI ED cases. It slants the proportion of the control group but provides a reasonable assessment for risk-stratification.

It may be argued that all patients who are likely to have ED-AKI

will receive an early blood test anyway. In this case a working pre- diction model is interesting but unnecessary. In our clinical experi- ence less than 30% adults attending our ED had a creatinine test. As a result, AKI may be missed and even when discovered may go undertreated. It is currently unclear to what extent over- investigation or under-investigation of renal blood tests occur in the ED. Both our previous data [8] and the NCEPOD report suggest that over 40% cases of AKI are missed early and a proportion are

likely to fall in the domain of the ED. The diagnosis is often made

To Rule OUT AKI

Sensitivity (95%CI)

Specificity (95%CI)

Positive likelihood ratio (95%CI)

Negative likelihood ratio (95%CI)

late and consequently treatment is delayed. AKI also has a 30-day mortality that is higher than acute myocardial infarction, stroke or sepsis [5,8,14].

We have derived and compared six potential models based on the

Score < 2 97

(95-99)

Balanced probability of AKI

Score > 6 To Rule IN AKI	80 (77-84)	68 (64-72)	2.5 (2.2-2.9)	0.29 (0.2-0.3)
Score > 13	17 (14-21)	95 (93-97)	3.8 (2.5-5.7)	0.87 (0.8-0.9)

33

(29-37)

1.5

(1.4-1.5)

0.09

(0.05-0.1)

availability of data in a typical ED. The first four models using data avail- able early in the ED pathway have similar correct classification at 70%, good calibration and adequate but not excellent discrimination. Models 1 and 2 are the simplest and most practical. However, with a little fur- ther interrogation Model 3 achieves the most optimal discrimination at triage and is our preferred model. This requires no blood tests and there is no need to review past records. As an early screening test, it is most appropriate.

120

y = 0.229 + 9.257 x + -0.258 x2

n = 20

R2 = 0.93; P < 0.001

100

80

Percentage_Probability_of_ED_AKI

60

40

20

0

-20

0 5 10 15 20

ED_AKI_Score

Fig. 2. Probability of AKI based on Score from Model 3. This shows a least squares regression curve based on the probability of ED-AKI at each score from 0 to 19 in the 31-score model. There were very few patients with scores above 19 so these have been grouped into score ’19’. The trend was not linear so a least squares regression curve was chosen to present the data. R² 0.93. The coefficient for the slope is 4.4 (95%CI 3.5 to 5.3; p < 0.0001). Coefficient of Determination R2 = 0.85. F-ratio = 102.2; P < 0.0001.

For departments that have point of care sodium at triage, or on-site blood gas analysis, model 4 has further advantages and better discrimi- nation. The lower level of the 95% CI for the AUROC is also greater than the AUROC for model 3 suggesting that the improvement is at least sta- tistically significant.

These prediction models improve the objective accuracy for

predicting AKI at triage. However, although they raise the probability of AKI they do not confirm it. The diagnosis still depends on requesting a creatinine test. They do, however, reduce the risk that AKI would be missed by not requesting a blood test. The early diagnosis of AKI could be markedly enhanced by point of care creatinine tests. Current practice in most EDs relies on creatinine blood samples being sent to the central hospital laboratory for analysis. This process ensures high quality results and low cost. However, there is a potential delay of at least one to two hours before a diagnosis and therefore treatment can be initiated.

One further role for predicting risk is making sure clinicians scruti- nise results. Routine urea and electrolyte tests are often overlooked. Predicting risk makes sure that clinicians are more likely to chase the re- sults of requested investigations. This in turn makes early interventions in terms of medication review, fluid balance more likely.

The diagnosis of AKI is not just a change in biochemistry. Apart from

urea and creatinine blood tests the diagnosis may be made based on urine output. Whilst most patients in the ED may be evaluated for serum creatinine, few have their bladder volume or urine output measured. Urine output usually falls prior to changes in creatinine. Monitoring bladder volume and urine output could be an early warning sign for AKI and not just a late investigation for fluid balance.

Whilst this was a single-centre, retrospective study, it provides the

largest cohort of ED cases so far addressing patient assessment for AKI and risk-stratification at the hospital front door and in the emergency department. It has been conducted over a five-month period and in- cludes all AKI e-alert cases. The only other study addressing risk of AKI at the hospital front door has been conducted by the RISK investigators which provides a generalizable group assessed over a single 24-h period with a small number of hospital AKI cases.

Future studies should consider several important points. The study does not include medication history, physiological observations such as a NEWS score, physical assessment, laboratory blood in- vestigations, urine output or urine sodium analysis all of which may yield important risk-stratification information and could im- prove the discrimination of the model. However, the focus of this assessment is the triage station where lengthy and complex assessments are unwarranted. Consideration should be given to excluding certain patients who are obviously going to have a cre- atinine requested anyway, such as patients with critical illness.

We have generated a model that is simple and easily accessible for every patient. The fact that it is easy to use suggests that it is more likely to be adopted into practice. The more complex the Risk assessment tool and the more variables that are needed the less likely it is that it will be done at all. By the very nature of the time it takes to get all the information the object is defeated which is to get a very quick guide early on in the patients journey through A&E. An accurate medication history was not possible from the avail- able information but should be considered in future refinements and risk-stratification models.

1. Conclusion

A risk-stratification model for ED-AKI has been derived and inter- nally validated. This model is acceptable and objective but requires fur- ther refinement and external validation.

Author contributions

D.A.F., S.Z., T.H.R. and A.O.P. designed the study; D.A.F. and S.P. car- ried out the data collection; D.A.F., T.H.R. and A.O.P. analysed the data;

T.H.R. made the figures; D.A.F., S.P., S.Z., T.H.R. and A.O.P. drafted and re- vised the paper; all authors approved the final version of the manuscript.

Declaration of Competing Interest

The authors have no financial interest to declare and no conflict of interest disclosures.

Appendix A. Appendix

Table 5

Summary Table for estimated sensitivity and specificity for Model 3 for ED-AKI (N = 1119)

Estimated specificity at fixed sensitivity

Sensitivity Specificity 95% CI a Criterion

80.00	68.61	63.19 to 73.33	>6.05
90.00	50.65	44.58 to 56.39	>3.95
95.00	39.01	32.55 to 44.38	>2.71
97.50	30.58	22.51 to 36.88	>0.97
99.00	19.06	13.31 to 25.17	>0.22
Estimated sensitivity at fixed specificity
Specificity	Sensitivity	95% CI a	Criterion
80.00	60.09	52.44 to 65.79	>8.61
90.00	35.78	24.73 to 44.39	>10.66
95.00	18.35	12.39 to 23.92	>12.87
97.50	10.67	0.00 to 0.00	>14.55
99.00	7.38	0.00 to 0.00	>15.89

^a BC_a bootstrap confidence interval (1000 iterations; random number seed: 978).

References

1. Zeng X, McMahon GM, Brunelli SM, Bates DW, Waikar SS. Incidence, outcomes, and comparisons across definitions of AKI in hospitalized individuals. Clin J Am Soc Nephrol. 2014;9(1):12-20. https://doi.org/10.2215/CJN.02730313.
2. Thomas ME, Blaine C, Dawnay A, Devonald MA, Ftouh S, Laing C, et al. The definition of acute kidney injury and its use in practice. Kidney Int. 2015;87(1):62-73. https:// doi.org/10.1038/ki.2014.328.
3. Lafrance JP, Miller DR. Acute kidney injury associates with increased long-term mor- tality. J Am Soc Nephrol. 2010;21(2):345-52 ASN.2009060636 [pii] https://doi. org/10.1681/ASN.2009060636.
4. Lameire N, Van Biesen W, Vanholder R. Acute renal failure. Lancet. 2005;365(9457): 417-30 S0140-6736(05)17831-3 [pii] https://doi.org/10.1016/S0140-6736(05)1

7831-3.

1. Foxwell DA, Pradhan S, Zouwail S, Rainer TH, Phillips AO. Epidemiology of emer- gency department acute kidney injury. Nephrology (Carlton). 2020;25(6):457-66. https://doi.org/10.1111/nep.13672.
2. National Confidential Enquiry into Patient Outcome and Death [NCEPOD] Report Acute Kidney Injury: Adding Insult to Injury. 2009. available at www.ncepod.org.uk
3. National Institute for Health and Care Excellence. Acute kidney injury: prevention, detection and management, London. Available at; 2019https://www.ncbi.nlm.nih. gov/pubmed/31909927.
4. Holmes J, Rainer T, Geen J, Roberts G, May K, Wilson N, et al. Acute kidney injury in the era of the AKI E-alert. Clin J Am Soc Nephrol. 2016;11(12):2123-31. https://doi. org/10.2215/CJN.05170516.
5. Holmes J, Roberts G, Geen J, Dodd A, Selby NM, Lewington A, et al. Utility of elec- tronic AKI alerts in intensive care: a national multicentre cohort study. J Crit Care. 2018;44:185-90. https://doi.org/10.1016/j.jcrc.2017.10.024.
6. Phillips D, Young O, Holmes J, Allen LA, Roberts G, Geen J, et al. Seasonal pattern of incidence and outcome of Acute Kidney Injury: A national study of Welsh AKI elec- tronic alerts. Int J Clin Pract. 2017;71(9)doi:https://doi.org/10.1111/ijcp.13000.
7. Wilson FP, Shashaty M, Testani J, Aqeel I, Borovskiy Y, Ellenberg SS, et al. Automated, electronic alerts for acute kidney injury: a single-blind, parallel-group, randomised controlled trial. Lancet. 2015;385(9981):1966-74 S0140-6736(15)60266-5 [pii] https://doi.org/10.1016/S0140-6736(15)60266-5.
8. Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro 3rd AF, Feldman HI, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150(9): 604-12http://www.ncbi.nlm.nih.gov/pubmed/19414839.
9. Wonnacott A, Meran S, Amphlett B, Talabani B, Phillips A. Epidemiology and out- comes in community-acquired versus hospital-acquired AKI. Clin J Am Soc Nephrol. 2014;9(6):1007-14 CJN.07920713 [pii] https://doi.org/10.2215/CJN.07920713.
10. Holmes J, Geen J, Phillips B, Williams JD, Phillips AO, Welsh AKISG. Community ac- quired acute kidney injury: findings from a large population cohort. QJM. 2017; 110(11):741-6. https://doi.org/10.1093/qjmed/hcx151.