Risk factors associated with biochemically detected and hospitalised acute kidney injury in patients prescribed renin angiotensin system inhibitors

Abstract

Aims

Therapy with angiotensin‐converting enzyme inhibitors (ACEi) and angiotensin receptor blockers (ARB) is a mainstay of treatment for heart failure (HF), diabetes mellitus (DM) and chronic kidney disease (CKD). These agents have been associated with development of acute kidney injury (AKI) during intercurrent illness. Risk factors for AKI in patients prescribed ACEi/ARB therapy are not well described.

Methods

We captured the incidence of AKI in patients commencing ACEi/ARB during 2009–2015 using anonymised patient records. Hospital‐coded AKI was defined from hospital episode statistics; biochemical AKI was ascertained from laboratory data. Risk factors for biochemically detected and hospitalised AKI were investigated.

Results

Of 61,318 patients prescribed ACEi/ARB, with 132 885 person years (py) follow‐up, there were 1070 hospitalisations with AKI as a diagnosis recorded and a total of 4645 AKI events, including AKI episodes indicated by biochemical KDIGO‐based creatinine change criteria. Incidence of any AKI event was 35.0 per 1000‐py, hospital‐coded AKI was 7.8 per 1000‐py and biochemical AKI was 33.7 per 1000‐py. Independent risk factors in a multivariable model for hospital‐coded AKI events were age, male gender, HF, diabetes, cerebrovascular disease, lower estimated glomerular filtration rate, socioeconomic deprivation, diuretic or non‐steroidal anti‐inflammatory use (all P < 0.001).

Conclusion

In patients prescribed ACEi/ARB, the highest risk of AKI is associated with conditions which are considered strong evidence‐based indications for their prescription. Socio‐economic status is an under‐reported risk factor for AKI with these agents. Strategies targeted at prevention of AKI may be of benefit, such as enhanced awareness based on higher risk comorbidities.

What is already known about this subject

• Therapeutic inhibition of the renin angiotensin system (RAS) has led to improvements in survival in patients with heart failure.

• RAS inhibition has been associated with increased risk of acute kidney injury (AKI).

• Patients at highest cardiovascular risk may be at higher risk of AKI due to additional comorbid factors.

What this study adds

• We found risk factors for AKI in >60,000 patients prescribed RAS inhibitors were age, heart failure, diabetes, chronic kidney disease and comorbidity.

• Patients with greatest benefit from RAS inhibition are also at risk of AKI.

• This association does not suggest causation; awareness of AKI is required in these patients.

1. INTRODUCTION

Therapeutic inhibition of the renin angiotensin system with angiotensin‐converting enzyme inhibitor (ACEi) and angiotensin receptor blocker (ARB) drugs is a mainstay of therapy for conditions associated with increased cardiovascular risk including hypertension, heart failure (HF), diabetes mellitus (DM) and proteinuric chronic kidney disease (CKD). This approach has been established following landmark clinical trials demonstrating the efficacy of these agents in improving outcomes.1, 2, 3, 4, 5, 6, 7, 8

There is evidence that patients taking renin angiotensin system inhibitors (RASi) with ACEi/ARB in combination with non‐steroidal anti‐inflammatory drugs (NSAIDs) and diuretics are at increased risk of AKI.9, 10 In addition, it is commonly reported that RASi are associated with acute kidney injury (AKI), particularly in the setting of impaired renal perfusion.9, 10 A commonly described scenario for this is when a patient becomes dehydrated, for example, due to a diarrhoeal illness11 and background RASi treatment leads to failure of regulation of angiotensin II‐dependent glomerular perfusion and renal hypoperfusion leading to AKI. However, the degree to which RASi are causal for AKI in this setting is debated.12 AKI in association with RASi therapy is common and the incidence is rising, either due to more widespread prescribing of these agents in patients at risk of AKI, or alternatively due to better awareness of AKI, including coding AKI as a diagnosis during hospitalisation.13, 14

AKI is common and, when severe, may be life threatening, often requiring hospitalisation and potentially acute dialysis. The overall mortality for acute kidney injury is high, with 1‐year survival less than 50% even in mild AKI.15 In patients who recover, it is common (but not invariable) for renal function not to return to the baseline level, especially in the presence of pre‐existing CKD.16, 17 Nevertheless, longer term studies show very poor survival up to one year after discharge.16 Strategies that can identify people with high risk of AKI or poor outcomes are needed.

AKI, defined by the KDIGO criteria for diagnosis of AKI, based on small changes in creatinine, is associated with poor short‐ and long‐term outcomes, irrespective of hospitalisation.18 Electronic alerts (e‐alerts) for biochemical AKI have been proposed as a mechanism for improving detection and management of AKI, although e‐alerts need to be combined with education and clinical support to alter outcomes.19, 20

By linking demographic, clinical, prescribing and biochemical sources of patient data from both primary and secondary care, it is possible to ascertain the incidence of, and risk factors for, AKI, including biochemical‐only AKI detected by blood tests, and episodes of AKI requiring hospitalisation. The aim of this study, using novel linkage of electronic patient data in an area of social deprivation and high rates of cardiovascular disease,21 was to describe the incidence and risk factors for AKI among patients prescribed RASi therapy.

2. METHODS

2.1. Cohort

Greater Glasgow and Clyde National Health Service (NHS) provides healthcare to a population of approximately 1.2 million. The NHS Greater Glasgow and Clyde “Safe Haven” is a secure environment whereby health data from different sources can be linked together and made available in de‐identified form for analysis. It has been ethically approved by the ethics committee of NHS Greater Glasgow and Clyde. This specific project was approved by the Local Privacy Advisory Committee of the Safe Haven.

All health episodes in Scotland are linked by the Community Health Index (CHI), a unique identifier for all patients. In this study, we used data related to NHS care, including patient records, prescribing, hospitalisations and laboratory testing. We defined patients included in the cohort as those incident users of RASi encashing at least one prescription for ACEi or ARB from the prescription information system (PIS), which captures data on all “cashed” prescriptions in NHS Greater Glasgow and Clyde from January 2009 to December 2015.22 We excluded patients with prevalent use of RASi prior to 9 January 2009 and patients with a prior diagnosis of cancer (excluding non‐melanoma skin cancer) at the time of commencement of RASi therapy. Patients entered the cohort at the date of their first prescription, and exited the cohort at death, at the last prescription date plus 30 days, to allow for a washout period, or at the end of the data extract from PIS. The makeup of the cohort is shown in Figure 1.

Figure 1.

Flow diagram showing how cohort was generated for analysis from electronic patient records

Comorbidities at baseline were defined by the presence of a diagnosis of hypertension, heart failure (HF), diabetes mellitus (DM), cardiovascular disease (CVD) or cerebrovascular disease (CeVD), from general practitioner (GP) electronic records (termed “local enhanced service” [LES]) data, as well as from ICD‐10 codes from prior hospital admission records at cohort entry. A Charlson co‐morbidity index was calculated for all patients.23 PIS records for all patients were used to identify additional treatment with diuretics (loop, thiazide and potassium sparing) and NSAIDs. Patients were also classified as having CKD based on this being recorded by their GP within LES coding. We defined baseline kidney function as the mean estimated glomerular filtration rate (eGFR) calculated using the CKD‐EPI formula in the year prior to cohort entry.24 Serum creatinine measures were isotope dilution mass spectrometry aligned.

The Scottish Government provides online calculators allowing use of patient postcode to generate divisions of socioeconomic deprivation, the Scottish Index of Multiple Deprivation (SIMD) (http://www.gov.scot/Topics/Statistics/SIMD). Using patient postcode, deprivation quintiles of deprivation status were calculated and categorised into most deprived (quintile 1) to least deprived (quintile 5).

2.2. Definition of acute kidney injury (AKI)

Hospital‐coded AKI was defined as a hospital admission with ICD‐10 code N17 in any diagnostic position on hospital discharge coding in the Scottish Morbidity Records 01 (SMR01). SMR01 collects data on all non‐obstetric, non‐psychiatric hospital discharges since 1968. Since 1989, SMR01 has been used to plan financial management of hospitals in order to ensure high completion rate. Internal audit of this data supports overall 89% accuracy for main condition diagnosis and similar or greater accuracy has been demonstrated in AKI in the United Kingdom.25, 26 As a secondary event of interest, we captured the incidence, stage and severity of AKI episodes not associated with a hospitalisation episode, based on all creatinine measurements for individual patients from the laboratory database during the period of exposure. These were categorised as “community‐based AKI”, where the creatinine measure used to define AKI was taken from a blood sample during a period that did not coincide with any hospital admission and “all AKI” which encompassed hospital‐coded AKI, community‐based AKI and “other AKI events”, i.e. AKI episodes occurring during a hospital admission but which were not coded by hospital coding data on hospital discharge. Biochemical AKI was defined using an algorithm aligned to the NHS e‐alert AKI warning system currently implemented in NHS England for routine health care and as previously reported.27, 28 AKI was diagnosed from the following criteria:

1. Serum creatinine ≥1.5 times higher than the median of all creatinine values 8–365 days ago.

2. Serum creatinine ≥1.5 times higher than the lowest creatinine within 7 days.

3. Serum creatinine >26 μmol/L higher than the lowest creatinine within 48 h.

If one or more of these criteria were met, AKI was attributed to that date/measurement. Our sources only resolved measurements to the level of date. When more than one value was recorded on a given day, the highest value was considered.

Severity of AKI was based on the KDIGO definition applied to this algorithm.29 For every AKI identified above, a staging was assigned per the following rules:

• Stage 1: Serum creatinine ≥1.5 and <2.0 times AKI baseline or ≥26.0 μmol/L increase above AKI baseline.

• Stage 2: Serum creatinine ≥2.0 and <3.0 times AKI baseline.

• Stage 3: Serum creatinine 3.0 times AKI baseline or ≥354 μmol/L increase above AKI baseline.

To avoid confounding by early changes in serum creatinine following instigation of ACEi/ARB therapy, we discounted serum creatinine measured <14 days following commencement of therapy. E‐alerts were not used in the laboratory systems during the period of this study. We identified deaths and date of death by linkage to the National Records Scotland death certificates (NRS).

2.3. Statistical analyses

The primary outcome was defined as the incidence of first AKI—either biochemical AKI detected in the community, or hospitalisation for AKI. Patients who died during follow‐up without experiencing an AKI event were censored at death. Kaplan–Meier survival curves were generated for time to first AKI in relation to: age, sex, SIMD, eGFR, diuretics use at baseline, NSAID use at baseline, use of diuretics or NSAID at baseline, prescription groups at baseline, and history of co‐morbidities, namely: hypertension, heart failure, diabetes, CKD, cerebrovascular disease and Charlson index of co‐morbidities. Univariable Cox proportional hazard models were fitted to obtain estimates of the association between each covariate and incident AKI, reported as hazard ratios with 95% confidence intervals. The proportional hazards assumption was assessed using Schoenfeld residuals, and the assumption was not met for several variables in each model. However, visual inspection of Kaplan–Meier plots suggested that these deviations were quite subtle, and the hazard ratios may be interpreted as giving the average association over the follow‐up period.

Multivariable Cox regression models were fitted to further analyse the associations between covariates and incident AKI. A manual backwards selection procedure was used, with all covariates (except for the Charlson index excluded on the basis it is comprised of multiple co‐morbidities being tested in the Cox model) considered in the starting model. Covariates were sequentially excluded based on the P‐value, to obtain a final model with all predictors making a significant contribution (at a 5% significance level) to the model. All other predictors were categorical. No adjustments were made for multiple comparisons. All analyses were carried out using the statistical software package R.30

2.4. Data availability

The data that support the findings of this study are not publicly available due to privacy or ethical restrictions. Further information on the handling of electronic health record data used in this study are available at: https://www.nhsggc.org.uk/about-us/professional-support-sites/nhsggc-safe-haven/about-the-safe-haven/.

3. RESULTS

3.1. Demographics of cohort and incidence of AKI

Figure 1 summarises how the cohort of incident RASi users was generated. During the study period 61 318 patients were prescribed ACEi/ARB. The mean age of the cohort was 59.8 years (SD 13.9), and 51.9% were male. A total of 3302 (5.4%) had HF, 8807 (14.4%) had diabetes, and the mean eGFR was 86.1 mL/min/1.73m² (SD 18.2). There were 7993 deaths during follow‐up.

During a median follow‐up of 1.92 years there were 1070 hospital‐coded AKI events, and 4483 biochemical AKI episodes. In total, 4645 patients had at least one AKI event during 132 885 person years (py) of follow‐up. Hospital‐coded AKI and biochemical AKI overlapped, but were not mutually exclusive, as 162 patients had a hospital‐coded AKI, without confirmatory biochemistry, where the patient had no available baseline kidney function tests. The incidence of all AKI events was 35.0 per 1000‐py, hospital‐coded AKI was 7.8 per 1000‐py and biochemical AKI was 33.7 per 1000‐py.

3.2. Risk factors associated with AKI

The patients at highest risk of AKI were those with most comorbidities. Data are presented on all AKI events in Table 1 (biochemical or hospital‐coded AKI) as the overall pattern of risk factors associated with AKI were similar for both biochemical and hospital‐coded AKI. Data on hospital‐coded and biochemical AKI are presented separately in Tables 2 and 3. On univariate analyses of the association between baseline characteristics and incident AKI, the risk of AKI events increased with increasing age, socioeconomic deprivation (Figure 2), lower eGFR, diuretic use, heart failure, diabetes, a diagnosis of CKD, cerebrovascular disease, or increasing Charlson co‐morbidity index. On univariate analysis there was no association with NSAID use in isolation (Table 2 and 3), and combined NSAID and diuretic use did not confer higher risk than diuretic use alone.

Table 1.

Patient demographics and incidence of all AKI events with associated hazard ratios for each variable on univariable analysis

		No. eligible	No. events	Person yearsfollow‐up	Event rate(per 1000‐py)	HR (95% CI); P‐value
All		61 318	4645	132 885.0	35.0	–
	Female	29 468	2302	62 080.5	37.1	–
	Male	31 850	2343	70 804.5	33.1	0.90(0.85,0.95); P < 0.001
Age	≤50	15 603	517	35 424.9	14.6	–
	51–60	16 554	752	37 818.5	19.9	1.36(1.22,1.52); P < 0.001
	61–70	13 984	1027	31 116.5	33.0	2.25(2.03,2.50); P < 0.001
	≥ 71	15 177	2349	28 525.1	82.3	5.50(5.00,6.05); P < 0.001
SIMD quintile	1 (most deprived)*	21 087	1895	45 699.5	41.5	1.61(1.46,1.76); P < 0.001
	2	10 203	872	22 126.9	39.4	1.53(1.38,1.70); P < 0.001
	3	7848	618	17 327.9	35.7	1.39(1.24,1.55); P < 0.001
	4	6829	458	15 261.7	30.0	1.17(1.03,1.32); P = 0.014
	5 (least deprived)	10 065	581	22 614.7	25.7	–
	Missing	5286	221	9854.3	22.4	0.84(0.72,0.98); P = 0.026
Baseline eGFR (mL/min)*	≤ 29	281	91	361.4	251.8	7.61(6.18,9.37); P < 0.001
	30–59	4271	973	7673.9	126.8	3.97(3.70,4.27); P < 0.001
	> 59	45 408	3147	101 000.3	31.2	–
	Missing	11 358	434	23 849.3	18.2	0.58(0.52,0.64); P < 0.001
Diuretics	No	45 862	2683	100 391.6	26.7	–
	Yes	15 456	1962	32 493.3	60.4	2.25(2.12,2.38); P < 0.001
NSAID	No	37 672	3067	87 595.3	35.0	–
	Yes	23 646	1578	45 289.7	34.8	0.96(0.90,1.02); P = 0.188
Prescription groups	None	28 534	1772	67 136.8	26.4	–
	NSAID only	17 328	911	33 254.8	27.4	1.00(0.92,1.08); P = 0.991
	Diuretics only	9138	1295	20 458.5	63.3	2.39(2.22,2.56); P < 0.001
	Diuretic + NSAID	6318	667	12 034.8	55.4	2.02(1.85,2.21); P < 0.001
Hypertension	No	46 939	2978	100 661.3	29.6	–
	Yes	14 379	1667	32 223.7	51.7	1.76(1.66,1.87); P < 0.001
Heart failure	No	58 016	3840	126 568.0	30.3	–
	Yes	3302	805	6316.9	127.4	4.12(3.81,4.44); P < 0.001
Diabetes	No	52 511	3702	113 698.8	32.6	–
	Yes	8807	943	19 186.2	49.1	1.51(1.40,1.62); P < 0.001
CKD	No	59 330	4169	129 127.3	32.3	–
	Yes	1988	476	3757.6	126.7	3.86(3.51,4.25); P < 0.001
Cerebrovascular disease	No	58 567	4226	12 7671.1	33.1	–
	Yes	2751	419	5213.9	80.4	2.37(2.15,2.62); P < 0.001
Charlson index	0	44 429	2200	99 182.5	22.2	–
	1	10 477	1117	21 975.7	50.8	2.27(2.11,2.44); P < 0.001
	2	4013	668	7910.3	84.4	3.74(3.43,4.08); P < 0.001
	3+	2399	660	3816.5	172.9	7.43(6.81,8.11); P < 0.001

Table 2.

Patient demographics and incidence of hospitalised AKI events with associated hazard ratios for each variable on univariable analysis

		No. eligible	No. events	Person years follow‐up	Event rate (per 1000‐py)	HR (95% CI); P‐value
All		61 318	1070	137 874.6	7.8	–
	Female	29 468	515	64 508.1	8.0	–
	Male	31 850	555	73 366.4	7.6	0.95(0.84,1.07); P = 0.373
Age	≤50	15 603	94	36 166.1	2.6	–
	51–60	16 554	147	38 790.8	3.8	1.46(1.12,1.89); P = 0.004
	61–70	13 984	211	32 339.4	6.5	2.51(1.97,3.21); P < 0.001
	≥ 71	15 177	618	30 578.2	20.2	7.86(6.32,9.76); P < 0.001
SIMD quintile	1 (most deprived)	21 087	431	47 760.9	9.0	–
	2	10 203	214	23 030.1	9.3	1.03(0.87,1.21); P = 0.729
	3	7848	141	18 018.9	7.8	0.87(0.72,1.05); P = 0.141
	4	6829	106	15 740.6	6.7	0.75(0.60,0.92); P = 0.007
	5 (least deprived)	10 065	128	23 242.0	5.5	0.61(0.50,0.74); P < 0.001
	Missing	5286	50	10 082.1	5.0	0.56(0.42,0.75); P < 0.001
Baseline eGFR (mL/min)	≤ 29	281	35	429.7	81.5	–
	30–59	4271	311	8509.1	36.5	0.45(0.31,0.63); P < 0.001
	> 59	45 408	620	104 633.5	5.9	0.07(0.05,0.10); P < 0.001
	Missing	11 358	104	24 302.2	4.3	0.05(0.04,0.08); P < 0.001
Diuretics	No	45 862	558	103 495.5	5.4	–
	Yes	15 456	512	34 379.0	14.9	2.76(2.45,3.11); P < 0.001
NSAID	No	37 672	729	90 829.8	8.0	–
	Yes	23 646	341	47 044.8	7.2	0.91(0.80,1.04); P = 0.170
Prescription groups	None	28 534	375	69 202.2	5.4	–
	NSAID only	17 328	183	34 293.3	5.3	1.00(0.83,1.19); P = 0.971
	Diuretics only	9138	354	21 627.5	16.4	3.02(2.61,3.49); P < 0.001
	Diuretic + NSAID	6318	158	12 751.5	12.4	2.31(1.92,2.79); P < 0.001
Hypertension	No	46 939	637	103 767.6	6.1	–
	Yes	14 379	433	34 107.0	12.7	2.06(1.83,2.33); P < 0.001
Heart failure	No	58 016	861	130 763.7	6.6	–
	Yes	3302	209	7110.8	29.4	4.50(3.87,5.23); P < 0.001
Diabetes	No	52 511	828	117 630.7	7.0	–
	Yes	8807	242	20 243.9	12.0	1.70(1.47,1.96); P < 0.001
CKD	No	59 330	897	133 647.4	6.7	–
	Yes	1988	173	4227.2	40.9	6.09(5.18,7.17); P < 0.001
Cerebrovascular disease	No	58 567	961	132 322.9	7.3	–
	Yes	2751	109	5551.7	19.6	2.72(2.23,3.31); P < 0.001
Charlson index	0	44 429	446	101 681.2	4.4	–
	1	10 477	262	23 213.8	11.3	2.58(2.22,3.01); P < 0.001
	2	4013	185	8567.4	21.6	4.97(4.18,5.90); P < 0.001
	3+	2399	177	4412.1	40.1	9.29(7.81,11.06); P < 0.001

Table 3.

Patient demographics and incidence of biochemical AKI events with associated hazard ratios for each variable on univariable analysis

		No. eligible	No. events	Person years follow‐up	Event rate (per 1000‐py)	HR (95% CI); P‐value
All		61 318	4483	133 045.2	33.7	–
	Female	29 468	2231	62 153.8	35.9	–
	Male	31 850	2252	70 891.4	31.8	0.89(0.84,0.94); P < 0.001
Age	≤50	15 603	507	35 432.1	14.3	–
	51–60	16 554	738	37 834.1	19.5	1.36(1.22,1.53); P < 0.001
	61–70	13 984	995	31 170.9	31.9	2.22(2.00,2.47); P < 0.001
	≥ 71	15 177	2243	28 608.1	78.4	5.33(4.84,5.87); P < 0.001
SIMD quintile	1 (most deprived)	21 087	1837	45 770.1	40.1	–
	2	10 203	841	22 143.5	38.0	0.95(0.87,1.03); P = 0.183
	3	7848	597	17 345.2	34.4	0.86(0.78,0.94); P = 0.001
	4	6829	442	15 273.6	28.9	0.72(0.65,0.80); P < 0.001
	5 (least deprived)	10 065	556	22 645.9	24.6	0.61(0.56,0.68); P < 0.001
	Missing	5286	210	9866.8	21.3	0.51(0.44,0.59); P < 0.001
Baseline eGFR (mL/min)	≤ 29	281	87	364.9	238.4	–
	30–59	4271	914	7742.4	118.1	0.51(0.41,0.64); P < 0.001
	> 59	45 408	3070	101 061.2	30.4	0.14(0.11,0.17); P < 0.001
	Missing	11 358	412	23 876.7	17.3	0.08(0.06,0.10); P < 0.001
Diuretics	No	45 862	2589	100 478.2	25.8	–
	Yes	15 456	1894	32 567.0	58.2	2.25(2.12,2.38); P < 0.001
NSAID	No	37 672	2960	87 714.4	33.7	–
	Yes	23 646	1523	45 330.8	33.6	0.96(0.90,1.02); P = 0.181
Prescription groups	None	28 534	1716	67 199.4	25.5	–
	NSAID only	17 328	873	33 278.8	26.2	0.99(0.91,1.07); P = 0.778
	Diuretics only	9138	1244	20 515.0	60.6	2.36(2.20,2.54); P < 0.001
	Diuretic + NSAID	6318	650	12 052.0	53.9	2.03(1.86,2.22); P < 0.001
Hypertension	No	46 939	2890	100 733.8	28.7	–
	Yes	14 379	1593	32 311.4	49.3	1.73(1.63,1.84); P < 0.001
Heart failure	No	58 016	3707	126 706.9	29.3	–
	Yes	3302	776	6338.3	122.4	4.10(3.79,4.43); P < 0.001
Diabetes	No	52 511	3577	113 820.9	31.4	–
	Yes	8807	906	19 224.3	47.1	1.50(1.39,1.61); P < 0.001
CKD	No	59 330	4037	129 251.1	31.2	–
	Yes	1988	446	3794.1	117.6	3.71(3.36,4.09); P < 0.001
Cerebrovascular disease	No	58 567	4083	127 816.4	31.9	–
	Yes	2751	400	5228.8	76.5	2.34(2.11,2.59); P < 0.001
Charlson index	0	44 429	2132	99 248.4	21.5	–
	1	10 477	1078	22 025.8	48.9	2.26(2.10,2.43); P < 0.001
	2	4013	633	7933.0	79.8	3.64(3.34,3.98); P < 0.001
	3+	2399	640	3838.0	166.8	7.38(6.76,8.07); P < 0.001

Figure 2.

Kaplan–Meier curves for incidence of any AKI after first prescription of ACEi/ARB by SIMD quintile

Risk of AKI was highest in three distinct (though overlapping) groups. Patients with heart failure had an incidence of hospital‐coded AKI of 29.4 per 1000‐py and biochemical AKI of 122.4 per 1000‐py. Similar figures were observed for patients with CKD, with hospital‐coded AKI incidence of 40.9 and biochemical AKI incidence of 117.6 per 1000‐py. Using Charlson index of 3 or more as a measure of greatest comorbidity identified a group at extremely high risk of both hospital‐code AKI and biochemical AKI (40.1 and 166.8 per 1000‐py respectively, Tables 2 and 3).

Using a multivariate Cox proportional hazards model, independent predictors of AKI were male gender, increasing age, increasing socioeconomic deprivation, diuretic use, NSAID use, history of heart failure, diabetes mellitus, eGFR, and history of cerebrovascular disease, as presented in Table 4. It is notable that male gender was associated with higher risk of AKI based on multivariable analysis, despite female gender being higher risk on univariate analysis.

Table 4.

Multivariable model for association between predictors and risk of any AKI event. Reference group for SIMD: SIMD 1(most deprived)

	Hazard ratio	95% CI	P‐value
Gender (male)	1.20	(1.13,1.29)	P < 0.001
Age at entry (per 10 year increase)	1.31	(1.27,1.35)	P < 0.001
SIMD 2	0.88	(0.81,0.96)	P = 0.003
SIMD 3	0.82	(0.75,0.91)	P < 0.001
SIMD 4	0.69	(0.62,0.77)	P < 0.001
SIMD 5 (least deprived)	0.56	(0.51,0.62)	P < 0.001
Diuretics (Yes)	1.32	(1.21,1.44)	P < 0.001
NSAID (Yes)	1.16	(1.06,1.27)	P = 0.001
History of heart failure (Yes)	2.57	(2.37,2.79)	P < 0.001
eGFR at baseline (per 10 units increase)	0.80	(0.78,0.82)	P < 0.001
History of diabetes (Yes)	1.35	(1.25,1.46)	P < 0.001
History of CEVD (Yes)	1.46	(1.32,1.63)	P < 0.001

4. DISCUSSION

This is one of the first reports using routine clinical data to quantify risk factors for AKI in ACEi/ARB users. Patients with increasing comorbidity were at highest risk of AKI, with many conditions associated with increased risk of AKI, including HF, DM and CKD—conditions where there is high‐grade evidence for using these drugs in accordance with national guidelines.31, 32, 33 Patients concomitantly prescribed diuretics and NSAIDSs are at greater risk of AKI, as observed by others.9, 10 Amongst incident patients prescribed ACEi/ARB medication, we found a similar rate of AKI episodes associated with a hospital admission to a previous cohort of ACEi/ARB users.14 The incidence of biochemical AKI based on internationally recognised creatinine change criteria was approximately twofold higher than has previously been reported for the general population in the Grampian region.34

4.1. Risk factors for both biochemical and hospital‐coded AKI

The recognition of association of AKI with prescription of ACEi/ARB therapy in the setting of relative hypovolaemia is well established.11, 35 However, using observational prescribing data combined with biochemical flagging and hospitalisation coding records for AKI, we identify patients at highest risk of AKI whilst prescribed these agents. Caution is required in interpreting any association between ACEi/ARB therapy and AKI as causal. Recent studies using national primary care data did not demonstrate higher risk of hospitalisation with AKI overall, or following common infections including gastroenteritis, among users of ACEi/ARB compared to other antihypertensives.36 Our data demonstrate that patients with the most compelling evidence‐based indications for prescription of ACEi/ARB therapy are those at highest risk of subsequent AKI. These indications include HF, CKD and/or DM with proteinuria where there are data from high‐quality randomised controlled trials (RCTs) suggesting benefit with these agents.1, 4, 5, 6, 8, 37, 38 The evidence for benefits of ACEi are most compelling in patients with HF and reduced left ventricular ejection fraction.8, 38 These agents are strongly recommended in the recent European Society of Cardiology (ESC) and American Heart Association (AHA) guidelines for management of HF.31, 32 However, despite our observation that AKI is common among those with HF on ACEi/ARB, AKI is described as “rare” in HF patients in the ESC guidelines31 and monitoring of renal function is seen as “good practice” with no mention of AKI in the AHA guidelines.32 Biochemically detected AKI events may represent natural fluctuations in serum creatinine occurring in patients with HF taking RASi often in combination with diuretic therapy. These changes may not represent “AKI” with any intrinsic renal damage and may simply reflect changes in serum creatinine in the setting of haemodynamic perturbation as glomerular perfusion pressure responds to changes in hydration status. Nevertheless, greater rises in serum creatinine in patients requiring ACEi/ARB therapy highlights a group of patients at greater mortality risk during follow up.39 Therefore, we would simply state that increased awareness of AKI in patients with HF is required and sensitivity is needed in interpreting AKI alerts in these patients.

There is evidence that these agents delay progression of proteinuric CKD and/or improve outcomes in patients with diabetes and albuminuria.5, 40 These results suggest that the groups potentially deriving most benefit from these agents are at highest risk of being hospitalised with AKI, albeit in observational data with no control group. The clinical significance of these biochemically detected AKI events is unclear, and further studies are required to determine if these events confer any longer‐term risk of decline in renal function. Whilst hospitalised AKI increases risk of subsequent CKD,41, 42 it is less clear whether subtler acute, transient declines in renal function lead to longer‐term renal risk.

4.2. Multi‐morbidity and AKI risk

In an ageing population with increasing comorbidity, the association of Charlson index and AKI episodes is concerning. These patients in this cohort were prescribed ACEi/ARBs based on evidence from RCTs, which were performed over 15 years ago. It should be recognised that in an ageing society where multi‐morbidity is more common, these RCTs may no longer be representative of many contemporary patients prescribed these agents. This was highlighted in a report from a similar population as our study, whereby 23.2% were classified as “multi‐morbid”.43 On the other hand, undertreatment of multimorbid patients with HF is likely to be associated with poor survival. Multi‐morbidity is common in patients from a socially deprived background.43 We observed that social deprivation status was associated with increased AKI risk. Therefore, AKI risk in patients prescribed ACEi/ARB therapy is associated with a cluster of interrelated risk factors, including number of comorbid conditions, concomitant therapy and socioeconomic deprivation.

The optimal strategy to address the risk of AKI in the community in patients taking ACEi/ARB is unknown. General Practitioners should be aware that the most comorbid patients are most at risk and need close monitoring, particularly during acute illness. Initiatives to tackle this problem are currently being investigated such as “sick day rules”, where patients taking these drugs are advised to stop them during acute illness. This strategy requires resources for patient education and to date is not supported by clinical evidence of efficacy.44, 45

4.3. Strengths and limitations of this study

The strengths of these analyses include a large sample size, with excellent coverage of the population studied, avoiding sampling biases. The results demonstrating similar risk factors for hospital‐coded and biochemical AKI suggest that our analysis methods for assessing influence of comorbid variables were robust and give a consistent message. We do not have a control group, so the incidence of AKI in patients with these comorbid conditions not prescribed ACEi/ARB is unknown. Defining the most appropriate patients to study as a control group is challenging. There would be biases in selecting a group of patients commenced on an alternative class of antihypertensive medication such as calcium channel blockers. This has been explored in other studies with only small increases of AKI incidence with ACEi/ARB in comparison to patients exposed to antihypertensive regimes not including ACEi/ARB.14 In the case of HF, it would be unusual not to be treated with ACEi/ARB therapy.

We acknowledge further limitations with these analyses. AKI without clinical symptoms may be diagnosed more frequently in patients having frequent blood samples (ascertainment bias). We are unaware of the indication for taking the blood sample leading to a record in the laboratory database. The incidence of biochemical AKI in untested patients is unknowable. Whilst we use the term “hospital‐coded AKI”, it is possible and indeed likely, that AKI was one of a number of diagnoses coded during a hospitalisation episode, rather than the sole diagnosis. There may be coding bias in either direction with hospital‐coded AKI, where AKI is added as a diagnosis in the absence of biochemical evidence, or where AKI was present but not recorded as a diagnosis on hospital discharge. The incidence of biochemically diagnosed AKI was particularly high in patients with CKD. Whilst these patients are likely to be at high risk of AKI, the use of an algorithm based on serum creatinine, may lead to a higher incidence of AKI related to how the algorithm diagnoses patients with an AKI event. It is possible that some of the co‐morbid conditions have not been coded in the health care records, leading to an under‐reporting of the Charlson co‐morbidity index. This can be seen with chronic kidney disease, where only 1989 subjects have been coded as having CKD by their General Practitioner despite 4553 patients having a recorded GFR < 59 mL/min/m² (which is likely to be consistent with CKD).

The prescribing records indicate that a patient collected a prescription, rather than took the prescribed medication. Although we describe CKD as an “indication” for therapy, we do not have proteinuria data and therefore it is unclear how strong this indication was for ACEi/ARB therapy and or whether proteinuria alters risk of AKI.

5. CONCLUSION

We describe associations of various clinical variables with increased risk of AKI in patients prescribed ACEi/ARB therapy. However, we do not describe these relationships as causal. The overwhelming evidence demonstrates that these medicines have heralded remarkable improvements in survival in patients with HF, in particular.8, 38

We demonstrate that older patients with heart failure, diabetes, CKD, lower socioeconomic status and prior stroke are at highest risk of AKI, both hospital‐coded and biochemically detected in the community. Biochemical AKI may serve as a risk marker for future adverse events. Further work is required to identify strategies to minimise risk of hospital‐coded and/or biochemical AKI in patients receiving therapy with these agents.

COMPETING INTERESTS

P.B.M. reports research funding from Boehringer Ingelheim, paid advisory boards from AstraZeneca and Vifor‐Fresenius, lecture fees from Novartis, Pfizer, Bristol Myers Squibb and travel support from Pharmacosmos. C.B. reports grants from the Medical Research Council, Economic and Social Research Council, NHS Grampian endowments, and National Institute for Health Research during the conduct of the study. S.S. was supported by a research training fellowship from the Wellcome Trust (102729/Z/13/Z). L.A.T. is funded by a Wellcome Intermediate Clinical Fellowship (WT101143MA).

The results presented in this article have not been published previously in whole or part, except in abstract format.

CONTRIBUTORS

P.B.M., L.A.T., C.B. and C.M. conceived the study. N.R., R.P. and A.M. analysed the data. All authors interpreted the data, drafted and approved the final manuscript.

Mark PB, Papworth R, Ramparsad N, et al. Risk factors associated with biochemically detected and hospitalised acute kidney injury in patients prescribed renin angiotensin system inhibitors. Br J Clin Pharmacol. 2020;86:121–131. 10.1111/bcp.14141