Incidence and impact on outcomes of acute kidney injury after a stroke: a systematic review and meta-analysis

Study eligibility was determined using a standardized form (Additional file 2). Two authors, JA and KN, independently screened the list of studies generated by the search, with disagreements resolved by a third author, CF. Titles and abstracts of all studies were screened before obtaining full text versions of relevant studies. To improve generalizability, studies were included if they were a case control or cohort (prospective or retrospective) study and had a sample size greater than 500 adult subjects hospitalized with either an acute ischemic or hemorrhagic stroke [17]. Included studies had a clear statement regarding the definition of acute kidney injury - creatinine values alone were not sufficient. Subarachnoid hemorrhage was not included in this systematic review in view of the different aetiopathophysiology.

Data was collected using a standardized proforma (Additional file 2) by JA and KN. The following study details were recorded: authors, year of publication, country of publication, type of study, clinical setting, sample size, patient characteristics (age, sex, ethnicity, and comorbidities), definition, type and severity of stroke and definition of AKI. Clinical parameters on admission, including serum creatinine and/or GFR, exposure to radiocontrast media, where specified and number of patients who developed AKI were recorded. Outcomes including mortality, disability, length of stay, re-stroke or cardiac events were also recorded.

Study quality assessment was performed independently by two authors, JA and CF using the Newcastle-Ottawa scale [18]. A maximum of 9 points can be allocated to a particular study based on quality of selection, comparability and study outcome (including follow up). Scores were defined as poor (0–3), fair (4–6) and good (7–9) [17].

Data synthesis, meta-analysis and statistical analysis were performed using Review Manager v5.3.5 software (The Cochrane Collaboration, UK) and StatsDirect v3.0 (StatsDirect Limited, UK). Meta-analysis of proportions was carried out using the Stuart-Ord (inverse double arcsine square root) method in a DerSimonian-Laird (random effects) model. The Odds Ratio (OR) with accompanying 95% confidence intervals (95% CI) were used to report individual and summary effect measures for dichotomous data. Chi squared tests for heterogeneity were performed to examine if the degrees of freedom were greater than the Cochran Q statistic, with α of below 0.05 considered to be statistically significant. In addition, the I² statistic was calculated to provide the estimated percentage of heterogeneity observed. I² values of 25%, 50% and 75% correspond to low, medium and high levels of heterogeneity. Any heterogeneity was further explored. A two-sided P value of < 0.05 was considered significant for all analyses.

A total of 6173 potentially relevant citations were identified (Fig. 1), of which 816 were duplicates. A further 5309 articles were excluded after review of title and abstract and an additional 40 excluded after full text review. The characteristics of the eight included studies are displayed in Table 1 [19‐26] with the study outcomes summarized in Table 2. All eight studies were published in the English language between 2007 and 2015. Seven studies were considered to be of good quality and one of fair quality. The eight studies provided data on 12,325,652 patients (range 897 to 7,068,334) from four countries (five from the US [20, 22‐24, 26], and one each from China [21], Greece [25] and Romania [19]). The US studies overwhelmingly had the largest sample sizes, with a total of 12,319,724 patients representing 99.9% of all included patients. Three of the US studies used Nationwide Inpatient Sample (NIS) data [23, 24, 26]. Although the same database was used, one study included only patients with ischemic stroke [23], one study included only patients with hemorrhagic stroke [24] and a third included only patients who sustained AKI requiring dialysis treatment (AKI-D) [26]. Therefore it was considered appropriate to include only the first two in the meta-analysis [23, 24]. Five out of eight studies used the International Classification of Diseases-9th or 10th Edition (ICD-9/1CD-10) coding to define AKI [21‐24] and AKI-D [26]. Only one of the studies reported data on AKI-D in addition to overall AKI incidence [24]. Two studies used the Acute Kidney Injury Network (AKIN) classification [20, 25] and one the Risk/ Injury/ Failure/ Loss/ End-stage (RIFLE) classification [19]. None used urine output criteria. Although there are some differences in the grading of AKI severity between these classifications, both define the absolute incidence of AKI as an increase in serum creatinine > 150%. Stroke was determined using ICD coding in four studies [22‐24, 26]. Three studies utilized the World Health Organization (WHO) definition of stroke [27] and extracted clinical data from medical records prospectively [21, 25] or retrospectively [19]. Two studies [21, 25] further subclassified the etiology of ischemic stroke using TOAST criteria [28]. One study utilized stroke registry data as well as clinical records and ICD coding [20].

Five studies followed patients until discharge from hospital [20, 22‐24, 26], one for 30 days [19], one from hospitalization up to one year [21] and one from 30 days up to 10 years [25] (Table 2). Two studies excluded patients with known CKD [23, 24] and two studies excluded patients with end-stage renal disease (ESRD) [20, 26]. Ischemic and hemorrhagic stroke patients were included in four of the studies [19, 20, 25, 26], ischemic stroke alone in three [21‐23] and hemorrhagic stroke alone in one study [24] (Table 2).

Nadkarni et al. reported the incidence of AKI-D only [26]. This was 0.15% in hospitalizations with acute ischemic stroke and 0.35% in intracranial hemorrhage, with an overall incidence of 0.5%. Saeed et al. reported an overall incidence of AKI-D of 1.7% [24].

Using the remaining seven studies, the pooled proportion of AKI as a percentage was 9.61% (95% CI 8.33–10.98) (Fig. 2) with an I² statistic of 99.8% indicating high heterogeneity. Excluding Lin et al. [21], which reported a much lower incidence of AKI than any other study (0.82%), made no difference to the heterogeneity (I² 99.9%).

The pooled incidence of AKI in the studies that utilized ICD coding to define AKI [21‐24] was 4.63% (95% CI 3.65–5.72%). Heterogeneity was high (I² 99.9%). Excluding Lin et al. [21], the pooled incidence of AKI increased to 6.46% (95% CI 5.18–7.86%) and heterogeneity remained high (I² 99.9%). Further excluding the study of hemorrhagic stroke hospitalizations [24], the pooled incidence of AKI remained similar (6.42%; 95% CI 4.07–9.27%) with high heterogeneity (I² 90.6%). In comparison, the pooled incidence in studies using creatinine-based AKI definitions [19, 20, 25] was 19.51% (95% CI 12.75–27.32%), again with high heterogeneity (I² 97.4%).

The pooled incidence of AKI in ischemic stroke [19‐23, 25] was 9.62% (95% CI 4.20–16.96%; I² statistic 99.5%). Excluding Lin et al. [21], AKI incidence increased to 12.45% (95% CI 4.96–22.70%) and heterogeneity remained high (I² 99.5%). Using studies with a coding definition for AKI, the pooled incidence was 4.05% (95% CI 1.06–8.86%; I² statistic 99.1%). Pooled incidence of AKI in studies utilizing creatinine-based definitions was 17.33% (95% CI 9.42–27.05%) with high heterogeneity (I² 97.6%).

The pooled incidence of AKI in hemorrhagic stroke [19, 20, 24, 25] was 19.17% (95% CI 7.75–34.15%). Heterogeneity was high (I² 99.0%). Excluding Saeed et al. 2015 [24], the only study in this group to use a coding definition for AKI, the pooled incidence increased to 24.50% (95% CI 18.03–31.61%). Heterogeneity decreased but remained high (I² 84.9%).

Factors associated with the development of AKI after multivariate analyses are shown in Table 2. Three studies explored risk factors for the development of AKI after stroke [19, 20, 25]. Older age [19], worse renal function on admission [19, 20, 25], ischemic heart disease [19], heart failure [19, 25] and higher National Institutes of Health Stroke Scale (NIHSS) score on admission [20, 25] were all found to be associated the development of AKI in stroke patients. Use of angiotensin-converting enzyme inhibitors (ACEi) and angiotensin II receptor blockers (ARBs) were only marginally associated with AKI (OR 1.004; 95% CI 0.993–1.058, P = 0.057) in one study [19]. One study also tested the association between contrast-enhanced computerized tomography (CT) and AKI and found no relationship [20].

None of the studies presented data on the adjusted rates of AKI associated with cerebral angiography, thrombolysis or any vascular intervention (mechanical thrombectomy, carotid stenting or endarterectomy).

Two studies [19, 25] examined the relationship between stroke type and risk of developing AKI after adjustment for confounders. Covic et al. [19] reported an OR of 2.50 (95% CI 1.42–4.41; P = 0.001) in hemorrhagic stroke and Tsagalis et al. [25] an OR of 2.02 (95% CI 1.34–3.04; P = 0.001) with lacunar stroke used as the reference.

Two studies found an association between stroke severity (as determined by NIHSS score) and the development of AKI [20, 25]. In Khatri et al. [20] the adjusted OR per 5 point increase in NIHSS score was 1.13 (95% CI 1.07–1.19; P < 0.001). Tsagalis et al. [25] reported an OR of 1.02 (95% CI 1.01–1.03; P = 0.020) after adjustment for age, sex, presence of atrial fibrillation (AF), serum glucose, hematocrit and antihypertensive agent use in the first 48 hours of admission.

Five studies reported disability post stroke with varying definitions. Two studies [21, 22] recorded degree of disability post stroke, as measured by the modified Rankin Scale (mRS). However data from Lin et al. [21] could not be analyzed with respect to AKI. Mohamed et al. [22] found no association between AKI and degree of disability after multiple adjustments. Two studies [23, 24] used coded discharge destination from NIS data as a surrogate marker for disability. Discharge was categorized as none to minimal disability and any other discharge status (home health care, short-term hospital or other facility including intermediate care and skilled nursing home or death) as moderate to severe disability. Both studies found a higher incidence of moderate to severe disability in patients with AKI after adjustment for multiple confounders (Saeed et al. 2014, OR 1.3 (95% CI 1.3–1.4, P < 0.0001); Saeed et al. 2015, OR 1.2 (95% CI 1.1–1.3, P < 0.0001)). A further study [26] utilized an ‘adverse discharge’ category to classify patients as being discharged to a nursing care facility, hospice or long-term care hospital. Here AKI-D was associated with increased odds of adverse discharge (adjusted OR 1.18, 95% CI 1.02–1.37, P < 0.01 for ischemic stroke, adjusted OR 1.74; 95% CI, 1.34–2.24; P < 0.01 for ICH) after adjusting for baseline demographics, hospital-level characteristics, Charlson comorbidity index and concurrent diagnoses.

Five studies collected data on length of stay (LOS) [20, 22‐24, 26]. All studies reported that AKI was associated with an increased length of stay ranging from 2 to 18 extra days spent in hospital (Table 2). In Mohamed et al. [22], this finding persisted after adjustment for age, NIHSS score, previous stroke and insurance status (no OR given, adjusted P < 0.0001).

Three studies analyzed crude inpatient costs using NIS data [23, 24, 26] and all showed AKI was associated with increased inpatients costs ranging from 14,139 to 49,827 US Dollars (Table 2).

One study [25] examined the relationship between AKI after a stroke and long-term cardiovascular events. The probability of having a composite cardiovascular event during the 10-year period was higher in the AKI group than the non-AKI group (cumulative probability 66.8 (95% CI 56.6–76.9) vs 52.7 (95% CI 48.5–56.1); P = 0.001). In a Cox multivariable regression, AKI was an independent predictor of new composite cardiovascular events at 10 years (hazard ratio 1.22; 95% CI 1.01–1.48, P < 0.05) after adjustment for hypertension, diabetes, stroke subtypes, brain edema on imaging and hematocrit.

Saeed et al. 2014 [23] found that patients with AKI were more likely to suffer a post-thrombolytic ICH (OR 1.4; 95% CI 1.3–1.6, P < 0.001) after multiple adjustments including age, sex, race/ ethnicity, hypertension, diabetes, AF, dyslipidemia, congestive heart failure, chronic lung disease, myocardial infarction, gastrointestinal bleeding, sepsis and nicotine dependence.

Six studies [19, 20, 23‐26] compared mortality in AKI versus non-AKI groups with all reporting increased mortality in patients who developed AKI. Two studies reported higher crude mortality rates associated with severity of AKI [19, 20].

Four studies reported in-hospital mortality [20, 23, 24, 26] and two reported 30-day mortality [19, 25]. The OR for all-cause in-hospital mortality in patients with AKI was 2.11 (95% CI 1.09–4.07) with high heterogeneity (I² 100%; Fig. 3). Excluding Saeed et al. 2015, a study of ICH only, the OR increased to 2.67 (95% CI 1.86–3.83) and heterogeneity decreased but remained high (I² 84%). The OR for all-cause 30-day mortality in patients with AKI was 3.13 (95% CI 1.20–8.19), again with high heterogeneity (I² 95%; Fig. 3).

Two studies [20, 23] provided data on in-hospital mortality after ischemic stroke with a pooled OR of 3.30 (95% CI 2.56–4.26) and low heterogeneity (I² 31%; Fig. 4). Two studies [20, 24] provided data on in-hospital mortality after ICH with a pooled OR of 1.40 (95% CI 1.37–1.43) and low heterogeneity (I² 0%; Fig. 4).

Nadkarni et al. reported an adjusted OR for in-hospital mortality in AKI-D of 1.30 (95% CI 1.12–1.48; P < 0.001) in ischemic stroke and 1.95 (95% CI 1.61–2.36; P < 0.01) in ICH [26]. Since this study included AKI-D only it was excluded from the meta-analysis. Saeed et al. 2015 reported a higher crude mortality rate in patients with AKI-D than AKI without dialysis (50.2% vs 28.4%, P < 0.001) [24].

Only one study provided long-term mortality data up to 10 years [25], demonstrating higher cumulative mortality in the AKI group at one year (34.6 vs 22.1 in non-AKI group) and 10 years (75.9 vs 57.7; P = 0.001). In a Cox proportional hazards model AKI was an independent predictor of 10-year mortality (hazard ratio 1.24; 95% CI 1.07–1.44, P < 0.01) after adjustment for confounders including sex, hypertension, hypercholesterolemia, smoking, systolic blood pressure, brain edema, hematocrit, antihypertensive agent use after the event and ACEi/ARB and statin use on follow-up. The probability of 10-year mortality also increased with severity of AKI.