Visual Abstract
Abstract
Background There are few observational studies evaluating the risk of AKI in people with type 2 diabetes, and even fewer simultaneously investigating AKI and CKD in this population. This limits understanding of the interplay between AKI and CKD in people with type 2 diabetes compared with the nondiabetic population.
Methods In this retrospective, cohort study of participants with or without type 2 diabetes, we used electronic healthcare records to evaluate rates of AKI and various statistical methods to determine their relationship to CKD status and further renal function decline.
Results We followed the cohort of 16,700 participants (9417 with type 2 diabetes and 7283 controls without diabetes) for a median of 8.2 years. Those with diabetes were more likely than controls to develop AKI (48.6% versus 17.2%, respectively) and have preexisting CKD or CKD that developed during follow-up (46.3% versus 17.2%, respectively). In the absence of CKD, the AKI rate among people with diabetes was nearly five times that of controls (121.5 versus 24.6 per 1000 person-years). Among participants with CKD, AKI rate in people with diabetes was more than twice that of controls (384.8 versus 180.0 per 1000 person-years after CKD diagnostic date, and 109.3 versus 47.4 per 1000 person-years before CKD onset in those developing CKD after recruitment). Decline in eGFR slope before AKI episodes was steeper in people with diabetes versus controls. After AKI episodes, decline in eGFR slope became steeper in people without diabetes, but not among those with diabetes and preexisting CKD.
Conclusions Patients with diabetes have significantly higher rates of AKI compared with patients without diabetes, and this remains true for individuals with preexisting CKD.
Type 2 diabetes (T2D) is one of the leading causes of CKD and ESKD worldwide.1 A large proportion of patients who develop CKD experience prior episodes of AKI, with evidence suggesting that kidney function does not fully recover after the AKI event.1 Moreover, CKD is a well-known risk factor for AKI, with recent studies suggesting there is a considerable overlap between the pathophysiology underlying the two conditions.2 However, the relationship is likely to be complex and remains poorly understood.
T2D has been reported as an independent risk factor for AKI in previous observational studies,3,4 and progressive decline in kidney function has also been well described in this population.1 Both AKI and CKD have been identified as risk factors for cardiovascular disease,5 which is the most frequent complication in T2D. Despite the increased access to routinely collected healthcare data, there are few observational studies evaluating the risk of AKI in people with T2D,6,7 and even fewer simultaneously investigating AKI and CKD in this population.1 As a result, there is a limited understanding of the interplay between AKI and CKD in people with T2D, and how this compares with the nondiabetic population.8 Previously, quantification of AKI from routine healthcare data were limited to the use of hospitalization and death data using International Classification of Diseases (ICD) coding.6 More recently, the Kidney Disease Improving Global Outcomes (KDIGO) definition for AKI, based on changes in serum creatinine (SCr), has been universally adopted, enabling a more uniform approach.9,10 However, this approach comes with its challenges, which mainly relate to the application of the KDIGO definition. In clinical practice, AKI can only be identified when previous tests within a time window are available for comparison, which may not be the case when blood testing is infrequent. To overcome this, various time windows to define baseline creatinine have been proposed, including the use of both prior and postindex values.11⇓⇓⇓⇓–16 Despite the numerous definitions, the variation in the intensity of blood sampling may still lead to misclassification between AKI and CKD.10 This highlights the importance of accurate definitions for both AKI and CKD that can be used in database studies to help understand the contribution of AKI to CKD and CKD progression, and the risk of developing AKI in patients with CKD.
The aim of this study was to develop an algorithm to examine rates of AKI in patients with and without T2D depending on CKD status using routinely collected healthcare data, and to investigate whether the association between AKI on GFR decline is different in people with T2D compared with people without diabetes.
Methods
Study Population
The design is a retrospective, cohort study of people from the Tayside region of Scotland (n=402,641 on January 1, 2012), which represents about 8% of the Scottish population. People with and without T2D that were matched by age, sex, and general practice were recruited in the Genetics of Diabetes Audit and Research in Tayside Study (GoDARTS) from December 1998 to October 2012; recruitment occurred either at diabetes or eye-screening clinics or through their general practitioner.17 About 50% of the patients with T2D at that time from the Tayside region were recruited into GoDARTS.8 Participants attended a clinic at recruitment, where a serum sample was collected to allow a number of routine biochemical measures to be tested. Recruitment was treated as the baseline for this study, with participants being followed up until May 2017 using comprehensive electronic records.
This study includes participants from GoDARTS with T2D at baseline to form the diabetic group, and patients with no diabetes form the control group. To allow for an accurate estimation of AKI rate in patients without diabetes, patients from GoDARTS who developed diabetes later during the follow-up time were not included in the study. Also, patients without SCr measures on or after recruitment were not included. For the eGFR slope analysis, patients with three or more SCr values, with at least a 1-year gap between the first and last measure before the first AKI episode (if applicable), and three or more SCr measures after the AKI episode, with at least a 90-day gap between the first and last of these measures, were included. Patients with an AKI event before analysis were excluded.
Datasets and Variables
The GoDARTS study was linked through an individual-specific, anonymized identifier to the following clinical datasets: information on diabetes, including type of diabetes and date of diagnosis, was acquired from the Scottish Care Information–Diabetes Collaboration diabetes summary and longitudinal data.18 SCr values were obtained from the laboratory biochemistry system, which comprised SCr measures from both primary and secondary care. The Scottish Renal Registry was used to identify patients receiving RRT and date of therapy initiation.19 The Scottish Morbidity Records 01 for hospital admission was used to evaluate patient comorbidities, including coronary artery disease, congestive heart failure, peripheral vascular disease, cerebrovascular disease, and liver disease on the basis of ICD-10 codes at admissions before recruitment. The community prescribing data were used to assess whether the patient had been prescribed any of the following classes of antihypertensive drugs: diuretics, angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, β-blockers, and calcium-channel blockers.20 The demographics dataset was used to determine participant sex and date of birth, which was used to calculate age at recruitment. Patients who had moved out of Tayside health board were treated as lost to follow-up. The Community Health Index (the National Health Service [NHS] Scotland population register) death dataset was used to obtain date of death. Follow-up time was defined as the time from recruitment until May 2017, date of RRT, date of death, or date the patient moved out of Tayside health board, whichever occurred first.
Development of an Algorithm To Identify AKI Episode from SCr Tests
SCr measures from January 1988 to May 2017 were used in the analysis; measures obtained after initiation of RRT were not included. All assays in the region are done in the same regional laboratory, and SCr measures were adjusted for changes in assays over time. AKI was defined on the basis of the KDIGO criteria.9 Because testing was infrequent with large time gaps for some patients, leading to a lack of baseline being calculated, we developed an algorithm to calculate baseline creatinine which incorporated both prior and postindex creatinine measurements in the definition of baseline (Table 1). Severity of AKI (stages 1–3; Table 1) was defined using KDIGO criteria.9 To identify AKI episodes, SCr measures that were within 7 days apart were grouped into single episodes of care. Within the episode of care, a 1.2-fold increase in creatinine from baseline was used to evaluate SCr values measured before and after each SCr value that was flagged as an AKI case, to assess AKI initiation and recovery and determine the start and end of the AKI episode.21 Furthermore, if two AKI episodes were within 7 days apart, then the two episodes and SCr values taken between these episodes were grouped into one AKI episode.21 The length of AKI episode was calculated on the basis of start and end dates of the AKI episode, and this was used to assess whether AKI had progressed to acute kidney disease (AKD), defined as an AKI lasting >7 days.22 The highest AKI stage within the episode was used to define the stage of the AKI episode.
Definition of AKI cases, AKI episodes, and AKI stages
eGFR and CKD Status
The Chronic Kidney Disease Epidemiology Collaboration formula was used to estimate GFR from SCr.23 Development of CKD was defined according to the CKD-KDIGO guideline as an eGFR of <60 ml/min per 1.73 m2 present on at least two occasions at least 90 days apart.23 To avoid misclassification between AKI and CKD, eGFR values contained within AKI episodes were first removed from the longitudinal data. The variation in the intensity of blood sampling led to eGFR estimates either too distant (in healthy individuals) or too dense (in patients who were sicker) over time. As a result, a median smoother was applied to the remaining eGFR values on the basis of a set of rules derived from the CKD-KDIGO definition as follows: for each date of index blood test, three eGFR baseline values were calculated using the median eGFR for the period 365–91 days before the index date, then 7 days before 7 days after index, and 91–365 days after index date, respectively. CKD diagnostic date was established when at least two of the three medians were <60 ml/min per 1.73 m2 (Table 2). The CKD date was then compared against recruitment date to determine whether participants had prevalent CKD at recruitment or whether they developed incident CKD during follow-up.
Establishing CKD date and CKD status from the longitudinal eGFR data
Primary and Secondary Outcomes
The primary outcome was the number of AKI episodes per person during follow-up, which was used to calculate AKI episode rates per 1000 patients per year (including recurrent events) and AKI rate ratios (RRs) in people with T2D versus those without diabetes, depending on CKD status. The secondary outcome was eGFR decline over time, calculated as the eGFR slope of the linear regression model per 1-year unit increase. Other outcomes were number of patients experiencing AKI during follow-up, length of AKI episodes, and AKI stage.
Statistical Methods To Analyze AKI Rates Depending on CKD Status
Counts and proportions for categoric variables, and mean and SD or median and interquartile range for quantitative data were used to describe the demographic characteristics. These were reported in people with and without diabetes and by CKD status (no CKD at recruitment or during follow-up, pre-CKD to account for the period before CKD development for those that developed CKD during follow-up, and post-CKD to include the post-CKD period for those that had CKD at recruitment or developed CKD during follow-up). The difference between two independent proportions were calculated on the basis of the Wilson method.24 The negative-binomial model for counts, with log-link and follow-up time as offset, was used to analyze the primary outcome and to estimate rates of AKI episodes in patients and controls depending on CKD status. The relationship between the outcome and the explanatory variable (sex, age, and diabetes status) was assumed linear via the log-link function.25 Unadjusted AKI rates and rates adjusted for age and sex were provided together with the corresponding RRs for association. Further adjustment for comorbidities at recruitment was performed to investigate how much of the effect of diabetes on AKI incidence rates can be explained by preexisting comorbidities. The chi-squared test was used to investigate the association between diabetes and AKI stage, and the nonparametric Mann–Whitney test was used to investigate differences in the length of AKI episodes between the T2D versus control group.
Sensitivity analyses were conducted to evaluate and compare incidence rates for stage 2 and stage 3 AKIs, and AKIs >48 hours, respectively, and for AKIs occurring during hospital admission in people with diabetes versus controls.
Statistical Methods To Analyze of Longitudinal eGFR Data
eGFR values measured during AKI episodes were first removed from the data and replaced at the start of the episode with a baseline eGFR, which was calculated as the median eGFR for the 7 days before the AKI episode, if measures were available, otherwise the median eGFR of values measured between 365 and 8 days before the start of the AKI episode was used. A linear, mixed-effect model was used to analyze the association between AKI and eGFR decline from the longitudinal eGFR data. AKI was included into the model as a time-varying factor with three levels: no AKI for patients with no AKI event during the follow-up, pre-AKI for patients with an AKI event during follow-up for the period before the AKI, and post-AKI for the period after the AKI episode.26 To identify significant changes in eGFR slope pre- and post-AKI event, and whether these changes differ between people with T2D and controls, an interaction term between AKI, T2D status, and time was incorporated into the model. Baseline variables such as sex, age (treated as age groups), and presence of cardiovascular diseases were fitted into the model with both fixed intercept and slope. An interaction between these variables and T2D was also included, and the Akaike information criterion was used for variable selection. Given the strong interaction effects between AKI, diabetes, and CKD status, the analysis was conducted separately for people with no CKD at recruitment and those who had an established CKD diagnosis before recruitment. The mixed model was fitted with both random intercept and slope per individual before and after the AKI episode (when applicable), assuming an unstructured covariance matrix for the random effects.
Data linkage and analysis was carried out using SAS 9.4 (SAS Institute Inc., Cary, NC).
Ethics Approval and Consent to Participate
The GoDARTS study was approved by the Tayside Medical Ethics Committee, with informed consent being obtained for all participants (Research Ethics Committee reference number 053/04). Data provision and linkage was carried out by the University of Dundee Health Informatics Centre (HIC; https://www.dundee.ac.uk/hic), with analysis of anonymized data performed in an International Organization for Standardization 27001– and Scottish Government–accredited, secure safe haven. HIC standard operating procedures have been reviewed and approved by the NHS East of Scotland Research Ethics Service, and consent for this study was obtained from the NHS Fife Caldicott Guardian.
Results
The Cohort
A total of 18,306 participants were recruited into the GoDARTS cohort, of which 16,700 met the selection criteria. Of these, 9417 of the patients had T2D at recruitment, and 7283 did not have diabetes at recruitment, nor did they develop it later; they formed the control group. A total of 1606 patients were excluded from this study, of which 681 had other types of diabetes, 720 developed diabetes after recruitment, and 205 did not have SCr tests on or after recruitment (Figure 1). Table 3 shows baseline characteristics of the cohort. People within T2D were older than controls (66.9 versus 60.8 years old; difference, 6.0 years; 95% CI, 5.7 to 6.4) and 44.0% were females, compared with 51.4% in the control group (difference, 7.4%; 95% CI, 5.8% to 8.9%). People with T2D had a lower eGFR at baseline compared with controls (76.6 versus 84.3 ml/min per 1.73 m2; difference, 7.7; 95% CI, 7.1 to 8.29), 26.6% of people with T2D had CKD at recruitment compared with only 9.1% in the control group (difference, 17.5%; 95% CI, 16.3% to 18.6%), and there was a higher percentage of people with cardiovascular disease in the diabetic group compared with the control group (Table 3). The mean (SD) follow-up time from recruitment was 8.2 (3.5) years for people with T2D versus 9.6 (2.4) years for controls.
The GoDarts cohort flowchart. T1DM, type 1 diabetes mellitus.
Baseline characteristics of the cohort broken down by diabetes status
AKI and AKI Episodes
Table 4 shows summary statistics of SCr measures from recruitment and describes the frequency of AKI in the two groups. A total of 512,615 SCr tests were recorded from recruitment; of those, 387,657 (75.6%) were from patients with T2D. The median (interquartile range) of SCr measures per individual during the follow-up were 31 (19–51) in T2D versus 11 (4–21) in controls. Including post-AKI SCr measures to calculate a baseline value increased the yield of AKI cases from 28,306 to 40,567. A breakdown of AKI cases identified using the different baseline SCr definitions using pre- and postindex SCr measures is shown in Supplemental Table 1. After grouping successive tests into episodes, a total of 13,928 AKI episodes were identified from recruitment until the end of follow-up. Of these, 11,647 were experienced by patients with diabetes. AKI occurred in 5837 patients, representing 48.6% (n=4580) of patients with T2D versus 17.2% (n=1257) of controls (difference, 31.4%; 95% CI, 30.0% to 32.7%). More than 50% of patients with diabetes experiencing AKI had recurrent AKI, whereas the majority of patients in the control group with AKI had only one episode of AKI during follow-up (Table 5). Overall, 54.2% of AKI episodes lasted no more than 2 days, a further 26.5% between 2 and 7 days, and the remaining 19.3% of AKI episodes were >7 days, resulting in AKD. Less than five AKI/AKD episodes were >90 days; however, after inspection, it was revealed these occurred during hospitalization due to other complications. Of the AKI episodes, 76.3% were stage 1, with the rest being stage 2 or 3. Diabetes was significantly associated with increased AKI episode length (P<0.001) but not significantly associated with AKI stage (P=0.74).
Descriptive statistics showing the characteristics of AKI flagged SCr tests and AKI episodes
Descriptive statistics showing sex, age, follow-up time, and number of SCr tests, and number of patients experiencing AKI and number of AKI episodes in the diabetic versus control groups, depending on CKD status
AKI and CKD
Supplemental Figure 1 illustrates the complex interplay between AKI/AKD and CKD and the many trajectories evolving during the course of the disease. The way AKI initiates and develops can take many forms, ranging from one acute kidney insult that improves rapidly with full recovery within 7 days (Supplemental Figure 1, A2), to one or more acute kidney insults during the course of the disease that progress to AKD and require >7 days to resolve. There are also cases when SCr does not fully reverse after an AKI episode, leading to the development of CKD (Supplemental Figure 1, D2). This further shows that, although some patients fully recover after an episode of AKI and never develop CKD (Supplemental Figure 1, A1–A3), others may experience rapid kidney decline after an AKI episode (Supplemental Figure 1, D1–D3). At the same time, there is also the possibility to develop CKD without prior AKI episodes, and only experience AKI later as superimposed on CKD (Supplemental Figure 1, B1–C3).
Table 6 describes the characteristics of people with diabetes versus controls in terms of their sex, age, and follow-up time, and frequency of AKI during follow-up depending on CKD status. Of the people with diabetes, 26.6% (n=2504) had CKD at recruitment and a further 19.7% (n=1855) developed the condition during follow-up, leading to a total of 46.3% (n=4359) people with CKD in the diabetic group compared with 17.1% (n=1251) in the control group (difference, 29.2%; 95% CI, 27.8% to 30.4%). In people with diabetes and CKD, 50.3% were female (n=2192) compared with only 38.6% (n=1954) in those without CKD (difference, 11.7%; 95% CI, 9.7% to 13.7%). Also, people with diabetes developed CKD at a younger age compared with people in the control group (mean age 74.1 versus 77.6 years; difference, 3.5 years; 95% CI, 3.0 to 4.0). Of the people with diabetes who developed CKD, 66.1% (n=2883) experienced AKI superimposed on CKD in the diabetic group compared with 45.5% (n=569) in the control group (difference, 20.6%; 95% CI, 17.5% to 23.8%). Additionally, 26.6% (n=493) of people with diabetes who developed CKD after recruitment had at least one episode of AKI before development of CKD, the corresponding figure in the control group was 9.9% (n=58; difference, 16.7%; 95% CI, 13.3% to 19.8%). The proportion of people experiencing AKI was significantly higher in the diabetic group compared with the control group for those patients who did not have CKD at recruitment and who did not develop it later: 31.7% (n=1602) versus 10.8% (n=651) in the control group (difference, 20.9%; 95% CI, 19.4% to 22.4%).
AKI episode rates and RRs in the diabetic and nondiabetic groups, depending on the CKD status
Estimating AKI Episode Rates in People with and without Diabetes
Table 6 shows estimates of AKI episode incidence rates and RRs for people with diabetes versus control, unadjusted and adjusted for sex and age at recruitment. Regardless of CKD status, adjusted AKI rates were 4.7 times higher in people with diabetes compared with controls (adjusted rate 179.0 versus 38.4 per 1000 person-years; RR, 4.7; 95% CI, 4.3 to 5.0). In particular, people with diabetes and no CKD experienced AKI at a rate almost five times higher than people with no diabetes (adjusted rate 121.5 versus 24.6 per 1000 person-years; RR, 4.9; 95% CI, 4.4 to 5.5), whereas, in people with CKD, rate of AKI for those in the diabetic groups was twice as high than in the corresponding control group (adjusted rate 384.8 versus 180.0 per 1000 person-years; RR, 2.1; 95% CI, 1.9 to 2.4). Similarly, people with diabetes who developed CKD after recruitment experienced episodes of AKI at a rate twice as high than those in the control group (adjusted rate 109.3 versus 47.4 per 1000 person-years; RR, 2.3; 95% CI, 1.8 to 3.0). It is noteworthy that the AKI rate in people with diabetes in the absence of CKD was very close to the AKI rate before development of CKD (121.5 versus 109.0 per 1000 person-years).
Additional model adjustment for other comorbidities at baseline only partially reduced the association between diabetes and AKI incidence rates (in people with no CKD at recruitment or during follow-up, RR, 3.85; 95% CI, 3.44 to 4.32; in people with CKD at recruitment or during follow-up, RR, 2.01; 95% CI, 1.82 to 2.22; Supplemental Table 2).
Sensitivity Analysis for the AKI Rate Analysis
The sensitivity analysis conducted to estimate rates for stage 2 and 3 AKIs show consistent results with the main analysis (Supplemental Table 3). The results show that people with diabetes and no CKD experience stage 2 and 3 AKIs at a rate that is over five times higher than people in the control group (adjusted mean rate 30.6 versus 5.5 per 1000 person-years; RR, 5.5; 95% CI, 4.6 to 6.6), whereas, in people with CKD, rate of AKI for those in the diabetic group was twice as high than in the corresponding control group (adjusted mean rate 76.5 versus 38.9 per 1000 person-years; RR, 2.1; 95% CI, 1.8 to 2.5). Similarly, analysis of rates of AKIs lasting >48 hours or AKIs during hospital admission show consistent results with the main analysis (Supplemental Tables 4 and 5).
Estimating the Effect of AKI on eGFR Slope over Time
Of the 16,700 people included in the initial analysis, there were 3250 people with AKI before recruitment that were not included in the eGFR analysis. A further 2558 people did not meet the selection criteria, of which 1324 had an AKI postrecruitment (738 with T2D and 386 with no diabetes). As a result, a total of 10,892 people with 279,391 SCr measures were included in the eGFR longitudinal data analysis. Of these, 5665 had T2D and 5227 were from the control group (Figure 1). Of the 10,892 participants, 2470 people experienced an AKI during follow-up, of which 1859 had T2D and 611 had no diabetes. People with no CKD at recruitment had a significantly higher decline in eGFR in the period pre-AKI compared with those with no AKI, regardless of diabetes status, but rate of decline was significantly higher in people with diabetes (eGFR slope pre-AKI versus no AKI was −1.14 [95% CI, −1.24 to −1.03] in people with T2D and −0.29 [95% CI, −0.45 to −0.11] in controls; slope difference, −0.85; 95% CI, −1.05 to −0.65; Figure 2, Supplemental Table 6). A further decrease in rate was observed in the control group in the period post-AKI compared with pre-AKI in both the T2D and control group, the increase in rate of decline was only marginally significant in people with T2D (eGFR slope post-AKI versus pre-AKI was −0.29; 95% CI, −0.59 to 0.01), whereas it was significant in the control group (eGFR slope post-AKI versus pre-AKI was −0.55; 95% CI, −1.08 to −0.03); however, the difference between T2D group and control was NS (slope difference, 0.26; 95% CI, −0.34 to 0.86). Sex was significantly associated with eGFR decline, with males having a higher mean eGFR than females in those with T2D and a lower eGFR in the control group. No change in eGFR slope was observed between males and females in any of the subgroups. An increase in age was associated with a reduction in eGFR at baseline, regardless of diabetes status, but significant differences in eGFR slope among the different age groups were observed only in people with T2D. Furthermore, people with peripheral vascular disease and hypertension had a significant, further decline in eGFR slope, regardless of diabetes status.
Visual representation of the eGFR slope estimates in people without AKI (no AKI), before the AKI (pre-AKI), and after the AKI event (post-AKI), depending on diabetes status and CKD status at recruitment. aReference group includes: no AKI during follow-up, female, ≤49 years, and no cardiovascular disease. bReference group includes: no AKI during follow-up, female, 50–64 years, and no cardiovascular disease.
People with CKD at recruitment show a higher rate of decline in eGFR in the period pre-AKI compared with no AKI, and this result was significant in the T2D group and marginally significant in the control group, but the difference between the two groups was NS (eGFR slope pre-AKI versus no AKI was −0.79 [95% CI, −1.05 to −0.52] in people with T2D and −0.40 [95% CI, −0.85 to 0.05] in controls; slope difference, −0.38; 95% CI, −0.90 to 0.14; Figure 2, Supplemental Table 7). The decline in eGFR rate post-AKI compared with pre-AKI did not change in people with T2D diabetes (eGFR slope post-AKI versus pre-AKI was 0.23; 95% CI, −0.24 to 0.71), whereas AKI was associated with further eGFR decline post-AKI compared with the pre-AKI period in controls (eGFR slope post-AKI versus pre-AKI was −0.84; 95% CI, −1.73 to 0.06), with the post-AKI effect being significantly different between T2D and control groups (slope difference, 1.07; 95% CI, 0.06 to 2.08). There was no significant eGFR difference between males and females in people with CKD at recruitment, regardless of diabetes status. An increase in age was associated with a reduction in eGFR at baseline, regardless of diabetes status, and older people with diabetes appeared to have a lower eGFR decline than those who were younger. None of the cardiovascular diseases were significantly associated with eGFR at baseline or eGFR slope; however, their effect was an important one as reflected in the Akaike information criterion used for variable model selection and, therefore, they were retained in the model.
Discussion
In our study, we have quantified rates of AKI in patients with and without diabetes, demonstrating the extent of the risk. Rates of AKI are significantly higher in patients with T2D compared with those without, with a 4.7-fold increase in AKI rate. In people with preserved renal function, rate of AKI is 4.9-fold higher in people with diabetes than in people without diabetes; whereas, in people with CKD, the rate of AKI for those in the diabetic group is two-fold higher than in patients without diabetes. More than 50% of the patients with diabetes who develop AKI will suffer from recurrent events. Rates of CKD are also higher in patients with T2D, with 46.3% developing CKD compared with 17.1% in those without diabetes.
Fall in eGFR slope before AKI was steeper in people with diabetes compared with those without diabetes. After AKI episodes, loss of eGFR became steeper in people without diabetes, but did not increase in those with diabetes and preexisting CKD.
In comparison to other studies, progressive decline leading to CKD has been well described in people with T2D,1 but AKI in diabetes mellitus has been less investigated.6,7,27,28 Girman et al.6 examined 119,966 patients with diabetes and 1,794,516 patients without diabetes from the General Practice Research Database. AKI incidence was markedly higher in their cohort: 198 per 100,000 person-years in patients with T2D compared with 27 per 100,000 patient-years among patients without diabetes (crude hazard ratio, 8.0; 95% CI, 7.4 to 8.7). They did not use a biochemical definition for AKI and relied instead on clinical coding, which can lead to significant under-ascertainment.29 In addition, a meta-analysis by James et al.30 showed the hazards ratios for AKI were higher in participants with diabetes compared with those without diabetes at any level of eGFR. Once again, the definition for AKI relied on administrative codes in these studies, thereby underestimating milder forms of AKI. There are very few studies that have examined AKI and CKD simultaneously and recurrent AKI in this group of patients.31
Our results are consistent with existing evidence indicating that diabetes is an independent risk factor for AKI.3,6,27 However, reported AKI rates in people with diabetes vary greatly depending on the population studied (e.g., different specialist settings, age range) and the methods used for AKI identification (e.g., medical history, ICD-10 coding, or changes in SCr).10 Most of the prior studies have reported AKI incidence of new AKI cases within a given time window and, therefore, estimates relate to number of patients experiencing AKI. The algorithm developed in this study allows for quantification of AKI rates on the basis of the number of AKI episodes, including recurrent AKI. Our findings have important clinical implications. AKI is associated with adverse patient outcomes, including increased mortality, future development of CKD, and increased length of hospital stays.3,32,33 Therefore, it places a significant financial burden on healthcare resources.34 In our study, >75% of AKIs were stage 1, reflecting a mild, transient increases in SCr. This may be of clinical significance because there is an increasing evidence showing that even mild, transient (lasting <24 hours) AKI is associated with poorer long-term outcomes,35,36 compared with those who do not have AKI. There are currently no effective treatments for AKI once it is established, and so earlier detection and prevention is vital. It is, however, important to note there may be misclassification of chronic decline in renal function in patients with diabetes that account for some of the observed increased rates of AKI. We have shown that rates of AKI are higher in patients with diabetes, both with and without CKD, with more than half developing recurrent episodes. To our knowledge, there has been no previous work looking at eGFR slopes before developing AKI. We found that those who develop AKI with diabetes have a greater decline in eGFR slope before developing AKI than those who do not. These findings are expected because a declining kidney would be more susceptible to episodes of AKI. However, it is surprising that there is less additional decline in eGFR in those with diabetes, compared with those without, after an episode of AKI compared with before an AKI episode. It remains unclear what the mechanism underlying AKI is in patients with diabetes. A predisposing factor in these patients may be generalized or intrarenal atherosclerosis. In addition, patients with diabetes are likely to have glomerular hyperfiltration that is masking structural renal damage. This renders them more susceptible to AKI than those without diabetes, due to their reduced repair capacity, and they are, therefore, susceptible to fluctuations in SCr. A further suggested mechanism is that tubular growth in response to hypergylcemia promotes inflammation, senescence, and tubulointerstitial fibrosis, which enhance the susceptibility of the diabetic kidney to episodes of AKI.37 It also remains unclear whether prevention of AKI in these patients would prevent or delay progression of CKD. However, it would seem sensible that these patients are monitoring more closely during intercurrent illnesses, with a greater awareness of avoiding high-risk medicines, such as nonsteroidal anti-inflammatories and aminoglycosides. There is currently a lack of awareness among patients with diabetes of the risk of AKI, and so patient education on the importance of hydration may play an important in role. We have also shown that, in patients with both hypertension and diabetes, there is an additional decline in eGFR, highlighting the importance of BP control in addition to ensuring good glycemic control in this patient group.
An important strength of the study is the refinement of the KDIGO definition, enabling a more sensitive estimation of AKI rates, which has allowed us to demonstrate the high risk of AKI in patients with diabetes, regardless of CKD status. We developed an algorithm to identify AKI episodes from SCr measures. A number of definitions to detect AKI cases on the basis of changes in SCr have been used previously,16 and NHS England has implemented an algorithm that applies the KDIGO definition to routinely collected SCr tests to automatically produce AKI alerts to support clinical investigations.15 This algorithm defines baseline creatinine levels on the basis of SCr 1 year before the index date, potentially leading to undetected AKI when such measurements are not available. The proposed algorithm uses SCr values both prior and after the index date. Although this may not be useful for AKI detection in clinical practice, it may improve AKI detection for epidemiologic purposes when applying it to routinely collected datasets, allowing for a more sensitive estimation of AKI incidence. Our study shows that at least one third of AKI cases remains undetected when baseline creatinine is only on the basis of tests before the index date. Previous epidemiologic studies of AKI from routinely collected SCr reported AKI cases in isolation, with episodes being defined using either fixed time periods, such as 30 days,14 or admission and discharge dates for patients who were hospitalized.31 This study is novel through the development of an algorithm that examines consecutive SCr measures to detect the start and the end of an AKI episode, which can be used to calculate the length of the episode and further assess whether the AKI has resolved quickly or if it has progressed to AKD. The grouping of AKI cases into AKI episodes was particularly important to allow an accurate estimate of AKI rates when applied to routinely collected data. Identification of the AKI episode start and end dates was also used in the study to clean the SCr data, to allow assessment of CKD status and correctly determine the CKD onset date, which represents another strength of the study. Another important strength of the study is the development of a statistical framework for the analysis of the eGFR longitudinal data to evaluate decline in eGFR before and after an AKI event, depending on diabetes and CKD status.
One of the main limitations of the study relates to the nature of routine healthcare data where blood measurements are infrequent, which makes it difficult to calculate baseline creatinine for assessment of AKI. As a result, some of the AKI in the longitudinal data might remain undetected, leading to misclassification between AKI and progressive CKD. This variation in the intensity of blood sampling may also lead to ascertainment bias in AKI estimation, due to more tests that are being performed in patients who are sicker. In our study, blood tests were performed, on average, three times more often in people with diabetes than people in the control group. This may partially explain the high AKI rate in people with diabetes compared with controls. It could, however, be argued that increased testing was performed in response to clinical indication and, similarly, lack of testing in those who were deemed well. The possibility that the increased AKI is being driven by the increased testing, rather than the other way around, is diminished by the analysis of more severe (stage 2 and stage 3) AKIs and AKIs lasting >48 hours, for which a high AKI RR between people with diabetes compared with controls in the absence of CKD was obtained. In addition, diabetes status confers a substantially increased risk for AKI in individuals with preexisting CKD, where the testing rate is high regardless of diabetes status. These results demonstrate a profoundly increased clinical burden of AKD in patients with diabetes. Another limitation of the study is the potential of selection bias due to the use of consented data from the GoDARTs cohort, a characteristic of most observational studies using consented data, which may lead to AKI rate estimates that are not generalizable. Furthermore, calculation of slopes required a number of creatinine measures over a specified time period, and so a significant number of patients were excluded from the analysis. This could have introduced selection bias, which may have affected our findings. However, it is difficult to eliminate this issue when examining eGFR slopes using observational data.
In conclusion, we have quantified the risk of AKI in patients with diabetes and its relationship as both a precursor and a consequence of CKD. The risk of AKI in this population of patients is currently underestimated and associated adverse outcomes after AKI are not well understood. Further work to evaluate the pathogenesis for AKI and the risk factors associated with the increased AKI rate in patients with diabetes, such as use of medication, is required to allow for development and implementation of interventions that both prevent the occurrence of AKI and reduce decline in eGFR, thereby improving patient outcomes.
Disclosures
All authors have nothing to disclose.
Funding
This project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement number 115974. This Joint Undertaking receives support from the European Union’s (EU’s) Horizon 2020 Framework Programme and European Federation of Pharmaceutical Industries and Associations with JDRF. The Wellcome Trust United Kingdom Type 2 Diabetes Case Control Collection (supporting GoDARTS) was funded by the Wellcome Trust, under grants 072960/Z/03/Z, 084726/Z/08/Z, 084727/Z/08/Z, 085475/Z/08/Z, and 085475/B/08/Z.
Acknowledgments
We acknowledge all consortium partners of the Biomarker Enterprise to Attack Diabetic Kidney Disease (BEAt-DKD) project for constructive discussions during project meetings. We acknowledge the support of the HIC, University of Dundee, for managing and supplying the anonymized data. We are grateful to all the participants who took part in the GoDARTS study, to the general practitioners, to the Scottish School of Primary Care for their help in recruiting the participants, and to the whole team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses.
The research outlined here was conducted as part of the BEAt-DKD project, an EU Seventh Framework Programme (FP7) and US consortia as part of the Innovative Medicine Initiative (see https://www.beat-dkd.eu/). The full list of BEAt-DKD partners is provided in Supplemental Appendix 1.
Dr. Simona Hapca designed the study, conducted the data processing and analysis, and wrote and revised the manuscript; Dr. Moneeza K. Siddiqui, Dr. Ryan S.Y. Kwan, Dr. Shona Matthew, Dr. Alex S.F. Doney, and Dr. Ewan R. Pearson contributed to the interpretation of the data and the revision of the manuscript; Dr. Samira Bell and Dr. Colin N.A. Palmer contributed to study design, the interpretation of the data, and the writing and revision of the manuscript; Dr. Simona Hapca and Dr. Colin N.A. Palmer are guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Supplemental Material
This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2020030323/-/DCSupplemental.
Supplemental Summary 1. List of BEAt-DKD partners.
Supplemental Table 1. Number of AKI cases identified using the NHS England algorithm and the modified algorithm broken down by the different criteria used in the definition of the AKI case.
Supplemental Table 2. AKI episode incidence rate ratios adjusted for sex, age and comorbidities at recruitment depending on CKD status.
Supplemental Table 3. Incidence rates of stage 2 and 3 AKI episode and rate ratios in the diabetic and non-diabetic groups depending on the CKD status.
Supplemental Table 4. AKI episode incidence rates and rate ratios for AKIs lasting more than 48hrs in the diabetic and non-diabetic groups depending on the CKD status.
Supplemental Table 5. AKI episode incidence rates and rate ratios for AKIs during a hospital admission in the diabetic and non-diabetic groups depending on the CKD status.
Supplemental Table 6. Parameter estimates of the longitudinal eGFR data analysis for people with and without T2D and no CKD at recruitment.
Supplemental Table 7. Parameter estimates of the longitudinal eGFR data analysis for people with and without T2D and no CKD at recruitment.
Supplemental Figure 1. The relationship between SCr and eGFR longitudinal data used to indentify AKI episodes and CKD diagnosis date.
Footnotes
Published online ahead of print. Publication date available at www.jasn.org.
See related editorial, “The Aftermath of AKI: Recurrent AKI, Acute Kidney Disease, and CKD Progression,” on pages 2–4.
- Copyright © 2021 by the American Society of Nephrology
References
In this issue
