Increased incidence of acute kidney injury requiring dialysis in metropolitan France

Background Acute kidney injury requiring dialysis (AKI-D) is associated with high mortality. Information about its epidemiology is nonetheless sparse in some countries. The objective of this study was to assess its epidemiology and prognosis in metropolitan France. Methods Using the French hospital discharge database, the study focused on adults hospitalized in metropolitan France between 2009 and 2014 and diagnosed with AKI-D according to the codes of the French common classification of medical procedures. Crude and standardized incidence rates (SIR) by gender and age were calculated. We explored the changes in patients’ characteristics, modalities of renal replacement therapy (RRT), in-hospital care, and mortality, along with their determinants. Trends over time in the SIR for AKI-D, its principal diagnoses, and comorbidities were analyzed with joinpoint models. Results Between 2009 and 2014, the AKI-D SIR increased from 475 (95% CI, 468 to 482) to 512 per million population (95% CI, 505 to 519). AKI-D was twice as high in men as women. Median age was 68 years. Over the study period, the AKI-D SIR steadily increased in all age groups, particularly in the elderly. The most common comorbidities were cardio-cerebrovascular diseases (64.8%), pulmonary disease (42.2%), CKD (33.8%), and diabetes (26.0%); all of these except CKD increased significantly over time. In 2009, heart failure (17.2%), sepsis (17.0%), AKI (13.0%), digestive diseases (10.7%), and shock (6.6%) were the most frequent principal diagnoses, with a significant increase in heart failure and digestive diseases. The proportion of patients with at least one ICU stay and continuous RRT increased from 80.3% to 83.9% and from 56.9% to 61.8% (p<0.001), respectively. In-hospital mortality was high but stable (47%) and higher in patients with an ICU stay. Conclusions This is the first exhaustive study in metropolitan France of the SIR for AKI-D. It shows this SIR has increased significantly over 6 years, together with ICU care and continuous RRT. In-hospital mortality is high but stable.

Introduction Acute kidney injury (AKI) is a common and severe complication in hospitalized patients [1,2]. Due to its prognosis and frequency, AKI has been identified as a global public health issue [3]. Whatever its etiology, AKI requiring dialysis (AKI-D) is associated with higher morbidity and mortality [4][5][6]. To develop preventive strategies and slow the growth of AKI-D, the International Society of Nephrology has recommended developing a better understanding of the epidemiology of AKI-D and its associated risk factors. Its global epidemiology, however, remains insufficiently characterized in the absence of accurate reporting of this event in many countries. Some studies from medico-administrative databases report dissimilar incidence rates over the past two decades [7][8][9]. Hsu et al. estimates that the crude incidence rate (CIR) of AKI-D rose from 222 to 533 per million population (pmp) between 2000 and 2009 in the United States [7], although a slightly more recent US study reports no increase in either the AKI-D incidence after adjustment for age and sex or in comorbidities between 2006 and 2014 [10]. In England, Kolhe et al. report that the CIR for AKI-D jumped from 15.9 to 208.7 pmp between 1998 and 2013 [8]. These increased incidence rates may be attributable to the growth of either susceptibility risk factors, such as advanced age, chronic kidney disease (CKD), and diabetes, or of exposure risk factors, such as nephrotoxic drugs or agents, sepsis, critical illness, and major surgery, together with changes in clinical practice related to indications for renal replacement therapy (RRT) [7,8,11,12]. Despite growing interest in AKI-D, national descriptions and prognostic data of these factors remain insufficient to enable the improvement of prevention and enhancement of healthcare organization.
In this context, our objective was to assess the trends in AKI-D incidence in metropolitan France between 2009 and 2014 and their determinants, from information in the French national hospital discharge database. Moreover, we explored the trends in patient characteristics, principal diagnoses, RRT modalities, hospital care, and mortality.

Study population
All hospitalizations for AKI-D from 2009 to 2014 were extracted from the comprehensive French hospital discharge database, which provides de-identified data with demographic and medical information about the admission diagnosis, underlying comorbidities, procedures performed, and complications. AKI-D was defined by three specific medical dialysis procedures and coded according to the French Common Classification of Medical Procedures (CCMP) [13] with the following codes for RRT: JVJF002 (RRT by intermittent hemodialysis, hemofiltration, or hemodiafiltration for AKI), JVJF005 (RRT by continuous hemodialysis, hemofiltration, or hemodiafiltration for AKI) and JVJB002 (RRT by peritoneal dialysis for AKI).
Patients aged younger than 18 years or not residents of metropolitan France (i.e., France excluding overseas districts and territories) were not included, nor were patients with a diagnosis of ESRD in the database before the AKI-D hospitalization. ESRD was defined by the codes presented in the supplementary data section. To reinforce the exclusion of ESRD patients, we applied a 3-month "wash-out" period and used the data for January-March, 2009, only to ensure that no patients from April had an ESRD diagnosis in the previous 3 months. Hospital stays for AKI-D in 2009 were estimated from the stays observed for the 9-month period that began in April after observing the absence of seasonal changes of AKI-D for years 2010-2014. To avoid bias from patients with recurrent AKI-D, new AKI-D episodes were retained only if they occurred at least six months after a previous episode, six months during which the patient had no stays for ESRD treated by dialysis.

Data
The Agence Technique de l'Information Sur l'Hospitalisation (ATIH) waived the need for consent (decree No. 94-666). All French hospitals caring for medical and surgical patients submit de-identified patient data to the French hospital discharge database annually. The linking of each discharge summary submitted to this database to a national grouping algorithm leading to a French diagnosis-related group (DRG) [14] allowed patient comorbidities to be recorded and linked [15]. The study was conducted according to the approval given by the ATIH. Authorization was also obtained from the Commission Nationale de l'Informatique et des Libertés (agreement No. 1375062).
Hospitalization information included age and gender, admission from home or transfer from another hospital, admission to intensive care unit (ICU), modality of RRT (continuous or intermittent or peritoneal dialysis), interval from hospital admission to RRT initiation, length of stay, and in-hospital mortality. The principal diagnosis was the admission diagnosis: the condition for which the patient was admitted. Eight principal diagnosis groups were defined according to the ICD-10 (i.e., shock, respiratory disease, digestive disease, cardiac failure, sepsis, AKI, CKD, and other diagnoses). Associated comorbidities were defined according to ICD-10 codes [16]. Ten groups of comorbidities were defined according to the methodology of the French national health insurance fund (i.e., cardio-cerebrovascular, pulmonary, hepatic, and psychiatric comorbidities, CKD, diabetes mellitus, malignancy, obesity, malnutrition, and dementia) [17]. Cardio-cerebrovascular comorbidity included the following subgroups: rhythm disorders, heart failure, ischemic heart disease, peripheral vascular disease, stroke, and myocardial infarction. Because RRT modality could change during the course of hospitalization, three modalities were defined: continuous RRT (CRRT), intermittent RRT, and peritoneal dialysis (S1 Appendix).
Outcomes. Annual crude incidence of AKI-D was defined by the ratio of new AKI-D hospital admissions of adult residents of metropolitan France each year from 2009 to 2014. The annual AKI-D SIR was standardized for age and sex by direct standardization, with the adult French population in 2014 as reference. Incidence rates were expressed per million population per year.

Statistical analysis
The characteristics of in-patients were expressed as absolute numbers and percentages according to calendar years. Qualitative values were compared with the chi-square test. Quantitative variables were described by their medians and interquartile ranges for non-Gaussian distributions and compared with the Wilcoxon score test and the Kruskal-Wallis test. The 95% confidence intervals of standardized incidence rates were calculated with a Poisson approximation.
Trends in standardized incidence rates of AKI-D, principal diagnoses, and comorbidities over the study period were analyzed with joinpoint models. A grid search was used to determine locations of no more than one joinpoint with a subsequent permutation test to determine superiority [18]. If the linearity of the trend was validated, the annual change was reported as the annual percent change (APC).
Patient characteristics and care were compared according to the outcome, either an ICU stay or in-hospital mortality. Univariate and multivariable logistic regression analyses were performed. The multivariable analysis used nested models containing several groups of variables, specifically, demographic data, year of admission, interval from admission to RRT initiation, comorbidities, principal diagnosis, and ICU admission, when applicable. Statistical analyses were performed by using SAS software version 9.3 (SAS Institute Inc.) and Joinpoint Statistical Software version 4.5.0.1 (Statistical Research and Applications Branch, National Cancer Institute).

Population
Between April 1, 2009, and December 31, 2014, 122 million hospital stays were recorded in metropolitan France, for an estimated population of 64 million inhabitants. These included 138,167 hospital stays coded AKI-D of patients aged 18 years or more (Fig 1).

Incidence trends of acute kidney injury requiring dialysis
Between 2009 and 2014, the absolute number of stays with AKI-D increased from 22,001 to 25,142. The AKI-D SIR increased significantly from 475 (95% CI, 468 to 482) to 512 pmp (95% CI, 505 to 519) during this period with an APC of +1.7% (Fig 2).

Patient trajectory and care
Most patients had at least one ICU admission during their hospital stay, with a significant increase from 80.3% to 83.9% over time. Globally, the use of CRRT (at least one use) rose significantly from 56.9% in 2009 to 61.8% in 2014 (Table 2), primarily in ICU, where use increased significantly from 67.9% to 71.3% (Table 3).

Fig 2. Population incidence of acute kidney injury requiring dialysis in metropolitan France from 2009 to 2014 (absolute number of stays and standardized incidence rate per million population).
For 2009, the absolute number of stays and the crude incidence rate were reported as an estimation over 12 months based on observations for 9 months. I bars represent 95% CIs for standardized incidence rates. � indicates that the Annual Percent Change (APC) is significantly different from zero at the alpha = 0.05 level.

Characteristics of the population according to ICU admission
Characteristics of the population according to ICU admission are shown in the supplementary data S1 Table and S3 Fig. In the multivariable analysis, younger age, later year of admission during the study period, principal diagnosis, and comorbidities (i.e., hepatic, pulmonary, and cardio-cerebrovascular comorbidities and cancer) were associated with a risk of ICU admission (S2 Table).

In-hospital mortality rate, trends, and risk factors
In-hospital mortality remained stable at 46.6% during the study period, significantly higher in men (47.4%) than in women (45.1%) (p<0.001). In the multivariable analysis, gender, age, interval between admission and RRT initiation, comorbidities, principal diagnosis, and ICU admission were significantly associated with in-hospital mortality (Table 4). Among patients with at least one ICU admission, mortality decreased significantly from the start to the end of the study period, from 54.3% to 51.7% (S4 Fig).

Patient characteristics and changes in management in patients with an ICU admission
For patients with at least one admission to ICU, the median IGS2 was statistically significantly higher at the end of the study period, rising from 59 (IQR, 45-75) to 60 (IQR, 46-77). The difference in the frequency of mechanical ventilation was also statistically significant, dropping from 77.9% to 76.1%. Conversely, the use of vasopressors was significantly higher at the end of the study period, as were the use of the CRRT modality and interval between admission and RRT initiation (Table 3). Epidemiology AKI requiring dialysis in metropolitan France

Discussion
For this first study of AKI-D in metropolitan France, using the national hospital discharge database, we showed a significant increase in the AKI-D SIR in adults from 475 to 512 pmp over the study period (2009)(2010)(2011)(2012)(2013)(2014). This corresponds to an APC of +1.7%. In 2014, the incidence of AKI-D was more than three times higher than that of ESRD (163 pmp) [19]. Note that this result is adjusted for population age and sex. We also reported an increase in several comorbidities, notably cardio-cerebrovascular, and described the modalities of dialysis used to manage AKI. As commonly reported, the principal diagnoses for these hospital stays were diverse: AKI, heart failure, sepsis, digestive diseases, respiratory diseases, and shock, with a  significant increase in heart failure and digestive diseases. In 2014, most patients experienced at least one admission to ICU (83.9%) and at least one treatment by CRRT (61.8%); both of these rates increased significantly during the study period. Overall in-hospital mortality was higher in ICU than in non-ICU patients, but it decreased significantly from 54.3% to 51.7% over the study period.

Comparisons with other studies
Our results are consistent with the results of previous studies in the United States of America, England, and Denmark [7][8][9]20,21]. In the US, the crude incidence of AKI-D was higher than in Europe and increased from 222 to 533 pmp between 2000 and 2009 [7]. This US incidence of AKI-D might, nonetheless, be biased because it was calculated from a nationwide inpatient sample and with a weighting system. It might also be overestimated because patients with ICD-9 codes of ESRD were not excluded. Because the US results by Kashani et al. found no increase in AKI-D incidence after age and sex adjustments, our results differ from theirs [10]. Unlike them, we used as the denominator of our incidence rates the "at risk" population, i.e., the population of metropolitan France, rather than hospitalized patients. The incidence rate of AKI-D in our study was somewhat higher than that observed in England and Denmark for the same period [8,9]. This result might be due in part to the use of different indications for RRT in clinical practice. Another point of difference is that our incidence rates were standardized. Others have showed an increase of crude incidence of AKI-D without standardization and without considering the aging of the general population [7][8][9]. Previous studies of ICU patients have reported that CRRT is the most widely-used RRT for AKI-D (between 52% to 80%) [9,[22][23][24]. We showed an increased incidence of AKI-D related to at least one CRRT modality in ICU. In our study, in-hospital mortality was stable at 47% and was higher than that either in England (41.1% in 2008-2013) [8] or in the US national administrative database study reporting that For 2009, the absolute number of stays and others characteristics were estimated from observations over 9 months. https://doi.org/10.1371/journal.pone.0211541.t003 Epidemiology AKI requiring dialysis in metropolitan France  [24].

Determinants of the increasing incidence rate of AKI-D
The rise in the SIR for AKI-D in metropolitan France might be due either to an increase in the population requiring acute RRT, or to a change in the indications for RRT in clinical practice, or both. The first hypothesis finds support in the increase in severe forms of AKI requiring RRT, associated with an increasing susceptibility to AKI due to several risk factors, including aging, cardio-cerebrovascular and hepatic comorbidities, diabetes, and obesity [7][8][9]26,27]. Furthermore, heart failure is a well-known risk factor for AKI-D, especially for patients with vascular overload meeting the definition of the cardiorenal syndrome; this too could account for the increasing AKI-D incidence [7,[28][29][30]. Another well-known AKI risk factor is sepsis [24,[31][32][33][34], but its frequency with sepsis decreased in our study, presumably associated with better management of sepsis and/or changes in RRT indications in recent years. Hsu et al. showed that septicemia, hypertension, respiratory failure, coagulation disorders, shock, and liver disease account for the increase in the temporal trend of AKI-D. Several of these factors were present in our study [35]. Furthermore, the stability of CKD comorbidity over 30% during the study period suggests that the increase is not explained by AKI-D among CKD patients, but by severe AKI requiring dialysis in patients with previously normal kidney function. Moreover, the interval between admission and RRT initiation increased, corresponding to an increase in hospital-acquired AKI compared to admission for community-acquired AKI [36].
Our second hypothesis suggests a change in RRT indications in clinical practice, more specifically, in ICUs, which are using RRT more liberally. CRRT in routine clinical practice began only in the late 1990s, together with improvement in technical modalities and new indications over time, such as sepsis, for which early CRRT is now indicated [37][38][39][40][41][42][43]. Our results suggest that intensivists increasingly consider RRT to be indicated for patients with circulatory failure or with multiple comorbidities. This, combined with better availability of the continuous modalities, may account for the increased use of CRRT in the ICU [7,37,38,[44][45][46]. In Denmark, increased incidence of AKI-D coincided with a growth in CRRT from 27.4% to 56.4% between 2000 and 2012; moreover, CRRT patients were all admitted to ICU [9]. The increase in AKI-D incidence could also be explained by an increase in the use of RRT for severe AKI in patients aged 70-89 years. In very elderly patients over 90 years, the incidence of AKI-D was much lower, despite its high rate of increase: because of their poor prognosis, these patients are rarely admitted to ICU and rarely undergo invasive procedures such as RRT when they are needed.

Strengths and limitations of the study
Our study has several strengths. First, it was based on the French hospital discharge database, which records all public and private hospitalizations and is complete for RRT. Although it is retrospective, this report is based on prospective recording of all hospitalizations over the study period and thus minimizes information bias. Second, it provides an accurate estimation of the SIR for AKI-D applicable to metropolitan France as a whole. Third, coding of RRT for AKI has appeared to be a robust criterion in hospital discharge databases that has proven in previous studies and in other countries that use it for billing and reimbursement purposes [47,48] to be sensitive (76.9%-90.3%) and highly specific (93.8%-99.9%). Finally, our study was based on an algorithm that identified all AKI-D cases. This algorithm excluded patients with an ESRD code in any hospitalization before the AKI-D hospital stay and any recurrent AKI-D stays less than 6 months after an included stay for AKI-D. This limited the potential overestimation of the AKI-D incidence rate.
Our study must nonetheless be interpreted in light of the following limitations. First, neither the cause of AKI nor the indication for RRT appeared as such in the discharge summary, which did, however, include the reason for admission. A second limitation is the risk of misclassification between AKI-D and ESRD when both AKI-D and ESRD codes were present for the same stay, as they were in 24% of stays. In the study by Hsu et al., among the AKI-D hospitalizations, 36% had concurrent discharge diagnostic codes of ESRD [35]. In a sensitivity analysis excluding hospitalized stays that had both AKI-D and ESRD codes recorded, we found that both crude and standardized incidence rates of AKI-D showed a rising temporal trend during the study period (2009)(2010)(2011)(2012)(2013)(2014). These hospital stays might correspond to ESRD patients starting RRT in emergency conditions, to AKI-D complicated with ESRD that indicated the need for post-discharge RRT, or to ESRD that was incorrectly coded. A study in California showed that many patients with AKI-D did not recover renal function and were considered to have reached ESRD [5]. Another limitation of our study is an information bias secondary to the quality of coding of comorbidities and principal diagnoses. Thus, in some cases, CRRT patients might have received intermittent dialysis during their hospital stay; however, only 20% of patients were treated with both modalities, and this rate decreased during the study period. Finally, we restricted the study to AKI-D patients because of missing information about KDIGO classification in the national hospital discharge database. However, it is essential to learn more about the epidemiology of AKI-D to develop preventive strategies for improving prognosis and decreasing costs.

Clinical implications
We provide additional support of the rising incidence of AKI-D from 2009 to 2014 in metropolitan France, further confirming the need for steps to be taken upstream to decrease risk factors and prevent AKI. Moreover, the indication for RRT in AKI must be appraised, specifically when associated with other conditions such as cardiac failure and sepsis. Furthermore, health services need to be alerted so that they can adapt and enrich the provision of care. Finally, a prospective study would be useful to better understand the indications for RRT in ICUs and thus improve the RRT initiation strategy there [49,50].