Regional Variation in the Incidence of Dialysis-Requiring AKI in the United States

Discussion

To our knowledge, this study is the first to explore geographic differences in the incidence of AKI in the general population in the United States. Its strengths include being nationally representative and having a large sample size, which translate into excellent power to estimate regional incidence rates accurately.

Understanding regional variations in disease incidence has provided useful insights for many medical conditions. For example, well characterized geographic variation in disease incidence and mortality of stroke and acute myocardial infarction (12–14) has helped to generate hypotheses on causation and has driven public health initiatives aimed at decreasing cardiovascular disease in high-risk regions (15,16). Since 2000, the USRDS Annual Data Reports have presented relevant statistics in an atlas format to emphasize geographic differences in disease occurrences and outcomes associated with ESRD (9,17). The use of mapping in publications by the institutions such as the Centers for Disease Control and Prevention and the Dartmouth Atlas Project has been instrumental in identifying geographic disparities in disease burden and mortality (12–14), practice patterns (18), and health care delivery (19).

Our investigations led to several novel findings. Using nationally representative hospital discharge data, we found significant geographic variation in the population incidence rate of dialysis-requiring AKI across the four US Census-designated regions, with highest incidence in the Midwest and lowest incidence in the Northeast in both crude and adjusted models.

We showed that, because of variations in hospitalization rates across regions in the country, variations in incidence of AKI per hospitalization do not accurately reflect true underlying regional variations in AKI in the population. This finding highlights the problematic way in which AKI epidemiology is historically performed (i.e., with AKI incidence expressed as cases per hospitalization rather than cases per underlying population). Inferences drawn by comparing single-center or even multicenter studies (2) that express AKI as a percentage of hospitalizations are confounded by varying thresholds for admission across hospitals/regions, and therefore, they arguably should be avoided as much as possible in future studies.

We wondered if the pattern of regional variation in population incidence of dialysis-requiring AKI may mirror the regional variation in ESRD incidence but found no correlation. We explored this comparison because of the numerous shared patient risk factors across the spectrum of kidney diseases (such as older age and diabetes mellitus) and the knowledge that underlying CKD itself is a potent risk factor for AKI (20). Our data suggest that determinants of incidence of dialysis-requiring AKI are distributed in a geographically distinct fashion from shared risk factors for kidney disease.

Interestingly, the one region with the most abundant nephrologist workforce is also the one with the lowest incidence of dialysis-requiring AKI (Northeast). At least one previous (nonrandomized) interventional study has shown that early nephrologist involvement in patients with AKI reduces the risk of a further decrease in kidney function (21). Density of nephrologists may also correlate with numerous other health care system attributes that affect incidence of AKI. A recent multicenter study examining the incidence of AKI associated with acute myocardial infarctions showed significant between-hospital variation in AKI incidence (as well as significant between-hospital variation in use of potentially nephrotoxic medications and intravenous hydration) (22,23). These data suggest that better understanding of practice patterns underlying regional variations in AKI incidence may provide very useful insights into effective measures to reduce AKI incidence, complementing advances made through local interventions (24).

It is notable that the regions with the lowest incidence of dialysis-requiring AKI have the highest case fatality. As we have defined it (and as it has been defined in the prior literature) (5,25), developing dialysis-requiring AKI not only involves the patient having severe AKI but also, the decision by the physician and the acceptance by the patient (and perhaps, family members) to initiate acute dialysis. (We use the term dialysis here to encompass both intermittent hemodialysis as well as continuous renal replacement therapy.) There are no universally accepted thresholds for when renal replacement therapy should be initiated for AKI (26,27). The inverse relationship between population incidence of AKI and the mortality associated with dialysis-requiring AKI suggests the possibility that there may be geographic practice pattern variation on the threshold to perform acute dialysis. Earlier and more pervasive use of acute dialysis on less severely ill patients with AKI may explain why there is higher incidence and lower mortality in some regions. Alternatively, we can also hypothesize that better practice in the care of patients with AKI, such as earlier nephrologist involvement and more optimal performance of dialysis, may be a factor in more favorable in-hospital mortality in some areas versus others. Our database is not sufficient to test these hypotheses.

Our study has several limitations. First, we do not have laboratory measurements, such as serum creatinine, to define and quantify the severity of AKI. Although we relied on the best-validated set of diagnostic and procedure codes currently in the literature to define our primary outcome of dialysis-requiring AKI, it would have been ideal to have clinical data such as serum creatinine to help validate the cases. We cannot exclude the possibility that there are regional differences in coding practices that introduced bias. However, we are not aware of data that indicate any such misclassification would occur systematically by US regions (i.e., consistently coded more so in Midwest versus Northeast). Another limitation of the study is that our results lack geographic granularity; only four regions can be separated out and analyzed. Conducting the analysis at the regional level may result in us missing potentially more dramatic contrasts in AKI incidence at the state level. This shortcoming is imposed by the stratified sampling scheme of the NIS dataset, which uses the Census-designated region rather than individual states as a stratification variable. It is possible that some of the AKI cases that we observed were patients with advanced CKD (in whom small changes in low baseline GFRs translated into large changes in serum creatinine concentration) who were admitted basically to start chronic dialysis, and that there were regional differences in frequency of this type of hospitalization. We could not separate out cases of acute or chronic renal failure versus de novo AKI. Another limitation is that we did not include comorbidities that are AKI risk factors (such as diabetes, CKD, proteinuria, and obesity) in our adjustment model. Our Poisson model approach to determine accurate population incidence prevented us from adding those comorbidities, because doing so would have required incorporation of regional population totals with those prevalent conditions. Determining those figures (i.e., number of men/women in specific age groups in one region with or without CKD, proteinuria, diabetes, and obesity) is not possible with currently available data. Our associations are of an ecological nature (e.g., we did not know that there was a high density of nephrologists working at the hospitals with fewer cases of AKI). Lastly, we quantified nephrology workforce using Board Certification data and do not have confirmation that these practitioners are clinically active.

To conclude, this study contributes novel and provocative information regarding the population of epidemiology AKI. It adds to the growing literature on the importance of geographic variation in kidney disease epidemiology (17,28,29). Additional research is needed to clarify whether geographic differences are because of AKI risk factors, practice pattern differences, or some other cause to develop potential preventative interventions.