Determination of number of AKI-D hospitalizations at state level
We used State Inpatient Databases (SID) to determine the number of AKI-D hospitalizations [
3]. SID are part of the family of databases and software tools developed for the Healthcare Cost and Utilization Project (HCUP), developed through a U.S. Federal-State-Industry partnership. The SID contain the universe of the inpatient discharge abstracts in participating States, translated into a uniform format to facilitate multi-State comparisons and analyses. The SID contain a core set of clinical and nonclinical information on all patients, regardless of insurance satus [
3‐
5].
Although there were 30 SID databases available for 2011, we only had financial resources to purchase 25 of them. We selected a convenience sample of 25 of the larger states by geographic region (which captured 90% of discharges): Arizona (AZ), Arkansas (AR), California (CA), Colorado (CO), Florida (FL), Iowa (IA), Kentucky (KY), Massachusetts (MA), Maryland (MD), Maine (ME), Michigan (MI), Mississippi (MS), North Carolina (NC), Nebraska (NE), New Jersey (NJ), New Mexico (NM), New York (NY), Nevada (NV), Oregon (OR), Rhode Island (RI), South Carolina (SC), Utah (UT), Vermont (VT), Washington (WA), and West Virginia (WV)(Additional file
1: Figure S1). The different state databases varied in capturing whether a diagnosis was present on admission or not and in indicating whether an admission represented a re-admission to the hospital for the same patient in a given calendar year.
We defined AKI-D as requiring both a diagnostic code for acute renal failure (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9] codes 584.5, 584.6, 584.7, 584.8, or 584.9) and a procedure code for dialysis (39.95, V45.11, V45.12, V56.0, or V56.2), along with the absence of procedure codes for arteriovenous fistula creation or revision (39.27, 39.42, 39.43, or 39.93) [
6‐
8]. This algorithm has been shown to be sensitive and specific, producing high positive and negative predictive values (all ≥90%) [
6‐
8].
To address potential ascertainment bias arising from the fact that states reported different numbers of diagnostic codes (range 9–60) and different numbers of procedure codes (range 6–30) for each individual in these databases, we first analyzed the number of actual diagnostic codes and procedure codes for each hospitalization. Of the 25 SID databases, 19 (AZ, AR, CA, CO, FL, IA, KY, MD, MI, NV, NJ, NM, NY, NC, OR, RI, VT, WA, and WV) reported > 15 diagnostic codes. In none of these 19 states did more than a quarter of the hospitalizations have > 15 diagnostic codes (range 4.1–24.5%). Of the 25 SID databases, 19 states (AZ, AR, CA, CO, FL, IA, KY, MD, MA, MI, NV, NJ, NY, NC, OR, RI, SC, VT, and WA) reported > 6 procedure codes. In none of these 19 states did more than a tenth of the hospitalizations have > 6 procedure codes (range 1.6–6.1%).
Thus in our primary analysis, we excluded the 2 states with fewer than 15 diagnostic codes (ME, NE) (Additional file
1: Figure S1). All states had at least 6 procedure codes. For the remaining states, we only analyzed the first 15 diagnostic codes and first 6 procedure codes listed for each hospitalization (i.e. for states whose database contained additional information, we ignored the diagnostic codes in position 16 and above and we ignored procedure codes in position 7 and above).
We did not count as AKI-D hospitalizations those hospitalizations with a diagnostic code for ESRD present on admission (585.6). Thus, in our primary analysis, we additionally excluded 5 states (CO, MS, NC, UT, WV) whose SID did not specify whether a diagnosis of ESRD was present on admission or not) (Additional file
1: Figure S1). (We did include AKI-D hospitalizations with diagnosis of ESRD only on discharge but not on admission.)
Therefore, our primary analysis was based on 18 states (AZ, AR, CA, FL, IA, KY, MA, MD, MI, NJ, NM, NY, NV, OR, RI, SC, VT, WA) (Additional file
1: Figure S1). In 2011, these 18 states accounted for 50% of the country’s incident ESRD cases [
9].
In sensitivity analyses, we used data from all 25 states we had SID data on but to make ascertainment more uniform, we only analyzed up to 9 diagnostic codes and up to 6 procedure codes for each state (all states reported at least these numbers of codes). In this sensitivity analysis, we used the same AKI-D definition as above, but we excluded all hospitalizations containing a diagnostic code for ESRD (585.6) regardless of whether it was present on admission or not (since for CO, MS, NC, UT, and WV, we could not tell if a given diagnosis was present on admission or not).
Statistical approach
The AKI-D incidence and rate of renal recovery among incident ESRD patients were both expressed as per million population (pmp) per year for each state. We used Pearson correlations to analyze the association between AKI-D incidence and rate of renal recovery across states. We repeated our analyses in subgroups defined by sex, age (45–64, 65–74, ≥75 years). We did not show results for age 0–44 years old due to small number of outcomes (for example, 8 of the 18 states had fewer than 10 observed cases of renal recovery among incident ESRD patients of this age range).
We also did not show results stratified by race/ethnicity as 10 of the 18 states had fewer than 10 observed cases of renal recovery among incident, non-Hispanic black ESRD patients. We also used partial correlation to analyze AKI-D incidence and the rate of renal recovery across states adjusted by sex and age groups. (We did not adjust for race as there was no correlation between race and AKI-D incidence and no correlation between race and rate of renal recovery.)
Data were analyzed using SPSS version 23.0 (IBM SPSS, Chicago, IL). Results were independently confirmed using SAS version 9.4 (SAS Institute Inc., Cary, NC) or STATA version 14.1 (College Station, TX) by separate analysts.