Acute kidney injury identification for pharmacoepidemiologic studies: Use of laboratory electronic acute kidney injury alerts versus electronic health records in Hospital Episode Statistics

Abstract

Purpose A laboratory‐based acute kidney injury (AKI) electronic‐alert (e‐alert) system, with e‐alerts sent to the UK Renal Registry (UKRR) and collated in a master patient index (MPI), has recently been implemented in England. The aim of this study was to determine the degree of correspondence between the UKRR‐MPI and AKI International Classification Disease‐10 (ICD‐10) N17 coding in Hospital Episode Statistics (HES) and whether hospital N17 coding correlated with 30‐day mortality and emergency re‐admission after AKI. Methods AKI e‐alerts in people aged ≥18 years, collated in the UKRR‐MPI during 2017, were linked to HES data to identify a hospitalised AKI population. Multivariable logistic regression was used to analyse associations between absence/presence of N17 codes and clinicodemographic features. Correlation of the percentage coded with N17 and 30‐day mortality and emergency re‐admission after AKI were calculated at hospital level. Results In 2017, there were 301 540 adult episodes of hospitalised AKI in England. AKI severity was positively associated with coding in HES, with a high degree of inter‐hospital variability—AKI stage 1 mean of 48.2% [SD 14.0], versus AKI stage 3 mean of 83.3% [SD 7.3]. N17 coding in HES depended on demographic features, especially age (18–29 years vs. ≥85 years OR 0.22, 95% CI 0.21–0.23), as well as sex and ethnicity. There was no evidence of association between the proportion of episodes coded for AKI with short‐term AKI outcomes. Conclusion Coding of AKI in HES is influenced by many factors that result in an underestimation of AKI. Using e‐alerts to triangulate the true incidence of AKI could provide a better understanding of the factors that affect hospital coding, potentially leading to improved coding, patient care and pharmacoepidemiologic research.

Burden of Acute Kidney Injury among Adult Hospital Patients in the Italian Lombardy Region: A 20-Year Real-World Data Analysis

“…The latter is close to our finding of AKI diagnosis in almost 90 per 100,000 population, indicating that the hospital staff considered mainly advanced stages for coding by ICD, even if dialysis was required in less than 8% of hospitalizations. The underreporting of total AKI incidence by ICD coding in administrative databases compared with laboratory-based detection has also been observed in other studies [13,14]. However, even though our analysis identified mainly patients with advanced stages, it provides important insights into the disease burden.…”

“…On the other hand, the observed between-province differences might also reflect ICD coding practice, that differed severalfold between hospitals [5]. The sensitivity of AKI detection in administrative databases was close to 30% in historical studies [25][26][27] but is reported to be higher in modern analyses where an average of 50% of stage 1, 70% of stage 2, and 80% of stage 3 AKI episodes were assigned ICD codes by the hospital personnel (with severalfold differences in coding frequency between hospitals) [5,13]. The rules for coding and reimbursement were the same for medical facilities in all the Lombardy provinces, and the exact reasons for the between-province differences need to be studied further.…”

The risk of acute kidney injury in colorectal cancer survivors: an english population-based matched cohort study

“…Furthermore, AKI staging and severity could not be determined as serum creatinine measures are not available from HES-APC. We used ICD-10 codes recorded as part of a hospital admission to define AKI, which have been validated and are considered to have high specificity in identifying AKI [ 27 ] but some AKI cases may still have been missed, especially less severe presentations. The journey of a colorectal cancer patient through the healthcare system is likely different to that of an individual with no history of cancer, as evidenced by the difference in follow-up time, a limitation which we acknowledge.…”

Defining measures of kidney function in observational studies using routine health care data: methodological and reporting considerations

“…The scientific literature reports varying algorithms to define AKI in health care databases (Table 4). 78 Below, we summarize common features of these definitions that require attention as they can affect the generalizability or validity of study findings. We illustrate this using our third example question: "After AKI, what is the causal effect of stopping versus continuing renin-angiotensin system inhibitors on the risk of recurrent AKI?"…”