INTRODUCTION: The Scottish Patient Safety Programme (SPSP) launched an 18-month national collaborative in August 2017 with the aim of reducing harm from Acute Kidney Injury (AKI). They have proposed using ICD-10 coding to measure any changes in AKI incidence, mortality and length of hospital stay that may occur over the course of the collaborative. However, administrative database coding for AKI has been shown to underestimate rates of AKI when compared to biochemically defined AKI. Our aim was to assess whether ICD-10 coding is a reliable method of measuring rates and mortality of AKI. METHODS: NHS Tayside introduced automated AKI e-alerts in April 2015 to improve the recognition of AKI. To determine how reliable ICD-10 coding was at measuring rates of AKI, we performed an observational cohort study of all NHS Tayside in-patient admissions aged 18 and over between 1st January 2013 and 30th April 2017, excluding those receiving renal replacement therapy prior to admission. 240,227 in-patients were included for analysis. We calculated the sensitivity, specificity, positive and negative predictive values for ICD-10 coding of AKI compared to biochemically defined AKI using the KDIGO definition, considering all stages of AKI as well as more severe stages (AKI 2 and 3). ICD-10 code N17 (acute renal failure) was used to define coded AKI. The relative risk (RR) for both 30-day and 90-day mortality was calculated for those with ICD-10 coded AKI and those with biochemically defined AKI. Additionally, we used interrupted time series (ITS) methodology to determine whether the introduction of e-alerts led to a change in rates of ICD-10 coding of AKI. RESULTS: Over the course of the study period there were 20,967 episodes of biochemically defined AKI (AKI 1: n=13,638; AKI 2&3: n=7,329) and 7,068 cases of ICD-10 coded AKI. Sensitivity of ICD-10 coding for AKI was low (<30%) for all AKI stages, and while it increased for more severe stages, was still poor (<50%). There was no significant difference between sensitivity, specificity, PPV and NPV when comparing pre and post-intervention data. For AKI mortality, there was a reduced risk of death following e-alert introduction in those with coded AKI (30-day mortality: RR=0.81 (0.73-0.90); 90-day mortality: RR=0.82 (0.76-0.89)) but there was no significant effect in those with biochemically defined AKI (30-day mortality: RR=1.00 (0.94-1.06); 90-day mortality: RR=0.99 (0.95-1.04)). Additionally, there was an increased proportion of coded AKI consisting of AKI stage 1 in the post-intervention group (+3.91%, p=0.0004), but a non-significant reduction in those with biochemically defined AKI (-1.26%, p=0.056). Segmented regression analysis of the ITS demonstrated a small but statistically significant decrease in rates of coded AKI following the e-alert introduction. CONCLUSIONS: ICD-10 coding of AKI was shown in our study to be unreliable for monitoring rates and outcomes from AKI. Using ICD-10 to measure outcomes of research may produce misleading results by showing a reducing AKI mortality subject to ascertainment bias and underestimating AKI rates, therefore alternative methods of measurement should be used. A defined e-alert algorithm for use in research may allow greater collaboration between regions but questions remain over the impact of e-alerts on clinical outcomes.