OBJECTIVES : In hospitalised patients with missing data on preadmission serum creatinine (SCr), the baseline renal function is commonly assumed to be normal or defined using surrogate values. In this study, we provide estimates of preadmission SCr, based on multiple clinical characteristics available in electronic clinical records, and assess their accuracy for AKI diagnosis in comparison with a) SCr at hospital admission and b) SCr value back-calculated from assumed estimated glomerular filtration rate of 75 ml/min/1.73m².

METHODS : A multivariate linear regression model was built in two approaches: Basic (sex, age, department and mode of admission, SCr at admission) and Complete (basic model + comorbidities). Potential bias due to differences in characteristics of patients with and without preadmission SCr was addressed using inverse probability weighting.

RESULTS : From 45,798 unique adult admission to Centro Hospitalar Universitário de São João in Porto, Portugal, between 2013 and 2015, we analysed 8,911 patients with preadmission SCr available. The actual cumulative incidence of AKI was 15.6%. Basic model had the best agreement for AKI diagnosis (Kappa=0.691, 15.5%) followed by Complete model (Kappa= 0.662, 18.8%). The admission SCr underestimated, while 75 ml/min-calculated SCr overestimated AKI occurrence (Kappa=0.585, 7.8%, and Kappa=0.628, 22.0%, respectively). The highest Youden´s index was achieved for Complete and 75ml/min calculated SCr (both 0.72), and the lowest for admission SCr (0.46).

CONCLUSIONS : Estimation of baseline SCr, taking into account elementary admission data improved misclassification in AKI diagnosis over commonly used surrogate methods.