Development of a nomogram model for the early prediction of sepsis-associated acute kidney injury in critically ill patients

Abstract

Abstract Sepsis-associated acute kidney injury is a common complication of sepsis, but it is difficult to predict sepsis-associated acute kidney injury. In this retrospective observational study, adult septic patients were recruited from the MIMIC-III database as the training cohort (n = 4764) and from Xiangya Hospital (n = 1568) and Zhang’s database as validation cohorts. We identified eleven predictors with seven independent risk predictors of sepsis-associated acute kidney injury [fluid input_day1 ≥ 3390 ml (HR hazard ratio 1.42), fluid input_day2 ≥ 2734 ml (HR 1.64), platelet_min_day5 ≤ 224.2 × 109/l (HR 0.86), length of ICU stay ≥ 2.5 days (HR 1.24), length of hospital stay ≥ 5.8 days (HR 1.18), Bun_max_day1 ≥ 20 mmol/l (HR 1.20), and mechanical ventilation time ≥ 96 h (HR 1.11)] by multivariate Cox regression analysis, and the eleven predictors were entered into the nomogram. The nomogram model showed a discriminative ability for estimating sepsis-associated acute kidney injury. These results indicated that clinical parameters such as excess input fluid on the first and second days after admission and longer mechanical ventilation time could increase the risk of developing sepsis-associated acute kidney injury. With our study, we built a real-time prediction model for potentially forecasting acute kidney injury in septic patients that can help clinicians make decisions as early as possible to avoid sepsis-associated acute kidney injury.