Development and Validation of a Risk Prediction Model of Post-stroke Acute Kidney Injury

Abstract

Background Acute kidney injury (AKI) is a common and serious complication that is closely correlated to a poor short-term or long-term prognosis in stroke patients. Therefore, it is necessary to develop a specific AKI screening tool to early identify patients at high risk of AKI. Objective To construct and verify a risk prediction model of post-stroke AKI and to develop a simple post-stroke AKI risk assessment scale. Methods Stroke inpatients with complete medical records were selected from the Second Affiliated Hospital Zhejiang University School of Medicine by use of convenience sampling, including 760 from neurology department treated during January to September 2021 (model group, 140 with AKI, and 620 without), and 310 treated during October to December 2021 (validation group, 53 with AKI and 257 without). Multivariate Logistic regression was used to identify factors associated with post-stroke AKI, then these factors were used to develop a risk prediction model. The Hosmer-Lemeshow test and receiver operating characteristic analysis were performed to assess the accuracy of fit and prediction value of the model, respectively. Then the model was verified in validation group, and based on the validation results, a simple post-stroke AKI risk assessment scale was developed. Results The prevalence of post-stroke AKI in the model group was 18.42% (P<0.05). Multivariate Logistic regression analysis showed that sex, history of hypertension, NIHSS score, history of use of loop diuretics, history of mechanical thrombectomy, serum levels of β2-MG, urea nitrogen, and sCysC were independently associated with post-stroke AKI (P<0.05). The post-stroke AKI risk prediction model constructed is y=1/ (1+e-a), in which a=-4.047+1.222× male + 1.386 × hypertension history + 1.716 × NIHSS score + 1.098 ×history of use of loop diuretics + 0.830 × mechanical thrombectomy history + 1.739 × β2-MG+1.202 × urea nitrogen + 2.160 × sCysC. The fit of the model was χ2=6.523, P=0.367. The AUC of the model for predicting post-stroke AKI in model group was 0.916 〔95%CI (0.891, 0.940) 〕, with 0.857 sensitivity, 0.832 specificity, and 0.689 Youden index when the optimal cut-off value was chosen as 12.8%. And the AUC of the model in predicting post-stroke AKI in the verification group was 0.906 〔95%CI (0.853, 0.960) 〕. The coefficients (β) derived from multivariate Logistic regression were rounded to the nearest integral value and weighted, then used to compile a simple scale with a total points of 11, whose AUC in predicting post-stroke AKI risk was 0.900〔95%CI (0.843, 0.957), P<0.001〕when the optimal cut-off value was determined as 4, and the accuracy rate of which in practical applications was 88.39%. Conclusion Our risk prediction model could effectively predict the risk of post-stroke AKI with high sensitivity and specificity, and the risk assessment scale compiled based on the model is a simple, feasible, objective, and quantitative tool for identifying high-risk patients, and the assessment result may be a reference for doctors and nurses to take interventions to early prevent AKI in stroke patients.

Keywords

• stroke|acute kidney injury|risk prediction model|assessment instrument|screening|forecasting|root cause analysis