Study On Nephrotoxicity Of Intravenous Administration Of Iodinated Contrast Media And Establishment Of A Deep Learning Model

PartⅠTrend analysis of post-contrast acute kidney injury in hospitalized patients with intravenous administration of iodinate contrast mediaBackground:Post-contrast acute kidney injury（PC-AKI）is generally considered to be the third commonest cause of hospital-acquired acute kidney injury（AKI）and accounts for 11%of AKI cases.Moreover,the occurrence of PC-AKI can increase the risk of renal failure,dialysis and death.Currently,the treatment of PC-AKI is mainly supportive,including hemodynamic monitoring,glycemic control,discontinuation of all nephrotoxic agents when possible,dialysis and so on.Specific treatment to reverse the PC-AKI is still lacking.Therefore,risk stratification and prevention of PC-AKI have become the focus of researchers.With the deepening understanding of PC-AKI risk factors and the implementation of prevention measures,whether the incidence of PC-AKI in hospitalized patients has changed is unclear.Objective:To examine the yearly trends in PC-AKI incidence and risk factors associated with PC-AKI in hospitalized patients with intravenous administration of iodinated contrast medium（ICM）.Methods:Adult patients who underwent contrast-enhanced computed tomography（CT）or CT angiography（CTA）during hospitalization at the Second Xiangya Hospital of Central South University from January 1^st,2015,to December 31^st,2021 were retrospectively included.PC-AKI was defined according to the SCr criteria of the Kidney Disease:Improving Global Outcomes（KDIGO）Clinical Practice Guideline for AKI.Logistic regression analyses were used to analyze the yearly trends of PC-AKI and the incidence of dialysis dependence at discharge and in-hospital mortality of PC-AKI patients.Risk factors of PC-AKI were identified by multivariable logistic regression analyses.Cochran-Armitage trend test was used to analyze the yearly trends of PC-AKI risk factors.Multivariable logistic regression analyses were also used to analyze the yearly trends of intervening measures related to PC-AKI.Results:（1）A total of 45,013 hospitalized patients were included in the study,and the overall incidence of PC-AKI was 4.5%.The annual incidence of PC-AKI showed a significant declining trend during the 7 years（P<0.001）in both unadjusted and adjusted trend analyses.The incidence declined from 5.3%in 2015 to 4.1%in 2021.The annual incidences of PC-AKI stage 1 and stage 2-3 also showed similar decreasing trends（P<0.001 for both）.（2）Risk factors of PC-AKI included advanced age（>75 years）,congestive heart failure,liver cirrhosis,renal stone disease,diabetic nephropathy,hypertension,sepsis,baseline estimated glomerular filtration rate（e GFR）,mechanical ventilation,use of nephrotoxic drugs on the day of intravenous administration of ICM,low-osmolar ICM,and repeated ICM administration within 3 days.（3）Among the comorbidities associated with PC-AKI,the proportion of patients with congestive heart failure or hypertension showed a significant increasing trend during the 7-year period,while the proportion of patients with advanced age,baseline e GFR<30ml/min/1.73m²,diabetic nephropathy or renal stone disease showed a significant declining trend.No significant trend was observed in the proportion of patients with 30≤baseline e GFR<60 ml/min/1.73m²,sepsis or liver cirrhosis.（4）Among the intervening measures associated with PC-AKI,the proportion of patients using nephrotoxic drugs on the day of ICM administration showed a significant downward trend,while the proportion of patients using iso-osmolar ICM or receiving intravenous fluids≥1000 ml on the day of ICM administration showed a significant upward trend.The proportion of patients with repeated ICM administration within 3 days showed no significant trend change.After adjustment,the trend changes of each intervening measure remained the same.（5）The overall incidences of dialysis dependence at discharge and in-hospital mortality in hospitalized PC-AKI patients were both 3.1%.In both unadjusted and adjusted trend analyses,no significant trend was observed in the annual incidence of dialysis dependence at discharge,while a significant decreasing trend was observed in the annual incidence of in-hospital mortality.Conclusion:A significant declining trend in PC-AKI incidence was observed in hospitalized patients with intravenous administration of ICM from 2015 to 2021,which may be related to increased implementation of PC-AKI prevention measures.PartⅡImpact of intravenous administration of ICM on renal function and prognosis in AKI patientsBackground:Patients with AKI are generally considered vulnerable to contrast-associated nephrotoxicity and consequently withheld or delayed the use of ICM,which could result in delays in diagnosis or treatment of the disease.However,there is little evidence to support such practice.Objective:To examine the effects of intravenous administration of ICM on renal function deterioration,dialysis and mortality in hospitalized AKI patients.Methods:Adult patients who experienced AKI during hospitalization and underwent plain or enhanced CT scans within 7 days after AKI onset at Second Xiangya Hospital of Central South University between January 1^st,2015 and December 31^st,2019 were retrospectively enrolled.Patients were divided into ICM group and non-ICM group based on whether they received ICM or not.Propensity score matching was used to balance the differences in baseline characteristics between the two groups.The outcomes of the study were renal function deterioration,new receipt of dialysis and mortality in 7 and 30 days after CT examination,as well as dialysis dependency at 30 days after CT examination.Logistic regression analyses were used to analyze the association of intravenous administration of ICM with renal function deterioration and dialysis,and odds ratio（OR）and 95%confidence interval（CI）were calculated.Cox regression analyses were used to analyze the association between intravenous administration of ICM and mortality,and hazard ratios（HR）and 95%CI were calculated.Sensitivity analyses were also performed in patients stratified by SCr levels at AKI diagnosis,times of CT scan,time from first CT scan to AKI diagnosis,and in patients without CKD or dialysis requirement prior to CT scan.Results:A total of 4060 patients were included in the study,with1336 patients in the ICM group and 2734 patients in the non-ICM group.After 1:1 propensity score matching,1172 patients were included in each group,and the baseline characteristics between the two groups were balanced.No significant difference was observed in the incidences of outcomes between the ICM and non-ICM groups.Outcomes in 7 days after CT examination:renal function deterioration,7.8%vs 9.0%,OR（95%CI）was 0.83（0.62-1.11）,P=0.21;new receipt of dialysis,2.3%vs 3.3%,OR（95%CI）was 0.72（0.43-1.18）,P=0.19;mortality,3.9%vs 4.8%,HR（95%CI）was 0.83（0.56-1.22）,P=0.34.Outcomes in 30 days after CT examination:renal function deterioration,5.1%vs 5.4%,OR（95%CI）was 0.93（0.64-1.34）,P=0.69;new receipt of dialysis,4.2%vs 4.5%,OR（95%CI）was 0.95（0.64-1.41）,P=0.80;mortality,9.0%vs 10.2%,HR（95%CI）was 0.88（0.68-1.15）,P=0.36;dialysis dependence at 30days after CT examination,4.4%vs 5.1%,OR（95%CI）was 0.86（0.58-1.25）,P=0.46.Sensitivity analyses also showed that there was no significant difference in the incidences of 7-and 30-day outcomes between the ICM and non-ICM groups in all subgroup patients.Conclusion:Intravenous ICM administration during AKI duration has no effect on the risks of renal function deterioration,dialysis and mortality in 7 and 30 days after CT scan.The nephrotoxicity of ICM in AKI patients is overestimated.PartⅢEstablishment of a deep learning-based model to predict AKI following intravenous administration of ICM in hospitalized patients with chronic kidney diseaseBackground:Predicting the risk of AKI after intravenous administration of ICM in CKD patients（e GFR<60 ml/min/1.73 m²）and stratifying the risk is important for guiding the intravenous use of ICM and clinical decision-making.Machine learning provides a new tool for building medical models,but it is unclear whether it outperforms the traditional logistic regression model in predicting PC-AKI.Objective:To establish a risk prediction model for PC-AKI in hospitalized CKD patients with intravenous administration of ICM based on deep neural network（DNN）,and compare the model with that established by logistic regression algorithm.Methods:（1）The study patients were collected from three independent centers.Adult patients with e GFR<60 ml/min/1.73 m² and performed enhanced CT or CTA during hospitalization were enrolled.The derivation and internal validation cohorts were formed with patients from the Second Xiangya Hospital of Central South University from January 1^st,2015 to December 31^st,2020.The external validation cohort was formed with patients from the Xiangya Hospital of Central South University from January 1^st,2017 to August 31^st,2017 and the openly accessible database Medical Information Mart for Intensive Care-Ⅳ.（2）The outcome of the study was PC-AKI,which was defined based on the SCr criteria of the KDIGO Clinical Practice Guideline for AKI.（3）Variables that were considered to be associated with PC-AKI and available prior to CT/CTA examination were extracted from the Hospital Information System as candidate variables.（4）Six feature selection methods were used to identify the most predictive variables from the candidate variables,including P-value ranking,correlation coefficient ranking,stepwise regression selection,Least Absolute Shrinkage and Selection Operator,decision tree and random forest.In each feature selection method,variables were sorted from high to low based on their importance or the degree of correlation with PC-AKI,and the top 20 variables were selected.The number of times that each variable was selected was then counted,and if the variable was selected for more than three times among the six feature selection methods,it would be ultimately chosen to build the model.（5）Models were established based on a DNN or logistic regression algorithm,and validated in the internal and external validation cohorts.Model performance was evaluated by the area under the receiver operating characteristic curve（AUROC）,the area under the precision-recall curve（AUPRC）and decision curve.（6）Low-risk and high-risk cutoff points were set to stratify patients in the DNN model.The risk of PC-AKI in each risk group and diagnostic test characteristics of the model were calculated.Results:（1）A total of 4218 patients were enrolled in our study.2944,737and 537 patients were in the derivation,internal and external validation cohorts,respectively,with the PC-AKI incidence being 10.3%,10.4%and 11.4%,respectively.（2）14 variables were ultimately chosen to establish our model based on the six feature selection methods,which included neutrophil count,baseline e GFR,platelet count,serum calcium,serum sodium,blood urea nitrogen,the number of times diuretics were used,proteinuria,renal CTA,mechanical ventilation,serum albumin,total protein,serum total bilirubin and the number of times antimicrobial drugs were used.（3）The AUROC（95%CI）of the DNN model was 0.939（0.916-0.958）and 0.940（0.909-0.954）in the internal and external validation cohorts,respectively,which significantly outperformed that of the logistic regression model,with AUROC（95%CI）being 0.690（0.628-0.754）and0.745（0.683-0.808）,respectively.The AUPRC（95%CI）of the DNN model was 0.630（0.529-0.716）in the internal validation cohort and 0.619（0.523-0.715）in the external validation cohort,which also showed significantly better performance than that of the logistic regression model,with AUPRC being 0.201（0.138-0.273）and 0.290（0.187-0.403）,respectively.Decision curves showed that when the threshold probability was between 0 and 0.38,the DNN model had greater net benefits than the logistic regression model.（4）Significant differences were observed among the risk groups after setting the low-and high-risk cutoff points of the DNN model.The observed PC-AKI risks increased significantly from the low-to intermediate-to high-risk group（<1.0%to>50%）and the accuracy of patients not developing PC-AKI was 99%in the low-risk category in both the internal and external validation cohorts.Conclusions:A DNN model using 14 routinely available variables can accurately predict and stratify the risk of PC-AKI in hospitalized CKD patients following intravenous administration of ICM.PartⅣEffects of intravenous hydration in preventing AKI after intravenous administration of ICM in patients with CKD stage 4-5Background:Patients with CKD stage 4-5（e GFR<30ml/min/1.73m²）are recommend to have intravenous hydration before and after intravenous administration of ICM in guidelines.However,this recommendation is largely based on expert consensus,and direct evidence showing the effectiveness of intravenous hydration in preventing PC-AKI in these patients is still lacking.Objective:To explore the effectiveness of intravenous hydration in preventing PC-AKI in patients with CKD stage 4-5 after intravenous administration of ICM.Methods:Adult patients with baseline e GFR<30 ml/min/1.73m²who received intravenous administration of ICM during hospitalization at Second Xiangya Hospital of Central South University between January1^st,2015 and December 31^st,2021 were retrospectively enrolled.Patients were divided into the hydration group and non-hydration group based on whether they received intravenous hydration or not.Intravenous hydration was defined as giving 0.9%saline,1 ml/kg per hour,12 hours before and 12 hours after ICM administration or 500 ml 0.9%saline before and after ICM administration.1:1 propensity score matching and propensity score overlap weighting were both used to balance the baseline characteristics between the two groups.The outcomes of the study were PC-AKI,dialysis dependence at discharge,and in-hospital mortality.PC-AKI was defined according to the 2012 KDIGO Clinical Practice Guideline for AKI and the European Society of Urogenital Radiology（ESUR）guidelines.Logistic regression analyses were used to analyze the relationship between intravenous hydration and the outcomes.Sensitivity analysis was performed by restricting patients in the non-hydration group to those with no intravenous fluids administration on the day of ICM administration,in order to rule out the impact of intravenous fluids on the outcomes.Results:（1）A total of 794 patients were included in the study,with 284 in the hydration group and 510 in the non-hydration group.After 1:1propensity score matching,210 patients were included in each group,and the two groups achieved balance in the adjusted baseline characteristics and the distribution of propensity scores was almost the same.There was no significant difference in the incidences of the outcomes between the hydration group and the non-hydration group:KDIGO-defined PC-AKI,25.2%vs 24.8%,OR（95%CI）was 0.93（0.57-1.50）,P=0.76;ESUR-defined PC-AKI,31.0%vs 25.2%,OR（95%CI）was 1.34（0.86-2.08）,P=0.19;dialysis dependence at discharge,4.3%vs 3.3%,OR（95%CI）was 1.56（0.56-4.50）,P=0.40;in-hospital mortality,1.9%vs 0.5%,OR（95%CI）was 4.08（0.58-81.08）,P=0.21.（2）Overlap propensity score weighted analyses also resulted in between-group balance on adjusted baseline characteristics and the weighted distribution of propensity scores was similar between the two groups.No significant difference was found in the incidences of the outcomes between the hydration group and the non-hydration group:KDIGO-defined PC-AKI,24.4%vs 24.0%,OR（95%CI）was 0.94（0.53-1.66）,P=0.82;ESUR-defined PC-AKI,29.9%vs 26.6%,OR（95%CI）was 1.17（0.70-1.97）,P=0.55;dialysis dependence at discharge,4.3%vs 3.6%,OR（95%CI）was 1.36（0.41-4.75）,P=0.61;in-hospital mortality,2.3%vs 1.6%,OR（95%CI）was 1.48（0.26-9.79）,P=0.65.（3）In sensitivity analysis,a total of 567 patients were included in the study,with 284 in the hydration group and 283 in the non-hydration group.After 1:1 propensity score matching and overlap propensity score weighting,the adjusted baseline characteristics were almost balanced between the two groups.The results also showed that there was no significant difference in the incidences of the outcomes between the hydration group and the non-hydration group.Conclusion:intravenous hydration shows no effect on the risk of PC-AKI,dialysis dependence at discharge and in-hospital mortality in patients with CKD stage 4-5 undergoing intravenous administration of ICM.It may not be necessary for these patients to have intravenous hydration around intravenous administration of ICM.