Construction Of Early Alert System For Intradialytic Hypotension Before Initiating Hemodialysis Based On Artificial Intelligence

The application of artificial intelligence in predicting intradialytic hypotension: a systematic reviewObjective: Intradialytic hypotension is a serious complication of hemodialysis,which is associated with all-cause and cardiovascular mortality.The occurrence of IDH is related to many parameters,such as the comorbidities of patients and dialysis-related factors.AI allows the unification of different types of patient data into the same algorithm for training and validating models to predict disease states or prognosis.At present,some early warning models have been used to predict the risk of IDH,but there are shortcoming,such as poor reporting quality,inaccurate prediction results,and limited clinical application.We aim to provide an overview of the performance of these models based on AI for predicting IDH and to discuss the possibility of clinical application of these models.Methods: Construct a search using medical subject headings to search OVID,Pub Med,and Web of Science databases,and use PROBAST to evaluate the bias of alert models in development or external validation..Results: Machine learning has been applied to predict IDH risk.Several models have been developed in recent years,but there is currently no randomized clinical trial to verify them.Conclusion: There are already some early warning models for predicting the risk of IDH,which show good performance in internal verification.But they have some limitations,such as poor reporting quality,inaccurate prediction results,and limited clinical application.Prediction and evaluation of IDH risk in hemodialysis patients by nomogram modelObjective: To predict the risk of IDH in hemodialysis patients by constructing a nomogram.Methods: Hemodialysis Patients in Sichuan Provincial People’s Hospital from 2014 to 2020 were included in this study.The patients’ basic information,laboratory examinations and dialysis treatment records were collected,and a multivariate logistic regression model was designed.Apply R language to the training set data to build a nomogram model to predict the risk of IDH,the Bootstrap method was used for validation.The AUROC and the internal validation C-index were calculated to evaluate the performance of the model.The stability of the prediction model was judged by the calibration curve between the predicted results and the actual results.Results: A total of 3906 patients and 314,534 dialysis treatment records were included.After regression analysis,age,sex,pre-dialysis systolic blood pressure,pre-dialysis diastolic blood pressure,heart rate,preliminary ultrafiltration volume,hemoglobin,and blood calcium were identified and entered into the nomogram.The nomogram showed good discrimination,the AUROC was 0.733(95%CI 0.700-0.765),the internal validation C-index was 0.730,and AUROC for the test set was 0.727(95%CI 0.724-0.731).The calibration curve shows the agreement between the probabilities predicted and the actual probabilities.Conclusion: The nomogram we constructed can be used to identify high-risk patients with IDH,and it is helpful for early personalized interventions,suggesting that nomogram may have clinical utility.The model may be potential to help physicians make treatment recommendations.Construction of an early alert system for intradialytic hypotension before initiating hemodialysis based on artificial intelligenceObjective: To use artificial intelligence to establish an early warning system before hemodialysis to identify high-risk patients for IDH.Methods: A total of 314,534 hemodialysis records were included in the 2014-2020 Renal Disease Treatment Information System of Sichuan Provincial People’s Hospital.IDH was defined as a ≥20 mm Hg drop in systolic blood pressure during dialysis,a ≥10 mm Hg drop in mean arterial pressure,or a clinical hypotensive event requiring nursing intervention.After preprocessing,the data was randomly divided into training set(80%)and test set(20%).An early warning model was constructed using four filling methods,three feature selection methods,and 18 machine learning algorithms.AUROC was the main indicator to evaluate the performance of the model,while SHAP was used to explain the best prediction for each variable.Contribution of the model.Results: A total of 3906 patients and 314,534 dialysis records were included,of which 142,237 developed IDH(incidence rate 45.2%).19 parameters were identified through AI feature screening,including age,pre-dialysis weight,dry weight,pre-dialysis blood pressure,heart rate,prescribed ultrafiltration,blood counts(neutrophils,monocytes,lymphocytes,eosinophils,platelet count),hematocrit,serum calcium,creatinine,urea,glucose,and uric acid.216 models were constructed using 18 machine learning algorithms,among which RF(random forest),GB(gradient boosting),and logistic regression were the three best performing models,with AUROC of 0.812(95% CI),0.811-0.813),0.748(95% CI,0.747-0.749),and 0.743(95% CI,0.742--0.744).Conclusion: Our dialysis software-based artificial intelligence alert system can be used to predict IDH occurrence,there by contributing to initiate relevant interventions.