Abstract:Objective Through mining and analyzing the routine test data to develop its residual value and provide auxiliary predictive value for acute ischiic stroke (AIS) complicated with stroke-associated pneumonia (SAP). Methods AIS patients admitted to a hospital from June 2019 to June 2021 were retrospectively analyzed, divided into SAP group and non-SAP group according to whether they were complicated with SAP, and subdivided into trai-ning set and testing set at a ratio of 7 ∶3. AIS patients admitted to hospital from July 2021 to June 2022 were collected as the validation set. SAP-related test parameters were screened by the least absolute shrinkage and selection operator (LASSO). Nomogram model of the combined prediction was constructed and validated with training set, testing set and verification set. Discrimination and calibration of prediction model were assessed by receiver operating chara-cteristic (ROC) curve and calibration curve respectively, clinical practicability was assessed by decision curve analysis (DCA). Nomograph model was arranged into a web calculator to improve the clinical practical value. Results A total of 379 patients with AIS were taken as the basic population of the study, including 42 cases (incidence 11.08%) in SAP group. According to the 7 ∶3 distribution method, 265 cases were divided in training set and 114 in testing set.157 cases of AIS admitted from July 2021 to June 2022 were used as validation set, including 24 cases (incidence 15.29%) in SAP group. Five test parameters related to SAP were screened out by LASSO, namely neutrophil, lymphocyte, prealbumin, fibrinogen, and D-dimer. The calibration curve showed good calibration that the predicted probability of training set and testing set was consistent with the actual probability. DCA curve showed that the high risk threshold of training set was 0-0.75 and the net benefit was 0-0.11. The high risk threshold of testing set was 0-0.65, the net benefit was 0-0.11, with good clinical practicability. ROC curve showed that area under curve (AUC) predicted by full data set was 0.924, the sensitivity and specificity were 83.33% and 87.24% respectively. AUC predicted by training set was 0.922, the sensitivity and specificity were 79.31%, and 91.95% respectively. AUC predicted by testing sets was 0.919, sensitivity and specificity were 84.62% and 86.14% respectively, all of which had good discrimination performance. AUC predict by validation set was 0.850, sensitivity and specificity were 66.67% and 89.47% respectively. The model has good external applicability. The web calculator was arranged at https://ww-rstudiomn.shinyapps.io/SAP-nomgram/, which can be accessed via QR code. Test showed that the performance was stable. Conclusion The mining of routine test data provides a clinical value for the prediction of AIS complicated with SAP, thus provides a basis for early treatment and intervention.