Early predictors of high flow oxygen treatment failure for post-operation patients with hypoxemia / 中华急诊医学杂志

ABSTRACT

Objective:To explore the early predictors of high flow oxygen treatment failure for post-operation patients with hypoxemia.Methods:The post-operation adult patients with hypoxemia (100 mmHg<PaO 2/FiO 2≤300 mmHg) received high flow nasal cannula (HFNC) oxygen were retrospectively screened in the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. The patients were assigned to the treatment success or failure group according to whether receiving reintubation with 48 h after extubation. The risk factors of 48-h reintubation were screened and analyzed by extreme gradient boosting (XGBoost) algorithm. And the predictors were formulated according to the risk factors. The changes of predictors were collected from extubation to 48 h later. The predictors were compared at different time points after extubation between patients weaning successfully and failed with t test. The values at different time after extubation were also compared to the baseline data. The areas under the receiver operating characteristic (ROC) curve were calculated for 48-h reintubation prediction according to values at 4 h before and after extubation, which were compared with those of rapid shallow breathing index (RSBI) and ROX index. The RSBI was defined as the ratio of respiratory rate to tidal volume. The ROX index was defined as the ratio of SpO 2/FiO 2 to respiratory rate. Results:A total of 524 520 medical records were screened and 318 patients were included. There were 38 patients (11.95%) received reintubation within 48 h. According to the XGBoost model, the important features of 48-h reintubation were the duration before extubation, body mass index, simplified acute physiology scoring II, heart rate (HR), PaO 2, mean blood pressure, tidal volume, age, SpO 2 and respiratory rate. Thus HR/PaO 2 and HR/SpO 2 were formulated as predictors for 48-h reintubation according to the above features. The areas under the ROC of HR/PaO 2 and HR/SpO 2 were 0.640 and 0.617 for 48-h reintubation prediction according values at 4 h before extubation, which were larger than those of RSBI (0.537) and ROX index (0.539). According values at 4 h after extubation, the area under the ROC of HR/SpO 2 was 0.657, which was larger than that of ROX index (0.587). When the HR/SpO 2 reached 1.2 at 4 h after extubation, the specificity for 48-h reintubation was 92%. There was significant difference of HR/SpO 2 at 4 h after extubation between patients weaning successfully and failed (1.02 vs 0.92, P<0.05), and no significant difference of ROX index at the same time (8.14 vs 9.27, P>0.05). There were significant differences of HR/SpO 2 and ROX index at 8 to 12 h after extubation between the two groups (both P<0.05). Conclusions:HR/SpO 2 is more early and accurate in predicting HFNC failure than ROX index for post-operation patients with hypoxemia. However, both the predictors should be further evaluated.