Association of preoperative cerebrospinal fluid sTREM2 concentration with postoperative delirium and three-year mortality in patients total knee or hip arthroplasty: a prospective cohort study.

BACKGROUND

As a common postoperative neurological complication, postoperative delirium (POD) can lead to poor postoperative recovery in patients, prolonged hospitalization, and even increased mortality. However, POD's mechanism remains undefined and there are no reliable molecular markers of POD to date. The present work examined the associations of cerebrospinal fluid (CSF) sTREM2 with CSF POD biomarkers, and investigated whether the effects of CSF sTREM2 on POD were modulated by the core pathological indexes of POD (Aβ42, tau, and ptau). The association of presurgical CSF sTREM2 with 3-year mortality in POD cases administered total knee or hip arthroplasty was assessed.

METHODS

We enrolled 545 Chinese Han patients undergoing total knee or hip arthroplasty (aged 50-95, weighing 50-80 kg, and using ASA II-III) combined with epidural anesthesia between October 2020 and March 2022. In these participants, POD was identified using the Confusion Assessment Method (CAM) and the severity of POD was evaluated using the Memorial Delirium Assessment Scale (MDAS) at 1-7 days postoperatively (or before discharge) by an anesthesiologist. The levels of cerebrospinal fluid (CSF) POD biomarkers were measured by ELISA. Next, logistic regression models were used to analyze the association between sTREM2and POD, as well as between cerebrospinal fluid (CSF) biomarkers and POD. We used Stata MP16.0. to examine whether the association between sTREM2 and POD was mediated by CSF POD biomarkers. We also used potential predictive factors to built 5 models, including Logistic Regression (LR), Support Vector Machine (SVM), K Nearest Neighbours (KNN), AdaBoost and CatBoost to assess the predictive abilities of sTREM2. After that, we verified the performance of the 5 models in the set, plotting receiver operating characteristic (ROC) curve analysis and precision recall curve (PRC) were used to further evaluate whether the machine learning (ML) models were effective in supporting clinical decision-making. All POD patients were followed up for three years, and Kaplan-Meier (K-M) survival analysis was used to compare the 3-year mortality rates of high sTREM2 group and low sTREM2 group in patients with POD.

RESULTS

Finally, a total of 545 patients (122patients in POD group and 423in NPOD group) were included in our study. sTREM2 and CSF levels of tau and ptau in the POD group were higher than those in the NPOD group. CSF Aβ42, Aβ42/ tau, and Aβ42/ ptau in the POD group were lower than those in the NPOD group. CSF sTREM2 was negatively associated with Aβ42 (r = - 0.445, P < 0.001), Aβ42/ tau (r = - 0.350, P < 0.001) and Aβ42/ ptau (r = - 0.429, P < 0.001), CSF sTREM2 was positively associated with tau (r = 0.179, P = 0.048) and ptau (r = 0.311, P < 0.001). The relationship between sTREM2 and POD was partially mediated by tau and ptau, with the mediation proportion of 17.91% and 22.09%, respectively. The following 5 variables (sTREM2, age, tau, ptau and Aβ42/ptau) were significant predictive factors via Lasso regression. Meanwhile, univariable analysis demonstrated CSF Aβ42/ptau levels was the protective factor of POD and sTREM2, age, tau, ptau were the risk factors of POD. Upon adjusting for possible confounders, including education level, sex, MMSE score, as well as history of diabetes, smoking, drinking, and hypertension, multivariable analysis showed consistent results. Following two rounds of sensitivity analysis, our results remained robust.The ROC(AUC = 0.999,95%CI:0.999-1.000) and PRC(AUC = 0.998,95%CI:0.995-1.000) for CatBoost were significantly better than the other models. The dynamic online calculator can accurately predict the occurrence of POD by selecting POD patients for the internal validation study. The Kaplan-Meier curve showed no significant difference in survival probability between the low sTREM2 group and high sTREM2 group (log-rank P = 0.53), but age subgroup analysis revealed significantly between age≥80 plus sTREM2 ≥ 20 000 pg/ml subgroup and the other subgroups on mortality in patients with POD (log-rank P = 0.017).

CONCLUSION

Elevated CSF sTREM2 is a preoperative risk factor for POD, which is partially mediated by tau and ptau. The CatBoost model can accurately predict the occurrence of POD. Age≥80 plus sTREM2 ≥ 20 000 pg/ml could increase 3-year mortality in POD cases.