Aimed to investigate perioperative disorders associated to neurocognition, the PNDABLE research seeks to comprehend their pathogenesis, risk elements, and biomarkers, with a focus on identifying PND lifestyle risk factors among the Han people in the north of China. The study holds significant scientific and pragmatic utility for generating a standardized model for early prevention and diagnosis of these conditions in the non-demented Han individuals. Before the preoperative cerebrospinal fluid and blood extraction, all participants provided informed consent. The study was entered in the Chinese Clinical Trial Registry under the number ChiCTR2000033439 and has gained approval from the Ethics Committee of Qingdao Municipal Hospital.

In the PNDABLE study, Han Chinese individuals between the ages of 50 and 90 years, weighing 50–80 kg, and classified as ASA I–II, who received unilateral total knee arthroplasty with epidural anesthesia at Qingdao Municipal Hospital in the period between September 2020 to January 2023 were included as participants. The study had specific exclusion criteria, which were as follows: (1) Participants with a preoperative MMSE score lower than 24 points. (2) Individuals with a history of drug abuse or psychotropic drug abuse, as well as those on extended periods of steroid or hormone drug usage. (3) Patients with pre-operative stage III–IV liver encephalopathy. (4) Those who had not long past complex surgeries. (5) Individuals with considerable disabilities to see or hear. (6) Patients who showed to coagulate irregularly prior to the operation. (7) Subjects with multiple sclerosis, head injury, infections in the central nervous system, neurodegenerative conditions excluding AD (such as epilepsy, Parkinson's Disease), or other significant neurological disorders. (8) Individuals diagnosed with substantial psychological disorders. (9) Patients suffering from critical systemic diseases (e.g., malignant cancer) that could potentially influence CSF or blood values of AD biomarkers, including Aβ and tau. (10) Subjects with family records of genetic conditions.

From the PNDABLE study, 500 participants with normal cognitive function contributed data on various covariates. These individuals were then classified into two groups: POD and NPOD, according to the occurrence or absence of postoperative cognitive dysfunction. The patient recruitment process is depicted in the flow chart of Figure 1.

To prepare for the procedure, all participants received specific instructions not to consume any food for 8 h and to refrain from drinking for 6 h prior to the scheduled time. Upon admission to the operating room, the participants received a combined spinal-epidural block, targeting the puncture location at the gap between lumbar 3–4 spinous processes (L3–4). Following successful puncturing, 2 ml of cerebrospinal fluid was taken from the subarachnoid space, and ~2–2.5 ml of ropivacaine (0.66%) was injected over a period of nearly 30 s. To sustain stable blood pressure during the surgical procedure, oxygen was administered through a mask at a flow rate of 5 L/min, with the goal of keeping the blood pressure in the range of +/– 20% of the baseline value. If the non-invasive blood pressure (NBP) fell below 90 mmHg (1 mmHg = 0.133 kPa) or was reduced higher than 20% of the baseline level, a 5 mg intravenous infusion of ephedrine was administered to the patient. Similarly, if the heart rate (HR) dropped below 50 beats per minute, a 0.5 mg intravenous infusion of atropine was given. For managing acute postoperative pain, an intravenous patient-controlled analgesia approach was employed. This included a mixture of butorphanol (0.1 mg/ml) and tropisetron (50 g/ml), which was diluted with normal saline until reaching a volume of 100 ml. Subsequent to the surgery, patients were transferred to the post-anesthesia care unit (PACU) for further observation and care.

Twenty-four hours before the surgery, comprehensive interviews were conducted with all patients to obtain their baseline data, encompassing age, gender, body mass index (BMI), and years of education. Moreover, relevant information, including medical history and comorbidities, was meticulously gathered from the patients' medical records. The preoperative cognitive status was assessed by neurologists. The task of conducting medical history collection, physical evaluations, and cognitive assessments related to dementia was undertaken by an anesthesiologist.

The levels of serum albumin and serum prealbumin were measured from fasting venous blood within 24 h after admission. Analyses were performed by Olympus AU2700 automatic biochemical Analyzer.

Following the standard lumbar puncture, CSF samples were promptly processed within a 2-h window. Each sample underwent centrifugation at 2,000 × g during 10 min. The CSF samples underwent separation and were then stored in an enzyme-free EP (Eppendorf) tube (AXYGEN; PCR-02-C) at −80°C. These measures were taken under the international BIOMARKAPD project to facilitate their use in the following stages of this research project. CSF biomarkers of POD were quantified using ELISA with the aid of a microplate reader (Thermo Scientific Multiskan MK3). Specific ELISA kits, including Aβ₄₂ (BioVendor, Ghent, Belgium Lot: No.296-64401), T-tau (BioVendor, Ghent, Belgium Lot: No. EK-H12242), and P-tau (BioVendor, Ghent, Belgium Lot: QY-PF9092), were employed for this purpose. To ensure accuracy and consistency, all ELISA measurements were carried out by senior technicians who strictly adhered to the manufacturer's instructions. These technicians were kept blinded to the clinical data during the measurement process. To minimize variability between batches, each sample and standard underwent duplicate measurements, and the mean of the duplicates was utilized for subsequent statistical analyses. Each antibody and plate used in the measurements originated from a single lot. Additionally, the within-batch coefficient of variation (CV) was maintained below 5%, while the inter-batch CV was kept below 15%, ensuring the reliability of the results.

Neurologists performed a preoperative cognitive evaluation using the Mini-Mental State Examination (MMSE). Patients with an MMSE score below 24 points were excluded from the study.

Delirium assessment took place twice daily, between 9:00–10:00 am and 2:00–3:00 pm, starting from the first day until the seventh day after the operation (or prior to dismissal). An anesthesiologist was responsible for conducting the delirium assessment during these time slots. Simultaneously, the visual analog scale (VAS) was used to measure pain levels on a scale from 0 to 10, with lower scores indicating less pain Postoperative delirium (POD) was identified through the Confusion Assessment Method (CAM) (17), while the Memorial Delirium Assessment Scale (MDAS) was employed to measure the severity of POD (18). Furthermore, cognitive capabilities were evaluated within 6 months using the modified Telephone Interview for Cognitive Status (TICS-m). To gauge the quality of life, the World Health Organization Quality of Life brief version (WHOQOLBREF) was utilized.

A preliminary test revealed that four covariates were expected to be included in the Logistic regression model. Based on the assumption of a 20% rate of loss to follow-up and a POD incidence of 10%, the necessary sample size was calculated to be 500 participants. This sample size was determined to ensure sufficient statistical power and validity for the study's objectives. In this study, we tested normality using the Kolmogorov–Smirnov test. For measurement data following a normal distribution, it was represented as mean ± standard deviation (SD). On the other hand, for measurement data exhibiting a skewed distribution, we used the median (p25, p75) or a number (%) to describe the data. To compare the baseline characteristics between the POD group and the NPOD group, we employed the chi-square test for categorical variables while the Mann–Whitney U-test was used for continuous variables. These statistical tests were chosen based on the nature of the data and are suitable for analyzing the differences between the two groups effectively. Normality was assessed with the Kolmogorov–Smirnov test. Measurement data that conformed to a normal distribution were presented as mean ± standard deviation (SD); on the other hand, measurement data with a skewed distribution were expressed as the median (p25, p75) or in percentage form (%).

First, we analyzed the risk and protective factors for POD using the multivariate logistic regression. In order to eliminate the influence of confounding factors, we conducted multiple sensitivity analyses in this study and three correction models were constructed: (1) age (50–90 years), sex, years of education and MMSE scores were included as covariates for correction; (2) age (50–90 years), sex, years of education, MMSE scores, as well as the history of smoking, drinking, hypertension and diabetes were included as covariates for correction; and (3) age ≥65 years, sex, years of education, MMSE scores, as well as the history of smoking, drinking, hypertension and diabetes were included as covariates for correction.

Second, assess the predictive capabilities of serum prealbumin or serum albumin and the combination of serum prealbumin + serum albumin + CSF POD biomarkers for POD, we utilized the receiver operating characteristic (ROC) curve and the precision-recall curve (PRC). Additionally, we employed Calibration to validate the prediction model and developed a nomogram that incorporated these independent variables for further analysis.

Third, we sought to explore whether the relationship between serum prealbumin or serum albumin and POD was influenced by CSF POD biomarkers by means of Stata MP16.0 (Solvusoft Corporation, Inc, Chicago, Illinois, USA). We utilized logistic regression models based on specific methods for this analysis. The first equation involved regressing the mediator variable (CSF POD biomarkers) on the independent variable (serum prealbumin or serum albumin). Subsequently, in the second equation, we regressed the dependent variable (POD) on the independent variable (serum prealbumin or serum albumin). Finally, in the third equation, we regressed the dependent variable on both the independent variable and the mediator variable. To estimate the attenuation or indirect effect, we conducted 10,000 bootstrapped iterations while controlling for age, years of education, gender, and MMSE in each path of the model.

Finally, we conducted an internal validation study to verify the predictive accuracy of the combined use of serum prealbumin and serum albumin + CSF POD biomarkers for POD. The model was visualized using ROC curve and clinical decision curve analysis (DCA). DynNom and Shiny packages were exploited to generate an online calculator (https://www.shinyapps.io/). Ten patients with POD occurring from February 2023 to October 2023 were selected for internal validation.

SPSS statistical software, version 25.0 (SPSS, Inc. Chicago, IL, USA), R software version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria), Stata MP16.0 (Solvusoft Corporation, Inc, Chicago, Illinois, USA) and GraphPad Prism version 7.00 (GraphPad Software, San Diego, CA) were used for statistical analyses and figure preparation. P-value < 0.05 was considered significant, except where specifically noted.

We observed a POD incidence of 23.1% (n = 84 of the 364 patients) The POD group and NPOD group displayed significant differences (P < 0.05) in serum prealbumin and serum albumin and various CSF biomarkers, including Aβ₄₂, T-tau, P-tau, Aβ₄₂/T-tau, and Aβ₄₂/P-tau, as shown in Figure 2. Interestingly, there were no statistically significant variations in the preoperative MMSE score and the cognitive function at the 6-month postoperative follow-up visit between the two groups. A comprehensive overview of the participants' demographic and clinical data can be found in Table 1.

CSF levels of T-tau and P-tau emerged as significant risk factors for POD. Conversely, the univariate analysis revealed that serum prealbumin and serum albumin and CSF levels of Aβ₄₂, Aβ₄₂/T-tau, and Aβ₄₂/P-tau were protective factors against POD. After adjusting for potential confounding variables such as age, years of education, gender, MMSE score, as well as history of diabetes, smoking, drinking, and hypertension, the multivariate logistic regression assessment maintained the consistency of these results. Furthermore, sensitivity analyses were conducted, and the outcomes corroborated the findings, as shown in Table 2.

As shown in Figure 4, the relationship between serum prealbumin and POD was found to be partially mediated by T-tau and P-tau, with their mediation proportions of 12.28 and 20.61%, respectively. The relationship between serum albumin and POD was found to be partially mediated by T-tau and P-tau, with their mediation proportions of 18.28 and 21.55%, respectively (Figure 3).

ROC curve showed that the model combining serum prealbumin and serum albumin and POD biomarkers (AUC = 0.957; P < 0.001) exhibited a better discriminatory ability to predict POD compared to serum prealbumin alone (AUC = 0.836; P < 0.001), serum albumin alone (AUC = 0.848; P < 0.001) and the combination of serum prealbumin and serum albumin (AUC = 0.934; P < 0.001) (Figure 4). PRC curve showed that the combination of serum prealbumin and serum albumin and POD biomarkers (AUC = 0.854) exhibited a better discriminatory ability to predict POD compared to serum prealbumin alone (AUC = 0.496), serum albumin alone (AUC = 0.547) and the combination of serum prealbumin and serum albumin (AUC = 0.770; Figure 5).

A calibration plot comparing the prediction of POD between the model and actual observation was created with the result of the Hosmer–Lemeshow test (P = 0.563), indicating good predictive accuracy (Figure 6A). We established a nomogram incorporating these independent variables (Figure 6B). In addition, to assess the clinical usefulness of the nomogram, the R software was used to plot the clinical DCA. The DCA provided insight into the range of predicted risks, and the results showed that the model delivered a high predictive value for POD patients (Figure 7) and then translated it to a dynamic online calculator (https://byl-qdsslyy.shinyapps.io/dynnomapp-1) aiming at promoting the clinical utility of our findings. The dynamic online calculator can accurately predict the occurrence of POD by selecting POD patients for the internal validation study (Figure 8).