Construction and validation of a predictive model based on machine learning algorithm for the risk of subsyndromic delirium after hip arthroplasty in the elderly

Construction and validation of a predictive model based on machine learning algorithm for the risk of subsyndromic delirium after hip arthroplasty in the elderly | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Construction and validation of a predictive model based on machine learning algorithm for the risk of subsyndromic delirium after hip arthroplasty in the elderly Li Zhang, Juanqi Li, Fangyuan Zhang, Ziru Zhang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8504448/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Background: subsyndromal delirium (SSD) can gradually develop into complete delirium, leading to prolonged hospital stays, increased risks of complications and death. This study aims to construct and validate a postoperative subsyndromal delirium risk prediction model for elderly hip arthroplasty patients using machine learning algorithms based on our electronic health record data. Methods: Electronic data of older adults who underwent hip arthroplasty at Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University from January 2020 to December 2023 were retrospectively analyzed. The data were divided into a training set (n=442) and a validation set (n=190) according to a ratio of 7:3, and four machine learning algorithms: logistic regression (LR), extreme gradient boosting (XGB), Gaussian Naive Bayes (GNB), and random forests (RF) were used to construct a predictive model of postoperative SSD. The area under the receiver operating characteristic curve (AUROC), sensitivity, accuracy, and area under the precision-recall curve (AUPRC) were used to evaluate the above four prediction models, and the variables were selected based on the SHAP values for the prediction model based on the optimal machine learning algorithm. Results : A total of 632 elderly patients who underwent hip arthroplasty were included in this study, of which 265 (41.93%) developed postoperative SSD, with 23 (3.64%) progressed from SSD to POD. After a comprehensive analysis, we ultimately chose the optimal-performing XGB-based algorithm to develop the prediction model. The first 6 variables selected based on SHAP values were: operative time, ASA classification, cumulative duration of intraoperative hypotension, previous delirium, age, and postoperative use of analgesics. The postoperative SSD prediction model developed based on these 6 variables had an AUROC of 0.849 (0.769-0.929), and an AUPRC of 0.657 (0.600-0.714). In the external test set, the AUROC was 0.842 (0.754-0.930) and the AUPRC was 0.650 (0.621-0.679). Conclusion: We used machine learning techniques to select six variables and demonstrate a predictive model for the risk of postoperative SSD in older adults with hip arthroplasty. The model would provide a useful tool for identifying patients at high risk for postoperative SSD, aiding in the identification of those with elevated postoperative SSD risk. machine learning hip arthroplasty subsyndromal delirium elderly risk prediction model verification Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction Delirium is an acute, fluctuating cognitive disorder whose main features include inattention, decreased level of consciousness, cognitive dysfunction and fluctuating affect [ 1 ] . Delirium is often misdiagnosed as other psychiatric disorders because of its similarity in aetiology and symptoms to dementia. However, unlike other disorders, the symptoms of delirium can be effectively relieved by controlling the causative agent [ 2 ] . Postoperative delirium (POD) is a common surgical complication in the elderly population, with a prevalence of approximately 5.10%-21.37% in the patient population undergoing hip fracture surgery [ 3 – 4 ] . In recent years, with the aging of the population, the group of elderly hip fracture surgery patients has been increasing, and clinical workers have paid more attention to the hazards, risk factors and prevention of POD. POD can cause prolonged hospitalisation, impaired functional recovery, increased healthcare costs and increased risk of readmission, which can affect surgical outcomes and the recovery process, and even lead to life-threatening diseases and death [ 5 ] . The clinical picture is characterised by acute post-operative seizures, altered level of consciousness, inattention and thought disorders [ 6 ] . However, some patients present with only some of the symptoms of delirium, but not enough to meet the traditional diagnostic criteria for delirium, and are referred to as having subsyndromal delirium (SSD) [ 7 – 9 ] . SSD is defined as a state with one or more symptoms of delirium or a precursor state of clinical delirium, but not meeting all the diagnostic criteria for delirium [ 8 – 9 ] . Despite the relatively mild symptoms of SSD compared to delirium, its duration and long-term effects cannot be ignored and still have a significant impact on patients' recovery and long-term cognitive function. Compared with patients without delirium, patients with SSD have longer hospital stays, more severe cognitive impairment and higher mortality rates [ 10 ] . Studies have shown that the risk factors for SSD are similar to those for delirium, such as age, underlying disease, duration of preoperative fasting, use of psychotropic medications, duration of surgery, anesthesia modality, and postoperative pain management [ 11 – 12 ] . In addition, SSD is a dynamically progressive disorder and may gradually progress to full-blown delirium, which seriously affects patients' recovery and quality of life [ 13 – 14 ] . Therefore, it is necessary for healthcare professionals to identify the high-risk group of SSD at an early stage and provide effective interventions to slow its progression. Traditional assessment methods mainly rely on clinician experience and simple scoring systems, such as the Delirium Assessment Scale and the Confusion Assessment Method (CAM), which can identify high-risk patients to some extent, but suffer from subjectivity and low sensitivity [ 15 ] . In recent years, the use of machine learning techniques in medical prediction has increased, and many studies have investigated the use of machine learning algorithms to predict the risk of postoperative delirium in elderly patients [ 16 – 18 ] . These studies typically use large-scale electronic health record data, and feature selection and model training improve the accuracy and generalisability of the prediction. However, to date, there are few studies on postoperative SSD risk prediction models for elderly hip arthroplasty patients. Therefore, this study aims to construct and validate a postoperative SSD risk prediction model for elderly hip arthroplasty patients using machine learning algorithms based on our electronic health record data. 2. Methods 2.1 Study Subjects This retrospective study was conducted under the approval of the Review Committee of Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University (approval number: K20240219) and was exempted from the requirement of informed consent. We submitted the statistical analysis plan to Tangdu Hospital, the Second Affiliated Hospital of Air Force Military Medical University before accessing all the data. We followed the TRIPOD statement for reporting the development and validation of multivariate prediction models [19] . The researchers strictly followed the "Helsinki Declaration" in their work. 2.2 Inclusion criteria We used electronic data from older adults who underwent hip arthroplasty at Tangdu Hospital, the Second Affiliated Hospital of Air Force Military Medical University between January 2020 and December 2023. Inclusion criteria: (1) patients over 65 years of age; (2) patients undergoing hip arthroplasty for the first time; (3) American Society of Anesthesiologists (ASA) class I-IV; (4) patients with delirium assessment results.Exclusion criteria: (1) those undergoing emergency surgery; (2) those with preoperative cognitive deficits such as delirium; (3) those with unstable postoperative vital signs; (4) postoperative hospital stay <5 d. The flowchart for patient selection in the modeling queue is shown in Figure 1. 2.3 Outcome indicators The main observational index in this study was postoperative SSD. the Chinese version of the CAM was used to assess the occurrence of SSD [20] . The CAM is the most commonly used tool for detecting delirium and includes the four features of the DSM-5 diagnostic criteria: (1) acute onset or fluctuating course; (2) impaired attention; (3) disorganised consciousness; and (4) disorganised and disorganised thinking [20-21] . These features can effectively distinguish delirium from other types of cognitive impairment and help healthcare professionals to identify delirium quickly and accurately. This version has a sensitivity of 90%, specificity of 94% and high inter-rater reliability. POD is defined as individuals meeting criteria 1 + 2 + (3 or 4). SSD is defined as individuals exhibiting at least one additional CAM symptom but failing to meet POD diagnostic requirements [8] . In general, two uniformly trained, experienced orthopaedic nurses assessed patients' SSD and POD 1 d preoperatively and 1-5 d postoperatively, twice daily or as the patient's condition changed. We analysed the results of all assessments from the end of surgery until discharge. 2.4 Selection of exposure variables In this study, exposure variables were selected with reference to those already reported in other literature. The final exposure variables included general predictors (e.g. age, MMSE score), neuropsychological factors (e.g. previous delirium, depression), comorbidities (e.g. diabetes mellitus, chronic obstructive pulmonary disease), haematological indices (e.g. Hemoglobin), intraoperative indices (e.g. haemorrhage volume, operative time), and postoperative indices (e.g. use of analgesics). 2.5 Data analysis and pre-processing Our primary aim was to develop a predictive model for postoperative SSD using combined preoperative, intraoperative and postoperative data. The data were divided into a training set ( n= 442) and a validation set (n=190) according to a 7:3 ratio, with the training set used for model development and the validation set used for validation testing. We screened the missing rates of all potential predictors and excluded those with missing values ≥25%; predictors with missing values <25% were imputed. The missing rates for MMSE scores, frailty, previous history of delirium, and the lowest postoperative body temperature were 11.23%, 8.39%, 17.88%, and 12.97%, respectively. Assuming the data were missing at random, multiple imputation was performed using the mice package in R software. Five complete datasets were generated, with the imputation model incorporating all predictor variables as well as the outcome variable from this study. The final results for the missing data were derived from the average of the model outcomes across the five imputed datasets. 2.6 Model development In this study, we constructed a prediction model for postoperative SSD in elderly hip replacement patients based on four machine learning algorithms: logistic regression (LR), extreme gradient boosting (XGB), Gaussian naive Bayes (GNB), and random forest (RF), and compared their prediction performance. These methods were used to construct a prediction model for postoperative SSD in elderly hip arthroplasty patients, and their predictive performance was compared. Since the number of people who did not develop SSD after surgery was much larger than the number of people who developed SSD, there was an imbalance in the modelling data, so in this study hyperparameters were set to increase the weights of positive samples to solve the problem of data imbalance. We also performed parameter adjustment on the training set by the fivefold crossover method, which improved the learning effect of the model. 2.7 Model evaluation We used the area under the receiver operating characteristic curve (AUROC), sensitivity, accuracy, specificity and area under the precision-recall curve (AUPRC) to evaluate the above four prediction models. The AUPRC ranges from 0 to 1, focusing on the model's performance on the “positive” group. It typically provides a more informative assessment than AUC-ROC in highly imbalanced datasets. We also used SHAP summary plots to show the importance of variables in predicting SSD in the GBM model. The effect of each feature on postoperative SSD was expressed as SHAP values by deriving marginal distributions and weighted means, which were fixed for all variables except the variable of interest [22] . 2.8 External validation To confirm the validity of the model performance, external validation was performed using different datasets from Xijing Hospital, the First Affiliated Hospital of the Air Force University of Military Medicine, to validate the effectiveness of the final model in predicting SSD. 2.9 Statistical analysis R software (4.3.2), Python (3.4.3) language and SPSS 25.0 software were used to process the data. Measurement information that met the normal distribution test was expressed as (±) or M (P25, P75), and was compared using the independent samples t-test or Mann-Whitney U test. Count data were expressed as [n (%)], and the χ2 test or Fisher test was performed to compare the differences. p < 0.05 indicated that the differences were statistically significant. 3. Results 3.1 Baseline characteristics A total of 632 elderly patients who underwent hip arthroplasty at our institution were included in this study. Of all patients, 265 (41.93%) developed postoperative SSD, with 23 (3.64%) progressed from SSD to POD. The baseline characteristics of SSD and non- SSD patients are summarised in Table 1 .Patients with SSD were predominantly male, older, had a higher ASA classification, a higher rate of electrolyte abnormalities and were often accompanied by multiple comorbidities. Intraoperatively, SSD patients had longer operative time, more bleeding, higher rates of blood transfusion, and were more prone to hypotension and had longer duration of hypotension (Table 1 ). Postoperatively, SSD patients had a higher rate of analgesic use (Table 1 ). Table 1 Baseline characteristics of SSD and non- SSD patients items non- SSD(n = 367) SSD(n = 265) t/ χ 2 / Z P Sex[n(%)] 22.479 <0.001 Male 157(42.78) 164(61.89) Female 210(57.22) 101(38.11) Age (years) 72.51 ± 6.74 80.16 ± 10.35 10.038 <0.001 Anaesthesia[n(%)] 1.825 0.177 General anesthetic 167(45.50) 135(50.94) Intravertebral anaesthesia 200(54.50) 130(49.06) ASA[n(%)] 79.325 <0.001 Ⅰ 144(39.24) 35(13.21) Ⅱ 184(50.14) 139(52.45) Ⅲ 36(9.81) 81(30.57) Ⅳ 3(0.82) 10(3.77) BMI(kg/m 2 ) 25.52 ± 3.71 24.98 ± 4.29 1.455 0.146 Complication [n(%)] Diabetes 72(19.62) 89(33.58) 15.811 <0.001 Hypertensive 166(45.23) 157(59.25) 12.094 <0.001 Rheumatoid 33(8.99) 31(11.70) 1.238 0.266 Coronary heart disease 47(12.81) 69(26.04) 17.977 <0.001 Arrhythmia 57(15.53) 71(26.79) 12.082 <0.001 Chronic obstructive pulmonary disease 6(1.63) 15(5.66) 7.763 0.005 Chronic Kidney Disease 9(2.45) 17(6.42) 6.126 0.013 Previous delirium 7(1.91) 33(12.45) 28.866 <0.001 Debility 18(4.90) 34(12.83) 12.801 <0.001 Alcohol consumption[n(%)] 49(13.35) 48(18.11) 2.686 0.101 Smoking[n(%)] 77(1.580) 45(16.98) 1.580 0.209 Preoperative MMSE score (points) 26.00(23.00, 29.00) 24.00(22.00, 26.00) 7.129 <0.001 Preoperative haematological indices Hemoglobin(g/dL) 13.31 ± 1.87 12.25 ± 2.19 6.541 <0.001 Creatinine(mg/dL) 0.93 ± 0.81 1.32 ± 1.56 4.087 <0.001 Albumin(g/dL) 36.52 ± 4.85 32.17 ± 5.03 10.954 <0.001 Hypokalaemia 15(4.09) 31(11.70) 13.208 <0.001 Hyperkalaemia 7(1.91) 25(9.43) 18.135 <0.001 Hyponatraemia 11(3.00) 41(15.47) 31.731 <0.001 Hypernatraemia 3(0.82) 3(1.13) 0.162 0.687 Hypocalcaemia 35(9.54) 57(21.51) 17.735 <0.001 Hypercalcaemia 8(2.18) 1(0.38) 3.562 0.059 Intraoperative indicators Operative time(min) 112.50(85.00, 125.50) 135.00(97.50, 149.00) 11.256 <0.001 Hypotension[n(%)] 139(37.87) 142(50.53) 15.381 <0.001 Cumulative duration of hypotension(min) 0(0, 2) 4(1, 6) 14.561 <0.001 Haemorrhage volume(mL) 105.00(62.00, 200.00) 195.00(100.00, 250.00) 8.015 <0.001 Blood transfusion[n(%)] 18(4.90) 39(14.72) 18.056 <0.001 Postoperative indicators Use of analgesics[n(%)] 88(23.98) 98(36.98) 12.527 <0.001 Minimum body temperature(℃) 36.52 ± 4.59 35.90 ± 6.10 1.278 0.202 ASA: American Society of Anaesthesiologists; BMI: body mass index. 3.2 Development of Prediction Models The AUROC of the LR, XGB, GNB and RF algorithms were 0.826 (0.734–0.917), 0.891 (0.824–0.958), 0.862 (0.772–0.951), and 0.862 (0.773–0.950), respectively (Fig. 2 ). In terms of AUROC, accuracy, F1 score, and AUPRC, the XGB, GNB, and RF models showed similar performance (XGB: 0.891, 0.859, 0.421, 0.731; GNB: 0.862, 0.836, 0.619, 0.724; LR: 0.826, 0.799, 0.573, 0.635; RF: 0.862, 0.871, 0.497, 0.668), and the specific performance metrics are shown in Table 2 .Considering that the AUROC and AUPRC of XGB are higher than those of GNB and LR models, we finally chose the prediction model based on the XGB algorithm. Table 2 Performance metrics of different machine learning algorithms for predictive models Classification model AUROC (95%CI) AUPRC (95%CI) Accuracy (95%CI) Specificity (95%CI) Sensitivity (95%CI) NPV (95%CI) PPV (95%CI) XGBoost 0.891 (0.824–0.958) 0.731 (0.642–0.820) 0.859 (0.841–0.877) 0.854 (0.813–0.895) 0.870 (0.817–0.923) 0.925 (0.893–0.957) 0.761 (0.698–0.824) Logistic 0.826 (0.734–0.917) 0.635 (0.578–0.691) 0.799 (0.769–0.828) 0.794 (0.748–0.840) 0.808 (0.744–0.873) 0.899 (0.862–0.935) 0.648 (0.577–0.718) RandomForest 0.862 (0.773–0.950) 0.668 (0.630–0.707) 0.871 (0.843–0.898) 0.869 (0.830–0.909) 0.874 (0.823–0.926) 0.925 (0.893–0.957) 0.790 (0.730–0.850) GNB 0.862 (0.772–0.951) 0.724 (0.630–0.817) 0.837 (0.794–0.878) 0.856 (0.814–0.898) 0.801 (0.741–0.861) 0.872 (0.832–0.912) 0.778 (0.717–0.8403) 3.3 Development of predictive models using selected significant variables The XGB algorithm, an integrated learning model based on gradient boosting decision trees, achieves good performance on a variety of data sets, but it loses the interpretability of a linear model. In contrast, SHAP analysis intuitively explains the importance of the selected variables, in particular, the contribution of each feature to the model prediction is explained by the Shapley value [ 22 ] . We generated summary SHAP plots based on the results of the XGB model (only variables with SHAP > 0.05 are shown). In this case, the vertical coordinates are the factors of each variable, with darker red representing higher risk and darker blue representing lower risk, and the horizontal coordinates are the SHAP values. To make the prediction model more applicable to clinical practice, we selected the variables with SHAP value ranked in the top 6 to develop the final model. As shown in Fig. 3 , the top 6 variables were: duration of surgery (0.527), ASA classification (0.415), cumulative duration of intraoperative hypotension (0.327), previous delirium (0.306), age (0.219), and use of postoperative analgesics (0.201). We developed a new prediction model for postoperative SSD using these six selected variables and the new prediction model had an AUROC of 0.849 (0.769–0.929) (Fig. 4 ), an accuracy of 0.747 (0.678–0.806), an AUPRC of 0.657 (0.600-0.714), a specificity of 0.712 (0.633–0.791), and a sensitivity of 0.815 (0.721–0.910). 3.4 External validation of the prediction model We evaluated the new prediction model in an external test set, and the results show that the new XGB model achieves an AUROC of 0.842 (0.754–0.930) (Fig. 4 ), an accuracy of 0.799 (0.748–0.843), an AUPRC of 0.650 (0.621–0.679), a specificity of 0.779 (0.691–0.867), and a sensitivity of 0.808 (0.754–0.861), all of which are close to previous. 4. Discussion In this study, we developed and validated a predictive model for postoperative SSD in elderly hip arthroplasty patients. The incidence of postoperative SSD in the dataset was 41.93%. We selected six variables based on the XGB algorithm: operative time, ASA classification, cumulative duration of intraoperative hypotension, previous delirium, age, and postoperative analgesic use to develop the prediction model..The final developed prediction model achieved an AUROC value of 0.849 (0.769–0.929), which is a superior predictive performance. SSD is an acute state of clouded consciousness less severe than delirium, accompanied by mild emotional fluctuations, mild decrease in level of consciousness, mild changes in cognitive function, mild thought disorders and mild perceptual deficits [ 8 ] . Although most of these symptoms are reversible, their duration and long-term effects cannot be ignored and may progress to full-blown delirium, with serious and costly consequences such as prolonged hospitalisation, complications and increased risk of death [ 23 – 24 ] . Studies have shown that approximately 9.50%-20.70% of patients progress from SSD to full-blown delirium [ 25 – 26 ] . Our data showed that 8.68% of elderly hip arthroplasty patients progressed from SSD to full delirium, supporting that SSD is a precursor to full delirium and that the risk of clinical delirium is particularly high. Like POD, postoperative SSD is considered preventable and identification of those at risk is a prerequisite for modifying predisposing factors and implementing targeted interventions.In this study, we applied relatively strict criteria for the diagnosis of SSD, excluding patients with direct POD, thus distinguishing SSD from POD in the immediate postoperative period. Considering that advanced age is an independent predictor of postoperative SSD [ 12 , 27 ] , in this study we selected the clinical data of elderly hip arthroplasty patients for modelling, avoided the parameter estimation bias that might be caused by the direct use of whole-population data, improved the predictive accuracy of the model in elderly hip arthroplasty patients, and finally developed a prediction model that was more targeted and could effectively predict the occurrence of postoperative SSD in elderly patients. In addition, we validated the prediction model with data sets from different time periods, and the prediction model showed better predictive ability. Considering the convenience of data collection and the practicality of the model, this study optimised the XGB model constructed on the basis of 33 variables, and finally screened six variables that contributed the most to the XGB model and were clinically associated with SSD, and obtained a more simplified model, which made data collection and processing more efficient and was conducive to the dissemination and application of the prediction model. The occurrence of SSD is associated with a variety of factors, including age, underlying disease, duration of surgery, anaesthetic modality, postoperative pain management and medication use [ 12 , 27 – 28 ] . Several studies have proposed scoring systems based on traditional regression models to predict SSD [ 29 – 31 ] . However, traditional regression analysis methods are limited by the number of variables and are only applicable to the analysis of linear correlations. It is significantly less effective in dealing with non-linearly differentiable data, which in turn affects the predictive accuracy of the model. In this study, we used machine learning techniques, which can process and detect non-linear relationships and complex data structures, and can improve the reliability and accuracy of disease diagnosis systems [ 32 ] . After comparing several algorithms of machine learning techniques, we finally selected the XGB algorithm with optimal performance to construct a postoperative SSD prediction model for elderly hip replacement patients. In the medical field, the interpretation of the results of machine learning analyses should be done with full consideration of clinical applicability.SHAP scores as an interpretive tool can compensate for the problem of poor interpretability in machine learning by helping to understand the contribution of individual features to the predicted outcomes of the model, thus ensuring that the interpretation of the model meets the needs of clinical practice [ 22 , 26 , 33 ] . The SHAP analysis in this study showed that the factors influencing postoperative SSD in elderly hip arthroplasty patients were, in order of importance: duration of surgery, ASA classification, cumulative duration of intraoperative hypotension, previous delirium, age and use of analgesics in the postoperative period. Like delirium, SSD is caused by deterioration in homeostasis and physical status, which is more likely to occur in high-risk surgery of longer duration [ 13 , 34 ] . ASA grading is effective in assessing the preoperative health status of patients, and the higher the ASA grading, the worse the physiological function and health status of the patient, and the higher the risk of postoperative complications such as delirium [ 35 – 36 ] . Hypotension can lead to inadequate cerebral perfusion, the duration of which is closely related to damage to organs such as the heart and kidneys, and also increases the risk of postoperative SSD and POD [ 37 – 38 ] . Studies have shown that previous delirium is associated with a higher risk of postoperative SSD and POD [ 39 ] . In addition, age is the most widely accepted risk factor for SSD in any clinical situation [ 28 , 40 ] .Finally, postoperative analgesic use may affect cognitive function and the brain's response to stress [41] , and older patients receiving postoperative analgesics may be more prone to delirium. However, it has also been found that opioid analgesic regimens are not associated with postoperative delirium [42] . Our model is clinically interpretable with only six variables, is highly tractable and shows superior predictive performance. Specifically, the risk of postoperative SSD was higher in those with prior delirium than in those without prior delirium. The risk of postoperative SSD also increased with duration of surgery, cumulative duration of intraoperative hypotension and age, i.e. the longer the duration of surgery, the longer the cumulative duration of intraoperative hypotension and the older the patient, the higher the risk of postoperative SSD. The risk of delirium also increased with the ASA classification score and with increasing score. The risk of postoperative SSD was also higher in those who used postoperative analgesics than in those who did not. However, since this study did not explicitly distinguish the temporal sequence between postoperative analgesic use and SSD, a reverse causal relationship between postoperative analgesic use and SSD cannot be ruled out. Future studies will clearly document the timing of analgesic administration relative to SSD onset to further optimize predictive models and intervention strategies. Overall, the predictive model we developed aids in identifying patients at high risk for postoperative SSD, thereby enabling earlier intervention in the prevention and management of postoperative neurocognitive complications. Clinical healthcare providers can utilize the predictive model preoperatively to calculate postoperative SSD risk scores and flag high-risk patients in electronic medical record systems. During the intraoperative phase, efficient team collaboration should minimize surgical duration while prioritizing blood pressure management to prevent hypotensive events. Postoperatively, multimodal analgesia should be employed to avoid excessive opioid use, and delirium assessment tools should be routinely administered at least daily to enable early SSD detection. Furthermore, during the SSD phase, patients typically exhibit greater cooperation with milder disorientation and agitation. Introducing interventions such as cognitive training and early mobilization at this stage yields superior outcomes, preventing further deterioration and improving long-term functional recovery and quality of life. The following limitations must also be taken into account when interpreting our results. First, this is a retrospective study and causality cannot be confirmed. Second, the results cannot be generalised to other patient groups because this study included the entire Chinese population and did not include other races. In the future, we will conduct multicenter validation across different populations and healthcare settings to enhance the generalizability of this study's findings. Third, the data were unbalanced and the low prevalence of postoperative SSD affected the sensitivity of the model to some extent. However, this low prevalence is due to the fact that the diagnosis was only recorded when patients were specifically assessed for delirium status. As screening patients for postoperative SSD is not a clinically necessary intervention, many cases appear to be missed in the real world. This also justifies the clinical need for such a predictive model for screening for postoperative SSD and the need for future validation with other datasets. Despite the limitations, this study is the first to assess the risk of postoperative SSD in elderly hip arthroplasty using machine learning algorithms and a well-established predictive model, providing clinicians with an evaluation tool to identify high-risk individuals. However, the predictive model outputs the relative risk of individual patients developing SSD, aiming to assist healthcare providers in prioritizing preventive care resources. While the model demonstrates good performance, it is not perfect, and its clinical application should be combined with the judgment of clinicians. Future multi-center, prospective studies are needed to further validate and optimize this model.. 5. Conclusion We used machine learning techniques to select six variables and demonstrate a predictive model for the risk of postoperative SSD in elderly hip arthroplasty. The model provides a useful tool for identifying patients at high risk for postoperative SSD, aiding in the early detection of those with elevated risk. This enables the advancement of preventive measures for postoperative neurocognitive complications. Implementing interventions such as cognitive training and early mobilization for high-risk patients at an early stage may help prevent further deterioration of their condition. Declarations Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. Ethics statement This retrospective study was conducted under the approval of the Review Committee of Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University (approval number: K20240219) and was exempted from the requirement of informed consent. Author contributions Li Zhang designed the experiments, performed the study, and wrote the article;Juanqi Li performed the study and collected the data; Fangyuan Zhang performed the study, collected the data, and analysed/interpreted the data; Ziru Zhang: designed the experiments, revised the article, and supported the work. Conflict of interest All authors declare no competing interests. Funding Declaration There was no funding. References Wilson JE, Mart MF, Cunningham C, Shehabi Y, Girard TD, MacLullich AMJ, et al. Delirium. Nat Rev Dis Primers . (2020) 6:90. doi: 10.1038/s41572-020-00223-4. Ryan SL, Kimchi EY. Evaluation and Management of Delirium. Semin Neurol . (2021) 41:572-87. doi: 10.1055/s-0041-1733791. Li T, Li J, Yuan L, Jiang C, Daniels J, Mehta RL, et al. Effect of Regional vs General Anesthesia on Incidence of Postoperative Delirium in Older Patients Undergoing Hip Fracture Surgery: The RAGA Randomized Trial. JAMA . (2022), 327:50-8. doi: 10.1001/jama.2021.22647. Duan Y, Zhang R. Risk factors and prediction model of delirium in elderly patients after hip arthroplasty. Pak J Med Sci . (2024) 40:1077-82. doi: 10.12669/pjms.40.6.9306. Hughes CG, Boncyk CS, Culley DJ, et al. American Society for Enhanced Recovery and Perioperative Quality Initiative Joint Consensus Statement on Postoperative Delirium Prevention. Anesth Analg . (2020) 130:1572-90. doi: 10.1213/ANE.0000000000004641. Aldecoa C, Bettelli G, Bilotta F, Fleisher LA, Leung JM, McDonagh DL, et al. Update of the European Society of Anaesthesiology and Intensive Care Medicine evidence-based and consensus-based guideline on postoperative delirium in adult patients. Eur J Anaesthesiol . 2024, 41(2):81-108. Cole MG. Subsyndromal delirium in old age: conceptual and methodological issues. Int Psychogeriatr . (2013) 25:863-6. doi: 10.1017/S1041610213000434. Sepulveda E, Leonard M, Franco JG, Adamis D, McCarthy G, Dunne C, et al. Subsyndromal delirium compared with delirium, dementia, and subjects without delirium or dementia in elderly general hospital admissions and nursing home residents. Alzheimers Dement (Amst) . (2016) 7:1-10. doi: 10.1016/j.dadm.2016.11.002. Boettger S, Nuñez DG, Meyer R, Richter A, Schubert M, Jenewein J. Subsyndromal delirium in the intensive care setting: Phenomenological characteristics and discrimination of subsyndromal delirium versus no and full-syndromal delirium. Palliat Support Care . (2018) 16:3-13. doi: 10.1017/S1478951517000104. Shim J, DePalma G, Sands LP, Leung JM. Prognostic Significance of Postoperative Subsyndromal Delirium. Psychosomatics . (2015). 56:644-51. doi: 10.1016/j.psym.2015.05.002. Azuma K, Mishima S, Shimoyama K, Ishii Y, Ueda Y, Sakurai M, et al. Validation of the Prediction of Delirium for Intensive Care model to predict subsyndromal delirium. Acute Med Surg . (2018). 6:54-9. doi: 10.1002/ams2.378. Chen S, Tang L, Chen J, Cai L, Liu C, Song J, et al. Prevalence and risk factors of subsyndromal delirium among postoperative patients: A systematic review and meta-analysis. J Adv Nurs . (2024). 80:924-34. doi: 10.1111/jan.15871. Mailhot T, Cossette S, Maheu-Cadotte MA, Fontaine G, Denault AY. Subsyndromal Delirium in Cardiac Surgery Patients: Risk Factors and Outcomes of the Different Trajectories. J Cardiovasc Nurs . (2022) 37:41-9. doi: 10.1097/JCN.0000000000000793. Paulino MC, Conceição C, Silvestre J, Lopes MI, Gonçalves H, Dias CC, et al. Subsyndromal Delirium in Critically Ill Patients-Cognitive and Functional Long-Term Outcomes. J Clin Med . (2023). 12:6363. doi: 10.3390/jcm12196363. Oh ES, Fong TG, Hshieh TT, Inouye SK. Delirium in Older Persons: Advances in Diagnosis and Treatment. JAMA. (2017) 318:1161-74. doi: 10.1001/jama.2017.12067. Racine AM, Tommet D, D'Aquila ML, Fong TG, Gou Y, Tabloski PA, et al. Machine Learning to Develop and Internally Validate a Predictive Model for Post-operative Delirium in a Prospective, Observational Clinical Cohort Study of Older Surgical Patients. J Gen Intern Med . (2021) 36:265-73. doi: 10.1007/s11606-020-06238-7. Bishara A, Chiu C, Whitlock EL, Douglas VC, Lee S, Butte AJ, et al. Postoperative delirium prediction using machine learning models and preoperative electronic health record data. BMC Anesthesiol . (2022) 22:8. doi: 10.1186/s12871-021-01543-y. Song Y, Zhang D, Wang Q, Liu Y, Chen K, Sun J, et al. Prediction models for postoperative delirium in elderly patients with machine-learning algorithms and SHapley Additive exPlanations. Transl Psychiatry . (2024) 14:57. doi: 10.1038/s41398-024-02762-w. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med . (2015) 162:55-63. doi: 10.7326/L15-0078-4. Mei X, Chen Y, Zheng H, Shi Z, Marcantonio ER, Xie Z, et al. The Reliability and Validity of the Chinese Version of Confusion Assessment Method Based Scoring System for Delirium Severity (CAM-S). J Alzheimers Dis . (2019) 69:709-16. doi: 10.3233/JAD-181288. Inouye SK, van Dyck CH, Alessi CA, Alessi CA, Balkin S, Siegal AP, et al. Clarifying confusion: the confusion assessment method. A new method for detection of delirium. Ann Intern Med . (1990) 113:941-8. doi: 10.7326/0003-4819-113-12-941. Li X, Dvornek NC, Zhou Y, Zhuang J, Ventola P, Duncan JS. Efficient Interpretation of Deep Learning Models Using Graph Structure and Cooperative Game Theory: Application to ASD Biomarker Discovery. Inf Process Med Imaging . (2019) 11492:718-730. doi: 10.1007/978-3-030-20351-1_56. Cole MG, Bailey R, Bonnycastle M, McCusker J, Fung S, Ciampi A, et al. Frequency of full, partial and no recovery from subsyndromal deliriumin older hospital inpatients. Int J Geriatr Psychiatry . (2016) 31:544-50. doi: 10.1002/gps.4372. Paulino MC, Conceição C, Silvestre J, Lopes MI, Gonçalves H, Dias CC, et al. Subsyndromal Delirium in Critically Ill Patients-Cognitive and Functional Long-Term Outcomes. J Clin Med . (2023) 12:6363. doi: 10.3390/jcm12196363. Yamada C, Iwawaki Y, Harada K, Fukui M, Morimoto M, Yamanaka R. Frequency and risk factors for subsyndromal delirium in an intensive care unit. Intensive Crit Care Nurs . (2018) 47:15-22. doi: 10.1016/j.iccn.2018.02.010. Song C, Wang S, Liu J, Xu W, Yang Y, Li X. Predictive factors for the duration of subsyndromal delirium in the intensive care unit. J Clin Nurs . (2024) 33:3775-83. doi: 10.1111/jocn.17148. Chung KS, Lee JK, Park JS, Choi CH. Risk factors of delirium in patients undergoing total knee arthroplasty. Arch Gerontol Geriatr . (2015) 60:443-7. doi: 10.1016/j.archger.2015.01.021. Ma X, Cheng H, Zhao Y, Zhu Y. Prevalence and risk factors of subsyndromal delirium in ICU: A systematic review and meta-analysis. Intensive Crit Care Nurs . (2025) 86:103834. doi: 10.1016/j.iccn.2024.103834. Cheng J, Lao Y, Chen X, Qiao X, Sui W, Gong X, et al. Dynamic Nomogram for Subsyndromal Delirium in Adult Intensive Care Unit: A Prospective Cohort Study. Neuropsychiatr Dis Treat . (2023) 19:2535-48. doi: 10.2147/NDT.S432776. Gao Y, Gan X. A novel nomogram for the prediction of subsyndromal delirium in patients in intensive care units: A prospective, nested case-controlled study. Int J Nurs Stud . (2024) 155:104767. doi: 10.1016/j.ijnurstu.2024.104767. Qian J, Shen X, Gao X, Sun Q. Grip Strength is a Predictor for Subsyndromal Delirium Among Older Adults Following Joint Replacement. Clin Interv Aging . (2023) 18:1675-85. doi: 10.2147/CIA.S423727. Gabor JA, Feng JE, Schwarzkopf R, Slover JD, Meftah M. Machine Learning With Electronic Health Record Data Outperforms a Risk Assessment Prediction Tool in Predicting Discharge Disposition After Total Joint Arthroplasty. Orthopedics . (2022) 45:e211-e215. doi: 10.3928/01477447-20220225-02. Hsu WH, Ko AT, Weng CS, Chang CL, Jan YT, Lin JB, et al. Explainable machine learning model for predicting skeletal muscle loss during surgery and adjuvant chemotherapy in ovarian cancer. J Cachexia Sarcopenia Muscle . (2023) 14:2044-53. doi: 10.1002/jcsm.13282. Cregar WM, Goodloe JB, Gerlinger TL. Increased Operative Time Impacts Rates of Short-Term Complications After Unicompartmental Knee Arthroplasty. J Arthroplasty . (2021) 36:488-94. doi: 10.1016/j.arth.2020.08.032. Dodsworth BT, Reeve K, Falco L, Hueting T, Sadeghirad B, Mbuagbaw L, et al. Development and validation of an international preoperative risk assessment model for postoperative delirium. Age Ageing . (2023) 52:afad086. doi: 10.1093/ageing/afad086. Wang X, Yu D, Du Y, Geng J. Risk factors of delirium after gastrointestinal surgery: A meta-analysis. J Clin Nurs . (2023) 32:3266-76. doi: 10.1111/jocn.16439. Wachtendorf LJ, Azimaraghi O, Santer P, Linhardt FC, Blank M, Suleiman A, et al. Association Between Intraoperative Arterial Hypotension and Postoperative Delirium After Noncardiac Surgery: A Retrospective Multicenter Cohort Study. Anesth Analg . (2022) 134:822-33. doi: 10.1213/ANE.0000000000005739. Zarour S, Weiss Y, Abu-Ghanim M, Iacubovici L, Shaylor R, Rosenberg O, et al. Association between Intraoperative Hypotension and Postoperative Delirium: A Retrospective Cohort Analysis. Anesthesiology . (2024) 141:707-18. doi: 10.1097/ALN.0000000000005149. Gao Y, Gao R, Yang R, Gan X. Prevalence, risk factors, and outcomes of subsyndromal delirium in older adults in hospital or long-term care settings: A systematic review and meta-analysis. Geriatr Nurs . (2022) 45:9-17. doi: 10.1016/j.gerinurse.2022.02.021. Hwang H, Lee KM, Son KL, Jung D, Kim WH, Lee JY, et al. Incidence and risk factors of subsyndromal delirium after curative resection of gastric cancer. BMC Cancer . (2018) 18:765. doi: 10.1186/s12885-018-4681-2. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 25 Feb, 2026 Editor invited by journal 02 Feb, 2026 Editor assigned by journal 20 Jan, 2026 Submission checks completed at journal 20 Jan, 2026 First submitted to journal 02 Jan, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8504448","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":596686148,"identity":"4c393159-cd61-4b32-b157-31102da016f9","order_by":0,"name":"Li Zhang","email":"","orcid":"","institution":"Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University,No.1","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"Zhang","suffix":""},{"id":596686149,"identity":"580ece94-9c0f-4a57-9c85-40878d7b8827","order_by":1,"name":"Juanqi Li","email":"","orcid":"","institution":"Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University,No.1","correspondingAuthor":false,"prefix":"","firstName":"Juanqi","middleName":"","lastName":"Li","suffix":""},{"id":596686150,"identity":"480007ce-285b-4ba8-94e3-4d0ae463c2ac","order_by":2,"name":"Fangyuan Zhang","email":"","orcid":"","institution":"Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University,No.1","correspondingAuthor":false,"prefix":"","firstName":"Fangyuan","middleName":"","lastName":"Zhang","suffix":""},{"id":596686151,"identity":"be97dd6e-6750-42bf-acc2-52b712e9e363","order_by":3,"name":"Ziru Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyklEQVRIiWNgGAWjYBACfvbmAwc+GNgws7E3EKlFsudY4sMZBWns/DwHiNRicCPH2Jjnw2F+yRkJxLrszAEzyRkGadIGNx9vvMFQYxNNUAdje0OaBNAvxga304otGI6l5TYQ0sLMc+AYyJZkg9s5ZhKMDYcJa2GTSGyT5jE4XL/h5hkitfBIJDMbA7UwS87gIVKLBM8xxodAhzHz8wD9kkCMX+yP93848OEPKCoPb7zxocaGsBZkYCCRQIpyiBZSdYyCUTAKRsHIAABsDUJEPpR8XQAAAABJRU5ErkJggg==","orcid":"","institution":"Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University,No.1","correspondingAuthor":true,"prefix":"","firstName":"Ziru","middleName":"","lastName":"Zhang","suffix":""}],"badges":[],"createdAt":"2026-01-03 05:08:15","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8504448/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8504448/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103840823,"identity":"eab93743-7178-466c-a2e7-13092a4c5522","added_by":"auto","created_at":"2026-03-03 14:42:04","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":161124,"visible":true,"origin":"","legend":"\u003cp\u003ePatient Selection Flowchart for Modeling Cohort. This study included 632 elderly patients who underwent hip replacement surgery at our hospital, comprising 367 patients with SSD and 265 patients without SSD.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8504448/v1/d0fc2b2b2d7bfbf599c73722.png"},{"id":103840825,"identity":"1bb8150f-836b-4633-b841-61efab22aa2d","added_by":"auto","created_at":"2026-03-03 14:42:04","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":305896,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eROC curves of prediction models developed using different machine learning algorithms. The AUROC values for the LR, XGB, GNB, and RF algorithms were 0.826 (0.734–0.917), 0.891 (0.824–0.958), 0.862 (0.772–0.951), and 0.862 (0.773–0.950), respectively. The prediction model based on the XGB algorithm was the optimal model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8504448/v1/f152907fe7c0eb7a5e706128.png"},{"id":103840827,"identity":"933130c9-0b36-4a78-bac1-c2580ebc0c52","added_by":"auto","created_at":"2026-03-03 14:42:04","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1376113,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP summary plot based on the XGB algorithm. Distribution of SHAP values for each feature influencing the postoperative SSD prediction model output. As feature values increase, corresponding SHAP value points shift to the right side of the chart, indicating greater contribution to the model's prediction of postoperative SSD. Point colors transition from blue to red, reflecting increasing risk values from low to high. Age (years), duration of surgery, ASA physical status classification, cumulative intraoperative hypotension time, history of delirium, age, and postoperative analgesic use are the most significant predictors, exerting substantial influence on the model's output.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8504448/v1/9ad62ed5bee3a7eabf37303d.png"},{"id":103840826,"identity":"5ecabc1c-5186-4011-85a3-b437d3e883ca","added_by":"auto","created_at":"2026-03-03 14:42:04","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":852535,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eROC curves of prediction models developed using 6 variables selected by the XGB algorithm, based on internal (Top 6) and external validation datasets (Top 6*). The AUROC for the internal validation set was 0.849 (0.769–0.929), accuracy was 0.747 (0.678–0.806), AUPRC was 0.657 (0.600–0.714), specificity was 0.712 (0.633–0.791), and sensitivity was 0.815 (0.721–0.910). The external test set yielded an AUROC of 0.842 (0.754–0.930), accuracy of 0.799 (0.748–0.843), AUPRC of 0.650 (0.621–0.679), specificity of 0.779 (0.691–0.867), and sensitivity was 0.808 (0.754–0.861).\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8504448/v1/d740207635deb29c5cb32feb.png"},{"id":104401042,"identity":"ad553aca-5e7c-4ac9-845d-130c65f6cb3a","added_by":"auto","created_at":"2026-03-11 12:11:44","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3844869,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8504448/v1/d52fc495-9e47-4174-afce-16b2d4695d38.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Construction and validation of a predictive model based on machine learning algorithm for the risk of subsyndromic delirium after hip arthroplasty in the elderly","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eDelirium is an acute, fluctuating cognitive disorder whose main features include inattention, decreased level of consciousness, cognitive dysfunction and fluctuating affect \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. Delirium is often misdiagnosed as other psychiatric disorders because of its similarity in aetiology and symptoms to dementia. However, unlike other disorders, the symptoms of delirium can be effectively relieved by controlling the causative agent \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. Postoperative delirium (POD) is a common surgical complication in the elderly population, with a prevalence of approximately 5.10%-21.37% in the patient population undergoing hip fracture surgery \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. In recent years, with the aging of the population, the group of elderly hip fracture surgery patients has been increasing, and clinical workers have paid more attention to the hazards, risk factors and prevention of POD. POD can cause prolonged hospitalisation, impaired functional recovery, increased healthcare costs and increased risk of readmission, which can affect surgical outcomes and the recovery process, and even lead to life-threatening diseases and death \u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e. The clinical picture is characterised by acute post-operative seizures, altered level of consciousness, inattention and thought disorders \u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e. However, some patients present with only some of the symptoms of delirium, but not enough to meet the traditional diagnostic criteria for delirium, and are referred to as having subsyndromal delirium (SSD) \u003csup\u003e[\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eSSD is defined as a state with one or more symptoms of delirium or a precursor state of clinical delirium, but not meeting all the diagnostic criteria for delirium \u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e. Despite the relatively mild symptoms of SSD compared to delirium, its duration and long-term effects cannot be ignored and still have a significant impact on patients' recovery and long-term cognitive function. Compared with patients without delirium, patients with SSD have longer hospital stays, more severe cognitive impairment and higher mortality rates \u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e. Studies have shown that the risk factors for SSD are similar to those for delirium, such as age, underlying disease, duration of preoperative fasting, use of psychotropic medications, duration of surgery, anesthesia modality, and postoperative pain management \u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e. In addition, SSD is a dynamically progressive disorder and may gradually progress to full-blown delirium, which seriously affects patients' recovery and quality of life \u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. Therefore, it is necessary for healthcare professionals to identify the high-risk group of SSD at an early stage and provide effective interventions to slow its progression.\u003c/p\u003e \u003cp\u003eTraditional assessment methods mainly rely on clinician experience and simple scoring systems, such as the Delirium Assessment Scale and the Confusion Assessment Method (CAM), which can identify high-risk patients to some extent, but suffer from subjectivity and low sensitivity \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. In recent years, the use of machine learning techniques in medical prediction has increased, and many studies have investigated the use of machine learning algorithms to predict the risk of postoperative delirium in elderly patients \u003csup\u003e[\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. These studies typically use large-scale electronic health record data, and feature selection and model training improve the accuracy and generalisability of the prediction. However, to date, there are few studies on postoperative SSD risk prediction models for elderly hip arthroplasty patients. Therefore, this study aims to construct and validate a postoperative SSD risk prediction model for elderly hip arthroplasty patients using machine learning algorithms based on our electronic health record data.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cp\u003e2.1 Study Subjects\u003c/p\u003e\n\u003cp\u003eThis retrospective study was conducted under the approval of the Review Committee of Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University (approval number: K20240219) and was exempted from the requirement of informed consent. We submitted the statistical analysis plan to Tangdu Hospital, the Second Affiliated Hospital of Air Force Military Medical University before accessing all the data. We followed the TRIPOD statement for reporting the development and validation of multivariate prediction models\u003csup\u003e\u0026nbsp;[19]\u003c/sup\u003e. The researchers strictly followed the \u0026quot;Helsinki Declaration\u0026quot; in their work.\u003c/p\u003e\n\u003cp\u003e2.2 Inclusion criteria\u003c/p\u003e\n\u003cp\u003eWe used electronic data from older adults who underwent hip arthroplasty at Tangdu Hospital, the Second Affiliated Hospital of Air Force Military Medical University between January 2020 and December 2023. Inclusion criteria: (1) patients over 65 years of age; (2) patients undergoing hip arthroplasty for the first time; (3) American Society of Anesthesiologists (ASA) class I-IV; (4) patients with delirium assessment results.Exclusion criteria: (1) those undergoing emergency surgery; (2) those with preoperative cognitive deficits such as delirium; (3) those with unstable postoperative vital signs; (4) postoperative hospital stay \u0026lt;5 d. The flowchart for patient selection in the modeling queue is shown in Figure 1.\u003c/p\u003e\n\u003cp\u003e2.3 Outcome indicators\u003c/p\u003e\n\u003cp\u003eThe main observational index in this study was postoperative SSD. the Chinese version of the CAM was used to assess the occurrence of SSD \u003csup\u003e[20]\u003c/sup\u003e. The CAM is the most commonly used tool for detecting delirium and includes the four features of the DSM-5 diagnostic criteria: (1) acute onset or fluctuating course; (2) impaired attention; (3) disorganised consciousness; and (4) disorganised and disorganised thinking\u003csup\u003e\u0026nbsp;[20-21]\u003c/sup\u003e. These features can effectively distinguish delirium from other types of cognitive impairment and help healthcare professionals to identify delirium quickly and accurately. This version has a sensitivity of 90%, specificity of 94% and high inter-rater reliability. POD is defined as individuals meeting criteria 1 + 2 + (3 or 4). SSD is defined as individuals exhibiting at least one additional CAM symptom but failing to meet POD diagnostic requirements \u003csup\u003e[8]\u003c/sup\u003e. In general, two uniformly trained, experienced orthopaedic nurses assessed patients\u0026apos; SSD and POD 1 d preoperatively and 1-5 d postoperatively, twice daily or as the patient\u0026apos;s condition changed. We analysed the results of all assessments from the end of surgery until discharge.\u003c/p\u003e\n\u003cp\u003e2.4 Selection of exposure variables\u003c/p\u003e\n\u003cp\u003eIn this study, exposure variables were selected with reference to those already reported in other literature. The final exposure variables included general predictors (e.g. age, MMSE score), neuropsychological factors (e.g. previous delirium, depression), comorbidities (e.g. diabetes mellitus, chronic obstructive pulmonary disease), haematological indices (e.g. Hemoglobin), intraoperative indices (e.g. haemorrhage volume, operative time), and postoperative indices (e.g. use of analgesics).\u003c/p\u003e\n\u003cp\u003e2.5 Data analysis and\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003epre-processing\u003c/p\u003e\n\u003cp\u003eOur primary aim was to develop a predictive model for postoperative SSD using combined preoperative, intraoperative and postoperative data. The data were divided into a training set (\u003cem\u003en=\u003c/em\u003e442) and a validation set (n=190) according to a 7:3 ratio, with the training set used for model development and the validation set used for validation testing. We screened the missing rates of all potential predictors and excluded those with missing values \u0026ge;25%; predictors with missing values \u0026lt;25% were imputed. The missing rates for MMSE scores, frailty, previous history of delirium, and the lowest postoperative body temperature were 11.23%, 8.39%, 17.88%, and 12.97%, respectively. Assuming the data were missing at random, multiple imputation was performed using the mice package in R software. Five complete datasets were generated, with the imputation model incorporating all predictor variables as well as the outcome variable from this study. The final results for the missing data were derived from the average of the model outcomes across the five imputed datasets.\u003c/p\u003e\n\u003cp\u003e2.6 Model development\u003c/p\u003e\n\u003cp\u003eIn this study, we constructed a prediction model for postoperative SSD in elderly hip replacement patients based on four machine learning algorithms: logistic regression (LR), extreme gradient boosting (XGB), Gaussian naive Bayes (GNB), and random forest (RF), and compared their prediction performance. These methods were used to construct a prediction model for postoperative SSD in elderly hip arthroplasty patients, and their predictive performance was compared. Since the number of people who did not develop SSD after surgery was much larger than the number of people who developed SSD, there was an imbalance in the modelling data, so in this study hyperparameters were set to increase the weights of positive samples to solve the problem of data imbalance. We also performed parameter adjustment on the training set by the fivefold crossover method, which improved the learning effect of the model.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2.7 Model evaluation\u003c/p\u003e\n\u003cp\u003eWe used the area under the receiver operating characteristic curve (AUROC), sensitivity, accuracy, specificity and area under the precision-recall curve (AUPRC) to evaluate the above four prediction models. The AUPRC ranges from 0 to 1, focusing on the model\u0026apos;s performance on the \u0026ldquo;positive\u0026rdquo; group. It typically provides a more informative assessment than AUC-ROC in highly imbalanced datasets. We also used SHAP summary plots to show the importance of variables in predicting SSD in the GBM model. The effect of each feature on postoperative SSD was expressed as SHAP values by deriving marginal distributions and weighted means, which were fixed for all variables except the variable of interest\u003csup\u003e\u0026nbsp;[22]\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e2.8 External validation\u003c/p\u003e\n\u003cp\u003eTo confirm the validity of the model performance, external validation was performed using different datasets from Xijing Hospital, the First Affiliated Hospital of the Air Force University of Military Medicine, to validate the effectiveness of the final model in predicting SSD.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2.9 Statistical analysis\u003c/p\u003e\n\u003cp\u003eR software (4.3.2), Python (3.4.3) language and SPSS 25.0 software were used to process the data. Measurement information that met the normal distribution test was expressed as (\u0026plusmn;) or M (P25, P75), and was compared using the independent samples t-test or Mann-Whitney U test. Count data were expressed as [n (%)], and the \u0026chi;2 test or Fisher test was performed to compare the differences.\u003cem\u003e\u0026nbsp;p \u0026lt;\u003c/em\u003e 0.05 indicated that the differences were statistically significant.\u003c/p\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Baseline characteristics\u003c/h2\u003e\n \u003cp\u003eA total of 632 elderly patients who underwent hip arthroplasty at our institution were included in this study. Of all patients, 265 (41.93%) developed postoperative SSD, with 23 (3.64%) progressed from SSD to POD. The baseline characteristics of SSD and non- SSD patients are summarised in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.Patients with SSD were predominantly male, older, had a higher ASA classification, a higher rate of electrolyte abnormalities and were often accompanied by multiple comorbidities. Intraoperatively, SSD patients had longer operative time, more bleeding, higher rates of blood transfusion, and were more prone to hypotension and had longer duration of hypotension (Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). Postoperatively, SSD patients had a higher rate of analgesic use (Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eBaseline characteristics of SSD and non- SSD patients\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eitems\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003enon- SSD(n\u0026thinsp;=\u0026thinsp;367)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSSD(n\u0026thinsp;=\u0026thinsp;265)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003et/\u003cem\u003e\u0026chi;\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u003cem\u003e/\u003c/em\u003e\u003csup\u003e\u003cem\u003eZ\u003c/em\u003e\u003c/sup\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSex[n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e22.479\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e157(42.78)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e164(61.89)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e210(57.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e101(38.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e72.51\u0026thinsp;\u0026plusmn;\u0026thinsp;6.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e80.16\u0026thinsp;\u0026plusmn;\u0026thinsp;10.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10.038\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAnaesthesia[n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.825\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.177\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGeneral anesthetic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e167(45.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e135(50.94)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIntravertebral anaesthesia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e200(54.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e130(49.06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eASA[n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e79.325\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eⅠ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e144(39.24)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e35(13.21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eⅡ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e184(50.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139(52.45)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eⅢ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e36(9.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81(30.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eⅣ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3(0.82)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10(3.77)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBMI(kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25.52\u0026thinsp;\u0026plusmn;\u0026thinsp;3.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e24.98\u0026thinsp;\u0026plusmn;\u0026thinsp;4.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.455\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.146\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eComplication [n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDiabetes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e72(19.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e89(33.58)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e15.811\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHypertensive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e166(45.23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e157(59.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12.094\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRheumatoid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e33(8.99)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e31(11.70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.238\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.266\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCoronary heart disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e47(12.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e69(26.04)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e17.977\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eArrhythmia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e57(15.53)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e71(26.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12.082\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChronic obstructive pulmonary disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6(1.63)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15(5.66)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7.763\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.005\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChronic Kidney Disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9(2.45)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17(6.42)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.126\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.013\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePrevious delirium\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7(1.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e33(12.45)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e28.866\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDebility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e18(4.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34(12.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12.801\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAlcohol consumption[n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e49(13.35)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e48(18.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.686\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.101\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSmoking[n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e77(1.580)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e45(16.98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.580\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.209\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePreoperative MMSE score (points)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26.00(23.00, 29.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e24.00(22.00, 26.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7.129\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePreoperative haematological indices\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHemoglobin(g/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13.31\u0026thinsp;\u0026plusmn;\u0026thinsp;1.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12.25\u0026thinsp;\u0026plusmn;\u0026thinsp;2.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.541\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCreatinine(mg/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.93\u0026thinsp;\u0026plusmn;\u0026thinsp;0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.32\u0026thinsp;\u0026plusmn;\u0026thinsp;1.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.087\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAlbumin(g/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e36.52\u0026thinsp;\u0026plusmn;\u0026thinsp;4.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e32.17\u0026thinsp;\u0026plusmn;\u0026thinsp;5.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10.954\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHypokalaemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15(4.09)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e31(11.70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e13.208\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHyperkalaemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7(1.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25(9.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e18.135\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHyponatraemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11(3.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e41(15.47)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e31.731\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHypernatraemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3(0.82)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3(1.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.162\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.687\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHypocalcaemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e35(9.54)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e57(21.51)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e17.735\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHypercalcaemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8(2.18)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1(0.38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.562\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.059\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIntraoperative indicators\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOperative time(min)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e112.50(85.00, 125.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e135.00(97.50, 149.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11.256\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHypotension[n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139(37.87)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e142(50.53)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e15.381\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCumulative duration of hypotension(min)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0(0, 2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4(1, 6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14.561\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHaemorrhage volume(mL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e105.00(62.00, 200.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e195.00(100.00, 250.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e8.015\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlood transfusion[n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e18(4.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e39(14.72)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e18.056\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePostoperative indicators\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eUse of analgesics[n(%)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e88(23.98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e98(36.98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12.527\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMinimum body temperature(℃)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e36.52\u0026thinsp;\u0026plusmn;\u0026thinsp;4.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e35.90\u0026thinsp;\u0026plusmn;\u0026thinsp;6.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.278\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.202\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003eASA: American Society of Anaesthesiologists; BMI: body mass index.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Development of Prediction Models\u003c/h2\u003e\n \u003cp\u003eThe AUROC of the LR, XGB, GNB and RF algorithms were 0.826 (0.734\u0026ndash;0.917), 0.891 (0.824\u0026ndash;0.958), 0.862 (0.772\u0026ndash;0.951), and 0.862 (0.773\u0026ndash;0.950), respectively (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). In terms of AUROC, accuracy, F1 score, and AUPRC, the XGB, GNB, and RF models showed similar performance (XGB: 0.891, 0.859, 0.421, 0.731; GNB: 0.862, 0.836, 0.619, 0.724; LR: 0.826, 0.799, 0.573, 0.635; RF: 0.862, 0.871, 0.497, 0.668), and the specific performance metrics are shown in Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.Considering that the AUROC and AUPRC of XGB are higher than those of GNB and LR models, we finally chose the prediction model based on the XGB algorithm.\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePerformance metrics of different machine learning algorithms for predictive models\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClassification model\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUROC (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUPRC (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSpecificity (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSensitivity (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNPV (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePPV (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.891 (0.824\u0026ndash;0.958)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.731 (0.642\u0026ndash;0.820)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.859 (0.841\u0026ndash;0.877)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.854 (0.813\u0026ndash;0.895)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.870 (0.817\u0026ndash;0.923)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.925 (0.893\u0026ndash;0.957)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.761 (0.698\u0026ndash;0.824)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLogistic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.826 (0.734\u0026ndash;0.917)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.635 (0.578\u0026ndash;0.691)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.799 (0.769\u0026ndash;0.828)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.794 (0.748\u0026ndash;0.840)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.808 (0.744\u0026ndash;0.873)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.899 (0.862\u0026ndash;0.935)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.648 (0.577\u0026ndash;0.718)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandomForest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.862 (0.773\u0026ndash;0.950)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.668 (0.630\u0026ndash;0.707)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.871 (0.843\u0026ndash;0.898)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.869 (0.830\u0026ndash;0.909)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.874 (0.823\u0026ndash;0.926)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.925 (0.893\u0026ndash;0.957)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.790 (0.730\u0026ndash;0.850)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGNB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.862 (0.772\u0026ndash;0.951)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.724 (0.630\u0026ndash;0.817)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.837 (0.794\u0026ndash;0.878)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.856 (0.814\u0026ndash;0.898)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.801 (0.741\u0026ndash;0.861)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.872 (0.832\u0026ndash;0.912)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.778 (0.717\u0026ndash;0.8403)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003e3.3 Development of predictive models using selected significant variables\u003c/h2\u003e\n \u003cp\u003eThe XGB algorithm, an integrated learning model based on gradient boosting decision trees, achieves good performance on a variety of data sets, but it loses the interpretability of a linear model. In contrast, SHAP analysis intuitively explains the importance of the selected variables, in particular, the contribution of each feature to the model prediction is explained by the Shapley value \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e. We generated summary SHAP plots based on the results of the XGB model (only variables with SHAP\u0026thinsp;\u0026gt;\u0026thinsp;0.05 are shown). In this case, the vertical coordinates are the factors of each variable, with darker red representing higher risk and darker blue representing lower risk, and the horizontal coordinates are the SHAP values. To make the prediction model more applicable to clinical practice, we selected the variables with SHAP value ranked in the top 6 to develop the final model. As shown in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e, the top 6 variables were: duration of surgery (0.527), ASA classification (0.415), cumulative duration of intraoperative hypotension (0.327), previous delirium (0.306), age (0.219), and use of postoperative analgesics (0.201). We developed a new prediction model for postoperative SSD using these six selected variables and the new prediction model had an AUROC of 0.849 (0.769\u0026ndash;0.929) (Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e), an accuracy of 0.747 (0.678\u0026ndash;0.806), an AUPRC of 0.657 (0.600-0.714), a specificity of 0.712 (0.633\u0026ndash;0.791), and a sensitivity of 0.815 (0.721\u0026ndash;0.910).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n \u003ch2\u003e3.4 External validation of the prediction model\u003c/h2\u003e\n \u003cp\u003eWe evaluated the new prediction model in an external test set, and the results show that the new XGB model achieves an AUROC of 0.842 (0.754\u0026ndash;0.930) (Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e), an accuracy of 0.799 (0.748\u0026ndash;0.843), an AUPRC of 0.650 (0.621\u0026ndash;0.679), a specificity of 0.779 (0.691\u0026ndash;0.867), and a sensitivity of 0.808 (0.754\u0026ndash;0.861), all of which are close to previous.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eIn this study, we developed and validated a predictive model for postoperative SSD in elderly hip arthroplasty patients. The incidence of postoperative SSD in the dataset was 41.93%. We selected six variables based on the XGB algorithm: operative time, ASA classification, cumulative duration of intraoperative hypotension, previous delirium, age, and postoperative analgesic use to develop the prediction model..The final developed prediction model achieved an AUROC value of 0.849 (0.769\u0026ndash;0.929), which is a superior predictive performance.\u003c/p\u003e\n\u003cp\u003eSSD is an acute state of clouded consciousness less severe than delirium, accompanied by mild emotional fluctuations, mild decrease in level of consciousness, mild changes in cognitive function, mild thought disorders and mild perceptual deficits \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e. Although most of these symptoms are reversible, their duration and long-term effects cannot be ignored and may progress to full-blown delirium, with serious and costly consequences such as prolonged hospitalisation, complications and increased risk of death \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e. Studies have shown that approximately 9.50%-20.70% of patients progress from SSD to full-blown delirium \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e. Our data showed that 8.68% of elderly hip arthroplasty patients progressed from SSD to full delirium, supporting that SSD is a precursor to full delirium and that the risk of clinical delirium is particularly high. Like POD, postoperative SSD is considered preventable and identification of those at risk is a prerequisite for modifying predisposing factors and implementing targeted interventions.In this study, we applied relatively strict criteria for the diagnosis of SSD, excluding patients with direct POD, thus distinguishing SSD from POD in the immediate postoperative period. Considering that advanced age is an independent predictor of postoperative SSD \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e, in this study we selected the clinical data of elderly hip arthroplasty patients for modelling, avoided the parameter estimation bias that might be caused by the direct use of whole-population data, improved the predictive accuracy of the model in elderly hip arthroplasty patients, and finally developed a prediction model that was more targeted and could effectively predict the occurrence of postoperative SSD in elderly patients. In addition, we validated the prediction model with data sets from different time periods, and the prediction model showed better predictive ability. Considering the convenience of data collection and the practicality of the model, this study optimised the XGB model constructed on the basis of 33 variables, and finally screened six variables that contributed the most to the XGB model and were clinically associated with SSD, and obtained a more simplified model, which made data collection and processing more efficient and was conducive to the dissemination and application of the prediction model.\u003c/p\u003e\n\u003cp\u003eThe occurrence of SSD is associated with a variety of factors, including age, underlying disease, duration of surgery, anaesthetic modality, postoperative pain management and medication use \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e. Several studies have proposed scoring systems based on traditional regression models to predict SSD \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e. However, traditional regression analysis methods are limited by the number of variables and are only applicable to the analysis of linear correlations. It is significantly less effective in dealing with non-linearly differentiable data, which in turn affects the predictive accuracy of the model. In this study, we used machine learning techniques, which can process and detect non-linear relationships and complex data structures, and can improve the reliability and accuracy of disease diagnosis systems \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e. After comparing several algorithms of machine learning techniques, we finally selected the XGB algorithm with optimal performance to construct a postoperative SSD prediction model for elderly hip replacement patients.\u003c/p\u003e\n\u003cp\u003eIn the medical field, the interpretation of the results of machine learning analyses should be done with full consideration of clinical applicability.SHAP scores as an interpretive tool can compensate for the problem of poor interpretability in machine learning by helping to understand the contribution of individual features to the predicted outcomes of the model, thus ensuring that the interpretation of the model meets the needs of clinical practice \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. The SHAP analysis in this study showed that the factors influencing postoperative SSD in elderly hip arthroplasty patients were, in order of importance: duration of surgery, ASA classification, cumulative duration of intraoperative hypotension, previous delirium, age and use of analgesics in the postoperative period. Like delirium, SSD is caused by deterioration in homeostasis and physical status, which is more likely to occur in high-risk surgery of longer duration \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e. ASA grading is effective in assessing the preoperative health status of patients, and the higher the ASA grading, the worse the physiological function and health status of the patient, and the higher the risk of postoperative complications such as delirium \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/sup\u003e. Hypotension can lead to inadequate cerebral perfusion, the duration of which is closely related to damage to organs such as the heart and kidneys, and also increases the risk of postoperative SSD and POD \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e37\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/sup\u003e. Studies have shown that previous delirium is associated with a higher risk of postoperative SSD and POD \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. In addition, age is the most widely accepted risk factor for SSD in any clinical situation \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/sup\u003e.Finally, postoperative analgesic use may affect cognitive function and the brain\u0026apos;s response to stress \u003csup\u003e[41]\u003c/sup\u003e, and older patients receiving postoperative analgesics may be more prone to delirium. However, it has also been found that opioid analgesic regimens are not associated with postoperative delirium \u003csup\u003e[42]\u003c/sup\u003e. Our model is clinically interpretable with only six variables, is highly tractable and shows superior predictive performance. Specifically, the risk of postoperative SSD was higher in those with prior delirium than in those without prior delirium. The risk of postoperative SSD also increased with duration of surgery, cumulative duration of intraoperative hypotension and age, i.e. the longer the duration of surgery, the longer the cumulative duration of intraoperative hypotension and the older the patient, the higher the risk of postoperative SSD. The risk of delirium also increased with the ASA classification score and with increasing score. The risk of postoperative SSD was also higher in those who used postoperative analgesics than in those who did not. However, since this study did not explicitly distinguish the temporal sequence between postoperative analgesic use and SSD, a reverse causal relationship between postoperative analgesic use and SSD cannot be ruled out. Future studies will clearly document the timing of analgesic administration relative to SSD onset to further optimize predictive models and intervention strategies. Overall, the predictive model we developed aids in identifying patients at high risk for postoperative SSD, thereby enabling earlier intervention in the prevention and management of postoperative neurocognitive complications. Clinical healthcare providers can utilize the predictive model preoperatively to calculate postoperative SSD risk scores and flag high-risk patients in electronic medical record systems. During the intraoperative phase, efficient team collaboration should minimize surgical duration while prioritizing blood pressure management to prevent hypotensive events. Postoperatively, multimodal analgesia should be employed to avoid excessive opioid use, and delirium assessment tools should be routinely administered at least daily to enable early SSD detection. Furthermore, during the SSD phase, patients typically exhibit greater cooperation with milder disorientation and agitation. Introducing interventions such as cognitive training and early mobilization at this stage yields superior outcomes, preventing further deterioration and improving long-term functional recovery and quality of life.\u003c/p\u003e\n\u003cp\u003eThe following limitations must also be taken into account when interpreting our results. First, this is a retrospective study and causality cannot be confirmed. Second, the results cannot be generalised to other patient groups because this study included the entire Chinese population and did not include other races. In the future, we will conduct multicenter validation across different populations and healthcare settings to enhance the generalizability of this study\u0026apos;s findings. Third, the data were unbalanced and the low prevalence of postoperative SSD affected the sensitivity of the model to some extent. However, this low prevalence is due to the fact that the diagnosis was only recorded when patients were specifically assessed for delirium status. As screening patients for postoperative SSD is not a clinically necessary intervention, many cases appear to be missed in the real world. This also justifies the clinical need for such a predictive model for screening for postoperative SSD and the need for future validation with other datasets. Despite the limitations, this study is the first to assess the risk of postoperative SSD in elderly hip arthroplasty using machine learning algorithms and a well-established predictive model, providing clinicians with an evaluation tool to identify high-risk individuals. However, the predictive model outputs the relative risk of individual patients developing SSD, aiming to assist healthcare providers in prioritizing preventive care resources. While the model demonstrates good performance, it is not perfect, and its clinical application should be combined with the judgment of clinicians. Future multi-center, prospective studies are needed to further validate and optimize this model..\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eWe used machine learning techniques to select six variables and demonstrate a predictive model for the risk of postoperative SSD in elderly hip arthroplasty. The model provides a useful tool for identifying patients at high risk for postoperative SSD, aiding in the early detection of those with elevated risk. This enables the advancement of preventive measures for postoperative neurocognitive complications. Implementing interventions such as cognitive training and early mobilization for high-risk patients at an early stage may help prevent further deterioration of their condition.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis retrospective study was conducted under the approval of the Review Committee of Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University (approval number: K20240219) and was exempted from the requirement of informed consent.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLi Zhang designed the experiments, performed the study, and wrote the article;Juanqi Li performed the study and collected the data; Fangyuan Zhang performed the study, collected the data, and analysed/interpreted the data; Ziru Zhang: designed the experiments, revised the article, and supported the work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding Declaration\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThere was no funding.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWilson JE, Mart MF, Cunningham C, Shehabi Y, Girard TD, MacLullich AMJ, et al. Delirium. \u003cem\u003eNat Rev Dis Primers\u003c/em\u003e. (2020) 6:90. doi: 10.1038/s41572-020-00223-4.\u003c/li\u003e\n\u003cli\u003eRyan SL, Kimchi EY. Evaluation and Management of Delirium. \u003cem\u003eSemin Neurol\u003c/em\u003e. (2021) 41:572-87. doi: 10.1055/s-0041-1733791.\u003c/li\u003e\n\u003cli\u003eLi T, Li J, Yuan L, Jiang C, Daniels J, Mehta RL, et al. Effect of Regional vs General Anesthesia on Incidence of Postoperative Delirium in Older Patients Undergoing Hip Fracture Surgery: The RAGA Randomized Trial. \u003cem\u003eJAMA\u003c/em\u003e. (2022), 327:50-8. doi: 10.1001/jama.2021.22647.\u003c/li\u003e\n\u003cli\u003eDuan Y, Zhang R. Risk factors and prediction model of delirium in elderly patients after hip arthroplasty. \u003cem\u003ePak J Med Sci\u003c/em\u003e. (2024) 40:1077-82. doi: 10.12669/pjms.40.6.9306.\u003c/li\u003e\n\u003cli\u003eHughes CG, Boncyk CS, Culley DJ, et al. American Society for Enhanced Recovery and Perioperative Quality Initiative Joint Consensus Statement on Postoperative Delirium Prevention. \u003cem\u003eAnesth Analg\u003c/em\u003e. (2020) 130:1572-90. doi: 10.1213/ANE.0000000000004641.\u003c/li\u003e\n\u003cli\u003eAldecoa C, Bettelli G, Bilotta F, Fleisher LA, Leung JM, McDonagh DL, et al. Update of the European Society of Anaesthesiology and Intensive Care Medicine evidence-based and consensus-based guideline on postoperative delirium in adult patients. \u003cem\u003eEur J Anaesthesiol\u003c/em\u003e. 2024, 41(2):81-108.\u003c/li\u003e\n\u003cli\u003eCole MG. Subsyndromal delirium in old age: conceptual and methodological issues. \u003cem\u003eInt Psychogeriatr\u003c/em\u003e. (2013) 25:863-6. doi: 10.1017/S1041610213000434.\u003c/li\u003e\n\u003cli\u003eSepulveda E, Leonard M, Franco JG, Adamis D, McCarthy G, Dunne C, et al. Subsyndromal delirium compared with delirium, dementia, and subjects without delirium or dementia in elderly general hospital admissions and nursing home residents. \u003cem\u003eAlzheimers Dement (Amst)\u003c/em\u003e. (2016) 7:1-10. doi: 10.1016/j.dadm.2016.11.002.\u003c/li\u003e\n\u003cli\u003eBoettger S, Nu\u0026ntilde;ez DG, Meyer R, Richter A, Schubert M, Jenewein J. Subsyndromal delirium in the intensive care setting: Phenomenological characteristics and discrimination of subsyndromal delirium versus no and full-syndromal delirium. \u003cem\u003ePalliat Support Care\u003c/em\u003e. (2018) 16:3-13. doi: 10.1017/S1478951517000104.\u003c/li\u003e\n\u003cli\u003eShim J, DePalma G, Sands LP, Leung JM. Prognostic Significance of Postoperative Subsyndromal Delirium. \u003cem\u003ePsychosomatics\u003c/em\u003e. (2015). 56:644-51. doi: 10.1016/j.psym.2015.05.002.\u003c/li\u003e\n\u003cli\u003eAzuma K, Mishima S, Shimoyama K, Ishii Y, Ueda Y, Sakurai M, et al. Validation of the Prediction of Delirium for Intensive Care model to predict subsyndromal delirium. \u003cem\u003eAcute Med Surg\u003c/em\u003e. (2018). 6:54-9. doi: 10.1002/ams2.378.\u003c/li\u003e\n\u003cli\u003eChen S, Tang L, Chen J, Cai L, Liu C, Song J, et al. Prevalence and risk factors of subsyndromal delirium among postoperative patients: A systematic review and meta-analysis. \u003cem\u003eJ Adv Nurs\u003c/em\u003e. (2024). 80:924-34. doi: 10.1111/jan.15871.\u003c/li\u003e\n\u003cli\u003eMailhot T, Cossette S, Maheu-Cadotte MA, Fontaine G, Denault AY. Subsyndromal Delirium in Cardiac Surgery Patients: Risk Factors and Outcomes of the Different Trajectories. \u003cem\u003eJ Cardiovasc Nurs\u003c/em\u003e. (2022) 37:41-9. doi: 10.1097/JCN.0000000000000793.\u003c/li\u003e\n\u003cli\u003ePaulino MC, Concei\u0026ccedil;\u0026atilde;o C, Silvestre J, Lopes MI, Gon\u0026ccedil;alves H, Dias CC, et al. Subsyndromal Delirium in Critically Ill Patients-Cognitive and Functional Long-Term Outcomes. \u003cem\u003eJ Clin Med\u003c/em\u003e. (2023). 12:6363. doi: 10.3390/jcm12196363.\u003c/li\u003e\n\u003cli\u003eOh ES, Fong TG, Hshieh TT, Inouye SK. Delirium in Older Persons: Advances in Diagnosis and Treatment. JAMA. (2017) 318:1161-74. doi: 10.1001/jama.2017.12067.\u003c/li\u003e\n\u003cli\u003eRacine AM, Tommet D, D\u0026apos;Aquila ML, Fong TG, Gou Y, Tabloski PA, et al. Machine Learning to Develop and Internally Validate a Predictive Model for Post-operative Delirium in a Prospective, Observational Clinical Cohort Study of Older Surgical Patients. \u003cem\u003eJ Gen Intern Med\u003c/em\u003e. (2021) 36:265-73. doi: 10.1007/s11606-020-06238-7.\u003c/li\u003e\n\u003cli\u003eBishara A, Chiu C, Whitlock EL, Douglas VC, Lee S, Butte AJ, et al. Postoperative delirium prediction using machine learning models and preoperative electronic health record data. \u003cem\u003eBMC Anesthesiol\u003c/em\u003e. (2022) 22:8. doi: 10.1186/s12871-021-01543-y.\u003c/li\u003e\n\u003cli\u003eSong Y, Zhang D, Wang Q, Liu Y, Chen K, Sun J, et al. Prediction models for postoperative delirium in elderly patients with machine-learning algorithms and SHapley Additive exPlanations. \u003cem\u003eTransl Psychiatry\u003c/em\u003e. (2024) 14:57. doi: 10.1038/s41398-024-02762-w.\u003c/li\u003e\n\u003cli\u003eCollins GS, Reitsma JB, Altman DG, Moons KG. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. \u003cem\u003eAnn Intern Med\u003c/em\u003e. (2015) 162:55-63. doi: 10.7326/L15-0078-4.\u003c/li\u003e\n\u003cli\u003eMei X, Chen Y, Zheng H, Shi Z, Marcantonio ER, Xie Z, et al. The Reliability and Validity of the Chinese Version of Confusion Assessment Method Based Scoring System for Delirium Severity (CAM-S). \u003cem\u003eJ Alzheimers Dis\u003c/em\u003e. (2019) 69:709-16. doi: 10.3233/JAD-181288.\u003c/li\u003e\n\u003cli\u003eInouye SK, van Dyck CH, Alessi CA, Alessi CA, Balkin S, Siegal AP, et al. Clarifying confusion: the confusion assessment method. A new method for detection of delirium. \u003cem\u003eAnn Intern Med\u003c/em\u003e. (1990) 113:941-8. doi: 10.7326/0003-4819-113-12-941.\u003c/li\u003e\n\u003cli\u003eLi X, Dvornek NC, Zhou Y, Zhuang J, Ventola P, Duncan JS. Efficient Interpretation of Deep Learning Models Using Graph Structure and Cooperative Game Theory: Application to ASD Biomarker Discovery. \u003cem\u003eInf Process Med Imaging\u003c/em\u003e. (2019) 11492:718-730. doi: 10.1007/978-3-030-20351-1_56.\u003c/li\u003e\n\u003cli\u003eCole MG, Bailey R, Bonnycastle M, McCusker J, Fung S, Ciampi A, et al. Frequency of full, partial and no recovery from subsyndromal deliriumin older hospital inpatients. \u003cem\u003eInt J Geriatr Psychiatry\u003c/em\u003e. (2016) 31:544-50. doi: 10.1002/gps.4372.\u003c/li\u003e\n\u003cli\u003ePaulino MC, Concei\u0026ccedil;\u0026atilde;o C, Silvestre J, Lopes MI, Gon\u0026ccedil;alves H, Dias CC, et al. Subsyndromal Delirium in Critically Ill Patients-Cognitive and Functional Long-Term Outcomes. \u003cem\u003eJ Clin Med\u003c/em\u003e. (2023) 12:6363. doi: 10.3390/jcm12196363.\u003c/li\u003e\n\u003cli\u003eYamada C, Iwawaki Y, Harada K, Fukui M, Morimoto M, Yamanaka R. Frequency and risk factors for subsyndromal delirium in an intensive care unit. \u003cem\u003eIntensive Crit Care Nurs\u003c/em\u003e. (2018) 47:15-22. doi: 10.1016/j.iccn.2018.02.010.\u003c/li\u003e\n\u003cli\u003eSong C, Wang S, Liu J, Xu W, Yang Y, Li X. Predictive factors for the duration of subsyndromal delirium in the intensive care unit. \u003cem\u003eJ Clin Nurs\u003c/em\u003e. (2024) 33:3775-83. doi: 10.1111/jocn.17148.\u003c/li\u003e\n\u003cli\u003eChung KS, Lee JK, Park JS, Choi CH. Risk factors of delirium in patients undergoing total knee arthroplasty. \u003cem\u003eArch Gerontol Geriatr\u003c/em\u003e. (2015) 60:443-7. doi: 10.1016/j.archger.2015.01.021.\u003c/li\u003e\n\u003cli\u003eMa X, Cheng H, Zhao Y, Zhu Y. Prevalence and risk factors of subsyndromal delirium in ICU: A systematic review and meta-analysis.\u003cem\u003e Intensive Crit Care Nurs\u003c/em\u003e. (2025) 86:103834. doi: 10.1016/j.iccn.2024.103834.\u003c/li\u003e\n\u003cli\u003eCheng J, Lao Y, Chen X, Qiao X, Sui W, Gong X, et al. Dynamic Nomogram for Subsyndromal Delirium in Adult Intensive Care Unit: A Prospective Cohort Study. \u003cem\u003eNeuropsychiatr Dis Treat\u003c/em\u003e. (2023) 19:2535-48. doi: 10.2147/NDT.S432776.\u003c/li\u003e\n\u003cli\u003eGao Y, Gan X. A novel nomogram for the prediction of subsyndromal delirium in patients in intensive care units: A prospective, nested case-controlled study. \u003cem\u003eInt J Nurs Stud\u003c/em\u003e. (2024) 155:104767. doi: 10.1016/j.ijnurstu.2024.104767.\u003c/li\u003e\n\u003cli\u003eQian J, Shen X, Gao X, Sun Q. Grip Strength is a Predictor for Subsyndromal Delirium Among Older Adults Following Joint Replacement. \u003cem\u003eClin Interv Aging\u003c/em\u003e. (2023) 18:1675-85. doi: 10.2147/CIA.S423727.\u003c/li\u003e\n\u003cli\u003eGabor JA, Feng JE, Schwarzkopf R, Slover JD, Meftah M. Machine Learning With Electronic Health Record Data Outperforms a Risk Assessment Prediction Tool in Predicting Discharge Disposition After Total Joint Arthroplasty. \u003cem\u003eOrthopedics\u003c/em\u003e. (2022) 45:e211-e215. doi: 10.3928/01477447-20220225-02.\u003c/li\u003e\n\u003cli\u003eHsu WH, Ko AT, Weng CS, Chang CL, Jan YT, Lin JB, et al. Explainable machine learning model for predicting skeletal muscle loss during surgery and adjuvant chemotherapy in ovarian cancer. \u003cem\u003eJ Cachexia Sarcopenia Muscle\u003c/em\u003e. (2023) 14:2044-53. doi: 10.1002/jcsm.13282.\u003c/li\u003e\n\u003cli\u003eCregar WM, Goodloe JB, Gerlinger TL. Increased Operative Time Impacts Rates of Short-Term Complications After Unicompartmental Knee Arthroplasty. \u003cem\u003eJ Arthroplasty\u003c/em\u003e. (2021) 36:488-94. doi: 10.1016/j.arth.2020.08.032.\u003c/li\u003e\n\u003cli\u003eDodsworth BT, Reeve K, Falco L, Hueting T, Sadeghirad B, Mbuagbaw L, et al. Development and validation of an international preoperative risk assessment model for postoperative delirium. \u003cem\u003eAge Ageing\u003c/em\u003e. (2023) 52:afad086. doi: 10.1093/ageing/afad086.\u003c/li\u003e\n\u003cli\u003eWang X, Yu D, Du Y, Geng J. Risk factors of delirium after gastrointestinal surgery: A meta-analysis. \u003cem\u003eJ Clin Nurs\u003c/em\u003e. (2023) 32:3266-76. doi: 10.1111/jocn.16439.\u003c/li\u003e\n\u003cli\u003eWachtendorf LJ, Azimaraghi O, Santer P, Linhardt FC, Blank M, Suleiman A, et al. Association Between Intraoperative Arterial Hypotension and Postoperative Delirium After Noncardiac Surgery: A Retrospective Multicenter Cohort Study. \u003cem\u003eAnesth Analg\u003c/em\u003e. (2022) 134:822-33. doi: 10.1213/ANE.0000000000005739.\u003c/li\u003e\n\u003cli\u003eZarour S, Weiss Y, Abu-Ghanim M, Iacubovici L, Shaylor R, Rosenberg O, et al. Association between Intraoperative Hypotension and Postoperative Delirium: A Retrospective Cohort Analysis. \u003cem\u003eAnesthesiology\u003c/em\u003e. (2024) 141:707-18. doi: 10.1097/ALN.0000000000005149.\u003c/li\u003e\n\u003cli\u003eGao Y, Gao R, Yang R, Gan X. Prevalence, risk factors, and outcomes of subsyndromal delirium in older adults in hospital or long-term care settings: A systematic review and meta-analysis. \u003cem\u003eGeriatr Nurs\u003c/em\u003e. (2022) 45:9-17. doi: 10.1016/j.gerinurse.2022.02.021.\u003c/li\u003e\n\u003cli\u003eHwang H, Lee KM, Son KL, Jung D, Kim WH, Lee JY, et al. Incidence and risk factors of subsyndromal delirium after curative resection of gastric cancer. \u003cem\u003eBMC Cancer\u003c/em\u003e. (2018) 18:765. doi: 10.1186/s12885-018-4681-2.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"machine learning, hip arthroplasty, subsyndromal delirium, elderly, risk prediction, model verification","lastPublishedDoi":"10.21203/rs.3.rs-8504448/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8504448/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground: \u003c/strong\u003esubsyndromal delirium (SSD) can gradually develop into complete delirium, leading to prolonged hospital stays, increased risks of complications and death. This study aims to construct and validate a postoperative subsyndromal delirium risk prediction model for elderly hip arthroplasty patients using machine learning algorithms based on our electronic health record data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e Electronic data of older adults who underwent hip arthroplasty at Tangdu Hospital, Second Affiliated Hospital of Air Force Military Medical University from January 2020 to December 2023 were retrospectively analyzed. The data were divided into a training set (n=442) and a validation set (n=190) according to a ratio of 7:3, and four machine learning algorithms: logistic regression (LR), extreme gradient boosting (XGB), Gaussian Naive Bayes (GNB), and random forests (RF) were used to construct a predictive model of postoperative SSD. The area under the receiver operating characteristic curve (AUROC), sensitivity, accuracy, and area under the precision-recall curve (AUPRC) were used to evaluate the above four prediction models, and the variables were selected based on the SHAP values for the prediction model based on the optimal machine learning algorithm.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e: A total of 632 elderly patients who underwent hip arthroplasty were included in this study, of which 265 (41.93%) developed postoperative SSD, with 23 (3.64%) progressed from SSD to POD. After a comprehensive analysis, we ultimately chose the optimal-performing XGB-based algorithm to develop the prediction model. The first 6 variables selected based on SHAP values were: operative time, ASA classification, cumulative duration of intraoperative hypotension, previous delirium, age, and postoperative use of analgesics. The postoperative SSD prediction model developed based on these 6 variables had an AUROC of 0.849 (0.769-0.929), and an AUPRC of 0.657 (0.600-0.714). In the external test set, the AUROC was 0.842 (0.754-0.930) and the AUPRC was 0.650 (0.621-0.679).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003e We used machine learning techniques to select six variables and demonstrate a predictive model for the risk of postoperative SSD in older adults with hip arthroplasty. The model would provide a useful tool for identifying patients at high risk for postoperative SSD, aiding in the identification of those with elevated postoperative SSD risk.\u003c/p\u003e","manuscriptTitle":"Construction and validation of a predictive model based on machine learning algorithm for the risk of subsyndromic delirium after hip arthroplasty in the elderly","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-03 14:41:59","doi":"10.21203/rs.3.rs-8504448/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-02-25T05:39:58+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-02-02T05:22:41+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-01-20T07:18:50+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-20T07:17:42+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Informatics and Decision Making","date":"2026-01-03T04:56:29+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"851d4965-f463-41ea-9b9d-3d030c92d958","owner":[],"postedDate":"March 3rd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-03T14:41:59+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-03 14:41:59","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8504448","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8504448","identity":"rs-8504448","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}