Development of an individualized model for predicting postoperative delirium in elderly patients with hepatocellular carcinoma

Postoperative delirium (POD) is a common complication in older patients with hepatocellular carcinoma (HCC) that adversely impacts clinical outcomes. We aimed to evaluate the risk factors for POD and to construct a predictive nomogram. Data for a total of 1481 older patients (training set: n=1109; validation set: n=372) who received liver resection for HCC were retrospectively retrieved from two prospective databases. The receiver operating characteristic (ROC) curve, calibration plot, and decision curve analysis (DCA) were used to evaluate the performance. The rate of POD was 13.3% (148/1109) in the training set and 16.4% (61/372) in the validation set. Multivariate analysis of the training set revealed that factors including age, history of cerebrovascular disease, American Society of Anesthesiologists (ASA) classification, albumin level, and surgical approach had significant effects on POD. The area under the ROC curves (AUC) for the nomogram, incorporating the aforementioned predictors, was 0.798 (95% CI 0.752–0.843) and 0.808 (95% CI 0.754–0.861) for the training and validation sets, respectively. The calibration curves of both sets showed a degree of agreement between the nomogram and the actual probability. DCA demonstrated that the newly established nomogram was highly effective for clinical decision-making. We developed and validated a nomogram with high sensitivity to assist clinicians in estimating the individual risk of POD in older patients with HCC.


Statistical analysis
The data were analyzed using both IBM SPSS 24.0 (IBM Corp) and R software (version 4.1.1).Parameters with a normal distribution are expressed as mean ± standard deviation and analyzed using the Student t-test.Parameters not following a normal distribution are expressed as median and interquartile range and analyzed using the Mann-Whitney test.Categorical variables are presented as frequency and percentage, and their comparisons were conducted using the chi-squared test.The least absolute shrinkage and selection operator (LASSO) regression model was used to select the optimal predictive variables.Subsequently, the identified key features were integrated into multivariable logistic regression analysis.Forest plots, constructed using GraphPad Prism, were used to visualize the results.Predictors found to be statistically significant were used to establish a nomogram system for diagnosing POD.The discriminative performance of the nomogram was assessed using the area under the receiver operating characteristic curve (AUC).Calibration curves were generated to evaluate the concordance between predicted and observed probabilities.Additionally, a decision curve analysis (DCA) was conducted to assess the clinical utility of the nomogram.Statistical significance was defined as a two-tailed P-value < 0.05.

Construction and validation of the nomogram
The nomogram was created using these independent factors as its foundation (Fig. 5).The AUCs of the nomogram model were 0.798 (95% CI 0.752-0.843) in the training set and 0.808 (95% CI 0.754-0.861) in the validation set (Fig. 6).The calibration curves generated by the nomogram exhibited strong concordance between observed outcome frequencies and predicted probabilities in both sets (Fig. 7).The results of DCA are shown in Fig. 8, which demonstrated that the nomogram used in our study had superior effectiveness compared with treating all patients or providing no treatment.This advantage was observed when the threshold probability ranged from 9 to 91% in the training set and from 5 to 87% in the validation set.

Discussion
The incidence of POD was 13.3% in the training set and 16.1% in the validation set.Consistent with prior research, Nomi et al. reported a POD rate of 14.2% 29 .Yoshimura et al. found a POD rate of 17.0% in patients undergoing hepatectomy 30 .However, Ishihara et al. reported a lower POD incidence of 7.5% 6 , and Chen et al. found this to be 8.4% 31 .Variability in POD rates may be attributed to several factors: a lack of consistent definitions and assessment methods for POD by researchers, the diverse clinical characteristics of patients, and inaccuracies in estimation owing to the use of retrospective research methods.
This research involved developing and validating a new nomogram for predicting POD in older HCC patients.The model demonstrated strong capabilities in both discrimination and calibration, which showed its clinical value.The model's robustness was further enhanced by external validation, affirming its applicability across different patient groups and clinical settings.To our knowledge, this represents the inaugural predictive model for POD in individuals diagnosed with HCC.
Previous studies have indicated a correlation between the risk of POD and the emergence of postoperative complications 32,33 .However, such data are not accessible before or during surgery and thus cannot be integrated into predictive models.
In this study, a significant correlation between advanced age and an increased risk of POD was observed.Older patients experience a decline in physical capabilities, brain tissue integrity, and stress response regulation, along with diminished levels of key central neurotransmitters like acetylcholine and epinephrine 34 .Age-related constriction of blood vessels reduces cerebral oxygenation, which can potentially lead to postoperative cerebral impairment 35,36 .Moreover, alterations in drug metabolism and response owing to aging may increase the adverse effects of medications, thereby increasing the likelihood of POD 37 .
The present study identified that a history of cerebrovascular disease is an independent risk factor for POD.Cerebrovascular disease can lead to cognitive impairment, dementia, and neurocognitive deficits, which is postulated to increase delirium possibly through altered brain networks and a reduced ability to integrate sensory inputs 38 .Long-term susceptibility to delirium should be regarded as an integral aspect of the overall cerebrovascular disease burden 39,40 .Several studies have indicated that cognitive dysfunction and reduced functional capacity are associated with a heightened risk of POD [41][42][43] .
The ASA physical status classification system is commonly applied to evaluate a patient's ability to withstand anesthesia, primarily based on their overall compromised health and the presence of multiple comorbidities 44 .Research has indicated that an ASA classification ≥ 3 is associated with an increased risk of complications and decreased overall survival after hepatectomy [45][46][47] .Our study indicated that an ASA classification ≥ 3 is a risk factor for POD, as evidenced in numerous studies on this topic [48][49][50][51] .Whereas we found no statistically significant differences in common comorbidities such as diabetes and hypertension between the groups, it is considered that the cumulative impact of various comorbidities might heighten baseline vulnerability in older patients.This susceptibility, combined with the stress of surgery, could be a contributing factor to the development of POD 52,53 .
Numerous research findings indicate that a lower patient albumin level increases their likelihood of experiencing POD, a conclusion that aligns with the findings of our study 6,[54][55][56][57] .Hypoalbuminemia affects drug metabolism, antioxidant defense, and toxin processing because albumin is the primary transport protein in blood plasma.Reduced albumin levels may result in cognitive dysfunction owing to toxic effects and oxidative injuries 58,59 .Appropriate medical intervention can yield lower albumin levels, potentially aiding in the reduction of POD risk.
Vol:.( 1234567890   This study showed that an open approach independently increases the risk of POD.A laparoscopic approach may reduce operative stress and postoperative systemic inflammation, which are known to be linked to the occurrence of POD 29,[60][61][62] .
There are a number of limitations in this study.First, this research was a retrospective evaluation conducted using a prospectively registered database.Recognizing the intrinsic biases inherent in this type of study design is crucial.Prior studies have highlighted several risk factors linked to POD, including preoperative depression and anxiety [63][64][65][66][67][68] .Nevertheless, these factors were not incorporated into our analysis owing to certain constraints.Second, the experiment was conducted in only two centers, both of which are located in the same city.To further validate the model, it is necessary to use a more extensive sample size and conduct studies across various centers in different regions.

Conclusion
In older patients with HCC, factors such as age, cerebrovascular disease history, ASA classification, albumin levels, and the type of surgical procedure are identified as independent predictors of POD.In this study, we developed and externally validated a new, precise nomogram for personalized assessment and clinical decision-making.

Figure 2 .
Figure 2. Clinicopathologic feature selection using the LASSO regression model.(A) The smallest lambda is determined through tenfold cross-validation.(B) LASSO coefficient profiles of 40 signatures.When the smallest lambda equals 0.015, the eight coefficients with non-zero values are selected.

Figure 3 .
Figure 3. Receiver operating characteristic (ROC) curves of the age (A) and albumin level (B).AUC Area under the ROC curve.

Figure 4 .
Figure 4. Forest plot of independent predictors of postoperative delirium.ASA American Society of Anesthesiologists, HR Hazard ratio, CI Confidence interval.

Figure 5 .
Figure 5. Nomogram for estimating the likelihood of POD in older patients diagnosed with HCC.ASA American Society of Anesthesiologists.

Figure 6 .
Figure 6.Receiver operating characteristic (ROC) curves in the training set (A) and validation set (B). AUC Area under the ROC curve.

Figure 7 .
Figure 7. Calibration curves of the nomogram in the training set (A) and validation set (B).The horizontal axis depicts the anticipated likelihood of POD, and the vertical axis illustrates the actual occurrence of diagnosed POD relative to the total cases.The diagonal dashed line represents the perfect prediction of the ideal model.The solid line represents the prediction of the nomogram; a closer fit to the diagonal dashed line represents the result after bias correction by bootstrapping (1000 repetitions).

Figure 8 .
Figure 8. Decision curve analysis of the nomogram in the training set (A) and validation set (B).The net benefit is quantified along the y-axis.The red line denotes predictions from the nomogram, the green line signifies the assumption of POD occurrence in all patients, and the blue line signifies the assumption of no POD occurrence in any patient.

Table 1 .
Patients' background characteristics in the training and validation sets.ASA American society of anesthesiologists, BUN Blood urea nitrogen, BMI Body mass index, IQR Interquartile range, SD Standard deviation, ALT Alanine transaminase, AST Aspartate transaminase, FEV1/FVC Ratio of forced expiratory volume in the first 1 s to forced vital capacity.