INTRODUCTION
Postoperative delirium (POD) is one of the most frequent and serious complications in elderly surgical patients and is associated with prolonged hospitalization, functional decline, long-term cognitive impairment, institutionalization, and increased rates of mortality [1-3]. Despite advances in perioperative medicine, POD remains underrecognized and undertreated, particularly in patients undergoing non-cardiac surgeries. Laparoscopic gynecologic surgery is being increasingly performed in elderly women, who often have multiple comorbidities and age-related vulnerabilities; however, POD in this population has received less attention.
Neuroinflammation is considered the central mechanism linking perioperative stress and delirium [1,2]. Surgical trauma, anesthetic agents, and hemodynamic instability can trigger systemic inflammatory responses that may disrupt the integrity of the blood-brain barrier, activate microglia, and alter neurotransmission. Numerous studies have investigated the biomarkers of systemic inflammation, most notably the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), and systemic immune-inflammation index (SII), as predictors of delirium and other adverse outcomes [3,4]. Among these, the NLR, which is derived from a routine complete blood count, is particularly attractive because it is inexpensive, widely available, and easily calculated.
Recent studies have suggested that anesthetic techniques may modulate systemic inflammation [4-8]. Our group previously observed that remimazolam-based anesthesia attenuated postoperative inflammatory indices in comparison with volatile anesthesia in elderly gynecologic patients, but did not clearly reduce POD [9]. This apparent dissociation raises an important clinical question: even if modifying perioperative inflammation by anesthetic choice does not reliably prevent delirium, can inflammatory markers still serve as risk-stratification tools to facilitate prevention and monitoring?
Evidence from meta-analyses indicates that an elevated preoperative NLR is associated with POD in diverse surgical populations [10]. However, many of the existing studies were heterogeneous in design, frequently lacked standardized delirium assessments, and provided limited data on focused laparoscopic gynecological cohorts. Furthermore, only a few studies translated the data for biomarker associations into pragmatic prediction models with internal validation and clinically usable risk thresholds.
Therefore, we conducted a retrospective cohort study of elderly women who underwent laparoscopic gynecologic surgery under general anesthesia. Our primary objective was to evaluate whether the preoperative and early postoperative NLR, along with other inflammatory indices, could predict POD within 7 days postoperatively. The secondary objective was to develop and internally validate a simple multivariable prediction model and an NLR-based risk threshold that could be readily applied in routine clinical practice.
MATERIALS AND METHODS
Study design and ethics
This retrospective, observational cohort study was conducted at a tertiary university hospital. We reviewed the electronic medical records of patients aged ≥ 65 years who underwent elective laparoscopic gynecologic surgery under general anesthesia during the study period. For risk prediction and model development, we additionally performed an outcome-matched (1:1) propensity-score analysis between patients with and without POD, which represented a nested case-control sampling strategy within the cohort. The study protocol was approved by the Institutional Review Board (IRB) of Wonkwang University Hospital (WKUH IRB No. 2023-11-031), and the requirement for written informed consent was waived because the study involved analysis of de-identified data obtained from routine clinical care.
All the study procedures were conducted in accordance with the principles of the Declaration of Helsinki. This study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [11].
Setting and participants
We identified consecutive women aged ≥ 65 years who underwent elective laparoscopic gynecologic surgery under general anesthesia between January 2018 and December 2023. Eligible procedures included laparoscopic hysterectomy, oophorectomy, adnexal surgery, and staging procedures for suspected or confirmed gynecologic malignancies.
The exclusion criteria were documented preoperative delirium, psychosis, or advanced dementia; chronic antipsychotic therapy; emergency procedures or combined/open approaches; missing preoperative or postoperative day-1 laboratory data required to compute inflammatory indices; and the absence of any documented delirium assessment within the first 7 days postoperatively.
To assess the potential selection bias related to exclusions caused by missing laboratory data, the baseline demographic and clinical characteristics of the patients included in the final analytical cohort were compared with those of patients excluded for missing laboratory values. The results of this comparison are presented in Supplementary Table 1.
Perioperative management
All patients received general anesthesia in accordance with the institutional protocols. Anesthesia was induced and maintained using remimazolam-based total intravenous anesthesia, volatile-based inhalational anesthesia, or a balanced combination of remimazolam and volatile agents. Opioid analgesia was provided using remifentanil and/or fentanyl. The choice of anesthetic technique, intraoperative fluid and vasopressor management, and postoperative analgesia (including intravenous patient-controlled analgesia [PCA]) were at the discretion of the attending anesthesiologist (CL).
The perioperative variables analyzed in the study included anesthesia duration, estimated blood loss, intraoperative red blood cell (RBC) transfusion, intraoperative hypotension requiring treatment, intraoperative bradycardia, and postoperative 24-h opioid consumption.
Delirium assessment and outcome definition
POD was assessed using the Confusion Assessment Method (CAM) in surgical wards [11] and the CAM-intensive care unit (ICU) in high-dependency or intensive care settings [12] as part of a routine postoperative screening protocol. Trained nurses and physicians performed CAM-based assessments at least once daily from postoperative day 0 (day of surgery) to postoperative day 7; additional assessments were performed whenever delirium was clinically suspected.
The primary outcome was POD occurring within the first seven postoperative days (postoperative days 0-7, inclusive of both days). If any CAM or CAM-ICU assessment yielded positive results during this interval, the patient was classified as having POD.
Although the CAM and CAM-ICU are screening instruments, they were used for the operational definition of POD because formal Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) or International Classification of Diseases, 10th Revision (ICD-10)-based diagnostic adjudication was not performed systematically in routine clinical practice, and assessments based on these formal criteria were not consistently documented in the electronic medical records. In this retrospective cohort, restricting the outcome definition to DSM-based diagnoses would have resulted in substantial outcome misclassification and the loss of clinically recognized delirium cases. Accordingly, any positive CAM or CAM-ICU assessment during the predefined postoperative window was considered indicative of clinically relevant POD, an approach that has been widely adopted in observational studies of perioperative delirium. This pragmatic definition reflects real-world delirium detection and management while maintaining consistency across ward and intensive care settings [11,12].
Laboratory measurements and inflammatory indices
Peripheral venous blood samples were obtained within 24 h before surgery and again on postoperative day 1 as part of routine care. Complete blood count data and the data for differential counts and C-reactive protein (CRP) levels were obtained from electronic health records.
From these data, we calculated the following indices: NLR = absolute neutrophil count/absolute lymphocyte count, PLR = platelet count/absolute lymphocyte count, and SII = (platelet count × neutrophil count)/lymphocyte count.
Covariates
Prespecified clinical covariates were as follows: age, body mass index (BMI), and American Society of Anesthesiologists (ASA) physical status; comorbidities, namely, hypertension, diabetes mellitus, coronary artery disease, prior cerebrovascular accident, chronic obstructive pulmonary disease, and peripheral arterial occlusive disease; educational level (years), smoking status, and alcohol consumption; surgical indications (benign vs. malignant disease) and cancer type (uterine, ovarian, cervical); anesthesia and surgery duration, estimated blood loss, and intraoperative RBC transfusion; intraoperative hemodynamic instability (hypotension, bradycardia); cumulative PCA opioid consumption over the first 24 h postoperatively; and length of hospital stay. These variables were selected on the basis of previous literature on patient vulnerability, intraoperative events, and systemic inflammation during the development of delirium [1-3,13].
Propensity-score matching (PSM)
To create a balanced dataset for risk prediction and model development, we performed 1:1 PSM between patients with and without POD. Propensity scores were estimated using a logistic regression model that included age, ASA class, BMI, major comorbidities, malignancy, and anesthesia duration. We applied nearest-neighbor matching without replacement with a caliper width of 0.20 on the logit of the propensity score [14,15].
This outcome-based matching strategy was implemented to minimize baseline differences in vulnerability and facilitate the evaluation of biomarker-outcome associations rather than causal inference regarding biomarker exposure. Covariate balance was assessed using standardized mean differences (SMDs) before and after matching (Supplementary Data 1). Although the balance improved for most covariates, residual imbalance persisted for age and malignancy (SMD > 0.10); therefore, these variables were prespecified for forced inclusion in all subsequent multivariable logistic regression models to mitigate residual confounding.
Statistical analysis
Continuous variables were summarized as mean ± standard deviation, and categorical variables were summarized as counts and percentages. Between-group comparisons were performed using Student’s t-test or the Mann-Whitney U test for continuous variables and the chi-square or Fisher’s exact test for categorical variables, as appropriate.
Associations between inflammatory biomarkers and POD were evaluated using point-biserial correlation coefficients. Univariate and multivariate logistic regression models were constructed to assess the associations between the candidate predictors and POD. In all multivariate logistic regression models, covariates demonstrating residual imbalance after PSM, specifically age and malignancy, were forcibly included regardless of statistical significance. Potential nonlinear relationships between the preoperative NLR and POD were explored using restricted cubic spline functions within a multivariate regression framework.
Model discrimination was assessed using the area under the receiver operating characteristic curve (AUC), and calibration was evaluated using calibration plots and statistics. Internal validation was performed using 1,000 bootstrap resamples to estimate optimism-corrected performance measures. Because outcome-based matching distorts the absolute outcome prevalence, predictive values such as positive and negative predictive values were not interpreted from the matched cohort. The diagnostic performance at the selected cutoffs was summarized using sensitivity, specificity, and likelihood ratios (LRs). Optimal biomarker cutoff values were derived using Youden’s index.
Since this was a retrospective study, no a priori sample-size calculations were performed. To address sample-size adequacy, we conducted a post-hoc power analysis for the observed between-group differences in preoperative NLR in the overall cohort (No POD: 3.0 ± 2.8; POD: 7.5 ± 4.0; n = 245 vs. 111), which corresponded to a large standardized effect size (Cohen’s d ≈ 1.40) and > 99.9% power at a two-sided alpha = 0.05.
Sensitivity analyses included complete-case analyses, exclusion of patients with malignancy, and alternative definitions of POD restricted to the first three postoperative days. All statistical analyses were performed using IBM SPSS Statistics ver. 29.0 (IBM Co.). Two-sided P values < 0.05 were considered statistically significant.
RESULTS
Patient flow and incidence of POD
Among the 940 eligible patients, 356 met the inclusion criteria and constituted the overall analysis cohort. POD occurred within the first seven postoperative days in 111 patients (31.2%). PSM resulted in a balanced cohort of 222 patients (111 with POD and 111 without POD), which were used for subsequent biomarker association and prediction analyses (Fig. 1). A comparison of the baseline characteristics between the included and excluded patients owing to missing laboratory data is provided in the Supplementary Table 1.
Baseline and perioperative characteristics before and after matching
Before PSM, patients who developed POD were older than those without POD (70.2 ± 4.1 vs. 67.5 ± 3.8 years, P < 0.001) and had a higher prevalence of malignancy (40.5 vs. 22.9%, P = 0.001). Intraoperative hypotension (31.5 vs. 15.1%, P < 0.001) and bradycardia (17.1 vs. 5.7%, P = 0.001) were also more frequent in the POD group. The incidences of other comorbidities at baseline and BMI values were similar between the groups (Table 1; Supplementary Table 2).
After PSM, most baseline characteristics were well-balanced between the groups. The distribution of ASA physical status classes (I-III) did not differ significantly between groups, whereas age and the prevalence of malignancy remained modestly higher in patients with POD (age: 70.0 ± 3.4 vs. 67.2 ± 1.4 years; malignancy: 34.2 vs. 16.2%), corresponding to SMDs greater than 0.1 (Table 1; Supplementary Table 2). The covariate balance before and after matching is illustrated graphically (Supplementary Fig. 1). Perioperative hemodynamic instability was also more frequent in the POD group, whereas postoperative opioid consumption and the incidences of most comorbidities were comparable between the groups. Accordingly, age, malignancy, and other clinically relevant covariates were retained as adjustment variables in all multivariate analyses. The preoperative and postoperative inflammatory biomarker levels continued to differ substantially between the groups after matching (Table 1).
Differences in inflammatory biomarkers in relation to POD status
In the matched cohort, preoperative NLR was markedly higher in patients who developed POD than in those who did not (7.7 ± 4.1 vs. 3.2 ± 3.0, P < 0.001). Postoperative NLR demonstrated a similar pattern (9.1 ± 4.5 vs. 4.5 ± 3.1, P < 0.001). Pre- and postoperative SII values were also significantly higher in patients with POD, whereas PLR showed weaker discrimination between the groups. Preoperative CRP levels did not differ significantly, whereas postoperative CRP levels were higher in patients with POD (Table 1).
Correlation between biomarkers and POD
Point-biserial correlation analyses demonstrated that the NLR showed the strongest association with POD among all evaluated biomarkers. The preoperative NLR was strongly correlated with POD (r = 0.540, P < 0.001), followed by the postoperative NLR (r = 0.580, P < 0.001). The SII exhibited moderate correlation with POD, whereas PLR and CRP showed weak or non-significant correlations. These findings support the selection of NLR as the primary biomarker for multivariate modeling and prediction (Table 2).
Logistic regression analyses of biomarker predictors
Univariate logistic regression analyses showed that preoperative NLR, postoperative NLR, SII, and PLR, and postoperative CRP levels were associated with POD. After multivariate adjustment for age and malignancy, which showed residual imbalance after matching, as well as ASA physical status, anesthesia duration, intraoperative hypotension, and anesthetic technique, preoperative NLR remained independently associated with POD (adjusted odds ratio [OR]: 1.29 per unit increase; 95% confidence interval [CI]: 1.17-1.42), as did postoperative NLR (adjusted OR: 1.08 per unit increase; 95% CI: 1.01-1.15) (Table 3). The complete results of univariate and multivariate regression analyses, including all clinical covariates and inflammatory biomarkers, are presented in Supplementary Table 3.
Predictive performance of models incorporating NLR
The clinical-only prediction model demonstrated modest discriminative ability for POD, with an AUC of 0.68. The addition of the preoperative NLR improved the model discrimination (AUC, 0.78), and the inclusion of both preoperative and postoperative NLR values yielded the highest discriminative performance (AUC, 0.84). These findings indicate the incremental predictive value of NLR beyond conventional clinical variables (Fig. 2). The receiver operating characteristic (ROC) curves for the biomarker-only models are shown in Supplementary Fig. 2.
Nonlinear association between preoperative NLR and POD
Restricted cubic spline analysis revealed a monotonic increase in the adjusted odds of POD with increasing preoperative NLR values across the observed range. No clear threshold or plateau effect was identified, supporting a dose-response relationship between the systemic inflammatory burden and delirium risk (Fig. 3).
Biomarker cutoff values and model validation
Using Youden’s index, the optimal preoperative NLR cutoff for predicting POD was 7.1. Corresponding cutoff values were also identified for the SII, PLR, and CRP, although the NLR demonstrated superior overall performance. Internal bootstrap validation demonstrated good calibration and minimal optimism (optimism-corrected AUC ≈ 0.82), with calibration plots confirming agreement between predicted and observed risks (Supplementary Fig. 3). The final multivariable model incorporating clinical covariates and preoperative and postoperative NLR achieved an apparent AUC of 0.84 in the matched cohort. Internal bootstrap validation demonstrated good calibration and yielded an optimism-corrected AUC of approximately 0.82, indicating minimal overfitting. Sensitivity analyses, including the exclusion of patients with malignancy, complete-case analyses, and alternative POD definitions, yielded consistent effect estimates and model performance, supporting the robustness of the association between NLR and POD. At the optimal preoperative NLR cutoff of 7.1, the sensitivity was 62.2% and specificity was 78.4%, corresponding to a positive LR (LR+) of 2.88 and a negative LR (LR-) of 0.48 (Table 4).
Sensitivity and subgroup analyses
To address potential confounding factors by malignancy, we performed prespecified subgroup analyses stratified by surgical indication (malignant vs. benign disease). In patients with malignant disease, the preoperative NLR remained independently associated with POD after multivariate adjustment, with effect estimates comparable to those observed in the overall matched cohort. Similarly, in patients undergoing surgery for benign conditions, a higher preoperative NLR was consistently associated with increased POD risk, although the optimal cutoff value was modestly lower than that observed in the malignant subgroup.
When the final multivariate model incorporating clinical covariates and NLR was evaluated within the prespecified subgroups, model discrimination remained acceptable. In the malignant subgroup, the apparent AUC was 0.81, whereas in the benign subgroup it was 0.83. These subgroup-specific AUC values were not directly comparable to those shown in Fig. 2, which depicts the model discrimination across different predictor sets in the overall matched cohort. Internal bootstrap validation yielded optimism-corrected AUC values that were slightly lower, but consistent across subgroups, indicating stable model performance (Supplementary Table 4).
DISCUSSION
In this retrospective cohort of elderly women undergoing laparoscopic gynecologic surgery, simple inflammatory indices, particularly the NLR, were strongly associated with POD occurring within 7 days postoperatively. The preoperative NLR was substantially higher in patients who developed delirium, and this association persisted after adjustments for age, ASA class, malignancy, anesthesia duration, intraoperative hypotension, and anesthetic technique. Although the postoperative NLR provided incremental predictive value, the major improvement in discrimination was derived from preoperative measurements.
A multivariate model combining the preoperative NLR with key clinical covariates, specifically age, ASA physical status, malignancy status, anesthesia duration, intraoperative hypotension, and anesthetic technique, achieved good discrimination and calibration. In the matched cohort, the model demonstrated an apparent AUC of 0.84 in the ROC analysis, while internal bootstrap validation yielded an optimism-corrected AUC of approximately 0.82, reflecting minimal optimism and good model stability. An NLR cutoff of approximately 7.1 identified a high-risk subgroup in which delirium was common [15,16]. These findings suggest that the NLR, a low-cost and widely available biomarker, can meaningfully enhance delirium risk stratification in elderly women undergoing laparoscopic gynecological surgery.
Our results extend those of prior studies linking systemic inflammation and delirium in elderly surgical patients [1-3]. Several studies have reported that an elevated preoperative NLR is associated with POD across diverse surgical settings; however, many of these studies were limited by heterogeneous populations, variable approaches for delirium ascertainment, or lack of validation [15-18]. By focusing on a relatively homogeneous cohort of elderly women undergoing laparoscopic gynecologic procedures, employing systematic CAM-based delirium screening for up to 7 days, and using a prediction-oriented analytic strategy with internal validation, our study provides more tailored evidence for this population.
The NLR may outperform other composite inflammatory indices such as PLR or SII because it integrates two complementary biological processes that are central to the pathogenesis of POD [19,20]. Elevated neutrophil counts reflect acute innate immune activation in response to surgical stress, ischemia-reperfusion injury, and systemic inflammation, which can promote endothelial dysfunction, blood-brain barrier disruption, and microglial activation, leading to delirium [21,22]. Conversely, relative lymphopenia represents impaired adaptive immune regulation and reduced neuroimmune resilience, a state associated with frailty, poor stress tolerance, and vulnerability to acute cognitive dysfunction in older adults [23,24]. Therefore, the imbalance between the exaggerated inflammatory burden and diminished host regulatory capacity captured by NLR may more closely mirror the neuroinflammatory cascade underlying delirium than platelet-based indices [1,3,10]. In contrast, the PLR and SII incorporate platelet-related components that are more sensitive to perioperative fluid shifts, transfusion, and hematologic variability; therefore, they may be less specific markers of the neuroinflammatory processes driving POD [3].
We had previously observed that remimazolam-based anesthesia attenuated postoperative inflammation, but did not clearly reduce POD in comparison with volatile anesthesia [9]. This apparent disconnect underscores the multifactorial nature of delirium and indicates that modulating a single aspect of the inflammatory response may be insufficient to prevent inflammation. The present analysis showed that the NLR remains a robust risk marker after adjusting for anesthetic technique and intraoperative hemodynamic instability, reinforcing its potential value in prediction rather than as a direct therapeutic target.
Our findings are consistent with meta-analysis data indicating that an elevated preoperative NLR predicts POD and other neurocognitive outcomes [10]. Importantly, we demonstrated that preoperative NLR alone, measured before anesthesia induction, provides substantial discriminatory ability, and that postoperative NLR confers only modest additional improvement. This is clinically relevant because the preoperative NLR is available before surgery and can inform risk stratification and perioperative planning.
The most immediate implication of this study is that the preoperative NLR can serve as a practical trigger for focused delirium prevention and monitoring in elderly women undergoing laparoscopic gynecologic surgery. Because complete blood counts are almost universally obtained during preoperative assessments, calculating the NLR imposes no additional cost or logistical burden.
Clinicians could use the NLR in combination with a small set of clinical predictors to achieve the following objectives: (1) identify patients who may benefit from multicomponent delirium-prevention bundles, including orientation, early mobilization, sleep optimization, and sensory aids; (2) tailor anesthetic and analgesic plans to minimize oversedation and exposure to deliriogenic medications; (3) plan closer postoperative surveillance with more frequent CAM assessments; (4) trigger early involvement of geriatric or psychiatric consultation for high-risk individuals; and (5) support shared decision-making and realistic counseling regarding the postoperative course and discharge planning.
However, the NLR should not be interpreted in isolation. It is best integrated with a broader assessment that includes evaluations of frailty, baseline cognitive function, comorbidities, and psychosocial factors. Future studies should explore how NLR-based risk algorithms can be embedded in electronic health records and combined with structured delirium-prevention pathways.
This study had several strengths. We examined a clearly defined surgical population at a meaningful risk of delirium, used standardized CAM-based tools for POD detection over a 7-day window, and prespecified a prediction-focused modeling approach with PSM and bootstrap internal validation. The reliance on routinely available laboratory and clinical variables highlights the feasibility and scalability of the proposed risk model, even in resource-limited settings.
Nevertheless, our findings should be interpreted in the context of several limitations. First, the retrospective, single-center design introduced the potential for residual confounding factors and limited the generalizability of the findings. Second, although PSM was used to improve baseline comparability, the 1:1 outcome-matching strategy created an analytical dataset that was methodologically analogous to a nested case-control sample within the cohort and did not preserve the natural incidence of POD. Accordingly, we did not interpret the absolute predictive values (e.g., positive and negative predictive values) from the matched cohort and instead emphasized prevalence-independent measures of performance, including discrimination (AUC), calibration, sensitivity/specificity, and LRs. In addition, PSM did not fully balance baseline characteristics, particularly age (post-matching SMD ≈ 0.20), and residual confounding may have persisted despite forced adjustment for age and malignancy in multivariable models. Third, a substantial proportion of the screened patients were excluded due to missing laboratory data, raising the possibility of selection bias; however, the baseline characteristics of the included and excluded patients were broadly similar (Supplementary Table 1). Fourth, no a priori sample-size calculations were performed, because this was a retrospective study. To address this limitation, we provided a post-hoc power assessment based on the observed between-group differences in preoperative NLR in the overall cohort; however, post-hoc power analyses should be interpreted cautiously and should not replace prospective planning. Fifth, we lacked detailed measures of baseline cognitive function and frailty, which are important determinants of delirium risk that may interact with inflammatory pathways. Sixth, biomarkers were measured at only two time points (preoperatively and on postoperative day 1), precluding the assessment of more detailed inflammatory trajectories. Finally, although internal validation using bootstrap resampling demonstrated model stability, external validation in independent cohorts is required before routine clinical adoption. Finally, NLR can be influenced by acute infection, corticosteroid therapy, hematologic disorders, or other unmeasured factors; therefore, clinicians should interpret elevated NLR values within a broader clinical context rather than relying on a rigid numerical threshold.
Conclusions
In elderly women undergoing laparoscopic gynecologic surgery, the preoperative NLR is a strong and independent predictor of POD within 7 days postoperatively. Adding the NLR to a small panel of clinical variables substantially improved risk discrimination and identified a high-risk subgroup in which delirium was common. Thus, a simple NLR-based risk model may help clinicians target delirium prevention and monitor patients who are most likely to benefit. Future multicenter studies should externally validate this model, refine the NLR thresholds, and evaluate their integration into comprehensive perioperative brain health pathways.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement1
