Risk factors for surgical site infection after craniotomy for meningioma resection in a tertiary hospital in China

Abstract

BACKGROUND

Meningioma is the most common intracranial neoplasm and carries a non-negligible surgical site infection (SSI) risk (reported incidence: 4%-6%) despite its clean-wound classification. Most existing studies enroll heterogeneous neurosurgical cohorts spanning vascular, traumatic, and biologically distinct tumor entities, even though meningioma itself is an independent SSI risk factor, which calls for disease-specific investigation. Preoperative host physiologic reserve indices have shown prognostic value across oncology and neurosurgery, yet remain unevaluated as predictors of SSI after meningioma resection in Chinese cohorts. We hypothesized that precisely measured surgery time and preoperative immune-nutritional status are associated with SSI risk in this population.

AIM

To examine modifiable factors associated with SSI after meningioma resection, focusing on precisely measured surgery time and preoperative immune-nutritional indices.

METHODS

A 1:4 matched case-control study was conducted at a tertiary neurosurgical center between January 2022 and November 2025, with matching on age, sex, and year of surgery. SSI was defined according to United States Centers for Disease Control and Prevention criteria. Surgery time was defined as the interval from skin incision to wound closure. Preoperative indices, including neutrophil-to-lymphocyte ratio (NLR), prognostic nutritional index, nutritional risk index, and modified frailty index, were extracted from routine clinical assessments. Least Absolute Shrinkage and Selection Operator regression followed by conditional logistic regression was used, with restricted cubic splines examining the dose-response relationship between surgery time and SSI risk.

RESULTS

Longer surgery time was independently associated with SSI (odds ratio = 1.824 per hour; 95% confidence interval: 1.405-2.369), with a threshold of 3.04 hours beyond which SSI risk increased 5.7-fold. Preoperative NLR was also independently associated with SSI risk (odds ratio = 1.081; 95% confidence interval: 1.007-1.159). The model showed good discrimination (area under the curve = 0.828) with acceptable calibration.

CONCLUSION

Longer surgery time and elevated preoperative NLR are associated with SSI after meningioma resection. A 3.04-hours threshold offers a practical criterion for prospective risk stratification.

Key Words: Surgical site infection; Meningioma; Craniotomy; Risk factors; Perioperative assessment

Core Tip: This disease-specific case-control study is among the first to systematically evaluate preoperative host physiologic reserve indices - including the modified frailty index, nutritional risk index, neutrophil-to-lymphocyte ratio, and prognostic nutritional index - as predictors of surgical site infection following meningioma resection in a Chinese neurosurgical cohort. Using precisely measured operative time (skin incision to wound closure) and Least Absolute Shrinkage and Selection Operator-informed multivariable modeling, we identified a 3.04-hour operative threshold beyond which surgical site infection risk increases 5.7-fold, and established preoperative neutrophil-to-lymphocyte ratio as an independent immune-nutritional predictor. These findings provide clinically actionable criteria for prospective risk stratification and targeted perioperative optimization.

INTRODUCTION

Meningioma is the most common intracranial neoplasm, accounting for approximately 39% of all intracranial tumors[1]. Craniotomy remains the mainstay of treatment. Surgical site infection (SSI), occurring in 4% to 6% of cases, is a serious complication that prolongs hospitalization, raises costs, and delays adjuvant therapy[2].

Most prior studies of post-craniotomy SSI have pooled heterogeneous neurosurgical populations spanning vascular, traumatic, and biologically distinct tumor entities. This obscures the risk profile specific to clean, elective meningioma surgery, even though meningioma itself is an independent risk factor for SSI[3]. The disease-specific evidence that does exist derives almost exclusively from Western cohorts, in which low nutritional risk index and elevated frailty have been associated with higher SSI risk[4-6]. Baseline characteristics and perioperative practice differ substantially between Western and Chinese populations, so these estimates cannot be directly extrapolated. Equivalent evidence from Chinese meningioma cohorts is lacking.

We conducted a 1:4 matched case-control study to determine the incidence of SSI after craniotomy for meningioma in a Chinese cohort and to examine both modifiable (e.g., operative duration and preoperative physiological status) and non-modifiable (e.g., tumor location) factors associated with it, with the aim of informing risk stratification and targeted infection prevention strategies in this patient population.

MATERIALS AND METHODS

Study design and patients

This retrospective study included adult patients who underwent elective craniotomy for meningioma resection at our institution between January 1, 2022, and November 30, 2025. We employed a 1:4 matched case-control design, with matching performed on age, sex, and year of surgery. These three variables were selected to control for the known demographic distribution of meningioma and temporal variation in institutional infection control practices, while avoiding overmatching given the limited number of SSI cases. Matching on year of surgery was intended to control for secular changes in perioperative care protocols, antibiotic prophylaxis, and SSI surveillance over the study period, preventing calendar time from acting as a confounder. Clinically relevant variables of primary interest, including World Health Organization grade and skull-base location, were deliberately not used for matching, so that their association with SSI could be assessed in the regression analysis. Clinical and demographic data were collected through manual review of preoperative laboratory results, anesthesia records, nursing medication administration records, pathological reports, and imaging data. Sample size was determined using an empirical approach satisfying the criterion of 10 events per variable, which supports the stability and robustness of the regression model.

Data collection

All laboratory values and anthropometric measurements were obtained within one week prior to surgery. The prognostic nutritional index (PNI) was calculated as: Serum albumin (g/L) + 5 × total peripheral lymphocyte count (× 10⁹/L). The geriatric nutritional risk index (GNRI) was derived using the formula: 1.489 × serum albumin (g/L) + 0.417 × (actual body weight/ideal body weight)[7]. The modified frailty index-5 (mFI-5) was calculated from five clinical variables: Diabetes mellitus, hypertension requiring medication, chronic obstructive pulmonary disease or active pneumonia, congestive heart failure, and non-independent functional status[8], with each variable contributing one point to the total score. Skull base meningiomas were defined as those arising from the olfactory groove, sphenoid ridge, tuberculum sellae, cerebellopontine angle, or petroclival region. Non-skull base meningiomas were defined as those located at the convexity, falx, parasagittal region, or other supratentorial or infratentorial sites outside the skull base[9]. Surgery time was defined as the interval from skin incision to wound closure, as documented in the anesthesia records. Data were also collected on systemic corticosteroid administration during the two-week period before and after surgery, and on prophylactic antibiotic administration within 0.5-1 hour prior to skin incision.

Outcomes and definitions

The primary outcome was organ/space SSI, defined in accordance with the United States Centers for Disease Control and Prevention criteria[10]. For craniotomy procedures, the SSI surveillance period was 90 days postoperatively. During the first 7 postoperative days, active surveillance was conducted through structured telephone follow-up. Beyond this window and up to 90 days, SSI cases were ascertained through review of institutional readmission records at our center. Organ/space SSI was defined as infection extending deeper than the fascial and muscle layers, encompassing intracranial infection and meningitis/ventriculitis.

The diagnostic criteria for intracranial infection required at least one of the following: (1) Positive pathogen detection by culture or non-culture-based methods from brain tissue or dura mater; (2) Evidence of abscess or infection identified by gross anatomical or histopathological examination; or (3) Presence of at least two clinical features (headache, fever > 38.0 °C, focal neurological deficits, altered level of consciousness, or confusion), accompanied by either pathogen detection on microscopic examination of brain tissue or abscess aspirate, or definitive imaging evidence of infection or abscess on computed tomography or magnetic resonance imaging. The diagnostic criteria for meningitis/ventriculitis required at least one of the following: (1) Positive pathogen detection in cerebrospinal fluid (CSF) by culture or non-culture-based methods; or (2) Presence of relevant symptoms (fever > 38.0 °C or headache with meningeal or cranial nerve signs), together with at least one of the following: Abnormal CSF biochemistry (elevated white blood cell count, elevated protein, and decreased glucose); pathogen detection on CSF gram stain; or positive blood culture with high clinical suspicion of meningitis.

Statistical analysis

All statistical analyses were performed using R Studio (R Foundation, Vienna, Austria). Missing data were handled using multiple imputation by chained equations (10 imputations, 20 iterations), with the imputation model including the outcome and all analysis variables to preserve their relationships[11]; identifiers were excluded. Continuous variables were imputed using predictive mean matching, categorical variables using logistic or polytomous regression. Continuous variables were compared between groups using the Mann-Whitney U test and categorical variables using the χ² test or Fisher’s exact test, as appropriate.

Variables associated with SSI were selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression followed by conditional logistic regression. To accommodate multiple imputation, LASSO (lambda. 1-standard-error criterion) was applied separately to each of the ten imputed datasets, and variables were retained according to their selection frequency across datasets, an inclusion-frequency thresholding strategy originally formalized by[12] and recently evaluated for penalized models under multiple imputation. Following the conventional majority rule[13,14], variables selected in at least 50% of imputed datasets were carried forward into the final conditional logistic regression model. Restricted cubic spline (RCS) analysis was performed to assess the linearity of the relationship between surgery time and SSI risk. Receiver operating characteristic (ROC) analysis was used to determine the optimal surgery time threshold. Model discrimination was evaluated by the area under the ROC curve (AUC), and calibration was assessed using a calibration plot. Bootstrap resampling was applied for internal validation. All tests were two-tailed, with statistical significance defined as P < 0.05. Sensitivity analyses, including dichotomization of key variables, adjustment for potential confounders, and subgroup analyses restricted to skull base meningiomas, were conducted to evaluate the robustness of the primary findings.

RESULTS

Patient’s characteristics at baseline

A total of 1803 patients with meningioma underwent craniotomy for tumor resection at our institution, of whom 46 (2.55%) developed postoperative SSI. According to the matching criteria, 184 non-SSI patients were selected as controls. The proportion of missing data for each variable is presented in Supplementary Table 1. Missing rates were low across all variables, with tumor size having the highest proportion at 6.25%. A comparison of baseline characteristics between the SSI and control groups is presented in Table 1.

Table 1 Demographic differences of the study cohort (n = 230), n (%).

Variable	Non-SSI (n = 184)	SSI (n = 46)	P value
Age (years)	57.00 (43.00, 62.00)	57.00 (43.00, 62.00)	0.948
Sex			> 0.999
Female	128 (70)	32 (70)
Male	56 (30)	14 (30)
Year			> 0.999
2021	20 (11)	5 (11)
2022	16 (8.7)	4 (8.7)
2023	36 (20)	9 (20)
2024	68 (37)	17 (37)
2025	44 (24)	11 (24)
Preop days	2.50 (1.00, 4.00)	3.50 (2.00, 4.75)	0.161
ASA			> 0.999
I-II	169 (92)	43 (93)
III-IV	15 (8.2)	3 (6.5)
Wound class			0.789
I	155 (84)	38 (83)
II	29 (16)	8 (17)
Blood loss (mL)	20.00 (0.00, 100.00)	0.00 (0.00, 100.00)	0.505
Skull base	72 (39)	33 (72)	< 0.001
Recurrence	20 (11)	5 (11)	> 0.999
Tumor size (mm)	27.00 (19.00, 39.12)	40.00 (30.00, 52.75)	< 0.001
WHO grade			0.413
1	155 (84.2)	38 (82.6)
2-3	29 (15.8)	8(17.4)
Intraop_EVD/LD	19 (10)	16 (35)	< 0.001
Steroid_periop	50 (27)	29 (63)	< 0.001
Preop_antibiotics	133(72)	26 (57)	0.038
mFI-5			0.035
0	96 (52)	16 (35)
≥ 1	88 (48)	30 (65)
PNI	51.50 (46.39, 54.32)	49.15 (44.26, 54.34)	0.308
NLR	2.13 (1.63, 3.17)	2.49 (1.64, 3.26)	0.143
GNRI	105.06 (100.18, 108.91)	104.01 (100.48, 108.07)	0.455
Pre-exist HAI before SSI	5 (2.7)	3 (6.5)	0.200
Surgery time (hour)	2.42 (1.67, 3.00)	4.17 (2.83, 6.00)	< 0.001

SSI: Surgical site infection; ASA: American Society of Anesthesiologists; WHO: World Health Organization; EVD/LD: External ventricular drain/Lumbar drainage; mFI-5: Modified 5-item frailty index; PNI: Prognostic nutritional index; NLR: Neutrophil-to-lymphocyte ratio; GNRI: Geriatric nutritional risk index; HAI: Hospital-acquired infection.

Bivariate and multivariable analysis of SSI risk factors

Of the variables assessed, eight showed a statistically significant association with postoperative SSI (Table 2). Surgery time, skull base location, intraoperative external ventricular drain or lumbar drain placement, tumor size, perioperative corticosteroid use, NLR, and mFI-5 score were associated with higher SSI risk; preoperative antibiotic prophylaxis was associated with lower risk. American Society of Anesthesiologists classification, intraoperative blood loss, GNRI, PNI, length of preoperative hospital stay, tumor recurrence, World Health Organization grade, and wound classification were not.

Table 2 Univariable conditional logistic regression results.

Variable	OR (95%CI)	P value
Preop days	1.13 (0.95-1.34)	0.177
ASA (III-IV)	0.79 (0.22-2.81)	0.717
Wound class (II)	1.12 (0.48-2.64)	0.789
Blood loss (per 100 mL)	1.03 (0.78-1.38)	0.816
Skull base (yes)	3.87 (1.91-7.84)	< 0.001
Recurrence (yes)	1.00 (0.36-2.81)	1.000
Tumor size	1.05 (1.02-1.07)	< 0.001
WHO grade (2-3)	1.53 (0.61-3.82)	0.365
Intraop_EVD_LD (yes)	4.25 (1.98-9.13)	< 0.001
Steroid_periop (yes)	5.10 (2.45-10.63)	< 0.001
Preop_antibiotics (yes)	0.50 (0.26-0.98)	0.043
mFI-5 (≥ 1)	2.25 (1.10-4.60)	0.025
PNI	0.99 (0.97-1.01)	0.358
NLR	1.06 (1.00-1.13)	0.067
GNRI	1.00 (0.97-1.04)	0.769
Pre-exist HAI before SSI (yes)	2.60 (0.57-11.89)	0.217
Surgery time	1.92 (1.50-2.46)	< 0.001

OR: Odds ratio; CI: Confidence interval; ASA: American Society of Anesthesiologists; WHO: World Health Organization; EVD_LD: External ventricular drain_ lumbar drainage; mFI-5: Modified 5-item frailty index; PNI: Prognostic nutritional index; NLR: Neutrophil-to-lymphocyte ratio; GNRI: Geriatric nutritional risk index; HAI: Hospital-acquired infection; SSI: Surgical site infection.

Applying the inclusion-frequency procedure described above, three variables were the most stable predictors across the ten imputed datasets: Surgery time (selected in 100% of datasets), tumor size (50%), and NLR (50%). These were carried forward into the final conditional logistic regression model. The LASSO cross-validation curve and coefficient paths for the first imputed dataset are shown in Figure 1; a heatmap of selection frequencies across all imputed datasets is provided in Supplementary Figure 1. This parsimonious model, which serves as our final inferential model, retained three variables and corresponds to approximately 15 events per variable, satisfying the conventional threshold of 10.

Figure 1 Variable selection by Least Absolute Shrinkage and Selection Operator regression (representative results from imputed dataset 1). A: Least Absolute Shrinkage and Selection Operator coefficient profiles across the regularization path; B: Cross-validation curve for tuning the penalty parameter; 7 variables were selected under the minimum criterion and 4 under the 1-standard-error criterion. Final model variables were determined by selection frequency across all 10 imputed datasets (Supplementary Figure 1). lambda.min: Lambda. minimum; lambda.1se: Lambda. 1-standard-error; NLR: Nutritional risk index.

Estimates from the multivariable conditional logistic regression model were pooled across the ten imputed datasets using Rubin’s rules (Table 3). Surgery time showed the strongest association with SSI [odds ratio (OR) = 1.824; 95% confidence interval (CI): 1.405-2.369; P < 0.001]. NLR was independently associated with SSI risk (OR = 1.081; 95%CI: 1.007-1.159; P = 0.030). Tumor size did not reach statistical significance.

Table 3 Multivariable conditional logistic regression (estimates pooled by Rubin’s rules across 10 imputed datasets).

Variable	OR (95%CI)	P value
Tumor size	1.02 (0.991-1.05)	0.169
NLR	1.081 (1.007-1.159)	0.030
Surgery time	1.824 (1.405-2.369)	< 0.001

OR: Odds ratio ; CI: Confidence interval; NLR: Neutrophil-to-lymphocyte ratio.

RCS analysis

RCS analysis was used to characterize the dose-response relationships of NLR and surgery time with SSI risk. Neither variable showed evidence of non-linearity. The relationship between surgery time and SSI risk was stable and linear (Figure 2). The confidence interval widened at the upper tail, but a sensitivity analysis confirmed that extreme values did not materially influence the findings (Supplementary Figure 2). Surgery time was then dichotomized at the optimal cut-off from the Youden index (3.04 hours; Figure 3). The dichotomized variable remained significantly associated with SSI (OR = 5.668; 95%CI: 2.596-12.374; P < 0.001) (Supplementary Table 2).

Figure 2 Restricted cubic spline of the association between surgery time (hours) and surgical site infection risk. The solid line indicates the estimated odds ratio and the shaded area the 95% confidence interval. SSI: Surgical site infection; RCS: Restricted cubic splines; OR: Odds ratio; CI: Confidence interval; Rug: Rug marks.

Figure 3  Sensitivity and specificity across surgery time thresholds.

Sensitivity analysis

To assess the robustness of the primary findings, a series of prespecified sensitivity analyses were conducted. Repeating LASSO variable selection using the lambda. minimum criterion, used here solely as a sensitivity analysis rather than for inference, retained more predictors (n = 7); effect estimates for the core variables remained consistent with the main model, supporting the more parsimonious lambda. 1-standard-error model (Supplementary Table 3). Among patients with surgery time exceeding 3 hours, the proportion of SSI cases was significantly higher in those who received intraoperative antibiotic re-dosing than in those who did not (76.5% vs 32.8%) (Supplementary Table 4). For NLR, effect estimates were stable across log-transformed, winsorized, and complete-case analyses (Supplementary Figure 3). After additional adjustment for corticosteroid use and antibiotic prophylaxis, the odds ratios for the core variables changed by less than 10% (Supplementary Figure 4). Surgery time remained a significant risk factor in both skull base and non-skull base subgroups, with no significant interaction detected (Supplementary Figure 5). Across these analyses, the primary findings were consistent.

Model evaluation and clinical translation

The final multivariable model showed good discrimination (AUC = 0.828; 95%CI: 0.751-0.884), a calibration slope of 0.883 (Supplementary Table 5), and a significant trend across linear predictor quartiles (P < 0.001) (Supplementary Figure 6). To support clinical applicability, a supplementary logistic regression model was constructed and visualized as a nomogram (Supplementary Figure 7). The nomogram serves as a visualization aid rather than the primary risk model, which remains the conditional logistic regression. Age, sex, and year of surgery were retained to maintain consistency with the matched cohort structure; they were not predictive of SSI and should not be interpreted as risk factors. Decision curve analysis showed that across threshold probabilities of 20%-60%, the model-based strategy yielded a net benefit above zero, outperforming the treat-all approach (Supplementary Figure 8). These results support the use of the model for risk stratification and perioperative decision-making in meningioma surgery.

DISCUSSION

In the present study, longer surgery time was independently associated with SSI following meningioma resection, with each additional hour corresponding to a higher infection risk (OR = 1.824 per hour). This association does not imply that operative duration is itself the principal cause; it is best interpreted as reflecting procedural complexity, for which surgery time serves as a measurable surrogate. ROC analysis identified 3.04 hours as the optimal clinical threshold, beyond which SSI risk increased approximately 5.7-fold. The direction is consistent with the broader literature. One systematic review reported that each 60-minute increment in operative duration was associated with a 37% increase in SSI incidence[15], and a meta-analysis restricted to brain tumor patients found surgery time to be on average 64 minutes longer in SSI cases than in uninfected controls[2]. The biological rationale is well established: Prolonged surgery increases wound exposure time, amplifies tissue trauma, and promotes localized ischemia, impairing host immunological defenses[16]. Surgery time also serves as an indirect index of procedural complexity, integrating factors such as tumor vascularity, neurovascular adhesion, and anatomical accessibility that are difficult to quantify individually. Its association with SSI therefore reflect the cumulative burden of these complexity-related factors as well as the direct biological consequences of prolonged tissue exposure. One single-center study of 304 meningioma patients reported no significant difference in surgery time between SSI and non-SSI groups, which may indicate that the influence of surgery time is attenuated when other determinants, particularly nutritional status, are inadequately controlled, or when institutional practice optimization masks its independent effect[5]. A further dimension comes from Mao et al[17], who showed that the effect of prolonged surgery on adverse outcomes was moderated by the time of day at which the procedure began, suggesting that the harms of lengthy operations may be amplified during late-night or fatigue-prone shifts. This interaction warrants consideration in surgical scheduling. The fatigue-related dimension operates independently of operative duration and was not captured by our exposure, which was confined to skin-incision-to-closure time; we regard it as a complementary axis of risk requiring dedicated prospective study rather than one addressed here.

A finding worth noting is that among patients with surgery time exceeding 3 hours, those who received intraoperative antibiotic supplementation had higher SSI rates than those who did not (76.5% vs 32.8%; P < 0.001). This counterintuitive result most plausibly reflects confounding by indication. Redosing was protocol-driven (triggered by duration > 3 hours and/or blood loss > 1500 mL), but adherence was incomplete and the decision also incorporated surgeon judgment, so supplementary antibiotics were preferentially given in procedures perceived as more complex or higher-risk. The small supplementation subgroup (n = 17) precludes definitive inference. This association must not be read as evidence against intraoperative antibiotic redosing, which remains a guideline-endorsed, evidence-based practice; it reflects confounding by indication not harm from redosing. What the data do underscore is the need for prospective studies of protocol adherence - when and why redosing is administered relative to protocol thresholds - to separate its true effect from the elevated baseline risk of the procedures in which it is used. The observation also reinforces the clinical relevance of the 3-hour threshold as a trigger for heightened vigilance. These data support a perioperative risk stratification framework centered on surgery time, measured precisely from skin incision to wound closure, as the primary exposure, since it most directly reflects the duration of tissue exposure and wound openness. This metric is intended to complement rather than replace other temporal measures. Total anesthesia time, operating room occupancy, and surgical start time may capture additional, distinct risk dimensions - most notably those related to circadian or team fatigue - that duration alone cannot represent. As an illustrative workflow requiring prospective validation, we propose operationalizing the 3-hour threshold across the perioperative timeline. Preoperatively, an anticipated duration exceeding 3 hours - predicted from features such as large size, high vascularity, or anatomically complex skull-base location - would designate a case as elevated-risk, prompting scheduling during optimal-performance hours, confirmation of correctly timed and weight-appropriate antibiotic prophylaxis, and multidisciplinary planning to rationalize the surgical approach and minimize unnecessary intraoperative exploration. Intraoperatively, as elapsed time approaches 3 hours, the threshold would serve as a real-time prompt to verify protocol-concordant antibiotic redosing and to reinforce normothermia, glycemic control, and meticulous closure. Postoperatively, cases exceeding the threshold would enter an intensified surveillance pathway, with structured wound assessment and monitoring of systemic inflammatory indices within the first 48 hours. This staged framework is offered as a pragmatic translation of our findings rather than a validated protocol (Figure 4).

Figure 4  Proposed perioperative workflow for the 304-hour surgery-time threshold (illustrative; pending prospective validation).

In this cohort, NLR was independently associated with SSI (OR = 1.081 per unit increase). Elevated NLR reflects subclinical inflammation or immunosuppression, which may impair early local immune responses to surgical contamination[14]. This is consistent with evidence from spinal surgery and general neurosurgical populations, where a systematic review supports NLR as a useful adjunctive biomarker (AUC ≈ 0.84)[18-20]. The present study extends this evidence specifically to patients with meningioma, a population in which NLR has not previously been examined in relation to SSI. The absence of a significant NLR-SSI association in certain glioma studies[21] may reflect differences in tumor biology: Malignant gliomas induce profound immunosuppression through multiple tumor-intrinsic mechanisms, which may render NLR less discriminating, whereas the predominantly benign nature of meningiomas preserves a more interpretable relationship between baseline inflammatory status and infectious susceptibility. A related contributor is chronic hyperglycemia, which impairs neutrophil function and fosters a pro-inflammatory microenvironment conducive to bacterial colonization[22], converging with the immune dysregulation that an elevated NLR reflects. Two large-scale database studies have shown that baseline frailty status is an independent predictor of adverse postoperative outcomes, including SSI, in patients with meningioma[4,6]. Host physiological reserve and surgical exposure thus represent two distinct, independently relevant risk domains. The frailty index captures a composite of comorbidity burden, functional dependence, and nutritional depletion - substrates that may also underpin elevated NLR and compromised wound immunity. Yet none of the frailty or nutritional indices examined - mFI-5, PNI, or GNRI - was an independent predictor in our multivariable model, contrasting with reports from large Western database cohorts. We interpret this divergence as mediation rather than absence of effect. Because frailty and nutritional depletion share biological substrates with the inflammatory dysregulation captured by NLR, their influence appears to be conveyed largely through this more proximal marker and through operative exposure, so that their independent contribution was attenuated once NLR and surgery time entered the model. The limited number of events and the parsimonious selection criterion further reduced the likelihood of retaining weaker independent effects such as that of mFI-5. The predominantly benign case mix and relatively preserved nutritional status of our cohort may narrow the discriminative range of indices such as PNI and GNRI. These observations qualify rather than contradict the established relevance of host physiological reserve: Frailty and nutritional status remain clinically meaningful but, in this cohort, operated largely through more proximal determinants. Although prior studies have suggested that an NLR in the range of 2-3 may warrant heightened vigilance[23], we did not derive a cohort-specific threshold, given the limited number of events and the more modest, less stable contribution of NLR in our model. A robust, validated cut-off for meningioma-related SSI awaits larger prospective studies. Even so, an elevated NLR may serve as a useful flag for baseline inflammatory or immunological compromise[24], and targeted preoperative optimization of nutritional and immunological status should be considered in such patients.

Tumor size did not reach statistical significance as an independent predictor, consistent with published data[25]. The most plausible explanation is mediation rather than collinearity or a true null effect. Larger tumors act as an upstream determinant that raises SSI risk chiefly by prolonging operative duration and increasing surgical complexity, along with related factors such as perilesional edema. Once surgery time - the more proximal mediator - enters the model, it absorbs much of the effect of tumor size, which accounts for the loss of independent significance. This is mirrored in the selection frequencies: Surgery time was retained in all imputed datasets, tumor size in only half. Tumor size thus retains clinical relevance as an early marker of anticipated risk but is insufficient for risk stratification on its own, and should be interpreted together with features such as extensive cerebral edema and anticipated surgical difficulty. Antibiotic prophylaxis was not a significant predictor, most plausibly because its use was near-universal in this cohort. Perioperative corticosteroid use likewise showed no significant association with SSI, in contrast to reports from broader neurosurgical populations. This discrepancy may reflect a context-specific role of corticosteroids in meningioma surgery[26], where anti-inflammatory benefits may counterbalance immunosuppressive effects[27]. Because corticosteroid use was recorded only as a binary variable, we could not assess dose- or duration-dependent effects, and this interpretation should be regarded as hypothesis-generating.

Skull-base location was strongly associated with SSI on univariable analysis but was not retained in the final model. As with tumor size, this most plausibly reflects mediation through operative duration rather than a true null effect, since skull-base meningiomas typically require longer, more complex resections. Surgery time had near-identical effects in the skull-base and non-skull-base subgroups, with no significant interaction (Supplementary Figure 5), indicating that skull-base location is not an effect modifier and adds no mechanism independent of operative duration. It remains a useful early marker for risk stratification, since it is known preoperatively and reliably anticipates prolonged.

The conditional logistic regression model incorporating surgery time, NLR, and tumor size demonstrated good discrimination (AUC = 0.828). Internal validation showed only minimal optimism: The optimism-corrected AUC (0.818) closely approximated the apparent value, with a concordant cross-validated AUC (0.800). The modest bootstrap calibration slope (0.883) is best interpreted in light of the limited number of events, which constrains the precision of calibration estimates rather than indicating substantial overfitting. These results support the model's use as a risk stratification tool while underscoring the need for external validation in larger cohorts. The supplementary nomogram serves as a visualization aid rather than the primary risk model, translating the core findings into a platform for individualized preoperative risk communication. Decision curve analysis confirmed net clinical benefit across a clinically relevant range of threshold probabilities. The matching variables it incorporates (age, sex, year) are included for structural consistency and are not themselves predictive of SSI. The matched design of the primary analysis precludes direct causal inference from standard logistic estimates, but the nomogram still offers a transparent, accessible instrument to support clinician-patient communication and perioperative planning[28].

Several limitations warrant consideration. The retrospective single-center design limits generalizability, as institutional protocols - surgical technique, perioperative care bundles, antibiotic prophylaxis, and local microbiology - may differ across settings. Our tertiary-referral cohort also represents a higher-complexity case mix than community hospitals, which may influence baseline infection risk and restrict extrapolation to other populations. Prospective, multicenter studies are warranted to provide more robust evidence and to better inform clinical decision-making. Matching on year of surgery, while controlling for secular changes in surgical and perioperative practice, reduces the variability of time-related exposures within matched sets. Risk factors whose prevalence shifted across the study period may therefore have their effects attenuated toward the null with reduced precision. The design is conservative for such time-varying factors. Post-discharge surveillance combined active telephone follow-up within the first 7 days with passive capture through readmission records thereafter. Later-onset, milder, or externally managed infections may therefore have gone unrecorded, which could underestimate the true SSI rate. This, together with the limited number of SSI events, may have constrained the ability to detect additional significant predictors. Several clinically relevant confounders - smoking status, preoperative glycemia, intraoperative temperature, and antibiotic timing - could not be ascertained in this retrospective registry. The absence of perioperative glycemic data is the most consequential, since hyperglycemia both impairs neutrophil function and correlates with elevated NLR; residual confounding may therefore have inflated the inflammatory contribution we attributed to this biomarker. The lack of intraoperative temperature records may likewise have led us to overestimate the surgery-time effect, as longer procedures are more prone to unintentional hypothermia, itself an independent predictor of SSI. Unmeasured smoking and antibiotic-stewardship practices were undocumented rather than selectively recorded; any resulting misclassification is likely non-differential and thus more apt to attenuate than to manufacture associations. These residual confounders are inherent limitations of retrospective design, and prospective studies with standardized capture of these physiological and process-of-care variables are needed to validate and refine our estimates. A final limitation is that our exposure was confined to operative duration measured from skin incision to wound closure. Total anesthesia time, operating room occupancy, and surgical start time were not captured in this retrospective dataset; thus risk related to circadian rhythm and accumulated team fatigue, for example, could not be assessed. Perioperative corticosteroid exposure was captured only as a binary variable; cumulative dose and treatment duration were not recorded, precluding assessment of any dose-response relationship with SSI.

CONCLUSION

This single-center retrospective study provides clinically relevant evidence on modifiable risk factors for SSI following elective craniotomy for meningioma resection. By defining surgery time precisely as the interval from skin incision to wound closure and analyzing a homogeneous tumor population, we found longer surgery time to be independently associated with SSI, with each additional hour corresponding to increase infection risk. We further identified 3.04 hours as the optimal clinical threshold, beyond which SSI risk increased approximately 5.7-fold. Each unit increase in preoperative NLR was associated with an 8.1% increment in SSI risk, pointing to a close relationship between systemic inflammatory status and infectious susceptibility. These findings suggest that for patients with an anticipated operative duration exceeding 3 hours, a formal risk-stratification framework should be established, operative workflows optimized to minimize exposure time, complex procedures scheduled outside periods of staff fatigue, and preoperative nutritional and inflammatory screening coupled with enhanced postoperative surveillance. Careful identification of modifiable risk factors, together with targeted prevention strategies, may help reduce the incidence of SSI following meningioma surgery.

ACKNOWLEDGEMENTS

The authors thank Dr. Zeng and Dr. Pan for their guidance and valuable advice throughout this study.