Mortality and morbidity in surgically treated patients with petroclival meningiomas: a systematic review and meta-analysis of case series

Abstract Background Reports on petroclival meningioma (PCM) surgical mortality and morbidity often deviate from established standards; as such, a comprehensive summary is lacking. Methods Eligibility/sources. Peer-reviewed case series of at least 10 PCM patients identified from PubMed, Web of Science, Ovid, or Google Scholar. Outcomes. Primary: mortality, tumor recurrence, any cranial nerve deficit (CND); other: individual CNDs, other complications. Data synthesis. Random-effects meta-analysis/meta-regression [effects: surgical approach (supratentorial, S; infratentorial, I; combined, (C), average age and follow-up, sample size, and percent of patients with gross-total resection (GTR)] of logit-transformed proportions. Results Data. 73 case-series/3553 patients. Mortality. Adjusted predicted mortalities of 2.4%, 2.5%, and 1.2% (50-month follow-up) for the S, I, and C approaches, respectively, with the upper limits of the 95% credibility intervals at 3.3%, 3.7%, and 3.6%, respectively. Recurrence. Adjusted predicted recurrences of 5.5%, 11.1%, and 12.0% (50-month follow-up and 57% GTR) for the S, I, and C approaches, respectively; recurrence was positively associated with follow-up period and negatively associated with having received GTR. At all covariates at median values but at GTR 90% predictions: 3.1% (95%CI 3.1–9.8), 6.3% (3.8–10.4), and 6.9% (3.4–13.2) with the S, I, and C; prediction credibility intervals 1–4% and 22.4%. Any CND. Adjusted predicted probabilities of 37.2%, 23.4%, and 29.5% (at median covariate values) for the S, I, and C approaches, respectively; prediction credibility intervals ranged from <10% to 78%. Other outcomes. The most common individual CNDs were nVII (14.4%), nV (11.5%), and nIII (10.2%); other common complications included motor deficit (10.8%), infection (9.8%), and CSF leak (7.5%). Conclusion This is the first systematic review on PCM surgical mortality, recurrence, and morbidity. Outcomes differ between surgical approaches and reporting quality varies greatly.


Introduction
Petroclival meningiomas (PCMs) account for approximately 2% of intracranial meningiomas and for less than 0.15% of primary intracranial tumors. 1,2 They present a neurosurgical challenge due to the deep anatomical location and proximity to vital neurovascular structures. This complexity is reflected in the abundance of viable surgical approaches and non-surgical treatment modalities for PCM, both of which are continuously evolving. Nevertheless, the mortalities and morbidities of these approaches are high. 2 A recent systematic review addressed the problem of postoperative cranial nerve deficits (CNDs) in PCM surgery. 3 However, a broader overview of the topic is missing and would be of value to guide the decision-making of both patients and neurosurgeons dealing with PCMs. Therefore, we conducted a comprehensive systematic review of surgery-related mortality, morbidity, and tumor recurrence in PCM patients.

Methods
We conducted a systematic review of a case-series of surgically treated PCM patients published until February 7, 2021, and performed a meta-analysis in an attempt to estimate the risk of post-surgical mortality, tumor recurrence, CNDs and other morbidity. We included a previously non-reported cohort of 11 PCM patients depicted as 'Mrak 2021' (all patients operated-on by one of the co-authors, GM).

Eligibility criteria
The inclusion criteria for relevant articles were as follows: (a) a case series of at least 10 patients with PCMs defined as 'true PCMs' or (simply) 'PCMs'. One study could report more than one cohort of patients (e.g. with different surgical approaches) or PCM patients could be a subset of a larger series including other tumor locationsas long as data on different procedures/PCM patients were reported separately with the same scrutiny regarding the outcomes and comprised at least 10 patients, each cohort was treated as a separate item; (b) a full-text article published in the English or German language; (c) numerical data on at least one of the outcomes of interest -(i) mortality; (ii) tumor recurrence; (iii) CNDs reported in respect to 'any' cranial nerve (i.e. subjects with at least one cranial nerve affected) or specifically in respect to individual cranial nerves nI to nVIII and 'lower cranial nerves', i.e. nIX-nXII cumulatively; (iv) other morbidity, more specifically, speech impairment, need for tracheostomy, gait disturbance, motor deficit, cerebrospinal fluid (CSF) leak, infection, hydrocephalus, seizures, stroke/intracranial hematoma, deep venous thrombosis and pulmonary embolism.

Literature search and study selection process
We searched PubMed Medline, Web of Science, Ovid (all resources), and Google Scholar using the search term 'petroclival meningioma' with no filters (the exception was Google Scholar, for which it was requested that the search term appeared in the title) to achieve a non-specific, yet highly sensitive search. We also manually searched the reference lists of published reviews for any relevant research that may not have been identified through our literature search. Following duplicate removal, the identified articles were independently screened (titles, abstracts) by four reviewers to exclude non-eligible studies; the final lists were compared and any disagreements were resolved. This process was repeated to complete the eligibility assessment of fulltext reviews.

Data extraction
In the first step, four reviewers each extracted data from a subset of included studies. In the next step, each subset was assigned to a different reviewer for re-extraction; this was done to ensure the reliability of the process. Any disagreements were resolved by reaching a consensus. The following data were extracted: (a) number of patients (denominator); (b) number with outcomes throughout the observation period; (c) predominant surgical approach used in the cohort (supratentorial, infratentorial, combined supra-/infratentorial); (d) other potentially relevant covariates, i.e. (i) average patient age, (ii) proportion of women, (iii) tumor histology, (iv) average tumor volume (if not reported, tumor size data were used to calculate tumor volume, if three diameters were reported, an ellipsoid shape was assumed and the volume was calculated accordingly, if two diameters were reported, the third was assumed to be an average of the first two, if only one diameter was reported the tumor was assumed to be in the shape of a sphere, with volume calculated accordingly), and (v) proportion of patients for which gross total resection (GTR) was achieved. A few studies reported average extent of resection instead. The two values were considered interchangeable when proportion with GTR was addressed as a moderator.
In cases in which PCM patients were a subset of a larger cohort 4-10 (other tumor locations were also reported) and covariate data were not specifically reported for the PCM patient subset, averages for the entire cohort were considered applicable. In the case that data on mortality, tumor recurrence, and morbidity other than CND were not explicitly stated, including a lack of an explicit statement about 'no events', it was considered that the study did not report these outcomes.
In the case that data concerning an individual CND were reported, but deficits of other cranial nerves were not mentioned (including a lack of an explicit statement of 'no other cranial nerve deficits'), it was assumed that no deficit regarding the nonaddressed cranial nerves had been observed. If CNDs were not mentioned at all, it was considered that the study did not report these outcomes.
Continuous variables reported as medians and ranges were converted into mean values, along with the corresponding standard deviations, to achieve uniform summaries across studies. 11

Study quality
We addressed several aspects suggested in a recently proposed tool for evaluating the quality of case-series/case reports. However, we later restricted the evaluation to the question of non-reporting and provided an overall comment about variability and inconsistency (across included studies) in the choice of outcomes and modes of their reporting (see Supplemental Methods for explanation). 12,13 Data synthesis We considered mortality, tumor recurrence, and occurrence of any CND (number of patients with at least one CND) as outcomes of primary interest. The occurrence of individual CNDs and non-CND complications were considered secondary outcomes, in part due to the anticipated problems with (in) adequate reporting.
For primary outcomes, we first generated random-effects pooled estimates (with 95% prediction intervals) separately for each surgical approach/technique by fitting generalized linear mixed (GLMM) models with a random intercept to logit-transformed proportions (binomial-normal model) with ad hoc Hartung-Knapp variance correction. Individual study proportions are presented as Clopper-Person confidence intervals (a continuity correction in studies with no events only to generate single study estimates, but not implemented in data pooling). [14][15][16] Funnel plots were generated, after which Peters' asymmetry test 17 was implemented, as suggested for single proportions 18 . Next, we implemented the Copas selection model analysis 19 to generate publication bias-corrected estimates 20 , after which we performed random (mixed)-effects meta-regression (GLMM, maximum likelihood for s 2 , t-distribution) to obtain predicted probabilities (with 95% CI) and 95% credibility (prediction) intervals for each surgical approach adjusted for average patient age, cohort size, average length of follow-up (in months) and percent of patients with GTR. For exploratory purposes, the meta-regression estimates were used to evaluate differences between surgical approaches (as odds ratios). Regarding secondary outcomes, pooled and meta-regression estimates were generated. We used the R packages meta 21 , metasens 22 and metafor 23 during the data synthesis.

Study selection and characteristics
After duplicate removal and title/abstract screening, 121 out of the 1663 initially identified articles underwent full-text eligibility assessment ( Figure 1), with 52 studies being excluded for various reasons (Figure 1, see Supplemental Results, Table S2 for details) and 69 1,4-10,24-84 publications (72 case-series) included in the review ( Figure 1). Together with the current case-series, this amounted to 73 patient series ( Figure 1): in 71 series patients had received open surgeries (20 supratentorial approach, 39 infratentorial approach, and 12 combined approach), while two series reported on endoscopically treated patients ( Figure 1). The outcomes from the two endoscopic series (total N ¼ 61) were summarized, but the series did not involve enough data to be included in any other analysis.
Several study characteristics (e.g. tumor histology) were reported only sporadically, which prevented any systematic overview, while others were reported more regularly and are summarized by approach: predominantly supratentorial (Table  1); infratentorial (Table 2); or combined supra-/infratentorial or endoscopic (Table 3). Data concerning patient age, length of follow-up or tumor size were occasionally missing and the reporting mode greatly varied and appeared uninformative at instances (Tables 1-3). The proportion of patients with GTR was missing from four studies (Tables 1 and 3), while two studies reported the mean extent of resection instead (Table 3). ). There was no residual heterogeneity; hence, the CIs for adjusted predicted probabilities were identical to the credibility intervals which indicated mortalities of 2.4%, 2.5%, and 1.5% across the three subsets of case-series at median covariate values, respectively ( Figure  2(C)). None of the covariates was associated with the outcome.   GTR: gross total resection; n/a: data not available. a Numbers are mean/or mean ± SD/or median (range)/or range/or mean (95% CI). b Numbers refer to diameters (in cm) or to tumor volume (in cm 3 ) and are mean/or mean ± SD/or median (range). GTR: gross total resection; n/a: data not available. a Numbers are mean / or mean ± SD/or median (range)/or range /or mean (95% CI). b Numbers refer to diameters (in cm) or to tumor volume (in cm 3 ) and are mean/or mean ± SD/or median (range). Higher proportion of patients with GTR was associated with lower probability of recurrence, while the length of the follow-up was associated with a higher probability of recurrence (Supplemental Figure S7). Other covariates were not associated with the outcome. Odds of recurrence for the infratentorial and combined series were twice as high as what was calculated for the supratentorial series (Supplemental Figure S8). Residual heterogeneity was high; hence, the credibility intervals were wide ( Figure 3C).

Any cranial nerve deficit
Any CND was reported in 20/20 of the supratentorial series (N ¼ 964), 37/39 of the infratentorial series (N ¼ 1953), and 12/12 of the combined approach series (N ¼ 544) (Figure 4(A), Supplemental Figures S9-S11 for forest and funnel plots). The point estimates indicated probabilities of 38%, 23%, and 31%, respectively, but included wide CIs and 95% prediction intervals that extended from 10% to 76% (Figure 4(A)). The biasadjusted estimates were somewhat higher (31-38%) and equally as imprecise (Figure 4(B)). The meta-regression included 58 series (N ¼ 3032, 881 events) (Figure 4(C)). At median values of covariates, the predicted point-estimates were 37% for supratentorial series, 23% for infratentorial series, and 30% for the combined approach series; however, the confidence intervals were wide and credibility intervals extended from <10% to 78% (Figure 4(C)). No covariate appeared to be associated with the outcome. The odds of any CND appeared lower in the infratentorial series than in the supratentorial series (Supplemental Figure S8).

Individual cranial nerve deficits
Meta-analyses of the proportions of patients with individual CNDs revealed considerable, and understandable, variability across individual nerve-by surgical approach subsets in terms of observed proportions and proportion of cohorts with no events, and by considerable heterogeneity within each nerve-by approach subset (Supplemental Table S3). Moreover, each of the metaregression models (one for each individual cranial nerve) demonstrated a high level of heterogeneity (Supplemental Table S4). The pattern of adjusted predicted probabilities of individual CNDs in the supratentorial series (nIII most commonly affected; declining probabilities with higher numbered nerves) ( Figure 5) was almost a mirror image of the pattern observed for the infratentorial series (nVII most commonly affected; low probabilities of nIII-nVI deficits) ( Figure 5), but most estimates were relatively imprecise (wide CIs) ( Figure 5).

Other morbidity
The reporting of other morbidities included such a high degree of variability that no meaningful summaries could be drawn. Table 3. Summary of series in which the majority of patients were operated on using a combined supra/infrantentorial approach, or an endoscopic approach (only two case series GTR: gross total resection; EOR: extent of resection; n/a: data not available. a Numbers are mean/or median (range). b Numbers refer to diameters (in cm) or to tumor volume (in cm 3 ) and are mean/or mean ± SD/or median (range).

Discussion
To the best of our knowledge, only one previous systematic review has addressed PCM surgery-related outcomes; however, the review primarily focused on CNDs and included 12 studies with a total of 334 PCM patients. 3 The present review embraced 73 case-series and a total of 3553 surgically treated PCM patients. Therefore, it could be justifiably viewed as the first comprehensive overview of surgery-related morbidity, mortality, and tumor recurrence in this patient population. Until the microsurgical era pioneered by Yasargil, 85 PCM surgical morbidity exceeded 50%, a proportion which rendered these patients inoperable. 2 When considering the historical perspective, the fatal natural history of the disease, its complexity, and the overall mortality and morbidity from the present analysis, it is clear that global progress in neurosurgical practice has been made. In other words, a disease that was considered fatal only 40 years ago is today controllable in most patients.
The current pooled estimates of mortality ('simple' and metaregression-based at the 50-month follow-up) ( Figure 2) were rather precise, low (in the 1.1% to 2.5% range), comparable across different approaches (with only slight differences), and apparently not affected by study-level covariates. Incidence of tumor recurrence, on the other hand, increased as the follow-up period extended and was lower in patients who had received gross total resection (GTR) relative to other patients. This finding is in line with the established principles of meningioma treatment strategysurgery is the gold standard of therapy (particularly considering the natural history of PCMs) and aimed at a high extent of tumor resection. 2,[86][87][88] The adjusted point-estimates (representing an average age of 50 years, case-series with 33 subjects, a 50-month follow-up, and [median] %GTR of 57%) of 5.5%, 11.1%, and 12.0% tumor recurrence for the supratentorial, infratentorial, and combined approach series, respectively, suggest that the latter two approaches demonstrate odds of tumor recurrence which are double than that associated with the supratentorial approach. This dynamic may be explained by tumor characteristics (better accessibility) that could not have been adequately adjusted for in the meta-regression analysis.
The suprantentorial, infratentorial, and combined approaches demonstrated 37%, 23%, and 30% incidences, respectively, of any (novel) CNDs (both 'simple' and meta-regression-based). The incidence of novel CNDs was not found to be affected by any of the study-level covariates. A similar estimate (34% -a raw proportion, not a pooled estimate) was reported 10 years ago based on 19 case-series with a total of 1000 patients. 88 The lower odds of CND incidence associated with the infratentorial approach in the present review may be explained by a similar dynamic as was noted for the lower recurrence rates among patients who had received GTRa more aggressive resection might also result in higher morbidity. This notion was recently emphasized in a series of posterior fossa meningiomas, more specifically, a higher resection rate appeared to be associated with new-onset CNDs. 89 Among individual cranial nerves (CN), CNVII was most Shown are adjusted predicted probabilities of events (95%CI) with 95% credibility intervals by surgical approach and residual heterogeneity indicators (I 2 , s 2 ). Adjustments are for average age and length of follow-up, series size and proportion of patients in whom gross total resection (GTR) was achieved. Estimates are given at median values of each covariate and also at median age, follow-up, sample size but at 90% GTR. k: number of included case series; N: total number of included patients; Events: number of subjects who died; No event k: series with no events.
commonly affected in the infratentorial or combined approaches, and CNIII was most commonly affected in the supratentorial approach. This can be expected when considering the anatomical corridor and its neural content; for example, the descending ladder-like structure of CN morbidity in the supratentorial approach is roughly mirrored by the ascending structure of the remaining two approaches ( Figure 5). Similar findings were reported in the only other published meta-analysis of PMC surgery-related CNDs 3 ; notably, CNVII deficits were the most common adverse effect, and the incidence (13.9%) was comparable with what was estimated for the combined approach (14.0%) in the current review, and higher than the present estimates for the supra-(5.3%) and infratentorial (7.7%) series.

Limitations
Al Mefty first introduced the definition of 'true' PCMs, which has been adopted throughout the literature. 25 However, other authors have challenged the definition and many fail to adhere to the current standards. 3 To overcome this issue, we included all PCM cohorts regardless of whether they adopted any diagnostic criteria. Study quality was not assessed due to reasons discussed in the methods section, but substantial heterogeneity and variability in the quantitative and qualitative reporting of results was obvious across the identified studies. This leads to the exclusion of some covariates (e.g. tumor volume) or outcomes (e.g. morbidities other than CNDs) from the quantitative synthesis. Also, missing or alternatively reported variables were handled based on assumptions (see methods), which might have introduced bias.
Radiosurgery as an adjunctive treatment was not considered in the analysis of recurrence, as the variable was reported sporadically across the identified studies. This important limitation has to be considered when interpreting the results.
The cohorts were grouped according to the prevalent approach (supra-or infratentorial, or combined) for pragmatic reasons, yet caution should be exercised when interpreting the analyses, especially the exploratory results that compare approaches. First, in individual studies the approach was conflated into a single (predominant) approach, even though the majority of studies reported diverse approaches. Second, the choice of approach is primarily a function of the anatomical span of a tumor rather than the surgeon's preference. Therefore, differences in the outcomes across approaches should only be considered as descriptive, and not as evidence of the superiority/ inferiority of one approach over another. This holds true for the majority of individual cases encountered in clinical practice. Nevertheless, the present data might inform decision-making in ambiguous cases.

Implications for practice and future research
This is the first comprehensive report on surgery-related mortality, morbidity, and tumor recurrence among PCM patients. The presented research is relevant because the results may inform patients about surgery-related risks and guide surgeons in tailoring their approach to the tumor characteristics. The presented  Figures S4, S5 and S6 show forest plots and funnel plots). Shown are random-effects estimates (95%CI) and 95% prediction intervals (PI), heterogeneity indicators (inconsistency index, I 2 and variance across case series, s 2 ) and results of the Peters test of asymmetry of the funnel plots (publication bias). (B) Summary of the Copas selection model analysis by surgical approach. Shown are p-values (critical alpha ¼ 0.1) from a test of the null-hypothesis that no selection remains unexplained, approximate number of unpublished studies suggested by the model and (bias)-adjusted estimates. (C) Summary of randomeffects meta-regression analysis. Shown are adjusted predicted probabilities of events (95%CI) with 95% credibility intervals by surgical approach and residual heterogeneity indicators (I 2 , s 2 ). Adjustments are for average age and length of follow-up, series size and proportion of patients in whom gross total resection (GTR) was achieved. Estimates are given at median values of each covariate and also at median age, follow-up, sample size but at 90% GTR. k: number of included case series; N: total number of included patients; Events: number of subjects who died; No event k: series with no events.  Figures S8, S9 and S10 show forest plots and funnel plots by approach). Shown are random-effects estimates (95%CI) and 95% prediction intervals (PI), heterogeneity indicators (inconsistency index, I 2 and variance across case series, s 2 ) and results of the Peters test of asymmetry of the funnel plots (publication bias). (B) Summary of the Copas selection model analysis by surgical approach. Shown are p-values (critical alpha ¼ 0.1) from a test of the null-hypothesis that no selection remains unexplained, approximate number of unpublished studies suggested by the model and (bias)-adjusted estimates. (C) Summary of random-effects meta-regression analysis. Shown are adjusted predicted probabilities of events (95%CI) with 95% credibility intervals by surgical approach and residual heterogeneity indicators (I 2 , s 2 ). Adjustments are for average age and length of follow-up, series size and proportion of patients in whom gross total resection (GTR) was achieved. Estimates are given at median values of each covariate and also at median age, follow-up, sample size but at 90% GTR. k: number of included case series; N: total number of included patients; Events: number of subjects who died; No event k: series with no events. Figure 5. Adjusted predicted probabilities (as percentages, 95%CI) of individual cranial nerve deficits (cumulative data for nIX-XII, i.e. lower cranial nerves) from metaregression models (Supplemental Table S4) by predominant surgical approach. Point-estimates are depicted numerically.
findings also indicate that there is a strong need for establishing reporting standards for case-series of PCMs.