Sex-specific impact of patterns of imageable tumor growth on survival of primary glioblastoma patients

Background Sex is recognized as a significant determinant of outcome among glioblastoma patients, but the relative prognostic importance of glioblastoma features has not been thoroughly explored for sex differences. Methods Combining multi-modal MR images, biomathematical models, and patient clinical information, this investigation assesses which pretreatment variables have a sex-specific impact on the survival of glioblastoma patients. Pretreatment MR images of 494 glioblastoma patients (299 males and 195 females) were segmented to quantify tumor volumes. Cox proportional hazard (CPH) models and Student’s t-tests were used to assess which variables were associated with survival outcomes. Results Among males, tumor (T1Gd) radius was a predictor of overall survival (HR=1.027, p=0.044). Among females, higher tumor cell net invasion rate was a significant detriment to overall survival (HR=1.011, p<0.001). Female extreme survivors had significantly smaller tumors (T1Gd) (p=0.010 t-test), but tumor size was not correlated with female overall survival (p=0.955 CPH). Both male and female extreme survivors had significantly lower tumor cell net proliferation rates than other patients (M p=0.004, F p=0.001, t-test). Additionally, extent of resection, tumor laterality, and IDH1 mutation status were also found to have sex-specific effects on overall survival. Conclusion Despite similar distributions of the MR imaging parameters between males and females, there was a sex-specific difference in how these parameters related to outcomes, which emphasizes the importance of considering sex as a biological factor when determining patient prognosis and treatment approach.


Introduction
Glioblastoma (GBM) is the most common primary malignant brain tumor, with a median overall survival of 9 to 15 months, depending on the given course of treatment (1) (2) (3) . According to Ostrom et al. (4) , only 35% of patients survive more than one year and 4.7% of patients survive more than five years after diagnosis. Factors such as age at diagnosis, Karnofsky performance score (KPS), extent of surgical resection, and tumor location have been found to play a significant role in determining the duration of patient survival (5) (6) (7) , but there is still limited insight into which underlying biological features contribute to a patient becoming a "survival outlier." To date, there is minimal research on the utility of using pretreatment (pretx), imagebased volumetric and kinetic variables to identify potential extreme and shortterm survivors. Additionally, while it has been consistently identified that GBM incidence is higher among males (8) (9) (10) (11) (12) and females GBM patients have better outcomes (8) (13) (14) (12) , little to no research has focused on sexspecific predictors of extreme and shortterm survival. The ability to pinpoint relevant predictors of the duration of overall survival has clinical value and identifies areas for future research. By using variables derived from patient clinical information and routinelyobtained, noninvasive MR images, we can establish predictors of survival duration that can be readily assessed in a pretx setting. Knowing whether these factors affect males and females in the same way will guide research efforts towards bestpractice, individualized patient care.
The purpose of this study was to determine whether there are sexspecific predictors of survival outcomes among glioblastoma patients. Using patient data from our multiinstitutional brain tumor repository, we tested the significance of eight pretx volumetric, kinetic, and clinical variables in predicting extreme and shortterm survival. We also tested whether these variables and additional categorical variables, including tumor laterality, extent of resection (EOR), isocitrate dehydrogenase 1 (IDH1) mutation status, and O(6)methylguanineDNA methyltransferase promoter (MGMT) methylation status, significantly impacted the overall survival of male and female patients. Throughout the analysis, males and females were tested separately as distinct population groups and their results were compared, allowing us to identify sexspecific impactors of survival outcome among GBM patients.

Biomathematical Models and PatientSpecific Tumor Kinetics
An extensive literature has been generated over the last two decades applying a biomathematical model to simulate patientspecific glioblastoma growth (16) (17) (18) (15) . The primary model is referred to as the ProliferationInvasion (PI) model and is based on two key parameters: the net rate of proliferation, ϱ, and the net rate of invasion, D. By assuming that the T1Gd abnormality coincides with one threshold of tumor cell density relative to the maximum carrying capacity, and the T2 or T2FLAIR (T2/FLAIR) abnormality coincides with another, lower threshold, one can obtain patientspecific estimates of these two parameters using standard pretx imaging (19) . This calibration uses two sets of MRI data: one set consists of two pretx time points of either T1Gd or T2/FLAIR images and the second set consists of one time point with both T1Gd and T2/FLAIR images. The first set is used to calculate a velocity and the second set is used to estimate the degree of invasion. These estimates have been shown to be prognostic of benefit from resection (18) , survival (16) , and radiation efficacy (20) and can be used to examine therapeutic response (21) (22) . However, two pretx time points of imaging are not always available, necessitating the use of a second mathematical model. This mathematical model incorporates hypoxia, necrosis, and the angiogenic cascade and is referred to as the ProliferationInvasionHypoxicNecroticAngiogenesis (PIHNA) model (23) . This model similarly relies on an analogous net rate of invasion and net rate of proliferation. In this calibration approach, the relative size of necrosis is used as a surrogate to the velocity estimate from the two time points of imaging .  For this calibration method, a lookup table was created, with each entry being an output from a unique D and ϱ  PIHNA simulation. The lookup table contains the estimated PI D/ϱ value, the size of necrosis, the T1Gd visible  portion of the tumor (assumed at >80% cell density) and the T2/FLAIR visible portion (assumed at >16% cell  density), throughout each simulation. Given a patient's T1Gd volume, T2/FLAIR volume, necrosis size, and the  PI estimated D/ϱ at one time point, the lookup table points to a subtable of simulation points that match the T2/FLAIR size on MRI within a small measurement tolerance. A D and ϱ is then chosen based on the simulations that match the PI D/ϱ and necrosis size within a small tolerance. If multiple D ϱ pairs exist that satisfy the aforementioned criteria, the pair that minimizes the T1Gd size is chosen. As a number of the patients in our cohort were not eligible for the traditional calibration method, we elected to use this method for all patients. Thus, in this paper, when the parameter D or ϱ is discussed, it is the D and ϱ corresponding to the PIHNA calibration.

Patient Population
Our research lab has amassed a large multiinstitutional repository over the last 15 years under institutional review board approvals. The repository consists of the clinical patient data (curated from medical records) and serial, multimodal MR images of over 1400 glioblastoma patients. From this repository, we identified all newlydiagnosed glioblastoma patients with necessary clinical information (sex, age, and overall survival) and a calculated pretx (prior to biopsy or resection) tumor volume from a T1Gd MRI. This cohort was comprised of 494 primary GBM patients (299 males and 195 females). All patients in this cohort were adults at time of diagnosis, with the exception of three adolescent patients (2 male and 1 female). Since the calculation of PIHNA D, PIHNA ϱ, and PI D/ϱ requires both T1Gd and T2/FLAIR images, a subcohort of patients with sufficient imaging was created from the main cohort in order to study the effect of these variables on survival (223 males and 141 females).
We defined extreme survivors (EXS) as those with overall survival (OS) of 5 years (1825 days) or longer. EXS typically make up less than 5% of glioblastoma patients (4) . However, due to the data collection efforts of a multicenter collaboration researching extreme survival among GBM patients (ENDURES), about 9.5% of patients in this cohort were EXS. EXS were compared to NonEXS, which consisted of all patients who survived less than 5 years. For further comparison, we identified patients from our cohort that were shortterm survivors (STS), which we defined as patients with a confirmed death and an overall survival shorter than 7 months (210 days) (24) . STS were compared to NonSTS, which consisted of all patients who survived longer than 7 months.
The breakdown of the main cohort and the subcohort by sex and survival group is shown in table 1 .   Table 2 outlines the eight quantitative volumetric, kinetic, and clinical variables that were explored in our investigation. Student's ttests with Welch's corrections were used to test whether there were significant differences in the eight quantitative variables between the following survival groups: EXS vs NonEXS, EXS vs STS, and STS vs NonSTS. CoxProportional Hazards models (CPH) were used to assess which of the quantitative variables were significant predictors of OS. Parameters that were significant or almost significant (p<0.10) in univariate analysis were compared in multivariate analysis. KaplanMeier survival analysis (logrank tests) and CPH models were used to assess the impact of the categorical variables on survival. The following categorical variables were included: IDH1 mutation status, MGMT methylation status, tumor laterality, and EOR. Ttests and KaplanMeier survival curves were generated using Prism (25) and the CPH models were generated using R studio (26) . All statistical analyses were performed separately for the male and female populations. There was no significant difference in the distribution or mean values of these variables between males and females (Supplement 11) .

Decision Trees
The decision trees (DT) in this study were created using R (26) , accompanied by a package called rpart (28) , which allows effective decision tree pruning. Six DT were produced in total, grouped into 3 pairs. The first pair of DT sorted EXS and NonEXS, the second pair sorted EXS and STS, and the third pair sorted STS and NonSTS. Within each pair, one tree was created using the male population and the other was created using the female population. All six trees were constructed using the eight quantitative pretx variables: age, T1Gd radius, necrosis radius, CE thickness, T2/FLAIR radius, PIHNA D, PIHNA ϱ, and PI D/ϱ. The PI and PIHNA subcohort of patients (223 males and 141 females) were used to create the testing and training groups. 70% of the each population was placed in the training set and 30% in the testing set and 10fold cross validation was used to ensure the generalizability of the results. For each tree, accuracy is reported for the training group, testing group, and the full cohort (training + testing). For the EXS vs NonEXS and STS vs NonSTS trees, EXS and STS were considered the condition positive groups, respectively, and the sensitivity was reported for the testing group, training group, and full cohort.

Variables associated with extreme and shortterm survival
Student's ttests were performed separately on males and females and compared the following groups: EXS vs NonEXS, EXS vs STS, and STS vs NonSTS. The results of this analysis can be found in Table 3.
When compared to the rest of the male population, EXS were significantly younger (p=0.005) and STS were significantly older (p<0.001). Male EXS, with a mean age of 51.33 years, were significantly younger than male STS, who had a mean age of 65.33 years (p<0.001). Additionally, male STS had significantly larger T1Gd radii compared to male EXS (p=0.041) and male NonSTS (p=0.011). However, there was no significant difference in T1Gd radii between male EXS and male NonEXS (p=0.485). Male STS also had significantly larger CE thickness when compared to male NonSTS (p=0.031). Male STS had significantly larger D than male NonSTS (p=0.017), with no difference between the D of male EXS and the D of male NonEXS (p=0.992). For ϱ, male EXS had significantly smaller ϱ when compared to male NonEXS (p=0.004) and when compared to male STS (p=0.047).
When compared to the rest of the female population, female EXS were significantly younger (p=0.032) while female STS were significantly older (p<0.001). The mean age for female EXS was 48.29 years, which was significantly younger than the mean age for female STS, 65.26 years (p=0.003). Female EXS had significantly smaller T1Gd radii compared to female NonEXS (p=0.010) and female STS (p=0.010). Female STS did not have significantly different T1Gd radii compared to female NonSTS (p=0.307). Compared to the rest of the female population, female EXS had significantly smaller D (p=0.008) and female STS had significantly larger D (p=0.018). Female EXS had significantly smaller ϱ compared to female NonEXS (p=0.001) and female STS (p=0.027). There was no significant difference in ϱ between female STS and female NonSTS (p=0.535).

Decision Trees
The six decision trees used in this analysis were constructed using the following pretx variables: age, T1Gd radius, necrosis radius, CE thickness, T2/FLAIR radius, PIHNA D, PIHNA ϱ, and PI D/ϱ.
In the female EXS vs NonEXS DT, the nodes that predicted EXS with 100% sensitivity included T1Gd radius < 21.93 mm, necrosis radius < 8.68 mm, and age < 28.5 years. For males, the best predictors of EXS included CE thickness < 11.33 mm, PI D/ϱ ≥ 0.3687 mm 2 , age < 72 years, and age < 59.5 years. Notably, all male EXS had CE thickness shorter than 11.33 mm, PI D/ϱ above 0.3687 mm 2 , and age below 72 years. Figure 1 shows the female and male pruned DT that sort patients into EXS and NonEXS survival categories. In the female EXS vs STS DT, the nodes that best predicted female EXS were ϱ < 10.33 year 1 , CE thickness < 4.746 mm, and age < 47.5 years and the node that best predicted female STS was age ≥ 47.5 years. In the male DT, the node that best predicted EXS was ϱ < 118.2 year 1 and the node that best predicted STS was D ≥ 11.85 mm 2 /year. Figure 2 shows the female and male DT that sort patients into EXS and STS. The third pair of DT sorted males and females into STS and NonSTS groups. Among females, the nodes that best predicted STS included age ≥ 49.5 years, T2/FLAIR radius ≥ 23.76 mm, and D ≥ 41.23 mm 2 /year. In the male DT, the nodes that most accurately predicted STS were age ≥ 47.5 years, age ≥ 79.5 years, ϱ ≥ 10.33 year 1 , and CE thickness between 11.25 mm and 12.36 mm. Figure 3 shows the female and male STS vs NonSTS DT.

Variables associated with overall survival
Univariate and multivariate CPH analyses were utilized to determine which variables significantly influenced the overall survival of GBM patients. Variables that were significant or almost significant (p<0.10) in univariate analysis were analyzed in multivariate analysis.
Factors with significant prognostic value in the female univariate CPH (Table 4B) ) and females (B). Factors that were almost significant (p<0.10) or significant in univariate analysis were included in the multivariate analysis.

IDH1 Mutation
Since mutations of the isocitrate dehydrogenase 1 (IDH1) gene have been previously identified as significant predictors of longterm survival (14) , we analyzed the impact of sex and IDH1 status on the overall survival of our patient cohort. 120 patients in the main cohort had determined IDH1 status, consisting of 69 wildtype (wt) and 8 mutant (mut) male patients and 39 wt and 4 mut female patients. When looking at the entire population (both males and females), there was a trend towards IDH1 mutant patients having significantly better survival (logrank, p=0.071). Among females, IDH1 mut survived significantly longer than IDH1 wt patients (logrank, p=0.008), but among males, the survival difference was not significant (logrank, p=0.924) (Supplement 1) . All 4 IDH1 mut females survived at least three years, making them all longterm survivors (29) .
We also assessed whether IDH1 mut patients had the same features as the extreme survivors in this analysis (younger age, lower PIHNA D, lower PIHNA ϱ, and smaller T1Gd radii). Unlike the female EXS, IDH1 mut females did not have lower PIHNA D (ttest, p=0.402) or smaller T1Gd radii (p=0.584) compared to their wt counterparts, but they did have significantly lower PIHNA ϱ when compared to wt females (p=0.027). Males did not show significantly different PIHNA D (p=0.796) or PIHNA ϱ (p=0.461) between the two IDH1 status groups, but IDH1 mut males did tend to have smaller T1Gd radii (p=0.052) when compared IDH1 wt males. Both male and female IDH1 mut were significantly younger than their wt counterparts (Male p=0.024, Female p=0.007).

MGMT Methylation
Methylation of the O(6)methylguanineDNA methyltransferase (MGMT) promoter has been found to be significantly more common in longterm survivors (30) , so we also assessed the impact of MGMT methylation on the survival of our population cohort. Ninety patients from the main cohort had available MGMT methylation status, which comprised of 32 females (12 methylated and 20 unmethylated) and 58 males (18 methylated and 40 unmethylated). Methylated patients had significantly better survival than unmethylated patients among males (logrank, p=0.013), females (p=0.007), and the entire population (males and females) (p<0.001) (Supplement 4) . Multivariate CPH analyses were performed to assess the impact of MGMT status on survival, while accounting for age. These analyses showed that MGMT status significantly impacted survival for males (p=0.004) and females (p=0.037). Among EXS with available MGMT methylation status (n=15), 50% (n=5) of males and 60% (n=3) of females had MGMT methylation, while among NonEXS (n=75), 29% (n=14) of males and 33% (n=9) of females had MGMT methylation, suggesting that MGMT methylation was more common among both male and female EXS.
When we tested to see if MGMT methylated patients shared the features of extreme survivors (younger age, lower PIHNA D, lower PIHNA ϱ, and smaller T1Gd radii), we found that MGMT methylated females had significantly lower ϱ (ttest, p=0.026) and tended to have lower D (p=0.057) when compared to MGMT unmethylated females. There was no significant difference in the values of D (p=0.477) or ϱ (p=0.869) between MGMT methylated and unmethylated males. For both males and females, there was no significant difference in age (Male p=0.724, Female p=0.735) or T1Gd radii (Male p=0.397, Female p=0.241) between methylated and unmethylated patients.

Laterality
Using pretx T1Gd MR images, we determined the laterality of each patient's tumor, classifying the tumors as being located in the right hemisphere, left hemisphere, or both hemispheres (bilateral). The impact of tumor laterality on survival was assessed separately for males and females, and the results were compared. Among males, there were 129 left hemisphere GBMs, 154 right hemisphere GBMs, and 11 bilateral GBMs, and among females there were 86 left hemisphere GBMs, 96 right hemisphere GBMs, and 9 bilateral GBMs. Laterality could not be determined for 5 male and 4 female patients.
Males with GBMs in the left hemisphere had better survival outcomes than males with GBMs in the right hemisphere. Male patients with tumors on the left side trended towards significantly better survival than males with tumors on the right side (logrank, p=0.077) and had significantly better survival than males with bilateral tumors (p=0.010). In a multivariate CPH analysis that also accounted for extent of resection, tumor location in the left hemisphere was found to be a significant independent predictor of improved survival outcome for males (p=0.017) (Supplement 14) . There were more EXS than STS among males with tumors on the left side and there were almost twice as many STS as EXS among males with tumors on the right side. Laterality did not have a significant impact on survival for female patients. There was no significant difference in survival between females with left and right hemisphere tumors (logrank, p=0.218), and females with bilaterally located tumors did not have significantly worse survival when compared to females with nonbilateral tumors (bilateral vs left p=0.272, bilateral vs right p=0.471) (Supplement 6) . In CPH analysis, laterality was not a significant predictor of female survival (p=0.299) (Supplement 14) .

Extent of Resection
Our investigation evaluated whether the extent of initial surgical intervention, a known prognostic factor among GBM patients, had the same prognostic value for both male and female GBM patients. Patient extent of resection (EOR) status, categorized as gross total resection (GTR), subtotal resection (STR), or biopsy, was obtained from the patient records. From the main cohort of 494 patients, 211 males (83 GTR, 83 STR, and 45 biopsy) and 136 females (54 GTR, 55 STR, and 27 biopsy) had available EOR status.
EOR had a significant impact on the survival of male GBM patients. In univariate CPH analysis, EOR was a significant independent predictor of overall survival in males (p=0.002), with GTR being associated with the best survival outcomes ( Supplement 14) . In a KaplanMeier survival curve comparison, GTR males had significantly better survival than STR males (logrank, p=0.033) and males who received some surgical resection (GTR or STR) had significantly better survival than males who only received a biopsy (p=0.013) (Supplement 8) . CochranArmitage Trend Test showed that there was significant trend towards male EXS receiving more extensive resections and male STS receiving less extensive resections or biopsies (p=0.027). Female who received resection (GTR or STR) trended towards improved survival compared to biopsy females (logrank, p=0.077) (Supplement 8) , but there was no significant difference in survival between GTR females and STR females (p=0.992) (Supplement 9) . Additionally, EOR did not significantly impact female survival in univariate CPH analysis (p=0.180) (Supplement 14) . Trend test showed that there was a notable but insignificant trend towards female EXS receiving more extensive resections and female STS receiving less extensive resections or biopsies (p=0.098). (31) . The bottom portion of the outer ring shows the relevant quantitative variables (CE thickness, T1Gd radius, age, PIHNA ϱ, and PIHNA D) and the top portion shows the three aspects of survival that are associated with these variables (EXS, STS, and Overall Survival). Red ribbons indicate significant relationships for female patients and blue ribbons indicate significant relationships for male patients. Variables that were significant in multivariate CPH are connected to the Overall Survival segment and variables that were significant in ttests are connected to the relevant EXS or STS segments.

Patients receiving current standard of care
Due to the timespan over which they were collected, the patients in our cohort received a wide variety of treatment protocols. In order to ensure that our significant results still hold true for patients who receive the current standard of care (maximal safe resection followed by concurrent temozolomide and radiation therapy), we created a subset of patients who received this treatment protocol (Stupp protocol patients) (32) and tested which factors were associated with EXS, STS, and overall survival among those patients (Supplement 15) . In this limited subpopulation, we had 113 males and 66 females (Supplement 15A) . Among females, PIHNA D was a significant independent predictor of overall survival and among males, PIHNA ϱ was a significant independent predictor of overall survival (Supplement 15C) .

Discussion
While there are no differences in the distributions of these quantitative and categorical variables between males and females, this investigation found that there are sexspecific differences in the impact that these variables have on patient survival.

Impact of quantitative variables on survival Sex Differences
Among females, tumor cell diffuse invasion rate (PIHNA D) is strongly negatively correlated with overall survival for females across the various analyses. All three ttests found that the longestsurviving females had lower mean D and the shortestsurviving females had higher mean D when compared to other survival groups (Table 3) . Multivariate CPH analyses found that D was a significant independent predictor of survival (Table  4B) . High PIHNA D was a highly sensitive predictor of STS in the female STS vs NonSTS DT (Figure 3) . Notably, both when EOR was included in multivariate CPH analysis (Supplement 14) and when only Stupp protocol patients were considered (Supplement 15C) , PIHNA D was still an independent predictor of survival for females. In this analysis, high PIHNA D is a predictor for shortterm survival and low PIHNA D is a predictor for longterm survival among female GBM patients. The sexspecific impact of PIHNA D on patient overall survival was also shown in Yang et al. (33) , where high PIHNA D values were only associated with worse survival outcomes among females. Although it was not significant in the CPH multivariate analysis, it is notable that males had a significant positive association between overall survival and PI D/ϱ in univariate analysis (Table  4A) . This suggests that more nodular tumors at time of diagnosis are associated with worse prognosis for males, which is contrary to the finding that more diffusely invasive tumors are associated with worse prognosis for females.
Smaller total tumor size (T1Gd radius) is significantly associated with EXS for females. DT analysis showed that nodes isolating females with below average necrosis radii and CE thickness, both components of overall tumor size, were highly sensitive predictors of EXS (Figures 1 and 2) . In the female EXS vs NonEXS DT, T1Gd radius was a highly sensitive predictor of EXS and 81% of female EXS had a T1Gd radius below 21.93mm (Figure 1) . Similarly, when the mean T1Gd radius of EXS was compared to the mean T1Gd radius of other survival groups, the mean radius of EXS was significantly smaller (Table 3) . Univariate CPH found that T1Gd radius size was a significant predictor of survival (Table 4B) , but if EXS were excluded from the analysis, this relationship is no longer significant (p=0.503). These results suggest female extreme survivors have smaller pretx T1Gd radii, but T1Gd radius is not negatively correlated with overall survival for females in general.
Among males, total tumor size (T1Gd radius) is negatively correlated with overall survival across the statistical and DT analyses. EXS had a significantly smaller T1Gd radii, while STS had significantly larger T1Gd radii ( Table 3) . The pretx size of the T1Gd radii was a significant independent predictor of overall survival for males (Table 4A) . Larger CE thickness, a component of the total tumor size, was associated with shortterm survival for males. STS had significantly larger CE thickness than NonSTS (Table 3) and the male STS vs NonSTS DT showed that above average CE thickness was a highly sensitive predictor of STS ( Figure  3) .

Common to both sexes
Age is known to have a significant impact on the survival of glioblastoma patients (5) (6) (7) and this analysis confirmed that age significantly impacts the survival of both males and females. Across the analyses, older age at time of diagnosis is consistently associated with shorter survival, while younger age is associated with longer survival (Table 3 and 4) .
Lower tumor cell proliferation rates (PIHNA ϱ) are associated with EXS for both males and females, but the reciprocal was not found, higher proliferation rates are not associated with shorter survival. Within both male and female populations, EXS had significantly lower ϱ values than all other survival groups, but there was no significant difference in mean ϱ value between STS and NonSTS (Table 3) . Low ϱ was a highly sensitive predictor of EXS in both the male and female EXS vs STS DT (Figure 2) . In the male and female univariate CPH analyses, ϱ had an almost significant impact on survival (Male p=0.064, Female p=0.052) ( Table 4) . However, if only nonextreme survivors are considered for male and female CPH analysis populations, ϱ no longer significantly impacts survival (M p=0.253 and F p=0.194). This suggests that the large number of EXS in the inclusive analysis disproportionately impacted the significance of ϱ. Low tumor cell proliferation rates appear to be predictive of longterm survival for both males and females, but high rates do not appear to predict shortterm survival.

Impact of categorical variables on survival IDH1 Mutation
While Schiffgens et al. (34) found that only IDH1 mutant males demonstrate significantly improved survival compared to IDH1 wildtype males, our investigation found the opposite, that only IDH1 mutant females demonstrate significantly improved survival when compared to their wildtype counterparts (Supplement 1) . While our study does have a relatively small sample of IDH1 mutants, our finding is in concurrence with the findings of Yang et al. (33) , who grouped females by genetic similarities and found that the longestliving female cohort predominantly consisted of IDH1 mutant females. They did not see this effect for males. Our IDH1 mutant females were all longterm survivors and they demonstrated the same depression in PIHNA ϱ when compared to the wildtype females that the EXS females demonstrated when compared to NonEXS females. However, IDH1 mutant females did not have lower PIHNA D compared to the wildtype population. Meanwhile, IDH1 mutant males did not show improved survival, depressed PIHNA ϱ, or significantly different PIHNA D when compared to IDH1 wildtype males. In Baldock et al. (17) , IDH1 mutation was shown to be significantly correlated with lower ϱ and higher D/ϱ (lower ϱ/D) among contrastenhancing glioma patients. The sexes were not separated in this analysis, so there is a possibility that the effect of the depressed ϱ may have only existed for females. The findings of Schiffgens et al. (34) , Yang et al. (33) , and this investigation make a compelling case for the need to consider sex in IDH1related research.
It is possible that the age difference between IDH1 mutant and wildtype patients contributed to the significant difference in overall survival that was observed between IDH1 mutant and wildtype females. However, IDH1 mutant males and females were both significantly younger than their wildtype counterparts and the significant difference in overall survival was only observed among females. It is not likely that IDH1 status alone led to the association between age and longterm survival, as it has been previously proposed (35) ( 36) , because there was a significant negative association between age and overall survival among IDH1 wildtype patients (Supplement 2) .

MGMT Methylation
Previous studies have demonstrated that MGMT promoter methylation is a significant independent prognostic factor (37) and is more common among longterm survivors (38) (30) . Despite having a relatively small sample of patients with known MGMT methylation status, our analysis was able to confirm that, for both males and females, MGMT methylation was more common among extreme survivors and was a significant independent prognostic factor. Previous studies have also found that the survival benefit of MGMT methylation was stronger or only significant among female patients (34) (39) , but our analysis did not see any evidence of females benefiting more from MGMT methylation than males. However, our analysis did show that methylated females had some of the same characteristics as extreme surviving females, namely that methylated females had lower PIHNA D and significantly lower PIHNA ϱ when compared to unmethylated females.

Laterality
In this investigation, GBM laterality significantly impacted male survival, but had no impact on female survival. Even after accounting for EOR, males with tumors located in the left hemisphere had significantly better survival than males with tumors located in the right hemisphere. We did not consider the role of the lateralization of language dominance in this analysis, but a previous study analyzed righthanded highgrade astrocytoma patients, a group that would be predominantly leftbrain dominant for language (41) , and they did not find tumor location in the dominant or nondominant hemisphere to be a significant predictor of overall survival (42) . Ellingson et al. (40) found that patients who responded favorably to chemotherapy, patients with prolonged survival, and patients with specific genetic modifications, like MGMT promoter methylation and IDH1 mutation, had tumors that clustered in areas of the left hemisphere of the brain. Additional research will need to be conducted on the relationship between genetic modifiers, laterality, sex, and survival.

Extent of Resection
Previous literature has identified extent of resection as a significant predictor of overall survival for GBM patients (43) (44) (6) (18) , but whether EOR has the same impact on survival for males and females has not been clearly elucidated. Our analysis found that EOR has a significant impact on the survival of male GBM patients, with a more complete resection being associated with longer survival and potentially extreme survival. Among females, there was a survival benefit associated with receiving resection, but the extent of resection did not have a significant impact on survival. These findings suggest that EOR may have a sexspecific impact on survival, but further study will be required to fully understand the extent of this difference.

Limitations and Further Work
Due to the utilization of retrospective clinical data, it was not possible to control for all confounding factors and bias within our dataset. However, our utilization of a large cohort of almost 500 patients allows for the mitigation of some of these confounding effects. The findings presented in this investigation lay the groundwork for future research on the topic of sex differences in prognostic indicators of extreme survival in patients with GBM. Future work could control for course of treatment or investigate the impact of factors like radiation therapy and temozolomide dosing regimens. Considering and controlling for the impacts of other genetic modifiers, like p53 mutation, would also be necessary in future work, since these factors influence tumor behavior and survival, and potentially could be influenced by sex. Shinojima et al. (8) observed that their cohort of extreme survivors consisted entirely of females and had a disproportionately large number of giant cell glioblastoma cases. Future work could consider whether histological variations in GBM have sexspecific effects on survival. Additionally, considering more sensitive and individualized elements of the tumor, like the biological environment surrounding the tumor, could provide a more thorough understanding of what makes survival outliers unique.

Conclusion
Taken together, these results further validate the need to consider sex as a relevant biological factor in all glioblastomarelated research. Sex has been shown to significantly impact GBM incidence and prevalence (8) (9) (10) (11) , survival (8) (13) (14) , oncogenic gene expression (33) , glycolytic pathway gene expression (45) , and now the predictors of overall survival. Despite these findings, many studies do not specify patient sex and those that do often do not consider sex when reporting the results of their analysis. The consideration of the role of sex in tumor behavior, incidence, growth, and treatment response will only lead to higherquality, more individualized knowledge and care for glioblastoma patients.