Sarcopenia is associated with chemoradiotherapy discontinuation and reduced progression-free survival in glioblastoma patients

We screened consecutive adult patients undergoing radiotherapy planning CT scans for histologically proven primary glioblastoma (isocitrate dehydrogenase 1 [IDH1] wildtype) at our institution between January 1, 2010, and December 31, 2021. Patients with incomplete clinical data and those with incomplete visualization of the C1 vertebra (either due to cutoff or due to streak artifacts) were excluded. We also omitted patients with prior cranial radiotherapy. Only patients with a novel glioblastoma diagnosis were considered, while recurrences were not included. Hence, inclusion criteria were similar to before [11] and the prior cohort from the exploratory study was part of this follow-up analysis. A screening flowchart is demonstrated in Supplementary Fig. 1.

Glioblastoma was defined according to the current definition of the World Health Organization (WHO) [12]. Clinical data collected included patient demographics (age, sex, height, weight, comorbidities according to the Charlson Comorbidity Index [CCI], and postoperative Eastern Cooperative Oncology Group [ECOG] score), tumor characteristics (date of diagnosis, tumor volume, O6-methylguanine-DNA methyltransferase [MGMT] promotor methylation, and IDH1 mutational status), and course of treatment (surgery date, resection status, planned radiotherapy dose and fractionation, applied radiotherapy dose and fractionation, chemotherapy application, prescription of corticosteroids, posttreatment stay, and progression during chemoradiotherapy). Hypofractionation was defined as a daily radiation dose exceeding 2 Gy. ECOG score was determined postoperatively at the onset of chemoradiotherapy, as recommended [13]. We quantified tumor volume on preoperative contrast-enhanced T1-weighted magnetic resonance imaging (MRI). Finally, we also collected data on first-time postoperative progression according to the Response Assessment in Neuro-Oncology Criteria (RANO) criteria [14]. Progression was defined as the time interval from diagnosis to MRI-based progression or death, whichever occurred first, similar to previous studies [15]. Data were partly abstracted from the patient records as well as from a neurosurgical database and the cancer registry of the Western German Cancer Center (Westdeutsches Tumorzentrum, WTZ).

Non-contrast radiation planning CT scans were downloaded from the ARIA Oncology Information System (Varian Medical Systems, Inc., Palo Alto, USA). After upload onto 3D Slicer (version 4.13.0), we reformatted the CT scans along the frontal and vertical plains to ensure consistent measurements independent of patient positioning. This way, the nasal septum and the dens axis were both oriented in a vertical plain [11].

We measured the retrocervical muscle area on a single axial CT image at the level of the first cervical vertebra (C1), as previously established (Fig. 1a; [11]). Threshold-based semi-automated segmentations were performed with established cutoff values of −29 to +150 Hounsfield units [9]. Besides the autochthonous muscles, we also included the trapezoid, the sternocleidomastoid, and the levator scapulae muscles in the analyses. Quantifications resulted in a muscle cross-sectional area value, measured in square centimeters.

Descriptive statistics were used to analyze patient characteristics. We compared the incidence of bivariate outcomes between sarcopenic and non-sarcopenic patients using the chi² test. For progression-free survival (PFS), we chose a time-to-event analysis, starting at the time of histopathologic diagnosis. In univariable analyses, we used the Kaplan–Meier method and ran a log-rank test to evaluate statistical significance. In multivariable analyses, Cox proportional hazard regressions were applied. The same model previously developed for OS [11] was used for consistency for PFS and OS. Additionally, we generated a second multivariable PFS model restricted to patients who completed standard chemoradiotherapy (e.g., 60 Gy of radiation with concomitant chemotherapy). Statistics were performed using the STATA software package (version 13.0, StataCorp, College Station, TX). A p-value below 0.05 was considered significant.

We screened 372 glioblastoma patients treated at our institution and included 363 in the study (Supplementary Fig. 1). Median age was 63 years, median body mass index (BMI) was 26 kg/m², and most patients were male (58.4%). Median CCI was 4, indicative of a low comorbidity score. Roughly half of the tumors showed MGMT methylation. Most glioblastomas were fully (44.6%) or partially (40.2%) resected, while the remainder (15.2%) underwent biopsy only (Table 1).

Of 363 patients, 311 (85.7%) received the standard radiation dose of 60 Gy. 90.1% of patients were administered concomitant chemotherapy. Chemotherapy overwhelmingly included temozolomide according to Stupp et al. [16] (81.9%), or temozolomide and CCNU according to Herrlinger et al. [17] (14.1%). 52.6% of patients were prescribed corticosteroids during radiotherapy. 9.4% discontinued radiation. Overall, 303/363 (83.5%) of patients completed chemoradiotherapy (i.e., radiotherapy to a dose of 60 Gy and concomitant chemotherapy). 7.7% progressed during radiotherapy. After treatment completion, 86.3% continued outpatient care while 6.3% were transferred to inpatient palliative care treatment. Median PFS was 7.7 months (95% confidence interval 7.0–8.5 months). 89.5% of patients progressed during follow-up: 259 (71.3%) patients showed progression on follow-up imaging while 66 (18.2%) of patients died with no available prior evidence of progression. Finally, 38 (10.5%) patients did not progress during follow-up.

Median muscle measurement was 27.7 cm² (interquartile range 22.5–33.0 cm²). Men demonstrated a larger muscle area at 31.1 cm² (interquartile range 27.6–35.2 cm²) compared to females at 22.2 cm² (interquartile range 19.9–25.4 cm²; p < 0.001). Measurements of less than 27.6 cm² in males and less than 19.9 cm² in females were considered sarcopenic as they were in the lowest quartile of their peers (Fig. 1b, c). In sum, 91 patients (38 females and 53 males) were considered sarcopenic while 272 patients (113 females and 159 males) were considered non-sarcopenic.

Sarcopenic patients had a lower BMI (p = 0.006), were older (p < 0.001), and had higher CCI (p < 0.001) and ECOG scores (p < 0.001). Finally, tumors were larger at the time of diagnosis (p = 0.023). There were no differences in tumor characteristics otherwise, and resection status was not different between groups (Table 1).

Sarcopenic patients were more infrequently treated with a standard dose of 60 Gy (78.0% vs. 88.2% for non-sarcopenic patients, p = 0.016). Concomitant chemotherapy was also more rarely administered (79.1% vs. 93.7%, p < 0.001). Conversely, radiotherapy was more often applied in a hypofractionated treatment regimen (12.1% vs. 4.0%, p = 0.005, Fig. 2a).

During treatment, sarcopenic patients were more often prescribed corticosteroids due to intracranial pressure symptoms (63.7% vs. 48.9%, p = 0.014). Sarcopenic patients showed a higher rate of progression during therapy (13.2% vs. 5.9%, p = 0.024) and radiotherapy was more often discontinued before reaching the prescribed dose (15.4% vs. 7.4%, p = 0.023). In sum, 72.5% of sarcopenic patients received standard chemoradiotherapy (i.e., radiation to a dose of 60 Gy in 2‑Gy fractions with concomitant chemotherapy), compared to 87.1% of non-sarcopenic patients (p = 0.001). Among those patients treated with chemotherapy, there was no difference in the choice of agent (temozolomide, temozolomide plus CCNU, or other) between the groups (p = 0.2, Supplementary Table 1). 11.0% of sarcopenic and 4.8% of non-sarcopenic patients were transferred to inpatient palliative care (p = 0.035). Conversely, sarcopenic patients were less likely to remain in outpatient care (as opposed to inpatient treatment) after completion of radiotherapy (80.2% vs. 88.2%, p = 0.055, Fig. 2b).

PFS was markedly reduced in sarcopenic patients in univariable analyses in Kaplan–Meier assessment (Fig. 3a) and log rank test (p < 0.001). Median time to progression was 5.1 months (95% confidence interval 4.0–7.0 months) in sarcopenic patients and 8.4 months in the non-sarcopenic group (95% confidence interval 7.5–9.9 months). When assessing all muscle measurement quartiles separately, no significant differences in PFS were seen between the upper three quartiles (median time to progression was 9.1, 8.0, and 8.7 months for the second, third, and fourth quartile, respectively; Fig. 3b). We then performed multivariable modeling, describing hazard ratios (HR) and confidence intervals (CI) for the association between patient, tumor, or treatment parameters (including sarcopenia measurements) and PFS (Table 2). When including predefined risk factors, sarcopenia remained independently associated with PFS (HR 0.61 [CI 0.46–0.81] relative to non-sarcopenic patients, p = 0.001). Additionally, BMI, CCI, and bilateral tumors showed a negative, while extent of resection and methylated MGMT status showed a positive association with PFS (Table 2). We also tested continuous muscle measurements in a multivariable model, finding that increased muscle measurements (per cm²) were independently associated with prolonged PFS (HR 0.96 [CI 0.94–0.99], p = 0.002, Table 2). In a final analysis, we restricted the multivariable model to patients completing standard chemoradiotherapy treatment (i.e., treatment with 60 Gy in 2 Gy fractions and concomitant chemotherapy) to assess independence of muscle-related associations from treatment completion. The significance of muscle measurements remained unchanged both in analyses for sarcopenic/non-sarcopenic groups (HR 0.65 [CI 0.47–0.89], p = 0.008) and for continuous measurements (HR 0.97 [CI 0.94–0.99], p = 0.007) in this subgroup (Supplementary Table 2).

Our novel sarcopenia definition was also strongly associated with OS in univariable log-rank analyses (median survival 9.1 vs 15.2 months, p < 0.001, Supplementary Fig. 2) as well as in multivariable modeling (HR 0.64, CI 0.49–0.85, p = 0.002, Supplementary Table 3). This extends and reconfirms previous exploratory results [11].

The American Society for Radiation Oncology (ASTRO) guideline [19] suggests age-stratified glioblastoma treatment with hypofractionated radiotherapy recommended for patients older than 70 years. The guideline’s authors note that age cutoff values may differ substantially between studies and many subgroup analyses are small and retrospective in nature [19]. Age also remains an imperfect variable given variation in physical functioning, quality of life, and prognosis between patients of similar age [20, 21]. Elderly patients with good physical functioning may indeed profit from chemoradiotherapy as opposed to radiotherapy alone [22]. Thus, the authors also suggest stratifying by “performance status” to better individualize treatment recommendations [19]. However, Karnofsky index and ECOG score are prone to high interrater variability [23, 24].

As a quantitative, non-rater-dependent biomarker, sarcopenia measurements incorporate both age and performance status [25]. Sarcopenia becomes more prevalent with age [25] and is negatively associated with physical functioning [5]. Some studies refer to body composition parameters as measures of “biologic age” as they encompass nutritional factors, functioning, inflammation, and chronological age [26]. This is reflected in our findings, as sarcopenia was associated with these factors in our cohort: Significant associations were found for chronological age, BMI, comorbidities, and ECOG score. The strongest signal was found for ECOG, demonstrating a close association between sarcopenia and (reduced) physical functioning, as expected [27]. Notably, whether sarcopenia is the result of a limited physical status or vice versa remains unclear, as the retrospective nature of our study allows for tests regarding association, but not causation. In any case, sarcopenia is a quantitative marker related to numerous otherwise difficult-to-quantify parameters.

Importantly, as discussed above, sarcopenia is associated with key treatment stratification parameters from the ASTRO guideline. Consistent with this, sarcopenic patients were less likely to receive the standard chemoradiotherapy treatment and more likely to undergo hypofractionated irradiation without chemotherapy in our study. Interestingly, despite abbreviated and de-escalated treatment prescription, patients still discontinued radiotherapy at a higher rate. This underlines the need for reduced treatment paradigms in sarcopenic patients. Hence, sarcopenia may be a potential prospective marker that helps guide treatment decision-making.

It is well known that elderly patients are at risk for increased treatment toxicities, most prominently from chemotherapy [28]. Treating younger patients with poor performance status also remains a major concern, yet data are more limited [29].

Sarcopenic patients were more often prescribed corticosteroids during treatment (64% vs. 49%). Steroids may change body composition measures in the long term [30] and aggravate sarcopenia. However, this is unlikely to have affected our measurements, as radiation planning CTs are commonly performed within weeks after primary diagnosis and neurosurgical evaluation, and before application of radiotherapy. Our data indicate that sarcopenic patients may be more at risk for brain edema symptoms, as these trigger corticosteroid therapy during radiation.

While chemoradiotherapy discontinuation was increased among sarcopenic patients, 85% of sarcopenic patients still completed the prescribed therapy. We believe our data suggest that sarcopenic patients may profit from more intensive medical care during therapy. However, the relatively high completion rate of radiotherapy among sarcopenic patients may support continued use of postoperative radiotherapy (with the potential omission of chemotherapy) in this group, similar to findings from the ASTRO guideline for elderly patients or those with low performance status [19]. However, our study is retrospective in nature and this question may only be definitively answered in prospective trials.

Upon treatment completion, a small subpopulation of patients were directly transferred to in-patient palliative care, likely indicative of limited outcomes. While these patients were overrepresented among sarcopenic patients (11% vs. 5%), the vast majority of sarcopenic patients, 80%, returned home after treatment. Sarcopenia may nonetheless help to identify patients who could profit from intensified, and early, palliative care interventions.

The associations between sarcopenia and OS [31, 32] or complications [33‐35] have been well described in numerous malignancies. Most studies imply that reduced physical functioning (commonly associated with sarcopenia [5]) is likely to make patients more susceptible to developing complications, limiting survival [36].

In our study, sarcopenic glioblastoma patients were more likely to show early progression. This is an intriguing finding, as PFS (different from OS) is likely not directly associated with reduced physical functioning. We initially hypothesized that treatment discontinuation in sarcopenic patients may induce early progression, similar to other malignancies [37]. However, while treatment was more limited in some sarcopenic patients, nearly three quarters of sarcopenic patients received full standard chemoradiotherapy treatment. Thus, we considered differences in treatment completion or chemotherapy application as unlikely to account for the drastic difference in PFS across the entire sarcopenic population (median PFS 5.1 vs. 8.4 months). When assessing this hypothesis, we were reluctant to include chemoradiotherapy treatment completion in our multivariable PFS model as we aimed to only include parameters available at the onset of postoperative treatment. Chemotherapy discontinuation may be advised during treatment depending on blood test results [38]. Hence, we generated a second, separate multivariable model and only included patients who completed standard chemoradiotherapy. In this homogeneously treated cohort, the strong association between sarcopenia and PFS remained unchanged. Consequently, we concluded that the link between sarcopenia and PFS was not mediated via treatment discontinuation.

Similarly, sarcopenic patients did not show differences in tumor characteristics or resection status, making it unlikely that these factors confounded our findings. Notably, tumor volume at diagnosis was slightly increased in sarcopenic patients, raising questions regarding the interplay between increased tumor growth and a rising incidence of sarcopenia. This association has been confirmed in different tumor entities [39]. However, despite inclusion of tumor volume in our multivariable model, sarcopenia was independently associated with PFS. Interestingly, residual tumor volume was not significantly different between the groups after resection.

Thus, after careful consideration of potential clinical confounders, we believe that sarcopenia is instead likely to be associated with PFS via biological factors. In other tumor entities, these include sarcopenia-related immune senescence [40] and proinflammatory signaling [41]. In glioblastoma, immunosuppression increases treatment resistance [42, 43] and induces a distinct neuroinflammatory microenvironment that promotes tumor growth and invasion [44]. In lung cancer, immunotherapy (which is not part of standard treatment in glioblastoma patients) showed a substantially reduced efficacy in sarcopenic patients, potentially indicative of an altered baseline immune system in this group [45]. Prospective studies have linked sarcopenia to systemic inflammation and a distinct disbalance in the immune system [46].

We suspect that sarcopenic glioblastoma patients are a subgroup of patients with an unfavorable inflammatory and immune status. This may potentially promote tumor progression, leading to the reduction in PFS we see in the sarcopenic subgroup.

Clinically, this finding is meaningful, as the optimal follow-up intervals for glioblastoma patients remain unclear [47]. In the absence of specific evidence, MRIs are commonly scheduled every 12 weeks following intra-institutional pragmatic considerations [47, 48]. Novel computational studies try to predict individualized time to progression to optimize this decision [49]. Here, sarcopenia may be a valuable parameter. Our study indicates that it may be reasonable for sarcopenic patients to undergo more frequent follow-up imaging, e.g., every 8 weeks. However, given moderate tolerance of primary treatment and aggressive tumor growth in sarcopenic patients, the clinical benefit of sometimes intensive retreatment [50] in this vulnerable subgroup might be debatable and should be considered with care. Notably, our measurements—which become available only postsurgically—are neither designed nor intended to change the role of up-front surgical resection, which remains the gold standard for virtually all glioblastoma patients.

As expected, based on the previous exploratory study [11], our increased cohort again shows a convincing association with overall survival, both in univariable and multivariable modeling. Different from the previous study, we used our novel sarcopenia definition in this analysis (as opposed to simply dividing a cohort by the median, as done in the exploratory stage). The strong difference in median survival of roughly 6 months demonstrates that our sarcopenic measurements have identified an at-risk cohort with good selectivity.

CT-based measurements are universally attainable in radiotherapy patients. The diversification of the glioma diagnosis based on methylation profiling and gene expression [51, 52] will increase the need for commonly available parameters to identify populations in which treatment escalation or de-escalation is advisable. This was also reflected in a recent post-hoc analysis of the CATNON trial [53], questioning the concurrent use of TMZ in adjuvant glioblastoma radiotherapy, and the lack of a clear treatment standard for patients with relapsed disease. Identifying sarcopenia early might help to guide treatment decisions in different prognostic groups and stages of treatment. End-to-end pipelines for automation of measurements have been built and may facilitate clinical implementation [54]. Automated scoring results (e.g., percentiles based on the overall patient cohort) may then be considered for therapeutic decision-making. However, prospective data collection and validation needs to precede any assessment of clinical application.

Sarcopenia may be modifiable to some degree, mainly through changes in nutrition and physical activity [55]. However, whether these interventions improve outcomes remains unclear. Our study indicates that patients with very low muscle measurements would potentially profit most from interventions. However, these patients may suffer from neurologic symptoms, such as hemiplegia, limiting the potential for physical exercise. Nonetheless, we believe that past and ongoing physical exercise [56‐58] and nutritional investigations [59] in glioblastoma patients have merit.

Some study limitations should be noted. First, this is a retrospective monocentric analysis with the corresponding risk of bias. However, we included a large cohort of patients and patient characteristics were largely representative of the overall glioblastoma population. Nonetheless, we remain unable to address prospective questions such as the modifiability of sarcopenia. Second, multivariable modeling was not possible for treatment application and tolerance outcomes due to high treatment adherence and low complication rates. However, multivariable modeling was performed for PFS. Third, PFS was not available in all patients. However, a large majority of patients, more than 70%, had available PFS. Fourth, semi-automated segmentations required some manual correction, potentially limiting clinical application. However, body composition measurements have been shown to be automatable thanks to novel computational methods [60]. Finally, MRI-based radiation planning may alleviate the need for cranial planning CTs in the future, preventing opportunistic sarcopenia measurements. However, while technically more challenging, sarcopenia measures are also feasible on MRI imaging [61].