Background

Glioblastoma accounts for nearly half of malignant brain tumors, with an annual incidence of 7 per 100,000 [1]. Treatment advances over the past decades have increased survival, but prognosis remains dismal, with 5-year survival rates near 10%, and a median overall survival (mOS) of 15-24 months [2, 3]. Standard-of-care for newly diagnosed glioblastoma includes maximum safe resection followed by (i) radiotherapy with concomitant daily temozolomide, (ii) adjuvant temozolomide, and (iii) with or without Tumor Treating Fields [4,5,6]. Patients with recurrent disease who receive treatments including bevacizumab, lomustine, or experimental therapies still have an OS of only 9-12 months [7,8,9]. Although recent molecular characterization of glioblastoma tumors helps in stratifying outcomes, it does not account for important clinical factors that influence patient prognosis [10].

Oncologists rely on certain factors to prognosticate patient survival, including age, Karnofsky Performance Status (KPS), isocitrate dehydrogenase (IDH) genotype, and O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation status [11,12,13]. Although age and KPS are the only validated clinical factors for prognosis, the glioblastoma population skews elderly, and they often possess medical comorbidities that may influence treatment outcomes. Therefore, assessing underlying medical comorbidities may provide further guidance for clinical decision making. Patients with more comorbidities tend to have lower rates of surgical resection, chemotherapy/radiotherapy utilization, and clinical trial participation [14,15,16]. Additionally, patients with multiple comorbidities are likely to be more frail and experience worsened surgical outcomes and increased mortality [17,18,19,20]. Furthermore, the presence of specific comorbidities may alter tumor physiology, facilitating cancer progression [21,22,23]. It may also be that poorly controlled comorbidities are significant drivers of dismal patient outcomes [24,25,26]. The cumulative impact from multiple comorbidities may also be relevant [27, 28]. Tools like the Charlson Comorbidity Index (CCI) have shown a greater number of comorbid conditions to negatively impact survival and increase risk for treatment complications [17, 28, 29].

It remains to be seen how overall comorbidity burden and particular diseases influence patient outcomes. Therefore, we examined a database of 867 glioblastoma patients seen at Brown University Health during 2005-2022 and identified medical comorbidities that significantly impact their survival. Specifically, patients with metabolic comorbidities, larger overall comorbidity burden, and particular combinations of comorbidities had the highest risk of death.

Methods

Data collection

Between January 2005 and December 2022, 867 patients with diagnosed glioblastomas who presented to Rhode Island Hospital and Brown University Health Cancer Institute were available for this analysis, together with those initially diagnosed at a different facility or ultimately transferred elsewhere (Table 1). Ten patients with IDH-mutant tumors were excluded from analysis according to the new WHO classification [10, 30].

Table 1 Clinical characteristics of the glioblastoma population at Brown University Health and their prognostic and treatment-related factors.
Full size table

Diagnosis date, death or censoring date, patient demographics, and clinical data, including age, KPS, MGMT promoter methylation, dates and types of treatment, and extent of tumor resection were tabulated from the electronic health record. The extent of tumor resection was defined by the operating surgeon as “gross total”, “near gross total”, “partial”, or “biopsy only”; for this analysis, gross total and near gross total were considered synonymous. KPS at time of diagnosis was estimated based on patients’ clinical presentation when not available in physician notes. Medical comorbidities were documented based on ICD-9 or -10 codes from notes on or before the date of diagnosis. Comorbidities that only appeared in patient charts after glioblastoma diagnosis were excluded from analysis. A complete list of medical comorbidities is included in Supplementary Table 1. ICD codes for physiologically similar conditions were aggregated as a single comorbidity (Supplementary Table 2). Comorbidity burden was determined by totaling the number of comorbidities for a patient, and CCI scores were computed (Supplementary Table 3).

OS was calculated from diagnosis to death, rounded to the nearest whole number. Patients who could not be confirmed as deceased were censored from survival analysis at their last date of contact.

Statistical analysis

Statistical analyses were performed in R version 4.4.1. For each comorbidity and treatment analyzed, survival outcomes over time were determined using a Kaplan Meier estimator, and difference between curves was assessed using the log-rank test (R package survminer). Statistical significance was defined at p ≤ 0.05.

For multivariate analysis, to distinguish survival differences among patient populations, the proportional hazards assumption was verified for each medical comorbidity by fitting an Aalen additive effects model (R package survival) to the comorbidities of interest and plotting their relative effect on survival. Comorbidities and treatments notable from the univariate analysis were then fitted to a Cox proportional hazards model to assess group effects (R package survival).

To assess whether particular comorbidities had disproportionately strong survival effects in combination, the 25 most frequent medical comorbidities were selected if they shared at least 10 patients. For each pair, a Cox proportional hazards model was fitted to the four combined groups in the pairing (comorbidities +/+, + /−, −/ + , −/−). The ‘disproportionality’ of a pair of comorbidities was defined as the ratio between the hazard ratio of the group having two comorbidities and the maximum hazard ratio of having just one. The pair with the greatest disproportionality was selected for further analysis. Notable graphics packages used included ggplot2, ggfortify, forestploter, and igraph.

Results

Patient demographics are similar to previously reported populations

Patient demographics are listed in Table 1. The median age at diagnosis was 65 years (range <1–93). Over 75% of patients were 50-80 years. Eight patients were diagnosed at age ≤20. The male-female ratio was 1.3, with 493 (56.9%) males and 374 (43.1%) females. At the time of data extraction, 86 (9.9%) were alive and 781 (90.1%) were deceased or censored.

Prognostic factors correlate with overall survival

The mOS of the entire cohort was 11 months (95%CI 10–12, Fig. 1a). In the pediatric population with age ≤20 (n = 8), mOS was 25 months. In adults, mOS significantly decreased with increasing age (Table 1); mOS for adults 21-49 years (n = 84) was 22 months, 50–59 years (n = 183) was 17 months, 60–69 years (n = 279) was 11 months, 70–79 years (n = 195) was 7 months, and ≥80 years (n = 108) was 3.5 months (r2 = 0.9075, Fig. 1b and Supplementary Fig. 1).

Fig. 1: Survival characteristics and prognostic factors in the glioblastoma population.
figure 1

Overall Survival (OS) of the entire cohort was 11 months (95% CI 10–12) (a). Younger age (b), high Karnofsky Performance Status (KPS) (c), gross total resection (d), and O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation (e) are favorable prognostic factors. Patients treated with radiation and temozolomide (TMZ) (f), as well as those treated with bevacizumab (g), lived longer than those without.

Full size image

KPS correlated with survival in our cohort (Fig. 1c, Table 1). The mOS for those with KPS of 90–100 (n = 264) was 18 months, 70–80 (n = 294) was 12 months, 50–60 (n = 218) was 5 months, and 10–40 (n = 64) was 2 months (r2 = 0.9113, Supplementary Fig. 2). The KPS of 17 could not be determined.

The extent of tumor resection significantly impacted survival (Fig. 1d, Table 1). Patients who had gross total resection, including those with near total resection (n = 290), had a mOS of 16 months (95%CI 15-19). Those with partial resection (n = 276) and biopsy (n = 291) had mOS of 11 (95%CI 10–13) and 5 (95%CI 4–6) months, respectively (p < 0.0001).

Patients with methylated MGMT promoters (n = 265) had a longer mOS than those with unmethylated promoters (n = 212), 16 compared to 11 months (p < 0.0001, Fig. 1e, Table 1). Patients with IDH mutated status (n = 10) lived significantly longer than those with IDH wild-type (n = 508), with a mOS of 101 compared to 12 months, respectively (p < 0.0001, Table 1).

Patients who received radiotherapy with concomitant temozolomide (n = 667) had mOS of 14 months (95%CI 13-15), while mOS for those who did not (n = 190) was 2 months (95%CI 2-3) (p < 0.0001, Fig. 1f, Table 1). Importantly, the population that received concurrent temozolomide and radiation had median KPS of 80 while those who did not had median of 60, suggesting that poor functional status precluded standard-of-care treatment and contributed to an inferior outcome. Furthermore, bevacizumab was approved for use for glioblastoma in 2009, and patients who received bevacizumab (n = 260) lived longer than those without (n = 597), with respective mOS 19 (95%CI 17–21) and 8 months (95%CI 7–9) (p < 0.0001, Fig. 1g, Table 1).

Our collective analysis demonstrated that age, KPS, extent of resection, IDH status, MGMT promoter methylation, and radiation with temozolomide utilization are relevant prognostic factors in our patients. Therefore, our cohort is not dissimilar to the general glioblastoma population [12, 13].

Univariate analysis of medical comorbidities

The spectrum of medical comorbidities is displayed in Fig. 2a. The most frequent ones are hypertension (n = 499), hyperlipidemia (n = 394), and obesity (n = 244). By system, 666 patients had metabolic conditions, 342 had gastrointestinal/genitourinary, 255 had neurologic/psychiatric, 233 had cardiovascular, 136 had pulmonary, and 318 had others (Fig. 2a). The entire cohort had a median of 4 comorbidities (Fig. 2b). The proportion of patients with multiple comorbidities broadly increased with age (Fig. 2c). Higher comorbidity burden was associated with decreased survival (p < 0.0001, Fig. 2d). Patients with >5 comorbidities had at least double the hazard ratio of those with ≤1 (Fig. 2e). The collective data indicate that advanced age correlates with more medical comorbidities, and this results in shortened survival.

Fig. 2: Types and number of comorbidities in the glioblastoma population.
figure 2

The number of patients with a given comorbidity or condition within a grouping of comorbidities is listed (a). GI/GU is abbreviated for Gastrointestinal/Genitourinary. “Additional malignancy” includes patients with another active or resolved malignant condition, except basal or squamous cell carcinoma of the skin. A median of 4 comorbidities is noted in the population (b). Older patients have a greater number of comorbidities compared to younger ones (c), and the Kaplan-Meier survival decreased (d) while the hazard ratio for death increased (e) with a greater number of comorbidities.

Full size image

We next confirmed the relationship between age and number of comorbidities on survival using the CCI. The CCI is a well-established tool that incorporates age and pertinent medical comorbidities to prognosticate 10-year survival in those with advanced medical illness, including glioblastoma (Supplementary Table 3) [28, 31, 32]. Higher CCI correlated with decreased survival in our dataset. As expected, patients with CCI ≥ 5 (n = 401) had shortened survival compared to the lowest possible score of 2 (n = 79), with respective mOS of 6 (95%CI 5–7) and 23 months (95%CI 19–28) (p < 0.0001, Fig. 3a). To examine a possible age effect, we then calculated CCI without age. In this age-independent model, patients with CCI ≥ 5 (n = 40) still had a shorter survival than those with CCI of 2 (n = 580), with respective mOS of 5 (95%CI 4–8) and 12 months (95%CI 11–14) (p < 0.0001), but the difference is substantially reduced (Fig. 3b). There is also a linear correlation between age and the CCI score (r2 = 0.874, Supplementary Fig. 2). In sum, CCI mainly captures the age effect in glioblastoma patients, and less robustly the comorbidity burden. We next investigated the effects of individual comorbidities on survival to identify factors more prognostic than those included in the CCI.

Fig. 3: Charlson Comorbidity Index (CCI) is strongly influenced by age in the glioblastoma population.
figure 3

Increased CCI correlated with decreasing OS (a). When age was removed, CCI still correlated with decreasing OS, but the magnitude was much reduced (b).

Full size image

We analyzed individual comorbidities using Kaplan–Meier statistics, including those absent from the CCI, such as hypertension, hyperlipidemia, type 2 diabetes mellitus (T2DM), obesity, atrial fibrillation, stroke, chronic obstructive pulmonary disease (COPD), and other malignancies (Fig. 4 and Supplementary Table 4). First, the mOS for patients with and without hypertension (n = 499 and 358) was 9 (95%CI 8–11) and 13 (95%CI 12–15) months, respectively (p < 0.0001, Fig. 4a). For patients with and without hyperlipidemia (n = 394 and 463), the mOS was 10 (95%CI 8–11) and 12 (95%CI 11–13) months, respectively (p = 0.0024, Fig. 4b). The mOS for patients with and without obesity (n = 244 and 412) was 11 (95%11–13) and 11 (95%10–13) months, respectively (p = 0.3684, Fig. 4c). For patients with and without T2DM (n = 142 and 715), the mOS was 7 (95%CI 6–11) and 11 (95%CI 11–12) months, respectively (p < 0.0001, Fig. 4d). Hypertension, hyperlipidemia, and T2DM are all diseases of metabolism, indicating that metabolic disease may play an important role in glioblastoma.

Fig. 4: Univariate analysis revealed multiple comorbidities that influence OS.
figure 4

Hypertension (a), hyperlipidemia (b), diabetes (d), atrial fibrillation (e), stroke (f), Chronic Obstructive Pulmonary Disease (COPD) (g), and additional malignancy (h) all had a statistically significant influence on glioblastoma patient survival, while obesity (c) did not. Patients without Body Mass Index (BMI) data were excluded from obesity survival analysis (n = 201). mOS and respective 95% CI for all comorbidities are listed in Supplementary Table 4.

Full size image

Patients with and without atrial fibrillation (n = 57 and 800) had a mOS of 4 (95%CI 2–7) and 11 (95%CI 11–12) months, respectively (p < 0.0001, Fig. 4e). The mOS for patients with and without stroke history (n = 44 and 813) was 4 (95%CI 3–7) and 11 (95%CI 11–12) months, respectively (p < 0.0001, Fig. 4f). Together, these vascular risk factors appear to play a role in glioblastoma survival.

Patients with COPD are more likely to have a smoking history and may be predisposed to developing malignancies, possibly including glioblastoma [23]. Patients with COPD (n = 36) had a shorter survival than those without (n = 821), with respective mOS of 4 (95%CI 3–9) and 11 (95% CI 10–12) months (p < 0.0001, Fig. 4g). Additionally, those with glioblastoma may be predisposed to other malignancies [33]. We therefore compared patients with a malignancy history to those without (n = 137 and 720), and their respective mOS were 8 (95%CI 6–11) and 11 (95%CI 11–12) months (p = 0.0082, Fig. 4h).

Multivariate analysis of medical comorbidities

Since all comorbidities significant in the univariate analysis were near-uniform in their direction of effect, we determined that the Cox proportional hazards assumption is appropriate for combined effects analysis (Supplementary Fig. 3). Temozolomide/radiotherapy treatment, tumor resection, MGMT promoter methylation, and bevacizumab treatment were the most significant prognostic factors in multivariate analysis (Fig. 5a). At diagnosis, both KPS and age had a small but significant impact on survival, with lower KPS and advanced age having higher hazard ratios (Fig. 5a). Hypertension, additional malignancies, and atrial fibrillation all had significant detrimental effects on patient survival, while hyperlipidemia had a small beneficial effect (Fig. 5a).

Fig. 5: Multivariate analysis reveals the combination of obesity and atrial fibrillation having a strong influence on glioblastoma patient survival.
figure 5

The hazard ratios for dying were listed according to types of prognostic factors, clinical characteristics, and comorbidities (a). Paired comorbidities were analyzed for interaction strength by comparing hazard ratios of double-positive patients and single-positive patients; interactions with most disproportionate hazard ratios are highlighted in red color (b). Patients with both obesity and atrial fibrillation had a markedly worse, statistically significant impact on survival compared to those with only obesity, only atrial fibrillation or neither, as shown with Kaplan-Meier survival (c) and Cox proportional hazard ratio (d).

Full size image

To determine whether medical comorbidities have unbalanced effects in combination, we examined pairings with the most disproportionate hazard ratios (Fig. 5b). When obesity and atrial fibrillation were examined together, their combination had a much greater hazard ratio than either alone. Patients possessing both (n = 16) had a hazard ratio of 3.4 (95%CI 2.1–5.6), while patients with only atrial fibrillation (n = 25) or obesity (n = 228) had insignificant hazard ratios of 1.3 (95%CI 0.9–1.9) and 1.0 (95%CI 0.9–1.2), respectively (p < 0.0001, Fig. 5c and d). This may suggest an underlying severe obstructive sleep apnea in this population [34]. Indeed, there were 52 patients diagnosed with obstructive sleep apnea (OSA), but only 3 had the triad of OSA, obesity and atrial fibrillation (OS 0, 2, and 3 months).

Discussion

A remarkable finding from our dataset is the median of 4 comorbidities in patients. This high number compounds the poor outcomes from established clinical (age, KPS and extent of resection) and molecular (IDH mutation and MGMT promoter methylation) prognostic factors. Our high comorbidity burden can be explained by an older-skewing population, which is consistent with medical literature [35, 36]. The median age of our patients was 65 years, which is a decade older than subjects in clinical trials [4,5,6]. Therefore, our population had a significantly lower median survival (11 months) compared to the previously reported benchmark (15–24 months) [2, 4, 6]. Nevertheless, our patients benefited from radiotherapy/temozolomide and bevacizumab.

Common individual comorbidities in our real-world glioblastoma patients were hypertension, hyperlipidemia, and obesity, and the most frequently impacted systems were metabolic, gastrointestinal/genitourinary, and neurologic/psychiatric, all of which are found typically in the general population. Our comorbidity burden correlates inversely with survival, because the hazard ratio doubles when >5 comorbidities are present. Since the number of comorbidities in our population increases with advancing age, we sought to determine if age and comorbidities have dual influence on glioblastoma survival.

Age dominates CCI score

CCI is an accepted scale to estimate survival outcomes in the general population, including glioblastoma patients [28, 32]. It is a composite score that predicts 10-year survival by combining comorbid conditions affecting cardiovascular, neurologic, pulmonary, renal, metabolic, and hematologic systems with age and especially cancer [31]. Although our glioblastoma population diverged significantly among patients with CCI scores of 2, 3, 4, ≥5, the spread was less dramatic but still notable when the age effect was removed. Therefore, our patients with older age have a much higher risk for death, and risk is still high in those with many comorbidities, independent of age.

Common medical comorbidities alter glioblastoma patient survival

Hypertension is a negative prognostic factor in our cohort, and this may indirectly contribute to early death since blood pressure is regulated by nitric oxide (NO) and renin-angiotensin-aldosterone systems, and both are relevant to cancer. First, NO is a diffusible gas and a potent vasodilator with established roles in angiogenesis and wound healing [37, 38]. Those with hypertension experience higher oxidative stress and subsequent impaired NO bioavailability [39, 40]. This lower NO expression may promote a more hypoxic glioblastoma that shifts toward an invasive mesenchymal phenotype [41]. Atherosclerosis and atrial fibrillation also exacerbate this effect in at-risk patients [42]. Therefore, NO attenuation from inflammatory processes may contribute to glioblastoma progression. Second, angiotensin II, a signaling molecule in the renin-angiotensin-aldosterone system, induces vascular endothelial growth factor (VEGF) expression and subsequent angiogenesis in human malignancies [43]. VEGF is a well-known target in glioblastoma treatment due to its local effects on tumor vasculature [44,45,46]. Blockage of angiotensin II receptors by losartan decreases blood-brain-barrier breakdown and VEGF production, which could prolong survival [47].

Similar to vasoconstrictive-driven hypoxia in hypertension, COPD increases hypoxia-inducible factor-1α expression, which makes the tumor more aggressive [48]. Patients with vascular risk factors can be thought of in a similar manner as those with COPD; cerebral and coronary artery atherosclerosis increase risk for hypoxia, which may promote aggressive tumor phenotypes [41]. These postulations are supported by data showing significant prognostic value of internal carotid artery calcium scores in glioblastoma patients [49].

Metabolic syndromes like T2DM and hyperlipidemia likely impact survival in our patients through mechanisms distinct to those previously described. Insulin resistance and hyperinsulinemia seen in T2DM increases production of insulin-like growth factor (IGF), receptors for which are commonly overexpressed in human cancer cells, and this could further stimulate tumor growth [50]. Although IGF signaling in human gliomas is not firmly established, some data suggest IGF-1 signaling is relevant in glioma progression [22, 51]. In human tumors with MGMT promoter methylation, metformin use has shown a significant survival benefit for those with either preexisting diabetes or steroid-induced hyperglycemia [52]. Furthermore, the addition of metformin to buparlisib, a phosphatidylinositol 3-kinase inhibitor used in clinical trials for human glioblastoma, potentiates the drugs’ efficacy in mouse models [53]. As for hyperlipidemia, plaque deposition in vessels promotes systemic inflammation that can promote tumor progression through previously discussed mechanisms, and cholesterol may be a required survival factor for glioblastoma cells [49, 54]. The small but significant protective survival effect of hyperlipidemia in our population might be explained by higher utilization of statins [26, 55]. Therefore, metabolic syndrome can be a modulating factor on glioblastoma progression.

The last patient group of interest are those with a malignancy history. Many of them are likely immunosuppressed due to prior chemotherapy or supportive medications, and glioblastoma therapy may worsen this [56, 57]. An immunosuppressive environment is beneficial for neoplastic cells because glioblastoma evades the immune system by both local and systemic mechanisms [58]. Dexamethasone, radiation, and temozolomide, which are used ubiquitously in glioblastoma patients, severely reduce CD4 counts, and therefore may potentiate tumor progression [59]. For example, in the HIV population, immunosuppressed states may contribute to the development of multiple malignancies [60]. Therefore, the subpopulation of patients with a history of immunosuppression requires particular attention because this may facilitate unchecked glioblastoma growth.

Combinatorial analysis uncovers potential pathologic state that modulates outcome

Despite frequent co-occurrence with other metabolic comorbidities, obesity alone had no independent survival effect in our population. Other studies have shown mixed results, with some showing a negative effect, while others find it to be protective [61, 62]. Elevated BMI may signify increased adiposity, which can stimulate tumor growth by angiogenesis and systemic inflammation, although increasing BMI may also correlate with muscle mass, which can be a protective measure for cancer-related mortality [63, 64]. Atrial fibrillation did have a negative prognostic effect in our univariate and multivariate analyses. These patients may develop subclinical clots within the tumor microenvironment that trigger inflammation and subsequent tumor progression [65, 66]. Importantly, patients with both atrial fibrillation and obesity had a significantly worsened survival compared to those with either comorbidity alone. Both conditions contribute to a thrombophilic state, with increased risk of complications like stroke [67]. Neurological deficits from glioblastoma may mask any new symptoms from embolic events. Additionally, obesity and atrial fibrillation are associated with OSA, an association that is well-established in cardiology based on prior retrospective and prospective studies, as well as interventional observations [68,69,70,71]. The potential for intermittent hypoxia from OSA to fuel glioblastoma progression is a real possibility [41, 48]. Although patients with an existing diagnosis of OSA did not show decreased survival (Supplementary Table 4), three of these OSA patients had concurrent obesity and atrial fibrillation, and they had shortened survival. We suspect this is an underestimate, and many more glioblastoma patients may have undiagnosed severe OSA [34]. This relationship may be further obscured by interventions like continuous positive airway pressure (CPAP) for OSA or glucagon-like peptide-1 agonists for weight loss. Furthermore, intervention for OSA with CPAP has been shown to improve atrial fibrillation management [70, 71]. Therefore, we speculate that prompt intervention for OSA may prolong glioblastoma patient survival.

Future directions

A strength of this investigation was the large patient population and timespan. Our database captured the population of glioblastoma patients from most Rhode Island communities for nearly two decades. However, this study was a single-institution retrospective review, and as such requires validation with patient cohorts from other institutions. Furthermore, this study was an unbiased survey of medical comorbidities that did not include data on post-diagnosis complications, interventions for particular comorbidities, or social determinants of health. Future studies should focus on optimizing patient care based on relevant medical comorbidity profiles.

Conclusions

We have shown that CCI does not incorporate the most relevant comorbidities in our glioblastoma population. Individual conditions such as hypertension or additional malignancies, or combinations of conditions such as obesity and atrial fibrillation, are more prognostically useful than CCI and may be superior for risk stratification. Increased understanding of pertinent glioblastoma comorbidities will advance medical interventions and patient outcomes.