Association Between Hyperglycemia and Survival in Patients With Newly Diagnosed Glioblastoma

A growing body of evidence in the diabetes literature suggests that short-term, and even relatively mild, hyperglycemia leads to excess morbidity and mortality in acute illness.^1,2 In the setting of cancer, in particular, large epidemiologic studies have suggested that survival is shorter in patients with comorbid diabetes.^3–7

Patients with glioblastoma multiforme (GBM) are at particular risk for hyperglycemia, because their peritumoral edema is routinely treated with high-dose glucocorticoids, which are known to increase plasma glucose, due to impairing glucose transport.^8–10 Furthermore, because patients with GBM have a poor prognosis and are thus unlikely to be at risk for long-term diabetes complications, the goal of hyperglycemic management in these patients is, in general, the avoidance of acute hyperglycemic complications, rather than intensive management. In this study, we tested the hypothesis that hyperglycemia is associated with shorter survival in patients with newly diagnosed GBM.

This retrospective study, approved by the institutional review board, analyzed the association between hyperglycemia and survival among patients with newly diagnosed GBM accrued onto five New Approaches to Brain Tumor Therapy CNS Consortium phase II trials. These trials were conducted between 1999 and 2004, before the standard use of temozolamide. They assessed the effect on survival of noncytotoxic therapies, including penicillamine, carboxyamido-triazole, suramin, gadolinium-texaphyrin, and celecoxib, which were administered in combination with postoperative external-beam radiation therapy (60 Gy in 30 fractions).^11–15 None of the experimental agents are known to affect glucose homeostasis, and none were associated with improved survival.

Of the 191 patients studied, the mean age was 55.7 years (standard deviation, 11.2 years), the median KPS was 90 (range, 60 to 100), 64% were male, and 93% were white. Mean daily dexamethasone dose ranged between 0 and 48 mg (median, 5 mg). The mean time-weighted glucose varied between 65 and 459 mg/dL, and was divided into quartiles: quartile one (< 94 mg/dL), quartile two (94 to 109 mg/dL), quartile three (110 to 137 mg/dL), and quartile four (> 137 mg/dL).

Comparisons of demographic and clinical characteristics according to quartile of time-weighted mean glucose are listed in Table 1. Consistent with the glucose-elevating effect of glucocorticoids, mean daily dexamethasone dose increased with increasing quartile of mean glucose (P = .006 for trend). Additionally, mean age increased and KPS decreased with increasing quartile of mean glucose (P = .001 and .040 for trend, respectively). There was no difference in the distribution of sex, race, or surgical debulking versus biopsy with regard to quartile of mean time-weighted glucose.

Of the 191 patients accrued onto the studies, 187 (98%) died in the follow-up period, which encompassed 214.3 person-years. The mortality rate was 0.87 deaths per person-year, and the median survival was 11.0 months. The four surviving patients were observed between 35 and 82 months.

Median survival was 14.5 months in quartile one, 11.6 months in quartile two, 11.6 months in quartile three, and 9.1 months in quartile four. In an unadjusted Cox proportional hazards model, relative to quartile one, the estimated HRs were 1.30 for quartile two (95% CI, 0.86 to 1.95), 1.56 for quartile three (95% CI, 1.02 to 2.37), and 2.06 for quartile four (95% CI, 1.35 to 3.14, with a P value of .001 for trend; Table 2). For every 10 mg/dL increase in time-weighted mean glucose, the HR was 1.05 (95% CI, 1.02 to 1.07; P < .0001). In addition, mean daily dexamethasone dose, age, and KPS were each statistically significantly associated with survival in the univariate analysis (Table 2).

After we controlled for mean daily dexamethasone dose, age, and KPS in a multivariate Cox proportional hazards regression model, the association between hyperglycemia and survival persisted. Figure 1 displays the adjusted survival curves for the four quartiles of time-weighted mean glucose. Compared with patients in the lowest quartile of mean glucose, those in quartile two (adjusted HR, 1.29; 95% CI, 0.85 to 1.96), quartile three (adjusted HR, 1.35; 95% CI, 0.89 to 2.06), and quartile four (adjusted HR, 1.57; 95% CI, 1.02 to 2.40) were at progressively higher risk of dying (P = .041 for trend; Table 3). An incremental increase of 10 mg/dL in time-weighted mean glucose was associated with an adjusted HR of 1.03 (95% CI, 1.00 to 1.06; P = .035).

The strength of our findings was demonstrated by several sensitivity analyses. Our results were unchanged when we excluded 68 patients with gaps longer than 2 months between glucose or glucocorticoid recordings (adjusted HR associated with a 10 mg/dL increase in time-weighted mean glucose, 1.05; 95% CI, 1.00 to 1.11; P = .036), and when we excluded 18 patients with outlying survival times (HR, 1.03; 95% CI, 1.00 to 1.06; P = .051). The results were also largely unaffected when average glucose not weighted for time was used as the exposure variable (HR, 1.04; 95% CI, 1.00 to 1.07; P = .032), and when the standard deviation of each patient's mean glucose was adjusted for in the model (HR, 1.03; 95% CI, 1.00 to 1.06; P = .081).

The secondary outcome (infection) was experienced in follow-up by 27 patients (14%), 18 of whom had pneumonia. Patients who developed infections had worse survival compared with patients without infections, with an HR of 1.64 (95% CI, 1.08 to 2.48; P = .02). In a multivariate logistic regression analysis, there was a trend toward an association between infection and mean glucose, with an odds ratio for infection of 1.06 (95% CI, 0.99 to 1.13; P = .09) for every 10 mg/dL increase in time-weighted mean glucose. However, adjusting for infection in the multivariate model did not alter the significant association between mean glucose and survival (adjusted HR associated with a 10 mg/dL increase in time-weighted mean glucose remained 1.03; 95% CI, 1.00 to 1.06; P = .035).

In this study, we report a shorter survival time associated with hyperglycemia in a group of patients with GBM and good baseline KPS. Even after adjustment for mean daily dexamethasone dose and other prognostic variables, the HR increased with increasing quartile of time-weighted mean glucose.

Recent large population-based studies have suggested that diabetes and impaired glucose tolerance are risk factors for developing certain types of cancer.^21,22 Furthermore, pre-existing diabetes at the time of cancer diagnosis has also been shown to be associated with increased all-cause mortality and cancer recurrence, most definitively in colon and breast cancers.^4–7,23 In the specific case of glioma, small studies have not shown that premorbid diabetes increases cancer incidence,^24–28 but the effect of hyperglycemia on survival after GBM diagnosis has never before been assessed.

One mechanism that has been proposed to explain the association between diabetes and worse survival in colon and breast cancers is that hyperinsulinemia resulting from hyperglycemia stimulates tumor growth. Insulin is a member of a family of growth factors, and similar to its cousins (insulin-like growth factors 1 and 2), may promote tumor proliferation.²⁹ Patients with colorectal cancer have demonstrated higher fasting insulin levels than have control patients,³⁰ and a prospective study has shown that patients with breast cancer with higher fasting insulin concentrations have decreased survival and higher rates of recurrence.³¹ In vivo studies have shown that high insulin levels enhance colorectal and breast cancer cell proliferation via receptors on the tumors.^32,33 Furthermore, treatment with metformin, which lowers insulin levels, decreases the incidence and size of mammary adenocarcinomas in tumor-prone transgenic mice.³⁴ Human glial tumors possess insulin receptors with binding activities and characteristics identical to peripheral insulin receptors, and insulin has been shown to stimulate glucose uptake in cultures of human GBM cells.³⁵

Another mechanism may be that hyperglycemia itself promotes tumor growth. Glucose is the major substrate for cerebral metabolism, and high-grade brain tumors have high glucose consumption.^36–38 The glucose content in the brains of healthy mice increases minimally after an intraperitoneal bolus of glucose, whereas mice with gliomas experience an increase in intratumoral glucose content of 2.5 times after induction of hyperglycemia.³⁹ High glucose levels within GBMs could provide extra substrate for glycolytic metabolism, and support unchecked tumor growth. Unfortunately, we were not able to directly evaluate this proposed biologic mechanism in this study, because radiation therapy for GBM commonly results in pseudoprogression, which would have complicated this assessment.^40,41

It is also possible that the association between hyperglycemia and worse survival is mediated not by tumor growth, but rather by other adverse effects of hyperglycemia, such as increased infection rates. Hyperglycemia is known to increase the risk of infection in outpatients and inpatients,^42–44 and this risk is likely magnified when glucocorticoids are received, because glucocorticoids suppress the immune system and predispose patients to infection.^45,46 Although we did observe a trend toward a relationship between hyperglycemia and infection in our study sample, infection had no impact on the association between hyperglycemia and survival. This finding suggests that the mechanism by which hyperglycemia may shorten survival does not include increasing infection rates.

To our knowledge, this is the first study to evaluate the association between glucose levels during cancer treatment and survival. An important strength of our study was our large study population, which was homogeneous with regard to eligibility criteria and treatment. Data collection was reliable and relatively uniform, because all patients were receiving research protocols. Moreover, patients with GBM, in particular, are at high risk for hyperglycemia, given the routine use of glucocorticoids in their management. Because glucocorticoid dose would also be expected to reflect the extent of residual disease and peritumoral edema, the validity of our results was strengthened by carefully adjusting for glucocorticoid dose in the analysis.

This study had several limitations inherent to its retrospective design. As with any observational study, there remains a possibility that unmeasured confounders influenced our findings. For example, residual tumor burden, which would have shortened survival and might have increased glucose because of physiologic stress, may have been incompletely controlled for in this analysis, despite our adjustment for glucocorticoid dose. Also, there was heterogeneity in the frequency of glucose and glucocorticoid recordings, although a sensitivity analysis did not suggest that this biased our results. Furthermore, the glucose measurements available were performed at random. Hemoglobin A1c levels would have provided a better index for evaluating long-term glycemia. Other potentially important variables that were not assessed in this retrospective study of patients diagnosed and treated from 1999 to 2004 include O⁶-methylguanine–DNA methyltransferase methylation status, exogenous insulin dose, and serum insulin level. These will be measured in a future prospective study aimed at confirming our results in the post-temozolomide era.

Our findings raise the question of whether intensive management of hyperglycemia would lead to improved survival in patients with newly diagnosed GBM. To answer this question, a randomized trial is being planned to compare survival between hyperglycemic patients with GBM whose hyperglycemia is managed intensively by an endocrinologist, and hyperglycemic patients with GBM whose hyperglycemia is managed by an oncologist according to the standard of care.

In conclusion, we observed significantly shorter survival in patients with GBM who experienced hyperglycemia, compared with patients with GBM who did not experience hyperglycemia, after controlling for several important confounders, including glucocorticoid therapy. Future studies evaluating whether lowering glucose improves cancer survival are warranted, especially considering the relatively benign nature of diabetes interventions.

The author(s) indicated no potential conflicts of interest.