Impact of postoperative dexamethasone on survival, steroid dependency, and infections in newly diagnosed glioblastoma patients

Abstract

Background

We examined the effect of dexamethasone prescribed in the initial 3 postoperative weeks on survival, steroid dependency, and infection in glioblastoma patients.

Methods

In this single-center retrospective cohort analysis, we electronically retrieved inpatient administration and outpatient prescriptions of dexamethasone and laboratory values from the medical record of 360 glioblastoma patients. We correlated total dexamethasone prescribed from postoperative day (POD) 0 to 21 with survival, dexamethasone prescription from POD30 to POD90, and diagnosis of an infection by POD90. These analyses were adjusted for age, Karnofsky performance status score, tumor volume, extent of resection, IDH1/2 tumor mutation, tumor MGMT promoter methylation, temozolomide and radiotherapy initiation, and maximum blood glucose level.

Results

Patients were prescribed a median of 159 mg [109-190] of dexamethasone cumulatively by POD21. Every 16-mg increment (4 mg every 6 hours/day) of total dexamethasone associated with a 4% increase in mortality (95% confidence interval [CI] 1%-7%, P < .01), 12% increase in the odds of being prescribed dexamethasone from POD30 to POD90 (95% CI 6%-19%, P < .01), and 10% increase in the odds of being diagnosed with an infection (95% CI, 4%-17%, P < .01). Of the 175 patients who had their absolute lymphocyte count measured in the preoperative week, 80 (45.7%) had a value indicative of lymphopenia. In the POD1-POD28 period, this proportion was 82/167 (49.1%).

Conclusions

Lower survival, steroid dependency, and higher infection rate in glioblastoma patients associated with higher dexamethasone administration in the initial 3 postoperative weeks. Nearly half of the glioblastoma patients are lymphopenic preoperatively and up to 1 month postoperatively.

Dexamethasone, a corticosteroid, is commonly prescribed to newly diagnosed glioblastoma patients primarily to control tumor-associated edema and reduce the symptoms caused by the mass effect on the surrounding normal brain.^1,2 It is a potent immunosuppressant.^3–5 Other indications for its use are postoperative nausea and vomiting prophylaxis^6–8 and management of postoperative aseptic meningitis.⁹ Despite its widespread use since the 1960s,¹⁰ only a few studies have explored the impact of that dexamethasone on survival of glioblastoma patients.^11–22 Most of these studies report dexamethasone use during chemoradiation compromises survival.

Here, we studied the relationships of survival, steroid dependency, and diagnosis of an infection within 90 days after the diagnostic operation with the cumulative dexamethasone dose prescribed to newly diagnosed glioblastoma patients from postoperative day (POD) 0 to 21. Our focused period from POD0-POD21 is unique for several reasons: (1) The effects of dexamethasone prescribed during this period are unknown, yet virtually all patients receive this potent medication before treatment with radiation and/or temozolomide. (2) Guidelines for postoperative dexamethasone use are not available. (3) Analysis of our data showed by POD21 the least number of patients were on dexamethasone (Figure 1A). (4) Administration of dexamethasone in this initial postoperative period is a modifiable intervention that neurosurgeons could undertake.

Figure 1.

Dexamethasone administration in 360 glioblastoma patients. (A) Proportion of patients on and (B) the median dose with interquartile range (gray shading) of dexamethasone prescribed is plotted against their postoperative day.

Materials and Methods

Study Design

We conducted an observational retrospective analysis on a cohort of newly diagnosed glioblastoma patients treated at a single academic tertiary center.

Primary Study Variable

The primary study variable was the cumulative amount of dexamethasone prescribed to newly diagnosed glioblastoma patients from the day of the initial diagnostic operation (POD0) to POD21. To minimize bias associated with retrospective collection, we retrieved dosage (in mg), route (enteral or intravenous), and the date and time of each dexamethasone administration in 2 ways: (1) electronically from the medication order records that are part of the electronic medical record. We matched these medication order records with nursing medication administration records and only included those confirmed as administered per nursing record for analysis. (2) We retrieved any outpatient receipt of dexamethasone up to POD21 from dexamethasone prescriptions ordered in the electronic medical record.

Our institutional practice was to use dexamethasone as the sole corticosteroid in the treatment and management of glioblastoma patients. Individual treating neurosurgeons decided the amount and schedule of the dexamethasone they prescribed; however, an initial tapering prescription was universal, commonly lasting 1 or 2 weeks.

Patient Selection

Institutional Review Board approved this study (IRB #191181), and a formal patient consent was waived. From a prospectively maintained institutional registry, we identified 453 patients aged 18 years or older with pathologically proven glioblastoma (World Health Organization grade 4). We excluded patients lacking electronically retrievable dexamethasone data (n = 77; all treated before 2007) and overall survival less than 21 days (n = 16). Therefore, we analyzed a final dataset of 360 newly diagnosed glioblastoma patients from 2007 to 2019 who were followed until November 2019.

Study Outcomes

All outcomes were collected blinded to the primary study variable. Primary outcome was overall survival defined as time from the diagnostic operation to death or last follow-up if the patient’s death was not confirmed. Survival was collected prospectively as part of the institutional registry. Secondary outcomes were (1) dexamethasone dependency defined as receipt of a dexamethasone prescription from POD30 to POD90 and (2) diagnosis of an infection within 90 days after the diagnostic operation. We defined infection as microbial growth in tissue sample culture or any condition or clinical test highly suggestive for an infection, for example, wound dehiscence noted on physical examination or chest radiograph concerning for pneumonia. We disregarded the initiation of anti-infective therapy with the intention of prophylaxis as infection.

Co-Variables

We collected the following co-variables from the registry: age at diagnosis, preoperative Karnofsky performance status score (KPS), tumor volume, extent of resection, administration of postoperative temozolomide and radiation treatment, and tumor IDH1/2 mutation and MGMT gene promoter methylation status. We calculated tumor volume using the semi-automated volume rendering function in OsiriX Lite software (version 9.4, Pixmeo, Geneva, Switzerland) after delineating the outer edge of a tumor’s contrast enhancement on the preoperative post-contrast magnetic resonance imaging. In cases of multifocal tumors, we summed their volumes. A neurosurgeon and neuroradiologist independently assessed the extent of resection by evaluating residual tumor enhancement on the immediate (less than 48 hours) post-contrast magnetic resonance imaging. We categorized the extent of resection as biopsy for less than a 50% resection, subtotal for 50%-95% resection, near total for 95%-99% resection or if a neuroradiologist could not exclude a minimal residual, and gross total for the absence of any enhancing tumor.

In addition to these co-variables, we calculated the volume of tumor edema using an unbiased and completely automatic segmentation algorithm called Brain Tumor Image Analysis (BraTumIA)^23–25 utilizing 4 specific MRI sequences of the preoperative MRI: pre-gadolinium T1-weighted, T2-weighted, T2-weighted fluid-attenuated inversion recovery, and post-gadolinium T1-weighted MRI sequences. Of the 360 patients, 166 had a preoperative MRI with the abovementioned 4 sequences in good quality without motion artifact and were successfully segmented. We also electronically retrieved the following laboratory values for all patients: any measurements of total white blood counts, absolute neutrophil and lymphocyte counts made from preoperative day 7 to POD28, and the maximum blood glucose level from preoperative day 7 to POD90.

Statistical Analyses

Continuous and categorical variables were reported as medians with an interquartile range and proportions, and between-group differences in them were tested using nonparametric t tests and chi-square tests, respectively. R software version 4.0 (R Foundation for Statistical Computing, Vienna, Austria) and accompanying statistical R packages were used to conduct all analyses and generate plots. We set the statistical significance for a 2-sided α at ≤0.05 for all general tests. However, because in this study we tested 3 simultaneous outcomes and hypotheses, for their chief analyses, we set the statistical significance for a 2-sided α at ≤0.017 (0.05/3).

The primary analysis was to quantify an association between the cumulative amount of dexamethasone prescribed to glioblastoma patients from POD0 to POD21 and survival using Cox regression, adjusting for the above-listed co-variables. We conducted 3 different Cox regression analyses to handle missing data: (1) restricted to patients with complete data; (2) inclusive of all patients designating missing values as unknown; and (3) imputing the missing values. The latter was done randomly using multivariate imputation by chained equations (mice²⁶ v3.11 and mitools v2.4 R packages), generating 100 complete datasets and pooling their results. We presented the results of Cox regressions with a hazard ratio with 95% confidence intervals (CI) and a P value. When the analyses revealed a significant correlation, we performed a follow-up analysis to identify a “cut point” in the range of total dexamethasone prescribed that was associated with the greatest divergence in survival. This “cut point” was used to dichotomize the cohort into 2 groups. For visualization, we plotted survival data of the 2 groups using right-censored Kaplan-Meier curves, which were censored at the time of last follow-up for patients whose death was not confirmed. We used the survminer v0.4.0 R package for this analysis and plot generation. The 2 groups were then matched to minimize between-group differences in all the co-variables using nearest neighbor matching based on propensity scores (MatchIt²⁷ v4.0 R package). We compared survival curves between groups using the log-rank test.

Secondary analyses were to quantify associations of steroid dependency from POD30 to POD90 and diagnosis of an infection within 90 days with the cumulative dexamethasone prescribed to glioblastoma patients from POD0 to POD21 using logistic regression. We adjusted these outcomes for age, KPS, initiation of radiotherapy, and temozolomide. Additionally, the former outcome was also adjusted for extent of resection and the latter outcome for maximum blood glucose level. Because early death excludes the possibility for these secondary outcomes, we repeated these analyses in a subgroup that excluded patients with an overall survival of less than 90 days. We presented the results of these logistic regressions with an odds ratio with 95% CI and a P value.

Results

Dexamethasone Administration

Table 1 lists the patient characteristics. Of the 360 newly diagnosed glioblastoma patients, 353 (98.1%) and 357 (99.2%) patients received dexamethasone on POD0 and POD1 (Figure 1A). On these days, they received a median of 18 mg [12-24] and 20 mg [16-32] of dexamethasone (Figure 1B), respectively. By POD21, 94 (26.1%) patients were on dexamethasone, receiving a median dose of 6 mg [4-8]. Cumulatively from POD0 to POD21, patients were prescribed a median of 159 mg [109-190] of dexamethasone.

Table 1.

Characteristics of 360 Glioblastoma Patients Analyzed

Patient Characteristic	Value
Age, median [interquartile range], years	62.3 [52.4-69.7]
KPS, median [interquartile range]	80 [70-90]
Temozolomide, n (%)	296 (82.2%)
Radiation, n (%)	322 (89.4%)
Extent of resection, n (%)
Biopsy	51 (14.2%)
Subtotal	177 (49.2%)
Near total	75 (20.8%)
Gross total	57 (15.8%)
IDH1/2 tumor mutational status, n (%)
Wild-type	262/269 (97.4%)
Mutated	7/269 (2.6%)
Unknown	91 (25.3%)
MGMT gene promoter status, n (%)
Methylated	108/252 (42.9%)
Unmethylated	144/252 (57.1%)
Unknown	108 (30.0%)
Preoperative tumor volume, median [interquartile range], mL	31.0 [15.6-52.0]
Total dexamethasone received POD0 to POD21, median [interquartile range], mg	159 [109-190]

Abbreviations: KPS, Karnofsky performance status score; POD, postoperative day.

The cumulative dose of dexamethasone prescribed to patients did not correlate with their age at diagnosis or extent of resection but very weakly correlated with their KPS (Pearson R = −0.16, P < .01) and tumor volume (Pearson R = 0.18, P < .01) (Figure 2). Although these 4 co-variables individually were statistically significant (P < .02) in explaining total dexamethasone administered in a multivariable linear regression model, the adjusted R² statistic of the model was a meager 0.09, indicating only 9% of the variation in the total dexamethasone prescribed was explained by the linear combination of these 4 co-variables. Cumulative dose of dexamethasone did not correlate with tumor edema in the 166 patients who had these values successfully segmented (Figure 2).

Figure 2.

Correlation of total dexamethasone with patient characteristics. Correlation of total dexamethasone prescribed to 360 glioblastoma patients from postoperative day (POD) 0 to 21 with their (A) age, (B) Karnofsky performance status score, (C) tumor volume, (D) tumor edema (n = 166), and (E) extent of resection, classified as biopsy, subtotal (STR), near total (NTR), and gross total (GTR) resection. A correlation line with a 95% confidence interval (gray shading) and Pearson correlation coefficient R and associated P value are noted in the top left corners in A–D. Boxplots representing median and interquartile range are plotted in (E) with a P value from a Kruskal-Wallis test.

Dexamethasone Dependency

We tested whether the total dexamethasone prescribed from POD0 to POD21 associated with being prescribed dexamethasone from POD30 to POD90 (steroid dependency). During this later period, 181 (50.3%) patients were prescribed dexamethasone. The median amount of dexamethasone prescribed from POD0 to POD21 to these 181 patients was significantly higher (in mg: 167 [127-207] vs 151 [104-179]; P < .01; Supplementary Figure1). The logistic regression analysis adjusted for age, KPS, initiation of radiotherapy and temozolomide, and extent of resection showed that the odds of being prescribed dexamethasone from POD30 to POD90 significantly correlated with the total amount of dexamethasone (odds ratio of 1.007 [1.003-1.011] per mg, P < .01). To convert these numerical results for a practical perspective, a day of dexamethasone prescribed 4 mg every 6 hours (16 mg total)—a common dose prescription—was associated with a 12% increase in the odds of being prescribed dexamethasone from POD30 to POD90 (odds ratio of 1.12 [1.06-1.19], P < .01). In this analysis, initiation of radiation was associated with a significant increase (odds ratio of 4.17 [1.34-13.5], P = .01) and extent of resection was associated with a significant decrease in the odds of being prescribed dexamethasone from POD30 to POD90 (odds ratios in reference to biopsy: subtotal 0.29 [0.12-0.61]; near total 0.19 [0.07-0.44]; and gross total 0.15 [0.06-0.37]; all P < .01).

The results were consistent in a subgroup analysis excluding patients whose overall survival was less than 90 days. Of the 320 patients in this subgroup, 163 patients were prescribed dexamethasone from POD30 to POD90, and the total dexamethasone prescribed to them was significantly higher (in mg: 163 [118-201] vs 151 [103-177]; P < .01). The adjusted logistic regression analysis showed that the odds of being prescribed dexamethasone from POD30 to POD90 in this subgroup significantly correlated with the total amount of dexamethasone: every 16 mg increment of total dexamethasone was associated with a 12% increase in the odds of being prescribed dexamethasone from POD30 to POD90 (odds ratio of 1.12 [1.05-1.20], P < .01).

Postoperative Dexamethasone Associates With Lower Survival

The cumulative amount of dexamethasone prescribed to patients from POD0 to POD21 correlated with a lower survival on univariable analysis (hazard ratio of 1.002 [1.000-1.004] per mg, P = .02). Due to missing data on IDH1/2 mutation status and MGMT gene promoter methylation status of tumors for some patients, we performed 3 multivariable analyses adjusting for the co-variables collected: a complete-case analysis excluding patients with an unknown MGMT gene promoter methylation and IDH1/2 status (n = 247), an inclusive analysis incorporating all patients (n = 360), and an analysis using imputed values (n = 360). All approaches revealed a consistent association of the total amount of dexamethasone prescribed to patients from POD0 to POD21 with lower survival: hazard ratio 1.004 [1.001-1.007] per mg, P < .01, for complete-case analysis; hazard ratio 1.003 [1.001-1.004] per mg, P < .01, for inclusive analysis; and hazard ratio 1.002 [1.000-1.004] per mg, P = .06, for imputed analysis (Table 2). In practical terms, a day of dexamethasone prescribed 4 mg every 6 hours (16 mg total) was associated with a 3%-6% increase in mortality.

Table 2.

Results of Multivariable Cox Regression Models

	Complete Cases (n = 247)		All Cases (n = 360)		Imputation (n = 360)a
Variable	Hazard Ratio (95% CI)	P Value	Hazard Ratio (95% CI)	P Value	Hazard Ratio (95% CI)	P Value
Total dexamethasone from POD0 to POD21
16 mg increment	1.06 (1.02-1.11)	<.01	1.04 (1.01-1.07)	<.01	1.03 (1.00-1.06)	.06
1 mg increment	1.004 (1.001-1.007)	<.01	1.003 (1.001-1.004)	<.01	1.002 (1.000-1.004)	.06
Age (1-yr increment)	1.03 (1.01-1.04)	<.01	1.02 (1.01-1.03)	<.01	1.02 (1.01-1.03)	<.01
KPS (10-point increment)	0.93 (0.82-1.05)	.23	0.85 (0.78-0.92)	<.01	0.89 (0.81-0.97)	.01
Temozolomide	0.39 (0.21-0.74)	<.01	0.43 (0.27-0.69)	<.01	0.49 (0.29-0.81)	<.01
Radiotherapy	0.24 (0.11-0.54)	<.01	0.37 (0.21-0.66)	<.01	0.32 (0.16-0.63)	<.01
Tumor volume (10 mL increment)	1.01 (0.96-1.07)	.69	0.97 (0.93-1.01)	.17	0.99 (0.94-1.04)	.61
Extent of resection
Biopsy	Reference		Reference		Reference
Subtotal	0.51 (0.31-0.84)	<.01	0.72 (0.47-1.10)	.13	0.64 (0.41-0.98)	.04
Near total	0.41 (0.22-0.77)	<.01	0.52 (0.32-0.83)	<.01	0.45 (0.28-0.73)	<.01
Gross total	0.27 (0.15-0.50)	<.01	0.36 (0.21-0.60)	<.01	0.34 (0.20-0.59)	<.01
MGMT gene promoter
Unmethylated	Reference		Reference		Reference
Methylated	0.46 (0.32-0.66)	<.01	0.56 (0.40-0.78)	<.01	0.56 (0.36-0.88)	.01
Unknown	—	—	0.57 (0.35-0.93)	.03	—	—
IDH1/2 genotype
Wild-type	Reference		Reference		Reference
Mutated	0.09 (0.01-0.71)	.02	0.10 (0.01-0.78)	.03	0.27 (0.06-1.25)	.10
Unknown	—	—	0.86 (0.52-1.42)	.55	—	—

Abbreviations: CI, confidence interval; KPS, Karnofsky performance status score; POD, postoperative day.

^aImputation of missing values was done randomly using multivariate imputation by chained equations (mice v3.11 and mitools v2.4 R packages), generating 100 complete datasets and pooling their results.

The association of total dexamethasone prescribed with mortality was continuous across the range of its values (Figure 3A). This is also visualized in Kaplan-Meier survival curves of patients grouped based on the quantity of total dexamethasone prescribed (Figure 3B). Median survival of patients who were prescribed (in mg) ≤75, 75-200, 200-300, and >300 decreased from (in days) 441, 421, 298, to 183, respectively (log-rank P = .03).

Figure 3.

The total amount of dexamethasone prescribed from postoperative day (POD) 0 to 21 associates with a lower survival. (A) The relationship of relative hazard ratio adjusted for multiple co-variables with the total amount of dexamethasone prescribed is plotted. Gray shading indicates a 95% confidence interval. (B–F) Kaplan-Meier survival plots with the number at risk table below of patients grouped according to the amount of total dexamethasone prescribed from postoperative days 0 to 21. (D–F) These 3 plots present 3 propensity-matched analyses, where the number of those who were prescribed 200 mg or more of dexamethasone by POD21 (n = 71) was matched increasingly with those who were prescribed less with a ratio of (A) 1:1, (B) 1:2, and (C) 1:3. Plus (+) ticks on the survival curves indicate censored values. The resulting P value from a log-rank test comparing the survival curves is shown underneath the legends.

The single cut-point value of total dexamethasone prescribed that dichotomized patients into groups with the most divergent survival profiles was 206 mg, which we rounded to 200 mg for practical purposes. Patients who were prescribed 200 mg or more of dexamethasone by POD21 (n = 71) had a median survival of about half that of those who were prescribed less (255 vs 421 days; log-rank P < .01; Figure 3C). These patients had lower KPS (70 [70-80] vs 80 [70-90], P = .01), larger tumor volumes (median, mL, 38.1 [21.9-54.7] vs 28.5 [14.0-50.5], P = .02) and proportionally had more biopsies and less gross total resections (Biopsy: 26.8% vs 11.1%; subtotal: 46.5% vs 49.8%; near total: 16.9% vs 21.8%; gross total: 9.9% vs 17.3%; P < .01; Table 3). There was no difference in age (median, years, 61.5 [52.2-68.9] vs 62.4 [52.5-69.7], P = .48) or the proportion of patients initiated on temozolomide (80.3% vs 82.7%, P = .76) and radiation (88.7% vs 89.6%, P < 1.00), suggesting that goals of care were not different between patients who were prescribed more and less than 200 mg of dexamethasone by POD21. In a Cox regression, we adjusted for the differences in all the co-variables between the 2 groups (inclusive analysis). The result was consistent with the univariate analysis (Figure 3C): those who were prescribed 200 mg or more of dexamethasone by POD21 had a 50% increase in mortality compared to those who were prescribed less (hazard ratio 1.52 [1.11-2.09], P < .01).

Table 3.

Differences Between Patients Who Received More and Less Than 200 mg of Total Dexamethasone After Propensity Matching

Total Dexamethasone From POD0 to POD21	≥200 mg	<200 mg
	All Cases (n = 71)	All Cases (n = 289)	P Valuea	1:1 Match (n = 71)	P Valuea	1:2 Match (n = 142)	P Valuea	1:3 Match (n = 213)	P Valuea
Age, median [IQR], years	61.5 [52.2-68.9]	62.4 [52.5-69.7]	.48	61.1 [51.3-67.7]	.63	62.0 [51.9-68.7]	.94	62.1 [51.7-69.3]	.81
KPS, median [IQR]	70 [70-80]	80 [70-90]	.01	80 [60-90]	.26	80 [60-90]	.26	80 [70-90]	.17
Tumor volume, median [IQR], mL	38.1 [21.9-54.7]	28.5 [14.0-50.5]	.02	46.4 [25.0-61.1]	.35	38.1 [17.4-59.8]	.79	36.0 [18.2-57.9]	.50
Temozolomide, n (%)	57 (80.3)	239 (82.7)	.76	59 (83.1)	.83	120 (84.5)	.56	177 (83.1)	.72
Radiation, n (%)	63 (88.7)	259 (89.6)	<1.0	62 (87.3)	<1.0	128 (90.1)	.94	191 (89.7)	<1.0
IDH1/2 tumor status			.45		.60		.80		.97
Mutant, n (%)	1 (1.4)	6 (2.1)		3 (4.2)		4 (2.8)		4 (1.9)
Wild-type, n (%)	48 (67.6)	214 (74.0)		47 (66.2)		93 (65.5)		143 (67.1)
Unknown, n (%)	22 (31.0)	69 (23.9)		21 (29.6)		45 (31.7)		66 (31.0)
MGMT promoter status			.02		.41		.64		.47
Methylated, n (%)	15 (21.1)	93 (32.2)		22 (31.0)		38 (26.8)		55 (25.8)
Unmethylated, n (%)	25 (35.2)	119 (41.2)		22 (31.0)		49 (34.5)		82 (38.5)
Unknown, n (%)	31 (43.7)	77 (26.6)		27 (38.0)		55 (38.7)		76 (35.7)
Extent of resection			.01		.97		.78		.17
Biopsy, n (%)	19 (26.8)	32 (11.1)		19 (26.8)		29 (20.4)		32 (15.0)
Subtotal, n (%)	33 (46.5)	144 (49.8)		32 (45.1)		71 (50.0)		112 (52.6)
Near total, n (%)	12 (16.9)	63 (21.8)		14 (19.7)		26 (18.3)		44 (20.7)
Gross total, n (%)	7 (9.9)	50 (17.3)		6 (8.5)		16 (11.3)		25 (11.7)

Abbreviations: IQR, interquartile range; KPS, Karnofsky performance status score; POD, postoperative day.

^aComparison is made to the ≥200-mg group. Statistical tests used are Wilcoxon for continuous and chi-square for categorical variables.

In an alternate approach, we minimized the above differences in the co-variables using 3 propensity-matched analyses, where the number of patients who were prescribed 200 mg or more of dexamethasone by POD21 (n = 71) was matched increasingly with patients who were prescribed less with a ratio of 1:1, 1:2, and 1:3. In these matched analyses, the 2 groups were similar (Table 3), and the differences in the co-variables were no longer statistically significant. These analyses showed that the survival of those who were prescribed 200 mg or more of dexamethasone by POD21 was lower. The numeric difference in median survival and the effect size of the adjusted hazard ratio obtained from a multivariable analysis (inclusive analysis) remained relatively precise, and the statistical significance increased with the matched sample size. The median survival between the respective groups were (in days and log-rank P value) 255 vs 460, P = .06 (1:1 matching); 255 vs 420, P = .03 (1:2 matching); and 255 vs 410, P = .02 (1:3 matching) (Figure 3D–F). The adjusted hazard ratios were 1.32 [0.89-1.96], P = .16 (1:1 matching); 1.36 [0.98-1.90], P = .07 (1:2 matching); and 1.41 [1.03-1.93], P = .03 (1:3 matching).

Perioperative White Blood Cell Counts

Because dexamethasone is a potent immunosuppressant,^3–5 we explored changes in white blood cell counts of glioblastoma patients from preoperative day 7 to POD28. We chose this period to incorporate preoperative baseline values and any change in them for up to a week after POD21 when most of the patients were off dexamethasone. Measurements of daily white blood cell counts in this period were not available for most of the patients. To increase the power of this analysis, we incorporated patients excluded in the prior analysis for lack of electronically retrievable dexamethasone data and whose survival was less than 21 days.

The median value of the total white blood cell count and absolute neutrophil count of patients increased immediately postoperatively and gradually declined (Supplementary Figure 2). Interestingly, the median value of the absolute lymphocyte count hovered near the lower normal limit of 1100 counts per microliter per our institution’s standard (Supplementary Figure 2), prompting us to determine the proportion of patients with lymphopenia. Of the 256 patients who had one or more measurements of the absolute lymphocyte count from preoperative day 7 to POD28, 141 (55.1%) patients had at least one value indicative of lymphopenia. In the preoperative period, this proportion was 80/175 (45.7%), and in the postoperative period up to POD28, it was 82/167 (49.1%) (Supplementary Figure 2). Given the lack of daily values for these patients, lack of data on preoperative steroid use, and reasons for measurement of these labs, we did not draw any statistical conclusions or correlations. The true value of these proportions may be higher.

Dexamethasone Associates With 90-Day Infection

Infections are more likely to occur in an immunocompromised state. Therefore, we probed the association of a diagnosis of an infection within 90 days after the diagnostic operation with the total amount of dexamethasone prescribed from POD0 to POD21. Seventy-two (20.0%) patients were diagnosed with at least one infection before POD90. The initial infections were 27 urinary tract infections, 20 respiratory infections (16 pneumonias), 8 thrush, 7 surgical site infections, 5 bacteremia, 3 gastrointestinal tract infections, 1 meningitis, and 1 shingles.

The median amount of dexamethasone prescribed from POD0 to POD21 to these 72 patients was significantly higher compared to the remaining patients (in mg: 168 [125-213] vs 156 [107-185]; P < .01; Supplementary Figure 1). The proportion of patients diagnosed with an infection nearly doubled when comparing patients who were prescribed less than 200 mg of dexamethasone (49/289 or 17.0%) by POD21 to those who were prescribed 200 mg or more (23/71 or 32.4%; P < .01).

We conducted a logistic regression analysis adjusted for age, KPS, initiation of radiotherapy and temozolomide, and the maximum blood glucose level recorded from preoperative day 7 to POD90. It showed that the odds of being diagnosed with an infection in our cohort significantly correlated with the total amount of dexamethasone (odds ratio of 1.006 [1.002-1.010] per mg, P < .01). In practical terms, a day of dexamethasone prescribed 4 mg every 6 hours (16 mg total) was associated with a 10% increase in the odds of being diagnosed with an infection (odds ratio of 1.10 [1.04-1.17], P < .01). Another significant variable was the maximum blood glucose level from preoperative day 7 to POD90 (median 212 mg/dL [177-263], minimum 73, maximum 713). A 10-point increase in this value was associated with a 3% increase in the odds of being diagnosed with an infection (odds ratio of 1.03 [1.00-1.06], P = .03).

The results were consistent in a subgroup analysis excluding patients whose overall survival was less than 90 days. Of the 320 patients in this subgroup, 54 patients were diagnosed with an infection, and the total dexamethasone prescribed to them was significantly higher compared to the remaining patients (in mg: 165 [120-208] vs 154 [107-184]; P = .05). The adjusted logistic regression analysis showed that the odds of being diagnosed with an infection in this subgroup significantly correlated with the total amount of dexamethasone: every 16 mg increment of total dexamethasone was associated with an 8% increase in the odds of being diagnosed with an infection (odds ratio of 1.08 [1.01-1.16], P = .02).

The 72 patients with known infections had a lower survival (median survival in days: 220 vs 423, log-rank P < .01). Therefore, we included the diagnosis of an infection in the multivariable survival analysis (inclusive analysis) to adjust for any of its confounding effects on survival with dexamethasone. Both the diagnosis of an infection (adjusted hazard ratio of 1.79 [1.31-2.46], P < .01) and the total amount of dexamethasone prescribed to patients from POD0 to POD21 (adjusted hazard ratio of 1.03 for every 16-mg increment [1.00-1.06], P = .025) remained significant.

Discussion

This single-center retrospective cohort analysis revealed that the cumulative amount of dexamethasone prescribed to 360 newly diagnosed glioblastoma patients in the initial 3 postoperative weeks correlated with lower survival, increased odds of being prescribed dexamethasone up to POD90, and increased odds of infection prior to POD90. These results warrant confirmatory clinical and biological investigations and further explorations of its mechanistic basis because the amount of dexamethasone administered to glioblastoma patients is a modifiable factor.

Immune system dysfunction and/or depletion may permit unchecked cancer growth and invite infections. Patients with glioblastomas are often immunosuppressed or immune-depleted^28,29 or may become so after temozolomide treatment.³⁰ Several recent reports have described lymphopenia in glioblastoma patients during chemoradiation.^31–36 Moreover, lymphopenia can occur in 25%-42% of treatment-naïve glioblastoma patients from sequestration of lymphocytes in the bone marrow, including in dexamethasone-naïve patients.^28,35,36 Our results showing that half of the patients are lymphopenic in the perioperative period (Supplementary Figure 2) agree with these prior reports.

Here, we present 2 hypotheses for the increased mortality associated with dexamethasone use.

First, dexamethasone impacts survival by mediating tumor oncogenicity, treatment resistance, and systemic immunosuppression. Glioma cells may increase their proliferation and migration under dexamethasone.^37,38 Importantly, dexamethasone can cause changes in glioma cell states that can lead to radiation and temozolomide resistance.^39–42 For example, in a murine glioblastoma model, dexamethasone administered daily for 3 days before radiotherapy resulted in a significantly lower survival.¹⁷ Dexamethasone is also a potent immunosuppressant^3–5 that exacerbates the immunosuppressed state in glioblastoma patients,^21,29 possibly allowing unchecked glioblastoma progression and diminishing the efficacy of immunotherapy.^43,44 Patients with an absolute CD4 lymphocyte count less than 200 have significantly lower survival.³⁶ Moreover, the immune system’s ability to infiltrate and combat tumors is increasingly being leveraged in the treatment of cancer. In the presence of dexamethasone, a lower number of immune cells, especially T cells, infiltrate glioblastoma, and the lymphocytes are functionally impaired; hence, checkpoint immunotherapy is less effective with concurrent dexamethasone in both mice and patients with glioblastoma.⁴⁴ The absence or decreased use of dexamethasone may allow the immune system in glioblastoma patients to be better leveraged therapeutically. For example, in a recent neo-antigen vaccine trial, glioblastoma patients who did not receive dexamethasone showed favorable neo-antigen-specific T-cell responses and an increase in T-cell infiltration in glioblastoma.⁴⁵

Second, dexamethasone may lead to a lower survival in glioblastoma patients by worsening a preexisting immunocompromised state, leading to increased infection rate. Although infection by itself may not be an independent predictor of survival in glioblastoma patients, infection is a common cause of death in glioblastoma patients.^46,47 Here, we report a direct association between the total dexamethasone prescribed from POD0 to POD21 and diagnosis of an infection before POD90 in glioblastoma patients. Some of these patients had multiple infections during this period, and in others, infection was present at the time of inpatient death. Our results agree with prior reports that suggest dexamethasone use associates with an increased risk of infections after craniotomies.^48,49

Our analysis suggests that higher cumulative dexamethasone prescribed in the initial 3 postoperative weeks is associated with increased odds of it being prescribed later, specifically from POD30-POD90, a time when patients undergo chemo- and radiation therapy. This steroid dependency may lead to prolonged duration or exacerbation of the harmful effects of dexamethasone. Indeed, concurrent administration of dexamethasone during chemoradiation is associated with lower survival.^16,17 In our cohort, the 181 patients who were prescribed dexamethasone from POD30 to POD90 had a significantly lower survival compared others (median survival in days: 339 vs 468, log-rank P < .01; adjusted hazard ratio 1.50 [1.16-1.94], P < .01).

While these reports and data presented here suggest that dexamethasone may be detrimental to patients with glioblastoma, it has been a primary medication used to control cerebral edema and patient symptoms since the 1960s.¹⁰ Therefore, further clinical investigations are needed to better understand the balance between the risks and benefits of dexamethasone use in glioblastoma patients in the postoperative period.⁵⁰

Limitations

We caution against generalizing our single-center data and accepting the precision of the effect size estimates, for biases inherent to a retrospective study design limit this study. The potential impacts of preoperative dexamethasone use and its use after POD21 are unaccounted for in this analysis because of the increased risk of inaccuracy in outpatient records. We also assumed that patients discharged before POD21 took dexamethasone as prescribed in the outpatient setting and did not transfer care to another center before POD21. Although neurosurgeon preference was the primary reason for the selection of different postoperative dexamethasone tapers, the retrospective study design limited capture of patient-specific reasons for variability in postoperative dexamethasone tapering and any changes in practice patterns over time.

Although we collected our secondary outcomes blindly, they were still collected retrospectively, and as such, are not free from inconsistencies related to loss to follow-up and outpatient documentation. Moreover, because they are not completely objective outcomes, they are not free from bias. The proportions of patients who were prescribed dexamethasone from POD30 to POD90 and had an infection before POD90 are very likely underestimated not only because they were retrospectively captured, but also because of the varying lengths of patient follow-up and patient self-selection in seeking care for the infection.

Preoperative tumor edema volumes were able to be derived in less than half of the patients in our cohort. Complete-case analyses incorporating tumor edema could only be conducted in less than a quarter of the patients in this cohort. Given that dexamethasone is used to manage edema, lack of analysis adjusted for tumor edema is a limitation. Further, any effect from investigational treatments that patients may have received is unaccounted for in this analysis (although a vast majority of the patients in this cohort did not receive investigational treatments within 90 days of diagnosis). Lastly, whether blood glucose levels were truly fasting could not be determined; hence, they could be inaccurate or confounded.

This is a relevant hypothesis-generating study. Hence, a prospective study design is needed to elucidate whether the patients who were prescribed higher doses of dexamethasone constitute a unique subgroup of patients or irrespectively experience increased mortality and risk of infection.

Conclusion

Guidelines for postoperative dexamethasone use are not available. In this retrospective analysis of 360 newly diagnosed glioblastoma patients treated at a single academic center in the United States, lower survival, steroid dependency up to POD90, and diagnosis of an infection before POD90 associated with higher dexamethasone administration in the initial 3 postoperative weeks. Nearly half of the glioblastoma patients are lymphopenic preoperatively and up to 1 month postoperatively. Therefore, neurosurgeons could consider limiting the amount of total dexamethasone prescribed in the immediate postoperative period. This study did not evaluate what postoperative dexamethasone regimen or taper is best. Further, these results warrant confirmatory clinical and biological investigations.

Funding

None received.

Conflict of interest statement. None declared.

Ethical approval. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee (Primary Institutional Review Board #191181) and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. For this type of study, formal consent is not required.

Prior presentation of results. Not applicable.