The relationship between repeat resection and overall survival in patients with glioblastoma: a time-dependent analysis

Abstract

OBJECTIVEPrevious studies assessed the relationship between repeat resection and overall survival (OS) in patients with glioblastoma, but ignoring the timing of repeat resection may have led to biased conclusions. Statistical methods that take time into account are well established and applied consistently in other medical fields. The goal of this study was to illustrate the change in the effect of repeat resection on OS in patients with glioblastoma once timing of resection is incorporated.METHODSThe authors conducted a retrospective study of patients initially diagnosed with glioblastoma between January 2005 and December 2014 who were treated at Memorial Sloan Kettering Cancer Center. Patients underwent at least 1 craniotomy with both pre- and postoperative MRI data available. The effect of repeat resection on OS was assessed with time-dependent extended Cox regression controlling for extent of resection, initial Karnofsky Performance Scale score, sex, age, multifocal status, eloquent status, and postoperative treatment.RESULTSEighty-nine (55%) of 163 patients underwent repeat resection with a median time between resections of 7.7 months (range 0.5-50.8 months). Median OS was 18.8 months (95% confidence interval [CI] 16.3-20.5 months) from initial resection. When timing of repeat resection was ignored, repeat resection was associated with a lower risk of death (hazard ratio [HR] 0.62, 95% CI 0.43-0.90, p = 0.01); however, when timing was taken into account, repeat resection was associated with a higher risk of death (HR 2.19, 95% CI 1.47-3.28, p < 0.001).CONCLUSIONSIn this study, accounting for timing of repeat resection reversed its protective effect on OS, suggesting repeat resection may not benefit OS in all patients. These findings establish a foundation for future work by accounting for timing of repeat resection using time-dependent methods in the evaluation of repeat resection on OS. Additional recommendations include improved data capture that includes mutational data, development of algorithms for determining eligibility for repeat resection, more rigorous statistical analyses, and the assessment of additional benefits of repeat resection, such as reduction of symptom burden and enhanced quality of life.

Keywords: CI = confidence interval; EOR = extent of resection; HR = hazard ratio; HSRT = hypofractionated stereotactic radiation therapy; KM = Kaplan-Meier; KPS = Karnofsky Performance Scale; OS = overall survival; glioblastoma; oncology; proportional hazards model; repeat surgery; survival analysis; time-dependent model.

Conflict of interest statement

Disclosures and Conflicts of Interest

None

Figures

Figure 1.. Flow chart for Patient Sample Selection.

Our initial sample consisted of 509 patients and patients were excluded if they had an outside resection (N=314), if tumor volume was unavailable (N=4), if they had a prior low grade glioma (N=27) or if they were incorrectly treated for a presumed metastasis that turned out to be GBM (N=1). The most common reason for exclusion was outside resection.

Figure 2.

(a) Simon and Makuch Modified Kaplan Meier Plot. In this figure, patients moved from “no resection” to “repeat resection”, if and when their repeat resection occurred. Survival probability was lower for patients who underwent repeat resection compared to those who did not. This figure is an illustrative tool for the formal time-dependent analysis, and cannot be used to make inference. (b) Traditional Kaplan Meier plot. This figure illustrates the relationship between repeat resection and OS if we were to ignore timing of repeat resection and treat repeat resection as a fixed effect.

Figure 3.. Kaplan Meier Survival Plots Landmarked at 8 and 14 months.

Plots were drawn from 8 months (a) and 14 months (b) from initial surgery, which were the approximate 50th and 75th percentile for time between 1st and 2nd resection. Patients were considered to have repeat resection if their repeat resection occurred by that time-point, and were considered to not have repeat resection if they did not have a repeat resection or if their repeat resection occurred after the specified time period (either 8 (a) or 14 months (b) from initial surgery). Patients who died before 8 (a) and 14 months (b) were excluded from these plots as per landmark methodology Figure 3a. Patients who had a repeat resection by 8 months (N=43) had a median OS of 8.1 months (95%CI: 5.1–11.0) compared with 13.9 months (95%CI: 11.8–17.7) for patients who did not undergo a repeat resection by 8 months (N=94). Figure 3b. Patients who had a repeat resection by 14 months (N=41) had a median OS of 9.0 months (95%CI: 4.7–11.7) compared with a median of 12.0 months (95%CI: 7.9–15.9) for those who did not undergo a repeat resection by 14 months (N=58). This figure is an illustrative tool for the formal time-dependent analysis, and cannot be used to make inference.