Background: Glioblastoma (GBM) is the most frequent intraaxial malignant brain tumour, in which recurrence management is a frequent and demanding issue. Recently, reintervention has emerged as a useful tool for treatment. However, some new evidence has shown that most of the articles published could have overestimated its effects. We aimed to analyse the effect on survival of reintervention considering it as a time-dependent variable and to compare it with classic statistical analysis.
Methods: We performed a retrospective study with GBM patients between 2007 and 2017. We compared the overall survival (OS) between reintervention and non-reintervention groups with time-dependent statistical methods (Simon-Makuch and landmarking methods and time-dependent multivariable Cox analysis) and compared them with those obtained with non-dependent time variable analysis.
Results: A total of 183 patients were included in the analysis and 44 of them were reoperated. The standard analysis with Kaplan-Meier and multivariable Cox regression of the cohort showed an OS of 22.2 months (95% CI 12.56-16.06) in the reintervention group and 11.8 months (95% CI 9.87-13.67) in the non-reintervention group (p < .001); and an HR 0.649 (95% CI 0.434-0.97 p = .035) for reintervention, demonstrating an increase in OS. However, time-dependent analysis with the Simon-Makuch test and the landmarking method showed that the relationship was not consistent, as this increase in OS was not significant. Moreover, time-dependent multivariable Cox analysis did not show that reintervention improved OS in our cohort (HR 0.997 95% CI 0.976-1.018 p = 0.75).
Conclusions: There has been a temporal bias in the literature that has led to an overestimation of the positive effect of reintervention in recurrent GBM. However, reintervention could still be useful in some selected patients, who should be individualized according to prognostic factors related to the patient, biology of the tumour, and characteristics of surgical procedure.
Keywords: Glioblastoma; Recurrent glioblastoma; Reintervention; Survival; Temporal bias; Time-dependent statistical analysis.