This retrospective data evaluation was approved by the institutional review board of BLINDED (approval number: S2022-1683-0001), and the requirement for written informed consent was waived.
Study population
Our institution’s database was retrospectively reviewed to identify consecutive patients confirmed with glioblastoma and who underwent MRI evaluation after surgery between January 2013 and August 2021. Patients were included if they: (i) had been histologically diagnosed with grade 4 IDH wild-type glioblastoma according to the 2021 World Health Organization (WHO) criteria for Central Nervous System (CNS) Tumors [8]; (ii) underwent CCRT after surgical resection; (iii) underwent evaluation with a brain tumor MRI protocol including both conventional and advanced MRI sequences (diffusion-weighted imaging [DWI] and dynamic susceptibility contrast imaging) before CCRT and within the first 12 weeks of completion of CCRT; (iv) had a measurable enlarging enhancing lesion on postcontrast T1-weighted images; and (V) had available information on O6-methylguanine-DNA methyltransferase (MGMT) promotor methylation status. Of 559 potentially eligible patients, 271 patients without an enlarging measurable enhancing lesion, 89 patients with inadequate MRI quality, and 31 patients without available MGMT methylation status information were excluded. A total of 168 consecutive patients met the above criteria and were enrolled. A flowchart of patient inclusion is shown in Fig. 1.
Imaging protocols
The brain tumor imaging protocol was acquired on a 3-T scanner (Ingenia 3.0 CX; Philips Healthcare, Best, the Netherlands) with a 16-channel head coil, and included T2-weighted, T2-weighted FLAIR, and pre and postcontrast T1-weighted images. The parameters for these sequences are described in Supplementary Materials 1.
Step-wise image analysis - Both conventional and advanced MRI sequences (DWI, dynamic susceptibility contrast imaging) were analyzed in a step-wise manner on both conventional and advanced sequences (DWI, DSC) by three neuroradiologists (---, ---, and ---, with 2, 8, and 23 years of experience in neuro-oncologic imaging, respectively) who were blinded to the clinical information.
Definition of contrast-enhancing mass - One radiologist (--- with 2 years of experience in neuro-oncologic imaging) reviewed the MRI examinations of all patients (n = 559) on a picture archiving and communication system. The definition of contrast-enhancing tumor was adopted from the Response Assessment in Neuro-Oncology (RANO) criteria [9] using measures of the product of the two largest cross-sectional diameters. The presence of contrast-enhancing tumor and the subsequent inclusion criterion was a newly appeared or enlarging (>25%) measurable contrast-enhancing mass greater than 1 × 1 cm raising clinical suspicion of tumor progression or treatment-related change.
Definition of positive findings on advanced MRI - Three neuroradiologists, who were blinded to any clinical information, assessed all the advanced MRI examinations in a random order on a picture archiving and communication system. For each contrast-enhancing mass, the readers assessed the corresponding area on FLAIR images, DWI, ADC maps, and rCBV maps using predefined subjective and objective imaging features (Supplementary Table 1). The imaging features included the presence of infiltrative T2 hypointense lesion and restricted diffusion, and/or a hot spot on rCBV maps. A classification of positive advanced imaging (first progression) was made when a contrast-enhancing lesion showed restricted diffusion and increased rCBV (hot spot foci in color-coded rCBV map) [10]. A low change in ADC and high change in rCBV between pre- and post-CCRT imaging also classified as positive advanced imaging [3].
A classification of negative advanced imaging (stable or response) was made when a contrast-enhancing lesion did not show restricted diffusion or increased rCBV (Figure 2, Supplementary Figure 1).
A consensus read was obtained from the three blinded readers and was defined as the collation of the independent reviews of the three readers on an MRI feature to help diagnose progression.
Definition of covariates and outcome
Clinical, radiologic, and pathologic data were retrieved from medical records. Baseline characteristics including sex, age, type of first surgery, Karnofsky performance score (binary, score >70 or ≤70) at diagnosis, MGMT methylation status, and second-line treatment were collected. The patients were followed up with regular consecutive MRI examinations at 2–3 month intervals.
The primary outcome of interest was OS, which was calculated from the day of histopathologic diagnosis until the day of death, as obtained from the national health care data linked to our hospital. Patients were censored at the date of medical record abstraction or the date of last imaging report, whichever came first.
Statistical analysis
All statistical analyses were performed using R software (version 4.0.2, R Foundation for Statistical Computing, Vienna, Austria) by an expert statistician (--- with 15 years of experience in statistics). P values of <.05 were considered statistically significant.
All results are reported as median with range or 95% confidence interval (CI) for continuous variables, and as frequencies or percentages for categorical variables. Differences in clinicopathological factors between positive and negative findings on advanced MRI were assessed using the chi-square test for discrete variables and Student’s t-test for continuous variables.
When positive MRI findings were observed on advanced MRI, patients were either continued on adjuvant TMZ or discontinued on adjuvant TMZ and switched to second-line treatment (i.e., surgery, antiangiogenic therapy, or clinical trial). Because patients were not randomized to a chemotherapeutic regimen, they were matched on the basis of their propensity score for continuing adjuvant TMZ, considering the tendency to pseudoprogression in the randomization. The propensity score is the conditional probability of receiving an exposure given a vector of measured covariates [11]. To address confounding factors, both propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) were applied [12]. The propensity for continuing adjuvant TMZ was estimated for each patient using the following baseline factors: age (<50 years, ≥50 years), type of first surgery (gross total resection, subtotal resection, or biopsy), KPS at diagnosis (<90, ≥90), and MGMT methylation status (methylated, unmethylated) [13]. For PSM, 1:1 matching was performed without replacement using a caliper width of 0.25 times the standard deviation of the logit of the propensity scores. For IPTW, stabilized weights were used [14]. The baseline characteristics in the two groups were regarded as balanced when the standardized mean differences were within ± 0.1. In the PSM analysis, a marginal Cox model with standard error, considering the matching pair as a clustering effect, was applied [15]. In the IPTW analysis, survival between the two groups was compared using a weighted Cox model with standard errors, accounting for the weighted nature of the participants.
The Kaplan–Meier method was used to draw survival curves for OS in all patients, PSM patients, and after IPTW analysis, and the log-rank test was used to compare curves.
Univariate and multivariate Cox proportional hazards regression analyses were performed on PSM patients to evaluate whether switching treatment regimen and other covariates were associated with OS.