A randomised trial to compare cognitive outcome after gamma knife radiosurgery versus whole brain radiation therapy in patients with multiple brain metastases: research protocol CAR-study B

CAR-Study B aims to assess, in a randomised design, change in cognitive performance after treatment with either GKRS or WBRT in patients with multiple (11–20) BM.

The primary objective is to determine the between-group difference in the percentages of patients with significant cognitive decline at 3 months after treatment as assessed by the Hopkins Verbal Learning Test-Revised (a memory task). The primary hypothesis is that the percentage of patients with reliable cognitive decline at 3 months will be significantly higher after treatment with WBRT in comparison to GKRS, in patients with 11–20 newly diagnosed BM.

CAR-Study B is a two-arm randomised trial. Adult cancer patients (n = 46), with 11–20 BM, Karnofsky Performance Status (KPS) ≥ 70 and a life expectancy of at least 3 months, are screened for inclusion and exclusion criteria (Table 1) by the radiation-oncologist. Eligible patients are invited for study participation at their first visit at the Gamma Knife Centre. During this first consultation, patients receive an information letter about the study and its procedures.

After signing a written informed consent statement, co-signed by the principle investigator or a formally delegated authorized person, a baseline neuropsychological assessment (NPA) is performed. Subsequently, patients are randomised by the method of minimisation 1:1 to either GKRS (n = 23) or WBRT (n = 23). The trial schema and randomisation factors are shown in Fig. 1. The trial has been approved by the local medical ethics review committee (METC Brabant, The Netherlands). Patients from both arms are followed up at 3, 6, 9, 12 and 15 months after treatment. High rates of attrition and noncompliance are very common in trials in patients with metastatic disease [14, 41]. In an attempt to maximize patient comfort and convenience, the administration of the test battery and additional questionnaires is combined with usual care clinical visits on site (3-monthly contrast MRI-scans and consult with the radiation-oncologist).

In both groups, chemotherapy is administered at the discretion of the primary physician and recorded by the research team. Type and duration of systemic therapy, use of steroids and other medication are accurately monitored and registered. Treatment side effects for both arms are recorded according to the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE version 4). Patients in both treatment arms may receive additional GKRS or WBRT, or salvage surgery when recurrences occur at any one of successive follow-ups; these additional treatments are recorded.

Patients who meet the inclusion and exclusion criteria (Table 1) are eligible for the study. It is projected to include 46 patients.

Gamma Knife Centre Tilburg, Department of Neurosurgery, Elisabeth-TweeSteden hospital, The Netherlands.

A software package (ALEA®) is used to support the online patient registration and randomisation, which is based on the minimization method [47]. Groups are balanced on various prognostic factors. This method has been proven to provide more balanced groups in smaller trials when compared with both restricted (stratified) and unrestricted (simple) randomisation, and is able to incorporate more prognostic factors [47‐49]. The Dutch Cancer Institute provides access to the online minimization program [50].

Eligible patients are assigned in 1:1 to either GKRS or WBRT. Prognostic factors included in the minimization algorithm are:

The Bayesian power analysis and interim analyses are based on the randomised trial by Chang and colleagues [45]. An independent statistician will do interim monitoring of this trial using Bayesian statistical methods [51, 52]. Each patient’s HVLT-R total recall score recorded at 3 months is assigned a binary outcome: A decline in the total recall score of 5 points or greater compared with baseline will be considered a failure (0). A stable or improved score, or a decline of 4 points or less compared with baseline will be considered a success (1). The failure rate for treatment k is designated qk. The prior failure rates for both treatment groups will be modelled as Beta(2.09, 2.91)-distributions, with a mean of 0.42 for both groups (for details see Appendix). During the trial, stopping rules specify that in the case of a probability greater than 0.975 for the event that the failure rate of one treatment group is higher than the failure rate of the other treatment group, we will stop randomising patients to that treatment-arm. In this case, the study is terminated prematurely and the central research question will be answered. If the effect sizes are comparable to earlier accounts in the literature (following Chang et al. an effect size of 0.30 is expected), the early stopping rule will likely come into effect when 46 patients are enrolled (23 patients at 3-months follow-up in both groups; for details see Appendix).

Group analyses are carried out on an intent-to-treat principle. Raw cognitive test scores are compared with published normative values according to age (and, if available, to education) and converted into standardized scores. Cognitive impairment is defined as test performance at or below − 1.5 SD from the normative mean [6, 53]. Reliable change indices (RCI), reflecting change at the individual level in the context of observed changes based on published normative data, correcting for measurement errors are calculated, since group results may mask the variability in individual responses to the intervention [44]. Number of patients, who have improved versus the number of patients who remained stable, or declined, will be counted for all follow-up assessments. These will be compared over conditions with chi-square tests.

Repeated measures analysis of variance with adjustment for potential confounders will be used, comparing subsequent follow-ups to baseline to assess cognitive change of group means over time and across treatment arms. These analyses are similar to those of the study of Chang et al. in which an identical cognitive endpoint was formulated [45].

Missing data, if not too many, will be explicitly or implicitly (dependent on the statistical technique of choice) imputed to facilitate intention-to-treat analysis. Multiple imputation may be used for explicit imputation of missing values. Alternatively, we may use linear mixed models that implicitly deal with missing data under the assumption of missing at random.

Type and duration of systemic therapy and medication use will be taken into account if necessary.

In case of new intracranial tumour activity, patients in both treatment arms may receive additional WBRT or GKRS at the discretion of the treating radiation-oncologist.

A copy of the current study protocol can be requested from Karin Gehring, PhD.