Evaluating the heterogeneity of hippocampal avoidant whole brain radiotherapy treatment effect: A secondary analysis of NRG CC001

Abstract

Background

Hippocampal avoidant whole brain radiotherapy (HA-WBRT) is the standard of care for patients needing WBRT for brain metastases. This study, using existing data from NRG Oncology CC001 including baseline tumor characteristics and patient-reported MD Anderson Symptom Inventory-Brain Tumor (MDASI-BT) scores, sought to identify subgroups of patients that demonstrate differential neuroprotective treatment response to HA-WBRT.

Methods

An exploratory analysis of NRG CC001, a phase 3 trial in which 518 patients were randomly assigned to WBRT plus memantine or HA-WBRT plus memantine, was performed. Rates of neurocognitive function failure (NCFF) were estimated between subgroups and stratified by arm. Covariate and subgroup interaction with differential treatment response were calculated.

Results

The benefit of HA-WBRT on decreasing NCFF was seen in patients living ≥ 4 months (HR 0.75, 95% CI: 0.58–0.97, P = .03), whereas patients living < 4 months derived no significant neurocognitive benefit. A significant association between baseline MDASI-BT cognitive factor and treatment response (interaction P = .03) was identified. Patients with lower MDASI-BT scores (less patient-reported cognitive impairment) derived significantly greater benefit (HR = 0.64, 95% CI: 0.48–0.85, P = .002) compared to those with highest MDASI-BT scores (HR = 1.24, 95% CI: 0.76–2.04, P = .39). Tumor histology also had a significant interaction (P = .01) with treatment response. Primary lung histology patients derived cognitive failure risk reduction (HR = 0.58, 95% CI: 0.43–0.77, P = .0007) from HA-WBRT, in contrast to nonlung primary histology patients (HR = 1.15, 95% CI: 0.78–1.50, P = .48).

Conclusions

Differential neuroprotective response to HA-WBRT was identified in this analysis. Patients surviving ≥ 4 months derived benefit from HA-WBRT. There is evidence of heterogeneity of treatment effect for patients with less severe patient-reported cognitive impairment at baseline and those with primary lung histology.

Brain metastases remain a major source of morbidity and mortality, with an estimated 20% of all adult patients with cancer developing brain metastases during the course of their disease.¹ Despite a historically high rate of mortality, recent advances in radiotherapy, immunotherapy, targeted therapeutics, and chemotherapy have led to improved patient outcomes.² Recognizing the varied characteristics of brain metastases patients, numerous efforts have been undertaken to accurately predict the differential effects of treatment strategies across patient subgroups. Such efforts not only aim to provide clinicians with valuable insights but also better inform clinical trial design.^3,4

Stereotactic radiosurgery (SRS) remains the favored treatment option for patients with limited brain metastases. However, patients with extensive brain metastases often require whole brain radiotherapy (WBRT), which is associated with a higher likelihood of cognitive dysfunction. The phase 3 clinical trial, NRG CC001, demonstrated that hippocampal avoidance during WBRT (HA-WBRT) using intensity-modulated radiation therapy (IMRT) combined with an NMDA receptor antagonist (memantine) provided a significant neuroprotective effect as compared with standard WBRT.⁵ Consequently, the use of HA and memantine during WBRT has now become the standard of care for brain metastases patients recommended to receive WBRT.^2,6

Given the heterogenous nature of brain metastases patients, practice patterns for administering HA-WBRT remain varied.⁷ In this context, we conducted a post hoc exploratory analysis of NRG CC001. Our objective was to identify subgroups that may derive the greatest neuroprotective benefits of HA-WBRT. The insights derived aim to guide future clinical trials in refining the application of HA-WBRT and also help clinicians better gauge the neuroprotective effects of HA-WBRT treatment.

Materials and Methods

Design and Participants

A post hoc exploratory analysis of NRG CC001, a phase 3 clinical trial in which 518 patients with brain metastases were randomly assigned to WBRT plus memantine or HA-WBRT plus memantine,⁵ was performed (Figure 1). All eligible patients enrolled in this trial from July 2015 through March 2018 were included in this secondary analysis. The study details of the NRG CC001 were reported in the primary outcome manuscript. No additional trial data was collected for this analysis. We obtained data from the National Clinical Trials Network (NCTN)/National Cancer Institute (NCI) Community Oncology Research Program (NCORP) data archive of the NCI’s NCTN. Our institutional review board (IRB) approved the project and confirmed exemption (HP-00099872) as they deemed that this secondary analysis of publicly available data met the definition of Not Human Subjects Research (NHSR) and as such did not require further IRB oversight or informed consent.

The primary endpoint for the trial was time to cognitive failure, which was defined as cognitive decline based on the reliable change index on at least one of the cognitive tests (Hopkins Verbal Learning Test-Revised [HVLT-R], Trail Making Test Part A and B [TMT-A, TMT-B], and Controlled Oral Word Association [COWA]). Secondary endpoints collected also included intracranial progression-free survival (PFS), overall survival (OS), patient-reported outcomes (EQ-5D-5L), symptom burden (MD Anderson Symptom Inventory-Brain Tumor [MDASI-BT] module) and cognitive function (individual tests and the Clinical Trial Battery [CTB] composite score).

Statistical Analysis

In this secondary analysis, the rates of neurocognitive function failure (NCFF) were estimated using cumulative incidence function (CIF). NCFF between arms and subgroups was tested using Gray’s test. Covariate and subgroup association with differential neuroprotective treatment response was calculated using competing-risk regression with the sub-distribution hazard function; these factors were then tested for interaction with the treatment arm and plotted on forest plots. The rates of OS and intracranial PFS were estimated using the Kaplan–Meier method. Comparisons of categorical arms (gender, Karnofsky performance score [KPS], neurologic function status, recursive partitioning analysis [RPA] class, and education level [stratified by highest level obtained ≤high school versus >high school]) were tested using χ² tests. A comparison of cognitive test scores was performed using the t test. Comparison of compliance was assessed using the χ² test. All statistical tests were 2-sided with 95% CI and related P-values; P < .05 was considered to be statistically significant. All analyses were done with SAS 9.4 Software.

Results

The benefit of HA-WBRT versus conventional WBRT with regards to the risk of cognitive failure was seen most robustly for those patients living ≥ 4 months (Gray’s test P = .03; Figure 2A), whereas patients living less than 4 months did not derive statistically significant benefit from HA-WBRT (Gray’s test P = .793; Figure 2B). Two hundred and thirteen out of the 518 (41%) patients enrolled on the trial survived for less than 4 months; overall median survival for the trial was 6.7 months. When assessed as a covariate, OS ≥ 4 months was found to be significantly associated with treatment benefit from HA-WBRT (HR 0.75, 95% CI: 0.58–0.97, P = .03; Supplementary Material).

There was significant heterogeneity of treatment effect based on exploratory subgroup analyses of patient and tumor characteristics (Figure 3). There was a significant association between baseline MDASI-BT cognitive factor scores (interaction test P = .03; Figure 3) and treatment response to HA-WBRT. Baseline characteristics for the MDASI-BT cognitive factor cohorts (less severe vs. most severe cognitive impairment [quartile 1–3 vs. 4]) are listed (Table 1). Baseline cognitive test scores and MDASI-BT factors were all significantly worse in the cohort with more severe patient-reported cognitive impairment (quartile 4). KPS (P = .02) and neurologic function status (P = .002) were also significantly worse in the most severe cognitive impairment group.

Patients with lower baseline MDASI-BT cognitive factor scores derived significantly greater differential treatment response (HR = 0.64, 95% CI: 0.48–0.85, P = .002; Supplementary Material) compared to those with the highest (most severe cognitive impairment) MDASI-BT cognitive factor baseline scores (HR = 1.24, 95% CI: 0.76–2.04, P = .39; Supplementary Material). Cumulative neurocognitive function failure incidence was significantly improved with HA-WBRT in the low baseline MDASI-BT cognitive factor cohort (Gray’s P = .002; Figure 4A) but not in the cohort of patients with severe baseline cognitive impairment (Gray’s P = .39; Figure 4B). The association of treatment effect remained significant for the cohort of patients living greater than 4 months (HR = 0.65, 95% CI: 0.48–0.87, P = .004; Supplementary Material).

Primary tumor histology was also found to be significantly associated with differential response to HA-WBRT (interaction test P = .01). Primary lung histology patients (HR = 0.58, 95% CI: 0.43–0.77, P = .0007; Supplementary Material) derived cognitive failure risk reduction from HA-WBRT, whereas nonlung primary histology patients did not (HR = 1.15, 95% CI: 0.78–1.50, P = .48; Supplementary Material). Cumulative neurocognitive function failure incidence was significantly improved with HA-WBRT in lung histology patients (Gray’s P = .001; Figure 5A) but not in the cohort of nonlung cancer patients (P = .48; Figure 5B). This differential response to HA-WBRT persisted in the patients with lung cancer that lived for greater than 4 months (HR = 0.71, 95% CI: 0.51–0.98, P = .036; Supplementary Material), similar to the interaction with MDASI-BT cognitive scores for longer-term survivors. Lung histology was also found to be correlated with cumulative incidence of neurocognitive failure (Gray’s P = .048), with primary lung histology patients having a higher cumulative incidence (Supplementary Material), irrespective of treatment arm.

Baseline characteristics for the primary lung and nonlung cohorts are listed (Table 2). There were more patients enrolled on the trial that had lung cancer (59.3%, 307) versus patients with a different primary histology (40.7%, 211). 45.5% (96) of the nonlung primary histology were breast cancer patients, 10.4% (22) were skin primaries, and 44.0% (93) were other histologies. The primary lung cohort was older than the nonlung cohort (median age 63.7 vs. 57.9 years, P < 0.001). With regards to the assignment of the treatment group, 49.8% of the nonlung primary cohort received HA-WBRT compared to 50.8% of the primary lung cohort. Due to breast primary patients, the nonlung histology subgroup had a higher proportion of female patients (P < .001), whereas gender was balanced in the lung histology subgroup. Baseline neurologic function and KPS were similar though when stratified by KPS 70–80 versus 90–100, there was a difference (P = .023). Education was significantly different between the groups; lung cancer patients enrolled on the trial tended to have less total education with 52% having ≤high school education compared to 35% of patients with nonlung primary cancers (P = .0003). In addition, patients in the lung primary cohort had significantly lower baseline HVLT-R total recall (P = .01), delayed recall (P = .008), and delayed recognition test scores (P = .02), but similar TMT, CTB as well as MDASI-BT factor scores.

Compliance with neurocognitive testing between lung and nonlung cohorts was also assessed; HVLT-R was the neurocognitive test chosen for testing compliance. There was no difference in 2-, 4-, or 6-month compliance between the 2 cohorts. However, when assessed by arm, there was a statistically significant difference in 6-month compliance in the HA-WBRT arm when stratified by lung versus nonlung (χ², P = .026) patients, with higher compliance in the nonlung cohort (Supplementary Material).

The median OS estimates of the lung and nonlung primary histology cohorts were 7.1 (95% CI: 5.8–8.5) and 5.8 (95% CI: 4.1–7.5) months, respectively (log-rank P = .50; Supplementary Material). In the patients who received WBRT, borderline nonsignificant differences in OS were observed between the lung histology (median OS 9.8 months, 95% CI: 6.9–12.7) and nonlung histology patients (median OS 5.5 months, 95% CI: 3.1–7.9) (P = .055; Supplementary Material). No such differences were noted in patients who received HA-WBRT; the median OS for the same groups was 6.1 and 5.9 months, respectively (P = .40; Supplementary Material). Intracranial progression-free survival was not different based on histology (P = .22; Supplementary Material).

For the lung patients, there was a statistically significant difference in median OS by treatment arm (9.8 months, 95% CI: 6.9–12.7 in the WBRT arm vs. 6.1 months, 95% CI: 5.1–7.2 in the HA-WBRT arm, P = .036; Supplementary Material). There were no significant differences between the cohort of lung cancer patients stratified by treatment arm with the exception of KPS (P = .04, Supplementary Material) in the HA-WBRT arm. There was a nonsignificant increase in the proportion of lung cancer patients with a KPS of 70–80 who were treated with HA-WBRT rather than WBRT (57.7% vs. 49.7%, P = .16). There was a significant difference in OST between the lung histology patients with KPS 70–80 vs. 90–100 (P < .01, Supplementary Material). There was no significant difference in peri-hippocampal failure between lung cancer patients treated with HA-WBRT versus WBRT (3.0 vs. 6.6%, P = 0.17).

Discussion

NRG CC001 was the first randomized clinical trial to demonstrate that HA-WBRT with memantine helped to preserve neurocognitive function and patient-reported outcomes, while showing no differences in toxicity, intracranial progression-free survival or overall survival in a large cohort of patients. Confirmatory results of the benefit of hippocampal avoidance have been demonstrated in subsequent prospective trials.^8,9 This post hoc secondary analysis of publicly available trial data from NRG CC001 sought to explore more nuanced patient and tumor characteristics that may help identify heterogeneity of treatment effect to validate in future clinical trials that could guide clinical decision-making.

This analysis suggests that the neurocognitive protective benefit of hippocampal avoidance was most robustly observed in the cohort of patients living greater than or equal to 4 months. Patients who lived less than 4 months did not have any significant benefit with HA-WBRT. This observation is similar to the data reported previously.^10,11 The mechanism of neuroprotection does not change with predicted longevity, and therefore one explanation for this could be that in patients with very short survival, the disease burden and the impact of any management (including supportive care medications) may preclude identification of cognitive benefit. Moreover, cognitive decline post-WBRT likely requires a window of time to manifest, and thus some patients may not live long enough to reap these neurocognitive protective benefits. It is reasonable to suggest that patients with poor expected overall survival would not significantly benefit from hippocampal avoidance added to WBRT given these analyses. Of note, there were no specific inclusion criteria for the trial relating to the life expectancy of participants, and in the oncology literature, the ability of practicing clinicians to accurately predict survival has been demonstrated to be challenging.^12–14 Future efforts should be undertaken to further prognosticate individual life expectancies. Disease-specific graded prognostic assessments can be helpful, but given the heterogeneous nature of patients with brain metastases, prognostication remains challenging and may become even more complex as newer immunomodulatory and targeted systemic agents are incorporated into clinical practice.

Additionally, this analysis offers preliminary evidence suggesting that the effectiveness of HA-WBRT treatment varies based on specific tumor and patient characteristics, which can be explored in future clinical trials. We systematically analyzed patient characteristics and found that there was significant heterogeneity of treatment effect when stratifying by baseline patient-reported cognitive scores, specifically the MDASI-BT cognitive factor score. The MDASI-BT module has been validated in patients with primary brain tumors and brain metastases.^15,16 There are 6 underlying constructs, 1 of which is a cognitive construct; this consists of difficulty in understanding, speaking, remembering, and concentrating. We chose to focus our analyses on this specific metric because of its overlap with domains of neurocognitive function.

Our data demonstrate that patients with less severe cognitive impairment at trial enrollment had greater benefits from HA-WBRT as compared to those with more severe symptoms at baseline. Although not readily studied in the brain metastases literature, the concept of cognitive reserve has long been postulated, first in Alzheimer’s research, to explain individual differences in coping with brain damage.¹⁷ The model of cognitive reserve predicts that there will be differences in the amount of pathologic damage to the brain required for the expression of clinical symptoms. Greater cognitive reserve directly relates to certain patients being better able to use or recruit alternative cognitive strategies and networks when dealing with neuropathological stress. When translated to brain metastases and patients’ responses to radiation (stress), it is possible to consider that patients with better underlying cognitive reserve may therefore be better able to delay the onset of symptomatology (neurocognitive impairment). If these patients go on to receive HA-WBRT, they may indeed have an even greater chance of preventing neurocognitive decline as compared to their counterparts who have lower cognitive reserve at baseline. Baseline cognitive testing may be an appropriate recommendation for patients needing WBRT.

The results of our post hoc analysis also highlight that there may be variability in treatment effects based on primary tumor histology. Specifically, patients with a primary lung histology in this study appeared to derive a greater benefit from HA-WBRT compared to those with a nonlung histology; the test for interaction between treatment arm and tumor histology was significant. This trend was consistent across the entire patient cohort, as well as those with a median survival time exceeding 4 months. The root cause of these findings is not entirely clear due to the post hoc nature of our analysis, but we did identify some differences between the 2 cohorts. While RPA class and KPS were balanced, there were significant differences in age, highest level of education, and baseline HVLT score between the 2 histology subgroups. Moreover, lung cancer patients were found to have a higher cumulative incidence of neurocognitive failure, as well as a lower HVLT-R compliance rate at 6 months when compared to the nonlung cohort. These imbalances may have contributed to heterogeneity of treatment effect observed based on histology. As such, these findings need to be approached with caution before influencing clinical practice.

Interestingly, we observed that lung cancer patients receiving WBRT experienced a longer median survival time compared to lung cancer patients who received HA-WBRT, 9.8 versus 6.1 months, respectively. This result was surprising as there is no clear rationale for HA-WBRT directly leading to decreased survival, especially given similar peri-hippocampal failure rates between the lung HA-WBRT and WBRT cohorts. While lung cancer patient cohorts, stratified by treatment arm, were largely balanced concerning known prognostic variables such as RPA class and age, potential unmeasured confounders in this heterogeneous patient population may have influenced outcomes. Moreover, the survival curves (Supplementary Material) between the treatment arms start to separate as early as 1 month, which may speak to these undetected intergroup differences. We did note a slightly higher proportion of patients with a KPS of 70–80 in the HA-WBRT arm (57.7% vs. 49.7%, P = .16). Although this difference was not significantly different, median survival times based on KPS (70–80 vs. 90–100) were significantly different in the lung cohort and could account for observed survival differences.

To address survival differences, the primary analysis of NRG CC001 as well as the post hoc analysis presented here considered death as a competing event for neurocognitive failure. Given that the greatest benefit of HA-WBRT regarding neurocognitive failure was most pronounced in the lung cohort—that also experienced a higher competing risk of death—it is conceivable that the advantages of HA-WBRT in reducing cognitive dysfunction might have been even more evident if the lung HA-WBRT and WBRT arms had comparable survival times.

Recent advances in immunotherapy and targeted agents for lung cancer patients emerged around the time when NRG CC001 was conducted. Unfortunately, the trial did not capture data on the use of immunotherapy, targeted oral agents, and chemotherapy, either before or after radiation therapy. Consequently, this post hoc analysis could not evaluate the cumulative effects of different systemic agents with WBRT or determine how they might have influenced the observed differences in this study. Considering that targeted agents and immunotherapy may be better tolerated than traditional cytotoxic systemic therapies, potentially as relating to neurocognitive side effects,¹⁸ future WBRT studies and those trials centering on evaluating neurocognitive outcomes should prioritize collecting this increasingly important data.

Our present study has limitations, the most noteworthy being that this was a post hoc analysis. This was an exploratory subgroup analysis examining potential heterogeneity of treatment effect among covariates, and thus, these subgroups were not prespecified. These contrasts with confirmatory subgroup analyses in which there is prespecification about subgroup effects and endpoints.^19,20 As such, it remains uncertain whether or not the nonlung histology or the high baseline cognitive impairment (MDASI) subgroups do not benefit at all from the neuroprotective effects of hippocampal sparing radiation techniques. The trial itself was not powered to identify such subsets. Future clinical trials should focus on better defining the role of HA-WBRT in these subgroups such that further personalization of care can be achieved.

Importantly, this is one of the first studies to use publicly available cooperative trial data from the NCTN/NCROP data archive of NCI’s National Clinical Trials Network for purposes of exploratory analyses. This may lead to more use of available existing trial data for the purposes of identifying and studying new clinical questions that may be tested in the future.

In conclusion, this work represents one of the first comprehensive efforts to evaluate the heterogeneity of the neuroprotective treatment effect of HA-WBRT, drawing from the NRG CC001 clinical trial data. Patients with a survival exceeding 4 months, milder cognitive impairment at baseline (as gauged by the MDASI-BT cognitive factor), and primary lung histology significantly benefitted from hippocampal avoidance during WBRT with memantine. The insights from these analyses hold potential value for shaping future HA-WBRT clinical trials.

Funding

This work was supported by the National Cancer Institute [grant number P30CA134274].

Acknowledgments

This manuscript was prepared using data from Datasets [NCT02360215-D1 and NCT02360215-D2] from the NCTN/NCORP Data Archive of the National Cancer Institute’s (NCI’s) National Clinical Trials Network (NCTN). Data were originally collected from clinical trial NCT02360215, NRG CC001. All analyses and conclusions in this manuscript are the sole responsibility of the authors and do not necessarily reflect the opinions or views of the clinical trial investigators not included as coauthors on this manuscript, NRG Oncology, the NCTN, the NCORP, or the NCI.

Conflict of interest statement

V.G. discloses honoraria from UpToDate. P.D.B. discloses an UpToDate contribution outside the submitted work. M.V.M reported receiving personal fees from Humanetics outside the submitted work. M.P.M. discloses consulting fees from Telia, Kazia, Novocure, Zap, Karyopharm, and Sapience, leadership role on the advisory board of Mevion and Board of Directors of Oncoceutics, Xcision (unremunerated), and stock in Chimerix. All other authors: none declared.

Authorship statement

Study conception and design: H.R.C, M.V.M. Data analyses: K.S., S.B., H.R.C., M.V.M. Data interpretation: all authors. Writing of manuscript: H.R.C., M.V.M. Contribution to and approval of manuscript: all authors.

Data availability

All data will be made available per the National Clinical Trials Network Data Archive rules. The link for the archive is https://nctn-data-archive.nci.nih.gov/.

Previously presented in part at American Society for Radiation Oncology (ASTRO) Annual Meeting 2022, Oral Presentation CNS Scientific Symposium, October 2022, San Antonio, Texas

Previously presented in part at the 2022 Annual Meeting of the American Society for Radiation Oncology in oral presentation format.

References