Hippocampal Avoidance During Prophylactic Cranial Irradiation for Patients With Small Cell Lung Cancer: Randomized Phase II/III Trial NRG-CC003.

METHODS

STUDY DESIGN
The study was a multicenter seamless randomized phase II/III trial led by NRG Oncology with collaboration from Alliance and ECOG.

TRIAL PATIENTS
Complete eligibility criteria are provided in the trial protocol (eProtocol) and included adult patients (≥18 years of age) with pathologic proven SCLC without brain metastases (by protocol-specified thin-sliced volumetric MRI with gadolinium contrast), Zubrod Performance 0–2, response to chemotherapy (± thoracic radiotherapy), and English- or French-speaking ability. Exclusion criteria included Hopkins Verbal Learning Test-Revised Delayed Recall (HVLT-R DR) raw score of 2 or less, radiographic evidence of brain metastases, hydrocephalus or other architectural distortion of the ventricular system, planned cytotoxic chemotherapy during PCI, or prior radiotherapy to the head and neck. Institutional review board approval and patient informed consent were required.

RANDOMIZATION AND MASKING
Patients were stratified according to stage (limited vs. extensive), age (<60 vs. older), and planned concurrent memantine usage (yes vs. no) and randomly assigned, using a permuted block procedure, to PCI or HA-PCI.

PROCEDURES
For both study arms, the prescribed PCI dose was 25 Gy in 10 fractions. Hippocampal contouring and HA-PCI planning directives have been described (Table 1; eProtocol).11,19 Before enrolling patients, all participating sites completed a credentialing exercise in which they would generate an HA-PCI IMRT treatment plan on a sample case to be reviewed centrally. For an investigator’s first HA-PCI case, rapid central review of hippocampal contours and HA-PCI planning was conducted in real-time before treatment initiation. If the initial plan was deemed acceptable, subsequent plans were reviewed post-treatment to provide close monitoring, assessment of plan quality, and ongoing feedback.
Memantine usage was optional and included as a stratification factor. Memantine was prescribed to start with radiotherapy for a total of 24 weeks with escalating doses over the first 4 weeks until reaching the target dose of 20 mg daily for the twice-daily dosing formulation or 28 mg daily for the extended-release formulation.20

OUTCOMES
Before randomization, each patient underwent baseline evaluation consisting of history and physical examination, Zubrod performance status, restaging systemic imaging, and thin-slice volumetric brain MRI with gadolinium contrast. All baseline evaluations along with adverse event assessments were repeated every 3 months until 12 months, every 6 months until 3 years, and annually thereafter until death. Before randomization and at month 3, 6, 12, 18 and 24, each patient also underwent assessments of neurocognitive function (NCF) and health-related quality of life (HRQOL). NCF testing was administered by a trained, certified site study team member (eProtocol). The same validated battery of NCF tests utilized in the prior HA-WBRT trial (NRG-CC001) was administered in this trial to assess learning and memory (Hopkins Verbal Learning Test-Revised [HVLT-R]), verbal fluency (Controlled Oral Word Association [COWA]), processing speed (Trail Making Test Part A [TMT-A]), and executive function (Trail Making Test Part B [TMT-B]). HRQOL was assessed by the EORTC QLQ-C30 and BN20 with higher scores indicating better HRQOL. All treatment-related toxicities and adverse events were recorded according to NCI Common Terminology Criteria for Adverse Events (CTCAE), version 5.0.
The primary endpoints were 12-month ICR rate for the randomized phase II component and 6-month HVLT-R DR failure, defined as decline using the reliable change index (RCI) from baseline to 6 months follow-up, for the phase III.21,22 Patients and clinicians were not blinded to treatment assignment, although the Neurocognitive Chair scoring the NCF tests was blinded. Secondary endpoints included first failure in any NCF test, failure in other NCF tests, HRQOL, overall survival (OS), and toxicity.

STATISTICAL ANALYSIS
Data from NRG/RTOG 0212 demonstrated HVLT-R DR failure in 29% of patients at 6 months following standard-dose PCI for SCLC. We hypothesized that HA-PCI will have a 14.5% absolute reduction, or 50% relative improvement, compared to PCI alone, which would require 98 analyzable patients per arm to ensure 85% statistical power with one-sided α=0.10. During the conduct of the study, a higher number of patients with extensive-stage disease were enrolled resulting in a higher rate of death by 6 months than initially estimated. Thus, the sample size was increased by 5% due to loss to follow up, 20% due to death (as opposed to 5% initially), and 25% due to patient non-compliance. The target sample size was 392 randomized patients to ensure 196 randomized evaluable patients.
As part of the phase III analysis, the randomized phase II primary endpoint of 12-month ICR risk was also assessed. With an estimated 287 evaluable patients and an assumed difference in proportions of 4.5% under the alternative, a 2-sample test of difference in proportions with a one-sided α=0.05 required 80% statistical power to evaluate a non-inferiority margin of 17.2%.
ICR risk at 12 months was tested between arm using a binomial test of difference in proportions at a one-sided significance level of 0.10 while HVLT-R DR failure at 6 months was tested using a Fisher’s exact test at a one-sided significance level of 0.05.
NCF failure, representing first neurocognitive test failure, was a secondary endpoint of interest and was defined as decline using RCI on at least one of the following tests: HVLT-R Total Recall, Delayed Recall, Delayed Recognition, TMT Part A, TMT Part B, or COWA.21,22 Death prior to NCF failure was treated as a competing risk and alive patients without NCF failure were censored at their last known follow-up time. The cause-specific Cox proportional hazards regression model was used to evaluate the effect of stratification variables (age, stage, and memantine usage) on time to first NCF failure.23 To examine each NCF test separately across time, a mixed effects model using maximum likelihood estimation with a random intercept was performed while adjusting for stratification factors. As a sensitivity analysis due to missing data, models were run using imputed data for alive patients who were missing NCF data24. Multiple imputation employed a Markov chain Monte Carlo (MCMC) method with 20 iterations. A two-sided type I error was used to determine significance in models.
A mixed effects model was used to examine the HRQOL outcomes which included global health status, physical functioning, role functioning, emotional functioning, cognitive functioning, social functioning, fatigue, and pain, across time while adjusting for stratification factors. These analyses were assessed using a two-sided significance level of 0.05.
OS was estimated using the Kaplan-Meier method, and differences between treatment arms were tested using the log rank test.25,26 OS was measured from the date of randomization to the date of death or the last follow-up date on which the patient was reported alive. Adjusted and unadjusted Cox proportional hazards models were used to obtain HRs and 95% confidence intervals (CI) for OS.23

RESULTS

STUDY PATIENTS
Between December 5, 2015, and June 21, 2022, 393 patients were randomly assigned to PCI or HA-PCI at 145 participating institutions in the USA and Canada (Figure 1). The study closed from October 2017 to January 2019 for the randomized phase II analysis and May 2020-July 2020 to increase the phase III sample size. Twenty-five patients were screened but not randomized, most commonly due to a baseline HVLT-R Delayed Recall score ≤ 2 (52%). Six patients, 3 in each arm, were found to be ineligible post-randomization but were included in the analysis under intent-to-treat. All but 13 patients (5 on the PCI arm and 8 on the HA-PCI arm) received radiotherapy (Figure 1). Baseline characteristics were well balanced between the study arms (Table 2). Median age was 64 years (min-max:34–85). Most patients had limited-stage disease (70.0%); 47% used memantine. The median follow-up was 17.0 months (min-max:0.1–86.9) for all and 30.8 months (min-max:0.1–86.9) for alive patients. Unacceptable protocol deviations in terms of hippocampal contouring and/or HA-IMRT planning was observed in 3.2% of patients on the HA-PCI arm (eTable A1). Compliance rates for NCF and HRQOL assessments are listed in eTable A2.

TREATMENT OUTCOMES

Primary Analysis and Cognitive and HRQOL Outcomes
HA-PCI had a non-inferior 12-month ICR rate as compared to PCI (14.7% vs. 14.8%, respectively, p<0.0001; Figure 2a). The difference in these proportions was 0.08% (90% CI: −5.8%, 6.0%). Six-month HVLT-R DR deterioration was not significantly different between arms (PCI 30.0% vs. HA-PCI 25.5%, p=0.28).
For the secondary endpoint of first failure on any NCF test, after adjusting for stratification factors of age, extent of disease, and actual memantine usage, the HA-PCI arm had a significantly lower risk of NCF failure as compared to the PCI arm (unadjusted HR= 0.80, 95% CI: 0.63–1.01, p=0.060, adjusted HR=0.78, 95% CI: 0.61–0.99, p=0.039; Table 3).
The HA-PCI arm had a higher baseline standardized COWA score as compared to the PCI arm (mean= −0.43 standard deviation [std dev] =1.04 vs. mean= −0.67, std dev= 1.05, p=0.028; Table 2), and greater decline in standardized COWA over time (least squares [LS] mean=−0.257, 95% CI −0.498 - −0.016, p=0.043; eTable A3). Similarly, on multiple imputation analyses, the association between the HA-PCI arm and decline in standardized COWA over time was significant (p=0.023; eTable A4). Isolated COWA failure was 3.3% of first NCF failure events, and COWA failure (isolated or in combination with other test failures) was 17.7 of first NCF failure events (eTable A5). Failure (isolated or in combination with other test failures) in HVLT-R (72.3%) and TMT (58.7%) represented the majority of first NCF failures in both arms (eTable A6).
At baseline, patients receiving HA-PCI had higher patient-reported cognitive functioning (mean=82.8 std dev=20.4 vs. mean=80.2 std dev=19.1, p=0.028; Table 2). Longitudinal modeling demonstrated no differences between treatment arms on any HRQOL domain (eTable A6).
Many correlations between NCF and HRQOL were significantly different from zero, but all were considered weak (ρ<0.35; eTable A7a–d).

Survival, Intracranial Progression, and Toxicity
There was no significant OS difference between arms (median PCI 24.9 months (95% CI 17.8–36.4) vs. HA-PCI 20.7 months (95% CI 17.0–34.7), unadjusted HR=0.97, 95% CI: 0.74–1.25, p=0.79, adjusted HR=0.88, 95% CI: 0.67–1.14 p=0.33; Figure 2b). There was no significant inter-arm difference in the proportion of patients developing intracranial progression in the hippocampus (PCI 13.6% vs. HA-PCI 20.5%, p=0.395). Extensive stage was associated with inferior OS (HR=2.46, 95% CI 1.86–3.25, p<0.0001, eTable A8) and predicted for inferior intracranial control (HR=2.44, 95% CI 1.56–3.82, p=0.0001) (eTable A9). There was no difference in grade 3+ toxicity without regard to attribution (31.4% vs. 30.7%, respectively, p=0.88) or related to treatment (11.0% vs. 10.1%, respectively, p=0.76) between the PCI and HA-PCI arms, respectively (eTable A10–A11).

DISCUSSION
Despite not meeting its primary endpoint, this phase III trial provides evidence that HA during PCI for small cell lung cancer reduces the risk of overall neurocognitive toxicity when defined as first failure on any neurocognitive test, an important secondary objective. The definition of first failure on any neurocognitive test has been considered a practice-changing endpoint in prior phase III trials that established the neurocognitive toxicity risk-reduction benefits of using stereotactic radiosurgery in lieu of conventional WBRT27,28 and adding memantine and HA during therapeutic WBRT for brain metastasis management.13,20 First failure on any neurocognitive test was not used as a primary endpoint in this trial, because historical neurocognitive data used to power the trial did not include follow-up neurocognitive testing prior to 6 months.7 HA during PCI also yielded non-inferior intracranial relapse risk and similar non-neurocognitive toxicity and overall survival in patients with small cell lung cancer. With these findings, HA should be considered for patients planning to receive PCI for small cell lung cancer.
When combined with the practice-changing results of NRG-CC001, a phase III trial that demonstrated reduction in neurocognitive toxicity with HA during therapeutic WBRT for patients with brain metastases, these findings add further support to the hypothesis of hippocampal stem cell radiosensitivity and build upon extensive preclinical work and prior clinical studies.9,10,13,15,29,30 Interestingly, the relative-risk reduction of neurocognitive toxicity with HA in this trial of PCI for small cell lung cancer (adjusted HR of 0.78) paralleled that of HA during therapeutic WBRT for non-small cell brain metastases on NRG-CC001 (adjusted HR of 0.74), highlighting the importance of HA in NCF preservation irrespective of histology (small cell or non-small cell) or treatment intent (prophylactic or therapeutic). Since HA uses IMRT to generate a several-fold reduction in radiation dose delivered to the mitotically active neural stem cells located in the hippocampal dentate gyrus and thereby permits better conservation of hippocampal neurogenesis (Supplementary Figure A1), these phase III trials provide further evidence that the hippocampal neural stem cell compartment is a radiosensitive structure-at-risk during brain-directed radiotherapy.12
While greater decline in COWA performance was noted on the HA-PCI arm in both the complete-case and imputation analyses, the HA-PCI arm did have higher baseline COWA scores. These findings raise the question of whether inter-arm imbalances may have impacted the COWA outcomes. Irrespective, isolated COWA failure was rare (3.3% of first NCF failure events), and its contribution to NCF failure was also minimal (17.6% of first NCF failure events included COWA failure). Instead, HVLT-R and TMT failures comprised the majority of first NCF failure events in both arms. Of note, baseline inter-arm differences in COWA were also observed on RTOG 0212, a randomized trial of various dose-fractionation PCI regimens for small cell lung cancer,7 and underscore the challenges of controlling for pre-PCI variability at baseline on all neurocognitive tests for trials of small cell lung cancer.
The addition of HA to PCI did not impact HRQOL at any time point. While similar findings were demonstrated in a small trial of HA-PCI, these findings contrast with patient-reported outcomes on NRG-CC001, which showed that patients who received HA during therapeutic WBRT for brain metastases reported less symptom interference at 6 months, fewer symptoms at 6 months and fewer cognitive symptoms over time.13 These discordant findings suggest that neurology-specific patient-reported outcomes, as opposed to generalized QOL assessments, may be the most appropriate patient-reported outcomes to include in neurocognitive toxicity risk-reduction trials.
Smaller trials of HA during PCI have been conducted but have yielded conflicting evidence. One trial observed lower rates of 3- and 6-month decline in delayed recall with the addition of HA to PCI,31 while another study identified no neurocognitive benefit from HA added to PCI.32 NRG-CC003 included a sample size at least two-fold larger than these prior trials and, importantly, a robust radiotherapy quality assurance infrastructure that has been demonstrated to significantly lower the incidence of unacceptable protocol deviations on prospective multi-center HA trials.33 The relatively low unacceptable protocol deviation rate (3.2%) on NRG-CC003 was a consequence of this infrastructure and likely contributed to this trial’s observation of neurocognitive toxicity risk-reduction with HA during PCI.
The results of this trial contribute to the evolving debate as to the use of PCI in SCLC in the modern era of MR imaging surveillance, which was not included in historical trials of PCI but likely has contributed to stage migration given the improved sensitivity of MR imaging to detect brain metastases. The risk-reduction in neurocognitive toxicity with HA augments the therapeutic window of PCI in SCLC and further supports the importance of the ongoing phase III trial of HA-PCI or PCI vs. observation in SCLC (S1827, NCT04155034).
Like most clinical trials evaluating different forms of radiotherapy,13,27,28 NRG-CC003 did not blind participants to treatment, as delivery of the HA-PCI and standard PCI are significantly different, and patient-blinding would introduce logistical challenges. Additionally, since the treatment team cannot be blinded, patient ignorance to treatment randomization introduces ethical issues as the treatment team would need to engage in active deception.34 Given the objectivity of NCF testing and the scoring of NCF tests by an independent Neuropsychology team blinded to treatment, lack of patient blinding is not expected to have any meaningful impact on the NCF results.35
In conclusion, while NRG-CC003 did not meet its primary endpoint of delayed recall preservation, the use of hippocampal avoidance during PCI reduces the risk of overall neurocognitive failure, which has been a practice-changing endpoint in prior neurocognitive toxicity risk-reduction trials for brain metastases. In addition, HA-PCI demonstrated non-inferior intracranial relapse and similar non-neurocognitive toxicity and overall survival compared to PCI.

Supplementary Material
PV ProtocolPV Data Supplement