192 - Estimating Neurocognitive Failure in the Presence of Competing Mortality: Kaplan-Meier and Competing-Risk Analyses from NRG CC001

11:05am - 11:15am ET

Room 109

Presenter(s)

Reza Zarinshenas, MD - University of Maryland Radiation Oncology, Baltimore, MD

R. Zarinshenas¹, K. Sun², H. R. R. Cherng¹, S. M. Bentzen², and M. V. Mishra³; ¹Department of Radiation Oncology, University of Maryland Medical Center, Baltimore, MD, ²Division of Biostatistics and Bioinformatics, University of Maryland Greenebaum Cancer Center, and Department of Epidemiology and Public Health, University of Maryland School of Medicine, Baltimore, MD, ³Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, MD

Purpose/Objective(s):

Patients with brain metastases receiving brain radiation face two concurrent risks: neurocognitive decline and death. How neurocognitive failure (NCF) risk is estimated depends on the clinical question. Kaplan–Meier (KM) analysis treats death as a censoring event, estimating what a patient's risk of NCF would be in the absence of competing mortality. This addresses the patient's question: "If I were to survive, what would my chance of cognitive decline be?" Cumulative incidence function (CIF) analysis accounts for death as a competing risk, estimating the real-world probability of NCF among all treated patients. This addresses the health-systems question: "Among all patients treated, how many will actually experience NCF?" When mortality is substantial, these complementary methods can yield meaningfully different estimates with implications for counseling and resource planning. We applied both to NRG CC001, where hippocampal-avoidance WBRT (HA-WBRT) reduced NCF vs WBRT (both with memantine), to examine how analytic method affects absolute risk estimates and treatment comparisons. We hypothesized that KM would yield higher absolute NCF estimates than CIF in this high-mortality population.

Materials/Methods:

This secondary analysis compared KM and CIF approaches in NRG CC001 (N=518). Data were obtained from the NCTN/NCORP Data Archive. Cognitive decline was defined by a decline in at least one cognitive test as determined by the reliable change index per the original trial. KM treated death as censoring, whereas competing-risk analyses modeled death as a competing event using a Fine–Gray subdistribution model. Given prior evidence of heterogeneity of HA-WBRT treatment effect by tumor histology, subgroup analyses were performed by tumor site (lung vs. non-lung).

Results:

At 6 months, KM yielded higher NCF estimates than competing-risk analyses in both arms. For WBRT, 6-month NCF risk was estimated at 86% (95% CI, 79–91%) by KM versus 68% (95% CI, 61–74%) by CIF. For HA-WBRT, estimates were 78% (95% CI, 70–84%) by KM and 59% (95% CI, 52–66%) by CIF. Despite these differences, relative treatment effects were similar: HR for HA-WBRT vs WBRT was 0.76 (95% CI, 0.60–0.97) by Cox model and 0.77 (95% CI, 0.60–0.98) using a Fine–Gray model. In subset analysis, among lung primaries (n=307), competing-risk modeling strengthened HA-WBRT benefit (Fine–Gray HR 0.58, 95% CI 0.43–0.80) vs Cox model (HR 0.71, 95% CI 0.52–0.98). Among non-lung primaries (n=211), point estimates diverged in direction but neither reached statistical significance (Cox HR 0.85, 95% CI 0.58–1.28; Fine–Gray HR 1.15, 95% CI 0.78–1.71).

Conclusion:

KM and CIF produced markedly different absolute NCF risk estimates, whereas relative treatment effects did not meaningfully differ between approaches. Reporting both measures in brain metastasis trials can strengthen shared decision-making by providing clinicians with complementary information for patient counseling and resource allocation.