Nodular Leptomeningeal Disease – A Distinct Pattern of Recurrence after Post-Resection Stereotactic Radiosurgery for Brain Metastases: A Multi-Institutional Study of Inter-Observer Reliability

Abstract

Treatment recommendations for relapsed intracranial disease are dictated by pattern of failure. SRS to the surgical cavity of resected brain metastases is an emerging standard of care that permits withholding or delaying whole brain radiotherapy. Post-resection SRS for brain metastases is associated with a specific and under-recognized pattern of intracranial recurrence, herein termed nodular leptomeningeal disease (nLMD), which is distinct from classical leptomeningeal disease (cLMD). We hypothesize that there is poor consensus with regard to the definition of LMD and that a formal, self-guided training module defining the patterns of intracranial recurrence following post-resection SRS will improve inter-rater reliability (IRR) and validity in the diagnosis of leptomeningeal disease (LMD). This study utilized pre- and post-treatment MRI studies from 30 patients with intracranial recurrence following post-resection SRS for brain metastases. We created two survey arms and disseminated them to ten physicians at multiple institutions with expertise in central nervous system (CNS) oncology, including CNS-dedicated radiation oncologists, neurologists, and neuroradiologists. In the “pre-training” arm, physicians labeled cases of recurrence using the 3 patterns commonly reported in prospective trials: 1. local cavity recurrence (LR); 2. distant parenchymal brain recurrence (DR); 3. LMD. We next supplied raters with a self-directed training module that provided guidance on distinguishing nLMD from other forms of intracranial recurrence. Raters next completed the “post-training” arm and labeled the same 30 cases (order randomized) using 4 labels including our recently described nLMD: 1. LR; 2. DR; 3. cLMD; 4. nLMD. Inter-rater reliability (IRR) increased following the training module. The fraction of cases with 100% agreement across all ten raters increased 84% (from 0.20 to 0.37). The inter-rater Fleiss’ Kappa (K) increased 42% (from K = 0.400 to K = 0.568, Wilcoxon paired rank sum p < 0.001). After training, 26% of cases originally labeled DR were reassigned to nLMD. Subgroup analysis confirmed that prior to training, IRR was lowest among LMD cases (K = 0.323). Inter-rater concordance improved across all subgroups but improved most in LMD (cLMD K = 0.444, nLMD K = 0.598). Physicians’ reported confidence in self-assigned labels also modestly improved after the training module (Wilcoxon p < 0.001). This study highlights the current need for consistent and reproducible identification of LMD. Assignment of nLMD, a dominant form of intracranial recurrence following post-resection SRS, to DR may contribute to discordant rates of LMD reported in studies of post-resection SRS. Our findings demonstrate that a brief self-guided training module distinguishing LR, DR, cLMD, and nLMD can significantly increase IRR while also producing modest increases in confidence for assigned labels.