Tumor Marker Test in Cerebrospinal Fluid for Leptomeningeal Metastasis Diagnosis and Response Assessment in Non-Small-Cell Lung Cancer

Abstract

Backgrounds: Current methods for diagnosis and disease monitoring for leptomeningeal metastasis (LM) face significant challenges, as they are relatively complex and lack the precision needed to detect subtle changes.

Methods: We conducted a retrospective study with the primary endpoint of assessing the diagnostic effectiveness of cerebrospinal fluid (CSF) tumor markers in LM patients with non-small-cell lung cancer (NSCLC). Secondary endpoints included evaluating the concordance with EANO-ESMO response assessment, determining optimal thresholds, and comparing costs against CSF ctDNA tests.

Results: We retrospectively included 368 patients (151 LM patients, 219 non-LM patients) as the training set, and another 137 patients (63 LM patients, 74 non-LM patients) on a consecutive basis as the validation set. The CSF tumor marker panel using a logistic model showed the best performance both in the training set (AUC [95% CI] = 0.992 [0.984-1.000]) and the validation set (AUC [95% CI] = 0.939 [0.891-0.986]). In the response assessment, 167 events were evaluated, with 33 classified as response, 59 as stable, and 75 as progression. We found a concordance with EANO-ESMO response assessment, and a threshold of ± 25% of the maximal CSF tumor marker level change (maxTML) yielded the optimal predictive performance. Finally, the cost of the CSF tumor marker test (0.76%, IQR (0.30%-1.67%)) was significantly lower in a single hospitalization than CSF ctDNA (62.45%, IQR (32.62%-85.81%)).

Conclusion: Our study demonstrated that the CSF tumor marker test was an accurate, easily interpretable, and cheap tool for both diagnosis and monitoring therapeutic response in LM patients with NSCLC.

Trial registration: Chinese Clinical Trial Registry (ChiCTR) number: ChiCTR2300078556.

Keywords: CSF tumor marker test; diagnosis; leptomeningeal metastasis; non‐small‐cell lung cancer; response assessment.

FIGURE 1

Study flow diagram. A total of 368 patients were retrospectively enrolled in the training set, comprising 151 patients with LM from NSCLC, 68 patients with BM from NSCLC, and 151 control patients. An additional 137 patients were consecutively enrolled as the validation cohort, including 63 patients with LM from NSCLC, 15 patients with BM from NSCLC, and 59 control patients. The diagnostic value of the CSF tumor marker test was analyzed and validated. Among the 151 LM patients in the training set, 71 patients (with 167 evaluable events) qualified for EANO‐ESMO response assessment. We assessed the concordance between the CSF tumor marker response and EANO‐ESMO response assessment. Finally, we compared the costs between the CSF tumor marker test and the CSF ctDNA test. BM, brain metastasis; LM, leptomeningeal metastasis.

FIGURE 2

Performance of CSF tumor makers in diagnosing LM. (A–C) ROC curves were generated to evaluate the performance of CEA, CA19‐9, CA125, and the CSF tumor marker panel in discriminating: (A) LM cohort versus control cohort, (B) LM cohort versus BM cohort, and (C) LM cohort versus non‐LM cohort in the training set. (D–F) ROC curves were generated to evaluate the diagnostic performance of CEA (Cutoff_CEA = 1.8 ng/mL), CA19‐9 (Cutoff_CA19‐9 = 1.2 U/mL), CA125 (Cutoff_CA125 = 0.5 U/mL), and CSF tumor marker panels (Panel 1, Panel 2 and Panel 3) in discriminating LM from control cohort (D), BM cohort (E), and non‐LM cohort (F) in the validation set. AUC, area under curve; CA19‐9, carcinoembryonic antigen 19‐9; CA125: carcinoembryonic antigen 125; CEA, carcinoembryonic antigen.

FIGURE 3

AUC of ROC curves for progression, stable, response, and DC prediction with different thresholds. We defined three patterns (progression, response, and stable) of CSF tumor marker response including progression, response, and stable based on maxTML, using varying thresholds: ± 15%, ± 20%, ±25%, ± 30%, and ± 35%. Our analysis revealed that a 25% threshold yielded the optimal predictive performance, with ROC curve AUC values of 0.929 for progression, 0.770 for stable, 0.910 for response, and 0.929 for DC. DC, disease control.