Accuracy of minimal residual disease detection by circulating tumor DNA profiling in lung cancer: a meta-analysis

Abstract

Background: The sensitivity and specificity of minimal residual disease detected by circulating tumor DNA profiling (ctDNA MRD) in lung cancer, with particular attention to the distinction between landmark strategy and surveillance strategy, for predicting relapse in lung cancer patients after definitive therapy has yet to be determined.

Methods: The prognostic value of ctDNA MRD by landmark strategy and surveillance strategy was evaluated in a large cohort of patients with lung cancer who received definitive therapy using a systemic literature review and meta-analysis. Recurrence status stratified by ctDNA MRD result (positive or negative) was extracted as the clinical endpoint. We calculated the area under the summary receiver operating characteristic curves, and pooled sensitivities and specificities. Subgroup analyses were conducted based on histological type and stage of lung cancer, types of definitive therapy, and ctDNA MRD detection methods (detection technology and strategy such as tumor-informed or tumor-agnostic).

Results: This systematic review and meta-analysis of 16 unique studies includes 1251 patients with lung cancer treated with definitive therapy. The specificity of ctDNA MRD in predicting recurrence is high (0.86-0.95) with moderate sensitivity (0.41-0.76), whether shortly after treatment or during the surveillance. The landmark strategy appears to be more specific but less sensitive than the surveillance strategy.

Conclusions: Our study suggests that ctDNA MRD is a relatively promising biomarker for relapse prediction among lung cancer patients after definitive therapy, with a high specificity but suboptimal sensitivity, whether in landmark strategy or surveillance strategy. Although surveillance ctDNA MRD analysis decreases specificity compared with the landmark strategy, the decrease is minimal compared to the increase in sensitivity for relapse prediction of lung cancer.

Keywords: Diagnostic accuracy; Liquid biopsy; Lung cancer; MRD; ctDNA.

Fig. 1

Flow diagram for selection of studies

Fig. 2

Performance of ctDNA analysis approaches for detecting MRD in non-small cell lung cancer. Size of dots is a visual representation for the weight of that study in the meta-analysis. Error bars, the 95% confidence intervals. A Summary of clinical sensitivity for ctDNA detection at the first posttreatment time point (ctDNA MRD landmark). Clinical sensitivity is defined as the percentage of patients who relapsed in the follow-up period and who were ctDNA positive at the landmark. B Summary of clinical specificity for ctDNA detection at the first posttreatment time point. Clinical specificity is defined as the percentage of patients who did not relapse in the follow-up period who were ctDNA negative at the landmark. C Summary receiver operating characteristic (SROC) curve of ctDNA detection at the first posttreatment time point. D Summary of clinical sensitivity for ctDNA detection with longitudinal monitoring posttreatment (ctDNA Surveillance). E Summary of clinical specificity for ctDNA detection with longitudinal monitoring posttreatment. F SROC curve of ctDNA detection with longitudinal monitoring posttreatment

Fig. 3

Clinical sensitivity and specificity for ctDNA detection. Error bars, binomial 95% confidence intervals. A Subgroup analysis of clinical sensitivity and specificity for ctDNA detection at the first posttreatment time point (ctDNA MRD landmark) for non-small cell lung cancer (NSCLC). B Subgroup analysis of clinical sensitivity and specificity for ctDNA detection with longitudinal monitoring posttreatment (ctDNA Surveillance) for NSCLC

Fig. 4

Subgroup analysis of clinical accuracy for different ctDNA detection methods. The ctDNA detection at the first posttreatment time point (ctDNA MRD landmark) or longitudinal monitoring posttreatment (ctDNA surveillance) for non-small cell lung cancer were indicated by different icons of different colors. Error bars, binomial 95% confidence intervals. Each icon represents a summary of the results of studies using a specific detection method under different monitoring methods. L refers to landmark while S refers to surveillance