Background: The advances in deep sequencing technologies makes circulating tumor DNA (ctDNA) detection in preoperative plasma a more and more significant indicator in early stage cancers. In particular, it allows us to learn a lot about tumor progression before surgery. Plasma variant allele frequency (VAF) of preoperative ctDNA mutations had been verified to correlate with tumor sizes in early lung cancers. Identification of lymph node (LN) metastasis status before surgery in patients with lung cancer is valuable for prognosis and treatment strategy decisions. To evaluate whether ctDNA predict the LN metastasis status, we conducted this analysis. Methods: Preoperative plasma and the matched leukocyte before treatment were collected from 38 early stage lung cancer patients, in which 8 cases with lymph node metastasis. 150-gene panel sequencing were performed on the specimens and the sequencing data were used as primary cohort. Multivariable logistic regression analysis was used to develop a prediction model based on the DNA and other clinical characteristics. The performance of the model was assessed by its calibration, discrimination and clinical usefulness in the public TRACERx 100 early lung cancers cohort. Results: The proportion of high variant allele frequency (HAF proportion) of mutations was significantly associated with LN status (p < 0.05, for both primary and validation cohorts). A prediction model was developed and predictors included the HAF proportion, smoking status and tumor size. The C-index of the model was 0.419 in primary cohort. In external validation cohort, the AUCs for LN positive was 0.678. Conclusions: Based on pan-gene panel deep sequencing, a big proportion of high variant allele frequency mutations in a single patient indicates advanced disease progression, which may reflect in LN involvement status. A logistic regression model was constructed using ctDNA and clinical characteristics. It showed considerable performance in the prediction of LN metastasis in early lung cancers.