Background: Early breast cancer (cT_1-2N₀) with one or two sentinel lymph node (SLN) involved may avoid axillary lymph node dissection (ALND) if follow by radiotherapy supported by Z0011 and AMAROS trials. However, only less than one-third of those patients have positive non-sentinel lymph node (nSLN) and can truly benefit from radiotherapy or ALND in those two trials. It is necessary to identify the risk of nSLN metastasis before local treatment decision. We previously developed a predictive model for nSLN involvement using circulating CK19 mRNA level combined with contrast-enhanced ultrasound (CEUS) score (ASCO2017 poster 239, NCT02992067) in a training set. To evaluate the predict effect of this model, we designed a further study using the model prospectively in a validation set (NCT03280134). Methods: We identified early breast cancer cases in Zhejiang Cancer Hospital from July 2017 to June 2018. The level of circulating CK19 mRNA tested by qRT-PCR and CEUS scores were collected before surgery in each case. Patients with 1~2 SLN involved were enrolled and continued for ALND. The estimated percentage of nSLN-involved were calculated both by our model formula and the Memorial Sloan-Kettering Cancer Center (MSKCC) nomogram. The predictive accuracy and false negative rates (FNR) were evaluated and the area under curve (AUC) was compared between two predictive models. Results: Totally, 235 patients diagnosed as early breast cancer with 1~2 SLN involved were enrolled and 35.36% of them were nSLN involved after ALND. The total accuracy and FNR by our model for nSLN-involved prediction was 94.89% and 6.02%, respectively. The AUC was 0.952 (95%CI, 0.922~0.982), significantly higher than that in MSKCC model 0.880 (95%CI, 0.833~0.927). Furthermore, only CK19 mRNA level (HR = 40.091, 95%CI, 13.663~117.635) and CEUS score (HR = 2.009, 95%CI 1.158~3.485) are significantly related to nSLN involvement in both univariate and multivariate analysis, adjusted by age, menopause statue, tumor size, histological grade, estrogen receptor, progesterone receptor and human epidermalgrowth factor receptor-2 expression. Conclusions: Our model using CK19 mRNA and CEUS score showed potential predictive value of nSLN before surgery in early breast cancer patients. Further validation in larger multicenter cohort is warranted before changing clinical practice.