1248P - Development of a high performance and noninvasive diagnostic model using blood cell-free microRNAs for multi-cancer early detection

Background

Despite recent progress, developing a low-cost and high-accuracy test for multi-cancer early detection (MCED) remains a challenge worldwide. Here, we report the development of a diagnostic model for MCED based on blood cell-free microRNAs (miRNA) using a training set composed of multiple cancer types.

Methods

We have previously assembled 3 large serum microarray datasets that assessed the expression of 2588 serum miRNAs from 6283 cancer patients (pts) across 13 cancer types and 5130 non-cancer controls, and developed a 4-miRNA model for MCED from a lung cancer training set ( Cancers 2022 ,14:1450). In the current study, an expanded training set including 1408 pts from 7 cancer types and 1408 age-, gender-matched non-cancer controls was compiled with the remaining participants organized into 3 validation sets. A new diagnostic model was built from the expanded training set by Linear Model for Microarray Data (limma) and 10-fold cross-validation.

Results

A new 4-miRNA diagnostic model was developed from the training set. In validation set 1 including 1358 lung cancer pts and 1501 non-cancer controls, the model achieved 99% sensitivity and 100% specificity. In validation set 2 comprising 1438 pts across 12 cancer types and 1623 non-cancer controls, the model achieved > 90% sensitivity for 8 cancer types and at least 75% sensitivity for 3 cancer types along with 99% specificity. In validation set 3 comprising 2079 pts across 4 cancer types and 598 non-cancer controls, the model achieved >90% sensitivity for all 4 cancer types and 98% specificity. Overall, the new model showed improved performance compared to the old model (p<0.001). Table: 1248P

Conclusions

Our study provided further evidence in demonstrating that circulating miRNA-based diagnostic models have the potential to be developed into inexpensive, highly accurate and noninvasive tests for MCED. Models trained using multiple cancer types outperformed those trained from a single cancer type.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

H. Hu: Financial Interests, Personal, Stocks or ownership: miRoncol Diagnostics. All other authors have declared no conflicts of interest.