How often is a non-cancer participant correctly classified?
Greater than 99% specificity means fewer than 1 in 100 non-cancer participants in the published validation received a false cancer signal.
Oncology · Published performance
Accuracy is a profile,not one percentage.
The meaningful question is not simply “How accurate is it?” It is how the model performed, for which cancer types, at what specificity, in what populations and under what study design.
Figures on this page describe the peer-reviewed research model. They should not be interpreted as guaranteed performance for every person or as a cancer diagnosis.
Reading performance correctly
Three measures answer different questions.
A single headline number hides the trade-offs. Specificity, sensitivity and predictive value each describe a different part of test performance.
How often is a cancer participant’s signal detected?
Sensitivity is not uniform. It varied by cancer type in the published datasets, which is why responsible reporting should show the range rather than one blended claim.
What does a positive signal mean in the population being tested?
Predictive value depends on how common cancer is in the tested population. Case-control performance cannot by itself establish real-world screening outcomes.
How the 2024 study was structured
Development first. Then independent validation.
The model was developed in one dataset and evaluated in three mutually exclusive validation datasets assembled from standardized serum microRNA data.
Model developmentTraining set
2,816
Participants used to develop the model
1,408 cancer patients across seven cancer types and 1,408 age- and gender-matched non-cancer controls.
Model evaluation3 independent sets
8,597
Participants used for independent validation
4,875 cancer patients across 13 evaluated cancer types and 3,722 non-cancer participants, with no overlap with the training set.
Why this distinction matters: separating model development from validation reduces the risk of reporting performance only on the data used to build the model. The papers also state that prospective screening trials in asymptomatic intended-use populations are still required to establish clinical utility.
The published performance profile
Strong results,reported with their range.
These figures come from the 2024 peer-reviewed research paper. They describe the published four-microRNA model, not a promise of individual outcomes.
>99%
Specificity in the principal validation findings
High specificity was an explicit design priority.
The diagnostic threshold was selected to limit false-positive signals in non-cancer controls. This is particularly important when a test is intended for people who may feel healthy.
>90%
Sensitivity for nine cancer types
Sensitivity was high for most validated types.
The paper reported greater than 90% sensitivity for lung, biliary tract, bladder, colorectal, esophageal, gastric, glioma, pancreatic and prostate cancers.
75–84%
Sensitivity for three additional cancer types
Greater than 90% sensitivity
Nine cancer types
Lung · Biliary tract · Bladder · Colorectal · Esophageal · Gastric · Glioma · Pancreatic · Prostate
75–84% sensitivity
Three cancer types
Liver · Ovarian · Sarcoma
Important boundary
Breast cancer
The model showed low sensitivity for breast cancer at the threshold chosen to maintain high specificity. It should not replace mammography or recommended breast screening.
Source: Zhang J, Rui H, Hu H. Noninvasive multi-cancer detection using blood-based cell-free microRNAs. Scientific Reports 14, 22136 (2024). Performance varies by dataset, cancer type, threshold and population.
What the evidence does,and does not,show
Published validation is a foundation, not clinical certainty.
Scientific credibility comes from reporting the boundaries with the results.
Retrospective case-control evidence
The published studies analyzed existing public microarray datasets. They were not prospective population-screening trials enrolling asymptomatic people in routine care.
Research model versus current laboratory test
Published performance supports the analytical approach. Product-specific methods, quality systems, intended use and current disclosures belong with the miCheckup test information.
Performance varies by cancer type
A high overall specificity does not mean every cancer is detected with equal sensitivity. Some cancers produced stronger signals in the studied datasets than others.
Neither result is diagnostic
A signal requires clinical follow-up. A result without a suspected signal does not rule out cancer and does not replace symptoms, physician judgment or recommended screening.
Prospective validation remains essential
Real-world studies in intended-use populations are needed to determine predictive value, clinical utility and performance under routine screening conditions.
Evidence before assertion
Read the peer-reviewed record.
The two open-access papers show how the research evolved from an initial proof of concept to broader model development and independent validation.
A Novel Blood-Based microRNA Diagnostic Model with High Accuracy for Multi-Cancer Early Detection
The foundational proof-of-concept study analyzed 7,536 serum samples and described the initial four-microRNA diagnostic model.
Read the open-access paperNoninvasive multi-cancer detection using blood-based cell-free microRNAs
The later study used a broader training architecture and three independent validation sets across 13 evaluated cancer types.
Read the open-access paperWhy this matters to the platform
Oncology established the first published branch.
The strategic value is larger than a single metric or a single test: high-dimensional microRNA measurements can be converted into disease-specific analytical models, validated, refined and extended as research grows.
Current test information
Research evidence belongs with complete product context.
For the current miCheckup cancer panel, intended use, limitations, laboratory information and testing pathway, use the dedicated test website.