Can a routine blood test triage gastric-cancer risk?
A six-variable model reached an AUC of 0.831 in 735 later patients at one Chinese hospital. It could become a low-cost prompt for endoscopy, but it has no external or prospective validation and age and sex may explain part of its performance.
The short version
3 key points
- 1.The study included 2,044 patients at Hanzhong Central Hospital: 477 with pathologically confirmed, newly diagnosed, treatment-naïve gastric cancer and 1,567 with benign gastric disease.
- 2.A random-forest model using age, sex, neutrophils, lymphocytes, monocytes and platelets achieved an AUC of 0.831 in a later 735-patient cohort; at the selected threshold sensitivity was 72.9% and specificity 77.2%.
- 3.Validation occurred later in time at the same hospital. The authors did not test an age-and-sex-only comparator, so the added value of the blood counts is not quantified.
The answer: useful early evidence for triage, not a cancer test
Six readily available variables separated gastric-cancer cases from benign gastric disease reasonably well in a later group of patients at the same Chinese hospital. The reference random-forest model produced an area under the receiver-operating curve of 0.831, where 0.5 is chance and 1.0 is perfect discrimination. At the threshold chosen from the training data, it identified 72.9% of cancer cases and correctly rejected 77.2% of benign cases. Those results justify external evaluation as an inexpensive pre-endoscopy risk flag.
They do not justify telling a patient that a complete blood count has diagnosed or excluded cancer. In the validation cohort, 34.8% of patients would have been flagged high risk. The positive predictive value was 50.4% and the negative predictive value 90.0% in a sample where 177 of 735 people had gastric cancer. Predictive values will change when the model is used in a primary-care population with a different disease prevalence. Endoscopy with biopsy remains the diagnostic reference standard.[1]
How the 2,044-patient study was built
Researchers retrospectively assembled 477 pathologically confirmed, newly diagnosed and treatment-naïve gastric-cancer cases and 1,567 patients assigned benign gastric diagnoses. Two reviewers checked eligibility, treatment history and major conditions that could disturb blood counts; patients were de-duplicated using national identity numbers. The blood test had to be drawn at admission, before surgery, chemotherapy or radiotherapy. The cohort was split by date into 1,309 development patients—300 cancers and 1,009 controls—and 735 later validation patients—177 cancers and 558 controls.
Eight algorithms were trained with five-fold cross-validation. The principal random forest used only age, sex and neutrophil, lymphocyte, monocyte and platelet counts. Its AUC of 0.831 was essentially the same as XGBoost's 0.832, while the random forest had the lowest Brier score, 0.133. Adding five derived inflammatory ratios did not improve discrimination: the expanded model's AUC was 0.821, a non-significant difference of minus 0.010. SHAP explanations ranked lymphocyte count, male sex and age as the leading contributors.[1]
The marker-negative subgroup is encouraging but narrow
The authors also asked whether the model could flag cancers missed by three conventional serum markers: CEA, CA19-9 and CA72-4. In the later validation cohort, 59 cancer patients had all three markers measured and below the hospital's reference thresholds. At the training-derived decision threshold, the model labelled 62.7% of them high risk, with a 95% confidence interval from 50.0% to 73.9%. That suggests the blood-count pattern is not simply reproducing a positive tumour-marker result.
It is not a head-to-head diagnostic comparison. Marker results were not routinely available for the benign controls, so the analysis measures recall only among confirmed cancers. It cannot show how many marker-negative people without cancer would be sent for unnecessary endoscopy. The subgroup is small, and recall fell as the probability threshold increased. A future comparison must measure both benefits and false alarms in the same consecutive clinical population.[1]
What clinicians and patients should do with the result
For clinicians in settings where endoscopy capacity is scarce, the attraction is practical: the proposed inputs are already collected in routine care, and the model might help prioritise people with persistent symptoms or other risk factors. A safe service would present the score alongside symptoms, history and local referral guidance, display uncertainty, and allow clinical override. It would not use a low score to dismiss alarm symptoms such as bleeding, weight loss, progressive swallowing difficulty or persistent anaemia.
For patients, the study offers no self-test. Age and sex were two of the strongest predictors, and blood-cell counts can change with infection, bleeding, nutrition, medicines and many chronic diseases. The authors excluded major count-altering conditions, so real-world performance could be worse when those exclusions are not clean. The score also distinguishes cancer from selected benign gastric disease within one hospital population; it has not been tested as screening for people without symptoms.[1]
The important evidence limits
The study is retrospective and single-centre. Moving the calendar forward is stronger than a random internal split because it tests later patients, but staffing, laboratory procedures and referral patterns remain largely shared. Benign diagnoses were based mainly on clinician assignments, so occult early cancers may have entered the control group. The authors had no external hospital, country or prospective workflow test, and unmeasured factors such as Helicobacter pylori, bleeding, nutrition, smoking, alcohol, medicines and infection could influence both blood counts and cancer risk.
A particularly important omission is an age-and-sex-only comparator. Because gastric-cancer cases were older and more often male, the paper does not quantify how much the four blood-cell counts add beyond those demographics. The study was supported by public research programmes in Shaanxi and Hanzhong, and the authors declared no commercial conflict. Funding declarations do not resolve the design limits, but no test manufacturer is reported as sponsoring the result.[1]
What would change the assessment
Confidence would rise with a locked model tested prospectively in several hospitals and primary-care pathways, including consecutive patients with realistic comorbidities. Investigators should compare the six-variable model with age and sex alone, established risk scores and ordinary clinician referral. Calibration and decision thresholds need to be recalculated for local cancer prevalence, while results should be reported by age, sex, stage and symptom group.
The decisive trial would measure care rather than only discrimination: time to endoscopy, early-stage detection, unnecessary procedures, missed cancers, workload and patient anxiety. Until that evidence exists, the study supports a research-grade triage hypothesis. It does not support replacing endoscopy, using a routine blood count as a stand-alone screen or reassuring an individual patient from a low machine-learning score.[1]
What this means for people
- Patients could eventually be prioritised for scarce endoscopy using information already present in a routine blood count, but the score cannot diagnose or exclude cancer.
- Clinicians need local calibration, an age-and-sex comparator and prospective workflow evidence before using the model to alter referrals.
- Health-service buyers should measure missed cancers and unnecessary endoscopies, not treat AUC as a patient outcome.
Global context
The data came from one hospital in Shaanxi, China. Gastric-cancer prevalence, screening practice, blood-test platforms and referral pathways differ across countries, so the operating threshold and predictive values are not portable without external validation.
What the evidence does not yet show
- Retrospective, single-centre design with only same-hospital temporal validation
- No age-and-sex-only comparator to isolate the incremental value of blood-cell counts
- Benign controls were based mainly on clinician-assigned diagnoses, allowing possible occult disease
- The marker-negative analysis included only confirmed cancers and cannot estimate false-positive referrals
- No prospective evidence that the score improves detection, outcomes or use of endoscopy
What to watch next
- Locked multicentre external validation
- Prospective testing in consecutive primary-care and endoscopy referrals
- Comparison with age and sex alone and established clinical risk tools
- Stage-specific sensitivity and local calibration
- Effects on missed cancers, unnecessary procedures, workload and time to diagnosis
Living evidence record
Impact record IAI-0RG5OQV
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
15 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what Frontiers in Immunology published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 15 October 2026
Research question, topics and regions
Whether age, sex and four pretreatment blood-cell counts can distinguish newly diagnosed gastric cancer from benign gastric disease
Topics: Gastric cancer · Complete blood count · Machine learning · Triage · Temporal validation
Regions: China · Asia
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Watch the background, then check the evidence
The Royal Society
AI-generated illustration · not a video frame
YouTube loads only when you choose. Press play in the player to start. If playback is unavailable, use “Watch on YouTube” below.
Healthcare, culture & human impact · Background explainer
Published 2026-07-01 · 44 min 28 sec
Is this the AI future we want? | The Royal Society
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Source checked 14 October 2026. Check available captions on the publisher’s video page.
The Impact Brief · Free
Keep following the stories that matter to you.
Get our weekday briefing on what changed and why it matters. Choose your topics below.
Continue the story
There’s more to this story
Health & Life Sciences
Can routine lab results distinguish two stroke types?
A ten-biomarker model reached an AUC of 0.902 in 495 later patients at the same Chinese hospital. The result is promising decision-support research, but unknown blood-draw timing, no stroke mimics and no external hospital mean it cannot replace emergency brain imaging.
6 min · 1 source
Health & Life Sciences
Can a six-protein AI score rule out serious fatty liver disease?
A multicohort study reports AUCs of 0.79–0.83 for a six-protein MASH score and a 0.99 negative predictive value in a low-prevalence UK cohort. Different reference standards, small biopsy cohorts and unavailable routine assays keep it from clinical readiness.
8 min · 1 source
Health & Life Sciences
Did a breast-cancer AI score survive external validation?
A 50-gene machine-learning signature looked moderately accurate in 101 development cases, but its transferable component fell to an AUC of 0.55 in 320 external cases and added no value beyond routine clinical factors.
7 min · 1 source
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.