What is the problem?
Immune checkpoint inhibitors are immunotherapies that help the immune system attack a tumour. They have changed care for some cancers, but not every patient benefits.
To estimate in advance who will respond, doctors have markers such as PD-L1 (a protein measured on the tumour) or tumour mutational burden (TMB, the number of mutations in the tumour). These tests need advanced analyses and are not always reliable.
Both markers require advanced laboratory analyses, which is why a tool based on routine tests is attractive.
What did the researchers do?
They developed SCORPIO, a machine learning system that uses routine blood tests (a complete blood count and a metabolic panel) and clinical characteristics. The study brings together 9,745 patients treated with immunotherapy, across 21 cancer types.
SCORPIO was trained on 1,628 patients with 17 cancer types from Memorial Sloan Kettering Cancer Center. It was then tested on two internal sets totalling 2,511 patients, then validated externally.
SCORPIO was asked to estimate two things: overall survival at several time points between 6 and 30 months, and clinical benefit, meaning tumour response or prolonged stability.
What were the results?
On the two internal test sets, SCORPIO scored time-dependent AUC values of 0.763 and 0.759 for overall survival, against 0.503 and 0.543 for tumour mutational burden. AUC is a score where 0.5 equals chance and 1 is perfect.
For clinical benefit (tumour shrinkage or prolonged stability), the AUCs were 0.714 and 0.641 against 0.546 and 0.573 for mutational burden. In external validation, on 10 global phase 3 trials (4,447 patients) and a real-world Mount Sinai cohort (1,159 patients), SCORPIO held up and beat PD-L1 staining.
The authors say SCORPIO maintained robust performance in the external cohorts, which supports its reliability and adaptability across cancer types and healthcare settings.
Key figures
- 9,745patients treated with immunotherapy, across 21 cancer types
- 0.763 and 0.759scores (AUC) on the two internal test sets, against 0.503 and 0.543 for tumour mutational burden
- 4,447 + 1,159patients from 10 phase 3 trials and a real-world cohort used for external validation
What does this change for patients?
In time, such a tool could help care teams anticipate who is most likely to benefit from immunotherapy, using simple blood tests already done in routine, so at low cost and without waiting for genetic analyses.
For now, it is a possible aid to thinking, not a decision test. It should not be used to refuse a treatment.
The authors highlight its potential to estimate outcomes across diverse cancer types and healthcare settings.
What are the limits?
The model was built and tested on existing data. The study does not show that using it to guide decisions improves survival or quality of life, which would need a prospective trial.
Performance remains modest on some measures (an AUC of 0.641, for example), and a tool like this must be tested in other hospitals and approved before any medical use.
Tumour mutational burden scored close to 0.5, which is close to chance, in these sets. SCORPIO's clinical-benefit AUC of 0.641 on one set also shows there is room to improve.
Immunotherapy: preparing your questions with Mima
SCORPIO is a research tool: your blood test is not read this way in routine care. In time, it could help teams anticipate who will benefit from immunotherapy.
If you are on immunotherapy, Mima helps you note tiredness, nausea and appetite day by day, and prepare questions for each appointment. She also explains side effects in plain words. A question: "Which signs should I report to you?"
Frequently asked questions
Can a blood test help choose immunotherapy?
In this study, an AI using routine blood tests and clinical data estimated response to immunotherapy better than the PD-L1 marker. The tool is not yet validated to guide treatment.
What is PD-L1?
It is a protein measured on the tumour to estimate whether checkpoint-inhibitor immunotherapy is likely to help. SCORPIO beat it in external validation.
What is AUC?
It is a score from 0 to 1 that measures how well a test separates two groups of patients. At 0.5 it is no better than chance, at 1 it is perfect.
Is SCORPIO used in hospitals?
The paper reports a research result and does not describe an approval or routine use.
The original study
Yoo SK, Fitzgerald CW, Cho BA, Samstein RM, Morris LGT, Chowell D, et al. “Prediction of checkpoint inhibitor immunotherapy efficacy for cancer using routine blood tests and clinical data.” Nature Medicine, 6 January 2025. DOI : 10.1038/s41591-024-03398-5
Read the full study in Nature MedicineThis article summarises a scientific study and is not medical advice.