A sweeping review of 90 studies has mapped how researchers around the world have tried to detect cancer by tracking how a patient’s blood test results change over time, rather than judging each result against a fixed abnormality threshold. The work, published in Diagnostic and Prognostic Research, is the first scoping review to systematically chart this emerging field, and its central message is double-edged: the idea is attracting growing scientific attention, but the evidence base remains fragmented, exploratory and heavily skewed towards just a handful of cancers.
The motivation is straightforward. Cancer is a leading cause of death globally, accounting for one in six deaths worldwide and more than one in four deaths in the UK in 2020. When caught early, the chances of successful treatment and survival are substantially higher, yet around half of all cancer cases in the UK, 45.5 percent, are diagnosed at advanced stages. Most cancers are diagnosed after symptomatic presentation to primary care, which places general practitioners on the front line of early detection. They face a difficult task: many cancers are rare, their symptoms are non-specific, and patients often have comorbidities that muddy the picture. Existing electronic risk assessment tools, which draw mainly on patient data such as family history and body mass index alongside symptoms, help GPs decide who to refer, but their usefulness for harder-to-suspect cancers such as myeloma and pancreatic cancer remains uncertain.
Standard blood tests already play a role in this process. Abnormalities in commonly available primary care tests, such as full blood counts and liver function tests, are crucial for detecting and risk stratifying possible cancer in symptomatic patients. Yet for most tests, the risk of an individual cancer associated with a single abnormal result is too low to justify urgent investigation, with anaemia for colorectal cancer and jaundice for pancreatic cancer the main exceptions. When all cancers are combined, low albumin, raised platelets, raised calcium and raised inflammatory markers can push the risk of cancer above the 3 percent referral threshold recommended by the National Institute for Health and Care Excellence, but these signals leave doctors uncertain about which specific cancer warrants investigation. This is where trends come in. A blood test trend refers to the pattern of change in repeated quantitative measurements of the same test within one person over time, rather than the interpretation of a single value against a binary threshold. Monitoring such patterns may better capture within-person changes, be more robust to measurement error, and add information beyond what a single test can provide. The review’s authors illustrate the concept with a simple example: a haemoglobin level in the low-normal range might not raise alarm on its own, but if it follows years of consistently high-normal results, that downward drift could signal an opportunity for cancer investigation.
To map the landscape, the research team, working with an experienced librarian, searched MEDLINE, Embase, ProQuest Dissertations & Theses Global and Overton for studies published from inception until 3 April 2025, with no language restriction, and manually inspected the reference lists of included studies. Eligible studies had to report change over time in repeated blood tests that included at least one test commonly available in primary care, covering full blood count components, liver function tests, renal function, inflammatory markers and a handful of other tests such as HbA1c, calcium and thyroid stimulating hormone. Studies had to investigate trends before or during cancer diagnosis; work on tests used after diagnosis, for prognosis or treatment monitoring, was excluded. Any study design counted, from cohort studies and case-control analyses to case reports, protocols, reviews and grey literature such as theses and conference abstracts, while news articles, websites and guidelines were left out.
The search identified 24,204 records after duplicate removal, and screening whittled these down to 89 included studies, with one more added through reference list screening, for a total of 90. Publication dates ranged from 1984 to 2025, and the field has clearly accelerated: only 4 studies appeared in 2004 or earlier and 2 between 2005 and 2009, compared with 9 between 2010 and 2014, 31 between 2015 and 2019, and 44 between 2020 and 2025. Most studies, 85 of them, were conducted in high-income countries, including 36 in the United States and 19 in the United Kingdom. Cohort designs were the most common, with 45 studies, followed by 33 case-control studies, 8 case reports, 3 case studies, 1 protocol and 1 review.
The concentration of research interest is striking. Colorectal cancer, studied in 26 papers, and pancreatic cancer, studied in 24, dominated the field, followed by haematological cancers with 16 studies and lung cancer with 7. Haemoglobin was the most frequently investigated blood test, appearing in 34 studies, with platelet count close behind at 28. Among colorectal cancer studies, haemoglobin and mean cell volume were the most commonly assessed tests, while HbA1c and fasting glucose led the way in work on pancreatic cancer. Many other cancer types and blood tests have barely been touched.
The review identified seven common themes across the literature. The largest, covering 64 studies, examined the association between blood test trends and cancer, using approaches that ranged from descriptive reporting of results at different time points and regression modelling of temporal changes, most often linear regression, to statistical comparisons between patients with and without cancer, quantified changes such as slopes of change, analyses of how risk ratios and hazard ratios varied across time intervals before diagnosis, group-based trajectory models, and individual-level trajectory modelling with joint or linear mixed-effect models. A second theme, addressed by 26 studies, involved prediction models incorporating blood test trends to derive cancer risk. Five of these studies developed 8 models in total, targeting colorectal, gastrointestinal, lung and pancreatic cancer, using machine learning methods such as XGBoost and random forest, logistic regression, and multivariate joint modelling. External validation was conducted in 14 studies, evaluating two established models: ColonFlag, with area under the curve values ranging from 0.74 to 0.84 across 10 studies, and ENDPAC, with AUC values from 0.69 to 0.75 across 4 studies. The point estimates of AUC for newly developed models ranged from 0.71 to 0.87. Notably, in two studies, blood test trends were considered as candidate covariates during model development but were excluded from the final models because they lacked statistical significance.
Other themes revealed both promise and persistent gaps. Sixty-two studies combined blood test trends with other cancer risk factors, most commonly adjusting for age, sex and body mass index, or stratifying analyses by sex and diabetic status, but integration of symptom information or other diagnostic tests was limited. Fifty-eight studies discussed implications for clinical practice, yet most offered only general statements about how particular trend patterns might act as signals warranting consideration of cancer, and no study provided recommendations for incorporating trends into clinical guidelines for general practitioners or reported actual implementation of a trend-based model in practice. When the review compared trend-based interpretation with conventional threshold-based abnormality, it found 22 studies that graphically illustrated changes often confined within the normal range without inferential comparison to abnormality thresholds, and only three studies explicitly highlighted the potential of a trend to pre-date an abnormality. Just one study compared the performance of a prediction model incorporating trends with one based on single time point values. Thirty-nine studies investigated combinations of trends in different blood tests, but most focused on components of the full blood count, and only one explicitly assessed the correlation between trends, without reporting the results in the manuscript. A further methodological concern, bias due to informative observation, was noted in 12 studies: because blood tests in real-world settings are often driven by emerging symptoms, chronic disease management or clinician concern, the timing and frequency of testing carry implicit information about a patient’s health, and no study proposed or applied a method to address or mitigate this bias.
The review’s authors conclude that blood test trends have been increasingly studied and reported in the context of cancer detection, but that knowledge remains fragmented and largely exploratory. Substantial methodological heterogeneity across studies, they suggest, reflects the exploratory nature of the field and the lack of consensus on how best to incorporate temporal information into cancer detection. Many studies relied on descriptive analysis, case reports or simplified representations of trends, often reducing changes to comparisons at fixed time points or using population-level aggregated data, while advanced longitudinal methods better suited to within-person trajectories, such as machine learning and joint modelling, were underrepresented. Prediction model development has been limited to a small number of cancer types with little external validation, raising concerns about generalisability. The authors call for future research to broaden the range of cancers and tests studied, prioritise external validation across diverse populations, directly compare trend-based models with standard threshold-based approaches, integrate symptoms and other clinical indicators, formally assess diagnostic performance including sensitivity, specificity and lead time, and investigate the acceptability, feasibility and cost-effectiveness of repeat testing from the perspectives of patients and clinicians. As a scoping review, the work did not conduct formal quality appraisal of the included studies, and the considerable heterogeneity in how trends were defined and applied precluded quantitative synthesis, so the findings primarily map the terrain rather than settle whether trends will ultimately improve early detection in the clinic.
Subject of Research: The use of blood test trends for cancer detection
Article Title: The use of blood test trends in cancer detection: a scoping review
Article References: Zhu, S., Friedemann Smith, C., Collins, K. K., Ziyenge, S. E., Murphy, J., de Vere Hunt, I., Brubert, Z., Roberts, N., Morris, E. J. A., Hobbs, R., Nicholson, B. D., & Virdee, P. S. (2026). The use of blood test trends in cancer detection: a scoping review. Diagnostic and Prognostic Research, 10(1), Article 24. https://doi.org/10.1186/s41512-026-00236-9
Image Credits: AI Generated
DOI: 10.1186/s41512-026-00236-9
Keywords: blood test trends, cancer detection, scoping review, primary care, colorectal cancer, pancreatic cancer, haemoglobin, prediction models, risk stratification, early diagnosis, full blood count, informative observation bias
News Source: Nathaniel Bowman. (October 28, 2026). Blood test trends could flag cancer earlier, but the evidence is still patchy, review finds. Scienmag.
