Unexpected weight loss (UWL) is a non-specific symptom that may signal underlying cancer but poses diagnostic challenges in primary care. Current risk stratification commonly uses the most recent blood test result and predefined abnormal thresholds. The evidence base has largely focused on unadjusted single-test abnormalities, which often give risks too low to trigger urgent investigation for a specific cancer. Monitoring temporal changes — blood test trends — could provide additional discrimination for undiagnosed cancer beyond single abnormal results.
The investigators conducted a retrospective cohort study using the Clinical Practice Research Datalink (CPRD) of English primary care between 1 January 2000 and 31 December 2018. Adults aged 18 years and older with a primary care record of UWL were included. Cancer outcomes (overall and site-specific) were ascertained through linked National Cancer Registrations Data. The final analytic sample comprised 275,205 patients with UWL, of whom 5.0% (n = 13,798) were diagnosed with cancer following the UWL presentation.
Quantitative results were extracted for 26 routinely recorded blood tests in primary care. For each patient, test histories up to 10 years before the UWL date were collected. The median number of tests per person over 10 years varied by analyte, mostly between 2 and 4 tests. The median interval between the first and last blood test per person was 5.2 years (IQR 1.5–8.2) in cases and 4.3 years (IQR 1.0–7.7) in non-cases. The median time from the last blood test to the UWL presentation was short: 0.1 years (IQR 0.0–0.6) for those with cancer and 0.3 years (IQR 0.0–1.1) for those without.
Trends were assessed using joint models over several retrospective windows (1-, 3-, 5-, and 10-year periods) prior to the UWL index date. Trend analyses were restricted to patients with at least two tests within the chosen trend window.
For single-test abnormality analyses, Cox models were fitted to estimate cancer risk associated with an abnormal most-recent test co-occurring with UWL. Joint models estimated the association between longitudinal test trajectories (trends) and subsequent cancer diagnosis. Models were presented unadjusted and adjusted for age and sex. Discrimination for each model was quantified using the area under the receiver operating characteristic curve (AUC) with 95% confidence intervals.
Adjustment for age and sex improved discrimination for both single-test abnormalities and trends across analyses. Comparing adjusted models, adjusted trends yielded higher AUCs than adjusted single-test abnormalities on 34 test–cancer combinations. The maximum reported AUC for trend-based discrimination was 0.82 (95% CI 0.81–0.83).
Notable examples where trend improved discrimination included:
Overall, 26 trend measures provided improved discrimination for cancer overall (that is, cancer as a composite outcome). The study emphasises that interpreting blood test results after age and sex adjustment improves triage for cancer investigation in patients presenting with UWL and that longitudinal trends can offer further incremental discrimination in select test–cancer pairings.
The authors note important limitations arising from using routinely collected clinical data. First, there is selection bias in who received blood tests in real-world care; the modelling strategy did not explicitly account for this testing bias. Second, trend analyses required at least two tests within each trend window, restricting trend estimation to patients with repeat testing and potentially reducing generalisability. Third, because the study uses observational electronic health records, timing and frequency of testing reflect clinical practice rather than a standardized protocol.
Other potential limitations—such as calibration of models, external validation, or the influence of comorbidities on trends—were not detailed in the summary and would require review of the full report for additional context.
For patients who present to primary care with unexpected weight loss, clinicians should consider interpreting blood test results adjusted for age and sex to improve selection for cancer investigation. Where prior test results are available, evaluating longitudinal blood test trends may further enhance discrimination in certain test–cancer combinations (for example, MCV, WBC/neutrophils, AST, RBC, haematocrit, and PLR for specific cancer sites). However, application of trend-based approaches will be limited to patients with prior testing and requires awareness of selection bias inherent in observational testing patterns. Further work to translate trend models into clinical decision support and to evaluate prospective performance and impact was not reported in the source article and would be needed before routine implementation.