This retrospective cohort study assessed whether monitoring historical blood test trends provides better discrimination between cancer and non-cancer than interpreting a single recent abnormal result in adults presenting to primary care with unexpected weight loss (UWL). The primary objective was to compare discrimination for undiagnosed cancer provided by blood test abnormalities co-occurring with UWL versus the trajectory of blood test values over preceding years.
The study population comprised patients aged 18 years or older with a record of UWL between 01/01/2000 and 31/12/2018 identified in the Clinical Practice Research Datalink. Overall and site-specific cancer diagnoses were ascertained through linked National Cancer Registrations Data. The final cohort included 275,205 patients, of whom 5.0% (n = 13,798) were diagnosed with cancer.
Researchers extracted all quantitative results for 26 blood tests performed prior to the date of UWL. Reported summary testing patterns indicated that the median number of blood tests per person over the 10-year period varied by test type, with medians reported as low as 2 tests for some and up to 4 tests for others, together with interquartile ranges. The median time between an individual’s first and last blood test over the observation period was 5.2 years (IQR 1.5–8.2) for patients subsequently diagnosed with cancer and 4.3 years (IQR 1.0–7.7) for those not diagnosed with cancer. The median interval from the last blood test to the recorded UWL was 0.1 years (IQR 0.0–0.6) for cancer cases and 0.3 years (IQR 0.0–1.1) for patients without cancer.
Two complementary modeling strategies were applied. Cox proportional hazards models estimated cancer risk associated with blood test abnormalities observed at or around the time of UWL (for example, low albumin or raised platelets). Joint models assessed longitudinal trends in each blood test over multiple retrospective windows (1-, 3-, 5-, and 10-year periods before UWL). Both unadjusted and age- and sex-adjusted models were reported. Discrimination was measured 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 compared with unadjusted models. Across comparisons, adjusted trend models produced higher AUCs than the equivalent adjusted abnormality models on 34 occasions. The highest reported AUC for a trend model was 0.82 (95% CI 0.81–0.83). These findings indicate that, for certain blood tests and cancer sites, historical trajectories added discriminative value beyond a single abnormal result.
The study identified specific combinations of test trends and cancer sites where trend-based discrimination was notably improved after adjustment. Examples reported included:
These specific pairings indicate that not all tests benefit equally from trend analysis; benefits were observed for selected test–cancer combinations.
The cohort-level timing metrics show that patients who developed cancer had slightly longer median intervals between first and last tests and a shorter median interval between the last test and UWL onset compared with those without cancer. The distribution of tests and timing underscores that historical data were available for many patients, but the number and spacing of tests varied by test type and by patient.
The authors reported two principal limitations related to the modeling strategy. First, the analyses did not account for potential selection bias in who was offered and received testing (that is, bias in who was tested). Second, the trend analysis required at least two test results within each trend window, restricting trend estimation to patients with repeat testing. The source abstract does not report how these limitations might quantitatively affect the results, and further details of sensitivity analyses or approaches to mitigate selection bias were not reported in the abstract.
The study concludes that interpreting blood test results in patients with unexpected weight loss after age and sex adjustment improves triage for cancer investigation. Additionally, evaluating historical blood test trends can provide further discriminative value for some specific test–cancer combinations beyond adjusted single-test abnormalities. The findings support the potential utility of incorporating trend information into primary care risk assessment for cancer in patients presenting with UWL, while acknowledging limitations related to testing selection and availability of serial measurements.