Blood test trend versus single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss: A retrospective cohort study

Brian D Nicholson; Clare R Bankhead; Sufen Zhu; Jason Lee Oke; Cynthia Wright Drakesmith; Rafael Perera; Frederick D Richard Hobbs; Pradeep S Virdee

doi:10.1371/journal.pmed.1004956

. 2026 Sep 17;23(9):e1004956. doi: 10.1371/journal.pmed.1004956

Blood test trend versus single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss: A retrospective cohort study

Brian D Nicholson ^1,^*, Clare R Bankhead ¹, Sufen Zhu ¹, Jason Lee Oke ¹, Cynthia Wright Drakesmith ¹, Rafael Perera ¹, Frederick D Richard Hobbs ^1,², Pradeep S Virdee ¹

Editor: Eleanor L Watts³

PMCID: PMC13585255 PMID: 42752402

Abstract

Background

Unexpected weight loss (UWL) is a non-specific cancer symptom. Abnormalities in the most recent blood test contribute to cancer referral decisions in primary care. Interpreting historical changes (trends) in blood tests could enhance discrimination for undiagnosed cancer. We aimed to assess the discrimination of blood test trend for undiagnosed cancer and compare this to that of blood test abnormality in patients presenting to primary care with UWL.

Methods and findings

We conducted a retrospective cohort study of patients presenting to English primary care with UWL aged ≥18 years between 01/01/2000 and 31/12/2018 in the Clinical Practice Research Datalink. We extracted all quantitative results for 26 blood tests prior to the date of UWL. Overall and site-specific cancer diagnosis was ascertained using linked National Cancer Registrations Data. Cox models, unadjusted and adjusted for age and sex, estimated cancer risk based on blood test abnormalities (such as low albumin or raised platelets) co-occurring with UWL and joint models assessed trends in each blood test over 1-, 3-, 5-, and 10-year periods prior to UWL. The area under the curve (AUC) was used to assess discrimination, with 95% confidence intervals (CIs). Amongst 275,205 UWL patients, 5.0% (n = 13,798) with cancer, the median (interquartile range (IQR)) number of blood tests per person over the 10 years ranged from 2 (IQR [12–3]) for 3 types of tests to 4 (IQR [2–7]) for 12 types of tests. The median (IQR) time between the first and last blood test per person was 5.2 (IQR [1.5–8.2]) years cases and 4.3 (IQR [1.0–7.7]) years for those cancer-free. The median time between the last blood test and UWL per person was 0.1 (IQR [0.0–0.6]) years for cases and 0.3 (IQR [0.0–1.1]) years for those cancer-free. Adjusted blood test abnormalities and trends were more discriminative than the unadjusted equivalent. Adjustment resulted in higher AUCs for trend compared to the equivalent blood test abnormality on 34 occasions, with highest AUC of 0.82 (95% CI [0.81, 0.83]). This included 26 trends for cancer overall, mean cell volume trend for bowel cancer and lymphoma, white blood count and neutrophil trend for lung cancer, and aspartate aminotransferase, red blood cell count, haematocrit, and platelet-to-lymphocyte ratio trend for prostate cancer. A limitation of our modelling strategy is that it did not take into account the bias in who was tested and our trend analysis was also restricted to patients with 2 or more tests in each trend window.

Conclusions

Blood test abnormalities should be interpreted after age and sex adjustment to improve triage for cancer investigation in patients attending primary care with UWL. Evaluating historical trend offers further discrimination after adjustment for some test-cancer combinations.

Author summary

Why was this study done?

Abnormal blood test results, such as low haemoglobin or raised platelets, support cancer risk assessment in primary care.
Monitoring trends (changes over time) in blood test results could enhance cancer risk assessment.
No study has assessed the potential of blood test trends in comparison to blood test abnormalities in patients who present with non-specific symptoms, such as unexpected weight loss (UWL).

What did the researchers do and find?

We conducted a cohort study using electronic health records data for 275,205 patients with UWL attending English primary care.
We compared the discrimination of trends in 26 blood tests prior to UWL for a new cancer diagnosis with the abnormality for the same blood test that is used in clinical practice.
We found that age- and sex-adjusted blood test abnormalities and trends were more discriminative than unadjusted blood test trends and abnormalities, and for 34 test-cancer combinations adjusted trends were more discriminative than the equivalent adjusted blood test abnormality, including 26 trends for cancer overall, mean cell volume trend for bowel cancer and lymphoma, white blood count and neutrophil trend for lung cancer, and aspartate aminotransferase, red blood cell count, haematocrit, and platelet-to-lymphocyte ratio trend for prostate cancer.

What do these findings mean?

Blood test results should be adjusted by age and sex to improve patient selection for cancer investigation in patients attending primary care with UWL.
Blood test trends over repeat tests could further enhance patient selection in some test-cancer combinations.
Our analysis uses routinely collected data so is limited to when patients had blood tests in clinical practice and to patients who had blood tests.

In this retrospective cohort study, Brian D. Nicholson and colleagues evaluate the potential of blood test trends in comparison to blood test abnormalities for cancer risk assessment in patients who present with non-specific symptoms, such as unexpected weight loss.

Introduction

The current evidence base for using non-specific blood tests to risk stratify for cancer in primary care is almost entirely limited to unadjusted blood test abnormalities of single tests [1–3]. For most blood test abnormalities, the associated risk of a single cancer is too low to warrant urgent cancer investigation. Low albumin, raised platelets, raised calcium, and raised inflammatory markers increase the risk of cancer overall above the 3% threshold recommended by the National Institute for Health and Care Excellence (NICE) for urgent investigation for cancer in England and Wales [1]. Our recent work identified the blood test abnormalities associated with increased risk of cancer overall in patients with unexpected weight loss (UWL) [4,5], who often experience diagnostic challenges [6–8], but blood tests are common [9]. For non-specific symptoms such as UWL, these abnormalities give general practitioners (GPs) an indication that cancer may be present, but uncertainty remains over which cancer(s) should be investigated.

Current primary care clinical guidelines focus on selecting patients for cancer investigation based on combinations of risk factors, signs, symptoms, and blood test abnormalities [1]. Monitoring temporal changes (or trends) in blood tests may enhance cancer risk stratification over and above acting on the only most recent blood test result [10–12]. A patient with a low-normal haemoglobin may not be considered high-risk if the haemoglobin result is interpreted in relation to a binary threshold, but a low-normal result following a period of high-normal results may represent an opportunity for cancer investigation. However, we have found no study that compared the cancer risk associated with blood test trends to the current standard of using the most recent blood test abnormality and no study exploring the role of blood test trend in patients presenting with non-specific symptoms in primary care, such as UWL [13].

Our aim was to explore the discrimination of blood test trends for a new diagnosis of cancer compared to the equivalent blood test abnormality in patients presenting with UWL to primary care, by cancer type and time period of trend.

Methods

Study reporting follows the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE), REporting of studies Conducted using Observational Routinely-collected health Data (RECORD), and Standards for Reporting of Diagnostic Accuracy Studies (STARD) guidelines. Populated checklists for each are in the Supplementary files.

Study design

A retrospective cohort study was performed using data from 01/01/2000 to 31/12/2018 in the English primary care Clinical Practice Research Datalink (CPRD) Aurum database. The data were linked at a patient-level to the National Cancer Registration and Analysis Service (NCRAS) Cancer Registration, Office of National Statistics (ONS) Death Registration, Hospital Episode Statistics (HES), and Small Area Index of Multiple Deprivation (IMD) databases. The end date of 2018 coincided with the most recent NCRAS data download for cancer outcomes.

Participants

We included patients with UWL, with their first coded UWL selected. Patients were aged at least 18 years at UWL, were eligible for linkage to NCRAS, HES, and ONS, and had at least 12 months of registration with their practice and at least one blood test within the 10-year period prior to UWL. Patients with a prescription of weight-reducing medication (orlistat) or bariatric surgery in the six months prior to UWL or a history of cancer prior to UWL were excluded. Systematized Nomenclature of Medicine-Clinical Terms (SNOMED-CT) codes are in https://github.com/PradeepVirdee/CPRD_UWL_TrendVsAbno_Cancer [14].

Test methods

Outcome.

The primary outcome was any primary cancer diagnosed within six months following the UWL. Six months was chosen because previous research demonstrated that the risk of cancer diagnosis after six months following UWL returned to the background levels expected at the patient’s age and sex [15]. Non-melanoma skin cancer, in situ, benign, ill-defined, or uncertain cancers were excluded. Cancers were also categorised into 20 groupings based on cancer site. Cancer cases were determined from the NCRAS database, with additional cases identified as the earliest diagnosis among CPRD, HES, and ONS. Cancer International Classification of Diseases, 10th Revision (ICD10) codes were in the format ‘C###’ and SNOMED-CT codes are in https://github.com/PradeepVirdee/CPRD_UWL_TrendVsAbno_Cancer [14].

Time to diagnosis was defined as the number of days from UWL to cancer diagnosis within 6 months. Patients without a diagnosis in this 6-month period were censored at the earliest of death, de-registration with the practice, data cut date for the practice, 31/12/2018 (study end date), or six months following UWL.

Clinical features.

Sociodemographic features closest recorded prior to UWL were extracted for each patient. Pre-existing comorbidities were identified. Co-occurring signs and symptoms, included in NICE guidance [1], closest to UWL in the three months before to one month after the UWL date were identified. We extracted blood tests commonly available in primary care: components of the full blood count, liver function tests, and inflammatory markers (Table A in S1 Appendix). These tests were decided a priori, as our goal was to study commonly available tests, to maximise uptake and impact of abnormality and trend, and tests that are triggered by UWL presentation. Continuous results of these blood tests available over 10 years prior to UWL were obtained, with values less than zero dropped, as these were data entry errors, and standardised to the same unit of measure following the approach used in previous work [16]. Previous research has highlighted the potential of inflammatory marker scores to improve cancer risk stratification in patients with UWL, so we also derived the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), and modified Glasgow Prognostic ratio (mGPR) [17]. Each continuous blood test result was converted into a categorical result (normal/abnormal) using reference ranges used in clinical practice (Table A in S1 Appendix), as the categorical score is used in clinical practice: neutrophil-to-lymphocyte score (NLS), platelet-to-lymphocyte score (PLS), and modified Glasgow Prognostic Score (mGPS), referring to them as tests hereafter.

Analysis

Data analysis.

Analyses were planned a priori and performed using RStudio (R version 4.5.2). Mean with standard deviation (SD) or median with range/interquartile range (IQR) were used to describe continuous data and counts and proportions were used to describe categorical data. Locally Weighted Scatterplot Smoothing (LOWESS) smoothing was used to describe trends in blood tests over time by age and sex graphically.

Joint modelling of longitudinal and time-to-event data was used to assess trends for cancer diagnosis using the JoineRML package. Joint modelling uses mixed-effects modelling for trends and a linked Cox model to provide hazard ratios (HRs) for outcomes [18,19]. A model was developed for each blood test. Mixed-effects models included time from blood test to UWL (years), age at UWL (years), and sex as fixed effects. An interaction between age and time and between age and sex was included to allow for differences in blood test trends between age groups and sex. Each patient was modelled using a random intercept and time to UWL using a random slope, with an unstructured covariance matrix to account for correlation in repeated measures and restricted maximum likelihood estimation. Time to UWL and age at UWL were modelled using piecewise linear splines, with knots at 6 months and 1, 2, and 3 years before UWL for each test. Linear splines and knot locations were chosen based on visual inspection of LOWESS plots and comparison of Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) of models using fractional polynomials, restricted cubic splines, linear splines, and different knot locations. The linked Cox model then included the blood test trend (as latent shared random effects from the mixed-effects component) as a predictor for cancer diagnosis.

Univariable analysis included only the trend in the joint model, using blood test trends measured over 1, 3, 5, and 10 years before UWL to assess how the association and discrimination change based on how much pre-UWL trend was used. Our LOWESS plots suggest that cancer-related trends begin to diverge from those in patients without cancer around 3–5 years before diagnosis, so we opted to investigate 1-, 3-, and 5-year trend windows prior to UWL. We expected that earlier blood tests were unlikely to be helpful and chose to assess this also by including a 10-year trend window. In each longitudinal window, we included patients with at least two blood tests to derive a trend, where at least one was co-occurring with UWL (measured in the 3 months prior). The multivariable analysis subsequently adjusted cancer risk for age at UWL (years) and sex, as these are easily accessible in primary care systems. Age and sex were available for all patients. Other data, such as ethnicity and BMI, were considered only to describe the sample, with patients with missing data included in a “Missing” category.

Cox modelling was used to assess the association between blood test abnormality co-occurring with UWL and subsequent cancer, for comparison with trend. Univariable Cox models included blood test abnormality (high = yes/no or low = yes/no, depending on the test) and multivariable models adjusted for age at UWL (years) and sex. Blood test abnormality on the co-occurring test was defined using reference ranges reported in the Oxford Handbook of Clinical Medicine (Table A in S1 Appendix) [20]. For each blood test in the 1-year trend models and abnormality models, we examined the proportional hazards assumption using plots of Schoenfeld residuals—no model violated the proportional hazards assumption (plots are available from the authors).

Performance of trend and abnormality was assessed using the fixed-time area under the curve (AUC) at 6 months following UWL as a high-level summary metric of discrimination across multiple comparisons. As exploratory analyses, we looked at the AUC of abnormality and trend by cancer site and by the number of repeat tests used to derive the trend.

Sample size.

We performed a sample size calculation for an AUC for five blood tests (albumin, AST, haemoglobin, CRP, and plasma viscosity—AST and plasma viscosity cohorts are the smallest cohorts) in the 1-year trend window and 2+ blood test cohort (Table B in S1 Appendix). For each test, the null AUC was 0.5 and target AUC was 0.7. A two-sided 0.048% alpha, accounting for multiple testing (a Bonferroni adjustment to a 5% significance level considering 104 test-trend window combinations), and 80% power was used. The calculation also uses the ratio of the number of patients without cancer to the number of patients with cancer, which we based on our observed data. Based on these inputs, 578–625 patients, depending on the test, were required. We repeated the calculation using 90% and 99% power due to our large sample size, with the latter increasing the sample size to 1,055–1,125.

Results

Summary of patient data

We included 275,205 eligible patients with UWL (Fig 1). Mean age at UWL was 61.8 (SD [20.1]) and 58.0% (n = 159,616) were female (Table 1). The most commonly available blood tests were haemoglobin (95.9%, n = 264,012 patients), white blood cell count (WBC) (95.3%, n = 262,153), and platelets (95.2%, n = 261,985) (Table C in S1 Appendix). The least commonly available were plasma viscosity (9.8%, n = 27,022) and mean platelet volume (MPV) (17.5%, n = 43,257). Among those included in the trend analysis (those with 2+ tests) (sample sizes are in Table C in S1 Appendix), there was a median of 2 (IQR [1, 3]) blood tests per person over the 10 years for three types of tests and 4 (IQR [2, 7]) for 12 types of tests (results are available on GitHub) [14]. The median number of blood tests per patient generally increased with increasing age, body mass index (BMI), number of co-morbidities, and presence of symptoms. The median number of tests per patient was comparable between men and women and between IMD groups. The most common blood test abnormalities on the most recent blood test were low red blood cell count (RBC) (12.4%, n = 34,156 patients), low haemoglobin (8.7%, n = 23,919), and high C-reactive protein (CRP) (7.7%, n = 21,241) (Table A in S1 Appendix).

Table 1. Summary (n (%)) of patient demographics and co-occurring clinical features at UWL.

Patient data	Overall	Cancer	No cancer
Sex
Female	159,616 (58.0%)	5,474 (39.7%)	154,142 (59.0%)
Male	115,589 (42.0%)	8,324 (60.3%)	107,265 (41.0%)
Age (years)
18–39	47,027 (17.1%)	116 (0.8%)	46,911 (17.9%)
40–49	30,307 (11.0%)	314 (2.3%)	29,993 (11.5%)
50–59	36,759 (13.4%)	1,122 (8.1%)	35,637 (13.6%)
60–69	42,529 (15.5%)	2,752 (19.9%)	39,777 (15.2%)
70–79	56,650 (20.6%)	4,944 (35.8%)	51,706 (19.8%)
80+	61,933 (22.5%)	4,550 (33.0%)	57,383 (22.0%)
Ethnicity
Asian	13,407 (4.9%)	241 (1.7%)	13,166 (5.0%)
Black	9,004 (3.3%)	291 (2.1%)	8,713 (3.3%)
Mixed	1,831 (0.7%)	39 (0.3%)	1,792 (0.7%)
Other	3,266 (1.2%)	100 (0.7%)	3,166 (1.2%)
White	230,196 (83.6%)	12,084 (87.6%)	218,112 (83.4%)
Missing	17,501 (6.4%)	1,043 (7.6%)	16,458 (6.3%)
IMD group
1 (least deprived)	49,128 (17.9%)	2,579 (18.7%)	46,549 (17.8%)
2	52,739 (19.2%)	2,709 (19.6%)	50,030 (19.1%)
3	52,283 (19.0%)	2,534 (18.4%)	49,749 (19.0%)
4	55,886 (20.3%)	2,779 (20.1%)	53,107 (20.3%)
5 (most deprived)	64,746 (23.5%)	3,170 (23.0%)	61,576 (23.6%)
Missing	423 (0.2%)	27 (0.2%)	396 (0.2%)
Smoking status
Current	89,297 (32.4%)	4,647 (33.7%)	84,650 (32.4%)
Past smoker	67,342 (24.5%)	4,463 (32.3%)	62,879 (24.1%)
Non-smoker	43,651 (15.9%)	1,972 (14.3%)	41,679 (15.9%)
Missing	74,915 (27.2%)	2,716 (19.7%)	72,199 (27.6%)
Alcohol status
Current	131,453 (47.8%)	7,397 (53.6%)	124,056 (47.5%)
Past drinker	293 (0.1%)	16 (0.1%)	277 (0.1%)
Non-drinker	76,037 (27.6%)	3,537 (25.6%)	72,500 (27.7%)
Missing	67,422 (24.5%)	2,848 (20.6%)	64,574 (24.7%)
BMI
Underweight	17,059 (6.2%)	617 (4.5%)	16,442 (6.3%)
Normal	122,654 (44.6%)	5,622 (40.7%)	117,032 (44.8%)
Overweight	65,940 (24.0%)	4,305 (31.2%)	61,635 (23.6%)
Obese	38,218 (13.9%)	1,926 (14.0%)	36,292 (13.9%)
Missing	31,334 (11.4%)	1,328 (9.6%)	30,006 (11.5%)
No. of comorbidities
0	38,364 (13.9%)	1,431 (10.4%)	36,933 (14.1%)
1	53,321 (19.4%)	2,538 (18.4%)	50,783 (19.4%)
2	53,569 (19.5%)	2,838 (20.6%)	50,731 (19.4%)
3	45,012 (16.4%)	2,478 (18.0%)	42,534 (16.3%)
4	33,486 (12.2%)	1,886 (13.7%)	31,600 (12.1%)
5+	51,453 (18.7%)	2,627 (19.0%)	48,826 (18.7%)
Clinical feature (10 most common)
Appetite Loss	7,707 (2.8%)	773 (5.6%)	6,934 (2.7%)
Cough	19,514 (7.1%)	1,215 (8.8%)	18,299 (7.0%)
Dyspepsia	6,100 (2.2%)	581 (4.2%)	5,519 (2.1%)
Fatigue	10,268 (3.7%)	698 (5.1%)	9,570 (3.7%)
Infection Chest	13,614 (4.9%)	918 (6.7%)	12,696 (4.9%)
Pain Abdomen	15,561 (5.7%)	1,530 (11.1%)	14,031 (5.4%)
Pain Back	15,953 (5.8%)	1,034 (7.5%)	14,919 (5.7%)
Pain Chest	7,341 (2.7%)	523 (3.8%)	6,818 (2.6%)
Shortness Of Breath	16,615 (6.0%)	1,185 (8.6%)	15,430 (5.9%)
Urinary Tract Infection	8,615 (3.1%)	506 (3.7%)	8,109 (3.1%)
Total	275,205	13,798	261,407

Open in a new tab

Abbreviations: IMD, index of multiple deprivation; BMI, body mass index.

There were 13,798 (5.0%) patients diagnosed with cancer within 6 months following UWL, with lung (1.13%, n = 3,120) and colorectal (0.78%, n = 2,150) most commonly diagnosed (Table 2). There were 6.5% (n = 17,909) of patients who were censored before reaching the end of 6-month follow-up (4.4%, n = 12,221 deaths; 2.1%, n = 5,688 de-registration with the practice, data cut date for the practice, 31/12/2018, or 6 months following UWL). The median time between the first and last blood test per person was 5.2 (IQR [1.5, 8.2]) years for patients with subsequent cancer and 4.3 (IQR [1.0, 7.7]) years for patients without (Fig A in S1 Appendix). The median time between the last blood test and UWL per person was 0.1 (IQR [0.0, 0.6]) years (or 1.2 months) for patients with subsequent cancer and 0.3 (IQR [0.0, 1.1]) years (or 3.6 months) for patients without.

Table 2. Summary of cancer diagnosis within 6 months following UWL.

Cancer in 6 months	Cancer in 6 months
Yes	13,798 (5.0%)
No	261,407 (95.0%)
Bladder	356 (0.13%)
Bowel	2,170 (0.79%)
Brain	68 (0.02%)
Breast	399 (0.14%)
Head and neck	218 (0.08%)
Kidney	532 (0.19%)
Leukaemia	291 (0.11%)
Liver	369 (0.13%)
Lung	3,120 (1.13%)
Lymphoma	722 (0.26%)
Melanoma of skin	85 (0.03%)
Myeloma	232 (0.08%)
Oesophagus	773 (0.28%)
Other	732 (0.27%)
Ovarian	244 (0.09%)
Pancreas	1,226 (0.45%)
Prostate	1,197 (0.43%)
Stomach	920 (0.33%)
Thyroid	51 (0.02%)
Uterine	93 (0.03%)

Open in a new tab

Abbreviations: UWL, unexpected weight loss.

Graphs of blood test trends over the 10-year study period are in Figs 2 and B–AA in S1 Appendix. Trends in patients with cancer began to diverge from patients without cancer around 3–5 years before the UWL date. Divergent trends were on average confined within the normal reference range, except for haemoglobin and RBC in patients aged 70+ years and inflammatory markers CRP, erythrocyte sedimentation rate (ESR), and plasma viscosity. Haematocrit and red cell distribution width (RDW) trends also exceeded the normal range limit in younger adults aged 50 years or below.

Fig 2 — LOWESS trends are presented, which represent the observed population-level trend in the study sample. The overlaid grey box is the reference range used in practice for that for that test. Abbreviations: mGPR, modified Glasgow Prognostic ratio; UW, unexpected weight loss.

Association between blood test trend and cancer diagnosis

Unadjusted analyses.

Unadjusted blood test abnormalities for 25 (96.2% of 26) blood tests were associated with cancer overall (95% confidence intervals (CIs) did not contain 1), with no association for alanine transaminase (ALT) (Table 3). However, unadjusted blood test abnormalities were poorly discriminative for cancer overall, with AUCs ranging from 0.5 AUC for mean cell haemoglobin concentration (MCHC) to 0.63 AUC for plasma viscosity (Table D in S1 Appendix).

Table 3. Hazard ratio (95% CI) of blood test abnormality and trend for increased 6-month cancer risk following UWL.

Blood test¹	Model²	Abnormality	1-year Trend	3-year Trend	5-year Trend	10-year Trend
Albumin (g/L) declining	Unadjusted	2.622 (2.492, 2.758)	1.076 (1.067, 1.086)	1.062 (1.050, 1.073)	1.053 (1.041, 1.064)	1.044 (1.032, 1.056)
Albumin (g/L) declining	Adjusted	1.951 (1.853, 2.055)	1.081 (1.068, 1.094)	1.074 (1.059, 1.088)	1.067 (1.053, 1.082)	1.054 (1.038, 1.070)
Alkaline phosphatase (iu/L) increasing	Unadjusted	2.210 (2.108, 2.317)	1.006 (1.006, 1.007)	1.006 (1.005, 1.007)	1.005 (1.004, 1.006)	1.005 (1.004, 1.006)
Alkaline phosphatase (iu/L) increasing	Adjusted	1.933 (1.844, 2.027)	1.007 (1.006, 1.008)	1.006 (1.005, 1.008)	1.006 (1.005, 1.007)	1.006 (1.004, 1.007)
ALT (iu/L) declining	Unadjusted	1.018 (0.951, 1.089)	1.004 (1.000, 1.008)	1.004 (0.999, 1.008)	1.004 (1.000, 1.009)	1.002 (0.998, 1.007)
ALT (iu/L) declining	Adjusted	1.135 (1.060, 1.215)	1.005 (1.000, 1.010)	1.002 (0.996, 1.008)	1.004 (0.998, 1.010)	1.000 (0.994, 1.007)
AST (iu/L) declining	Unadjusted	1.368 (1.191, 1.571)	1.002 (0.994, 1.009)	1.006 (0.999, 1.012)	1.003 (0.998, 1.009)	1.004 (0.998, 1.009)
AST (iu/L) declining	Adjusted	1.428 (1.242, 1.640)	1.000 (0.992, 1.008)	1.001 (0.994, 1.009)	1.001 (0.994, 1.009)	1.001 (0.993, 1.008)
Bilirubin (umol/L) declining	Unadjusted	1.091 (1.016, 1.172)	1.032 (1.022, 1.041)	1.019 (1.010, 1.029)	1.019 (1.010, 1.029)	1.019 (1.009, 1.030)
Bilirubin (umol/L) declining	Adjusted	1.258 (1.171, 1.352)	1.020 (1.009, 1.032)	1.022 (1.010, 1.034)	1.022 (1.010, 1.035)	1.021 (1.008, 1.034)
RBC (10¹²/L) declining	Unadjusted	1.948 (1.872, 2.026)	1.004 (0.959, 1.053)	1.040 (0.989, 1.092)	1.020 (0.970, 1.073)	1.022 (0.968, 1.078)
RBC (10¹²/L) declining	Adjusted	1.203 (1.153, 1.255)	1.025 (0.962, 1.092)	1.030 (0.970, 1.094)	1.009 (0.950, 1.071)	1.006 (0.945, 1.071)
Haemoglobin (g/L) declining	Unadjusted	2.768 (2.660, 2.880)	1.017 (1.015, 1.020)	1.017 (1.014, 1.020)	1.012 (1.009, 1.015)	1.013 (1.010, 1.016)
Haemoglobin (g/L) declining	Adjusted	1.786 (1.712, 1.863)	1.017 (1.013, 1.021)	1.015 (1.012, 1.019)	1.014 (1.011, 1.018)	1.014 (1.010, 1.017)
Haematocrit (L/L) declining	Unadjusted	2.093 (2.004, 2.186)	1.001 (0.993, 1.008)	1.002 (0.993, 1.011)	1.000 (0.991, 1.008)	1.006 (0.997, 1.015)
Haematocrit (L/L) declining	Adjusted	1.738 (1.663, 1.817)	1.001 (0.991, 1.011)	1.006 (0.994, 1.018)	1.007 (0.997, 1.017)	1.003 (0.992, 1.014)
MCH (pg) declining	Unadjusted	2.359 (2.234, 2.491)	1.135 (1.119, 1.151)	1.120 (1.101, 1.139)	1.121 (1.100, 1.142)	1.098 (1.078, 1.119)
MCH (pg) declining	Adjusted	2.510 (2.377, 2.651)	1.140 (1.120, 1.161)	1.160 (1.136, 1.183)	1.127 (1.104, 1.152)	1.119 (1.094, 1.143)
MCHC (g/L) declining	Unadjusted	1.396 (1.285, 1.517)	1.001 (1.000, 1.001)	1.001 (1.000, 1.001)	1.001 (1.000, 1.002)	1.000 (0.999, 1.000)
MCHC (g/L) declining	Adjusted	1.437 (1.322, 1.561)	1.001 (1.000, 1.002)	1.000 (0.999, 1.001)	1.001 (1.000, 1.002)	1.001 (1.000, 1.001)
MCV (fL) declining	Unadjusted	2.973 (2.676, 3.304)	1.053 (1.047, 1.058)	1.049 (1.043, 1.056)	1.031 (1.024, 1.038)	1.026 (1.018, 1.033)
MCV (fL) declining	Adjusted	3.463 (3.117, 3.848)	1.055 (1.047, 1.063)	1.047 (1.038, 1.055)	1.053 (1.044, 1.062)	1.048 (1.038, 1.057)
RDW (%) increasing	Unadjusted	1.847 (1.743, 1.957)	1.071 (1.046, 1.095)	1.078 (1.056, 1.100)	1.063 (1.038, 1.089)	1.056 (1.031, 1.081)
RDW (%) increasing	Adjusted	1.485 (1.400, 1.575)	1.062 (1.037, 1.087)	1.076 (1.048, 1.105)	1.064 (1.034, 1.095)	1.066 (1.037, 1.095)
Platelets (10⁹/L) increasing	Unadjusted	2.910 (2.760, 3.068)	1.005 (1.004, 1.005)	1.004 (1.003, 1.004)	1.004 (1.003, 1.004)	1.004 (1.003, 1.005)
Platelets (10⁹/L) increasing	Adjusted	2.973 (2.819, 3.135)	1.005 (1.004, 1.005)	1.004 (1.003, 1.005)	1.004 (1.003, 1.004)	1.003 (1.003, 1.004)
MPV (fL) declining	Unadjusted	1.420 (1.141, 1.767)	1.136 (1.062, 1.215)	1.078 (1.024, 1.135)	1.080 (1.025, 1.139)	1.056 (1.001, 1.114)
MPV (fL) declining	Adjusted	1.287 (1.034, 1.602)	1.149 (1.070, 1.235)	1.105 (1.040, 1.174)	1.093 (1.029, 1.161)	1.094 (1.030, 1.161)
WBC (10⁹/L) increasing	Unadjusted	2.486 (2.357, 2.622)	1.197 (1.177, 1.217)	1.176 (1.152, 1.200)	1.152 (1.126, 1.178)	1.179 (1.154, 1.203)
WBC (10⁹/L) increasing	Adjusted	2.471 (2.342, 2.606)	1.221 (1.194, 1.248)	1.194 (1.164, 1.224)	1.178 (1.148, 1.208)	1.161 (1.129, 1.193)
Basophil count (10⁹/L) increasing	Unadjusted	1.093 (1.049, 1.140)	3.384 (0.929, 12.321)	3.025 (0.790, 11.583)	9.487 (2.425, 37.110)	3.860 (1.055, 14.124)
Basophil count (10⁹/L) increasing	Adjusted	1.096 (1.052, 1.143)	1.936 (0.389, 9.628)	1.017 (0.197, 5.263)	3.028 (0.630, 14.560)	1.824 (0.371, 8.970)
Eosinophil count (10⁹/L) increasing	Unadjusted	1.125 (1.040, 1.217)	1.175 (0.762, 1.813)	1.012 (0.713, 1.437)	1.030 (0.715, 1.481)	1.103 (0.752, 1.616)
Eosinophil count (10⁹/L) increasing	Adjusted	1.013 (0.936, 1.096)	1.513 (0.984, 2.326)	1.460 (0.935, 2.283)	1.613 (1.017, 2.558)	1.321 (0.835, 2.092)
Lymphocyte count (10⁹/L) increasing	Unadjusted	1.694 (1.586, 1.811)	1.043 (0.980, 1.111)	1.118 (1.045, 1.196)	1.121 (1.043, 1.204)	1.182 (1.106, 1.262)
Lymphocyte count (10⁹/L) increasing	Adjusted	1.191 (1.113, 1.273)	1.043 (0.954, 1.139)	1.096 (1.003, 1.198)	1.110 (1.022, 1.206)	1.061 (0.972, 1.158)
Monocyte count (10⁹/L) increasing	Unadjusted	2.474 (2.373, 2.580)	4.869 (4.024, 5.892)	4.097 (3.294, 5.096)	4.878 (3.891, 6.116)	3.812 (3.028, 4.798)
Monocyte count (10⁹/L) increasing	Adjusted	1.988 (1.907, 2.074)	6.628 (5.151, 8.528)	5.345 (4.073, 7.015)	5.120 (3.869, 6.777)	4.341 (3.277, 5.752)
Neutrophil count (10⁹/L) increasing	Unadjusted	2.659 (2.528, 2.796)	1.255 (1.230, 1.280)	1.233 (1.203, 1.263)	1.205 (1.173, 1.239)	1.198 (1.164, 1.233)
Neutrophil count (10⁹/L) increasing	Adjusted	2.506 (2.383, 2.636)	1.251 (1.218, 1.286)	1.226 (1.188, 1.265)	1.217 (1.178, 1.258)	1.230 (1.189, 1.272)
CRP (mg/L) increasing	Unadjusted	3.020 (2.820, 3.234)	1.021 (1.019, 1.024)	1.023 (1.021, 1.025)	1.023 (1.021, 1.026)	1.024 (1.020, 1.027)
CRP (mg/L) increasing	Adjusted	2.472 (2.304, 2.653)	1.021 (1.018, 1.023)	1.023 (1.020, 1.026)	1.024 (1.021, 1.026)	1.024 (1.021, 1.026)
ESR (mm/h) increasing	Unadjusted	3.079 (2.900, 3.270)	1.021 (1.019, 1.024)	1.023 (1.021, 1.025)	1.023 (1.021, 1.026)	1.023 (1.020, 1.026)
ESR (mm/h) increasing	Adjusted	2.620 (2.466, 2.783)	1.021 (1.018, 1.023)	1.023 (1.020, 1.026)	1.024 (1.021, 1.026)	1.024 (1.021, 1.027)
Plasma viscosity (mPa.S) increasing	Unadjusted	3.689 (3.173, 4.291)	10.649 (6.252, 18.136)	2.092 (1.393, 3.143)	2.085 (1.671, 2.601)	2.083 (1.635, 2.653)
Plasma viscosity (mPa.S) increasing	Adjusted	3.112 (2.673, 3.623)	9.992 (5.545, 18.005)	9.754 (5.513, 17.256)	15.429 (8.890, 26.775)	25.259 (14.294, 44.635)
NLR increasing	Unadjusted	1.707 (1.596, 1.826)	1.179 (1.153, 1.206)	1.141 (1.108, 1.175)	1.144 (1.107, 1.182)	1.100 (1.064, 1.137)
NLR increasing	Adjusted	1.196 (1.118, 1.281)	1.226 (1.185, 1.268)	1.152 (1.112, 1.193)	1.122 (1.082, 1.164)	1.134 (1.092, 1.179)
PLR increasing	Unadjusted	1.725 (1.611, 1.846)	1.003 (1.002, 1.004)	1.002 (1.002, 1.003)	1.002 (1.001, 1.003)	1.001 (1.000, 1.002)
PLR increasing	Adjusted	1.202 (1.123, 1.288)	1.003 (1.002, 1.004)	1.003 (1.002, 1.003)	1.002 (1.001, 1.003)	1.002 (1.001, 1.003)
mGPR increasing	Unadjusted	2.771 (2.539, 3.024)	1.923 (1.712, 2.159)	2.153 (1.916, 2.418)	2.131 (1.944, 2.336)	2.165 (1.962, 2.388)
mGPR increasing	Adjusted	1.998 (1.827, 2.184)	1.991 (1.713, 2.314)	2.124 (1.891, 2.385)	2.202 (1.967, 2.465)	2.240 (2.000, 2.509)

Open in a new tab

1 HRs are presented for a one-unit change in the blood test per year and the associated hazard rate of cancer within 6 months.

2 Adjusted HRs for trend have been adjusted for age at UWL and sex.

Abbreviations: UWL, unexpected weight loss; ALT, alanine aminotransferase; AST, aspartate aminotransferase; RBC, red blood cell count; MCH, mean cell haemoglobin; MCHC, mean cell haemoglobin concentration; MCV, mean cell volume; RDW, red cell distribution width; MPV, mean platelet volume; WBC, white blood cell count; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; NLR, neutrophil-lymphocyte ratio; PLR, platelet-lymphocyte ratio; mGPR, modified Glasgow Prognostic ratio.

Unadjusted 5-year trends for 21 (80.8% of 26) blood tests were associated with cancer overall (Table 3). The trend was declining for seven (26.9% of 26) tests and increasing for 14 (53.8% of 26). Unadjusted associations were maintained when extending the trend to 10 years, except for mean cell haemoglobin (MCH), and after shortening the trend to 3 and 1 years, except for lymphocyte count over 1-year, and MCHC over 1 and 3 years. HRs for blood test trend were rescaled to represent unit changes per year that are more clinically meaningful (Table E in S1 Appendix).

Unadjusted blood test trends had AUCs for cancer overall ranging from 0.50 (95% CI [0.49, 0.51]) for 10-year ALT to 0.66 (95% CI [0.61, 0.70]) for 1-year plasma viscosity (Table D in S1 Appendix). In general, increasing the trend period resulted in a reduction in the AUC. Unadjusted trends over 1, 3, 5, and 10 years had a higher AUC than the equivalent unadjusted abnormality for 12 (46.2% of 26), 7 (26.9%), 6 (23.1%), and 5 (19.2%) blood tests, respectively (Fig 3).

Fig 3 — This figure shows the AUC of unadjusted and age- and sex-adjusted blood test abnormality and blood test trend. The AUC often favoured unadjusted trend than the unadjusted abnormality, particularly for trend over 1 year prior to UWL, with decreasing AUC as increasing earlier trend prior to UWL was used. The AUC was often comparable between adjusted trend than the adjusted abnormality. Abbreviations: UWL, unexpected weight loss AUC, area under the curve; ALT, alanine aminotransferase; AST, aspartate aminotransferase; RBC, red blood cell count; MCH, mean cell haemoglobin; MCHC, mean cell haemoglobin concentration; MCV, mean cell volume; RDW, red cell distribution width; MPV, mean platelet volume; WBC, white blood cell count; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; PV, plasma viscosity; NLR, neutrophil-lymphocyte ratio; PLR, platelet-lymphocyte ratio; mGPR, modified Glasgow Prognostic ratio.

The largest differences in AUC between unadjusted abnormality and unadjusted trend were observed for mGPR (abnormality: 0.55 (95% CI [0.53, 0.56]); 1-year trend: 0.62 (95% CI [0.59, 0.64])); RBC (abnormality: 0.50 (95% CI [0.49, 0.51]); 1-year trend 0.57 (95% CI [0.56, 0.57])), and mean cell volume (MCV) (abnormality: 0.51 (95% CI [0.51, 0.51]); 1-year trend: 0.58 (95% CI [0.57, 0.60])).

Adjusted analyses.

Adjustment of all blood test abnormalities and trends by age and sex resulted in higher AUCs compared to the equivalent unadjusted estimate (Table D in S1 Appendix). Adjusted blood test abnormalities for 25 (96.2% of 26) blood tests were associated with cancer overall, with no association for eosinophil count (Table 3). Adjusted blood test abnormalities had AUCs for cancer overall ranging from 0.64 (95% CI [0.63, 0.65]) for ALT to 0.71 (95% CI [0.66, 0.75]) for plasma viscosity (Table D in S1 Appendix).

There were 21 (80.8% of 26) adjusted blood test trends over 5 years associated with cancer diagnosis overall, with no association for ALT, aspartate aminotransferase (AST), RBC, haematocrit, or MCHC (Table 3). Associations were often maintained when reducing the trend to 1 and 3 years and extending the trend to 10 years. HRs for blood test trend were rescaled to represent unit changes per year that are more clinically meaningful (Table E in S1 Appendix). Adjusted blood test trends had AUCs for cancer overall ranging from 0.64 (95% CI [0.63, 0.65]) for ALT over 1 year 0.71 (95% CI [0.70, 0.72]) for CRP over 3–10 years (Fig 3).

There were higher AUCs (95% CI) for trend over 1, 3, 5, and 10 years for cancer overall compared to the equivalent adjusted abnormality for 0, 4 (15.4% of 26), 11 (42.3%), and 11 (42.3%) tests, respectively (Fig 4). There was no difference in AUC between the remaining 26 (100.0% of 26), 22 (84.6%), 15 (57.7%), and 15 (57.7%) abnormality and trend combinations, respectively.

Fig 4 — This figure portrays the instances where the AUC favoured the age- and sex-adjusted trend than the adjusted abnormality counterpart for cancer overall but also by cancer site. Only instances for cancer overall and by cancer site in the UWL cohort are shown; instances in subgroups, such as by age or sex, are not shown. Abbreviations: AUC, area under the curve; ALT, alanine aminotransferase; AST, aspartate aminotransferase; RBC, red blood cell count; MCHC, mean cell haemoglobin concentration; MCV, mean cell volume; WBC, white blood cell count; CRP, C-reactive protein; Neutc, neNLR, neutrophil-lymphocyte ratio; PLR, platelet-lymphocyte ratio; mGPR, modified Glasgow Prognostic ratio.

However, differences in the in AUC between adjusted trend and adjusted abnormality were small, with the largest differences observed for mGPR (abnormality: 0.66 (95% CI [0.63, 0.68]); 3-year trend: 0.70 (95% CI [0.69, 0.72])); CRP (abnormality: 0.68 (95% CI [0.66, 0.69]); 3-year trend 0.71 (95% CI [0.70, 0.72])), and MCHC (abnormality: 0.65 (95% CI [0.64, 0.66]); 10-year trend: 0.68 (95% CI [0.67, 0.68])) (Table D in S1 Appendix).

Subgroup analyses

Cancer site.

After adjustment, declining MCV trend was favoured for bowel cancer (0.72 (95% CI [0.70, 0.74])) and lymphoma (0.66 (95% CI [0.63, 0.69])) and rising WBC (0.72 (95% CI [0.70, 0.73])) and neutrophil trend (0.66 (95% CI [0.65, 0.67])) for lung cancer, and AST, RBC, haematocrit, and PLR trend for prostate cancer (Table F in S1 Appendix).

Number of tests.

For each blood test trend, the AUC decreased slightly as the number of blood tests used to derive the 1-year trend increased until 4 tests and remained comparable thereafter (Table G in S1 Appendix). However, confidence intervals grew wider as the number of repeat tests increased. Similar patterns were seen between unadjusted and adjusted analyses.

Discussion

We investigated whether trends in 26 commonly used blood tests might offer improved risk stratification for cancer compared to the most recent blood test abnormality in patients attending primary care with UWL. We explored unadjusted and adjusted associations between blood test abnormalities, and blood test trends over 1, 3, 5, and 10 years, and the diagnosis of cancer overall and by cancer site. Unadjusted analyses demonstrated that blood test abnormalities are weakly associated with cancer and poorly discriminative. Unadjusted blood test trends over 1 and 2 years demonstrated stronger associations with cancer and improved discrimination for cancer compared to unadjusted abnormality. Adjustment for age and sex strengthened associations and improved discrimination. Adjustment resulted in superior discrimination for blood test trend than blood test abnormality for 34 test-cancer combinations. This includes cancer overall across 12 tests over 3 time intervals, 1-year trend in MCV for bowel cancer and lymphoma, 1-year trend in WBC and neutrophils for lung cancer, and 1-year trend in AST, RBC, haematocrit, and PLR for prostate cancer.

This is a large-scale analysis comparing the discrimination of blood test trends for cancer over varying intervals with commonly used blood test abnormalities in a primary care population. We included over 275,000 patients with UWL, and close to 14,000 cancer cases, ensuring reliability and precision of analyses for cancer overall, although confidence intervals remained relatively wide for some tests. This was further demonstrated when we assessed the influence of multiple testing on our findings: we adjusted the 95% CIs to 99.952% CIs, based on a Bonferroni adjustment of a 5% significance level to 0.048% due to 104 test and trend window combinations, but found that the CIs only changed from the 3rd or 4th decimal point onwards. We therefore retained 95% CIs. Precision was retained for lung, bowel, and prostate cancer subgroup analyses as the number of events ranging from 1,197 for prostate to 3,120 for lung, demonstrating that some adjusted trends had greater discrimination than blood test abnormality. Weaker associations were found in subgroups with smaller sample sizes. Future work investigating the clinical utility of trend in more common symptoms than UWL may lead to more precise associations with trend, overall and by subgroup.

A further strength is that we utilised dynamic modelling to assess the association between blood test trend and cancer diagnosis, which can accommodate the unstructured nature of primary care records. Joint modelling for repeated measures data includes a mixed-effects model component to derive the trend and subject-specific random effects to account for sporadic and irregular testing and the correlation between repeat measurements [18,19]. It does not require a specific testing pattern to derive a trend as it accounts for the variation in the timing and frequency of testing and so is appropriate for unstructured routinely collected data. Prior to the last decade, such investigations were not possible, as many statistical software packages did not include these capabilities. Our modelling strategy did not take into account the bias in who was tested, with patients with chronic disease and greater symptom burden more likely to attend primary care and therefore undergo blood testing. Our trend analysis was also restricted to patients with 2 or more tests in each trend window. This could introduce selection bias, as patients with multiple blood tests must be alive for both tests and may have utilised healthcare services compared to those with one blood test. Patients with 2 or more tests could potentially be more likely to be diagnosed with cancer in primary care. However, when conducting our data checks, analyses showed that the cancer incidence in patients with only one test was comparable to patients with two or more tests (results are available from the authors). In future work, we intend to incorporate blood test trends into a prediction models and this would include all patients with UWL who have a test, not just those with 2 or more tests.

We found that trends over 1 year prior to UWL were most discriminative for individual cancers. The median number of test results in a 1-year trend was two, arguably representing a change rather than a trend across three or more measurements. As the latter test result was the same result used to define test abnormality, any additional discrimination is likely based on the nature of the data used in the trend analysis, the timing of the first test, how ‘normal’ and when the first measurement was, and how ‘abnormal’ the second measurement was. This combination of data structure and study design may explain why we did not see a marked additional utility of adjusted trends in the majority of comparisons with adjusted abnormality. We assessed whether cases with more than two tests within a year led to greater utility and found that discrimination decreased between two and four tests and often when incorporating data earlier than 1-year, signalling that the final measurement in the trend is the most discriminative, though confidence intervals grew wider. When UWL is associated with cancer cachexia, blood test abnormalities are expected [21], reflecting underlying metabolic and inflammatory processes, and explaining why blood test abnormalities at the time the patient presents with UWL offer greatest utility to risk stratify patients for cancer investigation. Blood test trends may hold greater utility in patients presenting with symptoms without accompanying underlying metabolic derangement, or in analyses where the symptomatic presentation is not taken into account.

A further limitation is that we relied on patient data from electronic health records to define UWL. However, we build on previous analyses using a validated list of UWL Read codes [22], where UWL code usage was cross-referenced with change in quantitative weight values over time [15], and validated cancer diagnoses recorded primarily in NCRAS, increasing the reliability of the cohort used and cancers diagnosed [14]. As GPs preferentially code clinical features they assess to be associated with cancer, our reported associations may be stronger than if unstructured free-text data were included [23].

A recent systematic review included 23 studies that assessed the association between blood test trend and cancer diagnosis [10]. However, no study compared performance between blood test abnormalities and the equivalent blood test trends to understand in which cases incorporating blood test trend might enhance cancer risk stratification. Existing studies have estimated the diagnostic performance of commonly used blood test abnormalities to risk stratify for cancer in primary care [3–5]. Commonly, the risk of cancer is estimated for a lower bounded age-range [24]. Risk estimates for multiple age-bands, such as decades of age, more closely align to the continuous age-adjustment of blood test abnormalities used in our analysis, with some studies reporting cancer risks within age strata or by sex separately as an actionable proxy for adjustment [25]. Model-based approaches including more specific and static cancer markers such as Cancer Antigen 125 and the Faecal Immunochemical Test (FIT) are being introduced and incorporated into clinical guidance [26–28]. Our analysis adds to these studies by considering age as a continuous variable, blood test trend, by focussing on the UWL cohort, and by considering a broader range of tests commonly used in clinical practice.

We identified scenarios in which blood test trends may warrant further evaluation to focus or augment the diagnostic work-up in patients presenting with UWL. Changes over a short 1-year period were most discriminative for specific cancers. Whilst these blood tests are commonly used in practice, they are not commonly used to test for the cancers we report an association for trend. Many of these cancers often have an existing, more specific, primary care test. Further research should therefore investigate associations between MCV trend and colorectal cancer in combination or as triage to FIT testing, AST, RBC, haematocrit, and PLR trend in relation to Prostate Specific Antigen (PSA) testing for prostate cancer, and WBC trend as triage to thoracic imaging for lung cancer. As adjusted trend outperformed abnormality most often for cancer diagnosis overall, as opposed to individual cancer sites, this offers a potential approach to enhance patient selection for multi-cancer testing when these technologies are ready for clinical evaluation in primary care [29].

Our findings illustrate that blood test abnormalities and trends have greater discrimination for cancer diagnosis overall and for specific cancers after adjustment for age and sex. In clinical practice, most blood test results are interpreted around a fixed threshold without adjustment [1]. Adjustment could be performed at the laboratory level and an adjusted risk estimate could be reported to primary care with the unadjusted test result, or unadjusted results received from the laboratory could be adjusted within the primary care electronic health record to produce the associated risk estimate. Our findings highlight test-cancer pairings for which trend has the potential to enhance the selection of patients with UWL for cancer investigation in primary care. Dynamic prediction models that utilise age, sex, and the combinations of blood test abnormalities and blood test trends identified in this manuscript could be combined with additional covariates, such as ethnicity, comorbidity, deprivation score, and smoking status. We did not adjust for these additional variables, as we prioritised complete and accurate data electronic health records data (age, sex, and blood tests) and our analysis is exploratory. Such an approach could be implemented into primary care electronic systems after validation to support GP decision-making. These changes would require modifications to current electronic health record functionality and implementation evaluations should consider their impact on clinical workflows, risk communication, clinician behaviour, cost-effectiveness, and clinical outcomes [30].

Adjusted trends over three or more years in 12 blood test components showed superior discrimination to adjusted abnormality. However, the improvements were modest and there are significant implementation considerations to analysing blood test trends within the primary care electronic health record that need to be studied to permit real-time trend analysis. Our results will also need external validation, by deriving performance measures in further primary care datasets. It may be that the more immediate and substantial gains made by focussing on age and sex adjustment of blood test abnormalities should be prioritised over the smaller additional gains made by adding trend analysis. This may not be the case for symptomatic presentations with less metabolic derangement, or for analysing blood test trend without accounting for symptoms, but this needs further investigation, but this needs further investigation.

Blood test abnormalities should be interpreted after age and sex adjustment to optimise patient selection for cancer investigation in patients attending primary care with UWL. Evaluating historical trend within 1 year prior to UWL offers further discrimination, after adjustment by age and sex, for some test-cancer combinations, but the utility of incorporating trend to improve patient triage requires further investigation. Simple decision rules or a risk prediction model incorporating this data could support patient selection in primary care.

Ethics approval

Ethical approval was provided by the Research Data Governance and Independent Scientific Advisory Committee through the CPRD on behalf of the Health Research Authority (protocol number: 22_001798).

Supporting information

S1 Appendix. Supplementary figures and tables.

(DOCX)

pmed.1004956.s001.docx^{(2.7MB, docx)}

S1 Checklist. STROBE+RECORD checklist (https://doi.org/10.1016/j.jclinepi.2007.11.008).

Checklist is protected under Creative Commons Attribution (CC BY) license.

(DOCX)

pmed.1004956.s002.docx^{(19.6KB, docx)}

S2 Checklist. STARD checklist.

(DOCX)

pmed.1004956.s003.docx^{(29.8KB, docx)}

Acknowledgments

The authors would also like to thank the patients who provided data that contributed to the analysis. They also thank patient and public involvement representatives Sue Duncombe, Alton Sutton, Bernard Gudgin, Clara Martins de Barros, Emily Lam, Ian Blelloch, Julian Ashton, Margaret Ogden, Shannon Draisey, and Susan Lynne for applying a patient perspective on the relevance and acceptability of blood test trends for cancer detection in practice.

Abbreviations

AIC: Akaike Information Criterion
ALT: alanine transaminase
AUC: area under the curve
BIC: Bayesian Information Criterion
BMI: body mass index
CIs: confidence intervals
CPRD: Clinical Practice Research Datalink
CRP: C-reactive protein
ESR: erythrocyte sedimentation rate
FIT: Faecal Immunochemical Test
GPs: general practitioners
HES: Hospital Episode Statistics
HRs: hazard ratios
ICD10: International Classification of Diseases, 10th Revision
IMD: Index of Multiple Deprivation
IQR: interquartile range
LOWESS: Locally Weighted Scatterplot Smoothing
MCV: mean cell volume
MCH: mean cell haemoglobin
MCHC: mean cell haemoglobin concentration
mGPR: modified Glasgow Prognostic ratio
mGPS: modified Glasgow Prognostic Score
MPV: mean platelet volume
NCRAS: National Cancer Registration and Analysis Service
NICE: National Institute for Health and Care Excellence
NLR: neutrophil-to-lymphocyte ratio
NLS: neutrophil-to-lymphocyte score
ONS: Office of National Statistics
PLR: platelet-to-lymphocyte ratio
PLS: platelet-to-lymphocyte score
PSA: Prostate Specific Antigen
RBC: red blood cell count
RDW: red cell distribution width
RECORD: REporting of studies Conducted using Observational Routinely-collected health Data
SD: standard deviation
SNOMED-CT: Systematized Nomenclature of Medicine-Clinical Terms
STROBE: STrengthening the Reporting of OBservational studies in Epidemiology
UWL: unexpected weight loss
WBC: white blood cell count

Data Availability

The data used in this study is not freely or publicly available. The data is owned and available from the CPRD. Due to privacy laws and the data user agreement between Oxford and CPRD, researchers are not authorised to share individual patient data. Data access is subject to approval via CPRD’s Research Data Governance Team, who, under the Health Research Authority Review Board, are required to review and approve access to the patient data before it can be shared. A data request application for protocol number 22_001798 can be made on the CPRD website: https://www.cprd.com/ [31] or by email: enquiries@cprd.com. The R code used in the analysis is available from GitHub: https://github.com/PradeepVirdee/CPRD_UWL_TrendVsAbno_Cancer (DOI: https://doi.org/10.5281/zenodo.20752173) [14].

Funding Statement

BDN was funded by the NIHR Policy Research Programme on Cancer Awareness, Screening and Early Diagnosis (Reference PR-PRU-NIHR206132). The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care. BDN and PSV were also funded by a Cancer Research UK Clinical Careers Committee Postdoctoral Fellowship (RCCPDF\100005). FDRH and CWD were supported by the NIHR ARC OTV (https://www.arc-oxtv.nihr.ac.uk/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The views expressed are those of the authors and not necessarily those of these funders.

References

1.NICE. Suspected cancer recognition and referral: site or type of cancer. 2020. [cited 21 Jan 2026]. Available from: https://www.nice.org.uk/guidance/ng12 [Google Scholar]
2.Watson J, Mounce L, Bailey SE, Cooper SL, Hamilton W. Blood markers for cancer. BMJ. 2019;367:l5774. doi: 10.1136/bmj.l5774 [DOI] [PubMed] [Google Scholar]
3.Rafiq M, Renzi C, White B, Zakkak N, Nicholson B, Lyratzopoulos G, et al. Predictive value of abnormal blood tests for detecting cancer in primary care patients with nonspecific abdominal symptoms: a population-based cohort study of 477,870 patients in England. PLoS Med. 2024;21(7):e1004426. doi: 10.1371/journal.pmed.1004426 [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Nicholson BD, Virdee P, Aveyard P, Price SJ, Hobbs FDR, Koshiaris C, et al. Prioritising primary care patients with unexpected weight loss for cancer investigation: diagnostic accuracy study (update). BMJ. 2024;387:e080199. doi: 10.1136/bmj-2024-080199 [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Nicholson BD, Rafiq M, Koshiaris C, Duncombe S, Bankhead C, Lyratzopoulos G, et al. Combining symptoms and blood tests to enhance risk stratification of cancer in patients presenting with unexpected weight loss: a prediction model derivation and validation study using English primary care data. Preprints with The Lancet. 2025. doi: 10.2139/ssrn.5355363 [DOI] [Google Scholar]
6.Pearson C, Poirier V, Fitzgerald K, Rubin G, Hamilton W. Cross-sectional study using primary care and cancer registration data to investigate patients with cancer presenting with non-specific symptoms. BMJ Open. 2020;10(1):e033008. doi: 10.1136/bmjopen-2019-033008 [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Lyratzopoulos G, Neal RD, Barbiere JM, Rubin GP, Abel GA. Variation in number of general practitioner consultations before hospital referral for cancer: findings from the 2010 National Cancer Patient Experience Survey in England. Lancet Oncol. 2012;13(4):353–65. doi: 10.1016/S1470-2045(12)70041-4 [DOI] [PubMed] [Google Scholar]
8.Renzi C, Lyratzopoulos G, Card T, Chu TPC, Macleod U, Rachet B. Do colorectal cancer patients diagnosed as an emergency differ from non-emergency patients in their consultation patterns and symptoms? A longitudinal data-linkage study in England. Br J Cancer. 2016;115(7):866–75. doi: 10.1038/bjc.2016.250 [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Nicholson BD, Hamilton W, O’Sullivan J, Aveyard P, Hobbs FR. Weight loss as a predictor of cancer in primary care: a systematic review and meta-analysis. Br J Gen Pract. 2018;68(670):e311–22. doi: 10.3399/bjgp18X695801 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Virdee PS, Collins KK, Friedemann Smith C, Yang X, Zhu S, Roberts SE, et al. The association between blood test trends and undiagnosed cancer: a systematic review and critical appraisal. Cancers. 2024;16(9):1692. doi: 10.3390/cancers16091692 [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Virdee PS, Collins KK, Smith CF, Yang X, Zhu S, Roberts N, et al. Clinical prediction models incorporating blood test trend for cancer detection: systematic review, meta-analysis, and critical appraisal. JMIR Cancer. 2025;11:e70275. doi: 10.2196/70275 [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Virdee PS, Marian IR, Mansouri A, Elhussein L, Kirtley S, Holt T, et al. The full blood count blood test for colorectal cancer detection: a systematic review, meta-analysis, and critical appraisal. Cancers (Basel). 2020;12(9):2348. doi: 10.3390/cancers12092348 [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Virdee PS, Bankhead C, Koshiaris C, Drakesmith CW, Oke J, Withrow D, et al. BLOod Test Trend for cancEr Detection (BLOTTED): protocol for an observational and prediction model development study using English primary care electronic health record data. Diagn Progn Res. 2023;7(1):1. doi: 10.1186/s41512-022-00138-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.GitHub/PradeepVirdee. CPRD UWL Trend Vs Abno. 2025. [cited 21 Jan 2026]. Available from: https://github.com/PradeepVirdee/CPRD_UWL_TrendVsAbno_Cancer [Google Scholar]
15.Nicholson BD, Hamilton W, Koshiaris C, Oke JL, Hobbs FDR, Aveyard P. The association between unexpected weight loss and cancer diagnosis in primary care: a matched cohort analysis of 65,000 presentations. Br J Cancer. 2020;122(12):1848–56. doi: 10.1038/s41416-020-0829-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Virdee PS, Fuller A, Jacobs M, Holt T, Birks J. Assessing data quality from the Clinical Practice Research Datalink: a methodological approach applied to the full blood count blood test. J Big Data. 2020;7(1). doi: 10.1186/s40537-020-00375-w [DOI] [Google Scholar]
17.Nicholson BD, Oke JL, Aveyard P, Hamilton WT, Hobbs FDR. Individual inflammatory marker abnormalities or inflammatory marker scores to identify primary care patients with unexpected weight loss for cancer investigation?. Br J Cancer. 2021;124(9):1540–2. doi: 10.1038/s41416-021-01282-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Henderson R, Diggle P, Dobson A. Joint modelling of longitudinal measurements and event time data. Biostatistics. 2000;1(4):465–80. doi: 10.1093/biostatistics/1.4.465 [DOI] [PubMed] [Google Scholar]
19.Rizopoulos D. Joint models for longitudinal and time-to-event data with applications in R. Chapman and Hall/CRC; 2012. [Google Scholar]
20.Oxford Academic. Oxford Handbook of Clinical Medicine - Reference intervals, etc. 2017. [cited 21 January 2026]. Available from: https://academic.oup.com/book/37194/chapter/326364279?login=true [Google Scholar]
21.Cao Z, Zhao K, Jose I, Hoogenraad NJ, Osellame LD. Biomarkers for cancer cachexia: a mini review. Int J Mol Sci. 2021;22(9):4501. doi: 10.3390/ijms22094501 [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Nicholson BD, Aveyard P, Bankhead CR, Hamilton W, Hobbs FDR, Lay-Flurrie S. Determinants and extent of weight recording in UK primary care: an analysis of 5 million adults’ electronic health records from 2000 to 2017. BMC Med. 2019;17(1):222. doi: 10.1186/s12916-019-1446-y [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Price SJ, Stapley SA, Shephard E, Barraclough K, Hamilton WT. Is omission of free text records a possible source of data loss and bias in Clinical Practice Research Datalink studies? A case-control study. BMJ Open. 2016;6(5):e011664. doi: 10.1136/bmjopen-2016-011664 [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Koo MM, Hamilton W, Walter FM, Rubin GP, Lyratzopoulos G. Symptom signatures and diagnostic timeliness in cancer patients: a review of current evidence. Neoplasia. 2018;20(2):165–74. doi: 10.1016/j.neo.2017.11.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Mounce LTA, Calitri R, Hamilton W, Rafiq M, Emery JD, Giannakeas V, et al. Improving the clinical utility of platelet count for cancer detection in primary care: a cohort study in England, Canada, and Australia. Cancers (Basel). 2024;16(17):3074. doi: 10.3390/cancers16173074 [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Wu R, Arendse KD, Hamdani T, Walter FM, Crosbie EJ, Mihaylova B, et al. Cost-effectiveness of CA125- and age-informed risk-based triage for ovarian cancer detection in primary care. Br J Cancer. 2025;133(10):1492–500. doi: 10.1038/s41416-025-03166-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Crooks CJ, West J, Jones J, Hamilton W, Bailey SER, Abel G, et al. Aliment pharmacology therapeutics. 2025;61(5):852–64. doi: 10.1111/apt.18459 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Tamm A, Shine B, James T, Withers J, Salih H, Noble T, et al. External validation of the COLOFIT colorectal cancer risk prediction model in the Oxford-FIT dataset: the importance of population characteristics and clinically relevant evaluation metrics. BMC Med. 2025;23(1):503. doi: 10.1186/s12916-025-04339-w [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Nicholson BD, Oke J, Virdee PS, Harris DA, O’Doherty C, Park JE, et al. Multi-cancer early detection test in symptomatic patients referred for cancer investigation in England and Wales (SYMPLIFY): a large-scale, observational cohort study. Lancet Oncol. 2023;24(7):733–43. doi: 10.1016/S1470-2045(23)00277-2 [DOI] [PubMed] [Google Scholar]
30.Dowrick A, Ziebland S, Rai T, Friedemann Smith C, Nicholson BD. A manifesto for improving cancer detection: four key considerations when implementing innovations across the interface of primary and secondary care. Lancet Oncol. 2024;25(8):e388–95. doi: 10.1016/S1470-2045(24)00102-5 [DOI] [PubMed] [Google Scholar]
31.Clinical Practice Research Datalink. 2025. [cited 29 April 2026]. Available from: https://www.cprd.com/ [Google Scholar]

PLoS Med. doi: 10.1371/journal.pmed.1004956.r001

Decision Letter 0

Heather Van Epps

5 Feb 2026

Dear Dr Nicholson,

Thank you for submitting your manuscript entitled "Blood test trend vs single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss in primary care: a diagnostic accuracy, longitudinal cohort study." for consideration by PLOS Medicine.

Your manuscript has now been evaluated by the PLOS Medicine editorial staff, and I am writing to let you know that we would like to send your submission out for external peer review.

However, before we can send your manuscript to reviewers, we need you to complete your submission by providing the metadata that is required for full assessment. To this end, please login to Editorial Manager where you will find the paper in the 'Submissions Needing Revisions' folder on your homepage. Please click 'Revise Submission' from the Action Links and complete all additional questions in the submission questionnaire.

For clinical studies, please upload a copy of your trial study protocol as a supporting information file. The study protocol should be the version submitted for approval to the institutional review board or ethics committee, should include any amendments to the study protocol, as well as the date of their approval by the institutional review or ethics committee. Please also detail any deviations from the study protocol in the Methods section of your manuscript. The editors will consider the protocol and study conduct prior to a final decision for external review.

Please re-submit your manuscript within two working days, i.e. by Feb 09 2026 11:59PM.

Once your full submission is complete, your paper will undergo a series of checks in preparation for peer review. Once your manuscript has passed all checks it will be sent out for review.

Feel free to email us at plosmedicine@plos.org if you have any queries relating to your submission.

Kind regards,

Heather Van Epps, PhD

Consulting Editor

PLOS Medicine

PLoS Med. doi: 10.1371/journal.pmed.1004956.r002

Decision Letter 1

Alexandra Tosun

13 Apr 2026

Dear Dr Nicholson,

Many thanks for submitting your manuscript "Blood test trend vs single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss in primary care: a diagnostic accuracy, longitudinal cohort study." (PMEDICINE-D-26-00455R1) to PLOS Medicine. The paper has been reviewed by subject experts and a statistician; their comments are included below and can also be accessed here: [LINK]

As you will see, the reviewers agree that the topic is important, but the statistical review identifies major methodological and interpretive weaknesses, while the subject reviewers note that the clinical relevance is insufficiently clear and the presentation potentially overly complex. After discussing the paper with the editorial team and an academic editor with relevant expertise, I'm pleased to invite you to revise the paper in response to the reviewers' comments. We plan to send the revised paper to some or all of the original reviewers, and we cannot provide any guarantees at this stage regarding publication.

In addition to these revisions, you may need to complete some formatting changes, which you will receive in a follow up email. A member of our team will be in touch with a set of requests shortly. If you do not receive a separate email within a few days, please assume that checks have been completed, and no additional changes are required.

When you upload your revision, please include a point-by-point response that addresses all of the reviewer and editorial points, indicating the changes made in the manuscript and either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please also be sure to check the general editorial comments at the end of this letter and include these in your point-by-point response. When you resubmit your paper, please include a clean version of the paper as the main article file and a version with changes tracked as a marked-up manuscript. It may also be helpful to check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper.

We ask that you submit your revision by May 04 2026. However, if this deadline is not feasible, please contact me by email, and we can discuss a suitable alternative.

Don't hesitate to contact me directly with any questions (atosun@plos.org).

Best regards,

Alexandra

Alexandra Tosun, PhD

Senior Editor

PLOS Medicine

atosun@plos.org

-----------------------------------------------------------

Comments from the academic editor:

This is a potentially important topic, particularly as the volume of data in national health registries continues to expand and clinical models can become increasingly complex. The findings are intuitive and warrant dissemination to the scientific community, even if their practical implementation may be challenging.

-----------------------------------------------------------

Comments from the reviewers:

Reviewer #1: Major Comments

1) The primary estimand is not clearly defined. It remains ambiguous whether the objective is etiological (association between blood test trend and cancer diagnosis) or predictive/diagnostic (risk stratification and discrimination). The manuscript interchanges hazard ratios and AUC metrics without clearly specifying whether the modelling framework is intended for risk prediction or hypothesis testing.

2) The manuscript does not report any evaluation of the proportional hazards assumption (e.g., Schoenfeld residuals), which is essential when Cox modelling is used.

3) The use of Cox proportional hazards models with a fixed six-month follow-up window requires stronger justification. Given the short and bounded follow-up period, logistic regression could provide equivalent discrimination assessment without reliance on proportional hazards assumptions.

4) The handling of time origin and potential immortal time bias requires clarification. Patients were required to have at least one blood test within 10 years prior to UWL and, for trend analyses, at least two tests including one near UWL. This selection may introduce conditioning on future information relative to earlier time points. The temporal structure should be more clearly articulated to ensure no forward-looking bias is introduced into the risk models.

5) The joint modelling specification requires more detail and justification. Although mixed-effects models with random intercept and slope are described, the functional form of the association between the longitudinal process and the survival outcome is not explicitly stated (e.g., current value parameterisation, slope parameterisation, shared random effects). The exact structure linking the longitudinal sub-model to the Cox model should be clarified.

6) The use of piecewise linear splines for time-to-UWL and age-at-UWL requires methodological justification. The rationale for knot placement at 6 months and 1, 2, and 3 years is not provided. It is unclear whether alternative knot configurations were explored or whether model fit diagnostics were conducted.

7) The decision to model each blood test separately limits interpretability and may inflate Type I error due to extensive multiplicity. Twenty-six blood tests across multiple time windows and cancer subtypes generate a large number of comparisons. No correction for multiple testing or false discovery control is reported.

8) Discrimination assessment using AUC derived from Cox models requires clarification. It is not specified whether time-dependent ROC curves were used or whether a fixed-time AUC at six months was estimated. Standard AUC methods are not directly applicable to censored survival data without appropriate modification.

9) No internal validation strategy is described. Given the extensive modelling and comparison of trends versus abnormalities, overfitting is a concern. There is no mention of bootstrapping, cross-validation, shrinkage, or optimism correction.

10) Missing data handling is insufficiently described. Several baseline covariates (e.g., ethnicity, BMI, smoking) show substantial missingness. It is unclear whether these were excluded from modelling, included as missing categories, or addressed using multiple imputation. Even if not included in final models, their exclusion requires justification.

11) The requirement of at least two blood tests within a given window introduces selection bias toward patients with greater healthcare utilisation. This may induce collider bias if healthcare contact is associated with both blood testing frequency and cancer risk. The implications of restricting to 2+ tests for trend modelling should be discussed and sensitivity analyses considered.

12) The comparison between "trend" and "abnormality" may not be methodologically equivalent. The abnormality model uses a binary threshold, whereas the trend model uses continuous longitudinal modelling with random effects and spline functions. This structural difference may inherently favour trend-based modelling due to greater modelling flexibility. A fairer comparison would require aligning model complexity.

13) Effect sizes for trends (e.g., HRs close to 1.00-1.08) suggest very small per-unit effects. Interpretation of the unit of change (per SD? per raw unit? per year?) is not sufficiently clear. Clinical interpretability of these HRs requires more explicit scaling.

14) Competing risks (e.g., death within six months) are not addressed. Given the age distribution and comorbidity burden, death may preclude cancer diagnosis, potentially biasing hazard estimates.

15) The rationale for limiting outcome ascertainment to six months should be better justified. A shorter or longer window could materially affect discrimination performance and hazard estimates.

16) Model calibration is not reported. While AUC evaluates discrimination, calibration metrics (e.g., calibration slope, calibration-in-the-large, calibration plots) are necessary to assess clinical utility in a diagnostic context.

Minor Comments

17) The LOWESS graphical exploration is described but not integrated into model specification decisions. It would be helpful to clarify whether non-linear patterns observed informed spline selection.

18) The manuscript does not specify the software and packages used for joint modelling. Given the complexity of joint models, reproducibility requires software identification and key tuning parameters.

19) It is unclear whether laboratory measurements were standardised across time and laboratories beyond unit harmonisation. Changes in assay methodology over an 18-year period could introduce systematic measurement shifts.

20) The exclusion of negative laboratory values is mentioned but not justified. It should be clarified whether these represent data errors or physiologically implausible values.

21) Age is modelled linearly (with splines in longitudinal sub-models), but no justification is provided for not exploring non-linear age effects in the Cox models for abnormality comparisons.

22) The approach to deriving inflammatory ratios (NLR, PLR, mGPS) is described, but the impact of categorisation versus continuous modelling is not discussed.

23) The description of censoring could more clearly distinguish administrative censoring from event-based censoring.

24) Sub-group AUC analyses (by age, sex, stage, number of repeat tests) are mentioned, but statistical uncertainty and interaction testing procedures are not described.

25) There is no reporting of effective sample size within each trend window after applying the ≥2-test restriction. This is particularly important for shorter windows (e.g., 1-year trend).

26) The manuscript would benefit from a flow diagram specific to the trend-analysis cohorts for each time window to clarify denominators.

Reviewer #2: PMEDICINE-D-26-00455R1 Blood test trend vs single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss in primary care: a diagnostic accuracy, longitudinal cohort study.

Line 35: spell out interquartile range (ICR)

Line 85: Would the authors be more specific about how UWL was measured? First patient report of UWL? Or indication in the chart of WL between the last and most recent measure? In the case of the latter, does the duration influence the date of measurement, or is it taken from the previous measure, or the most recent measure indicating UWL? In either case what is the magnitude of loss to qualify as UWL that is taken seriously? Not need to answer these questions specifically, but it would be good to operationalize the measure of UWL for those of us who are perhaps aware that there is a clear, well understood definition.

Line 167: spell out RBC and CRP.

Line 180: spell out ESR

Line 200: spell out ALT.....the authors could spell out some of these abbreviations, say in Table 2 if they are blood tests, but I won't mention this again since I suspect the list will grow longer. However, the reader who are not clinicians need to know what these abbreviations represent.

Line 310: The authors assert that the unstructured nature of primary care records can be seen as a strength....is it the records, or the variability of encounters and the test that are chosen during those encounters? I would presume that whatever is done is somewhat recorded, especially laboratory tests. The statement implies that the unstructured nature of records is a strength over a highly structured record. Is this what is meant? I see the point made beginning on line 313, so maybe rather than a strength, it is not a data liability?

In the discussion, where implications for research and clinical practice are discussed, the authors report that their findings illustrate that blood test abnormalities and trends have greater discrimination for cancer diagnosis overall and for specific cancers after adjustment for age and sex. If I'm a GP, and considering the implications of these data, I believe my first thought would be that these findings are potentially very useful, but "a lot of moving parts that would be hopeless for me to manage on my own. The authors then say it would not straightforward to provide the adjusted rates by the laboratory, or if not, the EHR could do the math, providing the GP, and perhaps go further to provide the GP with an alert about the possible indication for cancer. The authors then caution that "Although these changes would require relatively simple modifications to existing processes and electronic health record functionality, implementation evaluations should consider their impact on clinical workflows, risk communication, clinician behaviour, cost effectiveness, and clinical outcomes." Then, in closing, they say "Blood test abnormalities should be interpreted after age and sex adjustment to improve triage for cancer investigation in patients attending primary care with UWL. Evaluating historical trend leading up to the UWL offers further discrimination after adjustment for some test-cancer combinations."

It seems to me that this study has found something very important (A) and of immediate practical use (B), but it is not clear how to get from A to B, and they caution, we should not until consideration has been given to clinical workflows, risk communication, clinician behaviour, cost effectiveness, and clinical outcomes.

I would encourage the authors to be bolder and stress that these observations highlight the potential for a new tool to reduce delay in diagnosis, but have too many elements for GPs to process on their own. Thus, the need for EHR algorithms and perhaps an AI tool, which should be a very high immediate priority in order to proceed to the next step to determine the feasibility of their use in community health settings. It still may not be clear whose responsibility it would be to take up the charge, but at least you would have made it clear that some health authority should embrace that responsibility.

Reviewer #3: This is a well-written paper with clear abstract and introduction, clear methods and mostly clear results section, and a substantial amount of supplementary data (248 pages). Authors investigated whether historical changes in blood tests (trends) enhanced cancer risk discrimination (abstract)/stratification (methods) in patients presenting in primary care with unexpected weight loss (UWL) - compared to abnormalities in the most recent blood test. Unadjusted and adjusted (age and sex) AUCs were calculated. Strengths include confirming cases using linked national cancer registration data, and the longitudinal cohort design using robust primary care data from the Clinical Practice Research Datalink (CPRD). It is very clear that a great amount of time was spent to prepare the submitted files, and care was taken to ensure consistent formatting across all tables and figures. I admire the level of transparency as authors shared as much as they could within the constraints of the used CPRD data. However, I do wonder if the amount of reported information was at times detrimental to understanding the key results and the key messages from the manuscript. I would recommend some revisions to the main text (and potentially supplementary data), as described below.

Major Issues

1.Most of the results presented require the reader to check supplementary data, as often the summary in text is quite brief. I was wondering if narrowing down the focus of this paper would help? There are results on median N of tests per patient, median N of tests per cohort, 1,3,5 and 10-year trends (increasing or decreasing, for specific blood tests), discriminative ability for all cancers, and for specific cancers, diagnostic performance for abnormalities and trends, with and without adjustment for sex and age. I am quite happy to get a rebuttal about this, but the amount of information at times made it challenging to understand the results. If narrowing down the focus is not possible, I think a good compromise would be to highlight in the supplementary figures/tables some of the key information described in the manuscript. For example, if trends were favoured for some tumour types, these "favoured trends" could be shown in bold, or highlighted in a different colour. This would help the reader navigate the supplementary data. Often, AUCs are very similar across all comparisons, and being able to clearly see the results being emphasised in the manuscript would be very helpful.

2.In the results section, line 177, it is stated that "graphs of blood test trends over the 10-year study period are in the figures S2-S27". The evidence from these figures is then summarised in three sentences. It was only after looking at the supplementary data (and I thought that these figures were fantastic and very informative) that everything made more sense. I understand that these figures need to be supplementary, but I would recommend adding a bit more information to the paragraph (lines 177-182), perhaps adding some values, and explaining what was meant by limited data.

3.Lines 368-375: I think these sentences synthesise quite well the key findings re. investigated trends, but they are a bit lost in the long paragraph. I did wonder if these should be the first paragraph in this subsection (implications) as this information is more closely linked with the study aim compared to the information provided in the previous paragraph.

4.Lines 375-385: I would consider rewriting some of this. It may be a bit so soon to suggest that adjusted trends over 3 or more years may improve triage to specialist services, or future triage for multi-cancer testing. I would perhaps remove this bit and just acknowledge that any improvements seen for adjusted trends were modest (as stated later in the paragraph) and it would not be appropriate to recommend implementation (what is meant by socio-technical?) before their utility can be better ascertained, whether for all cancers, or for specific tests and tumour types. In lines 384-385, it may be worth reiterating that this possibility needs to be further investigated.

5.Lines 388-390: The conclusion could have been a bit punchier. The last sentence could be expanded a bit (389-390): improvements related to trends are often seen one year prior to UWL for some cancers, but the utility of incorporating trends to improve triage for patients with UWL need to be further investigated.

Minor issues

6.Authors and affiliations: I have noticed a typo you may wish to correct. Should it be Clare Bankhead instead of Clare Bankead?

7.Abstract, methods: Apologies if I missed this in the main manuscript, were the 26 blood tests decided a priori or do they refer to all different blood tests identified in CPRD records? If the former, what were the criteria for choosing the tests?

8.Introduction, lines 55-56. For the interest of the audience outside the UK, please specify that NICE guidelines are used in England and Wales

9.Methods, lines 137-139: Please provide the rationale for choosing these timeframes. Were these based on clinical knowledge, previous studies, natural history of specific cancers, etc.?

10.Methods, lines 147-148: Table S1 is very useful, but please also consider clarifying which organisation/country/health system established the norms/reference ranges used for the blood tests.

11.Table 1: Were the ethnicity categories the ones provided by CPRD? Consider adding a footnote about what is included in each category. Each of these may include heterogenous groups, with varying cancer risk (I understand that larger groups are needed to carry out the analyses).

12.Table 1: Please clarify if IMD1 is most and IMD 5 is least deprived, just to avoid any confusion from the reader. This also applies to some supplementary tables.

13.Table 2.: While this is relevant, I wonder if just providing some information in text would be sufficient? There are many excellent supplementary tables that could be in the manuscript instead.

14.Page 20, line 298: "Strengths and limitations" instead of "Strengths and limitation"?

15.Discussion, lines 304-307: The statement "Weaker associations were found in small subgroups with smaller sample sizes" seems neutral, but since it follows sentence about stronger associations with larger sample sizes being identified for larger cancer groups, it implies that if sample sizes were larger for other cancer types, stronger associations were likely to be found. Perhaps consider a more neutral approach - if samples were larger, one would be able to test whether there were also stronger associations for other cancers..

16.Table S4 - Can the authors please clarify what is meant by symptom co-occurring?

17.Discussion: It may be worth adding somewhere that perhaps results could differ if trends were adjusted for other (or additional) patient characteristics such as lifestyle behaviours or ethnicity - bearing in mind the limitations regarding the amount of missing data for these variables.

18.In table S9 CIs are missing for some tests - a footnote explaining this would be useful. Is it because numbers are too small?

19.In Table S12 the AUCs and CIs are very similar for CRP and ESR. This may just be a coincidence, but since they are so close to each other in the table could the authors please check in case there is some repeated information there?

20.Overall: what made the authors choose to report on AUCs other than other measures of diagnostic performance?

Any attachments provided with reviews can be seen via the following link: [LINK]

--------------------------------------------------------- ---

General editorial requests:

(Note: not all will apply to your paper, but please check each item carefully)

*We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. Please do not add or remove authors without first discussing this with the handling editor. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

*Please upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, we strongly recommend that you use PLOS's NAAS tool (https://ngplosjournals.pagemajik.ai/artanalysis) to test your figure files. NAAS can convert your figure files to the TIFF file type and meet basic requirements (such as print size, resolution), or provide you with a report on issues that do not meet our requirements and that NAAS cannot fix.

After uploading your figures to PLOS's NAAS tool - https://ngplosjournals.pagemajik.ai/artanalysis, NAAS will process the files provided and display the results in the "Uploaded Files" section of the page as the processing is complete.

If the uploaded figures meet our requirements (or NAAS is able to fix the files to meet our requirements), the figure will be marked as "fixed" above. If NAAS is unable to fix the files, a red "failed" label will appear above.

When NAAS has confirmed that the figure files meet our requirements, please download the file via the download option, and include these NAAS processed figure files when submitting your revised manuscript.

*Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information (web or email address) for obtaining the data. Please note that a study author cannot be the contact person for the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

*We expect all researchers with submissions to PLOS in which author-generated code underpins the findings in the manuscript to make all author-generated code available without restrictions upon publication of the work. In cases where code is central to the manuscript, we may require the code to be made available as a condition of publication. Authors are responsible for ensuring that the code is reusable and well documented. Please make any custom code available, either as part of your data deposition or as a supplementary file. Please add a sentence to your data availability statement regarding any code used in the study, e.g. "The code used in the analysis is available from Github [URL] and archived in Zenodo [DOI link]" Please review our guidelines at https://journals.plos.org/plosmedicine/s/materials-software-and-code-sharing and ensure that your code is shared in a way that follows best practice and facilitates reproducibility and reuse. Because Github depositions can be readily changed or deleted, we encourage you to make a permanent DOI'd copy (e.g. in Zenodo) and provide the URL.

FORMATTING - GENERAL

*Abstract: Please structure your abstract using the PLOS Medicine headings (Background, Methods and Findings, Conclusions). Please combine the Methods and Findings sections into one section.

*At this stage, we ask that you include a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract. Ideally each sub-heading should contain 2-3 single sentence, concise bullet points containing the most salient points from your study. In the final bullet point of 'What Do These Findings Mean?', please include the main limitations of the study in non-technical language. Please see our author guidelines for more information: https://journals.plos.org/plosmedicine/s/revising-your-manuscript#loc-author-summary.

*Please express the main results with 95% CIs as well as p values. When reporting p values please report as p<0.001 and where higher as the exact p value p=0.002, for example. Throughout, suggest reporting statistical information as follows to improve clarity for the reader "22% (95% CI [13%,28%]; p</=)". Please be sure to define all numerical values at first use.

*Please include page numbers and line numbers in the manuscript file. Use continuous line numbers (do not restart the numbering on each page).

*Please cite the reference numbers in square brackets. Citations should precede punctuation.

FIGURES AND TABLES

*Please provide titles and legends for all figures and tables (including those in Supporting Information files).

*Please define all abbreviations used in each figure/table (including those in Supporting Information files).

*Please consider avoiding the use of red and green in order to make your figure more accessible to those with color blindness.

SUPPLEMENTARY MATERIAL

*Please note that supplementary material will be posted as supplied by the authors. Therefore, please amend it according to the relevant comments outlined here.

*Please cite your Supporting Information as outlined here: https://journals.plos.org/plosmedicine/s/supporting-information

REFERENCES

*PLOS uses the numbered citation (citation-sequence) method and first six authors, et al.

*Please ensure that journal name abbreviations match those found in the National Center for Biotechnology Information (NCBI) databases (http://www.ncbi.nlm.nih.gov/nlmcatalog/journals), and are appropriately formatted and capitalised.

*Where website addresses are cited, please include the complete URL and specify the date of access (e.g. [accessed: 12/06/2023]).

*Please also see https://journals.plos.org/plosmedicine/s/submission-guidelines#loc-references for further details on reference formatting.

STUDY TYPE-SPECIFIC REQUESTS

(Note: not all will apply to your paper, but please check each item carefully)

*Abstract: Please include the study design, population and setting, number of participants, years during which the study took place (enrollment and follow up), length of follow up, and main outcome measures.

*Please ensure that the study is reported according to the RECORD guideline (available from https://www.record-statement.org) and include the completed checklist as Supporting Information. Please add the following statement, or similar, to the Methods: "This study is reported as per the Reporting of Studies Conducted using Observational Routinely-Collected Data (RECORD) guideline (S1 Checklist)." When completing the checklist, please use section and paragraph numbers, rather than page numbers.

*In the manuscript text, please indicate: (1) the specific hypotheses you intended to test, (2) the analytical methods by which you planned to test them, (3) the analyses you actually performed, and (4) when reported analyses differ from those that were planned, transparent explanations for differences that affect the reliability of the study's results. If a reported analysis was performed based on an interesting but unanticipated pattern in the data, please be clear that the analysis was data driven.

*Please state in the Methods section whether the study had a prospective protocol or analysis plan. If a prospective analysis plan (from your funding proposal, IRB or other ethics committee submission, study protocol, or other planning document written before analyzing the data) was used in designing the study, please include the relevant document(s) with your revised manuscript as a Supporting Information file to be published alongside your study and cite it in the Methods section. A legend for this file should be included at the end of your manuscript. If no such document exists, please make sure that the Methods section transparently describes when analyses were planned, and when/why any data-driven changes to analyses took place. Changes in the analysis, including those made in response to peer review comments, should be identified as such in the Methods section of the paper, with rationale.

*Please ensure that the study is reported according to the STARD guideline (https://www.equator-network.org/reporting-guidelines/stard/) and include the completed STARD checklist as Supporting Information. Please add the following statement, or similar, to the Methods: "This study is reported as per the Standards for Reporting of Diagnostic Accuracy (STARD) guideline (S1 Checklist)." When completing the checklist, please use section and paragraph numbers, rather than page numbers.

*Please structure your Abstract according to STARD for Abstracts (https://www.equator-network.org/reporting-guidelines/stard-abstracts/).

*Please structure the Methods section using the following sub-headings: Study design, Participants, Test methods, Analysis.

*Please include a diagram to describe the flow of participants through the study (typically figure 1).

PLoS Med. 2026 Sep 17;23(9):e1004956. doi: 10.1371/journal.pmed.1004956.r003

Author response to Decision Letter 2

5 May 2026

Attachment

Submitted filename: Reviewer responses V1.0.docx

pmed.1004956.s006.docx^{(59.2KB, docx)}

PLoS Med. doi: 10.1371/journal.pmed.1004956.r004

Decision Letter 2

Katarzyna Starowicz

3 Jun 2026

Dear Dr. Nicholson,

Thank you very much for re-submitting your manuscript "Blood test trend vs single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss in primary care: a diagnostic accuracy, longitudinal cohort study." (PMEDICINE-D-26-00455R2) for review by PLOS Medicine.

I have discussed the paper with my colleagues and the academic editor and it was also seen again by two reviewers. There are remaining editorial and production issues that need to be addressed in full and are listed at the end of this email. You must provide point-by-point response to each of them. Some of them might not be applicable, in that case please respond "not applicable". Not providing this in form of rebuttal will significantly delay your manuscript.

Any accompanying reviewer attachments can be seen via the link below. Please take these into account before resubmitting your manuscript:

[LINK]

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

In revising the manuscript for further consideration here, please ensure you address the specific points made by each reviewer and the editors. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments and the changes you have made in the manuscript. Please submit a clean version of the paper as the main article file. A version with changes marked must also be uploaded as a marked up manuscript file.

Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. If you haven't already, we ask that you provide a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract.

We expect to receive your revised manuscript within 1 week. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript.

Please note, when your manuscript is accepted, an uncorrected proof of your manuscript will be published online ahead of the final version, unless you've already opted out via the online submission form. If, for any reason, you do not want an earlier version of your manuscript published online or are unsure if you have already indicated as such, please let the journal staff know immediately at plosmedicine@plos.org.

If you have any questions in the meantime, please contact me or the journal staff on plosmedicine@plos.org.

We look forward to receiving the revised manuscript by Jun 10 2026 11:59PM.

Sincerely,

Katarzyna Starowicz, PhD

Associate Editor

PLOS Medicine

kstarowicz@plos.org

------------------------------------------------------------

Comments and Requests from Editors:

Please state why you looked at data only to 2018, and not more recent data. Did any of the blood tests change over that period? You should comment on how these conclusions could be validated. Is the detection window of 6mo also a limitation? Maybe you should have considered within 1 yr?

*We think it would be improve clarity if you could add what the main objective of the study is to the Abstract. The Methods and Findings section of the Abstract does not (to my read) crystallize what they aimed to find out, and what you found.

* Please confirm that your title complies with PLOS Medicine's style. Your title must be nondeclarative and not a question. It should begin with main concept if possible. "Effect of" should be used only if causality can be inferred, i.e., for an RCT. Please place the study design ("A randomized controlled trial," "A retrospective study," "A modelling study," etc.) in the subtitle (ie, after a colon).

* Please confirm that your abstract complies with our requirements, including format (three sections: Background, Methods and Findings, and Conclusions) and providing all the information relevant to this study type https://journals.plos.org/plosmedicine/s/submission-guidelines#loc-abstract

* Please ensure that the Introduction ends with a clear description of the study question or hypothesis.

* Please ensure that all abbreviations are defined at first use throughout the text (including STROBE, RECORD, and STARD).

* Please confirm that all numbers presented in the abstract are present and identical to numbers presented in the main manuscript text.

* Please disclose any use of AI tools and technologies to the study or article content at the end of the Methods. Please specify the name(s) and version(s) of any tool(s) used and a description of their use, including how the AI-assisted or generated content was evaluated by authors. Please also include a statement specifying who retains responsibility for the final manuscript. Please see our guidelines for more information: https://journals.plos.org/plosmedicine/s/ethical-publishing-practice#loc-artificial-intelligence-tools-and-technologies.

* The funding statement should include: specific grant numbers, initials of authors who received each award, URLs to sponsors’ websites. Also, please state whether any sponsors or funders (other than the named authors) played any role in study design, data collection and analysis, the decision to publish, or preparation of the manuscript. If they had no role in the research, include this sentence: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.”

* Please review your text for claims of novelty or primacy (e.g. 'for the first time') and remove this language. In addition, please check that any use of statistical terms (such as trend or significant) are supported by the data, and if not please remove them.

* Please remove the 'conclusions' subheading from the discussion. Please also remove any other subheadings from the discussion.

* Statistical reporting: Please revise throughout the manuscript, including tables and figures.

- Please report statistical information as follows to improve clarity for the reader ""22% (95% CI [13,28]; p</=)"".

- Please separate upper and lower bounds with commas instead of hyphens as the latter can be confused with reporting of negative values.

- Please repeat statistical definitions (HR, CI etc.) for each set of parentheses."

* In the last sentence of the Abstract Methods and Findings section, please describe the main limitation(s) of the study's methodology.

* In the abstract, please include the important dependent variables that are adjusted for in the analyses.

* PLOS Medicine requires that the de-identified data underlying the specific results in a published article be made available, without restrictions on access, in a public repository or as Supporting Information at the time of article publication, provided it is legal and ethical to do so. Please see the policy at

http://journals.plos.org/plosmedicine/s/data-availability and FAQs at

http://journals.plos.org/plosmedicine/s/data-availability#loc-faqs-for-data-policy "

* PLOS defines the “minimal data set” to consist of the data set used to reach the conclusions drawn in the manuscript with related metadata and methods, and any additional data required to replicate the reported study findings in their entirety. Authors do not need to submit their entire data set, or the raw data collected during an investigation. Please submit the following data:

The values behind the means, standard deviations and other measures reported;

The values used to build graphs;

The points extracted from images for analysis.

* The Data Availability Statement (DAS) requires revision. For each data source used in your study:

a) If the data are freely or publicly available, note this and state the location of the data: within the paper, in Supporting Information files, or in a public repository (include the DOI or accession number).

b) If the data are owned by a third party but freely available upon request, please note this and state the owner of the data set and contact information for data requests (web or email address). Note that a study author cannot be the contact person for the data.

c) If the data are not freely available, please describe briefly the ethical, legal, or contractual restriction that prevents you from sharing it. Please also include an appropriate contact (web or email address) for inquiries (again, this cannot be a study author).

* For studies in which a novel model is central to the manuscript's findings, as is the case here, authors are responsible for providing the source code needed to replicate the study's findings in a repository (such as GitHub, SourceForge or Bitbucket) or a cloud computing service (such as Code Ocean). Please include a citable digital object identifier (DOI) or other persistent identifier of the code (such as through linkage with Zenodo). Protection of authors’ intellectual property will not be cause for exception. Please explain in the manuscript’s Data Availability Statement how readers can access the shared code.

* Thank you for agreeing to make your data available. At this time, please provide the link to the data repository and accession numbers required for access.

* Please include the statement on code availability in the data availability statement. Please see guidance here:https://journals.plos.org/plosmedicine/s/code-sharing-guidance#loc-data-availability-statement-examples

* Please provide the name(s) of the institutional review board(s) that provided ethical approval.

* Please revise for use of patient-centered language. Please note that patient-centered language is constructed with the use of post-modified nouns (e.g. 'patients with psoriasis’ (or similar) instead of ‘psoriasis patients’) putting the person first in the sentence structure.

* Please include an acknowledgment of study participants.

* Please provide titles and legends for all figures and tables (including those in Supporting Information files). Please define all acronyms used in each figure or table in its corresponding legend.

* Please ensure that where relevant figures include 95% CIs.

* Please show graph axes beginning at zero. If this is not possible, please show a break in the axis.

* When a p value is given, please specify the statistical test used to determine it in the legend.

* Where data points are discrete, please ensure that they are depicted in the figures as discrete data and not as a continuous line.

* Please provide the unadjusted comparisons as well as the adjusted comparisons in all relevant Tables

* Please specify the variables controlled for in all relevant Tables

-------------------------------------------

Comments from Reviewers:

Reviewer #1: Thank you for thoroughly addressing all my comments and providing detailed clarifications in your revised manuscript. I appreciate the effort you have put into enhancing the rigor and transparency of your analyses.

The manuscript now reflects a robust and valuable contribution to subject, I am happy to recommend this work for publication and look forward to seeing its positive impact in the field.

Reviewer #3: The authors have addressed my queries very well. I have read through the responses, and checked the manuscript again. I only have one minor comment that the authors may wish to check:

Figure 1 - Please check if the date here is correct, i.e."31/12/2019". Should it be 2018 instead?

Any attachments provided with reviews can be seen via the following link:

[LINK]

PLoS Med. 2026 Sep 17;23(9):e1004956. doi: 10.1371/journal.pmed.1004956.r005

Author response to Decision Letter 3

8 Jun 2026

Attachment

Submitted filename: Reviewer_responses_V1.0_auresp_3.docx

pmed.1004956.s007.docx^{(35.2KB, docx)}

PLoS Med. doi: 10.1371/journal.pmed.1004956.r006

Decision Letter 3

Katarzyna Starowicz

15 Jun 2026

Dear Dr Nicholson,

On behalf of my colleagues and the Academic Editor, Eleanor L Watts, I am pleased to inform you that we have agreed to publish your manuscript "Blood test trend vs single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss in primary care: a diagnostic accuracy, longitudinal cohort study." (PMEDICINE-D-26-00455R3) in PLOS Medicine.

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email. Please be aware that it may take several days for you to receive this email; during this time no action is required by you. Once you have received these formatting requests, please note that your manuscript will not be scheduled for publication until you have made the required changes.

In the meantime, please log into Editorial Manager at http://www.editorialmanager.com/pmedicine/, click the "Update My Information" link at the top of the page, and update your user information to ensure an efficient production process.

PRESS

We frequently collaborate with press offices. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximise its impact. If the press office is planning to promote your findings, we would be grateful if they could coordinate with medicinepress@plos.org. If you have not yet opted out of the early version process, we ask that you notify us immediately of any press plans so that we may do so on your behalf.

We also ask that you take this opportunity to read our Embargo Policy regarding the discussion, promotion and media coverage of work that is yet to be published by PLOS. As your manuscript is not yet published, it is bound by the conditions of our Embargo Policy. Please be aware that this policy is in place both to ensure that any press coverage of your article is fully substantiated and to provide a direct link between such coverage and the published work. For full details of our Embargo Policy, please visit http://www.plos.org/about/media-inquiries/embargo-policy/.

Thank you again for submitting to PLOS Medicine. We look forward to publishing your paper.

Sincerely,

Katarzyna Starowicz, PhD

Associate Editor

PLOS Medicine

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 Appendix. Supplementary figures and tables.

(DOCX)

pmed.1004956.s001.docx^{(2.7MB, docx)}

S1 Checklist. STROBE+RECORD checklist (https://doi.org/10.1016/j.jclinepi.2007.11.008).

Checklist is protected under Creative Commons Attribution (CC BY) license.

(DOCX)

pmed.1004956.s002.docx^{(19.6KB, docx)}

S2 Checklist. STARD checklist.

(DOCX)

pmed.1004956.s003.docx^{(29.8KB, docx)}

Attachment

Submitted filename: Reviewer responses V1.0.docx

pmed.1004956.s006.docx^{(59.2KB, docx)}

Attachment

Submitted filename: Reviewer_responses_V1.0_auresp_3.docx

pmed.1004956.s007.docx^{(35.2KB, docx)}

Data Availability Statement

[pmed.1004956.ref001] 1.NICE. Suspected cancer recognition and referral: site or type of cancer. 2020. [cited 21 Jan 2026]. Available from: https://www.nice.org.uk/guidance/ng12 [Google Scholar]

[pmed.1004956.ref002] 2.Watson J, Mounce L, Bailey SE, Cooper SL, Hamilton W. Blood markers for cancer. BMJ. 2019;367:l5774. doi: 10.1136/bmj.l5774 [DOI] [PubMed] [Google Scholar]

[pmed.1004956.ref003] 3.Rafiq M, Renzi C, White B, Zakkak N, Nicholson B, Lyratzopoulos G, et al. Predictive value of abnormal blood tests for detecting cancer in primary care patients with nonspecific abdominal symptoms: a population-based cohort study of 477,870 patients in England. PLoS Med. 2024;21(7):e1004426. doi: 10.1371/journal.pmed.1004426 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref004] 4.Nicholson BD, Virdee P, Aveyard P, Price SJ, Hobbs FDR, Koshiaris C, et al. Prioritising primary care patients with unexpected weight loss for cancer investigation: diagnostic accuracy study (update). BMJ. 2024;387:e080199. doi: 10.1136/bmj-2024-080199 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref005] 5.Nicholson BD, Rafiq M, Koshiaris C, Duncombe S, Bankhead C, Lyratzopoulos G, et al. Combining symptoms and blood tests to enhance risk stratification of cancer in patients presenting with unexpected weight loss: a prediction model derivation and validation study using English primary care data. Preprints with The Lancet. 2025. doi: 10.2139/ssrn.5355363 [DOI] [Google Scholar]

[pmed.1004956.ref006] 6.Pearson C, Poirier V, Fitzgerald K, Rubin G, Hamilton W. Cross-sectional study using primary care and cancer registration data to investigate patients with cancer presenting with non-specific symptoms. BMJ Open. 2020;10(1):e033008. doi: 10.1136/bmjopen-2019-033008 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref007] 7.Lyratzopoulos G, Neal RD, Barbiere JM, Rubin GP, Abel GA. Variation in number of general practitioner consultations before hospital referral for cancer: findings from the 2010 National Cancer Patient Experience Survey in England. Lancet Oncol. 2012;13(4):353–65. doi: 10.1016/S1470-2045(12)70041-4 [DOI] [PubMed] [Google Scholar]

[pmed.1004956.ref008] 8.Renzi C, Lyratzopoulos G, Card T, Chu TPC, Macleod U, Rachet B. Do colorectal cancer patients diagnosed as an emergency differ from non-emergency patients in their consultation patterns and symptoms? A longitudinal data-linkage study in England. Br J Cancer. 2016;115(7):866–75. doi: 10.1038/bjc.2016.250 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref009] 9.Nicholson BD, Hamilton W, O’Sullivan J, Aveyard P, Hobbs FR. Weight loss as a predictor of cancer in primary care: a systematic review and meta-analysis. Br J Gen Pract. 2018;68(670):e311–22. doi: 10.3399/bjgp18X695801 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref010] 10.Virdee PS, Collins KK, Friedemann Smith C, Yang X, Zhu S, Roberts SE, et al. The association between blood test trends and undiagnosed cancer: a systematic review and critical appraisal. Cancers. 2024;16(9):1692. doi: 10.3390/cancers16091692 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref011] 11.Virdee PS, Collins KK, Smith CF, Yang X, Zhu S, Roberts N, et al. Clinical prediction models incorporating blood test trend for cancer detection: systematic review, meta-analysis, and critical appraisal. JMIR Cancer. 2025;11:e70275. doi: 10.2196/70275 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref012] 12.Virdee PS, Marian IR, Mansouri A, Elhussein L, Kirtley S, Holt T, et al. The full blood count blood test for colorectal cancer detection: a systematic review, meta-analysis, and critical appraisal. Cancers (Basel). 2020;12(9):2348. doi: 10.3390/cancers12092348 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref013] 13.Virdee PS, Bankhead C, Koshiaris C, Drakesmith CW, Oke J, Withrow D, et al. BLOod Test Trend for cancEr Detection (BLOTTED): protocol for an observational and prediction model development study using English primary care electronic health record data. Diagn Progn Res. 2023;7(1):1. doi: 10.1186/s41512-022-00138-6 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref014] 14.GitHub/PradeepVirdee. CPRD UWL Trend Vs Abno. 2025. [cited 21 Jan 2026]. Available from: https://github.com/PradeepVirdee/CPRD_UWL_TrendVsAbno_Cancer [Google Scholar]

[pmed.1004956.ref015] 15.Nicholson BD, Hamilton W, Koshiaris C, Oke JL, Hobbs FDR, Aveyard P. The association between unexpected weight loss and cancer diagnosis in primary care: a matched cohort analysis of 65,000 presentations. Br J Cancer. 2020;122(12):1848–56. doi: 10.1038/s41416-020-0829-3 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref016] 16.Virdee PS, Fuller A, Jacobs M, Holt T, Birks J. Assessing data quality from the Clinical Practice Research Datalink: a methodological approach applied to the full blood count blood test. J Big Data. 2020;7(1). doi: 10.1186/s40537-020-00375-w [DOI] [Google Scholar]

[pmed.1004956.ref017] 17.Nicholson BD, Oke JL, Aveyard P, Hamilton WT, Hobbs FDR. Individual inflammatory marker abnormalities or inflammatory marker scores to identify primary care patients with unexpected weight loss for cancer investigation?. Br J Cancer. 2021;124(9):1540–2. doi: 10.1038/s41416-021-01282-4 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref018] 18.Henderson R, Diggle P, Dobson A. Joint modelling of longitudinal measurements and event time data. Biostatistics. 2000;1(4):465–80. doi: 10.1093/biostatistics/1.4.465 [DOI] [PubMed] [Google Scholar]

[pmed.1004956.ref019] 19.Rizopoulos D. Joint models for longitudinal and time-to-event data with applications in R. Chapman and Hall/CRC; 2012. [Google Scholar]

[pmed.1004956.ref020] 20.Oxford Academic. Oxford Handbook of Clinical Medicine - Reference intervals, etc. 2017. [cited 21 January 2026]. Available from: https://academic.oup.com/book/37194/chapter/326364279?login=true [Google Scholar]

[pmed.1004956.ref021] 21.Cao Z, Zhao K, Jose I, Hoogenraad NJ, Osellame LD. Biomarkers for cancer cachexia: a mini review. Int J Mol Sci. 2021;22(9):4501. doi: 10.3390/ijms22094501 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref022] 22.Nicholson BD, Aveyard P, Bankhead CR, Hamilton W, Hobbs FDR, Lay-Flurrie S. Determinants and extent of weight recording in UK primary care: an analysis of 5 million adults’ electronic health records from 2000 to 2017. BMC Med. 2019;17(1):222. doi: 10.1186/s12916-019-1446-y [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref023] 23.Price SJ, Stapley SA, Shephard E, Barraclough K, Hamilton WT. Is omission of free text records a possible source of data loss and bias in Clinical Practice Research Datalink studies? A case-control study. BMJ Open. 2016;6(5):e011664. doi: 10.1136/bmjopen-2016-011664 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref024] 24.Koo MM, Hamilton W, Walter FM, Rubin GP, Lyratzopoulos G. Symptom signatures and diagnostic timeliness in cancer patients: a review of current evidence. Neoplasia. 2018;20(2):165–74. doi: 10.1016/j.neo.2017.11.005 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref025] 25.Mounce LTA, Calitri R, Hamilton W, Rafiq M, Emery JD, Giannakeas V, et al. Improving the clinical utility of platelet count for cancer detection in primary care: a cohort study in England, Canada, and Australia. Cancers (Basel). 2024;16(17):3074. doi: 10.3390/cancers16173074 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref026] 26.Wu R, Arendse KD, Hamdani T, Walter FM, Crosbie EJ, Mihaylova B, et al. Cost-effectiveness of CA125- and age-informed risk-based triage for ovarian cancer detection in primary care. Br J Cancer. 2025;133(10):1492–500. doi: 10.1038/s41416-025-03166-3 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref027] 27.Crooks CJ, West J, Jones J, Hamilton W, Bailey SER, Abel G, et al. Aliment pharmacology therapeutics. 2025;61(5):852–64. doi: 10.1111/apt.18459 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref028] 28.Tamm A, Shine B, James T, Withers J, Salih H, Noble T, et al. External validation of the COLOFIT colorectal cancer risk prediction model in the Oxford-FIT dataset: the importance of population characteristics and clinically relevant evaluation metrics. BMC Med. 2025;23(1):503. doi: 10.1186/s12916-025-04339-w [DOI] [PMC free article] [PubMed] [Google Scholar]

[pmed.1004956.ref029] 29.Nicholson BD, Oke J, Virdee PS, Harris DA, O’Doherty C, Park JE, et al. Multi-cancer early detection test in symptomatic patients referred for cancer investigation in England and Wales (SYMPLIFY): a large-scale, observational cohort study. Lancet Oncol. 2023;24(7):733–43. doi: 10.1016/S1470-2045(23)00277-2 [DOI] [PubMed] [Google Scholar]

[pmed.1004956.ref030] 30.Dowrick A, Ziebland S, Rai T, Friedemann Smith C, Nicholson BD. A manifesto for improving cancer detection: four key considerations when implementing innovations across the interface of primary and secondary care. Lancet Oncol. 2024;25(8):e388–95. doi: 10.1016/S1470-2045(24)00102-5 [DOI] [PubMed] [Google Scholar]

[pmed.1004956.ref031] 31.Clinical Practice Research Datalink. 2025. [cited 29 April 2026]. Available from: https://www.cprd.com/ [Google Scholar]

PERMALINK

Blood test trend versus single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss: A retrospective cohort study

Brian D Nicholson

Clare R Bankhead

Sufen Zhu

Jason Lee Oke

Cynthia Wright Drakesmith

Rafael Perera

Frederick D Richard Hobbs

Pradeep S Virdee

Roles

Abstract

Background

Methods and findings

Conclusions

Author summary

Why was this study done?

What did the researchers do and find?

What do these findings mean?

Introduction

Methods

Study design

Participants

Test methods

Outcome.

Clinical features.

Analysis

Data analysis.

Sample size.

Results

Summary of patient data

Fig 1. Patient flow diagram.

Table 1. Summary (n (%)) of patient demographics and co-occurring clinical features at UWL.

Table 2. Summary of cancer diagnosis within 6 months following UWL.

Fig 2. Trends in four blood tests in men aged 70 years at UWL (as examples).

Association between blood test trend and cancer diagnosis

Unadjusted analyses.

Table 3. Hazard ratio (95% CI) of blood test abnormality and trend for increased 6-month cancer risk following UWL.

Fig 3. AUC (95% CI) of blood test trend and abnormality for 6-month cancer risk following UWL.

Adjusted analyses.

Fig 4. Instances where the age- and sex-adjusted AUC favoured blood test trend over blood test abnormality.

Subgroup analyses

Cancer site.

Number of tests.

Discussion

Ethics approval

Supporting information

Acknowledgments

Abbreviations

Data Availability

Funding Statement

References

Decision Letter 0

Heather Van Epps

Roles

Decision Letter 1

Alexandra Tosun

Roles

Author response to Decision Letter 2

Decision Letter 2

Katarzyna Starowicz

Roles

Author response to Decision Letter 3

Decision Letter 3

Katarzyna Starowicz

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases