{"paper_id":"436c2353-c626-46c3-aa5f-8ba4b1863966","body_text":"1 \n \n \nCancer incidence and competing mortality risk following 15 presenting symptoms in primary \ncare: a population-based cohort study using electronic healthcare records \n \nMatthew Barclay, Senior Research Fellow 1 \nCristina Renzi, Associate Professor 1,2 \nHannah Harrison, Research Associate 3 \nAna Torralbo, Senior Research Fellow 4 \nBecky White, Senior Research Fellow 1 \nSamantha Ip, Research Associate 3,5 \nJuliet Usher-Smith, Clinical Lecturer 3 \nJane Lange, Staff Scientist 6 \nNora Pashayan, Professor 3,7 \nSpiros Denaxas, Professor 4 \nAngela Wood, Professor 3,5,8,9,10,11 \nAntonis C Antoniou, Professor 3 \nGeorgios Lyratzopoulos, Professor 1 \n \n1 Department of Behavioural Science and Health, Institute of Epidemiology and Healthcare, \nUniversity College London, London, United Kingdom,  \n2 Faculty of Medicine, University Vita-Salute San Raffaele, Milan, Italy,  \n3 Department of Public Health and Primary Care, School of Clinical Medicine, University of \nCambridge, Cambridge, United Kingdom,  \n4 Institute of Health Informatics, University College London, London, United Kingdom,  \n5 Victor Phillip Dahdaleh Heart and Lung Research Institute, University of Cambridge, Cambridge,  \nUnited Kingdom,  \n6 Cancer Early Detection Advanced Research Center, Oregon Health & Science University, Portland, \nOregon, United States of America, \n7 Department of Applied Health Research, Institute of Epidemiology and Healthcare,  University \nCollege London, London, United Kingdom,  \n8 British Heart Foundation Centre of Research Excellence, University of Cambridge, Cambridge, \nUnited Kingdom,  \n9 National Institute for Health and Care Research Blood and Transplant Research Unit in Donor \nHealth and Behaviour, University of Cambridge, Cambridge, United Kingdom,  \n10 Health Data Research UK Cambridge, Wellcome Genome Campus and University of Cambridge, \nCambridge, United Kingdom,  \n11 Cambridge Centre for Artificial Intelligence in Medicine, University of Cambridge, Cambridge, \nUnited Kingdom \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n2 \n \nSummary box \nWhat is already known on this topic \n• Evidence describing the diagnostic value of symptoms for cancer can help to assess \nwhich patients who present to primary care need urgent specialist assessment \n• Current evidence is limited as age is often handled categorically, smoking status is \nnot taken into account and study periods are historical. \n• Further, evidence is concentrated on assessing the risk of specific cancer sites, \nalthough the same symptom can be related to cancer of different organs. \nWhat this study adds \n• We present evidence on age-, sex-, and smoking status-specific estimates of risk of \ncancer of different organs and overall, alongside estimates of non-cancer death.  \n• Estimates relate to patients who present with one of 15 possible cancer symptoms, \nfrom a relatively recent time period.  \n• Certain symptoms such as jaundice and dysphagia are associated with high risk of \nnon-cancer death in older patients. \n• Other symptoms, such as unintended weight loss, fatigue and abdominal pain, are \nassociated with excess risk of a range of different cancers, and such evidence can \nguide the choice of diagnostic strategies and the design of multi-cancer diagnostic \nservices. \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n3 \n \nAbstract \n \nObjectives: Comprehensive assessment of risk of cancer diagnosis and non-cancer \nmortality following primary care consultation for 15 new-onset symptoms.  \n \nDesign: Cohort study. \n \nSetting: UK primary care (CPRD Gold), 2007 – 2017. \n \nParticipants: Patients aged 18-99, comprising a randomly-selected reference group and a \nsymptomatic cohort of patients presenting with one of 15 new onset symptoms (abdominal \npain, abdominal bloating, rectal bleed, change in bowel habit, dyspepsia, dysphagia, \ndyspnoea, haemoptysis, haematuria, fatigue, night sweats, weight loss, jaundice, breast \nlump, post-menopausal bleed). \n \nMain outcome measures: Risk of cancer diagnosis and risk of death in the 12 months \nfollowing index consultation. Time-to-event models were used to estimate outcome-specific \nhazards for site-specific cancer diagnosis and non-cancer mortality; results were combined \nusing the latent failure time approach to estimate cumulative incidence. \n \nResults: Data were analysed on 1,622,419 patients, of whom 36,802 had a cancer \ndiagnosis and 28,857 died without a cancer diagnosis within 12 months of first consultation. \nAbsolute non-cancer mortality risk exceeded cancer diagnosis risk in the reference group \nand in symptomatic patients with dyspnoea, dysphagia, weight loss, fatigue, or jaundice; \nabsolute cancer risk exceeded mortality risk for patients with breast lump or post-\nmenopausal bleed; for other symptoms the risk of a cancer diagnosis and non-cancer \nmortality were similar. \n \nEver-smoking was associated with raised cause-specific hazard for lung cancer (e.g., in \nwomen HR 4.8, 95%CI 4.2 to 5.6), and slightly raised hazards for upper GI and urological \ncancers. \n \nFor patients with red-flag symptoms, the risk of specific cancers exceeded the UK urgent \nreferral risk threshold of 3% from a relatively young age (e.g., for male smokers with \nhaemoptysis the risk of lung cancer exceeded 3% from age 55). For non-organ-specific \nsymptoms (such as loss of weight, or fatigue), while the risk of any cancer often exceeded \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n4 \n \n3%, the risk of any individual cancer type either did not reach this threshold at any age, or \nreached it only in older patients. \n \nConclusions: In patients with new-onset symptoms in primary care the risk of cancer \ndiagnosis and of non-cancer mortality are often comparable. Smoking-status is highly \ninformative for cancer risk in patients with respiratory or non-organ-specific symptoms. A \nholistic approach to risk assessment that includes the risk of multiple different cancer types \nalongside the risk of mortality due to consequential illnesses other than cancer, especially \namong older patients, is needed to inform management of symptomatic patients in primary \ncare, particularly for patients with non-organ-specific symptoms. \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n5 \n \nIntroduction \n \nMost patients with cancer are diagnosed after symptomatic presentation [1], and, given the \npaucity of effective tests to enable population-based cancer screening, this is likely to be the \ncase for the coming decade. Appropriately suspecting the diagnosis of cancer in \nsymptomatic patients is difficult, as symptoms may be caused by many other diseases. Even \nso-termed ‘alarm’ or ‘red-flag’ symptoms typically have positive predictive values for cancer \nthat do not exceed 5% in women of any age or in men younger than 70 [2]. In the UK, many \npatients with cancer experience diagnostic delays in the form of multiple pre-referral \nconsultations and prolonged intervals to diagnosis, despite practice guidelines issued by the \nNational Institute for Health and Social Care Excellence (NICE) that aimed to enable prompt \ndiagnosis of cancer in primary care [7,8]. Such delays are associated with adverse patient \nexperience and worse clinical outcomes [3–6], \n \nCurrently, most evidence supporting practice guidelines comes from case-control studies, \nexamining symptom-related risk of specific cancer sites. This study design ignores that \npresenting symptoms are often shared between different cancers and diseases other than \ncancer; there has been no comprehensive examination of the risk of the full spectrum of \npossible cancer types for most relevant presenting symptoms. Further, guideline \nrecommendations handle major cancer risk factors sub-optimally, as smoking status is \ntypically ignored as a risk stratifier, and age typically not considered as a continuous \nvariable, leading to information loss. Competing risk of death is also ignored, meaning that \nmanagement decisions centred on cancer risk ignore risks related to other diseases.  \n \nThis study is motivated by the need for evidence to support the updating of clinical practice \nguidelines for the primary care management of patients who present with symptoms of \npossible underlying cancer. Such evidence is needed both in terms of quantifying the \nabsolute risk of different cancer types and also the probability of patients dying without a \ncancer diagnosis. We also aim to aid the development of and complement the use of risk \nprediction tools by describing in detail the associations between symptoms and cancer risk \n[9,10]. We therefore provide a comprehensive assessment of risk of cancer diagnosis and \nnon-cancer mortality following consultation for 15 new-onset symptoms. \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n6 \n \nMethods \n \nStudy population \n \nWe used medical records from English National Health Service general practices that \ncontributed anonymized primary-care electronic health records to the Clinical Practice \nResearch Datalink Gold (CPRD), covering approximately 6.9% of the UK population [11]. \nPatients in CPRD are broadly representative of the UK general population with respect to \nage, sex, and ethnicity [11]. CPRD was linked to cancer diagnosis information from the \nEnglish national cancer registry [12].  \n \nWe first extracted a random sample of patients from CPRD for use as a reference group, \nchoosing index dates randomly from ‘valid’ follow-up during 2007-01-01 to 2017-12-31. We \nthen created a symptomatic cohort of all patients in CPRD Gold who had consulted for any \nof 15 presenting symptoms and who were not in the reference group, choosing the index \ndate as the date of their first ‘valid’ consultation for a symptom during 2007-01-01 to 2017-\n12-31. \n \nFor an individual patient, follow-up was judged to be ‘valid’ if: they had been registered at \ntheir practice for at least one year; their practice was judged by CPRD to be providing data \nof a suitable standard for use in research (i.e., after the practice’s “up-to-standard” date); it \nwas before the last data transfer to CPRD (i.e., the “last collection” date); the patient was \nregistered at a CPRD practice (i.e., before the patient’s “transfer out” date, and before their \ndeath); the patient was aged 30-99; and the patient had not yet had a recorded cancer \ndiagnosis in the cancer registry (excluding non-melanoma skin cancer). \n \nA study flowchart is given in Appendix 1 Table 1. \n \nOutcomes \n \nBoth mortality and cancer diagnoses were considered. Mortality was identified from the \nprimary care record; such information is highly concordant with the ‘gold standard’ official \ndeath registration records and is correct within one month 98% of the time [13]. Cancers \nwere split into seven groups for men and eight groups for women, summarised below and \nwith a full ICD10 codelist in Appendix 1 Table 2, guided by underlying body systems and \ncorresponding major clinical specialities receiving urgent referrals for suspected cancer in \nEngland [14]. Cancer diagnoses were sourced from linkages with the national cancer \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n7 \n \nregistry and only the first cancer diagnosis was considered; available cancer data covered \ndiagnoses up until 2018-12-31. As non-melanoma skin cancer is imperfectly registered and \nprimarily managed in primary care, diagnoses of non-melanoma skin cancer were not \nconsidered in this study. \n \nThe cancer groups considered were: \n• Breast cancer (women only), including invasive breast and in-situ breast cancers \n• Gynaecological cancer (women only), including invasive cervical, in-situ cervical, \novarian, uterine, and vulvar cancers \n• Lung, including lung cancer and mesothelioma \n• Upper gastrointestinal (GI), including liver, oesophageal, pancreatic and stomach \ncancers \n• Lower GI, including colon and rectal cancers \n• Urological, including bladder, in-situ bladder, kidney and other urinary tract cancers \n• Prostate cancer (men only) \n• Haematological, including Hodgkin lymphoma, non-Hodgkin lymphoma, acute \nmyeloid leukaemia, chronic lymphocytic leukaemia, other leukaemias, myeloma, and \nother haematological cancers \n• Other, including all other sites, specifically including melanoma, unknown primary, \nthyroid, and meningeal cancers, also including testicular cancer and male breast \ncancer \n \nThe first outcome (of cancer diagnosis or non-cancer death) experienced by each patient \nwas considered in the analysis. This means, for example, that in the analyses of cumulative \nincidence a patient who died shortly following a cancer diagnosis would only be considered \nto have had a cancer diagnosis, and their death would not contribute to the estimation of \nmortality risk irrespective of cause of death. Patients with a cancer diagnosis on the same \nday as their death (including, for example, death certificate only registrations of cancer) were \ntreated as having had a cancer diagnosis rather than having died, noting that death \ncertificate only registrations remained <0.4% through the study period [15]. \n \nSymptoms \n \nWe considered a subset of symptoms known to have an association with risk of specific \ntypes of cancer and that are already included in referral guidelines for symptomatic cancer \n[7,16]. The included symptoms form part of the presentation in 40% of all patients with \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n8 \n \ncancer England [1]. We identified symptoms from coded primary care data using existing \nRead v2 phenotyping algorithms [16]. The symptoms we considered were: \n• Abdominal symptoms \no Abdominal pain \no Abdominal bloating \no Rectal bleeding \no Change in bowel habit \no Dyspepsia \no Dysphagia \no Jaundice \n• Respiratory symptoms \no Dyspnoea \no Haemoptysis \n• Urological symptoms \no Haematuria \n• Non-specific symptoms \no Fatigue \no Night sweats \no Weight loss \n• Breast and reproductive organ symptoms \no Breast lump (including in men) \no Post-menopausal bleeding \n \nOnly the first presenting symptom for each patient was included, and each patient was \nincluded at most once in the analysis. For example, if a patient had a consultation for breast \nlump in 2007 that did not result in a cancer diagnosis and a consultation for abdominal pain \nin 2010 that did result in a cancer diagnosis, only the risk after the 2007 consultation for \nbreast lump would be included in analysis. If two or more of the examined symptoms \npresented on the same day, all were included as index symptoms (such occurrences were \nrare, see end of Results). \n \nSmoking status, sex, and age \n \nPatients were categorised as ever-smokers or never-smokers. Ever-smokers included all \npatients with a record of being a current or ex-smokers in their entire primary care record, \nincluding periods after cancer diagnosis or before their record became eligible for use in this \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n9 \n \nstudy; never-smokers included all other patients. Patients were classed as male or female \nbased on the recorded gender in their primary care record. Patients’ age was estimated as \nthe number of years between the mid-point of their year of birth and their index date. \n \nStatistical methods \n \nInitial analysis described the distribution of patients in the sample and counts of cancer \ndiagnoses and deaths within 12 months of any index symptom. \n \nHazards for specific cancers and non-cancer mortality were estimated using semi-parametric \n(Royston-Parmar) time-to-event models [17]. Follow-up for these analyses was censored at \nthe earliest of 18 months after the index symptom, at first event (i.e., cancer diagnosis or \ndeath), or at the end of the available cancer registry follow-up on 2018-12-31. Models were \nstratified by sex and included the following covariates: \n• Age (restricted cubic spline with six knots) \n• Smoking status (binary, ever record of smoking in primary care data vs never) \n• Index symptom (15 binary variables indicating the symptom(s) each patient had on \ntheir index date (all zero for patients in the reference group)) \n• An interaction with follow-up time in months for each index symptom, allowing the \nassociation between symptom and cause-specific risk to decay over time. This was \nmotivated by the fact that following many possible symptoms of cancer, excess risk is \nhighest in the first months following presentation (e.g., [18]) \n \nCumulative incidence of cancer group and non-cancer mortality was estimated by combining \neach of the cause-specific models using the latent failure time approach [19]. We report \ncumulative incidence for combinations of age-sex-smoking-symptom up to 12 months follow-\nup, with results focusing on estimated cumulative incidence at 12 months and age \nconsidered in five-year intervals. To sense-check these model-based estimates, we \nadditionally examined the crude cumulative incidence for each cancer group and non-cancer \nmortality within 12 months of each symptom by sex and smoking status using Aalen-\nJohansen non-parametric cumulative incidence curves [20,21]. \n \nConcordant with the methods and evidence that informed the development of NICE \nguidelines, we have considered the modelled cumulative incidence at 12 months to \nrepresent the positive predictive value for the outcome for the symptom [7]. Further, we \ncalculated the (sex/smoking/symptom-specific) age at which cancer risk exceeded the 3% \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n10 \n \nrisk threshold for referrals used in the UK. We additionally present similar estimates for each \nindividual cancer group. \n \nStatistical modelling used Stata 17 MP. Simulation of failure times was performed on a high-\nperformance cluster using Stata 16 MP. Survival models were fit using the merlin package \n[22], and multistate modelling was facilitated by the multistate package [23]. Data extraction \nand analysis code are available at \nhttps://github.com/MattEBarclay/cprd_symptom_cancer_1. \n \nPatient and public involvement \nThe study forms part of a programme of work examining the predictive value of symptoms \nfor cancer diagnosis using electronic health records data. To support this programme, we \nran three focus groups in August and September 2023 including a total of 15 patient and \npublic involvement volunteers. Study reporting was informed by PPI input, but no specific \nchanges were made.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n11 \n \nResults \n \nThe analysis cohort included 1,622,419 patients, 835,995 with an eligible first symptom \nrecorded between 2007 and 2017 (Table 1). More than half of the cohort (64%, 1,040,762) \nwere aged under 60 at index, with 24,731 (1.5%) patients aged 90 or older. The distribution \nof symptoms was uneven, with 14.4% of the cohort having abdominal pain as index \nsymptom, followed by dyspnoea (8.7%), fatigue (8.1%), dyspepsia (6.7%), rectal bleeding \n(3.0%), breast lump (2.4%), haematuria (1.6%), abdominal bloating (1.4%), weight loss \n(1.2%), change in bowel habit (1.1%), dysphagia (0.9%), post-menopausal bleeding (0.5%), \nnight sweats (0.5%), haemoptysis (0.4%), and jaundice (0.1%). The majority of patients \n(64%) had at least one smoking-related Read code in their records and were identified as \never-smokers. Within 12 months of their first recorded symptom, 36,802 patients had a \ncancer diagnosis and 28,867 patients died without a cancer diagnosis (a further 9,288 died \nfollowing a cancer diagnosis); both cancer and mortality risk were higher in older patients. \nEver-smokers had slightly higher cancer risk than patients without any smoking-related \ncodes.  \n \nAge-adjusted cancer-specific hazard ratios for smoking and each index symptom \n \nBoth male and female ever-smokers had far higher cancer-specific hazard of lung cancer \nthan non-smokers (Figure 1 and Appendix 4, HR 4.8, 95%CI 4.2-5.6, for women and HR 4.0, \n95%CI 3.5-4.6, for men), and elevated hazards of urological (e.g., for men: HR 1.4, 95%CI \n1.2-1.5, Appendix 4 Table 4) and upper GI cancers (e.g., for men: HR 1.4, 95%CI 1.2-1.5, \nAppendix 4 Table 1).  \n \nPatients consulting for symptoms of possible cancer had similar or greater cause-specific \nhazards for almost every cancer site than the reference population (Figure 1 and Appendix \n4). Yet for ten of the fifteen studied symptoms, the symptom was associated with lower \ncause-specific hazards for death than the reference group (the exceptions being dysphagia, \njaundice, dyspnoea, haemoptysis, and weight loss). Further, for many symptoms associated \nwith very high initial hazard of a specific cancer, while the hazard typically remained elevated \nat least to 12 months after the index consultation, it tended to reduce over time (Figure 1). \n \nAbdominal symptoms (abdominal pain, abdominal bloating, rectal bleeding, change in bowel \nhabit, dyspepsia, dysphagia, jaundice) \nFor both men and women presentations with abdominal symptoms were associated with \nincreased hazard of multiple types of cancer. At the same time, abdominal symptoms were \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n12 \n \nassociated with decreased hazard of death without a cancer diagnosis when compared with \nthe reference group, except for dysphagia and jaundice (Figure 1, and Appendix 4 Tables 2-\n3 and 12-13). Cause-specific hazard ratios at one month after presentation were highest \nregarding lower GI cancer for rectal bleeding and change in bowel habit (e.g., CIBH for men: \nHR 17.4, 95% CI 15.7-19.4) and highest regarding upper GI cancer for jaundice and \ndysphagia (e.g., dysphagia in women: HR 16.4, 95%CI 14.0-19.2); hazard ratios decreased \nsubstantially over follow-up for these symptoms. Abdominal pain and abdominal bloating \nwere associated with hazard ratios at consultation of around 4 for both upper and lower GI \ncancers (e.g., abdominal bloating in women with HR for lower GI cancer of 3.0, 95%CI 2.3-\n4.0), with abdominal bloating having a similar association for gynaecological cancers in \nwomen (HR 4.8, 95%CI 4.0-5.6), while dyspepsia was associated with a hazard ratio of \naround 4 for upper GI cancer. Patients with abdominal symptoms also appeared at elevated \nrisk for urological and haematological cancers, and for prostate and gynaecological cancers. \n \nRespiratory symptoms (dyspnoea, haemoptysis) \nRespiratory symptoms were primarily associated with lung cancer, but the strength of the \nassociation varied (Figure 1 and Appendix 4 Tables 1 and 11). Patients with haemoptysis \nhad a cause-specific hazard ratio of around 16 at consultation compared with the reference \ngroup (e.g., for men, HR 17.1, 95%CI 14.8-19.8), while the association with dyspnoea was \nweaker but still notable (e.g., for men, HR 2.6, 95%CI 2.4-2.9). Other types of cancer, \nnotably haematological cancers, also had elevated cause-specific hazards; (e.g., for men, \nthe HR for haematological cancer being 2.8, 95%CI 1.7-4.6, Appendix 4 Tables 6 and 15). \n \nUrological symptoms (Haematuria) \nHaematuria in women was primarily associated with urological cancers (HR 57, 95%CI 48-\n67) and with gynaecological cancers (HR 4.6, 95%CI 3.7-5.6) (Figure 1 and Appendix 4 \nTables 10 and 14). In men, it was associated with urological cancers (HR 45, 95%CI 40-50) \nand prostate cancer (HR 5.3, 95%CI 4.8-5.8) (Appendix 4, Tables 4 and 5). \n \nNon-specific symptoms (Fatigue, night sweats, weight loss) \nNon-specific symptoms were typically associated with elevated cause-specific hazard ratios \nfor all cancer groups considered (Figure 1 and Appendix 4), and generally HRs appeared \nrelatively similar in strength for each of the three non-specific symptoms. Weight loss had \nthe strongest associations overall (cancer-specific HRs general between 2 and 5), followed \nby night sweats (HRs generally between 1 and 4, though imprecisely estimated), followed by \nfatigue (HRs between 1 and 2). It often appeared that the strongest cause-specific \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n13 \n \nassociations were for haematological cancers, though confidence intervals tended to overlap \nwith those of other cancer groups. \n \nBreast and reproductive organ symptoms (breast lump, post-menopausal bleeding) \nPost-menopausal bleeding was associated with large cause-specific hazard ratios for \ngynaecological cancer (HR 43, 95%CI 39-47) and substantial cause-specific HRs for \nurological cancer (HR 4.1, 95%CI 2.6-6.4) (Figure 1 and Appendix 4 Tables 10 and 14). \nBreast lump in women was associated principally with breast cancer (HR 65, 95%CI 61-69) \nand to a lesser extent with haematological cancer (HR 2.6, 95%CI 1.80-3.6) (Appendix 4 \nTables 9 and 15). A small number of men present with breast lump, and these men had \ncause-specific hazard ratios for the ‘other cancers’ group, which included male breast \ncancer, of 7.1 (95%CI 5.0-10.0) (Appendix 4 Table 7). \n \nRisk of specific cancer sites by age, sex, and smoking status \n \nAfter symptom presentation for patients with single index symptoms, and based on \nsimulations combining the cause-specific models, we present simulated cumulative \nincidence of each cancer site and of death without cancer at 3 months (Appendix 2), 6 \nmonths (Appendix 3), and 12 months (Figures 2-5, Appendix 5). Hereafter in this section, we \ndiscuss cumulative incidence at 12 months after symptom consultation. Unlike the hazard \nratios presented above, estimates of cumulative incidence varied substantially by sex, as \nwomen have lower baseline cancer risk.  \n \n3% any cancer risk thresholds at 12 months \nPatients reaching a 3% risk of any cancer may not reach such a risk level for any specific \ncancer group, especially for symptoms associated with multiple types of cancer. For \nexample, female smokers presenting with weight loss had a 3% risk of cancer from age 60, \nbut did not reach the 3% risk threshold at any age when any of the individual cancer groups \nwere considered on their own (Table 2). For male non-smokers, risk of any cancer reached \nthe 3% threshold from the following ages and onwards: 45 for jaundice; 55 for dysphagia, \nweight loss, haematuria, and change in bowel habit; 60 for haemoptysis and rectal bleeding; \n65 for abdominal pain and bloating, night sweats and breast lump; and 70 for dyspepsia, \ndyspnoea, and fatigue (Table 2). For smokers, this threshold was often reached up to five \nyears younger. Conversely, compared with male patients presenting with the same \nsymptom, female patients reached the 3% threshold at an older age on average, with the \nmain exception being breast lump for which the 3% threshold (in women) was reached from \nage 40. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n14 \n \n \nNotably, male smokers in the reference group had a 3% risk of any cancer from age 75, and \nmale non-smokers from age 90; women in the reference group did not reach a 3% risk of \ncancer at any age. \n \nA summary of risk of individual cancers is given in Appendix 6, plus additional graphical and \ntabular results in Appendices 3 and 5. \n \nRisk of non-cancer mortality \n \nFor most of the studied symptoms, symptomatic patients were less likely to die (without a \ncancer diagnosis) than similar patients in the reference group (Figures 2-5). The three \nprincipal exceptions were jaundice, dysphagia and weight loss, for which post-presentation \nmortality exceeded that in the reference group, and also older patients with less-specific \nsymptoms for whom the risk of non-cancer mortality was often higher than the risk of any \ncancer. For example, for male smokers presenting with dyspnoea, around 6% who \npresented at age 80 would develop cancer within 12 months while 9% would die (Figure 3, \nAppendix 5 Table 1). \n \nPresentation with multiple symptoms \n \nAmong symptomatic patients, 1.2% (10,360 of 835,995) consulted for more than one of the \nfifteen studied symptoms on their index date, and a further 2.5% (21,167) consulted for an \nadditional studied symptom within 30 days of an index symptom but before a cancer \ndiagnosis. The proportion of patients with multiple index symptoms subsequently diagnosed \nwith cancer within 12 months of index (4.6%, 95% CI 4.2% to 5.1%) was higher than for \npatients with a single index symptom (3.5%, 95% CI 3.5% to 3.5%). This higher risk of \ncancer in patients with multiple index symptoms appeared applicable to many of the \nsymptoms considered, but sample size limitations meant proportions developing cancer \ncould often not be estimated precisely. \n \nThe cause-specific time-to-event models accommodated multiple index symptoms that were \nconsulted for on the same day, so for example the cause-specific hazard ratio for upper GI \ncancer for abdominal pain is already adjusted for the presence of dysphagia, for the \ninfrequent occasions (see above) where both were recorded – although possible interaction \neffects were not considered. Symptoms that were not consulted for on the same day as \nindex were not considered. In principle, estimates of cancer risk for any combination of \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n15 \n \nsymptoms can be estimated from the cause-specific models, but these have not been \nproduced due to computational limitations and the very large number of potential \ncombinations. \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n16 \n \nDiscussion \n \nUsing a cohort design, we comprehensively estimated the risk of different cancer diagnoses \nand non-cancer mortality following presentation in primary care with one of 15 index \nsymptoms, and in a reference group that was not selected based on symptom status and so \nshould approximate the risk in the general population. There was considerable variation in \nrisk by age and by sex. Smoking-status was highly informative for cancer risk for patients \nwith respiratory or non-organ-specific symptoms. Smokers typically reached the 3% \nthreshold warranting referral for cancer investigations up to five years younger than non-\nsmokers. The findings highlight the importance of including smoking status in clinical \nguidelines and referral decisions in patients with a new onset symptom. Even symptoms with \nstrong, well-established associations (e.g., dyspnoea and lung cancer) have notable \nassociations with other types of cancer (e.g., haematological cancers). We also provide \nestimates of cancer risk while considering the potential for non-cancer mortality. For the \noldest patients – and for those with symptoms such as dysphagia or jaundice – risk of death \nwithout a cancer diagnosis reached or exceeded the risk of cancer. Referral decisions based \non a universally applied 3% cancer risk threshold, as currently set out in UK clinical \nguidelines, may not be appropriate for these patients. \n \nStrengths and weaknesses \n \nKey strengths of the study are (a) the large representative dataset – allowing examination of \na range of both common and rare symptoms and outcomes – (b) the joint estimation of the \nrisks of the different outcomes, including of non-cancer mortality and risk of different types of \ncancer, and (c) the use of cancer registry data to ascertain presence of cancer, as cancer \nmay be under- or over-recorded in non-registry sources [24]. While this study represents the \nmost comprehensive and detailed description of risk of cancer in symptomatic patients to \ndate, there are various areas where future work could make further improvements. \n \nConsidering limitations, the study only considers deaths in patients without cancer, but it \nmay be important to understand if patients die quickly after a cancer diagnosis. Our measure \nof smoking status does not allow for a refined appreciation of smoking history and dose-\nresponse relationships. Additionally, our analytical approach only allowed each patient to be \nincluded once, not making full use of the longitudinal nature of EHR datasets [25]. We did \nnot consider interactions between symptoms and simulated outcomes for patients with a \nsingle symptom only, in part due to only few patients having multiple symptoms. We did not \nhave access to free-text data, despite evidence that coded data does not capture all \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n17 \n \nsymptoms [26,27]. Finally, we only examined 15 symptoms, ignoring the many other \nsymptoms and important health conditions that may be associated with risk of cancer \n[1,16,28]. A more detailed examination of potential limitations is given in Appendix 7. \n \nComparison with literature \n \nA large and growing literature describes risk of cancer following symptom presentations in \nprimary care; Moore and colleagues summarised the literature pre-2020 [16], and there are \nseveral recent papers [18,31–33]. Existing literature (a) rarely considers competing non-\ncancer mortality risk, (b) rarely considers smoking status, and (c) frequently provides no or \nonly limited information on the age-dependent and sex-specific nature of the risk of different \ncancers. Much of the previous evidence additionally considers either the risk of all cancers \ncombined or focuses on specific cancer sites judged to be of relevance to the specific \nexamined symptoms a priori. We improve on previous descriptive studies by presenting a \nbroad range of possible cancer diagnoses following presentation with wider spectrum of \nindex symptoms. Further research is needed to extend analyses similar to those reported \nhere to a wider collection of symptoms. \n \nSome existing evidence on so-called red flag symptoms such as rectal bleeding and \nhaemoptysis suggests the risk of cancer exceeds 3% for all ages, but did not examine the \nrisk in different age groups [16]; our findings indicate that risk of cancer following these \nsymptoms only exceeds 3% beyond certain age cut-offs. Furthermore, we show that for non-\nspecific symptoms, the risk of any cancer exceeds 3% at a considerably earlier age than the \nrisk of a specific cancer type, underscoring the need for studies that comprehensively \nexamine all major cancer types. Weight loss provides a cardinal example, where risk of any \ncancer exceeded 3% in male non-smokers from age 55 but risk of any individual site only \nreached 3% at age 85. \n \nOther studies have aimed to develop risk prediction tools for cancer intended for use in a \nprimary care setting (see for example, [34–36]), and in particular the QCancer risk prediction \ntool [9,10] already considers a range of symptoms and risk of diagnosis of different types of \ncancer. For decisions about the management of an individual patient, a risk prediction tool \nincluding multiple potential predictors may be more suitable than the results presented in this \npaper. We view our results as complementary; by describing what is effectively the average \nrisk in patients presenting with these symptoms (by age, sex, and smoking status), we can \ninform high-level policy decisions around symptomatic diagnosis of cancer such as clinical \nguideline recommendations, and help developers of more detailed risk prediction models by \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n18 \n \nhighlighting symptoms they may wish to consider. Further, our consideration of mortality risk \nprovides relevant information that is frequently missing from current risk prediction tools \n(including QCancer) and that is especially important in frail and elderly populations. \n \nImplications \n \nSymptoms recorded in primary care data can be highly informative about both cancer risk \nand short-term mortality risk. In some cases, for example lung cancer, smoking-status is \nvery strongly associated with the risk of cancer following a certain symptom. Risk of cancer \nand non-cancer mortality varies considerably by age; describing “overall” risk of cancer \nfollowing a symptom may be misleading if non-cancer mortality is not considered. Some \n(non-cancer) deaths will relate to as-yet undiagnosed disease which, like cancer diagnosis, \nnecessitates specialist assessment in secondary care, though this should be the subject of \nfuture enquiries. \n \nFor researchers, our results underline the methodological importance of accounting for the \nfact that symptoms may be associated with multiple different disease outcomes. Advanced \nstatistical modelling strategies are helpful in assessing diagnostic outcomes using EHR data, \nand current statistical packages allow for relatively straightforward handling of competing \nrisks either by directly modelling cumulative incidence (e.g., the Fine-Gray model [37]) or, as \nhere, by combining several cause-specific models [38]. Diagnostic research should adopt \nstrategies that allow consideration of risk of several potentially related diseases (e.g., \nmultiple types of cancer, as in this study), which can be done even with simple analytical \napproaches such as appropriate use of logistic regression [32]. \n \nFor clinicians and policy makers, our systematic assessment of risk of cancer (and of non-\ncancer mortality) in symptomatic patients in primary care raises two key questions. \n \nFirst, whether all age-sex-smoking status groups presenting with each of the studied \nsymptoms and with an estimated any-cancer risk of above 3% should explicitly be added to \nNICE referral guidelines. This may indeed be justified, though given the high mortality rates \nin the oldest patients, there might also be a risk of over-testing in older men in particular. \nHowever, the degree to which risk of over-testing is a concern relates to the exact causes of \nnon-cancer mortality and the extent to which it relates to pre-diagnosed or new non-\nneoplastic diseases which could benefit from specialist diagnostic assessment and earlier \ndiagnosis. As the components of non-cancer mortality due to pre-existing or new conditions \nis unclear, this should be addressed by future research. The current approach to cancer \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n19 \n \nreferral uses a normative threshold applicable to patients of any age and with any \nsymptoms, and the results highlight the importance of considering whether patients are likely \nto benefit from prompt diagnosis. \n \nSecond, whether current referral pathways are necessarily ideal. For example, many \nabdominal symptoms were strongly associated with lower GI, upper GI and gynaecological \ncancers, and some form of referral pathway offering combined multi-specialty assessment \nmay be justified for patients with these symptoms. Further, symptoms were often strongly \nassociated with less common cancers such as haematological neoplasms but, due to the \nlow incidence of these conditions, absolute risk rarely or never reached 3%; optimal \ndiagnostic management of these patients is clearly challenging. Our findings may be helpful \nin clarifying referral criteria for new non-specific cancer pathways. \n \nConclusions \n \nThe risk of cancer diagnosis and non-cancer mortality after symptomatic presentation can be \ncomparable and both should be considered in referral and investigation decisions – \nalongside age, sex, and smoking status. A holistic and stratified assessment of risk in \nsymptomatic patients, which considers the risk of a cancer diagnosis, the risk of a diagnosis \nof individual types of cancer, and the risk of non-cancer mortality is needed particularly for \npatients presenting with which are vague or non-specific symptoms associated with multiple \ncancer types and appreciable non-cancer mortality risk. Our results can support the updating \nof referral and management guidelines for symptomatic patients presenting in primary care. \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n20 \n \nEthics statements \nEthical approval \nThis study was approved by the UK Medicines and Healthcare products Regulatory Agency \nIndependent Scientific Advisory Committee (ISAC Protocol number 18_299), under Section \n251 (NHS Social Care Act 2006). This study is based on data from the Clinical Practice \nResearch Datalink obtained under license from the UK Medicines and Healthcare products \nRegulatory Agency. The data is provided by patients and collected by the UK National \nHealth Service (NHS) as part of their care and support.  \n \nData availability statement \nPotential concerns around patient confidentiality prevent open sharing of the underlying data \nfor this study. CPRD Gold data can be obtained from CPRD, subject to protocol approval via \nCPRD’s Research Data Governance Process. Further details can be found at \nhttps://cprd.com/data-access. Data extraction and analysis code are available at \nhttps://github.com/MattEBarclay/cprd_symptom_cancer_1. \n \nAcknowledgements \nThe work was supported by the International Alliance for Cancer Early Detection, a \npartnership between Cancer Research UK (C18081/A31373), Canary Center at Stanford \nUniversity, the University of Cambridge, OHSU Knight Cancer Institute, University College \nLondon, and the University of Manchester. SI is additionally supported by Cancer Research \nUK (EDDPMA-May22\\100062) and HH and MB by CRUK International Alliance for Cancer \nEarly Detection (ACED) Pathway Awards (EDDAPA-2022/100001 and EDDAPA-\n2022/100002, respectively). GL was supported by a Cancer Research UK (C18081/A18180) \nAdvanced Clinician Scientist Fellowship. CR acknowledges funding from Cancer Research \nUK Early Detection and Diagnosis Committee (grant number EDDCPJT\\100018). JUS is \nsupported by a National Institute of Health Research Advanced Fellowship (NIHR300861). \nACA is support by Cancer Research UK grant: PPRPGM-Nov20\\100002.   SI, AW and ACA \nare supported by the National Institute for Health and Care Research (NIHR) Cambridge \nBiomedical Research Centre (BRC-1215-20014; NIHR203312) [*]. AW is part of the \nBigData@Heart Consortium, funded by the Innovative Medicines Initiative-2 Joint \nUndertaking under grant agreement No 116074.  \n \nThe funders had no role in study design, data collection and analysis, decision to publish, or \npreparation of the manuscript. All authors had access to statistical reports, tables, and \nanalysis code. MB, CR, BW, SI and GL had full access to all of the data. \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n21 \n \nTransparency declaration \nThe lead author affirms that this manuscript is an honest, accurate, and transparent account \nof the study being reported; that no important aspects of the study have been omitted; and \nthat any discrepancies from the study as planned have been explained. \n \nCompeting interests \nAll authors have completed the ICMJE uniform disclosure form at \nhttp://www.icmje.org/disclosure-of-interest/ and declare: no support from any organisation for \nthe submitted work; MB has received personal fees from Grail Inc for membership of an \nIndependent Data Monitoring Committee; no other relationships or activities that could \nappear to have influenced the submitted work. \n \nContributors \nMB designed the statistical analysis, wrote analytical code, cleaned and analysed the data, \nand drafted and revised the paper. He is the guarantor. CR, HH and GL contributed to \ndrafting the paper. CR, JU-S, NP and GL provided clinical interpretation. HH, AT, BW, SI \nand SD contributed to data management and phenotyping. JL, AW and ACA contributed to \nthe design and interpretation of the analysis. All authors provided revisions to the paper and \ngave final approval to the submitted manuscript. \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n22 \n \nTable 1. Cohort summary.  \n   \nCohort  \n   \nCancers within 12 \nmonths  \n   \nDeaths within 12 \nmonths, no preceding \ncancer diagnosis  \n   \nDeaths within 12 \nmonths, following a \ncancer diagnosis  \n   \n   N  (col %)  N  (row %)  N  (row %)  N  (row %)  \nTotal  1,622,419    36,802  (2.3%)  28,867  (1.8%)  9,288  (0.6%)  \nAge at index (grouped)                   \n  30 to 39  395,313  (24.4%)  1,571  (0.4%)  426  (0.1%)  62  (0.0%)  \n  40 to 49  350,133  (21.6%)  3,063  (0.9%)  792  (0.2%)  235  (0.1%)  \n  50 to 59  295,316  (18.2%)  5,080  (1.7%)  1,343  (0.5%)  762  (0.3%)  \n  60 to 69  259,039  (16.0%)  9,014  (3.5%)  2,829  (1.1%)  1,970  (0.8%)  \n  70 to 79  185,854  (11.5%)  10,142  (5.5%)  6,007  (3.2%)  2,960  (1.6%)  \n  80 to 89  111,933  (6.9%)  6,818  (6.1%)  11,453  (10.2%)  2,720  (2.4%)  \n  90 to 99  24,731  (1.5%)  1,114  (4.5%)  6,017  (24.3%)  579  (2.3%)  \nSex                   \n  Women  880,888  (54.3%)  19,808  (2.2%)  15,671  (1.8%)  4,259  (0.5%)  \n  Men  741,531  (45.7%)  16,994  (2.3%)  13,196  (1.8%)  5,029  (0.7%)  \nIMD group                   \n  Least deprived  377,575  (23.3%)  8,934  (2.4%)  5,661  (1.5%)  2,001  (0.5%)  \n  2  356,859  (22.0%)  8,347  (2.3%)  6,177  (1.7%)  2,031  (0.6%)  \n  3  342,184  (21.1%)  7,755  (2.3%)  6,355  (1.9%)  1,889  (0.6%)  \n  4  294,638  (18.2%)  6,483  (2.2%)  5,559  (1.9%)  1,805  (0.6%)  \n  Most deprived  251,163  (15.5%)  5,283  (2.1%)  5,115  (2.0%)  1,562  (0.6%)  \nAny record of smoking                   \n  Never smoker  586,639  (36.2%)  10,390  (1.8%)  10,043  (1.7%)  2,259  (0.4%)  \n  Ever smoker  1,035,780  (63.8%)  26,412  (2.5%)  18,824  (1.8%)  7,029  (0.7%)  \nIndex symptom                   \n  Reference group 786,424  (48.5%)  7,536  (1.0%)  12,520  (1.6%)  2,034  (0.3%)  \n  Abdominal pain  233,933  (14.4%)  5,605  (2.4%)  2,163  (0.9%)  1,640  (0.7%)  \n  Abdominal bloating  22,629  (1.4%)  628  (2.8%)  261  (1.2%)  169  (0.7%)  \n  Rectal bleeding  48,515  (3.0%)  1,868  (3.9%)  860  (1.8%)  220  (0.5%)  \n  Change in bowel habit  17,212  (1.1%)  1,067  (6.2%)  163  (0.9%)  197  (1.1%)  \n  Dyspepsia  108,488  (6.7%)  2,120  (2.0%)  959  (0.9%)  609  (0.6%)  \n  Dysphagia  14,992  (0.9%)  1,036  (6.9%)  1,167  (7.8%)  451  (3.0%)  \n  Jaundice  1,817  (0.1%)  456  (25.1%)  217  (11.9%)  280  (15.4%)  \n  Dyspnoea  141,094  (8.7%)  3,945  (2.8%)  6,268  (4.4%)  1,490  (1.1%)  \n  Haemoptysis  5,859  (0.4%)  412  (7.0%)  146  (2.5%)  183  (3.1%)  \n  Haematuria  25,753  (1.6%)  2,770  (10.8%)  591  (2.3%)  378  (1.5%)  \n  Fatigue  141,932  (8.7%)  2,405  (1.7%)  2,212  (1.6%)  739  (0.5%)  \n  Night sweats  7,675  (0.5%)  133  (1.7%)  30  (0.4%)  35  (0.5%)  \n  Weight loss  19,617  (1.2%)  1,238  (6.3%)  1,173  (6.0%)  623  (3.2%)  \n  Breast lump  38,307  (2.4%)  4,789  (12.5%)  88  (0.2%)  185  (0.5%)  \n  Post-menopausal \nbleed  8,172  (0.5%)  794  (9.7%)  49  (0.6%)  55  (0.7%)  \n \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n23 \n \nTable 2. Modelled age at which the 3% referral threshold is crossed for any cancer and for each cancer site. \n \nCohort Symptom Any Breast Gynae. Lung Upper GI Lower GI Urological Prostate Haem. Other \nMale non-smokers Reference group 90 n/a n/a        \n Abdominal pain 60 n/a n/a        \n Abdominal bloating 65 n/a n/a        \n Rectal bleeding 60 n/a n/a   65     \n Change in bowel habit 55 n/a n/a   60     \n Dyspepsia 65 n/a n/a        \n Dysphagia 55 n/a n/a  60      \n Jaundice 45 n/a n/a  50     55 \n Dyspnoea 70 n/a n/a        \n Haemoptysis 60 n/a n/a 70       \n Haematuria 55 n/a n/a    55 65   \n Fatigue 65 n/a n/a        \n Night sweats 65 n/a n/a        \n Weight loss 60 n/a n/a     80   \n Breast lump 65 n/a n/a       75 \nMale smokers Reference group 75 n/a n/a        \n Abdominal pain 60 n/a n/a        \n Abdominal bloating 60 n/a n/a        \n Rectal bleeding 60 n/a n/a   60     \n Change in bowel habit 55 n/a n/a   60     \n Dyspepsia 65 n/a n/a        \n Dysphagia 55 n/a n/a  55      \n Jaundice 45 n/a n/a  50     55 \n Dyspnoea 65 n/a n/a        \n Haemoptysis 55 n/a n/a 55       \n Haematuria 50 n/a n/a    55 70   \n Fatigue 65 n/a n/a        \n Night sweats 60 n/a n/a        \n Weight loss 55 n/a n/a 70 75      \n Breast lump 60 n/a n/a       70 \nFemale non-smokers Reference group (n/a)       n/a   \n Abdominal pain 65       n/a   \n Abdominal bloating 65       n/a   \n Rectal bleeding 60     70  n/a   \n Change in bowel habit 60     70  n/a   \n Dyspepsia 75       n/a   \n Dysphagia 65    70   n/a   \n Jaundice 45    50   n/a  60 \n Dyspnoea (n/a)       n/a   \n Haemoptysis 65       n/a   \n Haematuria 60      65 n/a   \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n24 \n \nCohort Symptom Any Breast Gynae. Lung Upper GI Lower GI Urological Prostate Haem. Other \n Fatigue 75       n/a   \n Night sweats 75       n/a   \n Weight loss 65       n/a   \n Breast lump 35 40      n/a   \n Post-menopausal bleeding 30  30     n/a   \nFemale smokers Reference group (n/a)       n/a   \n Abdominal pain 65       n/a   \n Abdominal bloating 65       n/a   \n Rectal bleeding 60     70  n/a   \n Change in bowel habit 60     70  n/a   \n Dyspepsia 70       n/a   \n Dysphagia 60    70   n/a   \n Jaundice 40    45   n/a  55 \n Dyspnoea 70       n/a   \n Haemoptysis 55   60    n/a   \n Haematuria 55      60 n/a   \n Fatigue 70       n/a   \n Night sweats 70       n/a   \n Weight loss 60       n/a   \n Breast lump 35 35      n/a   \n Post-menopausal bleeding 30  30     n/a   \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n25 \n \nFigure 1. Modelled cancer and mortality risk at 12 months by index symptom, male non-smokers. \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n26 \n \nFigure 2. Modelled cancer and mortality risk at 12 months by index symptom, male smokers. \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n27 \n \nFigure 3. Modelled cancer and mortality risk at 12 months by index symptom, female non-smokers. \n \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n28 \n \nFigure 4. Modelled cancer and mortality risk at 12 months by index symptom, female smokers. \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n29 \n \nTable 3. Summary of cancer outcomes for patients with multiple different recorded symptoms at index \npresentation, and within 30 days of index symptom. \nIndex symptom Any other symptoms \nat index Patients Cancers within 12 months of index \nN % (95% CI) \nAny No 825,635 28,834 3.5% (3.5%, 3.5%) \n Yes 10,360 480 4.6% (4.2%, 5.1%) \n Within 30 days* 21,167 1429 6.8% (6.4%, 7.1%) \nAbdominal pain No 231,598 5,510 2.4% (2.3%, 2.4%) \n Yes 2,335 101 4.3% (3.6%, 5.2%) \n Within 30 days* 6,122 379 6.2% (5.6%, 6.8%) \nAbdominal bloating No 825,635 28,834 3.5% (3.5%, 3.5%) \n Yes 10,360 480 4.6% (4.2%, 5.1%) \n Within 30 days* 21,167 1429 6.8% (6.4%, 7.1%) \nRectal bleeding No 47,774 1,831 3.8% (3.7%, 4.0%) \n Yes 741 38 5.1% (3.8%, 7.0%) \n Within 30 days* 1,116 61 5.5% (4.3%, 7.0%) \nChange in bowel habit No 16,857 1,042 6.2% (5.8%, 6.6%) \n Yes 355 25 7.0% (4.8%, 10.2%) \n Within 30 days* 520 77 14.8% (12.0%, 18.1%) \nDyspepsia No 106,843 2,090 2.0% (1.9%, 2.0%) \n Yes 1,645 35 2.1% (1.5%, 2.9%) \n Within 30 days* 3,282 219 6.7% (5.9%, 7.6%) \nDysphagia No 14,760 1,021 6.9% (6.5%, 7.3%) \n Yes 232 17 7.3% (4.6%, 11.4%) \n Within 30 days* 1,054 56 5.3% (4.1%, 6.8%) \nJaundice No 1,759 450 25.6% (23.6%, 27.7%) \n Yes 58 9 15.5% (8.4%, 26.9%) \n Within 30 days* 81 17 21.0% (13.5%, 31.1%) \nDyspnoea No 139,758 3,899 2.8% (2.7%, 2.9%) \n Yes 1,336 61 4.6% (3.6%, 5.8%) \n Within 30 days* 2,655 173 6.5% (5.6%, 7.5%) \nHaemoptysis No 5,750 406 7.1% (6.4%, 7.8%) \n Yes 109 6 5.5% (2.5%, 11.5%) \n Within 30 days* 198 20 10.1% (6.6%, 15.1%) \nHaematuria No 25,438 2,749 10.8% (10.4%, 11.2%) \n Yes 315 22 7.0% (4.7%, 10.3%) \n Within 30 days* 636 76 12.0% (9.7%, 14.7%) \nFatigue No 140,212 2,353 1.7% (1.6%, 1.7%) \n Yes 1,720 58 3.4% (2.6%, 4.3%) \n Within 30 days* 3,132 157 5.0% (4.3%, 5.8%) \nNight sweats No 7,527 128 1.7% (1.4%, 2.0%) \n Yes 148 5 3.4% (1.5%, 7.7%) \n Within 30 days* 162 6 3.7% (1.7%, 7.8%) \nWeight loss No 19,168 1,193 6.2% (5.9%, 6.6%) \n Yes 449 52 11.6% (8.9%, 14.9%) \n Within 30 days* 725 90 12.4% (10.2%, 15.0%) \nBreast lump No 38,045 4,765 12.5% (12.2%, 12.9%) \n Yes 262 25 9.5% (6.5%, 13.7%) \n Within 30 days* 345 18 5.2% (3.3%, 8.1%) \nPost-menopausal bleed No 8,092 784 9.7% (9.1%, 10.4%) \n Yes 80 10 12.5% (6.9%, 21.5%) \n Within 30 days* 145 14 9.7% (5.8%, 15.6%) \n*subset of patients with no other symptoms at index \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n30 \n \nReferences \n1  Zakkak N, Barclay ME, Swann R, et al. The presenting symptom signatures of incident cancer: \nevidence from the English 2018 National Cancer Diagnosis Audit. Br J Cancer. 2023;1–11. \n2  Jones R, Latinovic R, Charlton J, et al. Alarm symptoms in early diagnosis of cancer in primary care: \ncohort study using General Practice Research Database. BMJ. 2007;334:1040–4. \n3  Neal RD, Tharmanathan P, France B, et al. Is increased time to diagnosis and treatment in \nsymptomatic cancer associated with poorer outcomes? Systematic review. British Journal of Cancer. \n2015;112:S92–107. \n4  Lyratzopoulos G, Neal RD, Barbiere JM, et al. Variation in number of general practitioner consultations \nbefore hospital referral for cancer: findings from the 2010 National Cancer Patient Experience Survey in \nEngland. The Lancet Oncology. 2012;13:353–65. \n5  Mendonca SC, Abel GA, Saunders CL, et al. Pre-referral general practitioner consultations and \nsubsequent experience of cancer care: evidence from the English Cancer Patient Experience Survey. \nEuropean journal of cancer care. 2015;25:478–90. \n6  Swann R, McPhail S, Witt J, et al. Diagnosing cancer in primary care: results from the National Cancer \nDiagnosis Audit. British Journal of General Practice. 2018;68:e63–72. \n7  National Institute for Health and Care Excellence. Suspected cancer: recognition and referral. NICE \nguidelines. 2015;NG12. \n8  Hamilton W, Hajioff S, Graham J, et al. Suspected cancer (part 2--adults): reference tables from \nupdated NICE guidance. Bmj. 2015;350:h3044. \n9  Hippisley-Cox J, Coupland C. Symptoms and risk factors to identify men with suspected cancer in \nprimary care: derivation and validation of an algorithm. British Journal of General Practice. 2013;63:e1–\n10. \n10  Hippisley-Cox J, Coupland C. Symptoms and risk factors to identify women with suspected cancer in \nprimary care: derivation and validation of an algorithm. The British Journal of General Practice. \n2013;63:e11. \n11  Herrett E, Gallagher AM, Bhaskaran K, et al. Data Resource Profile: Clinical Practice Research \nDatalink (CPRD). International Journal of Epidemiology. 2015;44. \n12  Henson KKE, Elliss-Brookes L, Coupland VH, et al. Data Resource Profile: National Cancer \nRegistration Dataset in England. International Journal of Epidemiology. 2020;49:16–16h. \n13  Harshfield A, Abel GA, Barclay S, et al. Do GPs accurately record date of death? A UK observational \nanalysis. BMJ Supportive & Palliative Care. 2020;10:e24–e24. \n14  NHS England. NHS Data Model and Dictionary. Two Week Wait Cancer or Symptomatic Breast \nReferral Type. 2023. \nhttps://www.datadictionary.nhs.uk/data_elements/two_week_wait_cancer_or_symptomatic_breast_refe\nrral_type.html (accessed 5 October 2023) \n15  NHS England,. Routes to Diagnosis, 2018. 2022. https://digital.nhs.uk/data-and-\ninformation/publications/statistical/routes-to-diagnosis/2018 (accessed 21 November 2023) \n16  Moore SF, Price SJ, Chowienczyk S, et al. The impact of changing risk thresholds on the number of \npeople in England eligible for urgent investigation for possible cancer: an observational cross-sectional \nstudy. British Journal of Cancer. 2021;125:1593–7. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n31 \n \n17  Royston P, Parmar MKB. Flexible parametric proportional-hazards and proportional-odds models for \ncensored survival data, with application to prognostic modelling and estimation of treatment effects. \nStatistics in Medicine. 2002;21:2175–97. \n18  White B, Rafiq M, Gonzalez-Izquierdo A, et al. Risk of cancer following primary care presentation with \nfatigue: a population-based cohort study of a quarter of a million patients. British Journal of Cancer. \n2022;126:1627–36. \n19  Prentice R, Kalbfleisch J, Peterson Jr A, et al. The Analysis of Failure Times in the Presence of \nCompeting Risks. Biometrics. 1978;34:541–54. \n20  Aalen O, Johansen S. An Empirical Transition Matrix for Non-Homogeneous Markov Chains Based on \nCensored Observations. Scandinavian Journal of Statistics. 1978;5:141–50. \n21  Wolkewitz M, Cooper BS, Bonten MJM, et al. Interpreting and comparing risks in the presence of \ncompeting events. BMJ. 2014;349:g5060–g5060. \n22  Crowther MJ. merlin—A unified modeling framework for data analysis and methods development in \nStata. The Stata Journal: Promoting communications on statistics and Stata. 2020;20:763–84. \n23  Crowther MJ, Lambert PC. Parametric multistate survival models: Flexible modelling allowing transition-\nspecific distributions with application to estimating clinically useful measures of effect differences. \nStatistics in Medicine. 2017;36:n/a-n/a. \n24  Arhi CS, Bottle A, Burns EM, et al. Comparison of cancer diagnosis recording between the Clinical \nPractice Research Datalink, Cancer Registry and Hospital Episodes Statistics. Cancer Epidemiology. \n2018;57:148–57. \n25  Keogh RH, Seaman SR, Barrett JK, et al. Dynamic Prediction of Survival in Cystic Fibrosis: A \nLandmarking Analysis Using UK Patient Registry Data. Epidemiology. 2019;30:29–37. \n26  Kostopoulou O, Tracey C, Delaney BC. Can decision support combat incompleteness and bias in \nroutine primary care data? Journal of the American Medical Informatics Association. 2021;28:1461–7. \n27  Price SJ, Stapley SA, Shephard E, et al. Is omission of free text records a possible source of data loss \nand bias in Clinical Practice Research Datalink studies? A case-control study. BMJ open. 2016;6. doi: \n10.1136/BMJOPEN-2016-011664 \n28  Skillrud D, Offord K, Miller R. Higher Risk of Lung Cancer in Chronic Obstructive Pulmonary Disease. \nAnnals of Internal Medicine. 1986;105:503. \n29  Barclay M, Renzi C, Antoniou A, et al. Phenotypes and rates of cancer-relevant symptoms and tests in \nthe year before cancer diagnosis in UK Biobank and CPRD Gold. PLOS Digital Health. \n2023;2:e0000383. \n30  Majano SB, Lyratzopoulos G, Rachet B, et al. Do presenting symptoms, use of pre-diagnostic \nendoscopy and risk of emergency cancer diagnosis vary by comorbidity burden and type in patients \nwith colorectal cancer? British Journal of Cancer 2021 126:4. 2021;126:652–63. \n31  White B, Renzi C, Barclay M, et al. Underlying cancer risk among patients with fatigue and other vague \nsymptoms: a population-based cohort study in primary care. British Journal of General Practice. \n2023;73:e75–87. \n32  Herbert A, Rafiq M, Pham TM, et al. Predictive values for different cancers and inflammatory bowel \ndisease of 6 common abdominal symptoms among more than 1.9 million primary care patients in the \nUK: A cohort study. PLOS Medicine. 2021;18:e1003708. \n33  Price SJ, Gibson N, Hamilton WT, et al. Intra-abdominal cancer risk with abdominal pain: a prospective \ncohort primary care study. British Journal of General Practice. 2022;72:e361–8. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint \n\n32 \n \n34  Williams TGS, Cubiella J, Griffin SJ, et al. Risk prediction models for colorectal cancer in people with \nsymptoms: a systematic review. BMC Gastroenterology. 2016;16:63. \n35  Harrison H, Usher-Smith JA, Li L, et al. Risk prediction models for symptomatic patients with bladder \nand kidney cancer: a systematic review. British Journal of General Practice. 2022;72:e11–8. \n36  Funston G, Hardy V, Abel G, et al. Identifying Ovarian Cancer in Symptomatic Women: A Systematic \nReview of Clinical Tools. Cancers. 2020;12:3686. \n37  Fine JP, Gray RJ. A Proportional Hazards Model for the Subdistribution of a Competing Risk. Journal of \nthe American Statistical Association. 1999;94:496–509. \n38  Putter H, Fiocco M, Geskus RB. Tutorial in biostatistics: competing risks and multi-state models. \nStatistics in Medicine. 2007;26:2389–430. \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted May 21, 2024. ; https://doi.org/10.1101/2024.05.21.24307662doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}