One in two people in the UK will be diagnosed with cancer during their lifetime. Almost 50% of cancers are diagnosed at a late stage and thus despite advancement in treatment regimens, cancer patients still face significantly higher death rates and impairments in their quality of life. Given that 38% of cancer cases are preventable, there is a clear need to streamline operations relating to cancer screening and care to deliver maximum value to the patients. Early diagnosis and personalised therapy are key if we are to improve the outcomes of cancer patients. However, we are still far from accomplishing this goal because 70% of cancer patients also have one or more pre-existing chronic illnesses that are not being taken into consideration in conventional diagnostic and treatment pathways. Patients with chronic illnesses should have increased exposure with healthcare systems and hence better opportunities for cancer detection. In contrast, pre-existing chronic conditions may also be associated with late cancer diagnosis due to competing demands for healthcare or competing causes of death, resulting in decreased likelihood for cancer screening. As people are living longer, there are greater opportunities for individuals to develop multiple diseases. Hence, studying one disease at a time may not yield results that are representative of real-life experiences at the population level. We propose to investigate the effects of chronic diseases on cancer outcomes. Results from this work will support new cancer screening strategies and clinical decision making to provide greater personalised cancer care.
Modelling cancer trajectories using large-scale electronic health records would help bring us a step closer to achieving personalised care and to improving resource allocation in healthcare systems. Using linked data from primary care, secondary care and the cancer registry, we will systematically analyse longitudinal differences in patients with cancer by deciphering pre- and post-cancer diagnostic routes in patients with multimorbidity. Cancer patients are routinely staged by their tumour burden/spread with limited consideration on other essential patient-centric risk factors such as the pathophysiology of comorbid illnesses. For instance, bacterial and viral infections are common risk factors for some cancers, in part due to chronic inflammatory processes that may initiate carcinogenesis. Since comorbidities in cancer patients may influence diagnosis, prognosis and treatment decision, we will investigate multimorbidity disease clusters to determine the associations between non-malignant comorbidities and cancer. We will analyse disease co-occurrences for diagnostic pairs. We will measure the period prevalence, the cumulative incidence and the median age at diagnosis for all cancer types stratified by sex, ethnicity and deprivation. Disease associations that are identified will fuel hypothesis-driven research for future studies. We will compare conventional statistical methods (Cox regression) and machine learning methods (Random Forest and Gradient Boosting Machine) for disease classification based on disease-specific risk factors. We will evaluate healthcare utilisation in primary and secondary care to ascertain the distribution of healthcare costs across different patient groups. This is particularly important as patients with multimorbidity have frequent encounters with the healthcare system and the analyses can inform solutions on improving the management of high-need patients. Details on how cancer phenotypes are choreographed throughout an individuals lifespan will have immense translational implications as this information can improve cancer diagnoses, prognosis stratification, treatment recommendation/efficacy and resource utilisation.
We will measure first occurrences of phenotypes for cancers recorded in primary care, secondary care and the cancer
registry; comorbidities in patients with cancer; all-cause mortality; all-cause hospitalisation; cancer-related mortality; healthcare utilisation; accident and emergency admissions. Phenotype definitions are available on the CALIBER data portal (https://www.caliberresearch.org/portal/phenotypes/chronological-map).
Harry Hemingway - Chief Investigator - University College London ( UCL )
Alvina Lai - Corresponding Applicant - University College London ( UCL )
Aasiyah Rashan - Collaborator - University College London ( UCL )
Alex Ho - Collaborator - University College London ( UCL )
Arturo Gonzalez-Izquierdo - Collaborator - University College London ( UCL )
Christopher Tomlinson - Collaborator - University College London ( UCL )
Constantinos Parisinos - Collaborator - University College London ( UCL )
Eloise Withnell - Collaborator - University College London ( UCL )
Jurgita Kaubryte - Collaborator - University College London ( UCL )
Michail Katsoulis - Collaborator - Farr Institute of Health Informatics Research
Sheng-Chia Chung - Collaborator - University College London ( UCL )
Stefanie Mueller - Collaborator - University College London ( UCL )
Vaclav Papez - Collaborator - University College London ( UCL )
Wai Chang - Collaborator - University College London ( UCL )
Yen Yi Tan - Collaborator - University College London ( UCL )
Zareen Thorlu Bangura - Collaborator - University College London ( UCL )
Nikolaos Papachristou - Collaborator - University College London ( UCL )
HES Accident and Emergency;HES Admitted Patient Care;HES Outpatient;NCRAS Cancer Registration Data;NCRAS National Radiotherapy Dataset (RTDS) data;NCRAS Systemic Anti-Cancer Treatment (SACT) data;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation