Development and validation of personalised risk prediction models for early detection and diagnosis of primary liver cancer among the English primary care population using the QResearch® database: research protocol and statistical analysis plan

Background and research aim The incidence and mortality of liver cancer have been increasing in recent years in the UK. However, liver cancer is still under-studied. The Early De tection of Hepatocellular Liver Cancer (DeLIVER-QResearch) project aims to address the research gap and generate new knowledge to improve early detection and diagnosis of primary liver cancer from general practice and at the population level. There are three research objectives: (1) to understand the current epidemiology of primary liver cancer in England, (2) to identify and quantify the symptoms and comorbidities associated with liver cancer, and (3) to develop and validate prediction models for early detection of liver cancer suitable for implementation in clinical settings. Methods This population-based study uses the QResearch® database (version 46) and includes patients aged 25-84 years old and without a diagnosis of liver cancer at the cohort entry (study period: 1 January 2008 to 31 December 2020). The team conducted a literature review (with additional clinical input) to inform the inclusion of variables for data extraction from the QResearch database. A wide range of statistical techniques will be used for the three research objectives, including descriptive statistics, multiple imputation for missing data, conditional logistic regression to investigate the association between the clinical features (symptoms and comorbidities) and the outcome, fractional polynomial terms to explore the non-linear relationship between continuous variables and the outcome, and Cox regression for the prediction model. We have a specific focus on the 1-year, 5-year, and 10-year absolute risks of developing liver cancer, as risks at different time points have different clinical implications. The internal-external validation approach will be used, and the discrimination and calibration of the prediction model will be evaluated. Discussion The DeLIVER-QResearch project uses large-scale representative population-based data to address the most relevant research questions for early detection and diagnosis of primary liver cancer in England. This project has great potential to inform the national cancer strategic plan and yield substantial public and societal benefits. Medical Subject Headings (MeSH) Liver Neoplasms; Carcinoma, Hepatocellular; Cholangiocarcinoma; Clinical Decision Rules; Early Detection of Cancer; Early Diagnosis; Symptom Assessment; Comorbidity