Examining methodology to identify patterns of consulting in primary care for different groups of patients before a diagnosis of cancer: a matched retrospective cohort study

Date of ISAC Approval: 
23/06/2020
Lay Summary: 
It is difficult for family doctors to spot which patients need testing for undiagnosed cancer, partly because its symptoms can be vague. For patients with long-term medical conditions this is even harder, particularly if their conditions share symptoms with cancer. For example, people with chronic chest disease will often be short of breath and have a cough, which are also symptoms of lung cancer. We will be developing new methods to find out if cancer diagnosis is significantly delayed in people with multiple health conditions. New methods are needed, because existing ones use the patient's first cancer symptom recorded in the medical notes. It is impossible to know if that symptom was from the cancer or the existing health condition We will study 3,000 people (cases) with cancer of the stomach or food gullet, and 3,000 people who do not have cancer (controls). Cases and controls will be paired on age, sex, their general practice, and their health burden. We will then divide our 6,000 patients into groups by their health burden. We will study three groups of different health burden: (1) low (no or only a few long-term conditions); (2) medium; and (3) high (many long-term health conditions). We will identify "normal" patterns for visiting the doctor in the two years before the case was diagnosed with cancer. From this, we will see when people start to visit their doctor more often than normal because of the undiagnosed cancer. We will develop a method for comparing this "pick-up" point in visits between the three groups, to test if it varies for patients with different health burdens.
Technical Summary: 
This matched, retrospective cohort study will be set in primary care, using observational data from the Clinical Practice Research Datalink (CPRD). We will study patients with CPRD diagnostic codes for oesophageal or stomach cancer recorded between 1 January 2010 and 31 December 2019. To quantify baseline consultation rates, we will include controls (1:1) to cases on age, sex, general practice and multimorbidity burden. Multimorbidity is defined as two or more coexisting chronic conditions, and multimorbidity burden as the overall impact of different diseases in an individual, taking their severity into account. We will develop methodology, based on time series and negative binomial/Poisson regression, to explore how the timeliness of cancer diagnosis varies between groups of patients with different multimorbidity burden. Previous work using such techniques has concentrated on the point at which the difference between consultation rates in cases and controls becomes statistically significant. This potentially introduces a bias, whose magnitude depends on sample size. Here we will build on preliminary work being done in partnership between Gary Abel and the UCL CanTest team to estimate the most likely inflection point and the uncertainty on that. Outcomes to be measured include: 1. Rates of face-to-face consultations between general practitioners and cases or matched controls in the 2 years before the case is diagnosed with cancer. 2. The time at which population-level consultation rate starts to increase before a diagnosis of cancer, across patient groups with different multimorbidity burden. Simulation suggests that 3,000 cases and 3,000 matched controls will give us 92% power to detect a 9% difference in monthly consultation rate between groups with differing multimorbidity burden.
Health Outcomes to be Measured: 
1. Rates of face-to-face consultations between general practitioners and cases or matched controls in the 2 years before the case is diagnosed with cancer. 2. The time at which population-level consultation rate starts to increase before a diagnosis of cancer, across patient groups with different multimorbidity burden. 3. In order to define (1) and (2) above, the date of cancer diagnosis is required.
Collaborators: 

Sarah Price - Chief Investigator - University of Exeter
Dr Bianca Wiering - Collaborator - University of Exeter
Dr Gary Abel - Collaborator - University of Exeter
Sarah Price - Corresponding Applicant - University of Exeter
Professor Willie Hamilton - Collaborator - University of Exeter