This study will look into different ways of measuring key values needed for effective input on designing public health programs (e.g., disease prevention programs) and public policy measures (e.g., projection of spending on healthcare and drug regulation) from databases such as CPRD. These values -- prevalence (how much disease exists in a population at a time) and incidence (how much new disease occurs in a population) -- are the bedrock of public health and policy. While databases such as CPRD offer a widely available and cost-effective way to make these measurements, there is not yet a universally-accepted approach to doing so. For instance, varying how long to look backwards in time in the data to determine how prevalent disease is, being more specific about what times patients are observable (and thus would have recorded disease), or changing the required amount of observable time to be eligible for study, may yield meaningfully different results. We will investigate whether these decisions and several others meaningfully affect what we would report as prevalence and incidence. In all cases, we will use publicly-available measurements done with more traditional techniques as a reference point to determine when a correct measurement has been made.
This study will investigate how different methodological assumptions of measuring prevalence and incidence may affect the observed prevalence and incidence of chronic disease in the CPRD database. Assumptions include how much time to look back to identify prevalent disease, different ways of calculating a denominator, and different ways of identifying incident cases. The prevalence and incidence observed under each set of assumptions will be compared with other measures of prevalence/incidence for the disease of interest in the UK, in order to determine a reasonable set of assumptions to apply in making these measurements in data such as CPRD. This will be performed for five different diseases, including cystic fibrosis, inflammatory bowel disease, schizophrenia, psoriasis, and chronic obstructive pulmonary disease. These diseases were selected as they have a range of different expected prevalences and incidences in the population, which will enable investigation into how the effects the methodological assumptions depend on the underlying prevalence/incidence. Incidence and prevalence will be estimated as proportions (expressed as percentages). Confidence intervals for proportions will be specified using the Wald method.
Cystic fibrosis; Psoriasis; Inflammatory bowel disease; Chronic obstructive pulmonary disease; Schizophrenia.
Dorothee Bartels - Chief Investigator - Boehringer-Ingelheim Germany
Dorothee Bartels - Corresponding Applicant - Boehringer-Ingelheim Germany
Jeremy Rassen - Collaborator - Aetion, Inc
William Murk - Collaborator - Aetion, Inc