Chronology of healthcare resource use and comorbidities in people with obstructive sleep apnoea and narcolepsy before and after diagnosis: a descriptive study protocol

We aim to measure healthcare resource use and medical conditions associated with Obstructive Sleep Apnoea (OSA) and narcolepsy over time before and after diagnosis. Additionally, we willvalidate coded records of these conditions in primary care and linked data. Our findings will informpolicies regarding the diagnosis and treatment of sleep disorders and methods for investigatingthese and other under-researched conditions.The primary cohort will include OSA, narcolepsy and comparison groups selected from CPRDprimary care data. The sleep disorder groups will have a first record of the condition between 01/01/1990 and the end of data collection. Comparison groups will be matched by sex, year ofbirth, practice, and registration time. A secondary cohort will be restricted to people eligible forlinkage to linked data sources. Main outcomes and estimates will be:- Positive Predictive Value of coded records for OSA and narcolepsy compared to hospitalconfirmed diagnoses measured using a GP questionnaire. - Annual point prevalence and incidence of OSA and narcolepsy.- Rates, rate ratios (RRs), rate differences (RDs), mean costs and differences in mean cost ofhealthcare resource use measured using primary care data and linked Hospital Episode StatisticsAdmitted Patient Care, Outpatient and Accident & Emergency data.- Rates, RRs and RDs of common or associated medical conditions measured in primary careand linked data. We will estimate RRs using generalised linear Poisson models with a classic log link(comorbidities) or a negative binomial link (resource use) including an interaction term betweenexposure status and time relative to diagnosis (categorical and restricted cubic spline) andadjusting for matching variables and time updating BMI (OSA only). We will predict adjusted rates and calculate RDs. Differences in mean costs will be estimated using generalised linear models.Analyses will be stratified by calendar time, demographics and BMI (OSA only).