On 3 September 2019, Ania Zylbersztejn from UCL Great Ormond Street Institute of Child Health, Jude Eze from the Epidemiology Research Unit of Scotland’s Rural College, and James Griffin from the Department of Statistics, University of Warwick participated in the Royal Statistical Society conference 2019 session on risk factors.
The session was organised by the Medical Statistics section of the Royal Statistical Society.
RSV infections, a leading cause of respiratory morbidity and mortality globally, were the topic of the first talk. Nearly all children are infected at least once before the age of two and currently no vaccine is for use preventing RSV. The aim of Ania Zylbersztejn’s study was to examine RSV risk factors by age in the community rather than within a hospital setting through a secondary analysis of blood from the born in Bradford study. 86 percent of children experienced RSV before their second birthday with older siblings and attending formal childcare identified as risk factors.
Jude Eze’s talk was on understanding the burden and epidemiology of Lyme disease through examining the ecology of the disease to understand the route of infection, focusing on how demographics were impacted by Lyme. Most infections are caused by ticks in the nymph stage of their life cycle and if left untreated, the disease can lead to serious illness, such as Bell’s palsy. Data from reference laboratories and the Scottish index of multiple deprivation were included in a binomial model of infection status with health boards included as a random variable. In 2010 regulations allowing GPs to diagnose Lyme were introduced, leading to less data sent to laboratories. The model showed risk effects of age, season, sex, and socio-economic status. The highest incidence was observed in the highlands’ health board area.
James Griffin examined patient-reported outcomes in clinical trials, things patients experience that are not directly measurable. Lots of pain measures suffer from floor and ceiling effects, making their data not quite continuous and not quite discrete, which is also known as 'spike and slab'. He illustrated how to model such data using two part models, first considering the probability of being in the spike of the distribution and thereafter considering the rest of the distribution. An application to the Roland Morris questionnaire for back pain was shown.
The discussions following the talks focused on modelling choices. Questions regarding the best method to capture the time-dependency of the RSV data were raised. For the tick modelling work, it was mentioned that while deprivation was fitted as a linear trend, the data itself showed more of a quadratic trend. Finally, the difference in assumptions between the two-part model and zero-inflated models was mentioned.