Biostatistics: Methods and Applications 1


Biostatistics: Methods and Applications 1 
Chair: Ehsan Karim (McGill University) 

JILL AINSWORTH, McGill University
Estimating Parameters from a Finite Mixture of Generalised Linear Mixed-Effect Models Using the Maximum Likelihood Method  [PDF]
Consider a dataset with the following features that the analysis should take into account: (i) discrete outcomes, (ii) the longitudinal measurements, and (iii) possibility of subpopulations across which covariate effects differ. In this paper, I propose the use of a finite mixture of generalised linear mixed-effect models (FinMix GLMM) as an appropriate analytic approach. The performance of maximum likelihood estimator of the model parameters are explored by a simulation study as well as an analysis of data from the Scottish Early Rheumatoid Arthritis (SERA) cohort. 
Methods for Interrupted Time Series Analysis of Count Data  [PDF]
This study evaluates different methods of analyzing count data in interrupted time series (ITS) designs. Fixed, random and zero-inflated Poisson models were fitted to an empirical data obtained from a previous ITS study evaluating the effect of an implementation tool to improve osteoporosis disease management. Extensive simulations with different scenarios were also performed to investigate performance. Preliminary results show that the fixed effects and zero-inflated Poisson models were best-fitting for estimating the effect of intervention on number of patients given BMD testing and osteoporosis medication, respectively. 
TANIA ALARCON FALCONI, Civil and Environmental Engineering, Tufts University
Characterization of Seasonality in Longitudinal Clinical Trials  [PDF]
The impact of disease prevention interventions can be affected by characteristic temporal patterns in probability and dose of exposures. Lack of proper accounting for such seasonal dynamics can affect observed disease incidence and lead to misinterpretation of intervention results. We characterized seasonality of respiratory and diarrheal infections in a longitudinal clinical trial of supplementation in children in Ecuador using mixed effects models. We accounted for cohort age effects and associated cross-immunity by considering different temporal components and children's age. We also explored temporal effects of climatic and meteorological parameters (temperature, precipitation, UV radiation, photoperiod, cloud cover) on cohort seasonal behavior. 
KHAN MOHAMMAD KAVIUL, Dalla Lana School of Public Health, University of Toronto
Investigation of Change in Methylation in Prostate Cancer Patients after Receiving Treatment  [PDF]
An experimental study was conducted on 33 prostate cancer patients undergoing chemotherapy. The patients were assigned to a treatment. Based on changes in their prostate-specific-antigen (PSA) responder or non-responder to the treatment was declared. Methylation status from blood sample was collected before treatment and weekly for seven weeks after treatment. We divided the whole genome into 1744 chromosome regions. We tested whether change in methylation status over time in a specific chromosome region varies in responders and non-responders using linear mixed effect model. Since there were 1744 such regions adjustment was done for multiple comparisons. 
Interim Data Sharing: What Interim Results/Information Should Be Shared During the Interim of a Randomized Controlled Trial with Those Responsible for the Conduct of the Trial? [PDF]
It is unclear what interim trial information a Data Safety Monitoring Board (DSMB) should share with non-DSMB members during a trial's conduct. To understand the opinions of those involved in clinical trials, an online survey targeted at trialists and statisticians was conducted in 2015. The total response rate was 17.8\%. Preliminary results suggest that the majority of respondents believe that the interim combined event rate should be shared (65\% [95\%CI: 60\% to 71\%]) with various parties responsible for a trial's conduct. Further analyses on the reasons for/against sharing interim information will be presented. Low response rate is a limitation. 
SHAHRIAR SHAMS, University of Toronto
Evaluate the Feasibility of Bayesian Inference as a Means of Reducing Scoring Uncertainty for the EQ-5D Health Utilities  [PDF]
\noindent Health utilities are the quality weights used in estimating quality-adjusted life years, and can be measured using the EQ-5D-3L questionnaire which classifies a respondent into one of 243 health states. The health states are converted to utilities through a prediction equation, developed in a valuation study that directly valued 43 health states. Prediction error is large and is currently ignored. Model mis-specification (δ) explains 84\% of the mean squared prediction error. Incorporating the posterior mean of δ substantially increased the precision of the valued-states. Furthermore, modelling the correlation structure of δs may increase the precision of the 200 unvalued-states.