Speaker: Juan Carlos Bazo Alvarez
Handling missing data in interrupted time series analysis.
In the interrupted time series (ITS) approach, it is common to average the outcome of interest at each time point and then perform a segmented regression (SR) analysis. The objective was to illustrate that such ‘aggregate-level’ analysis bias the estimates when data are missing at random, demonstrating alternative analysis options. Using electronic primary care records from the United Kingdom, we evaluated weight change over time induced by the initiation of olanzapine treatment. We contrasted ITS estimates from ‘aggregate-level’ SR analysis against estimates from mixed models with and without multiple imputation of missing data, using individual-level data. Finally, we conducted a simulation study to compare the methods in a ‘controlled’ environment. ‘Aggregate-level’ SR analysis suggested a higher average weight gain effect. For example, the short-term weight change was 822g /week from SR, 394g/week from mixed models, and 448g/week from mixed models with multiple imputation. Simulation studies confirmed that the ‘aggregate-level’ SR analysis biased estimates when data are missing at random. Mixed models gave less biased results, but in combination with multiple imputation gave unbiased results. The ‘aggregate-level’ SR analysis biases the ITS estimates when data are missing at random. Omitting the averaging-step and using alternative methods such as mixed models with or without multiple imputation are recommended.
Speaker: Kingshuk Pal
Time trends in incidence and prevalence of type 2 diabetes in the UK in the past decade: a retrospective cohort study
Type 2 diabetes (T2D) is a growing health problem across the world, affecting over 400 million people and estimates that it could affect nearly 700 million people by 2045. However patterns of change in the incidence and prevalence across different countries varies, with the incidence of type 2 diabetes dropping, remaining static or increasing in different populations. The incidence of diabetes in the UK increased significantly between 1996 and 2010, and prevalence of T2D doubled between 2000 and 2010. The cost of treating diabetes has been estimated around 2% of global GDP and in the UK spending on diabetes and related complications accounts for nearly 10 percent of the total NHS budget. Changes in the incidence and prevalence of T2D will have significant implications for healthcare services like the NHS.
However it is unclear whether recent changes in how T2D is diagnosed and the introduction of programmes to prevent diabetes have had any impact on the incidence and prevalence of this condition. Measures of glycosylated haemoglobin (HbA1c) were adopted by WHO as a diagnostic test in 2011, and subsequently it has been widely used in the UK to diagnose T2D. While HbA1c is more convenient than glucose tolerance testing, the relationship between glycaemia and HbA1c in populations previously categorized as impaired glycaemia (prediabetes) is less consistent and the impact of using HbA1c as a diagnostic tool on the incidence and prevalence of T2D is not yet clear.
Another potentially important factor affecting trends in incidence and prevalence is the increasing use of diabetes prevention programmes. Diabetes prevention programmes have been shown to successfully reduce risk factors for diabetes like weight, and a kilogram of mean weight loss has been associated with a 16% reduction in future diabetes incidence. The NHS started roll out of the Diabetes Prevention Programme in 2016 and it is now available nationwide. These interventions are heterogeneous and there is evidence that they have the potential to be cost-effective, however the impact of these interventions on T2D in the UK is currently unknown.
This study uses a previously validated algorithm to look at trend in incidence and prevalence of type 2 diabetes as recorded in UK primary care databases over the last 10 years.