Statistical Modelling & Data Science

Virtual cohorts are digital, non-identical, yet highly similar, synthetic data records that preserve the statistical properties of the original data. They can be used for simulation or to increase dataset sizes for example. Thus, virtual cohorts overcome many hurdles that may stem from low-powered samples, poor data sharing practices, or constraining administrative practices. Further, when attempting to integrate data, differences in existing datasets also impose challenges that limit opportunities for data integration. As a result, the pace of scientific advancements is suboptimal. Synthetic data and virtual cohorts generated using innovative computational techniques represent an opportunity to overcome some of these limitations and consequently, to advance scientific developments. In this paper, Muniz-Terrera et al demonstrate the use of virtual cohorts techniques, using the EPAD V1500 dataset, to generate a synthetic dataset that mirrors the deeply phenotyped sample of preclinical dementia research participants.

This paper introduces a two-stage approach for the modelling of longitudinal cognitive and clinical outcomes, biomarkers (baseline and longitudinal) and risk factors to analyse the data from the EPAD Longitudinal Cohort Study and shows its clinical and biological utility in the areas of trajectory stratification, subgroup identification and prediction. In the long term this approach may be applicable to precision medicine and secondary prevention in Alzheimer’s dementia research and practice.

This paper describes a method of machine learning which aims to provide modelling for predicting Alzheimer’s disease dementia. The model is based on SHARE and PREVENT datasets including 84,856 individuals, and achieved an accuracy of 87% with 99% specificity and 76% sensitivity. The model was capable of predicting dementia onset in a relatively younger population up to 14 years in advance.

This study sought to advance the understanding of conversion of Mild Cognitive Impairment to Alzheimer’s disease using big data modelling. Extracting data from the Alzheimer’s Disease

Neuroimaging Initiative (ADNI) six models were created with good to excellent predictive properties. Suggestions for furthering and enhancing the modelling are outlaid with the inclusion of additional biomarker data.

Around two thirds of people living with dementia live in low and middle income countries (LMIC). With no drugs yet that can cure dementia we know that we need to look at preventing or slowing down the diseases that cause dementia. Some risk factors such as our genetic make-up cannot be changed, but other risk factors such as diet, weight and activity levels can be changed and if those at risk made these lifestyle changes, it could delay the onset of disease. If we could delay the onset of the dementia by five years then half as many people at one time would be living with dementia. We have good data for this in high-income countries. However, Danso and colleagues argue that we need to estimate how many people in LMIC countries have these risk factors and develop a way of monitoring these risk factors so that people in LMIC can implement dementia prevention strategies too.

People living in LMIC do not have access to as many expensive resources as high-income countries, such as brain scanners. So it may be more difficult to track disease progress. Although, there may be other ways to do this. More and more people in LMIC do have access to mobile phones, so using the information from these, such as GPS monitoring, you may be able to tell if someone is getting lost more often, or language analysis of text messages and speech could tell us if their language style is changing. Using this data, and crunching it along with data generated across the world, we could be able to see patterns emerging and use this in LMIC to get a better picture of dementia risk and possible prevention interventions.