Depression and anxiety are highly prevalent mental health disorders both globally and in the UK. In any given week it has been estimated that 1 in 6 people suffer from them in the UK. Although these disorders have lifelong prevalence there is evidence that they have roots in childhood/young adulthood and that early identification could reduce overall burdens. Research using THIN data at Warwick/Birmingham University Nichols et al., 2018 designed a method for identifying early warning signs and predictors for depression amongst young people aged 15 to 24 years old. The purpose of the present research is to a) replicate this research by performing out of sample prediction on a new sample of THIN data, and b) expand on the previous research by using new state of the art modelling techniques c) consider the potential impact COVID-19 pandemic had on the depression prevalence, and d) expand original efforts by Nichols et al. (2018) to a broader age group. The first goal builds on the recent calls for more researchers to replicate results of machine learning studies (Beam et al., 2020). Despite the recent popularity of machine learning methods in health sciences, very little effort has been devoted to assuring that the findings are reproducible and replicable. This is why our primary objective is to revisit findings reported by Nichols et al. (2018). Our secondary objective is to expand on that study by evaluating additional machine learning techniques to improve performance of the original model. A third objective is to explore the potential consequences of the COVID-19 pandemic on the performance of models predicting depression in young adults people. In other words, we can test the robustness of models trained on the pre-COVID-19 times by fitting them to the data collected after the pandemic started in the UK. We expect that through the investigation of the errors of our model we will be able to gain an invaluable insight into the unique impact a global pandemic might have on increasing the risk of mental health problems. Finally, our goal is also to extend the age range to include children aged 10 and above to assess impact on model prediction. This could lead to a better understanding of how these disorders can be identified early and potentially reduce lifetime prevalence and personal impact.

Objectives

There are four key objectives for this research proposal: 

• Replicate core findings of Nichols et al. (2018) for 15 to 24 year olds including identifying concurrence/divergence.
• Identify if performance can be improved by, for example, using Random Forest or other machine learning methods and method ensembles.
• Identify impact of Covid-19 pandemic on model with regard to depression; how might Covid-19 related variables influence the model?
• Assess effect of extending the age range of the cohort on model performance."