In Public health in practice (Oxford, England)
Objectives : In the past months, many countries have adopted varying degrees of lockdown restrictions to control the spread of the COVID-19 virus. According to the existing literature, some consequences of lockdown restrictions on people's lives are beginning to emerge yet the evolution of such consequences in relation to the time spent in lockdown is understudied. To inform policies involving lockdown restrictions, this study adopted a data-driven Machine Learning approach to uncover the short-term time-related effects of lockdown on people's physical and mental health.
Study design : An online questionnaire was launched on 17 April 2020, distributed through convenience sampling and was self-completed by 2,276 people from 66 different countries.
Methods : Focusing on the UK sample (N = 325), 12 aggregated variables representing the participant's living environment, physical and mental health were used to train a RandomForest model to estimate the week of survey completion.
Results : Using an index of importance, Self-Perceived Loneliness was identified as the most influential variable for estimating the time spent in lockdown. A significant U-shaped curve emerged for loneliness levels, with lower scores reported by participants who took part in the study during the 6th lockdown week (p = 0.009). The same pattern was replicated in the Greek sample (N = 137) for week 4 (p = 0.012) and 6 (p = 0.009) of lockdown.
Conclusions : From the trained Machine Learning model and the subsequent statistical analysis, Self-Perceived Loneliness varied across time in lockdown in the UK and Greek populations, with lower symptoms reported during the 4th and 6th lockdown weeks. This supports the dissociation between social support and loneliness, and suggests that social support strategies could be effective even in times of social isolation.
Carollo Alessandro, Bizzego Andrea, Gabrieli Giulio, Wong Keri Ka-Yee, Raine Adrian, Esposito Gianluca
COVID-19, Global study, Lockdown, Loneliness, Machine learning, Mental health