Comparing lagged impacts of mobility changes and environmental factors on COVID-19 waves in rural and urban India: A Bayesian spatiotemporal modelling study

Abstract

Previous research in India has identified urbanisation, human mobility and population demographics as key variables associated with higher district level COVID-19 incidence. However, the spatiotemporal dynamics of mobility patterns in rural and urban areas in India, in conjunction with other drivers of COVID-19 transmission, have not been fully investigated. We explored travel networks within India during two pandemic waves using aggregated and anonymized weekly human movement datasets obtained from Google, and quantified changes in mobility before and during the pandemic compared with the mean baseline mobility for the 8-week time period at the beginning of 2020. We fit Bayesian spatiotemporal hierarchical models coupled with distributed lag non-linear models (DLNM) within the integrated nested Laplace approximation (INLA) package in R to examine the lag-response associations of drivers of COVID-19 transmission in urban, suburban and rural districts in India during two pandemic waves in 2020-2021. Model results demonstrate that recovery of mobility to 99% that of pre-pandemic levels was associated with an increase in relative risk of COVID-19 transmission during the Delta wave of transmission. This increased mobility, coupled with reduced stringency in public intervention policy and the emergence of the Delta variant, were the main contributors to the high COVID-19 transmission peak in India in April 2021. During both pandemic waves in India, reduction in human mobility, higher stringency of interventions, and climate factors (temperature and precipitation) had 2-week lag-response impacts on the [Formula: see text] of COVID-19 transmission, with variations in drivers of COVID-19 transmission observed across urban, rural and suburban areas. With the increased likelihood of emergent novel infections and disease outbreaks under a changing global climate, providing a framework for understanding the lagged impact of spatiotemporal drivers of infection transmission will be crucial for informing interventions.

Fig 1. COVID-19 cases, reproduction numbers and mobility changes in India during the pandemic.

(A) Number of daily new confirmed COVID-19 cases reported in India from March 15, 2020, to December 25, 2021. (B) Estimated mean and 95% confidence interval (CI) of the basic reproduction number ($R_0$) and instantaneous reproduction rate ($R_t$). (C) Relative weekly mobility of domestic travel by rural, suburban and urban areas in India as measured by the aggregated Google COVID-19 mobility research dataset. Relative mobility levels were standardized by the overall mean level of each type of flow in each region during the first 8 weeks of 2020. The red and grey vertical dashed lines indicate the date of the COVID-19 pandemic being declared by the WHO and the first date of each year, respectively.

Fig 2. Changes in community domestic travel networks of Indian districts across four time periods in 2019-2021.

(A) Communities (n=23) of domestic travel at district level during the pre-pandemic period from November 10, 2019, to February 22, 2020. (B) Communities (n=79) of domestic travel during the first lockdown on March 22 - May 2, 2020. (C) Communities (n=31) of domestic travel during the second lockdown on April 18 - May 29, 2021. (D) Communities (n=22) of domestic travel post-second lockdown period (8 weeks), from November 7 to December 31, 2021, after travel restrictions for COVID-19 had been lifted in India. In each panel, geographically adjacent areas of the same colour represent an internally and closely connected community in terms of human movement in India. The community structure was detected using the Louvain algorithm, based on the aggregated Google COVID-19 mobility research dataset. Circle size represents the relative volume of outbound travellers. The bigger the circle, the higher the level of outflow.

Fig 3. The lagged impact of different factors and scenarios on COVID-19 transmission during the Delta wave in 2021.

(A) The overall association between mobility changes and COVID-19 transmission dynamics under 0- to 3-week lags. The red/blue lines show ratio of Rt under the scenario of mobility below/above the overall mean level (0.99). The histogram with the secondary y-axis shows the frequency of data under different levels. (B) Contour plot of the association between mobility and risk of COVID-19 transmission. The deeper the shade of purple, the greater the increase in transmission risk, while the deeper the shade of green, the greater the decrease in Rt. (C) COVID-19 lag–response association for mobility level at 0.6, 0.8, 1.25, relative to the overall pre-pandemic mean level (1). The mean and 95% CI were presented. (D) – (F) Lag-response association between COVID-19 transmission and temperature (Temp) for cool (10°C), warm (20°C), and hot (30°C) weather, relative to the overall mean of 27.2°C. (G) – (I) COVID-19 lag-response association for precipitation (Prec) at 0.05, 0.5, and 1m, relative to the overall mean of 0.15m. (J) – (L) Lag-response association between COVID-19 transmission and the stringency of intervention policy at low (40), medium (60), and high (80), relative to the overall mean Stringency Index (68.2). Results are for the best fitting model with DLNMs (base model + mobility + temperature + precipitation + intervention policy; see SI Table S2) across the whole country.

Fig 4. The lag-response association between COVID-19 transmission and different factors in urban, suburban, and rural districts.

(A) – (D) COVID-19 lag–response association for different levels of mobility, temperature (Temp), precipitation (Prec) and the stringency of intervention policy in urban areas, relative to the overall mean level. Results are for the best fitting model with DLNMs (base model + mobility + temperature + precipitation + intervention policy) in urban districts. (E) – (G) Lag–response association between the risk of COVID-19 transmission and different levels of mobility, temperature (Temp), and the stringency of intervention policy in semi-urban areas, based on the best fitting model with DLNMs (base model + mobility + temperature + intervention policy; see SI Table S2) in suburban districts. (H) COVID-19 lag–response association for the Stringency Index of intervention policy at low (40), medium (60), and high (80), based on the best fitting model with DLNMs (base model + intervention policy) in rural districts. The mean and 95% CI of RR for each level were presented.

Fig 5. The lagged impact of different factors and scenarios on COVID-19 transmission during the initial wave of COVID-19 transmission in the second half of 2020.

(A) The overall association between mobility changes and COVID-19 transmission dynamics under 0- to 3-week lags. The red/blue lines show RR under the scenario of mobility below/above the overall mean level (0.99). The histogram with the secondary y-axis shows the frequency of data under different levels. (B) Contour plot of the association between mobility and relative risk (RR) of COVID-19 transmission. The deeper the shade of purple, the greater the increase in RR of transmission, while the deeper the shade of green, the greater the decrease in RR. (C) COVID-19 lag–response association for mobility level at 0.8, 1.2, 1.4 relative to the overall pre-pandemic mean level (1). The mean and 95% CI were presented. (D) – (F) Lag-response association between COVID-19 transmission and temperature (Temp) for cool (10°C), warm (20°C), and hot (30°C) weather, relative to the overall mean of 25°C. (G) – (I) COVID-19 lag-response association for precipitation (Prec) at 0.5, 1.5, and 2.5m, relative to the overall mean of 0.24m. (J) – (L) Lag-response association between COVID-19 transmission and the stringency of intervention policy at three different measures of stringency: 70, 75, and 80, relative to the overall mean Stringency Index (76.5). Results are for the best fitting model with DLNMs (base model + mobility + temperature + precipitation + intervention policy; see SI Table S2) across the whole country.