1222Estimating the causal effect of mobility on Dengue during the COVID-19 pandemic. (2nd September 2021)

• Record Type: Journal Article
• Authors: Roster, Kirstin; Connaughton, Colm; Rodrigues, Francisco A
• Is Part Of: International journal of epidemiology. Volume 50(2021)Supplement 1
• Publication Date: 2021-09-02
• Subjects: Epidemiology -- Periodicals; 614.4
• DOI: 10.1093/ije/dyab168.573
• Languages: English
• ISSNs: 0300-5771

Abstract

Abstract: Background: Research suggests that human mobility is associated with the spread of Dengue Fever [1]. The gold-standard in estimating a causal effect are randomized experiments. As these are neither feasible nor ethical in estimating mobility's impact on Dengue, we rely on methods that make use of observational data. During the COVID-19 pandemic, most of the world saw a sudden drop in long- and short-distance travel. In March 2020, inner- and inter-city transit dropped severely and uniformly across cities in Brazil. The drop was random (in time) and is not linked to any of the other factors that affect Dengue, such as the mosquito population size. This gives rise to a quasi-experimental situation to assess the impact of mobility reduction on Dengue Fever in Brazilian cities. The potential impact of the COVID-19 pandemic on Dengue Fever has been recognized globally [2, 3] and first attempts at estimating this impact have been published. Conceição et al. (2021) [4] find an association between social isolation and Dengue in the Brazilian state of São Paulo using a negative binomial regression model. Risk of Dengue infection in São Paulo decreases by 9% twenty days after isolation, defined as mobility reduction of 1-30% relative to baseline mobility prior to the pandemic. Lim et al. (2020) [5] use a Regression Discontinuity Design (RDD) to estimate the causal effect of social distancing policies on Dengue at the state-level in three South-East Asian countries. Results for Thailand indicate that social distancing raises Dengue incidence by 0.431 cases per 10, 000 people. The authors explain this effect by referring to the increased time spent at home, where the risk of infection is higher than in other locations, such as workplaces. No significant effect was found for Singapore and Malaysia. Figure 1 shows the time series of Dengue cases in cities in São Paulo over time. The top panel compares the evolution during the pandemic year 2020 with the average of prior distributions. We note that early 2020 saw greater incidence and an earlier peak than in prior years. This could be due to the stark mobility reduction that occurred in March (epidemiological weeks 11-13). However, the evolution of Dengue cases from 2020 is not unprecedented, as can be seen in the lower panel. Here we compare the time series in individual years and note that 2020 had a similar distribution as the year 2016. The peak in these years occurred earlier than in the years 2015 and 2019. In the following analysis, we aim to better understand the possible causal impact of mobility reduction on Dengue. Methods: We implement propensity score matching to assess the causal effect of mobility on Dengue during the COVID-19 pandemic in São Paulo state in Brazil. We match weeks during the peak pandemic period (March-June 2020) to comparable previous periods based instruments for the mosquito population size and human susceptibility to Dengue. The breeding conditions and thus mosquito population size is approximated using climate factors. We estimate the level of susceptibility within the human population by taking the average number of infections in the same month of the past three years, so as to approximate the duration of partial immunity after Dengue infection. By matching within a given city, we also control for city-level characteristics that may affect Dengue, such as landscape and environment factors, socio-economic situation, or population density. We use weekly data on Dengue cases and climate (rainfall, temperature, humidity) in 37 cities in the state of São Paulo, Brazil, from 2015 until 2020. The peak isolation period (March - June 2020) was determined using monthly traffic volume at toll stations, where isolation is the period in which passenger traffic was consistently below the pre-pandemic (2011-2019) minimum. We also leverage the regional COVID-19 Community Mobility Reports published by Google to assess the change in mobility due to the pandemic. We compute the propensity score using both a logistic regression model and a 100-tree Random Forest model with five-fold cross-validation. In both cases, we observe good overlap in the propensity scores among treated and control groups, suggesting that conditions for matching are met (see figure 2 ). We implement both one-to-one and one-to-many matching with calipers. After matching, we observe similar distributions of the control variables among the treated and matched control periods. Figure 3 illustrates this using the distributions of the one-to-one matching based on propensity scores of the logistic regression models. Results: We compare the Sample Average Treatment Effect on the Treated (SATT) across the four models and find variation in the direction of the causal effect. In 12 cities, mobility reductions are linked to more Dengue cases with results being robust to the propensity score estimation method and matching type. Fewer Dengue cases are reported in 9 cities during the pandemic, regardless of which model is chosen. The remaining cities are sensitive to the model chosen: in 6 cases, three of the four models produced a positive effect, while the majority indicated lower Dengue incidence in 5 cities. The diversity of results may be attributed to differing travel patterns across cities. Long-distance travel hubs like São Paulo (the state capital) and São Carlos (university city with flux of students across the country) demonstrate a negative relationship of a SATT of -18.0 and -18.8 cases per week, respectively. Cities that experience mostly local travel demonstrate the converse effect. This role of different mobility types should be explored further. Additional robustness checks should be performed to understand the role of additional control variables or varying time series length. Conclusions: The SATT of mobility on Dengue varies across the cities in our sample, with more cities experiencing an increase in cases during the COVID-19 pandemic. Key messages: The travel reduction due to the COVID-19 pandemic enables a quasi-experimental analysis of mobility on Dengue. Our results suggest that there is a a causal effect of mobility on Dengue that varies across cities in São Paulo state. Specific characteristics of cities may help explain where mobility leads to Dengue spreading and where home-based infections are the primary disease drivers.