Managing the spread of disease with mobile phone data

Abstract

While human mobility has important benefits for economic growth, it can generate negative externalities. This paper studies the effect of mobility on the spread of disease in a low-incidence setting when people do not internalize their risks to others. Using malaria as a case study and 15 billion mobile phone records across nine million SIM cards, this paper quantifies the relationship between travel and the spread of disease. The estimates indicate that an infected traveler contributes to 1.66 additional cases reported in the health facility at the traveler's destination. This paper develops a simulation-based policy tool that uses mobile phone data to inform strategic targeting of travelers based on their origins and destinations. The simulations suggest that targeting informed by mobile phone data could reduce the caseload by 50 percent more than current strategies that rely only on previous incidence.

Keywords: Big data; Epidemics; Health; Mobile phones; Mobility; Public policy.

Fig. 1

Malaria Incidence, Rainfall, and Population Mobility: Example of Health Post Areas in Richard Toll District. Notes: Panels show the values for each health post catchment area in Richard Toll. The y scale is rescaled for panel (b) to better illustrate the movement patterns in each health post area, given the large outliers in Diama Savoigne and Dabi Tiguette Djoudj. The scale in panel (c) is not rescaled to allow for comparison across health facility areas, but it is bounded at 2,000 (for Diama Savoigne and Dabi Tiguette Djoudj, the value in January goes up to around 4,000).

Fig. 2

Predicted, Predicted without Imported Cases, and Actual Incidence Averaged Across Health Post Areas, by District. Notes: The thin dashed lines represent the monthly predicted malaria incidence averaged across health post areas within a district. This was calculated based on values for the parameters of the model drawn from their distributions. I conducted 500 replications and used the mean monthly incidence value per health post area. Panel a compares the predicted values with the actual malaria incidence, where the solid line is actual incidence averaged across health posts within a district. In panel b, the predicted incidence is compared with a scenario where no cases were imported by travelers, shown in thick dashed lines. Incidence with 0 imported cases was calculated using the same 500 replications for parameter values, but imported cases were set to 0.

Fig. 3

Estimated Impact of Future, Current, and Past Expected Imported Malaria Incidence. Notes: The figure was constructed based on a regression of current malaria incidence on imported incidence of malaria two months later, one month later, currently, last month, and two months ago, controlling for time and location fixed effects and rainfall covariates and clustering errors at the health post area level.

Fig. 4

Effect on Malaria Cases of a Policy Targeting Migrant Workers at the Senegalese Sugar Company. Notes: The figure shows the number of cases of malaria and two types of schistosomiases seen at the health post of the Senegalese Sugar Company. The vertical line marks the timing of when a new policy was implemented by the company that tested every migrant worker for malaria and treated those who tested positive. Data were provided by the CSS.

Fig. 5

Targeting Policies: Cost and Benefit under Different Strategies. Notes: The panels show different strategies for targeting policies. In panels a and b, each symbol represents a health facility area-month. Targeting a health facility area-month means targeting all travelers entering that health facility area in that month. The strategies lay out which health facility area-months are targeted first. In panel c, each symbol represents a district-month. Targeting a district-month means targeting all travelers entering the five low-malaria districts from that district in that month. In panel d, a scenario where travelers from a particular district to a particular health facility area are targeted in a specific month is compared with previous scenarios of targeting particular health facility areas, targeting travelers from particular districts, and the scenario if no cell phone data are available of targeting travelers based on the incidence of districts in the previous year. The cost is calculated based on a variable cost of $2.85 per traveler and a fixed cost of $63,928 split proportionally between health facility areas based on population and number of facilities. The benefit is based on the parameters of the model to calculate the number of primary and secondary cases generated by travelers from each district in each month and summed for all travelers in a given health facility area-month. It is assumed that only 94 percent of those targeted are successfully tested.

Fig. A.1

Annual Malaria Incidence in Senegal in 2013. Notes: Data come from tested and confirmed cases at the health district level compiled by the National Malaria Control Program.

Fig. A.2

Senegal Health Districts and Location of Health Posts in the North Used in the Analysis. Notes: The five very low-malaria health districts used in the analysis are subdivided into health post catchment areas that group health posts and mobile phone towers together.

Fig. A.3

Average Monthly Health Post Catchment Area Malaria Incidence per 1000 Population and Average Monthly Rainfall, January 2013–December 2015, by Health District.

Fig. A.4

People Entering a Health Post Catchment Area as a Percent of the Population in the Area, Avg Across All Health Post Catchment Areas. Notes: Red dash lines represent some religious holidays and pilgrimages.

Fig. A.5

Median Duration of Trips, by Day of the Trip.

Fig. B.1

Testing the Approximation of the Entomological Innoculation Rate in Assumption 4. Notes: For each health post area in each month, the entomological inoculation rate is calculated as the vectorial capacity times transmission to mosquitoes times incidence (cCx). Based on assumption 4, in low-malaria districts it should be possible to approximate EIR with cCx. In the scatter plot, this means the points should be along the 45-degree line, which is shown in red. The graph shows that for the districts analyzed here, the assumption holds. The EIR and cCx were estimated using values from the literature for the various biological malaria parameters. Based on Gething et al. (2011), average transmission from infected humans to mosquitoes, c, is 0.161. The incubation period, τ, is nine days (Killeen et al., 2006) and the bites on humans per mosquito, a, is 0.3 (Ruktanonchai et al., 2016). Average mosquito lifespan of 12.5 is used.