Population movements from cholera-affected areas in Port-au-Prince and identification of communes at potentially increased risk of new outbreaks: Situation as of 23 November

About this report

Authors & contributors

This report was authored by the Flowminder Foundation. This report was authored by Thomas Smallwood, Linus Bengtsson and Veronique Lefebvre, with the contribution of Apphia Yuma, James Harrison, Robert Eyre and Sophie Delaporte.

Thomas Smallwood analysed, interpreted and wrote the report; Linus Bengtsson and Veronique Lefebvre directed the analysis and supported the interpretation of the report; Apphia Yuma directed the project and supported with writing and translation; James Harrison and Robert Eyre produced the aggregates and indicators derived from CDR data and supported the analysis; and Sophie Delaporte supported with writing, translation and data visualisation.

This study was made possible thanks to the anonymised (aggregated) Call Detail Records provided by Digicel Haiti, and cholera case data from the Directorate of Epidemiology Laboratory and Research (DELR).

Data

For billing purposes, mobile operators keep track of subscribers’ activity. These records are generated every time a subscriber makes or receives a call, sends or receives an SMS, or uses mobile data on their phone. They are called Call Detail Records (CDRs). CDRs contain information about the location of the cell tower that routes the call. These data can be analysed in near real time and provides insight into mobility patterns locally and across a country based on the location of those cell towers. For more information on CDR data, please visit our FlowGeek website.

This report is based on the results of an analysis of CDR data provided by Digicel Haiti. CDRs are aggregated and anonymised within Digicel Haiti’s firewall using FlowKit, and then analysed by Flowminder.

In this report, we used CDR data covering the period from 23 October to 22 November 2021. The current situation in Haiti and the impact of the ongoing crisis on the cellular network and, potentially, mobile device usage across the country have affected the quality of the CDRs from September to November 2022. However, we conducted a series of analyses to compare pre-crisis mobility in Haiti, from 01 August to 11 September 2022, to mobility for the same period in 2021. These analyses show mobility is consistent between the two years. Between 11 September and 22 November 2022 there is likely a decrease in mobility, compared to the same period in 2021, but it is difficult to differentiate reductions in mobility from the loss of cells or reductions in mobile device usage. Our analyses may therefore overestimate overall mobility, and likely overestimate the prevalence of long distance travel (>30km) compared to short distance travel (<30km).

Cholera case data were obtained from reports published by the Ministry of Public Health and Population of Haiti (MSPP). The number of new cases of cholera in each commune was calculated by subtracting the number of suspected and confirmed cases published in the report on the 22 November 2022 from the numbers of cases reported in the 15 November report. These reports can be found here. For some communes the number of samples sent to the lab is higher than the suspected cases. The MSPP explains this by the different speed of reporting for the two systems (case and lab reports). For any date, we have therefore chosen the highest of the two numbers (highest of suspected cases column 1 and lab samples column 2 in MSPP SitReps).

Data considerations

The estimates shown are our best current assessment of movements, given the considerations described above regarding having to use last year's data. However there are a number of uncertainties, some of which may be addressed in later analyses. The numbers given should not be interpreted as the truth and should be interpreted with other available evidence, notably derived from field surveys and reports.

CDR data has a number of inherent limitations, particularly the resolution and representativeness of the data.

Geographic variations in network coverage and phone use activity impede spatial and temporal comparisons

The spatial resolution of the data is limited by the density of cell towers. This means that the resolution of the data varies between different areas, particularly urban areas with a very high density of towers and rural areas with a low density. As a result, relatively small changes in location observable in urban areas may not be observed in rural areas, which could be interpreted as lower mobility, meaning areas with different coverage are not directly comparable. Furthermore, no observation can be made outside of the network coverage, though under normal conditions coverage in Haiti is high.

The temporal resolution of CDR data is dependent on the frequency with which subscribers use their mobile devices. Changes in subscribers’ locations will therefore only be observed if the subscribers use their mobile device while in each location. Variation in mobile device usage can therefore also be interpreted as a change in mobility and needs to be adjusted for. How active subscribers are may also vary regionally, as a result mobility may appear larger in a given region only because subscribers are more active there.

Our data in this report reflect the mobility of Digicel subscribers - not the mobility of the population

CDR datasets include a non-random sample of the population of interest. It is therefore important to assess biases in the representativity of the mobility of this sample compared to that of the population as a whole. In order to be included in a CDR dataset, an individual must therefore first own a mobile device and second subscribe to the mobile network operator(s) whose CDR data are being processed. Furthermore, a subscriber must use their mobile device often enough to generate sufficient calls for analysis.

As a result, there are several layers of filters affecting the sample of the population included in the dataset: mobile phone ownership, subscription to a participating mobile network operator, sufficient usage of the mobile device during the study period. For each of these filters, factors such as age, gender and socio-economic status may affect whether an individual is included in the dataset. Representativity of the sample may vary regionally, this means that a larger number of travelling subscribers in a specific region compared to another may not correspond to a larger number of travelling people, if the representativity of the mobility of subscribers varies between the two regions .

Variation in surveillance and reporting may affect the geographic distribution of reported cholera cases

There is also uncertainty in the case data provided by MSPP. No surveillance system is perfect and many infectious individuals are asymptomatic and not picked up by any reporting system. There is variation in the surveillance, diagnosis and reporting of cholera cases between different areas which may result in cases being more likely to be reported in some areas than others.

While these analyses replicate our earlier research, which were predictive of (with a level of uncertainty) where new outbreaks occurred in the 2010 outbreak, responders should also consider that communes across the country differ in access to water and sanitation as well as to other r​isk factors, which will influence the risk of new outbreaks occurring.

We welcome sharing of alternative aggregated case data to be shared with us for comparative analyses.

Not all observed trips are substantial opportunities for the transmission of cholera

Mobility aggregates derived from CDR data provide an estimate of how many subscribers were present in an area having previously been present in another area. However, the aggregates used in this report do not assess the length of time subscribers were in any given area. We therefore do not differentiate between subscribers who remain in an area for a sufficient period of time to facilitate the transmission of cholera and those who pass through an area en route to another destination without stopping. This limitation introduces a further source of uncertainty in our estimation of infectious pressure.

As a result, our results may overestimate the infectious pressure experienced by some communes, especially those along important transport corridors such as major highways. However, communes which are well-connected to areas with substantial cholera cases would be expected to be at higher risk of cholera spreading to the area, and we have observed such a pattern evolve as the outbreak has continued.

We hope to better evaluate the types of trips which are more likely to provide sufficient opportunity for infection in order to address this source of uncertainty, but this remains our best interpretation of the data at this time.

Methodology

Flows of subscribers from communes most affected by cholera and lying within the Port-au-Prince metropolitan area

Flows of subscribers from cholera-affected communes from within the Port-au-Prince metropolitan area to all other communes were calculated from the daily number of travellers between all pairs of communes between 23 October and 22 November 2021. Data from 2021 were used because common and expected travelling patterns between neighbouring communes could not be observed in the 2022 data due to partial or complete loss of network coverage in some regions (see Data section). However, we verified that 2021 and 2022 data were comparable pre-crisis, but noted that the number of subscriber travelling was probably larger in October to November 2021 than in October to November 2022, and that perhaps fewer subscribers travelled distances longer than around 30 km in 2022. A subscriber is determined to have travelled from one commune to another if the subscriber has a network event (here, a phone call) routed by a cell tower in the first commune and an another network event (whether subsequent or not) routed by a cell tower in the second commune within the same day. This means that for a subscriber travelling from A to B to C we count 1 traveller from A to B, 1 traveller from B to C and 1 traveller from A to C.

We summed the number of travellers from cholera-affected communes within the Port-au-Prince metropolitan area to all other communes for each day (24 hours) in the study period. This means that a small proportion of travelling subscribers have been summed more than once, if they have entered the focal commune from more than one other commune within a 24 hour period. We then adjusted the number of travellers each day by dividing by the number of subscribers observed within the cholera-affected communes to compute a probability for a subscriber to travel from a given commune (origin) to another (destination).

Finally, we averaged the adjusted daily travellers across the study period to estimate the mean daily flows of subscribers from the cholera-affected communes within the Port-au-Prince metropolitan area to all other communes across Haiti. We normalised the flows for the report by dividing the adjusted daily trips to each commune from the cholera-affected communes within Port-au-Prince by the largest flow, to give a value between 0 and 1, in which 1 is the largest flow of subscribers. The results of this analysis are provided in Map 1 and Table 1.

We did not need to adjust the daily numbers of travellers to account for variation in the cell network or mobile device usage as the number of subscribers, active cell transceivers, and calls were all constant over the considered period in 2021. Furthermore, as we averaged mobility over the study period we do not need to make adjustments for variation in mobility associated with the day of the week. However, we have not adjusted for variation in mobile device usage between regions, which may impact estimated mobility.

Estimated infectious pressure on communes with no recent confirmed cases

We estimated infectious pressure based on the methodology published by Bengtsson et al. (2015). Greater access to CDR data allowed us to improve our estimation of mobility by using the number of travellers between communes.

First, we estimated the flows of people between all pairs of communes using the daily number of travellers between all pairs of communes between 23 October and 23 November 2021. Data from 2021 were used because common and expected travelling patterns between neighbouring communes could not be observed in the 2022 data due to partial or complete loss of network coverage in some regions (see Data section). However, we verified that 2021 and 2022 data were comparable pre-crisis, but noted that the number of subscriber travelling was probably larger in October to November 2021 than in October to November 2022, and that perhaps fewer subscribers travelled distances longer than around 30 km in 2022.

We adjusted the numbers of travellers between communes each day by dividing by the number of subscribers observed at the commune of origin to compute a probability for a subscriber to travel from a given commune (origin) to another (destination). We then averaged the adjusted number of daily travellers across the study period to estimate the mean daily flows of subscribers between all pairs of communes in Haiti across the study period .

We estimated the number of new cases of cholera in each commune between 14 and 21 October from situation reports published by MSPP. We subtracted the cumulative number of suspected cases in each commune reported by MSPP on 08 November 2022 from the cumulative number of cases reported by MSPP on 14 November to estimate the number of new cases in this 7-day period (24-31 October for Report #1, 02-08 November for Report #2).

To estimate infectious pressure experienced by a commune, we multiplied each flow of subscribers into the commune by the number of new suspected cholera cases at the commune of origin and summed these together to give the total infectious pressure.This means that a small proportion of travelling subscribers has been summed more than once, if they have entered the focal commune from more than one other commune within a 24 hour period.

We normalised the infectious pressures for the report by dividing the infectious pressure experienced by each commune by the highest infectious pressure experienced by a commune with no confirmed cases of cholera. For communes with no confirmed cases, this gives a value between 0 and 1 in which 1 is the highest infectious pressure. The results of this analysis are shown in Map 2 and Table 2.