Spatio-temporal dynamics of measles outbreaks in Cameroon

Journal Pre-proof Spatio-temporal dynamics of measles outbreaks in Cameroon Alyssa S. Parpia, M.P.H, Laura A. Skrip, Ph.D, Elaine O. Nsoesie, Ph.D, Moise C. Ngwa, Ph.D, Aristide S. Abah Abah, M.D, Alison P. Galvani, Ph.D, Martial L. NdeffoMbah, Ph.D PII:

S1047-2797(19)30319-9

DOI:

https://doi.org/10.1016/j.annepidem.2019.10.007

Reference:

AEP 8758

To appear in:

Annals of Epidemiology

Received Date: 14 May 2019 Revised Date:

18 October 2019

Accepted Date: 31 October 2019

Please cite this article as: Parpia AS, Skrip LA, Nsoesie EO, Ngwa MC, Abah Abah AS, Galvani AP, Ndeffo-Mbah ML, Spatio-temporal dynamics of measles outbreaks in Cameroon, Annals of Epidemiology (2019), doi: https://doi.org/10.1016/j.annepidem.2019.10.007. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Elsevier Inc. All rights reserved.

Spatio-temporal dynamics of measles outbreaks in Cameroon Alyssa S. Parpia, M.P.H1, Laura A. Skrip, Ph.D2, Elaine O. Nsoesie, Ph.D3, Moise C. Ngwa, Ph.D4, Aristide S. Abah Abah, M.D5, Alison P. Galvani, Ph.D1, Martial L. Ndeffo-Mbah, Ph.D6,7✉ 1

Department of Epidemiology of Microbial Diseases, Yale School of Public Health, New Haven, CT, 06510, USA 2 Institute for Disease Modeling, Bellevue, WA 98005, USA 3 Department of Global Health, Boston University School of Public Health, Boston, MA 02118, USA 4 Department of International Health, Johns Hopkins Bloomberg School of Public Health, MD, 21205 , USA 5 Department of Epidemiological Surveillance, Ministry of Health, Yaoundé, Cameroon 6 Department of Veterinary and Integrative Biosciences, Texas A&M College of Veterinary Medicine and Biomedical Sciences, College Station, TX 77845, USA 7 Department of Epidemiology and Biostatistics, Texas A&M School of Public Health, College Station, TX 77845, USA

Corresponding author: Martial Ndeffo-Mbah, PhD Email: [email protected] Phone: +1 9798455646

✉

Abstract Purpose: In 2012, Cameroon experienced a large measles outbreak of over 14,000 cases. To determine the spatiotemporal dynamics of measles transmission in Cameroon, we analysed weekly case data collected by the Ministry of Health.

Methods: We compared several multivariate time-series models of population movement to characterize the spatial spread of measles in Cameroon. Using the best model, we evaluated the contribution of population mobility to disease transmission at increasing geographic resolutions: region, department, and health district.

Results: Our spatio-temporal analysis showed that the Power Law model, which accounts for long-distance population movement, best represents the spatial spread of measles in Cameroon. Population movement between health districts within departments contributed to 7.6% (range: 0.4%–13.4%) of cases at the district level, while movement between departments within regions contributed to 16.0% (range: 1.3%–23.2%) of cases. Long-distance movement between regions contributed to 16.7% (range: 0.1%–59.0%) of cases at the region level, 20.1% (range: 7.1%–30.0%) at the department level, and 29.7% (range: 15.3%–47.6%) at the health district level.

Conclusions: Population long distance mobility is an important driver of measles dynamics in Cameroon. These findings demonstrate the need to improve our understanding of the roles of population mobility and local heterogeneity of vaccination coverage in the spread and control of measles in Cameroon.

Keywords spatio-temporal analysis, multivariate models, measles outbreaks, Cameroon

Main Text Background: Despite the existence of a safe, inexpensive, and highly efficacious vaccine that provides lifelong immunity, measles is an important infectious disease causing an estimated 110,000 deaths worldwide in 2017 [1]. Sub-saharan Africa remains the most affected region with an estimated mortality of 4.23 deaths per 100,000 population and 355.68 disability adjusted life years per 100,000 population attributable to measles [2]. In 2011, the World Health Organization (WHO) African Region (AFR) established the goal of eliminating measles in Africa by 2020 [3]. It was initially expected that this goal could be achieved by both increasing national and health district level vaccination coverage to more than 95% and reducing measles incidence to less than 1 per 1 million population in all countries [3]. This ambitious goal appears elusive due to sporadic outbreaks across Africa. To optimize the impact of local, regional, and national measles vaccination efforts, it is paramount to improve our understanding of the spatiotemporal dynamics of measles transmission in order to inform the design of optimal targeted interventions in each endemic country. Despite a general decline in measles incidence since the implementation of vaccination campaigns [4,5], Cameroon persistently experiences sporadic measles outbreaks [6–9], the most devastating of which was the 2012 outbreak that rapidly spread across the country to infect over 14,000 people, as reported by the integrated disease surveillance system. Vaccination coverage in Cameroon remains well below the WHO target of 95%, with only 77% coverage in 2017 at the national level [10], and substantial heterogeneity at the regional level. Vaccination coverage prior to the 2012 outbreak varied from 51.7% and 52.4% in the North and Far North regions to 85.1% and 93.7% in Littoral and the North West regions, respectively [11]. This spatial heterogeneity of vaccination coverage across the country is a likely culprit for the recurrent measles outbreaks in Cameroon. Heterogeneity of vaccination coverage may be conducive to source-sink type dynamics [12] whereby pockets of the country are more susceptible to new infections. The potential of these sources to generate and sustain local or national outbreaks depends on spatio-temporal trends in disease spread. Recurrent measles outbreaks arise disproportionately in northern Cameroon[8,13–15], throughout which poverty is widespread. The Far North and North regions have lower access to healthcare than the rest of the country, with half as many nurses and doctors per capita (0.53 per 1000) than the region where the nation's capital is located (1.10 per 1000 population in Center)[16]. Furthermore, these regions have over half their population in the lowest national economic quintile (54.8% and 51.7%, respectively), with female illiteracy rates (61.0% and 50.4%, respectively) double the national average (25.5%)[11]. The proximity of the Far North region to the southern Chadian and northern Nigerian borders, terrorized by recent activities of the Boko Haram militant group, has made it difficult to implement vaccination campaigns. This ongoing security issue exacerbates the risk of disease outbreaks in northern Cameroon which was already precariously positioned. To evaluate the drivers of measles transmission and identify hotspots for disease spread in Cameroon, we analyze the spatiotemporal trends that characterized the 2011-2012 measles outbreak, the most devastating in recent years. In particular, we analysed the spatio-temporal dynamics of the outbreak at multiple geographical scales to determine the respective contribution of each scale to measles transmission.

Methods: Data Sources: The Cameroon Ministry of Health mandates measles reporting with case follow-up as part of a passive surveillance system[17] operated by the Epidemiology Service in the National Disease Control Department. Measles case and death data are collected through health district-level passive surveillance conducted by clinics, health centers, and hospitals, which are then reported to the Ministry of Health. We analyzed surveillance data from the Ministry of Health on weekly measles cases and deaths reported in 2011 and 2012, from each of the 183 health districts in Cameroon. Health districts are the smallest geographic scale for healthcare services and decision-making in Cameroon. To evaluate the robustness of our analysis at larger geographic scales, we also aggregated data at the departmental and regional levels. There are 60 departments and 10 regions across the country. Data on measles vaccination coverage at the region level were obtained from the 2011 Cameroon Demographic and Health Survey (Enquête Démographiqe et de Santé et á Indicateurs Multiples: EDS-MICS) [11]. We used region level vaccination coverage of measles and health district level estimates of Diphtheria, Tetanus, and Pertussis (DPT-3) vaccination coverage in Cameroon [18], to estimate the department and health district level vaccination coverage of measles. Specifically, we assumed the health district level of measles vaccination coverage relative to the region-level vaccination coverage was similar to that of DPT-3. We used this information to determine the expected heterogeneity of measles vaccination coverage at the health district and department level, while ensuring the region level coverage for measles was consistent with that described in EDS-MICS. Data on the number of health centers per health district as well as the number of health centers per health district that reported their number of measles cases each week were obtained from the Cameroon Ministry of Health. Spatial mapping of measles at the region, department, and health district levels was conducted using R, with a base map of health districts obtained from previous work on the spatial mapping of infectious diseases in Cameroon [19]. Case Definition: Suspected measles cases are defined as an illness characterized by a generalized maculopapular rash and fever, as well as one or more of the following symptoms: cough, coryza, and conjunctivitis; or as any case suspected to be measles by a clinician. For suspected measles cases identified within 30 days after the onset of symptoms, a laboratory confirmation test for measles-specific immunoglobulin (Ig) M antibody is requested by the Ministry of Health. Measles case confirmation is either defined by a positive result for measles IgM antibody testing or by epidemiological linkage. The data available from the Disease Control Department of the Ministry of Health were suspected cases. No information with which to classify them as laboratory confirmed or epidemiologically linked cases was available. These data were collected through the integrated disease surveillance system rather than the measles casebased reporting system. Analysis: Spatial autocorrelation of measles incidence across health districts was assessed by calculating Moran’s I statistic using Monte Carlo simulation (n=1000). While Moran’s I statistic provides a measure of spatial autocorrelation across all health districts, local Moran’s I statistics were calculated to assess spatial autocorrelation for each health district. To analyse the spatio-temporal trends of the 2011-2012 measles outbreak, we fitted a range of multivariate time-series models to identify which model most effectively describes the data, and used this model to assess the spatio-temporal dynamics of the outbreak and evaluate the contribution of population mobility to disease transmission. The analysis was successively conducted at the region,

department, and health district levels. The R package “surveillance” was used for development of the multivariate time-series model [20–22]. We generated multivariate time-series model for measles case counts, , , in geographic localities ( = 1, . . . , ) during weekly time periods ( = 1, . . . , ) for a total of two years [23,24]. The model accounts for (1) the contribution of local transmission to disease outbreak and (2) the spatial spread of disease from individuals’ movement between localities. This model was used to assess the spatio-temporal trends of the 2011-2012 measles outbreak in Cameroon at spatial scales of decreasing size: region ( = 1, . . . ,10), department ( = 1, . . . ,60), and health district ( = 1, . . . ,183). Measles vaccination coverage, the number of health centers per population per locality, and the weekly proportion of health centers that reported their number of measles cases to the national surveillance system were used as covariates in the models. The number of health centers per population was used to represent the capacity for individuals to access health care and have their measles infections diagnosed. Departments with fewer than 50 measles cases from 2011-2012 were excluded from the analysis, as well as the health districts located in these departments. These excluded departments and health districts were given weights of 0 in the first order model. Conditional on past observations, is assumed to have a negative binomial distribution with mean incidence and variance of = (1 + ), with an overdispersion parameter > 0: =

,

+

"

!# ,! $

! !,

,

ℎ & ≠ .

The mean incidence was decomposed into a within-locality ( ) component (i.e., the , generation of new measles cases from cases within locality ), and a spatial spread (i.e., between-locality interaction through population movement) component ( ∑!# ,!$ ! !, ) which accounts for the contribution of another locality, &, to disease transmission in locality . !, indicates the number of cases observed in locality & at time − 1. The spatial component includes weights ( ! ) which reflect the flow of infections from locality & to locality , and a within-locality parameter, , which captures the contribution of !, , where & ≠ , on , . The epidemic components indicate a strict dependence between events driven by the observed past cases, and captures occasional outbreaks in the time-series data. We used several models to capture the flow of infections between areas: first-order (nearest neighbour interaction), second-order, gravity, and power law models. Each model represents a different scenario for population movement between localities. An adjacency order matrix was generated to depict whether or not localities are neighbours based on sharing a common border, which was used to determine the weights. At each geographic level, we identified the type and structure of model that exhibited the optimal fit to measles data. The first order model assumes that spatial spread of disease may only arise as a result of cases in directly adjacent localities. Here, ! = 1 if & and are adjacent and 0 otherwise. The second order model includes weights that decay with distance for first and second-order neighbours only, eliminating the possibility of transmission of cases from localities more than 2 neighbours away. In the second order model, ! = 1 ⋅ 1(+! = 1) + , -. ⋅1(+! = 2), where 1 is the indicator function and +! is the adjacent order between localities & and , and 01 represents the relative contribution of the second order neighbour, relative to the first order neighbour. The spatial gravity model involves having attraction to a locality scale with its population size to reflect commuter-driven disease spread. This model is developed by multiplying the ability of a locality to import cases from neighbouring localities, i.e. 4+4 susceptibility (2 ), by the population fraction of the given locality (, 3 ). This model accounts for the idea that humans tend to travel further and preferentially to densely populated metropolises. Weights

in this model are the same as the first order model. The power law model accounts for potential longdistance transmission events between all localities. Weights ( ! ) are defined as a function of the adjacency order (+! ) between localities( ! = +! , for& ≠ and !! = 0), where 5 is decay parameter which represents the deterioration in the impact of locality & on cases in locality as the number of localities between the two areas increases, and are normalized such that ∑! ! = 1. At each level of geographic analysis, Akaike information criterion (AIC), an estimator of the relative quality of a model based on the maximum likelihood, was calculated for each model. The model with the lowest AIC, which indicates a combination of optimal goodness of fit with discouragement for overfitting, was identified at the region, department, and health district levels. Correlation coefficients between data and the model with the lowest AIC were calculated for all geographic levels of analysis in order to assess goodness of model fit to the data. At the health district level, the proportion of cases attributable to disease transmission within the district, within the department in which the district is located (excluding the district itself), within the region in which the district is located (excluding the department in which the district is located), and within the rest of the country (excluding the region in which the district is located), were calculated. The same was calculated at the department level, for transmission within the department, within the region in which the department is located (excluding the department), and within the rest of the country (excluding the region in which the department is located). Detailed descriptions of all models are provided in the Supplementary Material.

Results: Descriptive analysis A total of 14,806 measles cases and 73 measles-attributable deaths were reported during 2012 in Cameroon. The vast majority (97.4%) of cases were reported within the first six months of the year, with a peak in cases during week 12 (Figure S1). The North region reported the highest measles incidence, with 182.5 cases per 100,000 population, and the South West region had the lowest at 17.7 cases per 100,000 population. Health districts with the lowest cumulative measles incidence per 100,000 population (Figure 1A) are located in the Western (West, North West, South West, and Littoral) regions, which are correspondingly the regions with highest measles vaccination coverage (Figure 1B). Regions with low vaccination coverage have highly heterogeneous measles incidence rates (Figure 1A). The Adamaoua region in particular, which has a relatively low overall measles vaccination coverage of 64.0%, has starkly different cumulative incidences of measles across its health districts, ranging from 3.1 to 103.3 cases per 100,000 population. The East (74.4%) and South (69.5%) regions also have moderate to low vaccination coverages overall, with highly variable cumulative incidence in 2012 by health district. The Moran’s I coefficient (index = 0.206, p = 0.001) suggests that the measles outbreak in Cameroon was characterized by statistically significant spatial autocorrelation. This indicates spatial clustering of cumulative measles incidence across health districts. Upon calculation of local Moran’s I statistics for each health district, we identified two statistically significant clusters of measles incidence in districts within the North (Bibemi, Garoua I, Garoua II, Lagdo, Ngong, and Rey Bouba) and Far North (Bogo and Vele) regions.

Figure 1: Measles Incidence and vaccination coverage in Cameroon. (A) Cumulative incidence of measles per 100,000 population in Cameroon by health district in 2012. Red indicates a higher incidence and blue indicates a lower incidence of measles, with colour divisions by quantiles. Yaoundé (star), the capital, and Douala (circle), the economic capital, are the most populated cities in Cameroon. (B) Measles Vaccination Coverage (%) in Cameroon by region in 2011 [11]. Red indicates lower vaccination coverage and blue indicates higher vaccination coverage. Spatio-temporal analysis Region Level At the region level, the second-order model provided the best fit to the data when measles vaccination coverage, the number of health centers per population, and the weekly proportion of health centers that reported measles cases to the national surveillance system in each region were accounted for in the model. Compared to the first-order (nearest neighbour) model, the gravity model, and the power law model with these three covariates, the second-order model had an AIC that was 89.7, 41.7, and 29.6 lower, respectively. At this geographic scale, a second-order neighbour model is synonymous with a long-distance movement model given that there are on average three degrees of separation (sharing the same border) between regions. Correlation between the reported cases and the secondorder neighbour model exceeded 0.5 for all regions and exceeded 0.75 for 80% of regions, indicating that the model provides a good fit to the data. The model showed that the reported measles cases in each region were predominantly the result of within-region transmission with very little contribution from other regions (Figure 2). Measles transmission between regions had its highest contribution to measles outbreaks in the Littoral and Center regions, 59.0% and 37.7% of cases, respectively (Table 1). On average, 16.7% of cases were attributable to transmission from outside of the region (Table 1).

Figure 2: Second-order model fitted to weekly measles cases from 2011-2012 at the region level. The model-generated numbers of cases estimated to be driven by interaction with first and second-order neighbouring regions are represented in orange and the numbers of cases attributable transmission within each region are in purple. Black dots represent weekly measles cases at the region level as reported by the Ministry of Health. Measles vaccination coverage, the number of health centers per population, and the weekly proportion of health centers that reported measles cases to the national surveillance system in each region were adjusted for in the model. Department Level At the department level, we only considered the 38 departments that cumulatively reported more than 50 cases from 2011-2012. This restriction was made to minimize the impact of underreporting in our analysis. Compared to the first-order model, the gravity model, and the secondorder model when both measles vaccination coverage and the number of health centers per population are included as model covariates, the power law model with these two covariates had an AIC that was 41.15, 40.49, and 0.16 lower, respectively. The small difference between the AIC of the second-order model and the power law model suggests that there is substantial support for both models. Therefore, we could not preferentially choose one of the models over the other. Correlation between the data and the power law model exceeded 0.5 for 84.2% of departments and exceeded 0.7 for 42.1% of departments. Our analysis showed that across the country, disease transmission was primarily driven by within-department transmission (Figure 3A). On average, within-department transmission contributed to 63.93% of cases, within-region transmission contributed to 16.0% of cases, and transmission from other regions contributed to 20.1% of cases.

Transmission within-departments predominantly contributed to cases in the North (average: 72.7%, range: 51.7% in Mayo-Louti to 96.7% in Benoue), South (68.7%, 67.0 in Dja-et-Lobo to 70.4% in Ocean), and the Far North (70.2%, 47.6% in Mayo-Kani to 90.5% in Diamare) regions (Figure 3A and Table 1). Transmission between departments in the same region had the greatest impact in the Far North and Center regions where it contributed to 22.7% and 23.2% of cases, respectively (Table 1). In the Far North region, transmission between departments contributed to over 30% of cases in the LogonChari and Mayo-Kani departments (Figure 3B). In the Center region, it contributed to over 30% of cases in the Mfoundi and Lekie departments (Figure 3B). Transmission from other regions had the greatest impact in departments of the South and North West regions (Figure 3C and Table 1).

Figure 3: Proportion of cumulative measles cases from 2011-2012 in each department that are attributable to A) Department-level transmission, B) Region-level transmission excluding the department of interest, and C) Country-level transmission excluding the region in which the department of interest is located. Grey areas indicate departments with 50 or fewer cases over the two year period. Health District Level At the health district level, the power law model was also shown have the best fit to the data when both measles vaccination coverage and the number of health centers per population are included as model covariates. A total of 141 health districts are situated within the 38 departments reporting more than 50 cases from 2011-2012. In comparison to the first-order model, the gravity model accounting for commuter-driven travel, and the second-order model with these two covariates included, the power law model at the health district level with vaccination coverage and health centers per population had an AIC that was 246.23, 137.39, and 69.01 less, respectively. Among the 92 health districts that reported over 20 cases from 2011-2012, correlation between the data and the power law model results exceeded 0.5 in 65.2% of health districts and exceeded 0.7 in 26.1% of health districts. As observed at the region (Figure 2) and department levels (Figure 3), measles cases were also predominantly explained by within-district transmission at the health district level (Figure 4A). However, the contribution of between-district interaction in driving cases through population movement was higher at the health district level (Figure 4B–D), than between-department interaction in the department level of analysis (Figure 3). On average, transmission within districts contributes to 51.2% of cases in health districts, transmission between health districts in the same department contributes to

7.6% of cases, transmission between health districts in the same region contributes to 11.4% of cases, and other regions contribute to 29.7% of cases (Table 1). Transmission between districts within the same department had its greatest contribution to disease transmission in the South-West, with on average 13.4% (7.9% in Limbe to 28.0% in Konye) of cases attributable to population movement within a department (Figure 4B and Table 1). Transmission between districts within a region had its greatest impact on measles cases in the Far North region (Figure 4C and Table 1). It contributed to more than 27% of cases in 15 of the 29 health districts of interest in the region, including Tokombere and Meri, where within-region transmission contributed to 49.7% and 47.7% of cases, respectively (Figure 4C). Transmission from other regions contributed on average to 47.6% of cases in health districts in the South region (varying from 15.6% in Kribi to 46.5% in Sangmelima) and 43.1% of cases in the Littoral region (varying from 10.8% in Nylon to 76.2% in Melong) (Figure 4D and Table 1).

Figure 4: Proportion of cumulative measles cases from 2011-2012 in each health district that are attributable to A) Health District-level, B) Department-level transmission excluding the district of

interest, C) Region-level transmission excluding the department in which the district of interest is located, and D) Country-level transmission excluding the region in which the district of interest is located. Grey areas indicate health districts within departments with 50 or fewer cases over the two year time-period. Population movement across localities plays a greater role in disease transmission at the health district level than the department and region level. Between-district transmission contributed to more than 50% of cases in 44.0% of health districts, compared to 13.3% of departments in which betweendepartment transmission contributed to more than 50% of measles cases. Across the country, betweenlocality transmission contributed to 16.7%, 36.1%, and 48.8% of measles cases at the region, department, and health district levels, respectively (Table 1). At the region level, the proportion of cases attributable to transmission from outside a given region ranged from 0.1% in Adamaoua, the East, and the South to 59.0% in Littoral (Table 1). Other regions contributed to 20.1% of cases at the department level on average, ranging from 7.1% of cases in the Far North to 30.0% of cases in the South (Table 1). Finally, at the health district level, other regions excluding the region in which the health district is located contributed to 29.7% of cases on average, ranging from 15.3% of cases in the Far North to 47.6% of cases in the South (Table 1).

Table 1: Average contribution (%) of transmission at the region, department, and health district levels. At the region level, the proportion of cases attributable to the region of interest and that attributable to all other regions are reported. At the department level, transmission attributable to the department, other departments in the region the given department is located, and other regions that do not contain the department of interest are reported. At the health district level, transmission attributable to the health district, other health districts in the department within which the health district is located, other departments in the region in which the health district is located, and other regions that exclude that of the health district are reported. The economic capital, Douala, is located in Littoral and the political capital, Yaoundé, is located in the Center region. Region-level (%)

Department-level (%)

Health District-level (%)

Region

Region

Other Regions

Department

Other Departments in Health Other Health Health Health Departments other Regions District Districts in Districts in Districts in other within the the other Region Department Departments Regions within the Region

Adamaoua

99.9

0.1

60.4

19.9

19.7

68.6

0.4

5.8

25.2

Center

62.3

37.7

53.4

23.2

23.4

51.1

8.4

11.5

29.0

East

99.9

0.1

63.2

20.0

16.9

51.6

4.6

9.4

34.4

Far North

81.1

18.9

70.2

22.7

7.1

48.1

10.2

26.4

15.3

Littoral

41.0

59.0

63.3

9.4

27.2

40.2

6.2

10.5

43.1

North

98.3

1.7

72.7

6.8

20.5

64.5

6.5

3.8

25.2

North West

80.2

19.8

68.1

5.8

26.1

48.2

12.9

5.2

33.7

West

97.6

2.4

67.9

12.1

19.9

56.7

6.5

4.1

32.7

South

99.9

0.1

68.7

1.3

30.0

48.5

2.2

1.7

47.6

South West

96.8

3.2

63.2

11.6

25.2

43.1

13.4

7.4

36.1

Cameroon

83.3

16.7

63.9

16.0

20.1

51.2

7.6

11.4

29.7

Overall, the majority of cases in areas reporting large outbreaks (Figure S3) were driven by within-area transmission. In departments with over 200 cases, an average of 72.6% of cases were due to within-department transmission, and in health districts with over 100 cases, 73.5% of cases on average were due to within-district transmission.

Discussion: The power law model accounting for long-distance population travel was found to have the best fit at the health district and department levels. At the region level the second order model, which at this level also inherently considers long-distance travel, was the best fit. These models showed that local transmission was the main mechanism for disease transmission at the region, department, and health district levels. As expected, the contribution of between-locality interaction through population movement was shown to increase as the geographic scale was refined; however, the Littoral and Adamaoua regions are exceptions to this trend, where the contribution of population mobility to disease

transmission appears to be independent of spatial scale. On average, 63.9% (range: 15.1–97.9%) of cases in departments were attributable to within-departments transmission in contrast to 51.2% (range: 0.0–93.9%) of cases in health districts that were attributable to within-health district transmission. Interaction between departments contributed to 36.1% (range: 2.1–84.9%) of cases in departments and between-district interaction contributed to 48.8% (range: 6.1–100.0%) of cases in health districts, on average. These results indicate that population mobility between localities is an important risk factor for large scale measles outbreaks in Cameroon. While population mobility is integral to measles transmission in Cameroon, the country lacks substantial seasonal migrant populations, unlike other west African countries, where subnational seasonality has been shown to play an important role in measles dynamics [25]. For this reason, we assumed seasonality to be identical in all spatial units of study. Future work should investigate the potential impact of local seasonality on measles dynamics in migrant populations in Cameroon, such as the pastoral nomadic Mbororo populations [26]. As Littoral and Center contain the two most populated cities in the country, the economic and political capitals, Douala and Yaoundé, we find that substantial travel in and out of these cities, and in turn the regions, might be driving the measles outbreak far more so than in other regions in Cameroon, and that further work to elucidate the reasons behind this are necessary. At the region level, Littoral and Center had the highest proportion of cases attributed to transmission from outside the region, 59.0% and 37.7%, respectively, compared to all other regions which averaged 5.8%. Littoral in particular had a high vaccination coverage (85.1%) in 2011 compared to the country as a whole (70.6%), yet had a moderate regional measles incidence rate in 2012 (44.3 per 100,000 population) compared to the rest of the country (17.7 in the South to 182.5 per 100,000 population in the North). Our findings of the substantial contribution (36.7 to 59.0%) of between-locality transmission in driving measles cases in Littoral indicates the importance of maintaining high vaccination coverage country-wide in preventing measles cases in this highly traveled region. We also found this to be true in departments and health districts reporting large outbreaks (Figure S3). Our results agree with previous modeling studies that demonstrate that the epidemic trajectory of large-scale measles outbreaks in a community is virtually unaffected by immigrant infection [27,28]. These studies showed that epidemic dynamics were predominantly driven by within-locality transmission [27,28]. In Cameroon, within-locality movement is driven by numerous socio-economic factors including trade, farming, education, and family ties among others [29,30]. In some rural areas, children must travel regularly between their health district of residence and the nearest high school. Thus, schools may become hubs for the spread of diseases between the neighbouring health districts. The need to travel to regional and weekly markets also contributes significantly to rural-to-rural and urban-to-rural population movements within departments and regions [29]. Population movement for family-related reasons, such as visiting one’s locality of origin or relatives in other parts of the country, are a major cause of the movement of children during school breaks in Cameroon. This population mobility compounds the low vaccination coverage, especially in the northern regions, to exacerbate the risk of measles outbreaks in Cameroon. Low vaccination coverage is driven by a combination of poverty, inaccessibility to healthcare services, parents level of education, religious beliefs, and parents attitudes towards vaccination [31,32]. While our study focuses on the impact of population mobility within national borders on measles outbreak dynamics, cross-border movement is also an important instigator of the spread of measles [33]. Although the impact of cross-border movement was not assessed in our model, the current security crisis in the northern regions of Cameroon has increased the risk of measles outbreak in the country and its subsequent spread to Nigeria and Chad through mass migration. Furthermore, Chad experienced a large measles outbreak in 2011 which may have resulted in case importation in to Northern Cameroon. However, in the absence of data from Chad during this time, we are unable to access the impact of cross-border movement on case importation.

Our descriptive analysis demonstrates that the measles outbreak started in Adamaoua and the North, where the population is relatively dispersed, before moving to more densely populated regions near the national capital, Yaoundé, and the economic capital, Douala. This is consistent with northern regions being hotspots for measles outbreaks due to their lower vaccination coverage, health care providers to inhabitants ratio, income, and female education levels, than other regions. Moreover, the start of the 2012 outbreak, during late 2011 and early 2012, coincides with the Christmas to New Year holiday season which is marked by high population movement between urban, higher vaccination coverage areas and rural, lower vaccination coverage areas. While vaccination coverage is greater around the large cities in the western part of the country than in northern Cameroon, the higher population density in Yaoundé and Douala places more people at risk of measles infection, furthering the spread of the epidemic. Heterogeneity of vaccination coverage creates environments for outbreaks to take off [34] and weakens the impact of herd immunity [35]. Supplementary immunization activities (SIAs) have been conducted regularly in Cameroon in order to improve measles vaccination coverage. However, some of these SIAs have been localized and have historically missed key areas in Cameroon with particularly low vaccination coverage [36]. Before the 2012 measles outbreak, catch-up SIAs for children 9 to 59 months old or follow-up SIAs targeting children born since the most recent SIA had been conducted in Cameroon in 2001, 2002, 2006, 2007, 2009, and 2012 [6,37,38] The SIA in 2012 reached over 3.5 million children nationally such that 78% of targeted districts had ≥95% coverage [6]. However, SIA were initiated in April well after the start of the outbreak and did not target children aged 6 to 14 years old who have a low measles vaccination coverage and are at high risk of infection [6,39,40]. In Cameroon, population movement tends to occur in the direction of city centers [41], which informed our consideration of gravity models for fitting the epidemic trajectory. However, our findings have indicated that pull of individuals towards high-density areas, accounted for in the gravity model, does not sufficiently explain this measles outbreak. Accounting for long-distance transmission was shown to better explain the spread of measles in Cameroon at all geographic levels of analysis. Particularly as the resolution of data increases from the region to health district level, population movement was shown to play an increasingly important role in disease transmission. In the absence of population mobility data, the power law model provides a useful way to measure the influence of longdistance travel on disease spread [42]. A study on cases of measles in Cameroon in 2000-2001 identified distinct patterns of measles in the three northern regions compared to the seven southern ones, with annual major epidemics in the north and major epidemics only occurring in the south every 3 years. [13] This study identified that higher cumulative region-specific incidence rates were associated with higher birth and lower routine vaccination rates [13]. The Benakuma health district, located in North West Cameroon, experienced a measles outbreak in 2015 and was identified as a hotspot for future measles outbreaks due to a combination of poor vaccination levels, low socio-economic status overall, and environmental factors limiting ease of vaccination [43]. We identified that 61.0% of cases in the 2011-2012 outbreak in Benakuma were attributable to between-district interaction. These findings indicate the importance of identifying and then improving health district-specific vaccination coverage for curtailing future measles outbreaks. The measles surveillance system in Cameroon is mainly based on passive case diagnosis and reporting by local health facilities. Given inadequate infrastructure, the low number of health facilities, and the low ratio of health care providers to inhabitants in many health districts, case diagnosis and timely reporting is challenging in many parts of the country. This situation has surely resulted in underreporting of measles cases in some health districts, as exemplified by changes in the proportion of health centers that reported their number of measles cases to the national surveillance system. For example, in the Far North region, which is especially vulnerable to infectious disease outbreaks given

low vaccination coverage and conflict, the reports of very low annual measles incidence (<2 cases per 100,000 population) in some health districts neighbouring high incidence districts brings into question the ability of such districts to accurately diagnose and report cases. The potentially high level of underreporting of measles cases in some Far North health districts is likely to affect our results on the relative contribution of the different transmission routes on disease outbreaks not only within the health districts but also in some of their neighbouring districts. However, under-reporting of cases were not explicitly considered in our models due to lack of information. By using the weekly number of health centers in each district that reported the number of measles cases that occurred to the national surveillance system, as well as the number of health centers in each district per population as model covariates, we aimed to mitigate the impact of under-reporting. Future studies should focus on underreporting of infectious diseases in Cameroon and its impact on public health policy. Overall, our analysis shows that considering the role of population movement from higher order neighbours, in addition to directly neighbouring areas, is essential in understanding transmission dynamics of measles in Cameroon. On average 36.1 to 48.8% of cases occurring at the department or district level, respectively, originated from population movement outside of the residential locality. This indicates that substantial health benefits of improving measles vaccination within a particular district or department are likely to be realized beyond that locality’s borders. Improving vaccination coverage in rural and high risk transmission areas such as the North and Far North regions of Cameroon would not only benefit these regions but would also provide benefits to urban areas and the rest of the country.

Conclusions: This study found that the 2011-2012 measles outbreak in Cameroon was driven by a combination of local and long-distance transmission factors. The contribution of population movement to disease transmission was shown to be highly heterogeneous across the country. Improving our understanding of vaccination coverage at health district and department levels will be essential in mitigating measles cases originating due to long-distance movement of populations across artificial, administrative boundaries. Future research will aim to evaluate the impact of targeted increase of vaccination coverage on reducing measles outbreaks and spread of disease between regions, departments, and health districts across the country.

Authors' contributions: ASP and MLNM designed the study; ASAA acquired the data; ASP performed the analysis; ASP, EON, and MLNM interpreted the data. MCN contributed the health district level shapefile. ASP, LS, APG, and MLNM drafted the manuscript. All authors read and provided edits on the final paper. Acknowledgements: The authors would like to acknowledge the provision of data from the Cameroon Ministry of Health, and three anonymous reviewers for very constructive comments. Funding: This study was funded by the National Institutes of Health and a faculty startup funding from Texas A&M College of Veterinary Medicine and Biomedical Sciences. Martial L. Ndeffo-Mbah is supported by a faculty startup funding from Texas A&M. Alison P. Galvani is supported by NIH grant U01 GM087719. Elaine O. Nsoesie is supported by grant K01ES025438-04.

References: [1] World Health Organization. Fact Sheet: Measles. WHO 2018. https://www.who.int/newsroom/fact-sheets/detail/measles (accessed March 7, 2019). [2] IHME, University of Washington. GBD Compare Data Visualization. Institute for Health Metrics and Evaluation (IHME) 2018. https://vizhub.healthdata.org/gbd-compare/ (accessed January 25, 2019). [3] WHO Regional Committee for Africa. Sixty-first session of the WHO Regional Committee for Africa: Final Report. World Health Organization Regional Office for Africa; 2011. [4] World Health Organization. Progress towards measles control in WHO’s African Region, 2001-2008. Weekly Epidemiological Record 2009;84:397–404. [5] Masresha BG. Progress Toward Measles Elimination — African Region, 2013–2016. MMWR Morb Mortal Wkly Rep 2017;66. doi:10.15585/mmwr.mm6617a2. [6] Masresha BG, Kaiser R, Eshetu M, Katsande R, Luce R, Fall A, et al. Progress Toward Measles Preelimination — African Region, 2011–2012. Morbidity and Mortality Weekly Report (MMWR) 2014;63:285–91. [7] Otten M, Kezaala R, Fall A, Masresha B, Martin R, Cairns L, et al. Public-health impact of accelerated measles control in the WHO African Region 2000-03. Lancet 2005;366:832–9. [8] Luquero FJ, Pham-Orsetti H, Cummings DAT, Ngaunji PE, Nimpa M, Fermon F, et al. A long-lasting measles epidemic in Maroua, Cameroon 2008-2009: mass vaccination as response to the epidemic. J Infect Dis 2011;204 Suppl 1:S243–51. [9] Goodson JL, Masresha BG, Wannemuehler K, Uzicanin A, Cochi S. Changing epidemiology of measles in Africa. J Infect Dis 2011;204 Suppl 1:S205–14. [10] World Health Organization, UNICEF. Cameroon: WHO and UNICEF estimates of immunization coverage: 2017 revision. WHO; UNICEF; 2018. [11] Libite PR, Kelodjoue S, Dzossa AD, Fomo MA, Niekou R, Jazet E, et al. Cameroun: Enquête Démographiqe et de Santé et á Indicateurs Multiples (EDS-MICS) 2011. Institut National de la Statistique Ministère de l’Économie de la Planification et de l’Aménagement du Territoire Ministère de la Santé Publique; ICF International; 2012. [12] Amarasekare P, Nisbet RM. Spatial heterogeneity, source-sink dynamics, and the local coexistence of competing species. Am Nat 2001;158:572–84. [13] Cummings DAT, Moss WJ, Long K, Wiysonge CS, Muluh TJ, Kollo B, et al. Improved measles surveillance in Cameroon reveals two major dynamic patterns of incidence. Int J Infect Dis 2006;10:148–55. [14] ReliefWeb. Cameroon: Measles and Cholera Outbreaks - Nov 2015. ReliefWeb 2015. https://reliefweb.int/disaster/ep-2015-000156-cmr (accessed February 12, 2018). [15] World Health Organization. WHO in Cameroon Annual Report 2016. World Health Organization; 2016. [16] Tandi TE, Cho Y, Akam AJ-C, Afoh CO, Ryu SH, Choi MS, et al. Cameroon public health sector: shortage and inequalities in geographic distribution of health personnel. Int J Equity Health 2015;14:43. [17] Ngwa MC, Liang S, Mbam LM, Mouhaman A, Teboh A, Brekmo K, et al. Cholera public health surveillance in the Republic of Cameroon-opportunities and challenges. Pan Afr Med J 2016;24:222. [18] Mosser JF, Gagne-Maynard W, Rao PC, Osgood-Zimmerman A, Fullman N, Graetz N, et al. Mapping diphtheria-pertussis-tetanus vaccine coverage in Africa, 2000-2016: a spatial and temporal modelling study. Lancet 2019;393:1843–55. [19] Ngwa MC, Liang S, Kracalik IT, Morris L, Blackburn JK, Mbam LM, et al. Cholera in Cameroon, 20002012: Spatial and Temporal Analysis at the Operational (Health District) and Sub Climate Levels. PLoS Negl Trop Dis 2016;10:e0005105.

[20] Meyer S, Held L, Höhle M. hhh4: Endemic-epidemic modeling of areal count time series. 2017. [21] Meyer S, Held L, Höhle M. Spatio-Temporal Analysis of Epidemic Phenomena Using the R Package surveillance. Journal of Statistical Software, Articles 2017;77:1–55. [22] Paul M, Meyer S. hhh4: An endemic-epidemic modelling framework for infectious disease counts. University of Zurich; 2016. [23] Held L, Höhle M, Hofmann M. A statistical framework for the analysis of multivariate infectious disease surveillance counts. Stat Modelling 2005;5:187–99. [24] Paul M, Held L, Toschke AM. Multivariate modelling of infectious disease surveillance data. Stat Med 2008;27:6250–67. [25] Ferrari Matthew J., Djibo Ali, Grais Rebecca F., Bharti Nita, Grenfell Bryan T., Bjornstad Ottar N. Rural–urban gradient in seasonal forcing of measles transmission in Niger. Proceedings of the Royal Society B: Biological Sciences 2010;277:2775–82. [26] Pelican M. Mbororo on the move: from pastoral mobility to international travel. J Contemp Afr Stud 2011;29:427–40. [27] Bjørnstad ON, Grenfell BT. Hazards, spatial transmission and timing of outbreaks in epidemic metapopulations. Environ Ecol Stat 2008;15:265–77. [28] Bjørnstad ON, Finkenstädt B, Grenfell BT. Endemic and epidemic dynamics of measles. I. Estimating transmission rates and their scaling using a time series SIR model. Ecol Monogr 2002;72:185–202. [29] Mainet H. Town dwellers in their networks: urban-rural mobility and household strategies in Cameroon. Geografisk Tidsskrift-Danish Journal of Geography 2017;117:117–29. [30] Steel G, Birch-Thomsen T, Cottyn I, Lazaro EA, Mainet H, Mishili FJ, et al. Multi-activity, Multilocality and Small-Town Development in Cameroon, Ghana, Rwanda and Tanzania. The European Journal of Development Research 2019;31:12–33. [31] Russo G, Miglietta A, Pezzotti P, Biguioh RM, Bouting Mayaka G, Sobze MS, et al. Vaccine coverage and determinants of incomplete vaccination in children aged 12-23 months in Dschang, West Region, Cameroon: a cross-sectional survey during a polio outbreak. BMC Public Health 2015;15:630. [32] Wiysonge CS, Uthman OA, Ndumbe PM, Hussey GD. Individual and contextual factors associated with low childhood immunisation coverage in sub-Saharan Africa: a multilevel analysis. PLoS One 2012;7:e37905. [33] Sarkar S, Zlojutro A, Khan K, Gardner L. Measles resurgence in the USA: how international travel compounds vaccine resistance. Lancet Infect Dis 2019;19:684–6. [34] Durham DP, Ndeffo-Mbah ML, Skrip LA, Jones FK, Bauch CT, Galvani AP. National- and state-level impact and cost-effectiveness of nonavalent HPV vaccination in the United States. Proc Natl Acad Sci U S A 2016;113:5107–12. [35] Ndeffo Mbah ML, Liu J, Bauch CT, Tekel YI, Medlock J, Meyers LA, et al. The impact of imitation on vaccination behavior in social contact networks. PLoS Comput Biol 2012;8:e1002469. [36] Njim T, Aminde LN, Feteh FV, Ngum JM, Moustapha CA. Measles outbreak in a poorly vaccinated region in Cameroon: a case series study, public health challenges and recommendations. Pan Afr Med J 2015;22:163. [37] Masresha BG, Fall A, Eshetu M, Sosler S, Alleman M, Goodson JL, et al. Measles mortality reduction and pre-elimination in the African region, 2001-2009. J Infect Dis 2011;204 Suppl 1:S198–204. [38] Centers for Disease Control and Prevention (CDC). Progress toward measles control - African region, 2001-2008. MMWR Morb Mortal Wkly Rep 2009;58:1036–41. [39] CAMEROUN - Enquête Post-Campagne de vaccination contre la rougeole, d’évaluation de la distribution de la vitamine A et du mebendazole au Cameroun et d'évaluation des JLV contre la polio dans les régions septentrionales (2012), Phase 1 - Aperçu n.d. http://slmp-550104.slc.westdc.net/~stat54/nada/index.php/catalog/33/overview (accessed July 20, 2019).

[40] Ferrari M. J., Grenfell B. T., Strebel P. M. Think globally, act locally: the role of local demographics and vaccination coverage in the dynamic response of measles infection to control. Philos Trans R Soc Lond B Biol Sci 2013;368:20120141. [41] Meva’a Abomo D, Abessolo Nguema JR, Begoumenie B, Ba’ana Etoundi ML, Manga Engama E, Fotso Wougaing J, et al. Internal migration in Cameroon: Constraint for or driver of urban and health development? ACP Observatory on Migration; 2013. [42] Geilhufe M, Held L, Skrøvseth SO, Simonsen GS, Godtliebsen F. Power law approximations of movement network data for modeling infectious disease spread. Biom J 2014;56:363–82. [43] Njim T, Agyingi K, Aminde LN, Atunji EF. Trend in mortality from a recent measles outbreak in Cameroon: a retrospective analysis of 223 measles cases in the Benakuma Health District. Pan Afr Med J 2016;23:135. [44] Cheng Q, Lu X, Wu JT, Liu Z, Huang J. Analysis of heterogeneous dengue transmission in Guangdong in 2014 with multivariate time series model. Sci Rep 2016;6:33755. [45] Keeling MJ, Eames KTD. Networks and epidemic models. J R Soc Interface 2005;2:295–307. [46] Wen T-H, Hsu C-S, Hu M-C. Evaluating neighborhood structures for modeling intercity diffusion of large-scale dengue epidemics. Int J Health Geogr 2018;17:9. [47] Wesolowski A, Buckee CO, Pindolia DK, Eagle N, Smith DL, Garcia AJ, et al. The use of census migration data to approximate human movement patterns across temporal scales. PLoS One 2013;8:e52971. [48] Wesolowski A, O’Meara WP, Eagle N, Tatem AJ, Buckee CO. Evaluating Spatial Interaction Models for Regional Mobility in Sub-Saharan Africa. PLoS Comput Biol 2015;11:e1004267. [49] Sorichetta A, Bird TJ, Ruktanonchai NW, Zu Erbach-Schoenberg E, Pezzulo C, Tejedor N, et al. Mapping internal connectivity through human migration in malaria endemic countries. Sci Data 2016;3:160066. [50] Brockmann D, Hufnagel L, Geisel T. The scaling laws of human travel. Nature 2006;439:462–5.

Supplementary Material Methods We generated an endemic-epidemic multivariate time-series model for measles case counts, , in geographic localities ( = 1, . . . , ) during weekly time periods ( = 1, . . . , )for a total of two , years[23,24]. This model was used to assess the spatio-temporal trends of the 2011-2012 measles outbreak in Cameroon at spatial scales of decreasing size: region ( = 1, . . . ,10), department ( = 1, . . . ,60), and health district ( = 1, . . . ,183)geographic area levels. Our model accounts for within-locality transmission as well as transmission from neighbouring localities. Conditional on past observations, Yit, the time series of measles case counts, is assumed to have a negative binomial distribution with mean incidence: =

,

# "

,

+

,

, !

,

+" #

: Mean measles incidence : Within-area components for capturing occasional outbreaks. : Between-area driven effects, the influence of area $ on area : Weight matrix reflecting flow of infections from area $ on area : Number of cases observed in area $ at time − 1 : Endemic component which represents baseline cases and is temporally stable : Population fraction offset in each area at time

is the Past cases in other neighbouring areas are explanatory covariates in the model. , number of cases observed in area $at time − 1, and are weights reflecting flow of infections from area $to area . The mean incidence, , decomposes additively into a within-area transmission( ) component (i.e., the reproduction of measles within area ) and a between-area , transmission parameter affects the influence of , , $ ≠ , on , (i.e., the transmission of measles from other areas j to area i). The conditional variance of , is (1 + ' ), with an overdispersion parameter ' > 0. The unit of area, , refers to the region ( ),department( ),or health district ( ),depending on which geographic level of analysis is being considered, and the fraction of population living in area is used as the endemic offset. In the endemic components, risk of new events are driven by external factors independent of the history of the epidemic process such as seasonality, population density, and vaccine coverage. This component of the model explains a baseline rate of cases with a stable temporal pattern. The endemic component, ) , represents the baseline cases and is stable over time: *+,() ) = -

/

(.)

: Intercept : trend parameters

+/

(.)

+ 0

6

2

1 ,2 3 4(5 ,2 ) + 7 ,2 8+3(5 ,2 )9

1 ,2 , 7 ,2 : unknown parameters for modeling seasonal variation 5 ,2 :2;3/52, Fourier frequencies

The other two log-linear components are the within-area ( [44]: *+,(

/

*+,(

) = -

) = -

(>)

)and between-area driven ( ?

+ / (>) @

(A)

+ /

(A)?

@

(>)

) effects

,

(A)

: Intercept : Regression parameters. Area-level effects which account for heterogeneity between areas, such as vaccination coverage, number of health centers per locality, and the proportion of received measles case reports out of the total expected per locality from the national surveillance system. @ : Exogenous covariate vector of time effects

An × adjacency order matrix was generated to depict whether areas are neighbours based on sharing a common border. This matrix which was used in determining between-area interaction weights. At the regional, departmental, and health district levels, we generated first-order, second-order, gravity, and power law, models and identified the type of model at each resolution of measles data that demonstrated the best fit. Each of these four models were fit with covariates for measles vaccination coverage, the number of health centers per population, and the proportion of received reports on weekly measles cases among those expected by the national surveillance system were incorporated into the auto-regressive component of all models. Vaccination coverage and the number of health centers per population vary by locality, while the proportion of received reports varies by locality and week in the year. For each of the four models detailed below, model selection using Aikake’s Information Criterion was used to determine the model and covariates that provide the best fit to the data. First-order Model The first-order model involves terms for endemic log-linear transmission () ) with the multiplicative effect of seasonality, between-area driven effects ( ), and within-area transmission effects ( ). The endemic component, ) , represents the baseline level of cases following a stable temporal pattern. This model assumes epidemics can only arise in a given area as a result of cases in directly adjacent areas. In this model, developed for all geographic levels of analysis, all areas have the same coefficient for case importation from directly neighbouring areas (same susceptibility). Therefore, the sum of all case counts in adjacent areas enters the model as an explanatory variable. Conditional on past observations, , is assumed to have a negative binomial distribution. The within-area epidemic parameter = "CD(- (>) ) and between-area driven epidemic parameter = "CD(- (A) ) are both homogenous across areas and remain constant over time. Here, the weights, wji, are such that the epidemic can only arise as a result of cases in directly adjacent districts = 1($ ∼ ) = 1(+ = 1). First-order (or nearest neighbour) models have been used previously in network modeling to simulate the links between individuals [23,45]. Second-order model The second-order model includes weights that decay with distance for first and second-order neighbours only, eliminating the possibility of transmission of cases from areas more than two neighbours away. = 1 ⋅ 1(+ = 1) + " GH ⋅1(+ = 2), where 1 is the indicator function. Therefore,

travel between adjacent areas and neighbours of neighbours are considered. Second order models have been used to represent the spread of dengue epidemics between cities [46]. Spatial-interaction Gravity Model The spatial interaction gravity model involves having attraction to an area scale with population size to reflect commuter-driven disease spread. This model is developed by multiplying the area’s ability to import cases from neighbouring areas, i.e. susceptibility ( ), by the fraction of the total population JKJ of the given area (" I ) where = 1($ ∼ ). This model accounts for the idea that humans tend to travel further and preferably to metropolitan, densely populated areas, and was developed for all geographic areas of analysis. The gravity model has been used in studies of human migration and mobility in relation to transmission of infectious diseases in Sub-Saharan Africa [47–49]. Power law model The power law model accounts for potential transmission of cases as a result of long-distance travel events and involves estimating a parameter for the decay in epidemiological coupling as distance between areas increases. Weights ( ) are estimated as a function of the adjacency order (+ ) between areas ( =+ , for $ ≠ and = 0), where L is decay parameter which represents the deterioration in the impact of locality j on cases in locality i as the number of localities between the two = 1. This model was developed for all geographic areas increases, and are normalized such that ∑ areas of analysis. The power-law model has been used previously to model human travel [50].

Supplementary Figures

Figure S1: Weekly measles cases and cumulative cases across Cameroon from 2011 to 2012.

Figure S2: Weekly incident measles cases from 2011 to 2012 in Cameroon by region. The vast majority of all cases during this time period originated from the North and Far North regions.

Figure S3. Measles cases across Cameroon in 2012 at the (A) Region, (B) Department, and (C) Health District levels.