Heterogeneity in social and epidemiological factors determines the risk of measles outbreaks

Abstract

Political and environmental factors-e.g., regional conflicts and global warming-increase large-scale migrations, posing extraordinary societal challenges to policymakers of destination countries. A common concern is that such a massive arrival of people-often from a country with a disrupted healthcare system-can increase the risk of vaccine-preventable disease outbreaks like measles. We analyze human flows of 3.5 million (M) Syrian refugees in Turkey inferred from massive mobile-phone data to verify this concern. We use multilayer modeling of interdependent social and epidemic dynamics to demonstrate that the risk of disease reemergence in Turkey, the main host country, can be dramatically reduced by 75 to 90% when the mixing of Turkish and Syrian populations is high. Our results suggest that maximizing the dispersal of refugees in the recipient population contributes to impede the spread of sustained measles epidemics, rather than favoring it. Targeted vaccination campaigns and policies enhancing social integration of refugees are the most effective strategies to reduce epidemic risks for all citizens.

Keywords: human mobility; infectious diseases; multiplex networks; population dynamics.

Fig. 1.

Model structure and human mobility. (A) Schematic illustration of the model considered in this work. Each prefecture of Turkey is considered as a node of a metapopulation network of geographic patches. Two populations, namely, Turkish and Syrians, are encoded by different colors and move between patches following the inferred interpatch mobility pathways. Turkish and Syrian populations encode two different layers of a multilayer system (–33), where social dynamics and epidemics spreading happen simultaneously. (B) Mobility of Syrian refugees (Upper) and Turkish citizens (Lower) between the prefectures of Turkey as inferred from CDRs. Different colors are used to indicate the number of individuals moving from a prefecture to another.

Fig. 2.

Measles immunity levels. (A) Reported number of measles disease cases over time, during the 2016–2018 measles epidemics in Syria, as recently reported by the WHO (27); red bars correspond to data points used to derive the $R_e$ as a function of the exponential growth rate of the observed epidemic. (B) Obtained fit of the epidemic exponential growth between September 2016 and February 2017 in Syria: The red solid line represents the mean estimate; the orange shaded area represents 95% CI. (C) Observed distribution of measles cases across different ages during the 2016–2018 measles epidemics in Syria, as recently reported by the WHO (27). (D) Model estimates of the age-specific immunity profile in Syria at the beginning of 2018: Green bars represents mean values; vertical black lines represent 95% CI. (E) Observed age distribution of Syrian refugees in Turkey (light gray) (42) compared with the population age distribution in Syria (dark gray). (F) Estimated (blue) and observed (red) distribution of measles cases across different ages during the 2013 and 2018 measles epidemics in Turkey. (G) Model estimates of the age-specific immunity profile in Turkey at the beginning of 2018. The gray bar represents the level of measles immunity in infants, here assumed to be 50%, as a consequence of maternal antibodies. Vertical bars in the age bands 1 to 4 y and 5 to 9 y show the variability (minimum and maximum) across different Turkish provinces in the immunity level of these age segments. The vertical bar shown for individuals older than 10 y of age represents the 95% CI of the immunity level in this age segment, as obtained by fitting the model on the age distributions of cases shown in F. (H) Estimated percentage of susceptible individuals among Syrian refugees in Turkey. Blue and light blue boxplots represent the immunity estimates obtained with the baseline model and by using the catalytic method, respectively. The red bar shows the observed immunity level among Syrian refugees reported in ref. in a refugee cohort in Germany in 2015. Black lines represent the 95% CI.

Fig. 3.

Risk factors. (A) Percentage of measles-susceptible Turkish citizens as estimated across different prefectures (34). (B) Ratio between Syrian refugees and Turkish population—i.e., $N_k^{\left(R\right)}/N_k^{\left(T\right)}$—as inferred from CDRs. (C) Proportion of traveling refugees visiting the different Turkish prefectures.

Fig. 4.

The potential of widespread measles epidemics. (A) Cumulative number of cases among refugees at the national level as expected by considering scenarios where the measles epidemics start from different patches by assuming 20% of Syrian contacts with Turkish citizens, $R_0=18$ and that the percentage of susceptible refugees is 11.8%. For each patch, the color encodes the estimated incidence in the total population due to epidemics starting in that prefecture: This choice allows one to appreciate how each geographic area can potentially affect the whole country. (B) As in A, but for 40% of Syrian contacts with Turkish citizens. (C) As in B, but for 60% of Syrian contacts with Turkish citizens. (D) As in A, but considering cases among Turkish citizens. (E) As in D, but for 40% of Syrian contacts with Turkish citizens. (F) As in D, but for 60% of Syrian contacts with Turkish citizens.

Fig. 5.

The potential burden of measles epidemics. (A) Cumulative infections considering epidemics that exceed 20 cases in the entire population, as obtained in a worst-case scenario where $R_0=18$, the percentage of susceptible refugees is set at $11.8\%$, and epidemic onset randomly occurs in one of the 100 most at-risk prefectures in Turkey. Bars represent the average number of infections occurring among Syrian refugees (red) and Turkish citizens (blue) for the model projections as a function of the mixing parameter; black lines represent 95% CI. (B) Estimated cumulative infections in the case of 20% of Syrian contacts with Turkish citizens. Bubble size is proportional to the average number of measles cases in the Turkish prefectures per 10,000 individuals. B, Inset displays the Istanbul prefectures. (C) As in B, but for 40% of Syrian contacts with Turkish citizens. (D) As in B, but for 60% of Syrian contacts with Turkish citizens. (E) As in A, but as obtained when refugees are assumed to be geographically dispersed in each patch proportionally to the amount of local Turkish residents. (F) As in B, but as obtained when refugees are assumed to be geographically dispersed in each patch proportionally to the amount of local Turkish residents. (G) As in F, but for 40% of Syrian contacts with Turkish citizens. (H) As in F, but for 60% of Syrian contacts with Turkish citizens.

Fig. 6.

Spatiotemporal spread of potential epidemics. (A) Estimated number of prefectures affected by the epidemic as a function of the mixing parameter in the worst-case scenario, where $R_0=18$, the percentage of susceptible refugees is set at $11.8\%$, and epidemic onset randomly occurs in one of the 100 most at-risk prefectures in Turkey. Bars represent the average number of prefectures exceeding 20 cases among Syrian refugees (red) and Turkish citizens (blue); black lines indicate the 95% CI. (B) Percentage of the simulated epidemic that exceeds 20 cases per prefecture in the case of 20% of Syrian contacts with Turkish citizens. Red and blue bubbles refer to Syrian refugees and Turkish citizens, respectively. (C) As in B, but for 40% of Syrian contacts with Turkish citizens. (D) As in B, but for 60% of Syrian contacts with Turkish citizens. (E, Upper) Relative incidence over time, considering both the populations per prefecture in the case of total segregation. Prefectures are ranked in decreasing order at week 30. (E, Lower) Proportion of region affected in the initial phase of the epidemic, considering the first 25 prefectures affected by more than 10 cases. Border refers to the Hatay, Kilis, Gaziantep, Sanliurfa, Mardin, and Sirnak regions. (F) As in E in the case of 20% of Syrian contacts with Turkish citizens. (G) As in E in the case of 40% of Syrian contacts with Turkish citizens. (H) As in E in the case of 60% of Syrian contacts with Turkish citizens.