Epidemiology of seasonal influenza in the Middle East and North Africa regions, 2010‐2016: Circulating influenza A and B viruses and spatial timing of epidemics

Background There is a limited knowledge regarding the epidemiology of influenza in Middle East and North Africa. Objectives We described the patterns of influenza circulation and the timing of seasonal epidemics in countries of Middle East and North Africa. Methods We used virological surveillance data for 2010‐2016 from the WHO FluNet database. In each country, we calculated the median proportion of cases that were caused by each virus type and subtype; determined the timing and amplitude of the primary and secondary peaks; and used linear regression models to test for spatial trends in the timing of epidemics. Results We included 70 532 influenza cases from seventeen countries. Influenza A and B accounted for a median 76.5% and 23.5% of cases in a season and were the dominant type in 86.8% and 13.2% of seasons. The proportion of influenza A cases that were subtyped was 85.9%, while only 4.4% of influenza B cases were characterized. For most countries, influenza seasonality was similar to the Northern Hemisphere, with a single large peak between January and March; exceptions were the countries in the Arabian Peninsula and Jordan, all of which showed clear secondary peaks, and some countries had an earlier primary peak (in November‐December in Bahrain and Qatar). The direction of the timing of influenza activity was east to west and south to north in 2012‐2013 and 2015‐2016, and west to east in 2014‐2015. Conclusions The epidemiology of influenza is generally uniform in countries of Middle East and North Africa, with influenza B playing an important role in the seasonal disease burden.


| INTRODUC TI ON
The epidemiology of seasonal influenza is well defined in many regions of the world, particularly in developed countries of the Northern and Southern Hemispheres. In other regions of the world, much less is known about the epidemiology of influenza A and B, especially in the tropics and subtropics 1 . The number of publications related to the epidemiology of influenza in low-income countries 2-4 , including regional [5][6][7] and global analyses [8][9][10] , has been growing in recent years. This information is important as it can be used by public health authorities and national healthcare systems to define and manage country-specific prevention and control programmes for seasonal influenza (eg to define the optimal time to launch a vaccination campaign).
This study evaluates the epidemiology of seasonal influenza A and B in the Middle East and North Africa. These regions separate the tropical countries of Africa and Asia in the south from Europe, Russia and Central Asia in the North, and have a predominantly subtropical climate. Currently, annual vaccination campaigns focusing on the elderly and specific risk groups (including children, healthcare workers and Hajj pilgrims) are being recommended in most countries of the Middle East and North Africa 11 ; however, the number of doses distributed per 1000 population is still low and more efforts are needed to increase the vaccine uptake in this region 12 . A better knowledge of influenza epidemiology and patterns of spread will provide further support to these campaigns.
Several reports have appeared in recent years that have described the epidemiology of influenza and disease burden in single countries of this region [13][14][15][16][17] ; some others have tried to assess the impact of the annual Hajj pilgrimage to Saudi Arabia on the transmission of viral and bacterial respiratory infections, including influenza 18,19 .
However, no unified, comparative approach has been attempted to study the epidemiology of seasonal influenza in the Middle East and North Africa. The aim of our study was to compare countries in this region in terms of the circulating influenza A and B viruses and to assess the spatial timing of seasonal epidemics in this area.

| MATERIAL S AND ME THODS
Our analysis focused on a large world region encompassing twentyseven countries in the Middle East and North Africa (Figure 1).
These countries cover a combined population of nearly 800 million inhabitants. A preliminary check of the FluNet database revealed that most countries in North Africa and the Middle East had no or very little influenza surveillance data before the 2009 pandemic. Therefore, we F I G U R E 1 Countries included in the analysis (in red) and countries that were not included because of lack of data (in grey). Countries were marked using the two-letter ISO codes. AF, Afghanistan; AM, Armenia; BH, Bahrain; DZ, Algeria; EG, Egypt; GE, Georgia; IL, Israel; IQ, Iraq; IR, Iran; JO, Jordan; LB, Lebanon; MA, Morocco; OM, Oman; PK, Pakistan; QA, Qatar; TN, Tunisia; TR, Turkey opted to limit the study period to the post-pandemic period, in order to ensure greater comparability of information between countries.

| Source of data
Influenza epidemics usually occur during winter months and can bridge consecutive calendar years in the Northern Hemisphere. Therefore, we used as unit of analysis the "country season," defined as the period between 1 July of 1 year and 30 June of next year.

| Statistical analysis
For each country, the statistical analysis was limited to seasons with 50 or more influenza reported cases and at least 20 weeks of data reporting in order to enhance the reliability of results 9 . We determined the proportion of influenza cases that were caused by each virus type, subtype and lineage in each country and season, and determined the median values for each country (ie across seasons) and for the whole region (ie across all seasons and countries). Medians were compared between groups of countries using the Mann-Whitney nonparametric test.
We aimed to determine the typical timing and seasonality of the primary and secondary epidemic peaks in each country using the epipoi software 21 . The epipoi method first divides the weekly number of reported influenza cases by the maximum value in the same season: this is done to standardize all season-specific time series in each country, so that all seasons are given the same weight in subsequent analysis. It then models the trend and seasonality of each countryspecific influenza time series by summing annual, semi-annual and quarterly harmonics (obtained from Fourier decomposition): the curve generated in this way is named the periodic annual function (PAF) 21,22 . The timing of the primary peak refers to the month when influenza activity (as modelled by the PAF) reaches its maximum value; if the PAF is a bimodal curve, a secondary peak is also identified. The typical seasonality of influenza epidemics is quantified by calculating the amplitude of the primary and secondary peaks: this is defined as the ratio of the wave height (difference between the peak and the trough) in the PAF and the peak value, and it is an estimate of the degree to which influenza cases in each season tend to cluster around the typical peak timing extracted from the PAF (when the trough value in the model is below zero, the amplitude can exceed 100%). The peak and amplitude of the primary and secondary peaks were calculated for all influenza cases and separately for influenza A and B epidemics (for the latter analysis, for each country we only included seasons with 50 or more influenza cases caused by that virus type). For more details on how the PAF is fitted to the raw data and the timing and amplitude of the peaks are calculated using the epipoi software, the reader can refer to the paper by Alonso and McCormick where the software is presented 21 .
Finally, we aimed to detect whether seasonal influenza epidemics spread along geographical gradients in countries of the Middle East and North Africa. For this purpose, we fitted a linear regression model for each season using the longitude and latitude of the country's centroid as independent variables, and the timing of the typical (ie obtained from the PAF) primary epidemic peak (with weeks numbered consecutively from July of 1 year to June of next year) as the dependent variable 23 . A similar analysis was conducted in parallel also for Europe and Central Asia (ie countries included in the south-west Europe, northern Europe, eastern Europe and Central Asia Influenza Transmission Zones 24 ) to test whether the spatial timing of influenza epidemics in the same season is comparable in these two world areas (the Middle East and North Africa vs Europe and Central Asia).
Analyses were conducted using Stata version 14 (Stata Corp, College Station, TX) and epipoi 21 . Maps were obtained using freely available software (http://mapchart.net/). All tests were two-sided and considered as statistically significant when the P-value was .05 or less.  Table S1.

| RE SULTS
Influenza A and B accounted for a median 76.5% (IQR 60.4%-90.3%) and 23.5% (IQR 9.7%-39.6%) of all influenza cases in a season during the study period, respectively (Table 1). More specifically, influenza B accounted for fewer than 20% of reported influenza cases in 34 of the 83 seasons (41.0%) between 20% and 50% in 38 seasons (45.8%), between 50% and 80% in 10 seasons (12.0%) and above 80% in one season (1.2%). There were slight, non-significant differences (P-value = .96) in the median proportion of influenza cases that were caused by type B

| Timing of influenza epidemics
The inspection of the typical timing and amplitude of the primary and secondary peaks of influenza epidemics (Table 2 and Figure 2; the standardized time series and PAF for each country included in the analysis are available in the Figure S1) revealed the following prominent characteristics: • All countries except those situated in the Arabian peninsula had well-defined primary peaks taking place typically in January-March, and much smaller secondary peaks placed either in spring/ summer (May to July) or in autumn (September to November), with the sole exception being Jordan (whose secondary peak was typically in April and had an amplitude only slightly less than that of the primary peak);

| Spatial timing of influenza epidemics
The analysis of spatial spread of epidemics (Table 3)          In conclusion, we found that the epidemiology of influenza is substantially uniform in countries of the Middle East and North

| D ISCUSS I ON
Africa. In particular, the timing of influenza epidemics does not differ significantly between countries, with the partial exception of a couple of countries in the Arabian Peninsula. Influenza prevention may not be a priority at present in all countries in this area because of wars and persistent political instability. However, countries of the Middle East and North Africa are geographically contiguous and culturally interconnected, so it would be desirable that, as far as possible, influenza surveillance, preparedness and prevention activities are coordinated at a supranational level (eg regional) and implemented without disparities. According to WHO 11 , fourteen of the seventeen countries that we analysed implement seasonal influenza vaccination campaign: our findings confirm that the optimal time to vaccinate is similar to Europe and North America, that is the prewinter Northern Hemisphere months. The scientific evidence resulting from our regional analysis complements the one obtained from individual countries [13][14][15][16][17] ; our analysis could be further enhanced by strengthening influenza surveillance networks in this region and expanding the FluNet surveillance system to neighbouring countries which currently do not report.

ACK N OWLED G EM ENTS
We would like to thank WHO and laboratories reporting their influenza surveillance data to the FluNet database.