Mapping mHealth (mobile health) and mobile penetrations in sub-Saharan Africa for strategic regional collaboration in mHealth scale-up: an application of exploratory spatial data analysis

Sub-Saharan Africa is the region south of the Sahara desert and is located on the African continent. Known for its heavy burden of communicable, maternal, neonatal and nutritional diseases, sub-Saharan Africa has been suffering from limited access to care and poor health status compared to other continents [18]. In terms of mHealth interventions, sub-Saharan Africa has been considered as a region where applicability and potential of mHealth is promising with growing wireless network coverage and high mobile phone subscription rates [3, 19]. In 2015, the region’s average mobile cellular subscription rate was 82.9 per 100 people [20].

Table 1 illustrates socioeconomic and geographic information for all 48 sub-Saharan African countries. According to the World Bank and the United Nations Statistics Division, there are 48 countries located in the sub-Saharan African region, including 17 countries in the eastern Africa (Burundi, Comoros, Eritrea, Ethiopia, Kenya, Madagascar, Mozambique, Mauritius, Malawi, Rwanda, Somalia, South Sudan, Seychelles, Tanzania, Uganda, Zambia, Zimbabwe), 16 in western Africa (Benin, Burkina Faso, Cote d’Ivoire, Cabo Verde, Ghana, Guinea, the Gambia, Guinea-Bissau, Liberia, Mali, Mauritania, Niger, Nigeria, Senegal, Sierra Leone and Togo), 9 in middle Africa (Angola, Central African Republic, Cameroon, Democratic Republic of the Congo, Republic of the Congo, Gabon, Equatorial Guinea, Sao Tomoe and Principe and Chad), 5 in southern Africa (Botswana, Lesotho, Namibia, Swaziland and South Africa) and 1 in northern Africa (Sudan) [21, 22]. Among these, 6 countries (Comoros, Cabo Verde, Madagascar, Mauritius, Sao Tome and Principe and Seychelles) are island nations adjacent to the Atlantic and Indian Ocean.

Currently, the World Bank defines a low income country whose GNI (gross national income) per capita was $1025 or less in 2015; 90% of the world’s low income countries are located in sub-Saharan Africa [22]. In fact, there is only 1 high income country in this region (Seychelles). The eastern region has 1 high income country, 1 upper-middle income country, 2 lower-middle income countries and 13 low income countries. The western region has 5 lower-middle income countries and 11 low income countries. In the middle region, there are 3 upper-middle income countries, 3 lower-middle income countries and 3 low income countries. In southern region, 3 countries belong to upper-middle income group and 2 are lower-middle income countries. Sudan, in the north, is a lower-middle income country. The average GNI per capita for each region (current US$) in 2015 was: Eastern region− 2295.3, western region− 1055, middle region − 3148.9, South − 4484 and North − 2000.

The average population size is not too different between the east and the west 24,536,442 and 22,038,114, respectively; however, population gaps exist between the eastern Africa countries and the western Africa countries. For example, the average healthy life expectancy at birth in 2015 was 52.8 years (STD 4.6, Min 44.4, Max 64.2) for the western Africa countries whereas it was 54.9 years (STD 5, Min 47.8, Max 66.8) among the eastern Africa countries [23]. Also, the average adult literacy rate in 2015 was 50.8% (STD 17.6, Min 19.1, Max 88.5) in the west while it was 73.1% (STD 16.3, Min 32, Max 95.3) in the east [23, 24].

For the systematic data collection on the number of mHealth programs implemented in each sub-Saharan Africa country between 2006 and 2016, the following data sources were used: The WHO eHealth database, the United States Agency for International Development (USAID) mHealth database and the mHealth Working Group Inventory of Projects (Johns Hopkins University) were accessed. The variety of databases provides a comprehensive list of mHealth programs with project descriptions and basic information. To increase sensitivity, non-mHealth or non-eHealth specific databases were also accessed including the World Bank Projects & Operations database, the African Development Bank (AfDB) Projects & Operations database, the UK Department for International Development (DFID) Development Tracker, the Canadian International Development Agency (CIDA) and the Center for Health Market Innovation (CHMI) database. Since these databases contain mHealth projects as well as other international development projects, several search terms were used to secure specificity. Examples of search terms include mHealth, mobile health, mobile phone, cell phone, cellular phone, smart phone, mobile device, wearable device, tablet, PDA (personal digital assistant), SMS, Multimedia Message Service (MMS), text message, phone call and email.

From the above mentioned data sources, a comprehensive dataset was created after selecting mHealth projects based on inclusion/exclusion criteria. To be included in the study, mHealth projects must have been implemented between 2006 and 2016 in one of the sub-Saharan Africa countries. Other eHealth projects that do not explicitly involve the use of mobile devices were excluded. Geographic Information System (GIS) data was obtained from the GADM database (Global Administrative Areas), which is an open-source repository of spatial data developed by researchers at the University of California at Davis, the University of California at Berkeley and others [25]. Historical data on the mobile cellular subscriptions per 100 population and other country characteristics were obtained from publicly available data on the World Bank and WHO websites [23, 24].

This study attempted to investigate if the number of mHealth programs implemented in a country is spatially correlated with that in its neighboring countries, using exploratory spatial data analysis (ESDA). The same process has been performed for the average mobile cellular subscriptions per 100 population between 2006 and 2015. The average mobile subscription rate during the past decade was used for the analysis because it represents not only the current situation but the overall mobile network coverage trend for each country. To ensure that the average mobile subscriptions per 100 population between 2006 and 2015 were not statistically different from the current situation for all 48 sub-Saharan Africa countries, a Friedman test was performed between the rank order measures of the average mobile subscription rate per 100 population in 2006 and that in the year 2015. The Friedman chi-square value was 0.19 with a p-value of 0.67, thus the two ranks were not statistically different.

In conducting an exploratory spatial data analysis, it is possible to identify a local clustering within the study region. For the purposes of our study, two measures of spatial autocorrelation were investigated: Moran’s I and Local Indicator of Spatial Association (LISA). Moran’s I is used to measure global spatial autocorrelation by looking at the correlation between the variable of interest y (in this case, the total number of mHealth programs or mobile cellular subscriptions per 100 people) and the spatial lag of that variable y [26, 27]. Spatial lag of a variable y is derived from the average value of y for all the neighboring locations. The slope of the least squares linear regression line between y and lag-y is referred to as Moran’s I. The decision on whether a location j is a neighbor of a location i or not is introduced in the data by a weights matrix w _ij (i ≠ j) [12]. To decide what constitutes as a neighbor, a weights matrix had to be defined. There are several ways to designate spatial weights matrix based on contiguity or distance. This study adopted a weights matrix using queen contiguity where countries were considered neighbors when they shared a common boundary or a point [28]. w _ij was equal to 1 if country i and country j were neighbors and 0 if they were not neighbors according to the queen’s contiguity rule. Choosing a weights matrix scheme is often arbitrary and determined by the research topic and outcome of interest. This weights matrix, queen contiguity, was chosen for this study based on the literature review on exploratory spatial data analysis for spatial trends of incidence [26, 29‐31]. The Moran’s I generally ranges from −1 to 1, with 1 having a positive spatial autocorrelation and −1 having a negative spatial autocorrelation. If Moran’s I is 0 then there is no spatial autocorrelation, indicating spatial randomness.

LISA was evaluated to find out the spatial clustering for each location [29]. While Moran’s I is a global measure of spatial autocorrelation, LISA is a local Moran’s I and therefore, the sum of the LISA is proportional to a global indicator, Moran’s I [32]. However, Moran’s I has a major limitation in that it cannot identify meaningful clusters or spatial patterns at a local scale. On the contrary, LISA enabled us to identify hot spots where the variable of interest y _i (the total number of mHealth programs or mobile cellular subscriptions per 100 people) has higher value than average for the entire study region \( \overline{y} \) and where the same held true for its neighbors [30]. This is known as a high-high association or a hot spot. Similarly, in the same manner, cold spots or low-low associations were able to be identified using LISA. The proximity of the locations i and j is also represented by weights matrix w _ij defined by the queen contiguity rule (i ≠ j) [12]. ArcGIS software V10.5 was used for statistical analysis and mapping.

From our systematic data collection, the total unique number of mHealth programs implemented in sub-Saharan Africa between 2006 and 2016 was 487 (any duplicates were removed and any program that was implemented in multiple countries was counted separately for each country). Of these mHealth programs counted, the eastern and western regions had 287 and 145 for the past decade, respectively. More specifically, Kenya, Uganda and Tanzania in the eastern region were with the most mHealth programs implemented between 2006 and 2016 (71, 54, and 50, respectively).

Global Moran’s I for mHealth programs in sub-Saharan Africa was 0.16 with a pseudo p-value = 0.07, indicating an insignificant level of spatial autocorrelation at the global level. Figure 1a and b illustrate the choropleth map on the count of mHealth programs in each country and the corresponding statistically significant LISA clusters. For the LISA analysis, the significance level was filtered at a pseudo p-value = 0.05 and 999 permutations were performed for sensitivity analysis.

Despite relatively low levels of global Moran’s I, the analysis of LISA clusters allowed us to detect hot spots or high-high associations where a certain high value in a country correlated with a high value in its neighboring countries. Likewise, cold spots (low-low association), low-high associations and high-low associations were identified by LISA. As for mHealth programs, hot spots were identified along the eastern coastline, Kenya, Tanzania, Malawi and Uganda. South Sudan, Somalia and Rwanda are identified as low-high associations particularly because they had relatively few mHealth programs implemented but are contiguous to Uganda, Kenya or Tanzania where large numbers of mHealth programs had been identified. The Central African Republic was considered a low-low association due to limited experience with mHealth by both its neighboring countries and itself.

Global Moran’s I for average mobile subscription rates per 100 population between 2006 and 2015 is 0.29 with a pseudo p-value = 0.01, indicating the presence of geographic clustering to some extent. Figure 2a and b present the choropleth map of average mobile subscriptions per 100 population between 2006 and 2015 and the corresponding LISA clusters. The same significance filter and sensitivity analysis were applied as the LISA cluster analysis for mHealth programs. Average mobile penetration rates per 100 population were greater than 100 in Seychelles, Gabon, Botswana, South Africa and Mauritius for the past decade. In contrast, some of the countries with the lowest average mobile subscription rates had less than 50 subscriptions per 100 people during 2006 and 2016.

LISA analysis for average mobile subscriptions per 100 population between 2006 and 2015 identified extensive cold spots across northeastern and central parts of the sub-Saharan Africa (Sudan, South Sudan, Ethiopia, Democratic Republic of the Congo and Tanzania). In addition, 1 high-low association (Kenya) and 1 low-high association (Equatorial Guinea) were identified for average mobile subscription rates per 100 population. In the southern area, Namibia, Botswana, Zimbabwe and South Africa were identified as having had high-high associations where mobile subscriptions were high for themselves as well as for their neighbors.

Although it is well known that using mobile technology in health service delivery can be effective in resource-limited settings such as sub-Saharan Africa, challenges still remain in terms of scale-up efforts [4]. In addition, setting up reliable technology and infrastructure is essential to ensure sustainability of mHealth programs. Since sub-Saharan Africa is the region where investment in mHealth has been rapidly growing, it is the appropriate time to establish regional collaboration strategies to facilitate a synergy effect. To that end, this study attempted to lay a groundwork by investigating spatial distribution on the number of mHealth programs and mobile subscription rate. Specifically, the study identified geographic areas with relatively more/less mHealth programs or higher/lower mobile cellular penetration rates.

The main findings of our analysis suggest the following: Firstly, countries in the east of sub-Saharan Africa have a comparative advantage with more experiences in mHealth implementation than their counterparts, specifically for hot spot countries such as Kenya, Tanzania, Malawi and Uganda. These hot spots imply that they have the potential to take the lead role in moving towards a collaborative mHealth approach for scale-up in sub-Saharan Africa. Furthermore, low-high association countries such as South Sudan, Somalia and Rwanda can take advantage of their geographic location by closely collaborating with hot spot neighboring countries that can share knowledge and experience. For the low-low association case of Central African Republic, the inexperience in mHealth for itself and its neighboring countries coincides with overall underdevelopment in landlocked countries across the continent [9]. In fact, landlocked locations have long been blamed for their contribution to high transaction costs and less opportunities for economic growth [10].

Secondly, middle and northeastern areas of sub-Saharan Africa have been cold spots for mobile cellular penetration rates for the past decade. This phenomenon resonates with the disadvantages, mentioned earlier, attributed to the landlocked locations. Among the five countries identified as cold spots for mobile cellular subscription rates, South Sudan and Ethiopia are landlocked countries and the Democratic Republic of the Congo is also mostly landlocked. Our findings are in line with other research that pointed out the lower level of expansion in mobile coverage for the landlocked countries in the central and western region of Africa [33]. Kenya is identified as a high-low country that shows high mobile penetration rates but is surrounded by countries with low mobile penetration rates. Interestingly, Tanzania is one of the countries with the largest numbers of mHealth programs implemented between 2006 and 2016, but it is a cold spot for its average mobile subscription rate between 2006 and 2015. Tanzania makes a good case for further analysis on how it dealt with relatively low mobile subscription rates when implementing such a large number of mHealth programs. Hot spot countries in the south (Namibia, Botswana, Zimbabwe and South Africa) can lead the discussion on the regional collaboration strategies in improving mobile network coverage for effective mHealth scale-up.

Some of the limitations of this study include the following. Firstly, the number of mHealth programs is aggregated to the country level, making it difficult to identify a detailed regional distribution across the border. As for the mobile subscription rate per 100 population data, it may overestimate or underestimate the number of actual users of mobile phones because ownership of multiple subscriber identity module (SIM) cards by an individual is very common while sharing mobile phones among friends or family is also common in Africa [33]. Future spatial analysis on mHealth and mobile subscriptions can address this issue by looking at the data on a higher level of granularity. Secondly, there can be a risk of publication bias in that the mHealth project information data available online may not represent all mHealth implementation efforts within a country. To minimize this possibility, the study combined official data repositories from various organizations. Finally, this study can be further expanded for future research that takes into account for the temporal component of the data to investigate the trend in mHealth growth and the change in mobile technology penetration rates.

Despite the limitations, our study was the first to examine spatial heterogeneity in mHealth implementation effort along with the mobile subscription rates within the sub-Saharan Africa region. Identifying regional inequity in public health interventions such as mHealth can play a key role in informing stakeholders for strategic regional cooperation in terms of future resource allocation decision and new investment opportunities. In particular, the LISA analysis from our study demonstrates that the spatial differences are not randomly distributed but have some spatial pattern within the sub-Saharan Africa region. As the result of our study indicates, areas along the eastern coastline with much experience in mHealth can be a regional hub for collaboration whereas the northeastern and central parts of sub-Saharan Africa need more attention for improving access to mobile phones.

To go one step further, more research should be conducted on developing specific strategies for regional collaboration in mHealth and enhanced mobile penetration. In particular, there are several clusters of countries in sub-Saharan Africa that share common languages or common mobile network providers, which can be beneficial in regional collaboration for scaling up mHealth. Stakeholders from hot spot countries should actively participate in the discussion for plan of action. In addition, a root cause analysis should be performed on why cold spot countries lag behind in terms of mHealth implementation and access to mobile phones. In scaling up mHealth and tackling disparities in mobile penetration rates, regional collaboration or regional networks for mHealth and mobile technology can contribute to creating opportunities for exchanging hands-on knowledge and lessons learned among countries with different levels of experience.