Geographical distribution of risk factors for invasive non-typhoidal Salmonella at the subnational boundary level in sub-Saharan Africa

Samonella enterica, a single bacterial species, is a major public health concern in many parts of the world. In terms of human disease-causing serovars, Samonellae are divided into typhoidal and non-typhoidal groups: (1) Salmonella enterica serovars Typhi, Paratyphi A, B, and C are often collectively considered as typhoidal Salmonella, (2) other serovars including Salmonella enterica serovar Typhimurium and Salmonella enterica serovar Enteritidis are grouped as nontyphoidal Salmonella (NTS) [1, 2]. Typhoidal Salmonella strains are human-restricted organisms that cause enteric fever. While enteric fever is generally found in immunocompetent children and adults who are exposed to environmental or behavioral risk factors, invasive non-typhoidal Salmonella (iNTS) disease is more associated with differential host susceptibility to invasive disease rather than environmental factors [2, 3].

Globally, population-based surveillance data for iNTS including disease burden estimates, are scarce compared to that for typhoid and paratyphoid. In 2010, Ao et al. estimated the global incidence and deaths of iNTS based on a systematic literature review and extrapolation using key risk factors which are associated with the disease [4]. The highest incidence rate was observed in Africa followed by Europe where a high number of cases were mostly driven by Russia, Ukraine, and Estonia. The World Health Organization (WHO) and Kirk et al. also updated the global burden of iNTS by calculating disability-adjusted life years (DALYs) [5, 6]. Based on these studies, the geographical distribution of the global iNTS burden was roughly demonstrated [7]. Similar to Ao et al., the global burden of iNTS expressed by DALYs was greatest in Africa. More recently, the Institute for Health Metrics and Evaluation (IHME) included the global burden of iNTS in its cause list for the first time and reported the highest incidence in sub-Saharan Africa as well [8]. The IHME study also indicated that compared to the two previous studies for the same year, their estimates were slightly higher than those by WHO [5] and significantly lower than those by Ao et al. [4]. While all previous efforts were helpful to understand the geographical variation of the disease, it should be noted that the source data underlying the models were sparse, which resulted in the extrapolation of limited surveillance data to neighboring countries or regions [4, 6‐8]. This clearly indicates that more population-based surveillance data are urgently needed to improve our understanding of iNTS endemicity.

The current study attempts to identify potential high risk areas for iNTS at the subnational-level and to facilitate the site selection process for surveillance and vaccine studies. Because iNTS disease burden data are limited, the current study proposes the use of risk factors that are strongly associated with iNTS to identify high risk areas. Differential susceptibility of the host plays a major role in risk for iNTS disease [2]. As evidenced by previous studies [1, 3, 4, 9], the prevalence of iNTS is closely associated with the prevalence of malaria, HIV, and malnutrition. In addition, while it is debatable [10], several studies indicated that contaminated water sources could be partly associated with NTS as an environmental risk factor [11, 12]. Risk factors related to a disease are often available in a more standardized way at the country-level or subnational-level [13]. Incidence rates from a specific surveillance site in a country are frequently assumed to be representative of the whole country when estimating the burden of a disease or cost-effectiveness analyses. This approach is unavoidable due to the lack of population-based surveillance studies which require extensive human and financial resources. The use of standardized risk factors which are available at the subnational-level enables proper comparison of risk levels across countries or subnational boundaries in a more consistent manner. Understanding the endemicity of iNTS based on risk factors covers broader geographical areas where surveillance data are not yet available, and help researchers design future studies to examine incidence rates.

Previous surveillance studies are summarized in Table 1. Lunguya et al. collected the number of iNTS episodes in seven provinces in the Democratic Republic of Congo, and the proportion of iNTS (/1000) ranged from 0 (Kasai-Occidental) to 72 (Equateur) [20]. Marks et al. carried out surveillance studies in a multi-country setting [21], and three countries, where corresponding DHS datasets were available, were selected in the current study. The proportion of iNTS varied across the three sites with Asante Akim North in Ghana being the highest and Pikine in Senegal being the lowest. An additional surveillance study was conducted by Tapia et al. in Mali [22].

Figure 2 shows how the number of clusters was selected to reflect improved locality around each of the existing study sites. Each cluster included 17 to 34 households where the number of household members in each household ranged from 1 to 49. The density of DHS clusters within a country varies depended on the size of a country and other logistical issues. For example, the dispersion of DHS clusters is relatively lower in Burkina Faso compared to the dispersion in the Democratic Republic of Congo as shown in Fig. 2. The iNRF indices with an equal weight, as well as with relative weights were first constructed based on the four risk factors distributed within the radius per study site.

Correlations between the proportion of iNTS and the iNRF index with an equal weight were plotted in Fig. 3 (see Additional file 2 for the iNRF index with relative weights). A previous study indicated that young children were more vulnerable to Salmonella Typhimurium or Salmonella Enteritidis [20]. Given that the reported iNTS episodes were for younger age cohorts in Ghana and in Mali compared to other sites, these two studies were separately compared. Overall, the proportion of iNTS and the iNRF index with an equal weight was highly correlated showing the correlation coefficient of 0.821 at the 1% Bonferroni-adjusted significance level (p = 0.002). Nonetheless, the relationship between iNTS and the iNRF index in Kasai-Occidental appeared to be less consistent compared to that in other sites. The correlation coefficient with the iNRF index based on relative weights was 0.809 at the 1% Bonferroni-adjusted significance level (p = 0.003). Thus, the iNRF index with an equal weight was chosen for further analyses (simply named the iNRF index hereafter).

Sensitivity analyses were conducted to examine how extending a radius would affect the association between the iNRF index and the proportion of iNTS. Figure 4 shows the relationship between the proportion of iNTS and the iNRF index estimated with additional radii: 80 km and 100 km. It is clear to see that the wider the radius is, the weaker the correlation is. The correlation coefficients were estimated at 0.755 (p = 0.007) and 0.629 (p = 0.038) with 80 km and 100 km radii, respectively. This result was consistent with the presumption that risk factors distributed in adjacent areas to a study site would improve the accuracy of the association with the proportion of iNTS.

Given the association between the iNRF index and the proportion of iNTS episodes observed in the existing surveillance studies, the iNRF index was further estimated at the subnational boundary level for all 9 countries and shown in Fig. 5. At the country-level, Guinea appeared to be at the highest risk for iNTS followed by Cote d’Ivoire and Burkina Faso. On the other hand, Rwanda, Senegal, and Burundi were relatively at lower risk than other countries. At the provincial-level, the highest risk area was identified in Maniema, the Democratic Republic of Congo, whereas Dakar in Senegal was at the lowest risk. Supplementary table shows further details on individual index values and its percentiles at the subnational boundary level (see Additional file 1). The overall risk-level of iNTS was variable not only by country but also within a country. For example, while the iNRF index values were generally high (or low) across all provinces in Cote d’Ivoire (or Rwanda), the risk levels were variable by province in the Democratic Republic of Congo. The iNRF index values were low to moderate in some provinces such as Nord-Kivu, Sud-Kivu, and Kinshasa in the Democratic Republic of Congo, but the risk-level was much higher in Maniema and Ituri provinces in the same country.

The current study created a composite index called the iNRF index in order to identify potential high risk areas for iNTS. The iNRF index was generated based on four risk factors which were previously validated in literature, and the study outcome showed that the prevalence of iNTS was correlated to the iNRF index. In particular, the four risk factors used in this study were directly obtained from the survey data without imposing any unnecessary parametric assumptions for prediction. By using risk factors which were commonly available across countries, the risk-level of iNTS was estimated in a consistent manner at the subnational boundary level. The overall risk-level of iNTS was variable not only by country but also within a country, indicating that the actual burden of iNTS at the country-level would likely be different from a site-specific disease burden.

In the context of neglected tropical diseases where incidence rates are disproportionately high in developing countries, population-based surveillance data are scarce. In addition, data gathered in surveillance studies are mostly site-specific, e.g. within a single province of a country. Nonetheless, these site-specific surveillance data are often considered representative of an entire country and used for more complicated analyses such as global disease burden studies or cost-effective analyses. However, it is important to recognize that the burden of a disease is often variable not only between countries, but within a country. This variability within a country affects the overall disease burden estimate, as well as the cost-effectiveness of an intervention at the country-level.

iNTS has not yet been widely recognized compared to disease caused by S. Typhi or S. Paratyphi. Given the inadequacy of iNTS evidence, the current study constructed the iNRF index to understand the geographical distribution of iNTS risk factors in selected sub-Saharan African countries. While there is no doubt that population-based surveillance studies will provide more details on iNTS in the future, this kind of work is more resource and time intensive. Even if more surveillance studies became available, it would still be very difficult to obtain such information from all provinces within a country, especially in the developing-country setting. Thus, along with existing surveillance studies, the iNRF index presented in the current study can be used as a proxy to understand iNTS risk levels within a country and to support decision making for future iNTS research funding, clinical trial design/execution, and iNTS investment case development.

Some areas of uncertainty deserve attention. It should be noted that the 50 km radius used in this study is arbitrary. Nonetheless, the 50 km radius was chosen to ensure the minimum inclusion of DHS clusters in all countries given that the dispersion of DHS clusters was variable. In addition, the sensitivity analyses were conducted with varying radii and showed that the association between the iNRF index and site-specific iNTS estimates was more robust with a shorter radius than a longer radius. In general, while a longer radius would include more DHS clusters, it would weaken the precision to reflect the locality of study sites. On the other hand, a shorter radius would increase the precision of the estimate but reduce the number of cluster samples. Overall, the iNRF index was highly associated with the proportion of iNTS but did not correspond well with the observation in Kasai-Occidental. In Kasai-Occidental, there were only two blood cultures taken, making it difficult to generalize the outcome. Nonetheless, this may indicate that there might be additional risk factors which could explain the prevalence of iNTS. It should be noted that the number of adjusted iNTS episodes may have been estimated in a different way given logistical constraints. In fact, von Kalckreuth et al. indicated several challenges such as blood culture volume inadequacy or antimicrobial pretreatment which may have affected the limited diagnostic sensitivity of blood culture [23]. Identifying countries with a complete set of the four risk factors reduced the number of countries available for the current study. However, it was critical to include risk factors which had been frequently emphasized in existing literature (i.e., malaria and HIV) although these were not available in many of countries. On the other hand, focusing solely on risk factors which were more commonly available (i.e., drinking water source, or BMI) would increase the number of eligible countries, but these indicators alone were not particularly specific to iNTS but rather general, making the index less relevant to iNTS. This underscores the fact that it would be highly desirable for the DHS program to include all risk factors in more countries in the future. It is also worth noting that considering the current study aims at generating the index based on primary and standardized data across multiple countries, not all potential risk factors were included due to the absence of such data which met the inclusion standard. For example, sickle cell disease was also reported as a potential risk factor for iNTS [24] but not included due to the lack of available data. Nonetheless, having the four risk factors which met the strict inclusion criteria of the current study enabled us to generate robust outcomes to understand the endemicity of iNTS. Given the lack of existing surveillance data points and matching DHS datasets, it would be desirable to carry out more sophisticated statistical validation in the future as more population-based surveillance data and DHS countries with all four indicators will become available. Lastly, given the duration of a malaria episode is relatively short, the malaria indicator used in the current study might be less consistent with the actual level of burden.

Considering that a population-based surveillance study requires extensive human and financial resources, identifying potential high and low risk areas for iNTS prior to a study implementation would facilitate an appropriate site-selection process in the future. The iNRF index may also be used to adjust site-specific incidence rates when extrapolating them to other locations. It is worth noting that the accuracy of the global burden estimates can be further improved when employing risk factors directly obtainable at the subnational boundary level, rather than applying predicted values of risk factors with a set of assumptions.

Population-based surveillance studies for iNTS are limited in number, and more surveillance studies are urgently needed. The insufficient number of data hampers the accurate estimation of the burden of iNTS and makes it difficult to understand the geographical variation of the disease. Because conducting surveillance studies in all specific locations within a country will be highly challenging in a short-run, the iNRF index presented in the current study can be used to make standardized comparisons of iNTS risk levels at the subnational boundary level where there is no surveillance data available yet. Moreover, the use of the iNRF index summarizing the four risk factors can help decision makers interpret easily and monitor the overall progress over time as the DHS program periodically updates survey data. Further research is needed not only to increase population-based surveillance data, but also to make the risk factors of iNTS available in more eligible countries.