Using five years of reported typhoid data with spatial analytical techniques, this study is the first to explore the relationships between socio-environmental variables and typhoid occurrences in DMA. In the absence of regular surveillance, findings from this study in DMA not only provide insight about spatial-temporal patterns of typhoid but also suggested the socio-environmental factors associated with the disease.
Typhoid disease is very common in South Asia owing to the fact that this is one of the most impoverished regions of the world where poverty is consistently rising and a larger portion of population is lacking potable water and safe sanitation. A temporal epidemic curve reveals that yearly typhoid incidence rate was 8–11 persons per 100,000 people with the peak incidence rate in the period under consideration occurring in 2007/8. Monthly records demonstrated that almost half of the reported cases had occurred during the monsoon (July-October), indicating a distinct seasonal pattern. This finding supports an earlier clinical-based study conducted in the same area . Environmental factors are known to have impact on the distribution and transmission of typhoid in other endemic settings. Rainfall for instance, substantially affected the occurrence of typhoid by increasing the faecal contamination in the water supply in Pakistan , and the transmission of typhoid bacterium is to some extent influenced by rainfall, particularly in low lying areas where people rely on surface water for their daily needs, including drinking and domestic purposes . When natural runoff drains and transports rubbish, including human wastes to the surrounding water bodies during the monsoon, surface water becomes heavily contaminated, resulting in a higher number of cases of typhoid . Since water logging and flooding become pervasive during the monsoon in DMA, contamination of surface water  and tube wells  by flooding are likely to result in a peak incidence at that time. Furthermore, flooding, either natural or caused by human modification of the land surface could lead to the occurrence of typhoid , particularly in many wet locations like DMA. Not all the census tracts in the study area are equally susceptible to typhoid infection; generally areas with higher population density and inadequate provision of health infrastructure suffer from higher cases of typhoid infection, corroborating the results of an earlier study by Naheed et al. .
The spatial pattern of typhoid incidence indicated significant variation of the disease distribution in DMA (Figure 3). A close visual inspection of the incidence map suggested that census districts closer to large water bodies (e.g. river networks and lakes) are highly vulnerable to elevated incidence rate. This finding can be explained by the fact that both surface and groundwater water quality get severely degraded due to increased anthropogenic activities in DMA, which may have significant impact on the transmission and distribution of typhoid. In addition, low income people in the study area use surface water for cooking, bathing and other purposes. Consequently, a reasonable assumption is that contamination of these water bodies could directly influence the disease dynamics in the communities which is in agreement with a study conducted in Indonesia . As Salmonella bacteria can survive in water for days , contaminated surface water such as sewage, freshwater and groundwater could act as etiological agents of typhoid . It was generally observed that communities living in the proximity of the rivers Buriganga, Turag, and Balu had an elevated risk of typhoid compared with communities in other locations. These three rivers have found to have extreme pollution loads throughout the year in terms of coliform counts and other physio-chemical parameters [78–80], hence the probability of increasing of the disease burden is warranted. Also, risk factors investigations for typhoid have substantiated that all sources of drinking water, including piped water is highly contaminated in Dhaka [15, 19]. This accords with a study in Tajikistan  where contamination of piped water was found to have significant association with the occurrence of typhoid. These studies indicated that contaminated surface and piped water in DMA could amplify the likelihood of water borne infection among people living in that area. The transmission dynamics of typhoid in relation to water quality therefore remains a very promising area to explore.
A number of environmental, socioeconomic and demographic variables were combined through Principal Axis Factoring to classify each census tract according to three principal factors (e.g. environmental, economic and crowdedness), and to use the resulting score for risk area identification. The results demonstrated that QOL could serve as an important indicator as it was able to explain 73% variance in the model as an independent factor. This finding is in agreement with Khormi and Kumar  who found that neighbourhood quality provided the highest coefficient of determination in explaining the incidence of dengue disease in Saudi Arabia. Out of three factors extracted, factor 3 (e.g., so-called crowdedness index) had the highest coefficient of determination (r2=.63) followed by factor 2 (r2=.53) and factor 1 (r2=.60) based on individual GWR analyses, implying that population density, large households size and housing density have substantial impact on typhoid incidence. The study statistically substantiates the concept that areas with low risk of typhoid have a low mean population density (49069/km2), those with medium risk had a medium mean population density (633387/km2) while high risk areas had the highest mean population density (67464/km2). Similarly, literacy rate, water sources, unemployed population, percentage of slum area, sanitary facilities were higher in low risk areas than that of medium and high risk areas, illustrating the effect that socioeconomic status, water sources and sanitary facilities have on typhoid distribution in DMA . Crowdedness is regarded a sign of depressed socioeconomic conditions that facilitate person to person transmission  by sharing the same plate for food , cups and mugs for drinking, by being in contact with the infected person  or by residing in the same place . In addition, lack of education could put individuals at high risk as it is often related to poverty, poor housing condition, inadequate provision of safe sanitation and unemployment [8, 12, 18, 83]. We have also found that of the areas at high risk areas, 72.73% had low QOL, 18.19% medium QOL and 9.08% presented high QOL. Thus, it may be assumed that unplanned urbanization, higher population density, lack of critical urban infrastructures, particularly in DMA, have a considerable impact on the transmission and distribution of typhoid fever. While an advantage of the Principal Axis Factoring is that it reduces the complexity of correlated data and allows combining diverse data into fewer factors, a potential problem however is that it could lead to the loss of information through generalization  and a loss of direct causal relationships to raw predictor variables.
Spatial relationships were determined through global and local models, and the study recognized the efficacy of the GWR model to provide useful information about geographical heterogeneity. The GWR performed much better because the global model assumes the relationship between explanatory and dependent variables are consistent, and provides an average state of the phenomena being studied. The local model on the other hand, assumes the relationships are non-stationary. Since AICc is an effective way of comparing two models , the considerable difference in that measure implied an important improvement in the model fit . The results of r2 and AICc indicated GWR was a better model to predict typhoid risk in DMA.
Spatial statistics is gaining renewed interest as a means to attribute disease association and risk. Even though GWR has long been used in various studies including public health, crime and demography [86–89], there are some limitations of the model. One of such problems is the choice of appropriate kernel type and bandwidth to which the model is sensitive . Another notable problem is that the non-linear terms cannot be added to GWR models .
This study has a few limitations. First of all, the disease data that were acquired from hospitals may have underestimated or overestimated typhoid records. Because the data were historical records and documented from the record room of each hospital, we had no valid method to ascertain repeated hospitalizations of an individual patient. In addition, hospital-based surveillance may underestimate actual population at risk because only severely sick people tend to get admitted for treatment. Secondly, we only consider 11 major health service providers, the majority of which were public hospitals. The study could be improved by including data from private clinics where most of the affluent people seek health services. On balance, we believe that we have an underestimate of the occurrence. We do not believe that this affects the validity of our results since we have been able to develop a predictive model using what is effectively a sample of unknown size drawn from the true population of occurrences. Thirdly, we also could not separate cases into typhoid and paratyphoid groups. Isolation of these two types would allow us to estimate the disease dynamics and identify the most prevalent typhoid types in DMA. The etiology of the two diseases is similar but the morbidity rates are not. Again we believe this does not affect the validity of our results since we are dealing with disease occurrence, not disease outcome. Fourthly, a new method is needed to overcome the problems associated with GWR such as mixed geographically weighted regression proposed by Mei et al. . Finally, water source and sanitation data of each census tract could greatly improve future study since these variables are known to have considerable impact on the occurrence of typhoid.
Despite the limitations listed above, the major strength of this study is the derivation of the first regional risk map of typhoid infection which rigorously investigated a fine-scale spatial distribution of typhoid and its socio-environmental determinants. Moreover, the study determined that QOL could be an important indicator in identifying populations at risk of typhoid in a rapidly urbanizing megacity where high quality data is lacking. Although vaccination is available to prevent typhoid infection, it cannot be an alternative to sound environmental health infrastructures . Furthermore, DMA is likely to encounter rapid urban growth and more intense rainfall, driven by climatic change, in the coming years. These changes may put more people at risk of typhoid. Therefore, this study underscores the necessity of appropriate policies as well as critical public health infrastructures to curb the future spread of water borne diseases.