Geospatial analyses to identify clusters of adverse antenatal factors for targeted interventions

There are many established risk factors for adverse perinatal outcomes. These include late antenatal care, smoking during pregnancy, maternal infection [1], maternal hypertension [2], gestational diabetes [3] and social factors such as teen pregnancy [4] and lower socioeconomic status [5]. Late antenatal care and smoking during pregnancy are two important factors that are amenable to intervention.

Australia’s and UK’s NICE antenatal care guidelines recommend that the first antenatal visit should be before 10 weeks of gestation [6, 7]. Early first prenatal care is important for the detection of adverse pregnancy related outcomes and is vital for healthy perinatal outcomes in both mothers and infants [8–10]. The benefits include healthy birth weight [11–13], low risk of preterm delivery [12, 14] and lower neonatal mortality [12, 14]. Antenatal care programs not only monitor both maternal and foetal health, but also facilitate health promoting advice such as smoking cessation. Women of low socio-economic status [15, 16], younger women [15], primiparous women [17, 18] and indigenous women [19] are associated with late antenatal care.

Smoking during pregnancy is associated with a high risk of miscarriage, stillbirth and serious complications during delivery [20], all of which add to the overall health costs [21]. Younger women are more likely to smoke during pregnancy [22] and mothers of lower socio-economic status, who were primiparous and attended first antenatal care late were less likely to quit smoking during pregnancy [23].

Despite the negative health impacts of smoking during pregnancy and the health benefits of early first antenatal visit on both mother and unborn child, substantial proportions of women still smoke during pregnancy or have their first antenatal visit after 10 weeks gestation. For example, 20% of mothers who gave birth during 2002 to 2004 smoked during pregnancy in Western Australia [24] and in New South Wales (NSW), 41% of women had their first antenatal care late in their pregnancies [17]. Therefore, it is important for health care providers to ensure that both the care providers and the community are aware of the benefits of early prenatal care and the detrimental effects of smoking during pregnancy on maternal and infant health.

Although targeted preventative programs such as a home visiting program can improve maternal health-related behaviours during and after pregnancy [25, 26], it can be difficult to direct interventions to individual mothers at greatest need. The identification of small geographical areas with a high prevalence of poor maternal-related behaviour could allow for more targeted efforts at those most at risk.

The application of spatial statistics and geographic information system (GIS) to health outcomes is now being increasingly used to provide novel ways of examining disease patterns geographically [27–29]. SaTScan is a software for implementing spatial, temporal and space-time scan statistics that can be used to determine areas where an event of interest, for example, cancer incidence or preterm deliveries, appears inconsistent with the overall study area. This technique is not limited to existing administrative boundaries such as postal area, and can identify location of the clusters without a priori knowledge about their location and size [30].

To our knowledge, SaTScan has not been used for analysing geographical patterns of antenatal maternal-related risk factors. This study was undertaken to assess the usefulness of such geospatial methods in identifying communities at high risk of poor maternal-related behaviours, particularly, smoking during pregnancy and timing of the first antenatal visit, for which targeted interventions may be warranted and importantly feasible.

The study area was the south-western region of metropolitan Sydney, Australia. It is about 6382 square kilometres in area, consists of 15 local government areas (LGAs) and had a population of 1,460,847 in the 2011 Census [31]. The population in the 15 LGAs ranged from 32,423 to 187,766 persons, and the population density ranged from 17 to 6,349 persons per square kilometre. In the same census, Metropolitan Sydney covers about 12,368 square kilometres and had a population of 4,391,674 persons [31].

We identified four spatial clusters of maternal smoking during pregnancy and four spatial clusters of first antenatal visit occurring at or after 10 weeks of gestation. Following identification of the clusters, we were also able to characterise pregnant women within the clusters and compare those characteristics with women living outside of those clusters. For example, in the maternal smoking during pregnancy clusters, higher proportions of mothers, were aged less than 35 years, were born in Australia, had their first antenatal visit at or after 10 weeks of gestation and a lower proportion of mothers were primiparous. In the clusters of late first antenatal visit, a higher proportion of mothers lived in the most disadvantaged areas and a lower proportion of mothers were primiparous.

SaTScan is a widely used and accepted software for geospatial analytic technique to identify and describe geographic patterns of interest in the public health literature [38–41]. Our study used both spatial statistics and descriptive statistics to describe adverse maternal-related behaviours in the south western region of metropolitan Sydney. However, it should be noted here that, although we have demonstrated the usefulness of using SatScan to identify the spatial distribution of adverse risk factors, SaTScan can also be used to identify spatial distributions favourable risk factors or health outcomes, for example, areas with low rates of smoking or certain cancers. An identified spatial cluster where the causal mechanism is not readily apparent could have merely occurred by chance [42]. Therefore, any identified clusters need to be further investigated and the results and conclusions supported by other published studies [42]. For example our finding of significant differences in risk of smoking during pregnancy across socio-demographic groups is consistent with previous studies [22, 23]. Similarly, the identification of differences in risk of late antenatal care across our study area is consistent with previous studies that late antenatal care is related to mother’s age [15], area of residence [15, 43] and low socio-economic status [15, 16, 44]. This reinforces the robustness of the results obtained from SaTScan.

Demonstrating differences in the socio-demographic factors underlying the identified clusters can assist policy makers in understanding the epidemiology of particular diseases, risk factors or health issues, and to develop and implement interventions and reorient health services. For example, SaTScan has been used to implement anti-marlarial interventions at the household level [45], to identify clusters of hypoplastic left heart and assess genetic and environmental factors contributing to hypoplastic left heart [46], and planning regional tuberculosis prevention and control strategies by identifying clusters of high incidence of tuberculosis [47].

Our study, by identifying adverse maternal-related behaviours by spatial clusters or geography, provides valuable information about the geographical disparity of adverse maternal-related behaviour. It also provides additional insights to the characteristics of the cluster that contribute to adverse perinatal outcomes. Local knowledge and understanding of the physical characteristics of the identified geographical areas (for example, availability of public transport, location of health services) and socio-demographic characteristics of the women within the clusters can assist policy makers to focus the scope of prevention/intervention programs and changes to health care delivery, thus providing more effective programs to suit individual needs and public health resources can be delivered more efficiently.

We also overlaid the clusters of maternal smoking with the clusters of first antenatal visit to identify significant spatial clusters of both smoking during pregnancy and late antenatal visits. Therefore, if there are limited resources, targeting initially only those spatial clusters with both risk factors may be a cost-effective approach to improving maternal and birth outcomes.

Public health interventions can be directed at a number of geographic levels. They can be at a national level (for example, smoking cessation campaigns using the mass media) at the one extreme to very local interventions (for example, modifying the physical characteristics of a locality) at the other extreme. Interventions at both these geographic levels can be equally valid and important. SaTScan allows us to vary the size of the scan windows and hence the number and geographic dimensions of the clusters.

Clusters generated using a larger size scan windows produce larger clusters which can help policy makers make decisions at larger geographic levels, for example, at the state of regional level. However, these large spatial clusters will cover a large area with a larger and more heterogeneous population. Conversely, clusters generated using smaller circular windows will produce smaller clusters but will contain a more homogeneous population which can help policy makers in planning more focused community interventions [48, 49]. For example, Fang et al. used circular windows no more than 20% to identify hemorrhagic fever with renal syndrome clusters and smaller circular windows no more than 10% to identify possible subclusters for more efficient resource allocation for preventing hemorrhagic fever with renal syndrome [41].

The strengths of this study is the use of georeferenced address of each individual case (only 0.1% of the addresses were not be able to retrieved) which offered more geographic precision than studies based on administrative boundaries such as local government area, postal area or CCD. Using SaTScan enables us to analyse variations in any event of interest for many possible small and large geographic groups and without being restricted by administrative boundaries. In SaTScan, the shape of the scan window can be circular or elliptical. Circular windows provide good level of accuracy if the population at risk exists in circular shaped areas [29, 50, 51]. However, as population at risk do not exist in circular shaped areas, elliptical-shaped scan windows will provide slightly increased power for identifying non-circular shaped clusters, for example, long narrow clusters [52]. Tango and Takahashi recently proposed using flexible shaped spatial scan statistics to accommodate irregular shaped clusters [51]. Another advantage is that where point data are not available, the centroids of the smallest geographical spatial unit that are available can be used in SaTScan [52]. Lastly, SaTScan is also capable of performing space-time scan statistics to identify clusters existing in both space and time [34].

However, there are also a number of limitations of using SaTScan to identify spatial clusters. For those wishing to explore modelling based approaches, SaTScan only implements scan statistic methods. Also, whilst it can identify clusters, it cannot explain why the variations in the risks of the events of interest exist. SaTScan also lacks an interface to other graphing, mapping and statistical packages. Another limitation of SaTScan is that it does not provide guidance for choosing the maximum size of the scan windows [40]. We were also unable to link multiple births to a mother and therefore could not account for this in our analyses. This could underestimate the standard error and overestimate the significance of socio-demographic characteristics. Further, we were not able to include all potential predictors or confounders of adverse maternal –related behaviours in our models as they were not available in our dataset. Nevertheless, geospatial analytical techniques are useful tools for identifying geographic areas for intervention. However, the usefulness of these geospatial techniques is only as good as the quality of the data that are available. Finally, although not a limitation of the study, an important point to bear in mind is that due to the large number of women in our study, small differences between women in the spatial clusters and women outside of the spatial clusters were likely to be detected as statistically significant in the descriptive analysis. The epidemiological and clinical importance of such small differences should also be considered in decision making rather than simply relying on statistical significance in informing decisions.

This study provides a first attempt, as far as we are aware, to visually and quantitatively identify and describe the spatial characteristics of ante-natal risk factors. We have demonstrated that using existing georeferenced routinely collected health data, GIS and geospatial techniques can be useful tools to assist policy makers to implement targeted interventions in smaller geographic areas and which will also ensure that limited resources are used efficiently.