- 3.1 Included studies and their characteristics
The literature search identified 3,723 unique citations. After exclusion of irrelevant studies based on title and abstracts, 98 publications remained for full-text assessment. Of these, a further 55 papers did not meet the selection criteria and were excluded. Therefore, 43 articles were reviewed in this scoping review (figure 1). No study was found in the manual-search of the bibliographies of included studies.
Of the 43 included studies, 38 used data from a single country and five studies used data from two or more countries. Single-country studies originated from 13 different countries across three continents: 20 of these were within Europe, 10 in North America, six in Asia, and two were from the Oceania region (Figure 2). MS prevalence in these countries was greater than 68 people per 100,000 populations. Most of the European studies were from western and northern Europe, and five of the six studies in Asia were from Iran. North American studies were conducted in the USA and Canada. Both studies in the Oceania region were conducted in Australia.
As expected, the number of studies has increased over time, particularly since 2008 (Figure 3a). Figure 3b reveals that the geographical scale differed across included studies from a city to a whole country. Of the 43 studies, three (7%) collected data from city level, 19 (44%) from a state/provincial level, and 16 (37%) from whole country level. Five studies (12%) collected data from more than one country.
- 3.2 Type of GIS application
Since each study could employ more than one GIS application, the list of papers in each group was not mutually exclusive (Figure 4). Figure 4a highlights the number of studies in each GIS application group and the overlap between the groups. The majority of studies (n=30, 69%) employed more than one type of GIS applications: 29 studies (67%) employed two types, and one study (2%) employed three types. Thirteen studies (30%) employed only one type of GIS applications. Figure 4b shows the total number of papers in each group. Thematic mapping was the most commonly used GIS application, employed by 40 included studies (93%), followed by spatial cluster detection and risk factors detection, each used in 16 studies (37%), while health access and planning was applied in only two studies (4%).
Table 1 provides a summary and detailed information on the included studies and distribution of the studies across the defined groups.
Table 1: Characteristics of included studies.
ID
|
Author, (Year)
|
Type of GIS application
|
Visualisation complexity score
|
Location
|
Size, Time
|
Main spatial analysis
|
Main Finding
|
Thematic mapping
|
Spatial cluster detection
|
Risk factors detection
|
Health access and planning
|
1
|
Landtblom, (2002)(13)
|
Choropleth map
|
x
|
x
|
x
|
1
|
Sweden
|
5245, 1952-1992
|
Visualising mortality rate and disability pensioning rate of MS over the time
|
Varmland county had the highest mortality rate
|
2
|
Llorca (2005)(14)
|
Choropleth map
|
x
|
x
|
x
|
2
|
Spain
|
3084, 1975-1998
|
Visualising age-adjusted MS mortality
|
A north-south gradient in age-adjusted MS mortality exists in Spain
|
3
|
Yiannakoulias (2007)(15)
|
Choropleth map
|
x
|
x
|
x
|
1
|
Alberta, Canada
|
7602, 1994-2004
|
Visualising percentage change in distribution of MS cases between index date and 2004
|
Persons with MS are more likely to change their residence than normal population
|
4
|
Bostrom (2009)(16)
|
Choropleth map
|
x
|
x
|
x
|
1
|
Varmland, Sweden
|
465, 2002-2003
|
Visualising prevalence rate of MS
|
Support previous study reporting, Varmland as a high-risk zone for MS
|
5
|
Palmer (2013)(17)
|
Choropleth map
|
x
|
x
|
x
|
1
|
Australia
|
21283, 2010
|
Visualising age-adjusted prevalence rate of MS
|
Using prescription data to estimate MS prevalence provided similar result to other methods currently used
|
6
|
Ramagopalan (2011)(18)
|
Choropleth map
|
x
|
x
|
x
|
1
|
England
|
56681, 1999-2005
|
Visualising admission Rate of MS
|
Continued existence of a latitude gradient for MS
|
7
|
Risberg (2011)(19)
|
Choropleth map
|
x
|
x
|
x
|
1
|
Oppland, Norway
|
474, 1989-2001
|
Visualising prevalence and Incidence of MS
|
The highest prevalence rates of MS ever reported in Norway and possible influence of environmental factors
|
8
|
Bargagli (2016)(20)
|
Choropleth map
|
x
|
x
|
x
|
1
|
Lazio, Italy
|
7377, 2006-2011
|
Visualising age- and gender-adjusted MS prevalence rate
|
The Lazio region is a high-risk area for MS, although with an uneven geographical distribution
|
9
|
McDonald (2019)(21)
|
Choropleth map
|
x
|
x
|
x
|
1
|
Scotland
|
2569
|
Visualising incidence of MS
|
Incidence of MS seems to increase by latitude in Scotland
|
10
|
Wallin (2019)(3)
|
Choropleth map
|
x
|
x
|
x
|
1
|
International
|
I2221188, 1990-2016
|
Visualising age-standardised multiple sclerosis prevalence
|
Prevalence has increased substantially in many regions since 1990
|
11
|
Pugliatti (2002)(22)
|
Choropleth map
|
Bayesian hierarchical modelling
|
x
|
x
|
3
|
Northern Sardinia, Italy
|
686, 1997
|
Discover spatial hotspots using Bayesian mapping
|
Hot spot of MS exists in the southwestern part of the province of Sassari
|
12
|
Bergamaschi (2006)(23)
|
Choropleth map
|
Bayesian hierarchical modelling
|
x
|
x
|
3
|
Pavia, Italy
|
464, 2000
|
Discover spatial hotspots using Bayesian modelling
|
Bayesian methods can be used to obtain reliable maps of disease prevalence and to identify possible clusters of disease
|
13
|
Cocco (2011)(24)
|
Choropleth map
|
Bayesian hierarchical modelling
|
x
|
x
|
3
|
South-western Sardinia, Italy
|
292, 12/31/2007
,
|
Discover spatial hotspots using Bayesian modelling
|
The study confirms Sardinia as a high-risk area for MS and supports the relevance of both genetic and environmental factors
|
14
|
Fromont (2012)(25)
|
Choropleth map
|
Bayesian hierarchical modelling
|
x
|
x
|
3
|
France
|
28682, 2000-2007
|
Discover spatial hotspots using Bayesian modelling
|
Higher annual incidence levels in north-eastern France, and lower levels on the Atlantic coast as well as in ‘departments’ in the south on both sides of the Rhône River
|
15
|
Pivot (2016)(26)
|
Choropleth map
|
Bayesian hierarchical modelling
|
x
|
x
|
3
|
France
|
Not mentioned, 12/31/2004
|
Discover spatial hotspots using Bayesian modelling
|
Found heterogeneity in MS prevalence among the 21 departments of France, some with higher prevalence than anticipated from previous publications
|
16
|
Bezzini (2017)(27)
|
Choropleth map
|
Bayesian hierarchical modelling
|
x
|
x
|
3
|
Tuscany, Italy
|
7330, 12/31/2013
|
Discover spatial hotspots using Bayesian modelling
|
Using the Bayesian method, the study estimated area-specific prevalence in each municipality reducing the random variation and the effect of extreme prevalence values in small areas
|
17
|
Bergamaschi (2019)(28)
|
Choropleth map
|
Bayesian hierarchical modelling
|
x
|
x
|
3
|
Northern Italy
|
927, 12/31/2016
|
Discover spatial hotspots using Bayesian modelling
|
The Bayesian mapping highlighted area with a significantly higher/lower MS risk where to investigate etiologic hypotheses based on environmental and genetic exposures
|
18
|
Fromont (2010)(29)
|
Choropleth map
|
Bayesian hierarchical modelling
|
x
|
x
|
3
|
France
|
49417, 2003-2004
|
Discover spatial hotspots using Bayesian modelling
|
Higher prevalence of MS in North-Eastern France and a lower prevalence of multiple sclerosis in the Paris area and on the Mediterranean coast
|
19
|
Green (2013)(30)
|
Choropleth map
|
Bayesian hierarchical modelling, spatial scan statistics
|
x
|
x
|
3
|
Canada
|
2290, 1990-2006
|
Discover spatial hotspots using Bayesian modelling and spatial scan statistics
|
The incidence rates were lowest in northern Manitoba and highest in three areas
|
20
|
Torabi (2014)(31)
|
Choropleth map
|
Bayesian hierarchical modelling, spatial scan statistics
|
x
|
x
|
3
|
Manitoba, Canada
|
2290, 1990-2006
|
Discover spatial hotspots using Bayesian mapping and spatial scan statistics
|
Bayesian method can simultaneously identify geographical variations and control for possible confounders
|
21
|
Donnan (2005)(32)
|
Choropleth map
|
Spatial scan statistics
|
x
|
x
|
2
|
Tayside, Scotland
|
772, 1970-1997
|
Discover spatial and temporal hotspots using spatial scan statistics
|
There is clear temporal and geographical variation of prevalence which cannot be explained by over-ascertainment of clusters or confounding
|
22
|
Turabelidze (2008)(33)
|
Choropleth map
|
Spatial scan statistics
|
x
|
x
|
2
|
USA
|
321, 1998-2002
|
Discover spatial hotspots using spatial scan statistics
|
The estimates of MS prevalence in Mid-western community of USA appeared to be comparable to estimates from other areas of similar latitude in the United States and Western Europe
|
23
|
Saei (2014)(34)
|
Choropleth map, Heat map
|
Getis-Ord Gi*
|
x
|
x
|
3
|
Tehran, Iran
|
6027, 2001-2012
|
Discover spatial hotspots using Getis-Ord Gi* test
|
Heterogeneous geographical distribution of MS and its higher estimated incidence for northern zones in Tehran.
|
24
|
Bihrmann (2018)(35)
|
Heat map
|
Kernel regression
|
x
|
x
|
4
|
Denmark
|
12993, 1971-2013
|
Discover spatial hotspots using kernel regression
|
Small-scale geographical variation in the risk of MS suggests that local environmental risk factors could be at play and may be related to lifestyle factors
|
25
|
Wade (2014)(36)
|
Choropleth map
|
Local and global Moran’s, Getis-Ord General 𝐺
|
x
|
x
|
2
|
International
|
131 prevalence study
|
Using local and global hotspot analysis of Local Moran’s and Getis-Ord General 𝐺
|
They suggested a five-zone scale to categorize countries regarding MS prevalence
|
26
|
Gregory (2008)(37)
|
Choropleth map
|
x
|
Per capita income, and PM-10
|
x
|
1
|
Georgia, USA
|
6247, 2006
|
Bivariate linear regression to find association
|
The best predictive models for the MS prevalence in GA included both per capita income and PM-10 for females, but only per capita income only for males.
|
27
|
Sloka (2008)(38)
|
Choropleth map
|
x
|
Ultraviolet B (UVB)
|
x
|
1
|
Newfoundland, Canada
|
328, 1996-2003
|
Bayesian mapping to find geographical distribution and covariate analysis to find associations and interpolation to estimate unknown values of UVB
|
The study suggests that UVB radiation may contribute to the pathogenesis of MS
|
28
|
Beretich (2009)(39)
|
Proportional map
|
x
|
UV index
|
x
|
2
|
North America
|
Data obtained from other studies
|
bivariate linear regression
|
This analysis suggests a strong association between UV radiation and MS distribution, and an increase in risk for MS in those areas with a low UVI.
|
29
|
Amram (2018)(40)
|
x
|
x
|
UVB
|
x
|
0
|
Canada
|
3226, 1980-2005
|
Remote sensing and basic GIS analysis (Overlay)
|
Early life ambient UVB is not significantly associated with the age at MS onset.
|
30
|
Handel (2010)(41)
|
Heat map
|
x
|
UV index, smoking
|
x
|
3
|
Europe
|
Data obtained from other studies
|
Linear regression to find associations
|
Both HLA allele frequencies and UV index are highly correlated with MS prevalence
|
31
|
Taylor (2010)(42)
|
x
|
x
|
Latitude
|
x
|
0
|
Australia
|
330, 2003-2006
|
Poisson regression to find associations
|
The study stablished positive latitudinal gradient of first demyelinating events
|
32
|
Ramagopalan (2011)(43)
|
Choropleth map
|
x
|
UVB and infectious mononucleosis
|
x
|
1
|
England
|
Not mentioned, 1998-2005
|
Geographical Weighted Regression to find associations
|
UVB exposure and infectious mononucleosis together can explain a substantial proportion of the variance of MS.
|
33
|
Tsai (2013)(44)
|
Choropleth map
|
x
|
Arsenic (As), Mercury (Hg), Cadmium (Cd), Chromium (Cr), Copper (Cu), Nickel (Ni), Lead (Pb) and Zinc (Zn)
|
x
|
2
|
Taiwan
|
1240, 1997-2008
|
Interpolation (IDW method) and spatial regression to find associations
|
Exposure to lead (Pb) in soil positive associated with incidence of MS in Taiwan, especially in males. Exposure to arsenic (As) in soil negative associated with MS in Taiwan, especially in females.
|
34
|
Schuurman (2013)(45)
|
Bar graph map
|
x
|
Ultraviolet B (UVB)
|
x
|
1
|
Canada
|
4010, 2005
|
Raster Resampling and Overlay
|
This protocol provided a framework for researchers to more accurately estimate UVB for understanding etiology of MS and other chronic diseases.
|
35
|
Heydarpour (2014)(46)
|
Heat map, Dot density map
|
average nearest neighbour index
|
PM-10, SO2, NO, NO2, NOX
|
x
|
4
|
Tehran, Iran
|
2188, 2003-2013
|
Cluster Analysis by average nearest neighbour method and multiple regression to find associations
|
This study revealed the potential role of long-term exposure to air pollutants as an environmental risk factor in MS
|
36
|
Monti (2016)(47)
|
Dot density map
|
x
|
Heavy metals, UV, and Urbanization
|
x
|
2
|
South-Western Sardinia, Italy
|
282, 2011-2016
|
Multiple logistic regression were used to evaluate association of risk factors
|
Study suggests a relationship between environmental exposure to metals and MS.
|
37
|
Lavery (2017)(48)
|
Proportional map, Choropleth map
|
x
|
Environmental Quality Index (air, water, land)
|
x
|
2
|
USA
|
1170, 2008-2016
|
Logistic regression and Poisson regression used to find associations
|
Among environmental factors, air quality may contribute to the odds of developing MS in a paediatric population.
|
38
|
Ashtari (2018)(49)
|
Dot density map
|
x
|
Air quality index (AQI)
|
x
|
2
|
Isfahan, Iran
|
2000, 2008-2016
|
Unadjusted logistic regression to find associations and interpolation to estimate unknown values of AQI
|
Positive association found between EDSS of MS cases and the AQI level.
|
39
|
Ashtari (2018)(50)
|
Choropleth map
|
x
|
Height above sea, average annual rainfall, and land use
|
x
|
2
|
Isfahan, Iran
|
2000, 2001-2014
|
ANOVA, independent t-test and Kruskal-Wallis were used to find associations
|
With increase in the height above sea level and the average annual precipitation, the incidence of disease is decreased.
|
40
|
Bains (2010)(51)
|
Heat map
|
x
|
UV radiation (near-horizon sunshine)
|
x
|
2
|
Europe
|
62533, 1950-2010
|
Remote Sensing for data gathering and correlation analysis to find association
|
Direct solar damage to the optic nerve may be a trigger for MS.
|
41
|
Pakdel (2019)(52)
|
Choropleth map
|
x
|
Human Development Index (HDI), Income and education.
|
x
|
1
|
Iran
|
Not mentioned, 2006-2013
|
Linear regression analysis to find associations
|
Significant relationships were found between the prevalence of MS and HDI, income and educational level.
|
42
|
Culpepper (2010)(53)
|
Dot density map, Proportional map
|
x
|
x
|
|
3
|
USA
|
19219, 2007
|
Measuring availability and accessibility of Veterans Health Administrations for MS patients by travel time and network analysis
|
Reduction in the proportion of patients traveling more than two hours to the nearest MS clinic
|
43
|
Neven (2013)(54)
|
x
|
x
|
x
|
|
0
|
Belgium
|
Not mentioned
|
Using Global Positioning System (GPS) to assess physical activity of MS patients
|
Average number of travels per day for mild, moderate and severe MS patients are different from each other and compared to control group.
|
Most of the included studies (n=40, or 93%) employed thematic mapping, in 10 of which it was the only GIS application used (Table1, ID: 1-10). These studies used different types of thematic maps for visualisation, depending on their aim. Of these 40 papers, 31 studies (77%) used at least one Choropleth map: that is, a thematic map where geographic regions are coloured, shaded or patterned in relation to a value. Five articles (12%) used heat maps to represent the intensity of an attribute using colours, without using geographical boundaries to group them (Table1, ID: 23, 24, 30, 35, 40). The heat maps were used to visualize continuous variables such as air pollution or UV radiation. Four articles (9%) contained dot density maps to represent location of MS cases (Table 1, ID: 35, 38), geochemical sampling sites (Table 1, ID: 36) or specialty care providers (Table 1, ID: 42) by point symbols. The least common types of maps were proportional maps, in which a larger symbol represents a higher value (Table 1, ID: 28, 37), and bar graph maps, in which a bar graph represents distribution of a variable in a geographical area (Table 1, ID: 34). Figure 2 is an example of a combined choropleth and proportional map to visualize MS prevalence and number of studies in each country. The type of thematic maps used in each study is available in Table 1.
Most of the papers visualised variables related to disease including MS incidence/ prevalence (Table 1, ID: 3-5, 7-26, 28, 33, 35, 38, 39, 41), mortality/ morbidity rate (Table 1, ID: 1, 2, 17), or hospitalization rate (Table 1, ID: 6). Non-disease related variables were environmental variables (Table 1, ID: 27, 28, 30, 32, 34, 35, 37, 39, 40) and distribution of services (Table 1, ID: 41, 42).
Four studies employed smoothed incidence rates in their maps (Table 1, ID: 17-20) to improve the accuracy of incidence rates for small areas with few observations, and to remove noise (e.g. random variation) showing the geographical pattern clearer.
- 3.2.2 Spatial cluster detection
Sixteen papers used spatial cluster detection techniques to identify areas of high-risk (or low-risk) of MS (Table 1, ID: 11-25, 35). Only one study (Table 1, ID: 25) employed global clustering methods to assess spatial trends (the tendency of spatial clustering) across the entire study region (55). All of the 16 studies in this group used at least one local clustering method to assess significance of local statistics at each location and to identify the location of spatial clusters (hot spots, cold spots) and spatial outliers (56). Ten of these applied a Bayesian hierarchical modelling technique (Table1, ID: 11-20). Bayesian approach estimates posterior probability (PP) that indicates whether the prevalence (or incidence) rate for each area is significantly lower or greater than a given reference value (57). Also, four studies (Table 1, ID: 19-22) used spatial scan statistics, which is defined as the maximum of likelihood ratio test statistics over a collection of scanning windows (58). A few studies used other spatial local cluster detection techniques like Getis Ord Gi (Table 1, ID: 23, 25) or Anselin Local Morans I (Table 1, ID: 25). In addition, one study employed kernel regression that enables identification of hotspot areas (Table1, ID: 24). One study compared Bayesian modelling method with spatial scan statistics using MS incidence data and suggested that Bayesian modelling can identify geographical variations while controlling for possible confounders (Table 1, ID: 20). The type of spatial clustering techniques used in each study is available in Table 1.
- 3.2.3 Spatial risk factors detection
The association of environmental and geographical risk factors with spatial distribution of MS was investigated in sixteen studies. These risk factors were modifiable (such as air pollution, water pollution or heavy metals) and non-modifiable (such as ultraviolet (UV) radiation or latitude). UV radiation was the most common risk factor (9 articles) reported by the included studies (Table1, ID: 27-31, 32, 34, 36, 40). These studies used satellite data such as NASA Total Ozone Mapping Spectrometer (TOMS) dataset (59) (Table 1, ID: 27, 29, 30 , 34) or other local and national datasets (Table 1, ID: 28, 32) for measuring UV exposure of geographical areas. Air pollution (Table 1, ID: 26, 35, 37, 38) and heavy metals (Table 1, ID: 33, 36) were the next most frequently investigated risk factors. A full list of risk factors is available in Table 1.
Most studies used regression analysis to assess the association between risk factors and disease onset. Four studies used variations of linear regression (Table 1, ID: 28, 30, 35, 41) and three studies applied logistic regression to model a dichotomous dependent variable (Table 1, ID: 36-38). Also, Poisson regression was employed in one study to model count data (Table 1, ID: 37). Other less common methods were correlation analysis (Table1: 40), the Kruskal–Wallis test (Table1, ID: 39) and ANOVA test (Table 1, ID: 39). Interpolation methods like Inverse Distance Weighting (IDW) were employed in some studies (Table 1, ID: 27, 33, 38) for estimating unknown values of environmental exposures based on known values.
- 3.2.4 Health access and planning
Only two articles focused on health access and planning for MS using GIS tools. One study measured MS patients’ access to Veterans Health Administration (VHA) in USA (Table 1, ID: 42). Accessibility of specialty clinics was calculated by proximity analysis using Travel-time buffers.
One study identified the activity and related travel behavior of MS patients (Table 1, ID: 43). This study used Global Positioning System (GPS) tags to monitor MS patients’ outdoor activity and travel behavior in relation to disease-related disability.
- 3.5 Visualisation complexity
Figure 5 demonstrates the GIS visualisation complexity of the included papers. Thematic mapping with 40 papers was the most scored measure and buffer visualisation with one paper was the lowest (Figure 5a). Three studies which received a score of zero, did not have any GIS visualisation. 88% of studies were scored between one and three and only two studies received a score of four. None of the papers scored five or six (Figure 5b).