RIF V40 Manual

RIF_v40_Manual

User Manual:

Open the PDF directly: View PDF .
Page Count: 43

Download
Open PDF In Browser	View PDF

The Rapid Inquiry Facility (RIF)
Version 4.0
How to use the RIF 4.0 client

Authors (2017-2019):
Parkes, B., Morley, D., Hambly, P
Small Area Health Statistics Unit (SAHSU)
MRC-PHE Centre for Environment and Health
Department of Epidemiology and Biostatistics
School of Public Health
Imperial College London
Medical Faculty Building
St Mary's Campus, Norfolk Place
LONDON W2 1PG
Website www.sahsu.org

Contents
The Rapid Inquiry Facility (RIF).......................................................................................................................................... 1
Version 4.0 ........................................................................................................................................................................ 1
How to use the RIF 4.0 client ............................................................................................................................................ 1
1. Introduction to the RIF .................................................................................................................................................. 4
1.1 Purpose ................................................................................................................................................................... 4
1.2 Principal Features.................................................................................................................................................... 4
1.3 Current Limitations ................................................................................................................................................. 5
1.4 Input Facilities ......................................................................................................................................................... 5
1.5 Export Capability ..................................................................................................................................................... 6
1.6 Scope of this Manual............................................................................................................................................... 6
2. Background Considerations .......................................................................................................................................... 7
2.1 Disease mapping or risk analysis............................................................................................................................. 7
2.2 Geographical data issues ........................................................................................................................................ 8
2.3 Health and population database issues .................................................................................................................. 8
2.4 Exposure data ......................................................................................................................................................... 8
2.5 Statistics .................................................................................................................................................................. 8
2.6 Interpretation and Limitations ................................................................................................................................ 9
2.7 References ............................................................................................................................................................ 10
3. Starting up ................................................................................................................................................................... 11
3.1 Test data................................................................................................................................................................ 11
3.2 Logging in .............................................................................................................................................................. 12
3.3 RIF mapping tools.................................................................................................................................................. 12
4. Running a new RIF study ............................................................................................................................................. 13
4.1 Study details .......................................................................................................................................................... 14
4.2 Study area ............................................................................................................................................................. 14
4.3 Comparison area ................................................................................................................................................... 16
4.4 Investigation parameters ...................................................................................................................................... 16
4.5 Statistical methods................................................................................................................................................ 17
4.6 Saving and reloading studies ................................................................................................................................ 18
4.7 Study status ........................................................................................................................................................... 18
4.8 Run study .............................................................................................................................................................. 19
4.8 Messages ............................................................................................................................................................... 20
4.10 Reset ................................................................................................................................................................... 21
5. Data viewer ................................................................................................................................................................. 22
5.1 Choropleth map .................................................................................................................................................... 22
5.2 Data table .............................................................................................................................................................. 23
2

5.3 Population pyramid............................................................................................................................................... 24
5.4 Frequency distribution .......................................................................................................................................... 24
5.5 Risk Graphs............................................................................................................................................................ 25
5.5 Info Button ............................................................................................................................................................ 25
6. Mapping ...................................................................................................................................................................... 28
6.1 Choropleth maps................................................................................................................................................... 28
6.2 Disease map charts ............................................................................................................................................... 28
7 Export ........................................................................................................................................................................... 29
7.1 R Scripts................................................................................................................................................................. 30
7.2 Shapefiles .............................................................................................................................................................. 31
7.3 Generated Maps ................................................................................................................................................... 31
7.4 Reports .................................................................................................................................................................. 32
Appendices ...................................................................................................................................................................... 33
Appendix A. Statistical methods ................................................................................................................................. 33
Indirectly standardised risks ................................................................................................................................... 33
Empirical Bayes Analysis ......................................................................................................................................... 34
Full Bayesian smoothing ......................................................................................................................................... 35
R and R-INLA............................................................................................................................................................ 35
Appendix B. Descriptive analysis of Sahsuland ........................................................................................................... 37
Sahsuland population ............................................................................................................................................. 37
Sahsuland numerator data ..................................................................................................................................... 39
References .................................................................................................................................................................. 43

1. Introduction to the RIF
1.1 Purpose
The Rapid Inquiry Facility (RIF) is an automated tool to allow epidemiologists to rapidly address epidemiological and
public health questions using routines collected health and population data.
The RIF can perform risk analysis around putative hazardous sources and can be used for disease mapping. It generates
indirectly standardised rates and relative risks for any given health outcome, for specified age and year ranges, for any
given geographical area.
The RIF has been developed by the UK Small Area Health Unit at Imperial College London and funded by the US Centers
for Disease Control and Prevention (CDC) and the National Institute for Health Research Health Protection Research
Unit.
This manual describes version 4.0 of the RIF (last update March 2019). This version of the RIF supports the following
browsers:
•
•
•

Firefox 62.0 (64 bit) or greater: preferred as it handles the large memory requirements of high resolution
administrative geographies (e.g. UK Census Output Area) best;
Chrome 69.0 (64 bit) or greater;
Microsoft Edge 42.17134 or greater;

The RIF will work with Microsoft Internet Explorer (IE) 11.0.9600 or greater. Use of IE is not advised as it will run slowly
and crash at medium levels of resolution (e.g. US Counties).

1.2 Principal Features
•

The system is designed using a three tier architecture. The client runs entirely within the user’s browser and
does not require installation of any software on the client’s machine. All data is stored in a secure database
on the server with user access, security and data processing performed by the middleware, also running on
the server.

Figure XXX. The RIF Arctitecture.
•

In addition to the point source ‘risk analysis’ and ‘disease mapping’ options, it is also possible to import
detailed exposure data, such as output from dispersion modelling.
4

•
•
•
•

Within the risk analysis tool, the RIF performs test on the relative risks to assess for homogeneity and linear
trend with exposure.
Within the disease mapping tool, the RIF can perform empirical Bayes smoothing and full Bayes smoothing
using the r-INLA library assessed from the middleware.
The RIF can export data for further analysis in other (statistical) software packages such as SaTScan and
WinBUGS and GIS packages such as ArcGIS and QGIS.
Support for Information Governance via database security features (e.g. role-based access control, auditing).
Users can only access and utilise data for which they have been granted by the database administrator.

1.3 Current Limitations
•

•
•

•

No support for covariates embedded in numerator or denominator data. Data must be extracted into a
separate covariate table. Covariates must be merged into a single table and disaggregated (if required) by
year. It is not planned to remove these restrictions which were in the previous RIF.
External covariates must be quantilised. Support for continuous variable covariates (i.e. on the fly
quantilisation) may be added in future releases;
Denominator data is always used in indirect standardisation. There is no support currently in the RIF for
direct standardisation using standard populations. This was supported in previous versions of the RIF and
will be put back if required;
No support for ad-hoc SQL. This functionality will be partially re-implemented in future releases using user
specified conditions (pre-defined groups). The feature was removed as it cannot be implemented in a secure
manner (i.e. it permits SQL injection attacks);
Numerators are currently limited to ICD 10 coding only. Support will be added for:
o ICD 9 (Autumn 2018);
o ICD 11 (subject to the release of the 11th Edition in June 2018).

In the longer term it is expected that support will be added for:
•
•
•
•
•
•
•
•
•
•

ICD oncology (ICD-O-1);
UK HES oper and A+E codes;
User specified conditions (pre-defined groups), e.g. Low birthweight, complex groups of ICD codes, the all
record condition (the 1=1 ad-hoc SQL filter in the previous RIF);
The RIF currently lacks complex support for Information Governance beyond having strong role-based
permissions. An information governance tool is envisaged to assist;
More than one external covariate;
More than one investigation;
Multiple ICD field names (e.g. as used in hospital episode statistics);
Covariate geolevel cannot be of lower resolution than study geolevel;
Separate AGE_GROUP/AGE/SEX columns;
Support for more than current and previous version of a table outcome (e.g. ICD). This would allow ICD 9, 10
and 11 (or 8, 9 and 10) to be supported all at the same time together the the start and end year for each
version in a numerator table. Currently the RIF applies the same ICD filter to all years. This approach may
cause problems if there are coding incompatibilities between the version (i.e. the same code means
something different in two or more version).

1.4 Input Facilities
The RIF will be provided with a Data Loader tool that allows users to import their health and GIS data. For the present
there is a RIF Data Loading manual that describes how to manually process and load data into the RIF.

1.5 Export Capability
The RIF is a versatile tool for generating smoothed disease maps and for calculating relative risks in populations living
around putative sources of exposure. There are, however additional software packages that can also be used to
explore spatial and temporal trends in data, and to detect statistically significant clusters of disease that many users
will wish to employ to aid their investigations. The RIF has been designed to work alongside these programmes and
can currently export data in ZIP file format for use in statistical packages (R, WinBUGS and SaTScan), GIS packages via
shapefile import as well as to Microsoft Excel for further processing.

1.6 Scope of this Manual
This manual explains how a user, typically an epidemiologist, would use the RIF client to set up a new study, run the
study and examine the results. In addition, there is an appendix covering the statistical methods used by the RIF when
a study is run.

2. Background Considerations
This chapter will give a very brief overview of some of the considerations that should be made when planning,
undertaking and interpreting a RIF study. These considerations are not unique to studies undertaken using the RIF,
and although the RIF will help to speed up point source and mapping studies, users are cautioned to plan RIF studies
as carefully as they would any other epidemiological investigation they would undertake. More details on these issues
can be found in the following papers:
Beale L, Hodgson S, Abellan JJ, LeFevre S, Jarup L, 2010. Evaluation of spatial relationships between health and the
environment: The Rapid Inquiry Facility, Environmental Health Perspectives. Doi:10.1289/ehp.0901849
Beal L, Abellan J, Hodgson S, Jarup L, 2008. Risk assessment using spatial epidemiological methods, Environmental
Health Perspectives. Volume 116, number 8.
Ball W, LeFevre S, Jarup L, Beale L, 2008. Comparison of different methods for spatial analysis of cancer data in Utah,
Environmental Health Perspectives. Volume 116, number 8.

2.1 Disease mapping or risk analysis
The intention is to provide two types of study in the RIF, disease mapping and risk analysis. Initially the RIF v4.0 will
only have disease mapping available.
The disease mapping approach can be used to visualise mortality/morbidity rates and risks across and area. Disease
mapping can provide an invaluable tool to explore spatial patterns of health outcomes; identify potential issues
regarding data quality by geographical area; and identify areas which need additional resources or remediation.
The risk analysis approach can be used to explore whether a source or some particular exposure (risk factor) is having
an impact on health in a local population. To carry out a risk analysis study the geographical position of the putative
risk factor will need to be known (as a point or a plume for example), and some consideration should be given to what
distance the exposure of interest might be expected to have an impact. Thought should also be given to whether the
exposure is likely to have a short or long term effect, as this will determine which years of health data will be most
appropriate to study.
Careful consideration should always be given to the most appropriate scale of investigation, which will depend on local
circumstances (i.e. population density), and on the outcome of interest (i.e. whether this is a very rare outcome or
not). The most appropriate geographical resolution to be used in any particular study will depend on individual
circumstances and is often a compromise between having a high enough resolution to allow differences in disease risk
to be assessed by small area, and having a large enough area (or population) to ensure that disease rates are
sufficiently stable to permit interpretation. When mapping a rare disease across a sparsely populated area, thought
should be given to the value of mapping at the smallest units available; if these units lead to very unstable risk
estimates due to small populations, it may be preferable to lose some of the geographical resolution to gain more
stable disease rates. While there may be a basis for investigating the population living in very close proximity to a
putative pollution source, thought should be given to whether the size of this ’exposed’ population is sufficient to
provide a meaningful risk estimate.
When assessing potential disease clusters pot hoc, special care must be taken to avoid the ‘Texas sharpshooter’ effect,
where the cluster is tightly defined in space and time, thus minimising the population at risk, and maximising the
excess risk.

2.2 Geographical data issues
There are many different types of enumeration areas (e.g. administrative, health, electoral, postcode etc.) and
frequently their boundaries do not align. To use the RIF, however, the geographical data for any study must be
hierarchical, with the boundaries at higher resolution areas being subdivisions of the larger areal units. In most
countries census data are hierarchical. Since these boundaries tend to be defined administrative boundaries rather
than physical boundaries, the boundary locations can, and do, change over time. Area name and codes can also
change, which can be further complicated by the fact that different government departments can develop different
coding systems for administrative geographies, or use slightly different names for the same area.
Inconsistent geography is problematic for any temporal studies that span time periods when boundary changes have
occurred and are a major problem when trying to produce and compare meaningful statistics over time. The
Modifiable Areal Unit Problem (MAUP), as it is known, can affect any spatial study that utilises aggregate data sources
(Openshaw, 1984). Since enumeration areas are often arbitrary and can change spatially and temporally, they are said
to be ‘modifiable’. Many spatial datasets are collected at a fine resolution (i.e. a large number of small spatial units)
but are released only after being spatially aggregated to a coarser resolution (i.e. a smaller number of larger spatial
units). This is usual for census data which are collected from every household, but released as aggregated data for an
enumeration area. When values are averaged during the process of aggregation, variability in the dataset is lost and
values of statistics computed at different levels of spatial resolution will be different. This change is called the scale
effect. The aggregation or zonation effect must also be considered, which occurs due to variation in numerical results
that can occur due to the grouping of smaller areas into larger units (e.g. enumeration areas into census tracts). If EAs
were grouped into zones of similar size to census tracts, but in a different spatial arrangement, it is likely to produce
different statistical results between the two groupings of data.
Problems related to the ecological fallacy should also be considered. Users should be wary of interpreting results
solely from aggregate statistics and making assumptions about the nature of individuals from data that relates to
groups.

2.3 Health and population database issues
The appropriate statistical techniques and tools are available to calculate and map small area risks, but meaningful
results can only be achieved if the underlying health and population data are accurate and complete. Local variations
in ascertainment of health data, changes in health event recording over time (e.g. adoption of a new ICD revision),
errors in the denominator (population) data (e.g. due to migration), or incomplete/inaccurate geocoding of either
health or population data (e.g. greater positional errors for rural than for urban addresses) may introduce spurious
temporal or spatial patterns in risk.
Any underlying data problems are not corrected merely by running the analysis through the RIF. It is vital that any data
quality issues are known about, dealt with where possible, and where issues remain, that these are considered fully
when interpreting the results.

2.4 Exposure data
2.5 Statistics
One problem associated with investigating health risks in small areas is that small populations have a small number of
expected and observed events, which can lead to unstable risk estimates. This can result in misleading risk maps,
especially if the area with the smallest populations are quite large (rural areas for instance), as these areas with the
least stable risk estimates can dominate a map. In an attempt to overcome this problem and aid interpretation of the
disease mapping output, the RIF can perform both empirical and full Bayesian smoothing of the raw, relative risks to
account for sampling variability in the observed data. These methods can allow more meaningful risks to be calculated
at the small area level; however these statistical techniques need to be applied with due consideration and caution.
While raw risks can produce noisy maps that are difficult to interpret, over-smoothed maps may produce a
8

homogenous risk surface. Obviously there is a trade-off between high sensitivity (where true high risk areas can be
identified), and high specificity (where areas of no excess risk are correctly identified) (Richardson et al., 2004).
The RIF calculates standardised mortality (or incidence/morbidity) ratios (SMRs), however thse measures are not
directly comparable between different exposure groups as they are not based on the same standard population (i.e.
the age, gender and socio-economic make up between the populations being compared are not exactly the same).
This should only result in misleading comparisons where the population structure is significantly different between
the groups being compared (Goldman and Brender, 2000). An alternative to using indirectly standardised measures
would be to use directly standardised rates and assess comparative mortality figures (CMFs) (or incidence/admissions
figures) (Julious et al., 2001). The use of CMFs is advised for studies in which there are substantial numbers of cases in
each study area or exposure category; however at the small geographical level, the number of cases is usually so few
that directly standardised rates are unstable and the imprecision of this measure makes comparisons very difficult. In
such situations it is appropriate to use SMRs instead of CMFs, provided the stratum specific death rate for each
exposure class are proportional to the standard population rates, and bearing in mind that the rates in each exposure
group may not be directly comparable with each other (Jarup and Best, 2003). Given that the it is intended for use at
smaller geographical levels, the RIF uses indirect standardisation to calculate SMRs and does not currently perform
direct standardisation.
Currently the RIF does not perform any type of temporal analysis. If users are interested in time trends in rates or
relative risks, they might use the RIF to explore trends by running several annual (or other time length) periods and
then plotting the rates/risks obtained. This would be in spirit similar to moving average analysis. Although this could
be valid for explorative purposes, users should be aware that it is not a proper moving average analysis, and therefore
it lacks their properties, hence results should be interpreted carefully.
A full description of the statistical methods performed by the RIF is given in appendix A.

2.6 Interpretation and Limitations
Crucial to effective communication of spatial information is the use of suitable mapping techniques that convey results
objectively. Effective mapping requires both an understanding of the mapped phenomena as well as the mechanisms
to present the data appropriately. This is particularly true for maps that display data related to epidemiological risk in
order to avoid misinterpretation or to over- or under-emphasise particular results. The map displays in the RIF can be
configured with a wide range of base maps available. Additionally; it may be preferable for the user to export the data
in order to import it into a GIS system with greater data symbolisation capabilities.
The main advantages of undertaking spatial epidemiology at the small rather than large area level is increased
interpretability – small-area studies are less susceptible to ecological bias created by within-area heterogeneity; they
also allow local effects (such as impacts of point sources of pollution) to be investigated (Elliot and Wartenberg, 2004).
While analysis at the small area can help reduce components of ecological bias, unless the analysis is carried out at
the individual level it is impossible to rule out this bias entirely. Factors associated with disease in individuals
(Morgenstern, 1998), and while the RIF can help assess whether a reported cluster is statistically significant or can
demonstrate spatial trends in disease risk, the RIF cannot infer a causal relationship between an environmental factor
and a disease. If cause for concern around a particular site is confirmed, data should be checked and validated (for
completeness, diagnostic accuracy, etc.). Replication around other or multiple sites with similar discharges (if they can
be found) can be carried out or indeed etiologic studies at the individual level can be designed and carried out.
It should always be remembered that the RIF-type studies are subject to the limitations outlined above, and the user
should therefore always consider what impact inconsistent geography, health and population data, exposure
misclassification, ecological bias, and so on, will have on the study output. The RIF output therefore needs to be
interpreted with caution and with expert local knowledge.

2.7 References
Elliot P and Wartenberg D. 2004 Spatial epidemiology: current approaches and future challenges. Environmental
Health Perspectives 112(9):998-1006.
Goldman DA & Brender JD. 2000. Are standardized mortaility ratios valid for public health data analysis? Statistics in
Medicine 19(8):1081-1088.
Jarup L & Best N. 2003. Editorial comment on Geographical differences in cancer incidence in the Belgian Province of
Limburg by Bruntinx and colleagues. European Journal of Cancer 39(14): 1973-1975.
Julious SA, Nicholl J & George S. 2001. Why do we continue to use standardized mortality ratios for small area
comparisons? Journal of Public Health Medicine 23(10):40-46.
Openshaw S. 1984. The Modifiable Areal Unit Problem, CATMOG, Concepts and Techniques in Modern Geography, No
38, Norwich, GeoAbstracts.
Morgenstern H. 1998. Ecological Studies, in Modern Epidemiology, Second Edition, KJ Rothman & S Greenland, eds,
Lippincott Williams & Wilkins, pp.459-480.
Richardson S, Thompson A, Best N et al. 2004. Interpreting posterior relative risk estimates in disease-mapping studies.
Environmental Health Perspectives 112(9):1016-1025.

3. Starting up
3.1 Test data
Before using your own data, we recommend using the sample health, population and geography data sets provided
with the RIF software. These data give an idea of how the RIF works and help indicate what format data need to be
in before they can be used in the RIF. The test data are automatically installed with RIF software, and in this version
of the RIF/manual these data relate to a fictitious area known as Sahsuland.
NOTE: all these datasets are fictitious and may not reflect patterns observed in reality.
The data consist of:
•
•
•

Population data (by five year age group by gender1), for the period 1989-1996.
Cancer incidence data for the period 1989-1996.
Covariate data2 on socio-economic status, ethnicity, and proximity to Toxic Release Inventory (TRI) sites.

The example dataset ‘Sahsuland’, supplied with the RIF software, can be used to test the software setup and as a
template for database construction.

Sahsuland is approximately 32,869 km2. The area of Sahsuland (Figure 1) uses for different hierarchical enumeration
areas or levels of geography. Each area can be identified by a unique ID value. This also follows a hierarchical form,
so that LEVEL2 areas are unique by LEVEL1 area, LEVEL3 areas are unique by LEVEL 2m and so on. A unique ID at the
highest resolution of level 4 is a combination of the level 1 ID, the level 2 ID and the level 3 ID, and follows the
system used by many countries for their census data (e.g. FIPS in the USA, Output areas in the UK – see Table1).
Table 1. The census areas in Sahsuland

Sahsuland
Level 1
Level 2
Level 3
Level 4

Administrative area
UK
District
Standard table Wards
(ST Wards)
Tract
Super Output Areas
(SOAs)
Census Block Group
Census Output Areas
(OAs)
USA
State
County

Canada
Province (PR)
Census Division (CD)
Census-subdivision
(CSD)
Dissemination Area
(DA)

All level IDs are stored as text values with LEVEL1 areas using two characters, LEVEL2 use 3 characters, which is
joined with the LEVEL1 unique ID to make a LEVEL 2 unique ID of 2 and 3 characters separated by a dot. The Level 3
units use a 6 character value and the LEVEL4 is a single character. Again, unique IDs for each region are achieved by
concatenating each lower resolution area such that the proceeding level falls within each separated by a single dot.
Note. The data formats described in this section refer to Sahsuland data only. These data formats are not a
requirement by the RIF. Data requirements are covered in the RIF Data Loading manual.

Age groups are actually by one year age group for ages 0 to 4, then by five year age groups from ages 5 to 85, e.g. age groups 0,
1, 2, 3, 4, 5-9, 10-14, …,80-84, 85+
2
The RIF can handle ecological level covariate data

Screen shots and examples in this manual are based on this Sahsuland data. Descriptive statistics of Sahsuland can
be found in appendix B.

3.2 Logging in
Your RIF administrator should provide you with your user name, password and the correct URL to access the login
page of your RIF installation. Type the URL in the address bar of your web browser and log in using your username
and password.

Figure XXX. The RIF login screen.

3.3 RIF mapping tools
The RIF uses an internet browser based map viewer to display and select study areas and to visualise results. These
maps work in the same way as conventional map containers (e.g. Google Maps), the main difference being they use
open source map data.
The following controls are common to all RIF maps. Other tools are specific to certain RIF functionality and these will
be dealt with in the relevant sections.

Zoom in: Zoom map in.

Zoom out: Zoom map out.
Quick export map: Save the current map in view as a png file.
Zoom to selection: Zoom the map to the currently selected districts(s).
12

Zoom to study extent: Zoom the map to fit all districts used in this RIF study.
Zoom to full extent: Zoom the map to fit all districts in the current geography.
Clear selection: Deselect all currently selected districts.
Transparency: Change the transparency (opacity) of the district layer being mapped
Enter address: Zoom the map to a place name, geographical feature etc.
Full screen: Display the map in full screen mode (Esc to exit)

Attribution: Attribution (source, copyrights) information for the map layers
Hide or show selection shapes: Display the shapes used to select the study. Green when displayed
Base map: Opens the base map selection where the base map can be changed or removed.

Figure XXX. Base map settings.

4. Running a new RIF study
To create, then run a new RIF study, five steps will need to be completed:
•
•
•
•
•

Enter study details such as ‘geography’ and study type
Defining the study area
Defining the comparison area
Set the investigation parameters
Decide the statistical methods to be used

These steps can all be completed under the Study Submission tab. At any point all the details of the study can be
cleared by clicking the reset link.
13

Figure XXX. Study Submission screen

4.1 Study details
The study details can be completed using the fields along the top of the study submission tab
Study Name. The study must be given a name which is types into the study name field. The name cannot exceed 20
characters in length.
Health Theme. This is defined during data loading as a means to group your relevant data sets together for ease of
use. This will usually relate to a disease type, e.g. cancers.
Geography. Select the appropriate geography from the drop-down list of geographies (this will usually relate to the
either the country in which your study is for or to a predefined representation of districts e.g. tracts, wards). The list
consists of all the available geographies which you can access.
Numerator. Select the health outcome you are interested in mapping.
Denominator. The data to be used as a denominator. This cannot be changed and is auto-selected depending on
which numerator table is being used. Relevant numerator-denominator pairs are decided in the data loading process.
•

If you hover the mouse over the field name a detailed description will be displayed if available.

4.2 Study area
Clicking the study area link will load the study area selection screen

Figure XXX. Study area selection screen
The first thing to do is to set whether this is a disease mapping study or a risk analysis study using the switch at the
top-left of the window. (See section 2.1 for more information on these study types).
Initially the whole geography is displayed in the map area at the default level of resolution. The purpose of this screen
is to select which areas of the geography are to be investigated. For risk analysis, one to six bands (of multiple districts)
may be specified. Disease mapping studies are not banded in this way, so only one selection band is available as
default.
Area selection/deselection can be performed in a number of ways:
•
•
•
•
•

Use the band drop-down to specify the band number
Clicking directly onto the area in the map
Selecting areas from the list displayed on the left side of the screen
Selecting areas by defining a freehand polygon using the
button
Select area within concentric bands using the
button.

Note that selections using the boundary of another polygon based on intersection with a district's centroid. The
icon will toggle visibility of these centroid locations off and on.
The

icon will select all the districts within a geography

By clicking the 'Upload from list' button, a comma deliminated (csv) file can be used to select a predefined list of
districts and bands. This must have just thee columns ID, NAME and Band; e.g. seer_mainland_states.csv.
ID,NAME,Band
01779778,California,1
01779780,Connecticut,1
01705317,Georgia,1
01779785,Iowa,1
01779786,Kentucky,1
01629543,Louisiana,1
01779789,Michigan,1
01779795,New Jersey,1

00897535,New Mexico,1
01455989,Utah,1
01779804,Washington,1

A zipped shapefile can also be used to define study areas. The select by shapefile
icon brings up the open shapefile
dialogue. The file can be points or polygon (see below) and must be in a zipped folder with extension .zip.
The select by postal code/WGS/grid coordinates
icon allows the user to enter a single point as a postal (or ZIP)
code, WGS 84 (GPS) coordinate or using national coordinates. Postal codes are only available if the necessary lookup
data has been loaded and setup (see the data loading manual).
The
Hide or show selection shapes icon allow the user to display the shapes used to select the study. It is green
when shapes are displayed
Once displayed on the map, the uploaded layers can be removed with the
•
•
•

clear selection icon.

Multiple layers are supported.
When loading point shapefile (e.g. incinerator locations), radii for exposure band(s) in metres need specified.
When loading a polygon shapefile (e.g. output from an exposure model). Selections can be defined by the full
extent of the areas with the shapefile, by a cut-off from an attribute within the file (e.g. a threshold value for
a pollutant). This may be multiple (descending) cut-offs for risk analysis according to bands. In addition, in the
case of a risk analysis study, if the polygons have a band attribute, this may be specified as well.

Clicking ‘done’ will store the selected study regions which define the study area and return the user to the study
submission tab. The study area ‘tree’ should now be coloured to indicate that this part of the study submission is
complete.

4.3 Comparison area
Clicking the comparison area tab on the study submission tab will load the comparison area selection screen for
defining areas (populations) for the calculation of indirectly standardised risks. The comparison area screen is very
similar to the study area selection screen with all the same methods of selection study regions that will for the
comparison area. A comparison area is not banded in the same way as the study area, so by default only one selection
band is possible. Note that the type of study can only be defined via the study area window.
Clicking ‘done’ will store the selected study regions which define the comparison area and return the user to the study
submission tab. The comparison area tree should now be coloured to indicate that this part of the study submission is
complete.

4.4 Investigation parameters
Clicking the investigation parameters link brings up the investigation parameters selection screen.

Figure XXX. Investigation parameters screen.
ICD10 codes of interest can be selected from the list and search box on the left hand side of the screen. Selected codes
appear in the lower right side of the screen. An investigation name can be edited in the investigation name field. The
covariates section allows the user to select male, female or both sexes and define one further covariate field. The
range of ages and years the study is to cover are chosen in the age range and year range sections respectively.
Clicking ‘done’ will store the selected parameters and return the user to the study submission tab. The investigation
parameter tree should now be coloured to indicate that this part of the study submission is complete.

4.5 Statistical methods
Clicking the statistical methods link brings up a screen allowing selection of the statistical methods the RIF will for the
disease mapping process. By default, the RIF always calculates the indirectly standardised rates and the relative risk
ratios as well as performing Empirical Bayesian Smoothing. Either select one of the available procedures or the option
not to apply smoothing if you do not to run n additional Bayesian method.
For risk analysis only the default is supports, the user cannot perform an additional Bayesian method.
Details of the basic statistical methods and the full Bayesian smoothing options along with external references are
included in the Technical Appendix.
Clicking ‘done’ will store the methods selected and return the user to the study submission tab. The statistical methods
tree should now be coloured to indicate that this part of the study submission is complete.

Figure XXX. Statistical methods selection screen.

4.6 Saving and reloading studies
At any point during the process of defining a new study submission, the details of the study can be saved locally on
the user’s machine by clicking the save study link. This brings up a save as dialog box allowing the user to select a local
folder and file name, then click ‘save’ to write a local copy of the study.

Figure XXX. Open from file dialog box for reloading a study during the submission process
At a later date the user can load the study by clicking the open from file link and navigating to the location of a
locally stored file and clicking open. Depending on the browser being used, the name of the file can be changed.
The file is in “human readable” JSON5 format and can be edited with care. Saved study setup files are also produced
as a part of the export to ZIP functionality.

4.7 Study status
The status link in the bottom, centre of the study submission screen brings up the submission status screen which
lists all the user’s studies that have been previously submitted and are still available to access on the system. It also
shows any studies that have recently been submitted but are not yet fully processed yet. The study state column
provides the user with a short code to identify the status of the study:
•
•

Study state: S means the study results have been computed and are ready to be used.
Study state: F means the study failed in the R code. The trace button will display the R error in a popup.
18

•
•

Study state: C means the study has been created but not verified.
Study state: E means the study has been extracted but results have not been generated.

Entries that are highlighted in pink are not available for mapping as they are currently being processed or have failed
to be run.

Figure XXX. Submission status screen

4.8 Run study
Once the user is happy that all the study submission details are complete, the run study link can be clicked which
brings up the run study screen. Here an optional description can be added and the details of the study can be checked
by clicking the view study submission summary link. The study will be submitted when the user clicks ‘run’ and a
message will be displayed: ‘Success: study submitted’.

Figure XXX. Run study screen

Once the study has been submitted, the use will be returned to the study submission screen. The study may take some
minutes to run depending on the size and complexity of the study. Clicking the status link allows the user to see the
status of any recently submitted studies. When the status is listed as R the results of the study can be viewed under
the data viewer tab.
View study submission summary provides a report of the study as submitted

Figure XXX. View study submission summary screen

4.8 Messages
RIF messages appear as strips from the top of the screen and disappear after five seconds unless they are a serious
error and related to submitting or running a study. The messages screen allows the user to view the messages for
the session.

Figure XXX. Messages screen

4.10 Reset
The reset button clears the study selection.

5. Data viewer
The results of a study that has been submitted and run can be examined and analysed in detail under the data
viewer tab. The header of the data viewer has two dropdown boxes allowing the user to select the study to view and
to filter the results by sex. The main data viewer area is made up of four sections. Moving clockwise from top right,
these are:
•
•
•
•

configurable choropleth map of the study area
data table of the regions included in the study area
population pyramid of the whole study area
frequency distribution of the outcome currently displayed in the choropleth map.

Figure XXX. An example disease mapping study in the data viewer tab.

5.1 Choropleth map
The array of buttons above the map give the user several options to navigate the map and configure what
information is displayed.
The choropleth map icon
brings up the choropleth map symbology screen which allows the user to select which
field is displayed on the map and how the values in the field are represented using colours on the map.

Figure XXX. Choropleth map symbology screen.
The colour scale dropdown selection lets the use choose from many different colour scales as defined by the Color
Brewer project (http://colorbrewer2.org). The field to map dropdown lets the user select which data field is displayed
on the map and in the frequency distribution graph. The intervals dropdown lets the user define the number of breaks
in the data being displayed (the maximum is 11 but some colour scale have a lower limit). The classification dropdown
lets the user define how the breaks in the data are defined. In each case the number of classifications is defined by
the intervals dropdown.
•
•
•
•

•
•

Quantile. Divided the data into quantiles with the same number of data points in each quantile.
Equal interval. The difference between the lowest and highest values is divided equally into categories.
Jenks. Uses the Jenks natural breaks classification method designed to determine the best arrangement of values
into different classes (Jenks, 1967).
Standard deviation. Calculated the mean and standard deviation of the underlying data, then divides the data into
5 categories (>2 standard deviations above/below mean, >1,<=2 standard deviations above/below mean, <=1
standard deviation away from mean). Always 5 categories.
Atlas relative risk. The classification system used when displaying relative risks in the SAHSU Environment and
Health Atlas (ref). Always 9 intervals.
Atlas probability. The classification system used when displaying probabilities in the SAHSU Environment and
Health Atlas (ref). Always 3 intervals.

Additionally, the breaks defined by the classification method can be manually edited in the edit breaks section.

5.2 Data table
The data table shows one row for each of the regions that make up the study area as selected during the study
submission process. Clicking on rows in the data causes the corresponding regions to be highlighted in green on the
choropleth map. The data table shows the population and health data (Area Id, Ban Id, Observer, Population,
investigation id); basic statistics (Expected, Adjusted, Relative Risk, Lower95, Upper95); and the results of the
Bayesian smoothing (Posterior probability, Smoothed Smr, Smoothed Smr Lower95, Smoothed Smr Upper95).

The data can be sorted ascending or descending by clicking on the column headings. There are filter boxes directly
under the column names. Typing in a filter box will filter the results displayed in the data table. Note that the filters
work using string filtering, i.e. typing 10 in the ‘Band Id’ filter will show all the rows that have the string ‘10’ in the
band id (e.g. 10, 100, 101, 110).
The link to the right of the column heading gives the user an additional three menu options:
•
•
•

Clear all filters. Removes an filter strings that have previously been entered.
Export all data as csv. Allows the user to save a comma separated variable (csv) file of all the study data.
Export visible data as csv. Allows the user to save a comma separated variable (csv) file of all the records
currently in the data table.

Clicking on regions in the choropleth map or selecting the rows in the data table sets the value for that row in the
‘selected’ column to 1 (as well as highlighting the row). By filtering the data table so that it only shows the records
where ‘selected’ is 1, then choosing ‘export visible data as csv’, the user is able to effectively make a manual
selection of regions in the map and export only the data associated with those selected regions.

5.3 Population pyramid
The population pyramid section displays a population pyramid showing the age distribution of the residents of the
geography from which the study is taken. The dropdown box in the top right of the population pyramid shows the
years for which there is population data and allows the user to view the population pyramids for the available years.

Figure XXX. Population pyramid.

5.4 Frequency distribution
For disease mapping studies the frequency distribution histogram shows the distribution of the data field currently
being displayed in the choropleth map.

Figure XXX. Distribution histogram

5.5 Risk Graphs
For risk analysis studies the risk graph shows the distribution of the data field currently being displayed in the
choropleth map.

Figure XXX. Risk Graph

5.5 Info Button
The data viewer, mapping and info screens all have an info button to the right of the gender chooser:

This allows the user top select different reports for a study:
25

1.
2.
3.
4.

Summary
Covariate Loss Report
Homogeneity Tests
Risk Graphs

Option 2 requires the study to use covariates and is stratified by covariate name. It also provides information on
extract verification. Options 3 and 4 are for risk analysis studies only. Option 4 allows the users to interact with the
risk graph. Up to three risk factors can be displayed: band, overage exposure and distance from source. To view one
gender, set both gender selectors to the same choice. Hovering over the value circle displays the confidence limits.

Figure XXX. Study Summary

Figure XXX. Covariate Loss Report

Figure XXX. Homogeneity Tests

Figure XXX. Risk Graphs with confidence limit displayed

6. Mapping
The mapping tab allows the user to compare two studies side-by-side or different data from the same study in two
different maps. The mapping screen is divided vertically in two to give a left display and a right display. The header
for each display allows the user to select the study and sex of the data displayed in the area below.
It is usually used for disease mapping. Different studies may be compared but they must share the same geography.

Figure XXX. Disease mapping tab.

6.1 Choropleth maps
Choropleth maps are displayed for both the left and right display with the same navigation and selection functions as
the choropleth map in the data viewer tab (see section 5.1). In addition there additional functions to facilitate the
simultaneous use of two maps.
Lock and unlock selection. When locked, clicking on the map region in the left display will also highlight the same
region in the right display and vice-versa. If the studies selected for the left and right displays have been defined at
different geographical levels, lock selection will not work and a warning will be displayed.
Lock and unlock map extents. When locked, both maps will always display the same extents. Zooming or scrolling
on one map will cause the other map to move such that both maps display the same area.
Copy symbology. Copies the symbology settings from the left display to the right display.
The
Hide or show selection shapes icon allow the user to display the shapes used to select the study. It is green
by default when shapes are displayed
Maps are selection

Lock and extent

Locked by default.

6.2 Disease map charts
For disease mapping studies the charts displayed below the maps summarise the risk field data across the whole
study area. The charts display all the values of the risk field show in the map above as well as the upper and lower
28

confidence intervals. The data in the chart is ordered from lowest to highest risk (moving left to right). Clicking on a
point in the chart moves the red line to that data point, displays the risk and confidence intervals above the chart
and selects the same region in the map above. Similarly, clicking on the region in the map moves the red line to the
equivalent data point in the chart.

Figure XXX. Disease map chart.
The smaller chart displayed below the main chart acts as a navigation panel for the main chart above. By moving the
mouse so it is above the left or right edge of the shaded area in the navigation panel, the user can click and drag to
make the shaded area narrower or wider. This will increase or decrease the zoom in the main chart above.

Figure XXX. Disease map chart showing how to alter the zoom.
For risk analysis studies the risk graphs are shown for males and females.

7 Export
The export tab allows a user to export the results of a completed study as a zip file. Select the study to display using
the dropdown at the top of the page. Initially a preview of the extract (input data) and map (result) tables are
shown. Enter a range of rows and click the refresh button to preview further rows. The map container shows either
the study or comparison area.
On clicking 'Export study tables', the full map and extract tables are downloaded as csv files and the study and
comparison areas are downloaded in GIS format (geoJSON) at the specified detail level. All files are saved in a zipped
folder prefixed with the study name and date in your specified output folder (see the RIF set up instructions for how
to change this). Then button changes top ‘Exporting’ whilst the export is underway. When the export is completed
the button changes to ‘Download Study Export’.
‘Save completed study’ allows the user to save the completed study setup as a JSON5 file suitable for upload an
modification in another study.
29

Figure XXX. Export tab.
The export ZIP file contains the following data:
•
•
•
•
•
•
•
•
•

Study extract, results and adjacency matrix in CSV form;
R scripts to re-run statistics phase
Shapefiles of results and the 2nd geolevel (e.g. state boundaries)
Results as geoJSON;
Geography: study and comparison areas as JSON and as shapefiles;
Maps: data viewer and left and right disease mapping panes;
Reports: denominator population pyramids by year;
Saved study JSON5 file (to re-run study);
An HTML report to integrate the above data.

7.1 R Scripts
Supplied R scripts allows users to re-run statistics phase by running rif40_run_R.bat in the data directory

Figure XXX. RE-running the R scripts.
30

7.2 Shapefiles
ESRI standard shapefile are part of the export which contain males, females and both (males and females) data:
•
•
•
•

Column names are shortened to fit DBF file rules;
Results are rounded to 2 decimal places for sane quantizing;
Per map styling is supplied as style layer descriptors (.sld) files so maps can be easily re-created in GIS tools.
Beware the .sld file does not contain a filter. The user must supply their own for males, females or both;
Projection used is the original administrative boundary projection (i.e. usually the normal for the country).

Figure XXX. Creating a map from a shapefile.

7.3 Generated Maps
The Zip file contains generated maps with excellent resolution to aid the user in producing their own maps.

Figure XXX. Supplied maps.
31

Maps are produced in the following formats:
•
•
•
•
•

PNG;
GeoTIFF. Can be used as a raster layer in GISTools. (.prj and .tfw files World Map format files are also created
for each map). GeoTIFF can have copyright embedded;
SVG. This is currently a single layer and therefore not easily editable;
JPEG;
EPS (encapsulated PostScript) and Postscript.

Map by default are 7480 pixels wide @100 dpi (Elsevier full page: can be changed). The extent of the study is only is
mapped. Maps are scaled with 3% extra margins and 10-50% left margin (for the Legend) depending on the aspect
ratio, and then expanded to the grid resolution (e.g. 10 degrees). The grids can be turned off. The projection used is
the original administrative boundary projection (i.e. usually the norm for the country).
Printing setup is managed system wide. To manage the printing setup see section: 8.4.2 Printing Defaults of the RIF
Web Application and Middleware Installation manual.

7.4 Reports
Population denominators are provided as high-quality graphics in both tree and pyramid forms:

Figure XXX. Tree population pyramid.

Figure XXX. Population Pyramid.

Appendices
Appendix A. Statistical methods
Disease mapping
Disease maps aim at representing the geographical distribution of the incidence of disease. In the frame of the RIF
software, only maps produced with counts of data are considered. Counts of disease cases are reported for a list of
regions, denoted here areas, and delimited by geographical boundaries. The easiest way to map the geographical
variations of the disease would be to directly map the counts. However, these counts depend strongly of the age-sex
composition of the population at risk within each area, and cannot be directly compared. Consequently it is necessary
to apply the use of standardisation and standardised rates to exclude the effect of populations. Standardisation
requires the definition of a ‘comparison’ population associated to each area. The comparison population may be the
total population of all study areas, or subsets of study areas.
Standardised disease rates in populations can be calculated using direct and indirect standardisation. Direct
standardisation involves applying the disease rates found in the study areas to a standard population. This is not always
available so, currently the RIF version 4.0 does not attempt direct standardisation of disease rates. When applying
indirect standardisation, the standard disease rates from a comparison population are applied to the study population
to give the expected disease counts such as the standardised mortality rates (SMR) or the standardised incidence
rates (SIR).
Indirectly standardised risks
The standard disease rates are taken from the regions defined by the comparison area when the study was submitted.
Cases occurring in the comparison population are located for male and females in each five year age band and per
covariate. Standardised rates of disease in the comparison population are calculated for each gender, age and
covariate by dividing the number of cases in each group by the total population in each corresponding gender, age
and covariate stratum. The comparison area populations need to be large enough such that the age-sex-covariate
specific disease rates are reliable. These standardised rates (rj*) are applied to the study area population strata (j) to
calculate the expected counts (Ei),
𝐸𝑖 = ∑ 𝑁𝑖𝑗 𝑟𝑗∗
𝑗

Where Nij is the study population in strata j of area i. The SMR for the study population(s) is then simply given by the
ratio of the observed (Oi) to expected counts (Ei):
𝑆𝑀𝑅𝑖 =

𝑂𝑖
𝐸𝑖

Values of the relative risk larger than one indicate an excess of risk relatively to the underlying ‘comparison
population’, whereas values smaller than 1 indicate a deficit of risk. Since each observation, is divided to the expected
counts given the structure of the population, this variable has no unit, and comparisons between areas can be done.
The risks obtained for two or more study populations (e.g. different ‘bands’ of exposure around a putative source of
pollution), should not be directly compared as they are not based on the same standard population (i.e. the age,
gender and covariate make up between the population being compared are not exactly the same).
The uncertainty associated with the SMR estimate is quantified by calculating the 95% confidence. Here we note that
relative risk 𝑅𝑅𝑖 is the parameter of a Poisson distribution,
𝑂𝑖 ∼ 𝑃(𝐸𝑖 𝑅𝑅𝑖 ).
•

If 𝑂𝑖 < 100, confidence intervals are found by using the Chi-squared method.
33

𝛼

For instance, if 𝑅𝑅𝑖𝑈 is the upper bound, then, by definition, for any 𝑋𝑖 ∼ 𝑃(𝐸𝑖 𝑅𝑅𝑖𝑈 ), 𝑃(𝑋𝑖 > 𝑂𝑖 ) = 2 .
2
We know that 𝑃(𝑋𝑖 ≤ 𝑂𝑖 ) = 1 − 𝑃(𝑌 ≤ 2 𝐸𝑖 𝑅𝑅𝑖𝑈 ), with 𝑌 ∼ Χ 2𝑂
. Consequently,
𝑖 +2

2
𝜒2𝑂
+2

upper 95% CI = 2𝐸 𝑞𝛼/2 𝐼
𝑖

lower 95% CI =

2
𝜒2𝑂
1
𝐼+2
𝑞
.
2𝐸𝑖 1−𝛼/2

For confidence level 100(1-α)%.
•

If 𝑂𝑖 ≥ 100, a Gaussian approximation of the log relative risk is done, log(𝑂𝑖 /𝐸𝑖 ) is assumed to follow a
Gaussian distribution with mean log(𝑅𝑅𝑖 ), and variance 1/𝑂𝑖 . Then,
𝑂𝐼 /𝐸𝐼

lower 95% CI =

upper 95% CI = 𝐸𝐼 × exp (1.96√𝑂 )

exp(1.96√1/𝑂𝑖 )
𝑂

𝐼

𝑖

Empirical Bayes Analysis
The maps of the standardised mortality or incidence ratio may lead to misinterpretations, since the extreme values
are more often the consequence of small counts than a true extreme relative risk. Consequently, a non-significantly
positive standardized mortality risk may be higher than a significant one for which the population at risk is higher. To
reduce the influence of the small counts, Clayton and Kaldor (1987) proposed empirical Bayes estimates of the relative
risk. They are based on a Poisson-Gamma hierarchical model. By accounting for differential variability in the data, this
hierarchical approach provides more precise estimates of relative risk and more accurate assessments of significant
changes that the standard methods. The estimates are smoothed toward a global value by assuming that all the
relative risks are sampled in the same gamma distribution. Moreover, the smaller the count, the stronger the shrinkage
effect. In detail, relative risks 𝑅𝑅𝑖 are assumed to come from a single Gamma distribution of scale 𝛼 and shape 𝛽,
𝑂𝑖 ∼ 𝑃𝑜𝑖𝑠𝑠𝑜𝑛(𝐸𝑖 𝑅𝑅𝑖 )
𝑅𝑅𝑖 ∼ 𝐺𝑎𝑚𝑚𝑎(𝛼, 𝛽).
Approximation of the posterior mean of relative risk 𝑅𝑅𝑖 are given by the empirical Bayes estimates,
𝐸(𝑅𝑅𝑖 |𝑂𝑖 , 𝐸𝑖 , 𝛼̂, 𝛽̂ ) =

𝑂𝑖 + 𝛽̂
𝐸𝑖 + 𝛼̂

where 𝛼̂ and 𝛽̂ are the maximum likelihood estimates of 𝛼 and 𝛽.
In practice, these estimates of the relative risk are obtained through the following iterative procedure:
1. Start with initial values for the relative risk 𝑅𝑅𝑖 . For instance
̂ 𝑖 = 𝑂𝑖 .
𝑅𝑅
𝐸𝑖

2. Obtain estimators 𝛼̂ and 𝛽̂ using equations
𝑛

𝛼̂ 1
̂𝑖
= ∑ 𝑅𝑅
𝛽̂ 𝑛
𝑖=1

and
̂
𝛼
𝛽̂ 2

̂
𝛽

̂
𝛼

̂𝑖 − 2 )
= 𝑛−1 ∑𝑛𝑖=1(1 + 𝐸 ) (𝑅𝑅
𝛽̂
𝑖

3. Obtain new estimated values for the relative risks
̂𝑖 =
𝑅𝑅

𝑂𝑖 + 𝛽̂
𝐸𝑖 + 𝛼̂

4. Repeat steps 2 and 3 until estimated values for 𝛼̂ and 𝛽̂ do not change significantly.
Full Bayesian smoothing
If some degree of spatial dependence of the risk is assumed, meaning that the risks in close areas are similar, we can
estimate the risk in one area with information borrowed from its neighbours. Since these estimates are based on more
information than SMR estimates, they are more robust. The BYM model (Besag, York, and Mollié 1991) was developed
to address this issue of spatial dependence of risk. First a neighbourhood structure of the study area must be defined
which specifies the neighbour relationships between areas. Here, we denotes 𝜕𝑖 the set neighbours for area 𝑖. The log
relative risk is then split into two terms, 𝑢𝑖 a spatial term which accounts for the spatial variations within the risk, and
𝑣𝑖 a noise which accounts for independent local variations,
log(𝑅𝑅𝑖 ) = 𝑢𝑖 + 𝑣𝑖 .
The spatial term is modelled with an intrinsic conditional auto-regressive (CAR) model, meaning that for each area i,
𝑢𝑖 is given conditionally to its neighbours. Specifically, it is assumed to follow a normal distribution with mean, the
neighbour mean and variance, a variance parameter, 𝜎𝑢2 , divided by the number of neighbours, 𝑛𝑖 ,
𝑢𝑖 |𝑢−𝑖 ∼ 𝑁 (

∑𝑘∈𝜕𝑖 𝑢𝑘 𝜎𝑢2
, ).
𝑛𝑖
𝑛𝑖

The larger the number of neighbours, the smaller the variance. The Gaussian noises, denoted vi, are supposed to be
independent and identically distributed with variance 𝜎𝑣2 .
One can consider the map of the uniquely spatially structured term given by (exp(𝑢𝑖 )). They display uniquely the part
of the risk which has a smooth spatial distribution. The independent term is then seen as residual. But one can also
consider the estimate of the entire relative risk which leads to robust estimate thanks to the spatial CAR term, but also
allows individual variability.
Other models are also proposed in the RIF, the ‘CAR’ model, in which the log relative risk is only modelled by the
conditional autoregressive 𝑢𝑖 term. In this model, the relative risk is assumed to be spatially smooth without any local
independent variations. In the CAR and the BYM model, estimates are smoothed towards a local mean.
Finally the ‘HET’ (heterogeneous) model is also proposed. In this model, the log relative risk is only composed of the
independent term vi. As for empirical Bayes estimates, with this model the log relative risk estimates are smoothed
toward a global mean. These two models are thus similar, they only differ by their prior (Gaussian vs gamma), and
their inference method (fully Bayesian vs empirical Bayes).
Prior specification
Minimally informative independent inverse gamma 𝐼𝐺𝑎𝑚𝑚𝑎(0.5,0.0005) priors are assigned to the model
parameters 𝜎𝑢2 and 𝜎𝑣2 .
R and R-INLA
The statistical calculations described above are performed by the RIF server calling an instance of a R procedure (R
Core Team, 2015). The full Bayesian smoothing is performed using the integrated nested Laplace approximations
(INLA), proposed by Rue et al. (2009) . Whereas Bayesian inference often makes use of Markov Chain Monte Carlo
(MCMC) simulation methods (Casella and George, 1992), the increasing size and related high spatial resolution of the
datasets supported by the RIF mean that even state of the art, high powered servers would take several days to
35

perform Bayesian inference via MCMC. Since INLA uses a deterministic algorithm it produces accurate results much
faster than MCMC methods (Blangiardo and Cameletti, 2015). The INLA functionality is delivered through and R
package called R-INLA. The website www.r-inla.org is a useful source of further information; it provides many papers,
tutorials and examples that assist in the understanding and implementation of INLA.

Appendix B. Descriptive analysis of Sahsuland
Sahsuland is a fictitious island nation of approximately 32860 km2 comprising 4 different hierarchical levels of
geography.

Figure XXX. Sahsuland
Sahsuland population
Population data is available for the period 1989-1996 in five year age groups from ages 5 to 85 and one year age
groups for the ages 0 to 4.
37

Table XXX. Sahsuland population by age.
Year
1989
1990
1991
1992
1993
1994
1995
1996

0-4
622,270
631,650
642,260
644,900
639,190
637,090
628,860
617,540

5-9
600,646
602,524
607,206
616,862
631,042
645,432
655,334
661,308

10-19
1,283,290
1,261,454
1,239,456
1,223,550
1,216,676
1,212,252
1,220,428
1,235,714

20-39
2,893,970
2,899,864
2,914,374
2,941,456
2,949,324
2,966,698
2,979,138
2,964,174

40-59
2,380,576
2,416,484
2,450,368
2,477,394
2,508,448
2,548,060
2,579,936
2,610,400

60-79
1,961,654
1,960,170
1,961,088
1,954,206
1,942,956
1,931,200
1,925,502
1,922,848

80+
460,856
472,292
484,834
497,464
511,690
523,196
533,890
538,526

Total
10,203,262
10,244,438
10,299,586
10,355,832
10,399,326
10,463,928
10,523,088
10,550,510

Table XXX. Suhsuland population by level 2 geography (17 divisions, equivalent to US Counties)
Year

Population
Min

Max
1st

1989
1990
1991
1992
1993
1994
1995
1996

7,848
7,942
8,060
8,076
8,096
8,096
8,100
8,138

4,531,618
4,546,944
4,561,132
4,584,918
4,604,640
4,629,846
4,659,062
4,668,972

59,719
60,344
60,948
61,080
61,194
61,260
61,357
61,315

Divisions
Quartiles
Number with pop. less than:
Median
3rd
100,000 500,000 1 million
145,340 620,080
6
13
14
146,260 621,506
6
13
14
146,930 625,866
6
13
14
147,852 628,008
6
13
14
148,660 629,334
6
13
14
149,888 630,394
6
13
14
150,888 631,606
6
13
14
152,100 633,123
6
13
14

Table XXX. Suhsuland population by level 3 geography (202 divisions, equivalent to US Tract)
Year

Population
Min

Max
1st

1989
1990
1991
1992
1993
1994
1995
1996

2,760
2,820
2,856
2,880
2,900
2,950
2,980
3,016

203,848
206,622
209,214
208,694
208,924
209,526
210,038
209,140

26,000
26,271
26,484
26,621
26,612
26,843
27,207
27,397

Quartiles
Median
3rd
42,746
62,519
42,917
62,459
43,119
63,308
43,297
63,684
43,458
64,134
43,746
64,505
44,281
64,837
44,457
65,221

Divisions
Number with pop. less than:
10,000
50,000
100,000
9
122
183
9
123
183
8
124
183
8
123
183
8
123
183
8
121
182
8
121
182
8
121
182

Table XXX. Suhsuland population by level 4 geography (1230 divisions, equivalent to US Census Block Group)
Year

Population
Min

Max
1st

1989
1990
1991
1992
1993
1994
1995

128
144
146
146
154
160
156

33,588
33,662
34,150
34,226
34,516
34,622
34,806

3,567
3,578
3,610
3,636
3,648
3,702
3,744

Quartiles
Median
3rd
6,120
10,759
6,091
10,656
6,148
10,760
6,182
10,774
6,217
10,724
6,300
10,804
6,323
10,842
38

Divisions
Number with pop. less than:
2,500
5,000
20,000
58
517
1122
58
515
1121
58
512
1120
55
509
1119
52
506
1119
47
501
1117
46
499
1116

1996

146

35,092

3,750

6,347

10,894

495

1116

Sahsuland numerator data
Numerator data consists of cancer incidences. For the early years (1989 – 1994) the data is in 45 different ICD-9
codes. Until a taxonomy service is written for ICD-9 codes, this data is cannot be used. The years 1995 and 1996 have
cancer data for 41 different ICD-10 codes.
Table XXX. Total cases in Sahsuland, ICD-10 codes and health conditions covered by the Environment and Health
Atlas*.
ICD-10 Description
C220
C221
C223
C229
C33
C340
C341
C342
C343
C348
C349
C500
C501
C502
C503
C504
C505
C506
C508
C509
C64
C65
C670
C671
C672
C673
C674
C675
C676
C678
C679
C710
C711
C712
C713
C714
C715
C716
C717
C718
C719

Liver cell carcinoma
Intrahepatic bile duct carcinoma
Angiosarcoma of liver
Malignant neoplasm of liver, not specified as primary or secondary
Malignant neoplasm of trachea
Malignant neoplasm of main bronchus
Malignant neoplasm of upper lobe, bronchus or lung
Malignant neoplasm of middle lobe, bronchus or lung
Malignant neoplasm of lower lobe, bronchus or lung
Malignant neoplasm of overlapping sites of bronchus and lung
Malignant neoplasm of unspecified part of bronchus or lung
Malignant neoplasm of nipple and areola
Malignant neoplasm of central portion of breast
Malignant neoplasm of upper-inner quadrant of breast
Malignant neoplasm of lower-inner quadrant of breast
Malignant neoplasm of upper-outer quadrant of breast
Malignant neoplasm of lower-outer quadrant of breast
Malignant neoplasm of axillary tail of breast
Malignant neoplasm of overlapping sites of breast
Malignant neoplasm of breast of unspecified site
Malignant neoplasm of kidney, except renal pelvis
Malignant neoplasm of renal pelvis
Malignant neoplasm of trigone of bladder
Malignant neoplasm of dome of bladder
Malignant neoplasm of lateral wall of bladder
Malignant neoplasm of anterior wall of bladder
Malignant neoplasm of posterior wall of bladder
Malignant neoplasm of bladder neck
Malignant neoplasm of ureteric orifice
Malignant neoplasm of overlapping sites of bladder
Malignant neoplasm of bladder, unspecified
Malignant neoplasm of cerebrum, except lobes and ventricles
Malignant neoplasm of frontal lobe
Malignant neoplasm of temporal lobe
Malignant neoplasm of parietal lobe
Malignant neoplasm of occipital lobe
Malignant neoplasm of cerebral ventricle
Malignant neoplasm of cerebellum
Malignant neoplasm of brain stem
Malignant neoplasm of overlapping sites of brain
Malignant neoplasm of brain, unspecified
39

Cases
1995
162
186
2
74
22
422
1,164
120
572
74
4,258
220
506
430
190
1,596
352
50
118
3,914
1,014
78
26
14
186
44
88
42
64
142
2,500
92
188
98
146
32
10
50
26
42
134

1996
156
182
4
96
6
496
1,166
178
648
38
3,728
206
452
526
238
1,596
358
78
176
3,954
862
48
42
22
110
30
92
40
38
84
2,390
52
136
78
72
30
14
40
14
12
306

Total

19,448

18,794

EHA health conditions
C34
Lung cancer
6,610
6,254
C50
Breast cancer
7,376
7,584
C67
Bladder cancer
3,106
2,848
C71
Brain cancer
818
754
C22
Liver cancer
424
428
* There are no cases of prostate cancer, skin cancer, Leukaemia or mesothelioma in the Sahsuland data.
Table XXX. Summary of cancer cases for 1995, by level 2 geography. All cases, top 6 individual ICD codes and health
conditions covered by the Environment and Health Atlas
ICD-10
code

Cases
Min

Max
1st

All
C349
C509
C679
C504
C341
C64
EHA
C34
C50
C67
C71
C22

8
2
4
0
0
0
0

9,000
2,112
2,218
1,216
592
488
466

119
18
21
12
7
6
5

2
4
2
0
0

3,060
3,528
1,350
366
180

37
46
20
2
2

Quartiles
Median
3rd
10
298
1,255
1
60
279
2
56
219
2
22
172
3
26
125
6
20
67
5
10
67
7
88
120
44
10
8

419
444
235
59
28

1
1
3
7
9

Divisions
Number with cases less than:
50
150
500
1500
2
6
10
13
7
11
15
16
8
12
16
16
10
13
16
17
10
13
16
17
12
15
17
17
12
16
17
17
6
6
9
13
15

10
9
13
16
16

13
13
16
17
17

16
16
17
17
17

Table XXX. Summary of cancer cases for 1996, by level 2 geography. All cases, top 6 individual ICD codes and health
conditions covered by the Environment and Health Atlas
ICD-10
code
All
C349
C509
C679
C504
C341
C64
EHA
C34
C50
C67
C71
C22

Cases
Min

Max

10
2
4
0
0
0
0

8,962
1,794
2,174
1,250
690
542
380

1st
93
18
16
10
3
6
5

2
4
0
0
0

2,926
3,708
1,354
388
178

28
39
14
3
3

Quartiles
Median
3rd
228
1,172
68
242
46
248
28
154
14
122
16
53
10
52
96
72
44
8
6

10
0
2
3
2
7
6
8

360
513
177
41
31

Divisions
Number with cases less than:
50
150
500
1500
2
6
11
13
8
12
15
16
9
12
16
16
11
13
16
17
11
13
16
17
11
14
16
17
13
16
17
17

1
1
1
9
11

6
5
10
13
15

11
10
13
16
16

14
13
16
17
17

16
16
17
17
17

Table XXX. Summary of cancer cases for 1995, by level 3 geography. All cases, top 6 individual ICD codes and health
conditions covered by the Environment and Health Atlas
ICD-10
code
All
C349
C509
C679
C504
C341
C64
EHA
C34
C50
C67
C71
C22

Cases
Min

Max

0
0
0
0
0
0
0

412
106
90
50
38
36
24

5%
20
2
2
0
0
0
0

0
0
0
0
0

172
132
54
22
12

4
6
2
0
0

25%
52
10
8
4
2
2
2

Centiles
50%
81
16
14
10
6
4
4

75%
116
28
24
16
12
8
8

95%
262
65
56
38
25
18
14

16
20
8
2
0

26
30
12
4
2

41
46
20
6
4

100
91
44
12
8

Divisions
Number with cases less than:
1
5
25
100
1
2
14
127
3
24
145
201
4
26
154
202
14
54
182
202
39
93
192
202
39 113
198
202
40 122
202
202
2
1
8
49
87

12
6
38
141
176

99
68
171
202
202

192
195
202
202
202

Table XXX. Summary of cancer cases for 1996, by level 3 geography. All cases, top 6 individual ICD codes and health
conditions covered by the Environment and Health Atlas
ICD-10
code
All
C349
C509
C679
C504
C341

Cases
Min
4
0
0
0
0
0

Max
360
72
130
70
62
32

5%
22
2
2
2
0
0

25%
44
8
10
6
2
2

Centiles
50%
79
14
14
10
4
4

75%
115
24
24
16
12
8

95%
267
52
54
36
28
20
41

Divisions
Number with cases less than:
1
5
25
100
0
2
15
131
6
24
154
202
6
23
153
200
9
50
185
202
48 105
189
202
43 117
197
202

C64
EHA
C34
C50
C67
C71
C22

131

202

0
0
0
0
0

132
160
74
26
14

6
6
2
0
0

14
18
6
0
0

24
32
12
2
2

41
48
20
6
3

84
100
41
12
8

1
1
7
58
90

9
9
35
141
174

107
79
177
201
202

195
191
202
202
202

Table XXX. Summary of cancer cases for 1995, by level 4 geography. All cases, top 6 individual ICD codes and health
conditions covered by the Environment and Health Atlas
ICD-10
code
Min Max
All
C349
C509
C679
C504
C341
C64
EHA
C34
C50
C67
C71
C22

Cases

0
0
0
0
0
0
0

70
22
26
18
16
14
8

5%
2
0
0
0
0
0
0

25%
6
0
0
0
0
0
0

0
0
0
0
0

26
32
24
8
6

0
0
0
0
0

2
2
0
0
0

Centiles
50% 75%
12
22
2
6
2
4
2
4
0
2
0
2
0
2
4
4
2
0
0

8
8
4
2
0

95%
42
12
10
8
6
4
4
17
16
8
4
2

Divisions
Number with cases less than:
1
3
5
15
50
38
115
237
719 1204
374
729
918 1208 1230
402
736
952 1214 1230
573
897 1085 1228 1230
747 1041 1149 1229 1230
820 1115 1190 1230 1230
854 1126 1204 1230 1230
228
181
476
918
1,050

536
413
804
1154
1204

733
637
1025
1215
1224

1151
1150
1226
1230
1230

1230
1230
1230
1230
1230

Table XXX. Summary of cancer cases for 1996, by level 4 geography. All cases, top 6 individual ICD codes and health
conditions covered by the Environment and Health Atlas
ICD-10
code
Min Max
All
C349
C509
C679
C504
C341
C64
EHA
C34
C50
C67
C71
C22

Cases

0
0
0
0
0
0
0

70
22
26
18
16
14
8

5%
2
0
0
0
0
0
0

25%
6
0
0
0
0
0
0

0
0
0
0
0

26
32
24
8
6

0
0
0
0
0

2
2
0
0
0

Centiles
50% 75%
12
22
2
6
2
4
2
4
0
2
0
2
0
2
4
4
2
0
0

8
8
4
2
0

95%
42
12
10
8
6
4
4
17
16
8
4
2

536
413
804
1154
1204

733
637
1025
1215
1224

1151
1150
1226
1230
1230

1230
1230
1230
1230
1230

References
Besag J, York J, Mollie A. (1991). Bayesian image restoration, with applications in spatial statistics. Annals of the
Institute of Statistical Mathematics, 43, 1-59.
Blangiardo M. and Cameletti M. (2015). Spatial and spatio-temporal Bayesian models with R-INLA. John Wiley and
Sons Ltd.
Casella G. and George E. (1992). Explaining the Gibbs sampler. American Statistician, 46, 167-174.
Clayton DG and Kaldor J. (1987). Empirical Bayes estimates of age-standardized relative risks for use in disease
mapping. Biometrics 43, 671-681
R Core Team (2015). R: A language and environment for statistical computing. R Foundation for Statistical
Computing, Vienna, Austria. URL https://www.R-project.org/.
Rue H, Martino S, and Chopin N. (2009). Approximate Bayesian inference for latent Gaussian model by using
integrated Laplace approximations (with discussion). Journal of the Royal Statistical Society, Series B, 71, 319-392.

Source Exif Data:

File Type                       : PDF
File Type Extension             : pdf
MIME Type                       : application/pdf
PDF Version                     : 1.7
Linearized                      : No
Page Count                      : 43
Language                        : en-GB
Tagged PDF                      : Yes
XMP Toolkit                     : 3.1-701
Producer                        : Microsoft® Word for Office 365
Creator                         : Parkes, Brandon L
Creator Tool                    : Microsoft® Word for Office 365
Create Date                     : 2019:03:06 15:45:36+00:00
Modify Date                     : 2019:03:06 15:45:36+00:00
Document ID                     : uuid:61F8A885-F870-4005-86B3-13927EDF96BC
Instance ID                     : uuid:61F8A885-F870-4005-86B3-13927EDF96BC
Author                          : Parkes, Brandon L

EXIF Metadata provided by EXIF.tools

RIF V40 Manual

RIF_v40_Manual

Navigation menu

Versions of this User Manual:

Views

Navigation