1. Too far to walk – calibrating distances to maternity health facilities for women in Ghana using GIS tools
Zoë Matthews

2. Coauthors
Fiifi Amoako-Johnson, Social Statistics, University of Southampton
Faustina Frempong-Aiguah, (RIPS) University of Ghana
Peter Gething, Spatial Epidemiology, Dept Zoology, University of Oxford
Peter Atkinson, School of Geography, University of Southampton
Angela Baschieri, London School of Hygiene and Tropical Medicine
Philomena E. Nyarko, (RIPS) University of Ghana and GSS
Francis Nii-Amoo Dodoo, (RIPS) University of Ghana
Jane C. Falkingham, Centre for Population Change, University of Southampton
Patrick Aboagye, Ghana Health Services, Accra, Ghana

3. Acknowledgements ESRC/DFID – for funding project
IMMPACT – for use of survey data from Ghana
CERSGIS – University of Ghana, for use of GIS data
Ghana Statistical Service – for use of GIS and facility data

4. Outline Background Ghana Aims Methods
Stage 1: Build a model of physical access
Stage 2: Calibrate the model using survey data of actual physical access in a small area
Stage 3: Apply nationally
Building and calibrating a model to measure physical access to facilities
Data sources in Ghana
Results – not yet!
Why the results (when we get them!) will be an advance – and how they will be useful

5. Policy efforts to break down barriers to effective coverage
COST
2003 Fee Exemption Policy
4 regions targeted for the policy
Policy ended in 2005 but free care for all pregnant women via national insurance scheme announced 2007/2008
DISTANCE
CHPS strategy for reaching the unreached is part of the national poverty reduction strategy.
CHPS districts deploy professional health workers to provide community-based health care, including safe motherhood and family planning
27.6% of districts have brought health services to communities
QUALITY
MAF plans to upgrade family planning, skilled delivery and EmONC
Includes equipment, improved education and support for health workers as well as commodities and support for governance

6. Existing data: health system infrastructure
Average distance to nearest hospital
As measured on census 2000…but these are not maternity facilities

7. Existing data

8. Journey time (minutes)
Aim
To give an accurate picture of distances to maternal health facilities in Ghana – down to small areas – nationwide.
To provide ‘cost-surface’ maps of Ghana for maternity facilities as has been done in Kenyan districts for malaria
WHY?
To facilitate understanding of the extent to which distance, or distance related factors are a barrier to use
To monitor the success of policies
For planning and targetting
Journey time (minutes)

9. How?
Stage 1: Build a model of distance to facility using GIS catchment techniques
Stage 2: Calibrate model using survey data on a small area with actual travel distances and times
Stage 3: Apply nationally

10. Data sources Stage 1 Stage 2 Stage 3 Facility locations Land cover
Transport routes
Elevation map
Stage 2
Actual distances for a smaller area – survey data
Should include residence location, facility location and actual time/distance
Should be specific to maternity
Stage 3
As for Stage 1 – covering whole country
Accurately parameterised model from Stage 2 incorporating facility choice and bypassing

11. Stage 1: for Ghana we have
Facility map
Land cover including:
Road and path networks
Natural barriers – lakes, rivers, swamps,
nat.parks by season
Contour maps for gradient

12. Building a GIS model to measure physical access to health facilities
How far?
Which is nearest?
Facility A
Facility B 12

13. Simple approach is Euclidean distance
Catchment modelling
Simple approach is Euclidean distance
Facility A
Facility B 13

14. BUT Euclidean distance not a good measure Affected by:
Catchment modelling
BUT Euclidean distance not a good measure
Affected by:
Facility A
Facility B 14

15. Catchment modelling
…natural barriers
Facility A
Facility B 15

16. …elevation / gradient / slope
Catchment modelling
…elevation / gradient / slope
Facility B
Facility A
Facility B 16

17. …transport network Catchment modelling Facility B Facility A17

18. SO journey time likely to be a better metric
Catchment modelling
SO journey time likely to be a better metric
Facility B
Facility A
Facility B 18

19. + = Modelling step 1: production of ‘impedance’ grid Natural barriers
Catchment modelling
Modelling step 1: production of ‘impedance’ grid =
Natural barriers
Roads/paths
Impedance 19

20. Spatial analysis and GIS for the mapping of malaria in Kenya
Catchment modelling
Step 2: Incorporation of effect of gradient on speeds
Used variation of Naysmith’s rule
does your model use Naismiths rule to estimate the time it takes to walk a route? Yes, the original version we coded up for Kenya used a slight elaboration on Naysmith's rule to calibrate walking time and effect of gradient
Spatial analysis and GIS for the mapping of malaria in Kenya 20

21. Journey time (minutes)
Catchment modelling
Step 3: Design of algorithm to calculate journey-time
Assumes parameters based on number of gradient categories, base speeds by mode of transport and conversion of travel speeds to friction/impedance
Uses 100m by 100m raster grid
Uses region-growing approach from each facility
Calculates cumulative sum of journey time to each pixel
Incorporates a facility choice component
Journey time (minutes)
does the model assign a friction value to each pixel of the various raster image files according to surface type to generate the cost to move
across each pixel? Yep, exactly that. 'Friction' is defined more specifically as 'time in seconds to traverse the pixel' and it varied depending on which direction you were going in (because a sloped pixel is uphill in one direction and downhill in the other) and also whether you went diagonally across it (which is a further distance than if going straight across).
Does it use the cost module in Idrisi (Clark labs) applied to obtain travel cost image then scaled to minutes travel time????
Nope, our algorithm was written from scratch in C - no such algorithms were available in GIS at the time. Also the elaborations described above are not always coded in to t he GIS versions.. 21

22. Stage 2: Calibrate using real data from a survey
IMMPACT out-of-pocket-costs survey in two regions Volta and Central – over 2,000 women surveyed
Includes distances to maternity facilities and times as well as forms of transport
To calibrate: Repeatedly COMPARE different versions (with alternative parameterisations) of the MODEL applied to locations in the survey with the REAL TIMES from the survey… then choose the optimum best-performing model with least squared differences
Proceed to Stage 3…APPLY MODEL MORE WIDELY
Volta Central Total
Vaginal Delivery HF
Vaginal delivery H/TBA
C Section
Total
A two-stage approach was used to identify women for the household cost survey. The first stage selected
health facilities operating immunisation programmes and child welfare clinics in all the six
districts identified for the evaluation in Central and Volta regions and sampled women falling into the
sampling frame. In the second stage the sampled women were followed to their homes to administer
the household cost questionnaire.

24. Results: travel time

25. Results: % of women who live within 2 or 4 hrs travel time