Table 3 Detailed summary of the data extracted from the publications included in the review

Noor et al. [52]

Kenya

Subnational

Methods development and/or evaluation

Paediatric patients

Malaria/fever treatment seeking

174 government hospitals, health centres and dispensaries

ArcView

• Catchment areas estimated using Thiessen polygons (straight-line distance) and the point-in-polygon method to encapsulate enumeration area centroids within each polygon

• Catchment denominator estimated as the total census population for all included enumeration areas

Onsite survey of service users to collect origin/destination data for statistical comparison of estimated catchment areas and empirical utilisation patterns

Maps separately visualising catchment areas post-estimation and based on actual utilisation; distance decay curve for indicator

Gething et al. [51]

Kenya

Subnational

Methods development and/or evaluation

Paediatric patients

Febrile outpatient attendance

81 government hospitals, health centres and dispensaries

ArcView

• Preliminary catchment areas produced using Thiessen polygons (straight-line distance)

• Empirical utilisation patterns were gathered by onsite survey of service users and used to adjust the boundaries between adjacent health facilities based on a ‘fuzzy choice’ algorithm

Statistical comparison of estimated catchment areas and empirical utilisation patterns, including direct comparison of boundaries produced by Thiessen polygons and the ‘fuzzy choice’ adjustment

Mean polygon boundary positions following adjustment; distance decay curves for indicator

Noor et al. [15]

Kenya

Subnational

Methods development and/or evaluation

Under 5 s

Fever treatment seeking

173 government hospitals, health centres and dispensaries

ArcView

• Travel time/impedance modelled using road network, land cover and digital elevation data assuming walking to the nearest health facility

• Empirical utilisation patterns were gathered by onsite survey of service users and used to adjust the boundaries between adjacent health facilities based on a transect algorithm

Statistical comparison against the results produced by other catchment estimation methods: (i) travel time model without transect adjustment; (ii) straight-line distance to the nearest health facility

Map of one study district is presented to separately visualise catchment area boundaries and travel time strata produced by each model

Feikin et al. [61]

Kenya

Subnational

Estimation of catchment areas and/or population denominators

Under 5 s

Outpatient attendance

7 first level government or community health facilities

ArcView

• Denominators are enumerated population totals within polygons created by straight-line distance from geolocated residences to their nearest health facility

Statistical comparison of empirical utilisation patterns to denominators produced based on straight-line distance

Target population denominator for each named health facility; distance decay curve for indicator

Okiro et al. [59]

Kenya

Subnational

Other

Under 15 s

Malaria/all-cause admissions

17 district general hospitals

ArcGIS

• Admitted patients’ village of residence at defined timepoints were extracted from HMIS

• Hospital catchment areas were defined algorithmically by selecting a set of enumeration areas representing 90% of admissions, then adjusted to smooth their boundaries and capture unassigned enumeration areas

• Catchment denominators were estimated at defined timepoints using projected census enumeration area counts

None

Map visualising the locations of hospitals and their estimated catchment areas; baseline and projected target population denominator for each hospital

O’Meara et al. [66]

Kenya

Subnational

Model covariate computation

Under 5 s

Malaria-related admissions

10 public and 1 mission primary healthcare facilities

Not stated

• Travel time was modelled using road network data with nominal speeds for network walking time to the nearest health facility

• Catchment areas were defined by assigning all children in the study area to their nearest health facility

None

Map visualising travel time to the nearest health facility; distance decay curves for indicator

Moïsi et al. [68]

Kenya

Subnational

Model covariate computation

Under 5 s; some stratified analyses in < 1 year and 1–4 years subgroups

Child mortality

5 hospitals and 47 vaccine clinics

ArcGIS

• Travel time/impedance modelled using road network, bus route and topographic data

• Locations placed within specified travel time strata of the nearest health facility

Sensitivity analysis comparing multiple travel scenarios with nominal speeds: walking or walking/public transport, the latter assuming compound journey sequences

Maps separately visualising travel time to the hospital/nearest health facility by travel scenario

Okiro et al. [58]

Kenya

Subnational

Other

Under 15 s

Malaria/all-cause admissions

8 district general hospitals

ArcGIS

• Admitted patients’ village of residence at defined timepoints were extracted from HMIS

• Hospital catchment areas were defined algorithmically by selecting a set of enumeration areas representing 90% of admissions, then adjusted to smooth their boundaries and capture unassigned enumeration areas

• Catchment denominators were estimated at defined timepoints using projected census enumeration area counts

None

Projected target population denominator for each hospital

Moïsi et al. [69]

Kenya

Single health facility

Model covariate computation

Under 5 s

Child mortality

1 district hospital

ArcGIS

• Travel time/impedance modelled using road network, bus route and topographic data

• Locations placed within specified travel time strata of the nearest health facility

Sensitivity analysis comparing multiple travel scenarios with nominal speeds: walking or walking/public transport, the latter assuming compound journey sequences

Maps separately visualising travel time to hospital by transport scenario; distance decay curve for indicator

Alegana et al. [18]

Namibia

Subnational

Estimation of catchment areas and/or population denominators

Under 5 s

Fever treatment seeking

All public health facilities within the study area

AccessMod; ArcGIS

• Travel time/impedance modelled using road network, land cover and digital elevation data

• Travel time model assumes compound journey sequences to reach the nearest health facility

• Travel time used as main predictor in a three-parameter logistic regression model to estimate treatment seeking probability

• Threshold used to define catchment area extents was set based on review of the decay curve

• Catchment denominators estimated by intersecting boundaries with population surface

Statistical comparison against the results produced by estimating straight-line distance to nearest health facility

Map visualising probability of treatment seeking by health facility with overlaid catchment area boundaries; distance decay curve for indicator; target population denominators aggregated by region and travel time strata

Blanford et al. [43]

Niger

National

Estimation of catchment areas and/or population denominators

Children aged 12–59 months

Complete vaccination status prior to 12 months of age

50 hospitals, 400 integrated health centres and 54 maternity centres

ArcGIS

• Travel time/impedance modelled using road network, land cover, surface water and digital elevation data

• Travel times greater than 4 h to the nearest health facility taken as inadequate access

• Kernel density estimation used with settlement population data to explore health service accessibility in areas with the high/low population

Sensitivity analysis comparing multiple travel scenarios with nominal speeds. Separate scenarios created for each combination of walking/vehicular travel and wet/dry season

Maps separately visualising (i) travel time to nearest health facility for each travel scenario, (ii) settlements by travel time for each season, highlighting those with no access

Kadobera et al. [65]

Tanzania

Subnational

Model covariate computation

Under 5 s; some stratified analyses in < 1 year and 1–4 years subgroups

Child mortality

Public health facilities;

13 dispensaries, 2 health centres and 2 district hospitals

ArcView

• Road network and footpath data used to estimate the path distance from each geolocated household to its nearest health facility along the shortest route

Statistical comparison against the results produced by estimating straight-line distance from each geolocated household to its nearest health facility

None presented

Mathiu et al. [37]

Multi-country study

(The Gambia, Senegal)

Subnational

Estimation of catchment areas and/or population denominators

Under 5 s

Suspected meningitis incidence

Sentinel hospitals participating in the World Health Organisation meningitis network

OpenEpi

• District of residence retrieved from hospital records for each suspected meningitis case

• Districts rank-ordered by number of cases contributed to the cumulative total

• Catchment areas defined as the subset of rank-ordered districts contributing 80% of all cases

• Hospital denominator estimated using district-level population

None

Target population denominator for each hospital

Manongi et al. [56]

Tanzania

Single health facility

Model covariate computation

Children aged 2–59 months

Hospital inpatient admissions and deaths

1 district hospital

ArcGIS; ArcView/AccessMod

• Straight-line distance between villages and the hospital were calculated

• Catchment area was defined as the group of villages closer to the study hospital than any other

• Catchment denominator population was extracted from national census data

None

Distance decay curves for indicators; target population denominators aggregated by travel time strata

Zaman et al. [63]

The Gambia

Subnational

Model covariate computation

Children aged 6–51 weeks at trial entry

Mortality, pneumonia incidence and pneumococcal vaccine efficacy

The two largest health facilities in the study area: one hospital and one health centre

ArcGIS

• Straight-line distance from compound of residence to the nearer of two study health facilities was calculated

• Distance was categorised using a series of cut points

Statistical comparison against travel times derived from an isotropic time surface analysis

None presented

McLaren et al. [48]

South Africa

National

Model covariate computation

Separate subgroup analyses: over 18 s; under 5 s

Over 18 s: health consultation in the previous year; under 5 s: skilled attendant at birth

Public health facilities of all types

Not stated

• Straight-line distance from each geolocated household to its nearest health facility was calculated

None

Density curves for distance to the nearest health facility by population characteristics

World Health Organisation [39]

n/a

Single health facility

Estimation of catchment areas and/or population denominators

Under 5 s

Suspected meningitis incidence

Sentinel hospitals participating in the World Health Organisation meningitis network

n/a

• Hospital records to be reviewed for each suspected meningitis case (standard case definition) and district of residence retrieved

• Districts rank-ordered by number of cases contributed to the cumulative total

• Catchment area defined as the subset of rank-ordered districts contributing 80% of all cases

• Hospital denominator estimated using district-level population

n/a

n/a

Aoun et al. [55]

Rwanda

Subnational

Model covariate computation

Under 5 s

Height-for-age z-score

All health centres and district hospitals in the study area

AccessMod/ArcGIS

• Travel time/impedance modelled using road network and land cover data

• Catchment areas were defined in line with two constraints: (i) 1 h walking time to the nearest health facility, (ii) capacity of that facility

None

Maps separately visualising the location of health facilities, population density, land cover, travel time to the nearest health facility, and estimated catchment area boundaries

Afagbedzi [38]

Ghana

Subnational

Methods development and/or evaluation

All ages; subgroup under 5 s

Malaria and diarrhoea incidence (separately)

All health facilities within the study area

ArcGIS

• Catchment areas estimated by creating Thiessen polygons (straight-line distance)

• Catchment denominators were estimated using the point-in-polygon method to aggregate the population of all census enumeration areas whose centroid fell within the catchment area

None

None presented

Macharia et al. [70]

Kenya

Subnational

Other

Pregnant women and infants

Uptake/allocation of long-lasting insecticidal bed nets

888 health facilities distributing long-lasting insecticidal bed nets and operated by government, non-governmental or faith-based institutions

ArcView; AccessMod; ArcGIS; R (including ‘geoR’ package)

• Travel time/impedance modelled using road network, land cover and digital elevation data

• Travel time model assumes compound journey sequences to reach the nearest health facility

• Result used as the main predictor in logistic regression models estimating the probability of health facility attendance

• Additional model covariates were sourced from household survey data and converted to surface layers using ordinary kriging spatial interpolation

• Threshold used to define catchment area extents was set based on review of the decay curve

• Catchment denominators estimated by intersecting boundaries with population surface

Statistical comparison with independent data from the study area

Maps separately visualising travel time to nearest health facility and delineated catchment/unserved areas

Haidari et al. [45]

Mozambique

National

Estimation of catchment areas and/or population denominators

Girls aged 10 years

Access to immunisation centres

1337 routine immunisation centres for the World Health Organisation Expanded Program on Immunisation in Mozambique

SIGMA, a study-specific GIS platform

• Theoretical catchment areas defined by overlaying straight-line distance radii onto immunisation centre point locations

• Catchment denominators estimated by intersecting boundaries with population surface

• Population of the target demographic subgroup was approximated as a proportion of the total population figure

Sensitivity analysis to directly compare the results produced using several straight-line distance radii

Maps visualising the spatial coverage of progressively increasing catchment area radii

Smith et al. [49]

Uganda

National

Model covariate computation

Under 18 s; some stratified analyses in under 5 s and over 5 s subgroups

Unmet surgical need

All public health centres, general hospitals, district hospitals and regional referral hospitals with surgical capability

ArcGIS; QGIS; GeoDa

• Catchment areas estimated using a Voronoi diagram (straight-line distance) and the point-in-polygon method to encapsulate all enumeration areas centroids within each polygon

None

Maps visualising study outcomes and spatial associations in the indicator for surgical health facilities, aggregated to district level

Ouma et al. [19]

Kenya

Subnational

Model covariate computation

Under 5 s

Fever treatment seeking

All health facilities operated by government, faith-based and other non-profit organisations

ArcGIS; R (including ‘geoR’ package)

• Travel time/impedance modelled using road network, land cover and digital elevation data with nominal speeds

• Travel time model assumes compound journey sequences to reach the nearest health facility

• Result used as the main predictor in generalised linear mixed models to estimate treatment-seeking probability

• Threshold used to define catchment area extents was set based on review of the decay curve

• Catchment denominators estimated by intersecting boundaries with population surface

Candidate models were evaluated using a validation subset and accuracy metrics including classification error and receiver operating characteristic

Maps visualising separately visualising travel time to nearest health facility and probability of treatment seeking; distance decay curves for model covariates

Juran et al. [41]

Sub-Saharan Africa

Regional

Estimation of catchment areas and/or population denominators

All ages; subgroup under 15 s

Access to major hospitals providing surgical care

All major regional and district hospitals operated by local government, non-governmental or faith-based institutions

AccessMod

• Travel time/impedance modelled using road network, land cover and digital elevation data

• Catchment areas estimated using 30 min, 1 h and 2 h travel time strata under two travel scenarios: walking on land cover and motorised transport on the road network

• Catchment denominators estimated by intersecting boundaries with population surface adjusted for rate of surgical burden

None

Maps visualising total and percent population within travel time strata

Alegana et al. [40]

Sub-Saharan Africa

Regional

Model covariate computation

Under 5 s

Fever treatment-seeking

All major regional/district hospitals, dispensaries, clinics, health posts and health centres operated by government, local authority, faith based and non-governmental organisations

AccessMod; JAGS; R and rjags package

• Travel time/impedance modelled using road network, land cover and digital elevation data

• Travel time model assumes compound journey sequences to reach the nearest health facility

• Result used as the main predictor in a Bayesian model based on Item Response Theory to estimate treatment-seeking probability

Fitted model was evaluated using a validation subset and accuracy metrics including classification error and receiver operating characteristic

Maps separately visualising travel time to nearest hospital, health centre, and lower tier health facilities; distance decay curves for indicator

Kundrick et al. [47]

Malawi

National

Estimation of catchment areas and/or population denominators

Infants

Measles vaccination coverage; susceptible birth cohort; effective reproductive ratio

Unclear

Not stated

• Catchment areas estimated using a Voronoi diagram (straight-line distance)

• Catchment denominators estimated by intersecting boundaries with population surface

• Susceptible birth cohort denominator was approximated as 50% of the total population figure multiplied by the regional fertility rate

None

Maps separately visualising scaled measles vaccination coverage, susceptible birth cohort and effective reproductive ratio at health facility polygon level

Milucky et al. [57]

Burkina Faso

Single health facility

Estimation of catchment areas and/or population denominators

All ages; subgroups include children aged under 1 year, 1–2 years, 2–5 years, 5–15 years

Acute respiratory infection admission

1 district hospital

SAS, ArcGIS

• Commune of residence retrieved from hospital records for each admission

• Communes rank-ordered by number of admissions contributed to the cumulative total

• Catchment area defined as the subset of rank-ordered communes contributing 85% of all admissions

• Hospital denominator estimated using commune-level population projections from census

None

Map visualising the location of communes comprising the hospital catchment area

Hierink et al. [78]

Mozambique

Subnational

Other

Under 5 s

Access to nearest health facility pre-/post-cyclone

All health facilities within the study area

QGIS; R (spatial packages not stated); AccessMod

• Travel time/impedance modelled using road network, land cover, surface water and digital elevation data

• Travel time model assumes compound journey sequences to reach the nearest health facility

• Accessibility defined using a maximum travel time of 2 h to the nearest health facility

• Travel time extent intersected with population surface to estimate the number with access pre-/post-cyclone

Sensitivity analysis comparing travel scenarios based on upper/lower limits of motorised transport speed, and assumption that some floods waters are passable with reduced walking speed

Maps separately visualising areas with access to health facilities pre-/post-cyclone, and accessibility ratios at each timepoint

Joseph et al. [46]

Kenya

National

Model covariate computation

Children aged 12–23 months

DPT3 vaccination status; full immunisation status (BCG, measles, DPT3, polio and pneumococcal vaccines)

All public and private health facilities offering immunisation services

ArcMap; AccessMod

• Travel time/impedance modelled using road network, land cover and digital elevation data

• Locations placed within specified travel time strata of the nearest health facility

Sensitivity analysis comparing multiple travel scenarios with nominal speeds: walking or walking/second transport mode (varying by terrain and road type/infrastructure), the latter assuming compound journey sequences

Maps separately visualising travel time to the nearest health facility by transport scenario

Alegana et al. [71]

Kenya

Subnational

Methods development and/or evaluation

Children aged 1 month to 14 years

Malaria/severe malaria admission

4 major level 4 or level 5 hospitals

ArcGIS; AccessMod; R (including ‘R-INLA’ package)

• Admitted patients’ enumeration area of residence were extracted from HMIS

• Enumeration area-level probability of admission was modelled using a Bayesian hierarchical zero-inflated Poisson regression model

• Catchment areas comprised of enumeration areas for which Bayesian posterior probability estimate exceeded a specified threshold

Candidate models were evaluated using a validation subset and metrics including cross-validated mean logarithmic score and root mean square error

Maps separately visualising the spatial distribution of admissions by enumeration area and conversion to hospital catchment areas

Mpimbaza et al. [60]

Uganda

Subnational

Estimation of catchment areas and/or population denominators

Children aged 1 month to 14 years

Malaria admissions

5 public district hospitals

Stata; R (for statistical analysis)

• Admitted patients’ parish of residence were extracted from HMIS

• Catchment areas defined as the rural/peri-urban parishes contributing admissions that were located nearest to the hospital

• Catchment denominators estimated using parish-level population projections from census

None

Map visualising the location of hospitals and their catchment areas; target population denominators expressed in population years at risk

Cairo et al. [50]

Democratic Republic of the Congo

Subnational

Estimation of catchment areas and/or population denominators

Children requiring paediatric surgical services

Access to paediatric surgical services

The highest level hospital or primary referral centre in each health district (40 in total); most were operated by public or faith-based organisations, but one private sector and one non-governmental organisation

AccessMod

• Accessibility defined using a maximum travel time of 2 h to the health facility, itself measured using 15 km straight-line distance as a proxy

None

Maps separately visualising the proportion of the target population with access to health facilities of different types, aggregated to district level

Epstein et al. [72]

Uganda

Subnational

Methods development and/or evaluation

Children between 6 months and 10 years of age

Malaria incidence

Two malaria surveillance centres

R (including ‘malariaAtlas’ package)

• Catchment areas were defined as those villages nearer to the specified health facility than any other

• Travel time/impedance modelled using road network, land cover, surface water and digital elevation data

• Travel time model assumes compound journey sequences to reach the nearest health facility

• The mean travel time over all locations within the village was calculated, and used to quantify the effect of distance decay on the village-level probability of health facility utilisation

• Catchment area denominators were estimated by ‘downweighting’ their projected population totals in line with distance decay, modelled using service utilisation data extracted from the HMIS

Statistical comparison of population-level incidence rates calculated using HMIS data and the estimated catchment denominators against independent data collected from a separate cohort study in the same area and population

Maps visualising the location of villages comprising health facility catchment areas and the probability of attendance by village; distance decay curves for probability of attendance

Cotache-Condor et al. [44]

Somaliland

National

Estimation of catchment areas and/or population denominators

Under 15 s

Access to surgical care

15 hospitals capable of providing surgical care

ArcMap

• Road network distance from households to the nearest hospital were converted to travel time assuming nominal and constant travel speed

• The optimal catchment area extent was defined using a maximum travel time of 2 h, but locations were also placed within specified travel time strata of the hospital

• Catchment denominators estimated by intersecting boundaries with population surface

Sensitivity analysis comparing multiple travel scenarios: walking or public transport

Maps visualising catchment areas for each hospital type by travel scenario and travel time strata

Simkovich et al. [42]

Multi-country study

(Rwanda, Peru, India, Guatemala)

Subnational

Estimation of catchment areas and/or population denominators

Paediatric patients

Incidence of severe pneumonia

All public and private hospitals, health centres, health posts or other facilities in the study area treating paediatric patients

ArcGIS Pro, R (for statistical analysis and visualisation)

• Travel time/impedance modelled using road network and speed limit data assuming compound journey sequences

• Locations placed within specified travel time strata of the nearest health facility

• Strata boundaries intersected with population surface to estimate the proportion within each

Sensitivity analysis comparing multiple accessibility scenarios based on varying resource-/capability-based

health facility characterisations

Maps separately visualising travel time to the nearest health facilities of different types

Hyde et al. [64]

Madagascar

Subnational

Methods development and/or evaluation

All ages; subgroup under 5 s

Malaria incidence

19 public health centres

QGIS/QuickOSM plugin; R (including ‘gstat’, ‘spdep’, ‘sp’ packages)

• Catchment areas defined by matching administrative units to their nearest health facility by the shortest average path distance over all households therein

• Catchment denominators were compiled using administrative unit-level population estimates from the Ministry of Health

• Population of the target demographic subgroup was approximated as a proportion of the total population figure

Additional data extracted from health facilities’ HMIS were used to estimate service utilisation and malaria incidence rates, and these were compared to independent cohort data from the same area to evaluate these estimates in terms of biases introduced by geographical/financial barriers

Maps visualising malaria incidence at specified timepoints by Fokontany