Health population and place

A disease case can be put on a map as soon as it has a usable location, yet a case count alone is only a partial measure of disease. Ten cases in a suburb of 2,000 people mean something different from ten cases in a city of 100,000, so public-health mapping has to combine events with the population from which they arose. In New Zealand that brought notified-disease records into contact with addresses, Census geography, population tables and the awkward question of how much geographic detail could be shown without identifying a person.

The denominator created a second problem. Population data were organised into statistical areas whose boundaries had been designed for census collection and reporting, not for the natural geography of infection. A disease cluster could sit across a meshblock or area-unit boundary, while two nearby households could end up in different statistical units. Rates also became unstable when very small populations were divided into a small number of cases. The map could be numerically correct and still give a strong visual impression that depended heavily on the geography chosen for the calculation.

By the 1990s New Zealand public-health organisations had enough digital information for these problems to become routine rather than theoretical. Health Waikato had been collecting digital notified-disease data from general practitioners from 1993 under the Health Act framework. The records included geographic location, disease type and basic patient demographics. Public-health officers wanted to use the accumulated data to find areas with unusually high incidence, investigate possible causes and decide where to direct resources. Public-health staff needed timely information for investigation and planning. It was to turn a growing surveillance database into something that could be used quickly and repeatedly.

The best documented early case comes from work by Lars Brabyn and Duane Wilkins at the University of Waikato, published in 2001 as Mapping health events: a comparison of approaches. Their starting point was the Waikato Regional Public Health Unit’s notified-disease database. The paper describes a practical requirement for a rapid and routine system for visualising and analysing health events, rather than a one-off research map. ArcView Avenue was used to automate the map-production steps that would otherwise have required repeated pointing, clicking, selecting and calculation.

The automation was fast by the standards of the period. The paper reports that health-event maps for the whole Waikato Health Region could be produced in less than a minute at the most detailed scale. That speed changed what could reasonably be done with surveillance data. A map no longer had to be requested, assembled manually and treated as a finished cartographic product. It could be regenerated as new data arrived or a different disease, time period or geographic scale was selected.

The input data were already part of a national system. Notified cases were stored in EpiSurv, the computerised database installed in public-health services, while ESR collected regional data nationally on behalf of the Ministry of Health. Each record could contain a report date, patient name, home address, age, occupation and medical information. For the Waikato research, 500 health-event locations were extracted for analysis, but the locations were deliberately displaced by at least 500 metres in urban areas and 2,500 metres in rural areas before being used. Privacy was therefore part of the GIS method, not something added after the map had been made.

Geocoding was another part of the workflow. By 2001 EpiSurv had been extended with a GeoStan address-matching engine that allowed an operator to choose from possible matches, and the paper reported exact matching for almost 99 per cent of new cases with some user adjustment. Historical records were harder. Automatic geocoding of 11,564 Waikato records from March 1993 to February 2000 produced mixed results, with 4,015 records receiving a certainty score below 60 per cent. Those low-certainty records were excluded from the later analytical work rather than being placed on the map and quietly treated as reliable.

Duane Wilkins, F. Dumble and J. Woodham’s 2001–02 dental GIS work examined the allocation of dental-therapist resources and the targeting of health promotion. The project linked children’s records with homes, schools, water and fluoride information, income, NZDep96, school rolls and therapist rosters. ArcView supported geocoding, overlay analysis and reporting. Handwritten therapist forms were collated and scanned for the reporting process. The work was presented through the SIRC paper known as “Terrorising the Tooth Fairy” and a surviving 2002 Esri presentation, “Finding, Filling, Drilling and Closing the Gaps”. Mapping those relationships allowed the service to examine where provision and need differed. Public descriptions use the methods and aggregated findings while individual health records remain protected.

Brabyn and Wilkins compared two ways of turning disease events into rates. The conventional vector method aggregated cases into Census units, linked them with resident population and calculated events per 100,000 people. The alternative converted the information into raster surfaces and used focal-neighbourhood functions so the analysis could operate across a consistent moving area rather than being confined to irregular administrative polygons. Both methods started with the same events and population data.

The methods produced different geographic patterns. The paper reported differences of up to 100 per cent between the vector and raster rates at a given location and scale. Some of that difference came from the geometry of Census units, which varied greatly between densely populated Hamilton and large rural areas. At TLA scale the pattern became smoother; at finer scales the variation was much more visible. A user could therefore change the apparent distribution of disease without changing a single case record, simply by changing the spatial model used to organise the calculation.

For public-health work, staff needed to know how each local rate had been constructed. A polygon shaded dark red could reflect a genuine concentration of cases, a small denominator, an awkward boundary or a combination of all three. Raster methods allowed the analyst to control the neighbourhood more consistently while aggregation still required careful interpretation. The work was an early New Zealand example of GIS being used to interrogate how a result was represented as well as to display it.

The same paper preserved the link between research and operations. It did not attempt to explain why particular diseases occurred where they did. Its concern was whether public-health staff could represent notified events in ways that were fast enough for routine use and defensible enough for investigation and planning. That boundary kept the GIS focused on surveillance and representation while leaving epidemiological interpretation to the wider public-health process.

By January 2001 the Waikato work sat inside a broader national push. A contemporary United States CDC public-health GIS bulletin reported that EpiSurv had been distributed to all New Zealand Public Health Units. The system recorded notifiable diseases and incorporated address geocoding. The bulletin also reported that the State had financed one copy of a leading desktop GIS package and digitised topographic maps for each Public Health Unit.

That arrangement joined several components that had previously been separate. EpiSurv held the surveillance records. Address matching turned an address into coordinates. Desktop GIS gave local staff a way to work with those coordinates, while digital topographic mapping provided geographic context. The same surveillance information could also move into a national web presentation. Health GIS was therefore becoming an information chain rather than a collection of isolated maps.

Public Health Units developed their spatial capabilities at different rates. Staff experience, local data quality and the kinds of health problems faced in each region varied. Hardware and software still required support, and a GIS licence did not resolve weak addresses or sparse population data. What had changed was the baseline. By 2001 public-health services around the country could be expected to have access to digital surveillance data, geocoding and at least a basic GIS environment.

The national web component had begun earlier. At the May 2002 launch of the New Zealand Public Health Observatory, Health Minister Annette King described the Ministry of Health as having started to combine GIS and the Internet for communicable-disease information in 1999 through the Public Health Early Warning system, PHEW!. The system allowed a person with a web browser to create and view maps, tables and statistics from the national notifiable-disease surveillance system. The Ministry described the system as a world first.

ESR's own surveillance systems also kept moving. By May 2007 EpiSurv7 was operating on the SurvINZ infrastructure as a secure web-based national notification system. The contemporary system description lists enhanced GIS capability, local and national reporting, near-real-time information and integration of several surveillance streams. ESR's Bruce Adlam presented the system to the Pacific Public Health Surveillance Network coordinating body that March. GIS had moved from being something analysts did to extracts from the disease database to being an explicit capability of the national surveillance application itself.

The Waikato paper adds a useful limit to what PHEW! displayed. It describes the website as exposing a subset of the national EpiSurv database, with Territorial Local Authority geography as its most detailed representation. That scale was much coarser than the geocoded case locations available inside surveillance systems and research datasets. Public access and operational detail were therefore separated. The web service could make national patterns visible without publishing individual household locations.

Later technical sources describe PHEW! as carrying more than static choropleth maps. Users could obtain current information on notifiable-disease incidence and view detailed EpiSurv time series for mapped regions. The system also made an early attempt to flag anomalies in temporal and geographic patterns, and later outbreak-detection literature associates New Zealand surveillance with cumulative-sum methods for detecting possible outbreaks.

Government descriptions, research papers, professional reports and historic URLs describe PHEW!’s functions. The original web application is no longer available for readers to inspect.

Address quality could distort the analysis before any rate was calculated. In 2002 Chris Skelly and colleagues published a study of campylobacteriosis notifications from 1993 to 1997 that examined uncertainty in rural geocoding. They calculated minimum and maximum notification rates according to whether cases with uncertain locations were assigned to rural or urban areas. The estimated maximum rural rates were four times the estimated minimum rural rates, while the equivalent urban estimates changed much less.

The difficulty lay in interpreting and using the mapped information. It was that rural addresses were harder to place confidently in the first place. A case with an incomplete or ambiguous address could be assigned to the wrong type of area or left ungeocoded. Because rural populations were smaller, a relatively modest number of uncertain cases could alter the rate substantially. A perfect-looking dot could still be in the wrong paddock.

The study found wide variation between Public Health Service Regions in the relationship between ungeocoded and rural notifications. That made data quality a geographic characteristic of the surveillance system itself. An analyst comparing regions needed to consider disease incidence alongside how well cases could be located in each region. Improvements in national address data and geocoding would later reduce some of the friction, but spatial precision remained limited by the quality of the reference information used to create the coordinates.

For health GIS, this also constrained what could be inferred from absence. A place with few mapped cases could genuinely have low incidence, or it could have a larger share of records that failed to geocode. Spatial analysis could expose these differences, but it could not make poor source data disappear. The safer workflow was to retain geocoding quality, exclude or flag weak matches and make the uncertainty visible in the analysis.

Public-health GIS was also moving beyond disease surveillance. In November 2002 Lars Brabyn and Chris Skelly published a national study of access to public hospitals. The analysis used a road network to calculate minimum travel distance and estimated travel time from roughly 38,000 Census meshblock centroids to 63 public hospitals. Local population could then be related to the calculated accessibility rather than represented only by straight-line distance.

The method recognised that two places the same distance from a hospital can have very different access. A winding rural road, urban congestion, river crossings and the actual structure of the road network affect travel. The GIS cost-path model assigned travel characteristics to the network and calculated the least-cost route to the closest hospital. This turned a national road dataset and population geography into a health-service planning measure.

The division of work is documented clearly. Brabyn conducted the GIS analysis, while Skelly organised the collection of hospital information; both contributed to the conception of the research and the paper. Public Health Intelligence funded the work, and the computer analysis was undertaken in the University of Waikato Department of Geography. The study therefore linked a Ministry policy requirement with university GIS capability rather than placing the whole analytical process inside one agency.

The May 2002 Public Health Observatory launch had already signalled where this work was heading. The Ministry said the Waikato Geography Department had been an instrumental collaborator in developing national accessibility models and expected travel distances and times to hospitals, primary care and pharmacies to become available through the Observatory. Health GIS was extending from where disease had occurred to how easily populations could reach services.

Both disease rates and accessibility modelling depended heavily on population data. In the early 2000s that information was not automatically a free download. On 9 May 2002 Statistics Minister Laila Harré announced that detailed 2001 Census results would become available free through Census Table Builder. Until then, Statistics New Zealand had charged for detailed information through Supermap subscriptions costing a minimum of NZ$3,300 and up to about NZ$25,000 depending on the level of information required.

The government release described those charges as a major access barrier, particularly for communities and other users with limited budgets. It specifically named health boards among the organisations expected to benefit from free access. For a health analyst, the change affected more than a spreadsheet licence. Population denominators, age structures, ethnicity and other local demographic variables were part of the inputs needed to interpret health events and service needs.

Twenty days later, the connection was made explicitly at the launch of the Public Health Observatory. The Minister credited Statistics New Zealand with supplying the boundary information needed to draw maps and the population information needed for regional comparison. She also linked the new availability of free Census data to the viability of services such as the Observatory. The sequence provides a rare case where a change in government data pricing can be tied directly to a documented GIS-enabled public service.

In health, the 2002 episode is useful because it makes the dependency visible. GIS software could already map the disease records, but population-health analysis needed national demographic information at appropriate geographic scales. Removing a large recurring access charge widened the group of organisations able to perform that work without first buying a commercial demographic-data package.

The New Zealand Public Health Observatory launched on 29 May 2002 as a broader service than PHEW!. It provided a common online window into population-health information, with maps, tables and downloadable material intended for District Health Boards, researchers, health organisations and the public. The Ministry’s Public Health Intelligence group conceived and led the initiative, working with the New Zealand Health Information Service, ESR and Eagle Technology. National context data also came from Statistics New Zealand and LINZ.

PHEW! was described at the launch as a forerunner, not as an earlier name for the same application. Its centre of gravity was notifiable communicable disease. The Observatory aimed to bring together a wider range of population-health datasets and to let users download tables for use with their own information. That wider design placed web mapping inside a broader information-distribution service rather than making the map the only output.

The launch material also reveals the practical state of web GIS at the time. Internet mapping expertise was described as scarce enough that the Ministry used a bureau-service arrangement with Eagle Technology to obtain the technical capability needed for the project. The organisation could own the health purpose and data relationships without maintaining every part of the web-mapping stack internally. That was a common pattern as government agencies began moving specialist GIS functions into browser applications.

The Observatory also made collaboration more visible. Health data stewards remained responsible for hospitalisation, mortality and cancer information, while Statistics New Zealand supplied demographic context and other organisations contributed specialist data or analysis. The service drew information from several health systems. It was a published view across several maintained sources, joined by geography and a common web interface.

The same period produced a more explicit professional identity around health and geography. GeoHealth 2002 was held in Wellington in December, with the Ministry of Health Public Health Intelligence group as lead sponsor. Contemporary CDC material identifies Chris Skelly, then Senior Advisor, Health GeoInformatics, and Jan Rigby as organisers. The proceedings were edited by Rigby, Skelly and Peter Whigham and were issued alongside the 2002 Spatial Information Research Centre colloquium.

The programme brought together public-health practitioners, GIS researchers and international specialists. Disease mapping, environmental exposure, accessibility, spatial statistics and health-service analysis were no longer isolated papers scattered through unrelated meetings. New Zealand had enough operational work and research to support a conference organised specifically around geographic information and health decision-making. The later continuation of GeoHealth reinforced that emerging community.

The conference brought together practitioners working with systems already in use. EpiSurv, PHEW!, local analytical work, national accessibility modelling and the Public Health Observatory had created a body of practitioners who were dealing with the same technical and policy questions from different parts of the health system. The professional network grew around working infrastructure rather than preceding it.

By the time COVID-19 reached New Zealand, web mapping, mobile forms, hosted feature layers and dashboards were ordinary technologies. The geographic problem had also changed. Rather than publishing disease rates for public inspection, some response systems needed to collect and share place-based information while sharply limiting who could see individual records.

During the first national lockdown, the Caring for Communities workstream was established within the national response to focus on people, whānau and communities at greater risk of adverse health, social or economic outcomes. Government updates described the work as supporting access to services and developing regional leadership structures that connected local government, iwi, Pasifika and other communities with central-government leaders. The system needed information about needs and services that varied from place to place, but the underlying records could be sensitive.

The Public Service Commission’s 2020 medal citation for Duane Wilkins records that he was seconded from LINZ to the COVID-19 Operations Command Centre. It states that he developed a support website and a series of map-based tools for Māori and Pacific communities and NGOs during the response. The official account credits his development contribution.

A surviving May 2020 technical document identifies the geospatial environment as Manaaki and records how it was assembled. The core data-entry tool was a Survey123 Connect form designed so users could record place-based items through a simple mapping process rather than relying on addresses. Information was stored in a single ArcGIS Online feature layer, with a copy extracted each day for backup and analysis. The document also described maps, dashboards, web applications, a registration form and a small website integrated with the same hosted environment.

Access control was built into the data model. Each user could see only their own records by default, and even users from the same organisation did not automatically see one another’s entries. A public-facing registration form collected only the details needed to create an account, while the operational feature layer remained restricted. Up to 165 usernames had been made available without charge by Esri for COVID-19 use at the time the document was written.

The public side of Manaaki used another pattern. Several applications re-shared public datasets from other sources, and hosted layers were expected to retain source links and extraction dates. A tabbed StoryMap acted as a launch page for embedded applications, forms and viewers, while dashboards were built with the then-public beta Dashboard application. Sensitive operational records and public contextual information could therefore sit in the same broad platform without being exposed through the same permissions.

The document was written during planning for transfer or continued management of the system. Every available Drive revision from 13 to 14 May 2020 identifies Wilkins as the last modifying user. That revision history supports direct authorship or editing of the surviving technical design and migration record. Manaaki was developed through a collaborative response programme.

A year later the same platform lineage was being used for a more explicitly health-focused purpose. The Hauora-Manaaki governance document says that Hauora-Manaaki was built on Manaaki to support pro-equity implementation and real-time monitoring of the COVID-19 vaccination programme. The technical platform was still ArcGIS Online, but the document devotes more space to governance, confidentiality and permitted use than to map functions.

The agreement states that Māori Data Sovereignty principles apply and names Rangatiratanga, Whakapapa, Whanaungatanga, Kotahitanga, Manaakitanga and Kaitiakitanga. It says no iwi-level data will be stored on the platform, restricts use to pro-equity purposes unless additional use is explicitly agreed, and treats District Health Board information as confidential unless otherwise specified. National-level information already supplied as public open data is treated differently from restricted operational data.

Privacy controls were similarly explicit. The platform was not to collect personal information beyond what was required to maintain the system, and no identifiable health information was to be stored. Uploaded data had to be aggregated and anonymised before access and use. The agreement also provided for deletion of shared information on authorised request, required individual user accounts and two-factor authentication, and recorded that the platform was not rated to hold information subject to the Government Security Classification System.

The first two available revisions of the May 2021 document identify Wilkins as the modifying user, while later revisions no longer identify an individual modifier. That pattern fits an early drafting or editing contribution followed by a broader collaborative governance process. The document describes the system’s intended governance. The document records a system in which deployment was technically straightforward but aggregation, confidentiality, access and Māori authority over data had to be specified separately.

Across this period, the technology changed more quickly than the basic geographic questions. Health Waikato needed to know where notified cases were and how their frequency related to the population underneath them. EpiSurv and PHEW! connected local records with national surveillance and public presentation. The Public Health Observatory combined health information with Census boundaries and population context. Accessibility modelling treated travel through the road network as part of population health, while geocoding research exposed how uncertain addresses could distort rural rates.

By 2020 a small team could assemble forms, feature layers, maps, web applications and dashboards without building a dedicated server GIS from scratch. That did not make the information problem simpler. Manaaki needed per-user visibility and separation between operational and public data. Hauora-Manaaki required aggregation, anonymisation, confidentiality controls and explicit Māori Data Sovereignty principles. More capable delivery tools increased the need to decide exactly what should be visible, to whom and for what purpose.

Health GIS therefore became less identifiable as a separate kind of mapping system. A surveillance analyst might work with geocoded cases and Census rates. A policy team might use modelled travel times. A public website might serve maps and tables from several national datasets. A community-response team might collect place-based needs through a form whose users never opened a conventional GIS application. Geography persisted through all of them as a way of organising records, populations, services and access.

Policing, safety and emergency GIS used some familiar components: geocoded events, Census context, operational databases, web delivery and later public dashboards. The working pressures were different. Police and emergency users often had to decide where to deploy people and resources while events were unfolding, and spatial systems moved from specialist crime analysis into operational policing and emergency-management capability.