CareGap SG

Most of the time, we expect affordability of services to be the top concern for elderlies in Singapore due to the increasing cost of living. However, during our many community outreach projects during national service, we found out that healthcare accessibility was the concern feedback-ed by most elderlies. No one wanted to spend 30 minutes commuting to a check-up or simple blood draw when appointments are as frequent as weekly or fortnightly.

61 years of independence and Singapore has one of the most robust health infrastructures, from Medisave and CPF schemes that addresses affordability to decentralized health care services in heartlands for accessibility.

However, we realized one large gap which is that key infrastructures like polyclinics or specialist centres still aren't meeting the drastically different rates of ageing at each of Singapore's 40 planning areas. Additionally, the huge government expenditure tied to such infrastructural investments makes allocation of polyclinics and healthcare infrastructure ever so salient. We cannot keep relying on top-down projections that will never reflect the real demands on the ground. This can already be seen in some hugely underserved areas of Singapore like Jurong West and Sembawang.

So we thought, what if we could see where the next healthcare gap will appear before it actually happens?

Singapore is already expanding its community healthcare infrastructure. MOH plans to increase the number of polyclinics from 28 to 32 and add 10,600 nursing-home beds by 2030. But deciding where new capacity should go is difficult when national-level projections can hide large differences between individual estates.

We wanted to explore whether data could help make that decision more local, forward-looking and measurable.

CareGap SG is a healthcare planning tool that looks beyond where healthcare demand is today and asks where the greatest gaps could emerge over the next five years. Our goal is to forecast community-care demand from 2026 to 2031 for each of Singapore's 40 planning areas, then combine that forecast with how easily residents in each area can actually access healthcare services.

Instead of treating population growth alone as demand, we account for the age profile of each area. The basic idea is to forecast population by age, then estimate healthcare demand, measure existing access and lastly, identify the gap.

This allows us to produce a 2031 demand-versus-access map of Singapore and highlight areas where growing healthcare needs may not be matched by nearby capacity. But identifying a gap is only half the problem. We also want CareGap SG to answer the next question a planner would naturally ask: "If I can build the next facility somewhere, where should it go?"

Our site-picker evaluates potential locations and uses a greedy site-selection algorithm to identify locations that could bring the greatest number of currently underserved residents within an acceptable access distance. The end result is not meant to replace the judgement of healthcare planners. It is meant to give them another lens: one that lets them test scenarios, compare planning areas and quantify the potential impact of different decisions before infrastructure is built.

We designed CareGap SG around three things: demand, access and action. For demand, we bring together population and ageing data from SingStat and HDB with healthcare utilisation and capacity data from MOH. Rather than applying one national growth rate across Singapore, we forecast each planning area separately so that an estate ageing rapidly is not treated the same as one with a younger population.

We then translate the population forecast into estimated care demand by weighting each age group by its healthcare usage rate. For access, we combine this demand with the geographical locations of polyclinics, CHAS clinics and eldercare services. We use the Two-Step Floating Catchment Area (2SFCA) method to estimate how much healthcare capacity is realistically available to residents within a defined catchment. A higher score means more healthcare capacity is available relative to the demand surrounding a resident. A lower score helps us identify communities where access may be under greater pressure.

Finally, we turn those calculations into something planners can actually use. Our Databricks pipeline takes data from sources such as data.gov.sg, SingStat and OneMap through a Bronze → Silver → Gold architecture. Lakeflow handles ingestion and transformation, while MLflow tracks our forecasting and backtesting results.

The final outputs are designed around the decisions we want to support:

An AI/BI dashboard visualising projected 2031 healthcare demand and accessibility across Singapore, with different ageing scenarios.
A Site-Picker App where planners can test potential new healthcare locations and estimate how many additional seniors would gain access within 1 km.
A Genie space where planners can ask questions in natural language, such as "Which areas face the largest capacity shortfall by 2030 under accelerated ageing?"

One of the biggest challenges is that healthcare demand isn't directly observable at the level we need.

We can forecast how many people of different ages will live in an area, but turning that population into future healthcare demand requires assumptions about utilisation. Similarly, detailed capacity information such as daily appointment slots may not always be publicly available. Where necessary, our accessibility model therefore needs transparent proxies rather than pretending that missing data does not exist.

Another challenge is geography. A healthcare facility does not only serve the planning area it sits in. Residents cross planning-area boundaries, and several communities can compete for the same facility. This is why we chose 2SFCA rather than simply counting the number of clinics inside each area.

Forecasting itself presents another challenge. With a relatively small historical dataset, a more complicated model is not necessarily a better one. We therefore plan to start with linear regression and validate it through rolling backtests from 2021 to 2026, publishing the forecasting error by area and forecast horizon.

These limitations have shaped the project as much as the solution itself. We want CareGap SG to be useful because its assumptions are visible and testable and not because the model is a black box.

We want to know whether CareGap SG could lead to better allocation of limited healthcare capacity.

That means measuring whether recommended sites can reduce the accessibility gap between Singapore's best- and worst-served planning areas, increase the proportion of residents aged 65 and above living within 1 km of community healthcare, and serve more seniors per new facility compared with a baseline site-selection approach.

Ultimately, CareGap SG started with a fairly simple observation: Singapore isn't ageing as one country. It is ageing estate by estate. If healthcare demand is changing locally, we believe planning for it should be local too.