Global lessons from digitally identifying the poor for health in Cambodia, India, Rwanda

Digital technologies help in the development of health schemes, according to the findings of a new publication from the WHO. The paper offers concise summaries of three countries and concludes with a “proceed with a little caution” message. More digitization, the linking of more databases and identifying all citizens in a country to assess eligibility are the recommendations. Add some AI for good measure. As interest in the application of digital technologies, digital identity and biometrics in health systems grows, the evidence base remains small, according to the World Health Organisation (WHO). To provide better insight into the technologies’ contribution to universal health coverage, the agency has produced a report into issues such as targeting of individuals and identifying them as well as the financing. ‘The use of digital technologies to support the identification of poor and vulnerable population groups for health coverage schemes: Insights from Cambodia, India and Rwanda’ is a document review of published and grey literature. The countries chosen due to their systems and the challenges they face. The report focuses on the benefits of digital technology and digital identity for health schemes and health outcomes. It acknowledges the risks to data security and protection, which can be compounded as more government databases are linked, but notes that privacy and confidentially risks are discussed in more detail elsewhere.

Cambodia: community-based IDPoor system

The report provides interesting and highly readable summaries of the health financing system in the countries it examines. Cambodia began piloting its Health Equity Fund (HEF), a non-contributory health coverage scheme, in 2000. By 2015 all government health facilities were connected. “Cambodia is the first country known to have undertaken country-wide community-based poverty identification so regularly,” states the report. “A 2015 World Bank study found that most countries (60 percent out of 155 countries monitored) do not have any poverty data available at such regular intervals.” The system issued paper cards to eligible families which proved effective in reducing catastrophic health expenditure, but outcomes varied widely. The Ministry of Planning began a national standardized, formalized mechanism in 2005 that by 2007 became a national community-based poverty identification system – referred to as IDPoor. Supported digitally, it rolled out in all rural districts by 2013, home to 90 percent of people living below the poverty line. Villagers elect a representative group of peers who are trained to interview fellow villagers to determine poverty status. The list is displayed publicly then discussed and ratified by the commune council, typically of ten villages. A status of poor or extremely poor can be given for a household (not individuals). This data is entered into the national IDPoor database by a private company (deemed to make fewer mistakes). Initially, printed photos were sent in to be attached to profiles and added to the Equity Cards sent back to the villages. Government agencies and NGOs could access data, sent as PDFs or burnt to DVDs. The approach was deliberately low-tech, in line with IT literacy skills. In 2014 the database went online through the IDPoor Information System, granting real-time access to health facilities and government and NGOs which could extract target subsets. The scheme expanded to cities and incorporated new features in 2016 such as using smartphones to photograph the recipient and upload the photo. APIs allowed new patient management system and registration system, verification and reimbursement of food and transport costs to cardholders. In 2017 the government announced its wish to link IDPoor to other databases such as for disaster management and registration of people with disabilities. By 2017, 90 percent of all communities were covered, close to 13 million people. The three-year cycle to update profiles and (re)determine eligibility is being phased into on-demand. It will transition from households to individuals. Its data is being used for welfare voucher distribution, cash transfers and COVID-19 assistance. The report finds that the use digital technologies has proved convenient and time-saving and has improved accuracy while reducing costs. Despite digitization, the system still starts with face-to-face village-level assessment. However, the shift to on-demand rather than a village-wide assessment every three years “could lead to administrative burden on the poor, a process linked with high non-take-up.”

India: Aadhaar-backed PMJAY

Unlike Cambodia’s individual assessment of individuals and low-tech start to its health eligibility, India’s current health coverage scheme, Ayushman Bharat Pradhan Mantri Jan Arogya Yojana (PMJAY), uses the national biometric digital identity scheme, Aadhaar, for its identity data. Launched in 2018, PMJAY replaced the Rashtriya Swasthya Bima Yojana (RSBY) government- run health insurance scheme, the Senior Citizen Health Insurance Scheme (SCHIS), and some of the state health insurance schemes. It is offered to 40 percent of the population. The PMJAY patient database is based on the 2011 Socio-Economic Caste Census (SECC 2011) and contains the data of all household members that met one or more of the listed deprivation criteria during that census, plus all RSBY families. As well as being out-of-date, the SECC brings many errors. So instead of using the database for identifying patients, the system uses Aadhaar. When a patient presents him or herself with Aadhaar ID, an algorithm calculates a confidence score as to whether the person is the same as the database record, due to errors and location. If deemed eligible, an Ayushman card (proof of entitlement under PMJAY) is issued and allows immediate recognition. No Aadhaar? Go and register. Emergency care is handled differently. Data can be pulled from Aadhaar to register the patient easily. “The link to Aadhaar seems to have facilitated the enrolment process of people into the PMJAY since public awareness-raising has stressed the health coverage benefits of an Aadhaar identity smart card.” Updates to the SECC are being tackled in a piecemeal fashion: “A new survey across the country was considered impossible. Instead, the government is now also using beneficiary data from other social assistance systems to inform the PMJAY database of eligible beneficiaries.” Noting the accompany risk, the report finds “As Aadhaar makes use of biometric data and is linked to the telephone number and bank account of the individual, it greatly enhances the potential to triangulate data.” “For instance, interruptions in the treatment of patients with tuberculosis and AIDS were reported, as they feared such breaches of data confidentiality. A future question relates to the potential linkage between Aadhaar and the Unique Digital Health card, currently introduced through the National Digital Health Mission, and the benefits and risks this may create.”

Rwanda: community-based health insurance and the Ubudehe system

Rwanda marks something of a middle ground between the Cambodian and Indian approaches. It is community-based, but incorporates the entire population and requires differing levels of contribution. Increased government funding and digitization have helped improve the scheme and health outcomes considerably. The community-based health insurance scheme (CBHI) was introduced in 2000 to target most of the population. By 2016 81.6 percent of the population was covered and a further 6 percent were in other schemes. The poorest do not pay, but stratified contributions were subsequently introduced, ranging from US$2 to $4 per year, depending on earnings and assets. This change required greater administration to assess people and manage contributions. This new administration became known as Ubudehe, meaning mutual assistance. It registers the whole population via community-based assessment where volunteers collect data on household assets which is check and validated at a community assembly. In 2015 the CBHI was transferred to the Rwandan Social Security Board (RSSB). Three years later, the Board introduced the Mutuelles Members Management System (3MS), which was then linked to Ubudehe through an API, creating the first CBHI-Ubudehe link. The Ubudehe can provide the information on who has paid contributions to the RSSB via the 3MS. Linkages became tighter: “the 3MS system was connected with ‘IREMBO-Rwanda Online,’ an e-Government portal that also allows for making mobile telephone-based payments of contributions. Through this enhanced interoperability, it is possible to link all necessary information in order to check whether a household has paid the correct contribution amount on time.” Integrations continued. In 2019, the Ubudehe was linked to the Population Registry of the National Identification Agency (NIDA), which issues national ID numbers and credentials to over-16s. “Updates in the NIDA database are transferred to the Ubudehe system on a continuous basis, thus also improving the ability of the RSSB to manage its beneficiary database. In the same year, this interface was fully functional and 90 percent of all persons over 16 years in the Ubudehe database had been matched to a national identity profile.”

Conclusions: combine and cover everyone

All three could have worked as paper-based systems but would have become extremely cumbersome and labor-intensive, finds the report. All three countries are also conducting ongoing discussions to connect the health systems to more databases. Although new technology could help prevent issues with ever-larger systems, the authors warn: “While there are arguments for the potential power of social registries as gateways for inclusion, past failure of social registries to identify the beneficiaries of social programmes accurately raise concerns, because the registries suffered from large targeting errors, a frequently static nature and high costs.” AI could be deployed and future systems could even automatically enrol people based on what various databases hold on them. The report recommends combining efforts with other social protection schemes rather than trying to establish a new, siloed data set: “Even without a complete leap towards a unique digital identity, existing poverty data from multiple sources can be exchanged, re- used and triangulated, thus avoiding unnecessary duplication of data collection for targeting and identification.” Community-based approaches can help build trust into a new and potentially highly technologically advanced system. Only context-appropriate technologies should be used, and implemented in the right order. The report comes down on the side of Rwanda and India – to capture everyone: “Even when the aim is to identify the poor, it may be advisable to create or build on a database for all citizens, and not just have a database for the poor. Such a universal database could allow for more flexibility, which is especially relevant to individuals and households who move in and out of poverty over time.”