Leveraging Adaptive Machine Learning Algorithms for Patient Identification in Zimbabwe (2023)

Data Use Community, July 2023

The Data Use Community (DUC) is an open global community passionate about improving health data sharing and use. It is a forum of virtual meetings and an online discussion board for sharing and learning from peers and country experiences. On July 26, 2023, Blessing Manyiyo, Technical Lead, Simbarashe Chaputsira, Data Scientist and Patrick Mapuranga, System Architect with the Zimbabwe Technical Assistance, Training & Education Center for Health (Zim-TTECH), shared their experience with leveraging machine learning algorithms to support patient identification processes in Zimbabwe. Below is a summary written by the DUC Secretariat of what was understood at the time of sharing.

Background

Many health systems are grappling with the challenges associated with identifying and linking electronic health records (EHRs) across and within systems. Unique identification approaches and techniques can be used to improve the accuracy of integrating EHRs to better support patient management, allocation of resources, and continuity of care.

The Zimbabwe Technical Assistance, Training & Education Center for Health (Zim-TTECH) serves as the strategic information (SI) partner to the Ministry of Health (MoH) and manages the implementation of the national EHR in Zimbabwe, otherwise known as Impilo. Initially developed in 2016, Impilo includes more than 22 modules, and is available as a web-based mobile application. As efforts continue to scale Impilo, Zim-TTECH is charged with the task of developing technology-based solutions to advance patient identification processes to strengthen continuity of care and public health management. According to Zim-TTECH, the goals for these two use cases are to:

• Improve continuity of care at the patient level by ensuring that all relevant patient information is readily available to any healthcare professional involved in managing and providing care; and

• Strengthen public health management at the service delivery level through accurate identification of patients to help public health agencies track the spread of diseases, monitor outbreaks, and implement appropriate interventions to protect communities.

Technical Approach

The patient identification (ID) process starts with the registration of an individual at the facility level via Impilo, where a person ID that is system generated is used to identify patients. While this process happens in real-time for health facilities connected to the internet (online), health facilities without the infrastructure to connect (offline) may have a delay of several days before being able to register a patient – as patient identification can occur at both the facility and central level. Given that national identifiers are not considered mandatory as patient IDs, demographic data like full name and date of birth are used to assess whether an individual is a new or returning patient. The diagram below outlines the workflow for registering an individual when visiting an online or offline health facility (Diagram 1).

Diagram 1: Patient Registration Workflow at Health Facility

When information about an individual is not found via local Impilo, data is transmitted to the central database which contains EHR information from across different health facilities. The technical components involved for patient identification include an adaptive machine-learning algorithm based on probabilistic methods that use a mixture of attributes (13 identifiers) to identify and link patient records. To mitigate errors caused by similar identifiers, each attribute is weighted as part of the de-duplication process. The diagram below highlights the patient identification workflow outside of the local facility level (Diagram 2).

Diagram 2: Patient Identification Workflow

The technical components involved in the patient identification process are synthesized in the table below (Table 1).

Table 1: Technology Components

Impact & Challenges

Impilo has been deployed to nearly 70 percent of health facilities in Zimbabwe that includes information on 1.6 million individuals. Additionally, there are 800 web-based or mobile applications of Impilo in use outside of health facilities. After transitioning to an adaptive machine-learning algorithm, Zim-TTECH has been able to match individual identities with 95 percent accuracy.

With any technology implementation, there remain a few areas of improvement. The themes of challenges encountered include:

• Lack of unique patient identifiers -> Currently an array of identifiers is used due to the lack of unique patient identifiers, making the de-duplication process cumbersome and inefficient;

• Similar attributes/identifiers within the same geographic region -> Difficulties were encountered with de-duplicating identities due to patients having same names in two different Zimbabwean regions; and

• Limited infrastructure/connection -> Most facilities are offline making it difficult to de-duplicate in real-time.

Lessons Learned

Several lessons can be gleaned from Zim-TTECH’s experience leveraging an adaptive machine-learning algorithm for patient identification.

• Digitizing medical records and creating a workflow for both online and offline health facilities can improve the patient identification process;

• Continuous updates and enhancements are necessary to improve the performance of machine-learning algorithms; and

• Implementing a unique patient identifier versus using an array of information can streamline and increase the accuracy of patient matching.

Looking Ahead

Various steps have been taken towards having more facilities connected via Impilo, as well as advocating the use of unique patient identification in Zimbabwe. To address the challenges encountered, Zim-TTECH is exploring the use of biometrics with collection by fingerprint readers and iris scanners. Additionally, the MoH is currently pursuing legislation for instituting a National ID or patient ID system. Looking ahead, Zim-TTECH is working with various stakeholders to develop a national digital health platform. The diagram below highlights the components of the proposed architecture (Diagram 3).

Diagram 3: Proposed Zimbabwe Digital Health Platform

For more information on the experiences in Zimbabwe, please visit the DUC presentation here.

References

DUC Meeting July 26, 2023: Presentation by Blessing Manyiyo, Tech Lead, Simbarashe Chaputsira, Data Scientist, and Patrick Mapuranga, System Architect, Zim-TTECH