Empowering IDSR reporting and analysis through An Intelligent feedback System (IntelSurv): Dissemination Workshop Minutes

Activity Venue: Amaryllis Hotel, Blantyre, Malawi

Date: 28th March 2024

Moderator: Moses Gwaza

Workshop Discussion

Welcoming remarks: Dr Chikumba ~ HoD

• Welcomed everyone to the workshop and highlighted that the aim of the KAI lab is to promote Artificial Intelligence (AI) which is a product under one of MUBAS core value of Research and Innovation.
• Encouraged participants to freely engage and share ideas, experiences and lessons learnt throughout the project.

Dr Alinafe Mbewe ~ Deputy Director and Head of Digital Health in Malawi

• Embraced merging of basic hospital settings and health care with digital health.
• Revealed that the department of Digital Health and the entire Ministry of Health have plans to completely employ digital health into Malawi’s health systems and discussions are underway on how to make digital health useful in healthcare particularly to end users who are the patients.
• Emphasized the need for and importance of ensuring that whatever is being done in advancing digital health, must translate into addressing a specific need for patients and users.
• She also stressed the importance of ensuring sustainability of digital health systems for the benefit of patients especially in public health facilities.

Overview of the workshop: Dr Taylor

• Dr Taylor appreciated attendees for coming to the workshop and outlined the workshop's purpose.
• She highlighted that the IntelSurv was a 6-month pilot study aimed at solving problems encountered during data collection on priority diseases, disease surveillance, medical terms, and health scenarios among others using Generative AI.
• The development of the IntelSurv was based on an understanding of the challenges faced by health system users like clinicians, nurses, HSAs in data collection on disease surveillance.
• Such an understanding would help in developing the right AI tool which would then help in solving their problems and planning for future challenges and opportunities.
• She emphasized the importance of involving users in developing tools to avoid and/or minimize potential misuse and to ensure that the tools address real needs on the ground.

IntelSurv’s Journey from Concept to Reality: Personal Insights from Dr Liwewe

• In 2020, when the COVID-19 pandemic began, there was little/no data available and the Ministry of Health provided funding to address this issue.
• A training was done on case management due to the rise in COVID-19 cases. However, guidance on data collection and addressing data quality issues was lacking.
• Recognizing the value of surveillance data, efforts were made to incorporate it into training programs to facilitate early case detection and targeted interventions.
• Discussions about using the collected data started at the INSPIRE AGM in 2022.

The process of developing the tool involved the following:

• Proposal development for qualitative data exploration
• Organizing focus group discussions and in-depth interviews with key informants in Lilongwe and Blantyre.
• Data Analysis: Key issue that came out was trainings of health workers who did not understand the COVID-19 pandemic and related medical issues.
• Development of key questions from transcripts
• Evaluation of AI generated Questions and Answers from Policy document
• Development of the IntelSurv tool.

Lessons learnt.

• AI is a valuable resource in health.
• When developing tools, key users must be at the centre of development e.g. tools for health workers must be developed by health workers.
• Multisectoral collaboration is important for innovations in the health sector.
• Opportunities exist for more innovative ideas in health.

IntelSurv Demo: Dr Taylor and McPhail Magwira

• The tool is currently implemented at the local level, with plans to extend to other countries.
• It allows for comparison of data collection methods across different countries.
• Reference databases for IntelSurv architecture are official technical guidelines and documents by Ministry of Health and data collected by health workers.
• Offline access is supported, allowing users to view previous questions and responses.
• The development of the tool was driven by community needs especially by health care workers.
• The presenters emphasize the importance of acceptance and ownership by healthcare workers (HW) as crucial for the tool's success.

The Integrated Disease Surveillance and Response (IDSR) approach in Malawi: Mr Edward Tchado PHIM

• Mr Tchado provided an overview of the IDSR strategy, its core functions, roles, and linkages to other health initiatives.

• IDSR used for planning, implementing, and evaluating public health practices
• Includes passive and active surveillance and operates at national, district, community, and health facility levels.

Types of surveillance

• Indicator-Based Surveillance (IBS): Relies on structured data and signals, with examples like facility-based surveillance, laboratory-based surveillance, disease-specific surveillance (e.g., malaria, TB, HIV), case-based surveillance, syndromic surveillance, and sentinel surveillance.
• Event-Based Surveillance (EBS): Not based on routine monitoring, but on screening available information to detect any unusual events in the community, such as unusual diseases or clustering of cases.

• Both IBS and EBS contribute to early warning and response systems by providing real-time, organized, and reliable data.

• IDSR principles include coordination and integration, promoting efficient resource use, reliable data, and improved flow of surveillance information.

Food for thought

• How Can AI be employed to address prescription, prediction, and inscription in relation to IDSR??

Top of Form

Bottom of Form

Testing Generative AI – Human in the Loop - Interactive Session: Evie Chapuma & Chisomo Kankhwali

• A demo session was conducted with health workers (users) to test IntelSurv and gather feedback on installation, accessibility and speed.

Main findings were as follows:

• The interface was simple and easy to navigate.
• Most installations were successful, with few exceptions.
• The app provided useful information on the meaning of fields and their rationale.
• The history of interactions was found to be useful.
• Users were able to ask general questions, not only those predefined in the app.

Areas of improvement

• The application took time to respond.
• A database of knowledge and FAQs for offline mode is required.
• Some answers were too long, requiring shorter responses.
• The style of responses could be improved.
• Languages should be translated to local languages for easy accessibility.

Using GenAI to transform decision making for Public Health Officials: John Gitonga, Dalberg Data Insights (DDI)

• DDI aims at using data to advance change in Africa.
• Presented case studies on VIDA and AIA Health, showcasing AI tools for decision making in public health.

VIDA case study

• An innovative toolkit comprising comprehensive dashboard and a robust performance focusing on Expanded immunization program (EPI).
• Important for performance monitoring, logistics and communication.

Case study 2: AIA Health

• An AI analyst which transforms decision intelligence for PH officers at MoH.
• Works by integrating GPT into national HMIS thereby enabling users to query disparate data through a web app and transform it into actionable insights instantly.

• Both are based on data retrieval from HMIS and do not offer solutions to data quality issues. Data must be cleaned first before being integrated into the tool.

Opportunities and pitfalls of Applying Generative AI in Healthcare: Dr Taylor

Opportunities

• Language models (LLMs) understand language and generate content.
• They provide information that healthcare workers need for their work, including data collection, addressing gaps in the system.
• LLMs are capable of generating queries.
• These models are large and are algorithms based on machine learning.
• They undergo training to improve their performance and accuracy.

How LLMs work

• Training: process of creating massive database of knowledge chunked.
• Transformers: algorithms for searching and reconstructing text from the database
• Fine tuning
• Prompting

Challenges with LLMs

• Knowledge

• Intrinsic: insufficient, outdated, wrong, biased, not transparent to users
• Representation: knowledge becomes data, information loss, inability to capture context

• Algorithms related: tradeoff between efficiency and accuracy, black box aspects, difficult to test and validate.
• Humans in the loop: Feedback is essential, limitations are being communicated and understood by users, testing is inbuilt.
• Security and privacy: removing sensitive data from the training or fine-tuning data.

Key ingredients needed when developing AI tools

• Good quality documents/knowledge to train and fine-tune LLMs.
• State of the art natural language processing techniques.
• Incorporate feedback from healthcare professionals and patients.
• Provide transparency to users.
• Testing and validation of tools regularly over a significant time

Local Languages and IntelSurv: Paul Kazembe

• Almost all Bantu languages do not have standard vocabulary to match to certain medical terms.
• Descriptions often used which tend to be misleading.
• Solution would be use of the gadget with proper AI tools installed.

Digital Health Programs and opportunities and needs: Dr Mbewe

• Emphasized the importance of collaboration and dialogues among eHealth leaders in digital health.

• Highlighted lack of expertise and personnel in the department of digital health and a need for capacity building.
• The Malawi health information system plans to integrate various departments like labs and pharmacies to ensure effective patient tracking across facilities.
• Government funding for Digital Health initiatives is currently lacking, thereby largely depending on donor/partner support.
• Digital health initiatives should adopt a holistic approach to improve overall patient care, addressing all aspects of healthcare, not just specific diseases.
• All efforts should aim to make a meaningful impact on healthcare delivery and outcomes.
• Encouraged and emphasized on integrating new developments in digital health into already existing digital platforms like iCHIS.

Closing remarks Dr Chikumba

• Emphasized the importance of Malawi's digital health system for overall healthcare.
• He pointed out existing challenges in moving data generated from health facilities to higher levels like district level or sharing with stakeholders, which might be necessary for improving healthcare.
• MUBAS prioritizes partnerships and collaboration, and he highlighted that the University is committed to providing expertise to the digital health department and Ministry of Health through its Masters students, IT experts, and data analytics specialists.
• He expressed gratitude to all participants and especially to Dr. Taylor for organizing the workshop.

End of workshop