My current research project is on the use of mobile patient support tools to assist in the improvement of medication adherence for those suffering from tuberculosis in Sub-Saharan Africa. This is an ongoing project that I look forward to discussing with you more as the project evolves.

A Framework for AI-Driven Patient Support in Tuberculosis Treatment: The Role of Voice and Text Agents

Abstract

Tuberculosis (TB) remains a leading infectious killer worldwide, with a disproportionate burden on low- and middle-income countries, particularly in Africa. A major challenge to global TB control is low medication adherence, driven by factors such as treatment length, side effects, and insufficient patient education. This paper proposes a cost-effective and scalable solution: an AI-driven patient support tool that utilizes a hybrid of voice and text agents on mobile devices. The core concept is that leveraging the widespread availability of smartphones and a multilingual, accessible interface can significantly improve medication adherence and patient knowledge, thereby reducing the spread of TB. This paper will outline the scale of the TB challenge in Africa, detail the proposed solution, and discuss its potential impact from the perspective of a global health expert.

1. The Global and Regional Burden of Tuberculosis

The World Health Organization (WHO) consistently highlights TB as a persistent public health crisis. The WHO Global TB Report 2024 notes that approximately 10.8 million new TB cases and 1.25 million deaths occurred globally in 2023, making it the world's leading cause of death from a single infectious agent.

The burden is especially heavy in the African region, which accounts for approximately 24% of new global cases and 27% of all TB deaths. While the area has seen a notable decline in fatalities (42%) and cases (24%) since 2015, it still falls short of the WHO's ambitious End TB Strategy milestones. The progress is also uneven, with some countries making significant gains while others, particularly in Central and West Africa, lag behind.

A critical factor contributing to this high burden is the infectious nature of the disease. A single person with active pulmonary TB can infect an average of 10-15 other people per year through airborne droplets. This cycle of transmission is perpetuated by a range of factors, which a modern patient support tool could be designed to address.

2. Challenges in Tuberculosis Control: A Call for Innovative Solutions

The effectiveness of TB treatment is highly dependent on patient adherence to a complex and lengthy regimen of antibiotics, which can last for six months or more. Non-adherence is a primary driver of treatment failure, disease spread, and the emergence of drug-resistant strains (e.g., MDR-TB), which are more difficult and expensive to treat.

According to WHO reports and academic research, the reasons for non-adherence are multifaceted and include:

• Lack of Patient Knowledge: Patients often stop taking medication once they feel better, unaware that the bacteria may still be present in their bodies. They may not understand the duration of treatment or the consequences of not completing it.
• Socioeconomic Factors: Poverty, lack of transportation to clinics, and the financial burden of the disease itself are significant barriers to adherence.
• Logistical Challenges: Patients may forget to take their medication, especially with multi-pill regimens, or lack a consistent support system to remind them.
• Misinformation and Stigma: Fear of social stigma can cause patients to hide their diagnosis, leading to a lack of family support and a reluctance to seek or continue treatment.

Traditional public health strategies like Directly Observed Therapy (DOT) have been effective but are resource-intensive. With the growing prevalence of mobile technology, there is an opportunity for new, more scalable interventions.

3. The Proposed Solution: AI-Driven Patient Support Agents via Mobile Devices

This paper proposes the development of a cost-effective AI-driven patient support tool, accessible via mobile devices, to address the core issues of medication non-adherence and patient education. The tool would operate as a hybrid of voice and text agents to maximize accessibility across a diverse African audience.

• Leveraging Mobile Penetration: The foundation of this solution is the widespread use of mobile devices in Africa. Over 30% of the population has internet access, primarily through smartphones, even in countries with lower connectivity. The AI agent would function as an app or a service accessible through a messaging platform, requiring no new infrastructure.
• Hybrid Voice and Text Interface: A key feature of the tool would be its ability to operate via both voice and text. This is critical for reaching an audience with varied literacy levels. Voice functionality would be used for:
  • Automated Reminders: Delivering medication reminders in the user's preferred language.
  • Simple Q&A: Answering common questions about TB symptoms, side effects, and the importance of completing the treatment course.
  • User Check-ins: Allowing patients to confirm that they have taken their medication verbally.
• English as a Lingua Franca: The initial version of the tool could be launched in English. Despite the vast linguistic diversity of Africa, English is a prominent official language in 24 countries and is a widely understood lingua franca for many non-native speakers, particularly among the more digitally connected youth. This approach is a pragmatic and inexpensive first step, and the tool can be expanded to include other languages like Swahili and French in the future as resources permit.
• Addressing the "Sharing" Concern: To mitigate the risk of disease spread through a shared device, the tool would incorporate an essential educational component. Voice and text prompts would regularly remind users about the importance of personal hygiene and the risks associated with sharing a device, reinforcing lessons from in-person healthcare workers.

The cost-effectiveness of this solution is rooted in its scalability. A single AI agent can support thousands of patients at a fraction of the cost of a human-based DOT program, making it a viable and sustainable option for low-resource settings.

4. A Global Health Expert's Perspective to confirm that they have taken their medication verbally

As a global health advocate, I find the concept of an AI-driven patient support tool for TB to be both timely and promising. The proposal correctly identifies the most critical barriers to TB elimination: adherence and education, and it cleverly aligns its solution with the existing technological landscape in Africa, mobile phones. The hybrid voice and text model is a smart way to address the issue of varying literacy levels, which is often overlooked in tech-based health interventions.

However, there are a few points that would need rigorous attention to move from a promising concept to an effective intervention:

1. Pilot Study and Local Context: The assumption that a tool developed for a generic "African" audience will be effective is a risk. A pilot study in a specific, high-burden community is essential. This would allow for the customization of the AI's language, cultural context, and messaging to better resonate with the target population. For instance, a tool developed for a community in rural Nigeria would need to be very different from one for an urban area in South Africa.
2. Privacy and Security: The paper touches on privacy, but this is a paramount concern. Health data is highly sensitive. The AI system must be built with robust data encryption and de-identification protocols from the outset to protect patient privacy and build trust. A clear, concise privacy policy that is understandable to all users, regardless of literacy, would be a requirement.
3. Integration with the Healthcare System: The tool should not operate in a vacuum. Its greatest value will come from its integration with existing healthcare systems. Can the AI agent send adherence data to a healthcare provider's dashboard? Can a patient use the tool to schedule an in-person check-up? The AI is most effective as a complement to, not a replacement for, human healthcare workers.
4. Beyond English: While starting with English is a practical first step, the tool's true potential for large-scale impact will only be realized when it is translated and localized for the most prevalent indigenous languages. As noted in the paper, languages like Hausa, Yoruba, and Swahili are spoken by tens of millions of people who may have limited English proficiency. A phased rollout plan that prioritizes these languages would be critical for equitable access.

In conclusion, this proposal is a viable and important step forward. The emphasis on a cost-effective, adaptable, and mobile-first approach is precisely what is needed to tackle a complex public health issue like TB in Africa. By addressing the critical concerns of cultural context, privacy, and integration, this AI-driven tool could become a powerful new weapon in the global fight against tuberculosis.