Using AI in Early Prediction of Maternal Complication

Using AI-enabled system in Public Health Sector: Lessons from Mlinde Mama

From Data to Delivery: Practical, Ethical, and Human‑Centered AI in the Public Sector

Lessons from the Mlinde Mama experience on building AI for maternal health in Tanzania.

---

Introduction: Beyond Paper / Theory

Artificial Intelligence (AI) is often discussed in abstract terms, but PHIT’s work with Mlinde Mama shows how AI‑enabled systems can be applied responsibly in the public sector to improve service delivery and save lives. Rather than replacing human systems, AI in Mlinde Mama augmented public health capacity, focusing on antenatal risk prediction in Tanzania’s Geita Region.

Our goal was ambitious: to build a machine learning (ML) model that could predict pregnancy complications and integrate it directly into Tanzania’s national digital health infrastructure. This wasn’t an academic exercise; it was about creating a practical tool for nurses in rural clinics. Here’s what we learned.

How AI Was Used in Mlinde Mama

AI and automation supported the platform’s core functions, drawing from over 300,000 ANC records in the government’s Unified Community System (UCS):

• Personalized Message Scheduling: Based on gestational age and user data for stage‑specific health alerts.
• Pattern Recognition: Analyzing questions from two‑way communication to provide tailored guidance and flag urgent issues like danger signs.
• Risk Prediction: The ML algorithm, developed with MOH experts using methods like XGBoost, was trained to predict hypertensive disorders, achieving 91% accuracy.
• Data Insights for Monitoring: Cleaning and structuring data for continuous ML learning and public health surveillance.

The final model was deployed as a Software‑as‑a‑Service (SaaS) solution with well‑documented APIs, enabling other systems to connect to the ML engine and receive predictions, even offline.

Lesson 1: The Data phit_userity is Always Messier Than You Think

Our first and most profound lesson was about data. We had initially planned to tap into large electronic medical records, but we discovered:

• Fragmented Systems: A woman’s antenatal data was often stored separately from her delivery outcome, breaking the crucial link needed for prediction.
• Incomplete Records: Many facilities relied on paper. Even where digital systems existed, they focused on demographics, not the detailed clinical data we needed.
• Data Gaps: Key parameters were often only tested if a problem was suspected.

This forced us to pivot to the government’s UCS platform. We had to roll up our sleeves and do the painstaking work of cleaning, structuring, and connecting this data ourselves. Start with the data, not the algorithm.

Lesson 2: Build for Integration, But Plan for Flexibility

Our vision was to embed the ML module directly within the UCS. However, the Ministry of Health was in the process of updating UCS, and our timelines couldn’t be directly accommodated.

Instead of giving up, we became flexible. Working in close collaboration with the MoH, we developed the ML engine as a standalone “Prediction Module” using a SaaS model. This adaptability turned a potential failure into a strength. The tool is no longer limited to one platform; it can be used by private hospitals, research institutions, and other health programs, broadening its impact.

Lesson 3: Government Ownership is the Ultimate Goal

From the start, our mantra was sustainability. By choosing the government‑owned UCS and co‑creating the solution with MoH officials, we ensured the final product was not “PHIT’s tool” but “Tanzania’s tool.” This meant navigating bureaucracy patiently, building local capacity, and focusing on what the system needed, not what we wanted.

Key Principles for AI in the Public Sector

PHIT’s experience highlights critical lessons for responsible innovation:

1. Public Interest First: AI served citizens by reducing facility burdens and prioritizing high‑risk cases, not just chasing efficiency.
2. Transparency and Trust: Co‑creation with the MOH built adoption, and feedback loops allowed providers to correct and modify system output.
3. Human Oversight is Non‑Negotiable: Providers interpreted AI predictions, ensuring ethical use in health. The AI augments, not replaces, human judgement.
4. Equity and Inclusion: The system was designed for rural, low‑literacy users with SMS fallbacks, reaching over 5,900 women.
5. Data Protection and Ethics: We adhered strictly to MOH and NIMR standards, with clear data‑sharing protocols and no IP filings, ensuring citizen data was protected.

Why This Matters for Governments and Partners

When implemented responsibly, AI can:

• Extend the Reach of Public Services: In Mlinde Mama, it helped boost ANC4+ to 93.9%.
• Improve Consistency and Quality: Leading to higher test completion rates and a dramatic 88% reduction in adverse birth outcomes for women with four or more visits.
• Support Evidence‑Based Decisions: Insights from endline evaluations and published studies directly inform scale‑up plans.
• Strengthen Accountability: Monthly data reviews and dissemination at conferences like the IDM Symposium 2024 ensure continuous learning.

PHIT’s Vision for the Future

Mlinde Mama demonstrates that AI can be practical, ethical, and impactful in low‑resource public sector settings when grounded in phit_user needs and strong governance. PHIT is committed to advancing responsible digital and AI‑enabled public infrastructure that:

• Strengthens government systems for ownership and sustainability.
• Protects citizens’ rights by prioritizing equity and ethics.
• Delivers measurable social impact, as seen in Geita’s improved maternal outcomes.

Contact info@phit.or.tz to discuss AI applications in health.