Architecting Uganda’s Data-Driven Health Future

Kampala, Uganda | THE INDEPENDENT | As Uganda’s health system undergoes digital transformation, we speak with Peter Mukiibi, a senior health data scientist whose work spans HIV analytics, supply chain optimization, and disease surveillance. Trained at Makerere University in computer engineering with advanced studies in medical systems engineering from Germany, he offers insights on AI, interoperability, and the future of health systems and analytics in East Africa.

QUESTION: Your portfolio spans HIV cohort analytics, ARV stock-out prediction, and One Health disease surveillance. How has Uganda’s health data landscape evolved over the past decade?

ANSWER: The transformation has been fundamental but incomplete. A decade ago, district health officers compiled monthly reports manually from paper registers, often months in arrears. Today, DHIS2 provides near-real-time reporting from over 80% of facilities in many districts. UgandaEMR and OpenMRS track patient encounters beyond just HIV care. The national eLMIS digitizes commodity flows that were previously invisible.

We’ve shifted from data scarcity to data fragmentation. The challenge now is making systems communicate. When we built the ARV stock-out prediction system in 2016, we achieved 30% reduction in stock-out days and forecast accuracy with MAPE below 12% by integrating LMIS transactions with DHIS2 service volumes. But that integration required custom-built pipelines. Every partner solving similar problems rebuilds these connections from scratch, which is enormous waste of technical capacity.

The most exciting evolution is from retrospective reporting to predictive analytics. The stock-out predictor used XGBoost with lagged features to forecast facility-level needs 7-11 weeks ahead, enabling pre-emptive redistribution. Our HIV cohort dashboard in 2019 used Kaplan-Meier survival analysis to improve nine-month retention from 71% to 80% by identifying high-risk patient segments and struggling facilities. We’re moving from “what happened?” to “what will happen, and how can we prevent it?”

East Africa broadly mirrors this pattern. Kenya ahead on mobile health innovations, Tanzania investing heavily in DHIS2, Rwanda achieving impressive standardization. But everywhere faces the same core challenge: donor-driven vertical programs create parallel systems that don’t interoperate. We’re drowning in systems but thirsting for integrated intelligence.

QUESTION: You’ve deployed machine learning for supply chain and cohort analytics. What are the specific challenges in Uganda’s context, and what technical strategies work?

ANSWER: Challenges operate at three levels: data quality, infrastructure, and institutional trust.

Data quality is the persistent struggle. Building the HIV cohort dashboard required linking EMR encounters, lab results, and pharmacy refills, but these lived in separate databases with inconsistent patient identifiers. Without national ID integration, we had patients indexed by clinic numbers, phone numbers, none reliably unique. We spent as much time on probabilistic matching and de-duplication as on actual survival modeling.

For stock-out prediction, we merged asynchronous streams, weekly LMIS updates, monthly DHIS2 reports, manually recorded lead times. We used Prophet for time-series forecasting precisely because it handles missing data gracefully, and XGBoost with carefully engineered lagged features.

Infrastructure constraints are real. Internet reliability outside Kampala remains challenging. We design for offline-first operation, local server deployments with cloud synchronization when connectivity permits. For the integrated mortality review system in 2021, we used ensemble classification with spatial smoothing in R and GeoPandas, but deployed on district servers, not purely cloud-based.

Institutional trust determines whether models get used. When the stock-out system flagged high-risk facilities, supply officers initially distrusted it, “How does the computer know better than my experience?” We addressed this through radical transparency: publishing feature importance scores showing exactly what drove predictions, consumption trends, lead time variability, upstream stock levels. We built feedback mechanisms where managers could annotate predictions with local context. As predictions proved accurate and prevented crises, trust grew organically.

My technical philosophy for Uganda: Use proven, open-source tools over cutting-edge but fragile technologies. PostgreSQL over proprietary databases. Python and R over specialized languages. Invest in data quality and feature engineering more than exotic algorithms, gradient boosting on clean data beats deep learning on messy data every time. And design with the end user, the district health officer, at the center, not the data scientist.

QUESTION: Interoperability challenges pervade your projects. How should Uganda approach health data standards and integration?

ANSWER: Interoperability is perhaps our most critical unsolved problem, as much governance as technical challenge.

The current state is untenable. Uganda’s Ministry of Health operates a dozen major systems in parallel: DHIS2, UgandaEMR, OpenMRS, various lab systems, eLMIS, eIDSR. They rarely communicate. When they do, it’s through brittle custom integrations, built project-by-project. I’ve built several, the EMR-to-DHIS2 pipeline achieved 95% concordance with manual tallies using HAPI FHIR and Mirth Connect. But here’s the problem: that serves one program flow. Another partner connecting a different EMR must rebuild it entirely.

HL7 FHIR and OpenHIE provide the right architectural vision. FHIR is modern. RESTful APIs, JSON, aligned with how contemporary developers work. It’s composable, start with patient demographics and immunization records, expand gradually. OpenHIE defines how health information exchange should be structured: shared services, patient master index, facility registry, terminology services, accessible to all systems via standardized APIs.

When we built the One Health platform in 2018 for zoonotic disease surveillance, we merged human case data from DHIS2, animal disease reports from agricultural databases, and environmental observations. We used MongoDB for heterogeneous data and NetworkX for transmission network analysis. It worked for that case but wasn’t generalizable. If building today, I’d push hard for FHIR-based resource definitions for cross-sectoral reuse.

Implementation requires institutional structure. Uganda needs a national health information exchange, a funded entity with authority to mandate standards and provide shared services. Specifically:

1. Establish authoritative registries as public goods, facility master list, health worker registry, patient master index maintained centrally, accessible via APIs
2. Mandate FHIR compliance for any system procured with government or donor funds
3. Publish Uganda-specific FHIR Implementation Guides defining our programs, identifiers, workflows
4. Deploy an interoperability layer (OpenHIM or similar) for mediation, routing, security
5. Develop parallel data governance frameworks. The Data Protection Act of 2019 provides foundation, but we need health-sector-specific regulations

The One Health platform required data-sharing memoranda between Ministries of Health, Agriculture, and Water. That negotiation took longer than technical integration but was necessary.

We need local capacity to lead this. Makerere’s School of Public Health and Department of Computer Science should jointly offer a dedicated program in health data analytics training. Currently that expertise is learned on the job, if at all. Without Ugandan FHIR specialists who understand both the standard and our health programs, interoperability will remain consultant-dependent and unsustainable.

QUESTION: How should medical and public health education evolve in Uganda to prepare professionals for AI-augmented healthcare?

ANSWER: Current curricula are fundamentally inadequate for data-intensive health systems. Medical schools teach statistics in isolation, rarely integrated with clinical reasoning. We’re graduating doctors who can’t interpret epidemiological curves or question why an algorithm flagged their patient as high-risk.

As AI proliferates, and it will rapidly, this gap becomes dangerous. We risk a two-tier system: data scientists building models without clinical knowledge, and clinicians using tools without understanding their assumptions or limitations.

Three pillars are essential. First, data literacy as core competency for all health professionals. Every medical student, nurse, public health officer should graduate understanding data provenance, limitations, and critical interpretation. Not making everyone programmers, but cultivating healthy skepticism: Is this malaria spike real epidemiology or a reporting artifact? This requires longitudinal integration through clinical training, not standalone biostatistics courses.

Second, specialized health informatics training. We need professionals combining health expertise with technical skills in data engineering, statistical modeling, and system design. Currently this training essentially doesn’t exist in Uganda. People like me piece it together through international programs or self-directed learning.

Makerere should for instance establish a master’s in public health data science jointly between the school of Public Health and the department of Computer Science, including: health information systems architecture (DHIS2, EMRs, FHIR); applied machine learning (predictive modeling, survival analysis, causal inference); health economics; statistical programing using R, SQL, STATA, Python and SAS; data governance, privacy and ethics; implementation science and system design. Then all this should be augmented with practicum placements on real ministry of health priorities.

Third, continuous professional development. We can’t wait 5-10 years for curriculum reform. The Ministry’s Resource Centre should establish structured training with defined competencies: Basic track for all district health officers (data quality assessment, dashboard interpretation, evidence-based decision-making); Intermediate for HMIS coordinators (DHIS2 configuration, analytics); Advanced for national analysts (predictive analytics, system design).

AI literacy specifically requires urgent attention. Interesting timing. ChatGPT just launched last month and it’s revelatory. It demonstrates both the promise and peril ahead. The ability to generate human-like text, synthesize information, even write code has immediate applications. Imagine AI assistants helping clinicians draft patient summaries, search medical literature, or explain complex conditions to patients in local languages.

But ChatGPT also highlights critical challenges. It can confidently generate plausible sounding but incorrect medical information, “hallucinations” in AI parlance. It reflects biases in training data. It has no understanding of Ugandan health contexts unless explicitly provided. This is precisely why health professionals need AI literacy: understanding when to trust versus challenge AI recommendations, recognizing limitations, exercising appropriate clinical override.

Medical curricula should include case studies where AI tools failed and why. For example, image recognition models trained on Western populations underperforming on African patients with different skin tones or disease presentations. Or how language models like ChatGPT might confidently suggest treatments contraindicated in resource-limited settings.

Tools like ChatGPT could democratize access to medical knowledge, a rural clinical officer could query it for differential diagnoses or treatment protocols. But without training in critical evaluation, it’s dangerous. That training must start now, in medical schools and through continuous professional development.

QUESTION: Your systems required complex stakeholder navigation. What governance lessons have you learned about sustainability?

ANSWER: Sustainability is where most digital health projects die! Rarely for technical reasons. Key lessons:

Co-ownership from inception. Don’t build sophisticated analytics then “present” them for adoption. For the HIV cohort dashboard, we convened a technical working group from the start. ART program manager, district focal persons, facility in-charges, M&E officers, patient advocates. Every design decision was debated and consensus built. Slower upfront but created genuine ownership that drove usage.

Embed within existing structures. Donor-funded projects love creating new “data units” outside the Ministry’s formal structure. These evaporate when funding ends. We deliberately housed the stock-out predictor within the existing Logistics Management Unit at National Medical Stores and the National ART Program. We trained existing staff, not hiring external teams. The system outlasted project funding because it became institutional practice.

Document everything. For the mortality review system, we created governance charters, standard operating procedures, technical architecture diagrams, job aids with screenshots. When the original lead moved districts, the system continued because knowledge was distributed, not locked in one person’s head.

Design for median users, not power users. We initially built sophisticated dashboards with complex filters and visualizations which translated into minimal usage. Then we observed district health officers’ actual workflows. They needed three questions answered: “Which facilities struggle?” “Are we meeting targets?” “Where should I supervise?” We radically simplified to foreground those with red/yellow/green indicators and priority maps. Usage soared.

Create feedback loops. For stock-out prediction, we held quarterly reviews where managers could challenge the model: “Why did it miss this crisis?” “Can we add this data source?” We transparently investigated failures and updated models. This built trust, users saw a learning system, not an infallible oracle.

Secure budget lines. Unless operational costs, hosting, maintenance, staff time appear in annual Ministry budgets, systems die when donor projects end. This requires multi-year negotiation with Ministry finance teams but is non-negotiable for sustainability.

The meta-lesson: Sustainable data systems are socio-technical, not just technical. Code is maybe 30% of the challenge. The other 70% is governance, stakeholder buy-in, training, documentation, budget security, and change management.

QUESTION: Looking toward 2025, what AI applications are most promising for Uganda, and what prerequisites must be in place?

ANSWER: The promise is real but requires responsible deployment.

Most promising applications. Predictive resource allocation. We’ve proven this with 30% stock-out reduction. This generalizes, forecasting malaria surges based on rainfall and historical incidence, pre-positioning supplies before crises hit. Moving from reactive crisis response to proactive prevention.

Clinical decision support for non-specialists. Uganda has one radiologist per 2 million people. AI-assisted chest X-ray interpretation for TB can extend specialist expertise to Health Center IIIs, helping general clinicians who see TB rarely. The AI doesn’t replace clinical judgment, it augments it, reducing false negatives. Similarly, diabetic retinopathy screening via AI-analyzed retinal photos enables task-shifting to nurses with basic imaging equipment.

End-to-end supply chain optimization. Beyond facility-level forecasting to entire network visibility, analyzing National Medical Stores to regional hubs to facilities, identifying bottlenecks, predicting delays, dynamically routing shipments based on real-time conditions.

Diagnostic accuracy through data integration. Most algorithms rely on isolated findings. AI can integrate clinical presentation, labs, patient history, and local disease prevalence to improve diagnosis. For example, combining RDT results with symptoms, travel history, and outbreak status for malaria diagnosis where RDT accuracy is compromised.

ChatGPT and large language models introduce fascinating possibilities. Imagine AI assistants helping health workers draft patient notes, query treatment protocols in Luganda or Ateso, or synthesize complex medical literature into actionable summaries. For training, AI tutors could provide personalized clinical case simulations for medical students.

However, ChatGPT’s limitations are instructive. It generates plausible but sometimes incorrect information confidently. It lacks Uganda-specific context unless provided. It can’t replace clinical judgment. But as a tool for information access, especially in rural areas with limited specialist consultation, it could be transformative if health workers are trained to use it critically.

Prerequisites for responsible deployment. Data governance. Before scaling AI, we need clear protocols: Who owns health data? What consent for secondary use? How do we de-identify while preserving utility? The Ministry should establish a Data Governance Committee with authority to approve data sharing, audit compliance, and sanction violations.

Algorithmic transparency. Every deployed AI tool needs documentation: training data, accuracy metrics by subgroup, limitations, failure modes. Prefer interpretable models over opaque deep learning where possible. A 90% accurate explainable model often beats 93% accurate black box.

Local validation. We cannot import AI tools built on Western data. They must be rigorously validated on Ugandan populations with explicit attention to performance across urban/rural, age groups, disease stages. Validation by independent teams, results published transparently.

Local capacity. If all AI tools are built by foreign companies, we create permanent dependency. Makerere should offer advanced machine learning for health applications. Create fellowships for Ugandan data scientists on health AI. Prioritize open-source over proprietary.

Equity audits. If training data underrepresents refugees, rural communities, marginalized groups, models will underserve them. For every application, analyze: Does this perform equally across regions, gender, socioeconomic strata? Uganda hosts 1.4 million refugees, our models must serve them equitably or we risk algorithmic marginalization.

AI should be one tool in comprehensive strategy, not substitute for primary healthcare investment, clinician training, and infrastructure improvement. The fanciest model is useless if districts lack resources to act on recommendations.

QUESTION: One policy priority to accelerate Uganda’s data-driven health transformation over the next five years?

ANSWER: Establish a National Health Informatics Institute, a funded, authoritative institution responsible for standards, interoperability infrastructure, capacity building, and technical assistance.

Why this matters. Uganda’s digital health is fragmented, donor-driven projects solving narrow problems in isolation. PEPFAR funds HIV EMRs, Global Fund supports DHIS2, CDC invests in lab systems, UNICEF backs vaccine logistics. Each makes sense individually but together creates a dysfunction where data can’t flow and we constantly reinvent solutions.

As someone building integrations, I see enormous waste. Every partner building EMR-to-DHIS2 pipelines solves a problem that should be solved once, robustly, and made available to everyone. This is system architecture failure; we lack the connective tissue allowing pieces to function coherently.

What the Institute would provide. Standards development and enforcement. Authoritative body for health data standards. Publish Uganda-specific FHIR Implementation Guides. Maintain terminology mappings, ICD-10, RxNorm, LOINC, adapted to our context. Require standards compliance for any system serving public facilities.

Shared interoperability infrastructure. Operate national data services any authorized system can leverage: patient master index ensuring unique identification; facility registry providing authoritative location data; health worker registry; terminology services; interoperability layer (OpenHIM) mediating secure exchange.

Technical assistance. Maintain health informaticians, data engineers, FHIR specialists providing implementation support. District wants dashboards? Institute provides templates, training, troubleshooting. Partner needs DHIS2 integration? Institute has documented patterns and hands-on assistance.

Training and certification. Structured pathways from basic DHIS2 configuration through advanced analytics and system architecture. Create recognized credentials and professional career paths for health informaticians.

Why maximum leverage. Because it’s infrastructure that makes everything else effective. Train 100 data scientists, but if they’re building redundant integrations, you’ve wasted talent. Procure analytics tools, but if they can’t access integrated data, they’re paperweights. Fund 50 projects, but if they can’t interoperate, you’ve created 50 silos.

The Institute is the force multiplier, allowing individual investments to reinforce rather than duplicate each other.

QUESTION: What would be your closing reflection?

ANSWER: Three things sustain my optimism despite challenges:

The talent: Uganda produces remarkable young people from Makerere and other institutions combining technical brilliance with mission-driven purpose. If we create environments where they flourish locally rather than drain abroad, the future is extraordinarily bright.

The demonstrated impact: Stock-out predictor reduced ARV interruptions measurably, cohort dashboard improved retention demonstrably, mortality surveillance identified actionable hotspots. We’re past proof-of-concept. These are operational systems delivering value, creating momentum for further investment.

The generational shift: Emerging leaders in Ministry of Health and districts are digitally native and data-literate. They see dashboards and analytics as normal management tools, not exotic add-ons. This cultural shift is profound.

What tempers optimism: risk of moving too fast without foundations, deploying AI before sorting interoperability, collecting data before establishing governance, pursuing innovation before ensuring sustainability. Technology seduction can distract from less glamorous institution-building, workforce training, and standards establishment.

My hope is Uganda chooses the harder but durable path: investing in foundations even when they’re not flashy, prioritizing interoperability over vertical solutions, building local capacity even when hiring foreign consultants is faster.

Uganda can lead data-driven healthcare for Africa, not because we’re wealthiest or have most advanced technology, but because we’re building thoughtfully with sustainability, equity, and local ownership at the center. That’s the vision worth fighting for, and despite challenges, I believe it’s achievable.