About the Project

Supplementary Appendix to Forrest et al. Mapping AI Diagnostic Innovation Networks in Sub-Saharan Africa: A Social Network Analysis of Trials, Institutions, and Funders

Current Dataset Status

Last update:November 30, 2025
Next scheduled update:January 31, 2026
Cumulative trials screened:14 unique trials (from 101 total registry records across ClinicalTrials.gov, PACTR, and WHO ICTRP after deduplication)
Trials included in final analysis:11 trials (3 excluded during verification)
Exclusion reasons: (1) Insufficient Sub-Saharan African presence (single pilot site, primary operations in Europe); (2) Non-diagnostic application (treatment optimization algorithm); (3) Withdrawn registration before enrollment.

1. Living Systematic Review Platform: Governance and Quality Assurance

Platform Overview

We developed a Living Systematic Review Platform (LSRP) to maintain currency of AI diagnostic trial mapping in Sub-Saharan Africa through systematic, automated surveillance of trial registries combined with rigorous manual verification procedures. The publicly accessible domain https://africa-ai-trials.net will be maintained for a minimum of 1 year after publication, and the Github repository located at https://github.com/drjforrest/africa-ai-trials.net.git.

Update Protocol

Search frequency: Quarterly automated searches executed in January, April, July, and October of each year.
Registry monitoring:
  • Automated API queries against ClinicalTrials.gov, Pan African Clinical Trials Registry (PACTR), and WHO International Clinical Trials Registry Platform (which aggregates ISRCTN and other regional registries)
  • RSS feed subscriptions monitoring new trial registrations matching inclusion criteria
  • Email alerts from registry administrators for trials tagged with relevant disease and technology keywords
Search strategy consistency: Identical Boolean queries executed across all time points using registry-specific syntax. Search parameters remain constant to ensure longitudinal comparability while registry interface updates are tracked and search strategies adapted as needed to maintain semantic equivalence.
Publication tracking: Monthly automated PubMed/MEDLINE searches identify published results from registered trials using trial registry numbers as search terms. Published studies not linked to registry entries undergo manual investigation to determine registration status and eligibility.

Quality Assurance Procedures

Automated screening: Registry API responses filtered by:
  • Geographic inclusion criteria (trials conducted in Sub-Saharan African countries)
  • Technology keywords (artificial intelligence, machine learning, deep learning, neural networks, computer vision, natural language processing)
  • Diagnostic application (excluding therapeutic intervention trials, prevention trials, health systems strengthening without diagnostic component)
  • Trial status (planned, recruiting, active, paused, completed, published; excluding withdrawn or terminated prior to enrollment)
Manual verification: Two independent reviewers (JF, MT) verify each trial identified through automated screening meets full inclusion criteria:
  • Conducted entirely or partially in Sub-Saharan Africa (trials with mixed sites must include ≥1 SSA site)
  • Evaluates AI diagnostic technology as primary intervention or comparison
  • Involves prospective human participants (excludes retrospective analyses, simulation studies)
  • Registered in recognized clinical trial registry (excludes informal pilot studies, unregistered research)
  • Sufficient documentation to extract institutional relationships and funding sources
Source validation: All institutional relationships, funding sources, and technology specifications verified from official trial registration documents. Institutional affiliations extracted from "Sponsor," "Collaborators," and "Study Sites" registry fields. Funding sources extracted from "Funding Source" fields in official registrations, not from press releases or news coverage.

Version Control and Tracking

  • Git repository maintains all dataset versions with timestamped commits
  • Change log documents all additions, corrections, and exclusions with justification
  • Audit trail preserves original extraction for any corrected data with rationale for changes
  • DOI-versioned dataset releases correspond to manuscript submissions and publications

Data Accuracy Safeguards

No self-reported entries: Platform does not accept investigator-submitted trial data. All information extracted exclusively from authoritative registries ensures independence from investigator reporting biases.

Limitations Acknowledged

Registry data quality: Trial registries depend on investigator-reported information. Errors in institutional affiliations, funding sources, or intervention descriptions may exist in original registrations and are not detectable without direct investigator contact.
Late registration: Trials may be registered after recruitment begins, potentially missing early partnerships or funding sources present at trial initiation but not documented in final registration.
Incomplete registration: Some trials provide minimal detail in registry entries, particularly regarding institutional roles, funding amounts, or technology specifications. Trials with insufficient documentation are excluded from analysis.
Informal relationships: Registry data captures formal organizational participation but not informal collaborations, shared resources, or advisory relationships that may exist outside documented trial structures.
Platform limitations: Automated surveillance cannot detect trials never registered in monitored databases. Trials registered in national registries not aggregated by WHO ICTRP may be missed unless published results link to registry entry.
Betweenness centrality interpretation: The bipartite trial-institution network structure inherently suppresses betweenness centrality — zero betweenness is expected in a network where trials and institutions form isolated components, and does not necessarily indicate the complete absence of informal collaboration pathways outside the registry-documented trial structures.

2. Search Strategies and Registry-Specific Syntax

ClinicalTrials.gov Search Strategy

Interface: Advanced search using field-specific queries
Field specifications:
  • Intervention/Treatment: Contains AI technology terms
  • Condition: Any (broad capture given diverse disease applications)
  • Other terms: Contains diagnostic terms
  • Locations: Sub-Saharan African countries
Date range: All registered trials through November 30, 2025
Status filters: Recruiting, Active not recruiting, Completed, Enrolling by invitation, Not yet recruiting (excludes Withdrawn, Suspended, Terminated before enrollment)
Results: 47 trials screened → 7 trials included after manual verification

Pan African Clinical Trials Registry (PACTR) Search Strategy

Interface: Basic search (PACTR does not support advanced Boolean queries)
Search approach: Multiple individual searches combined:
  1. "artificial intelligence" + "diagnostic"
  2. "machine learning" + "screening"
  3. "deep learning" + "detection"
  4. "computer vision" + "diagnosis"
  5. "AI diagnostic"
  6. "neural network diagnostic"
Geographic filter: All trials registered as occurring in African countries
Date range: All registered trials through November 30, 2024
Results: 23 trials screened → 3 trials included after manual verification

WHO International Clinical Trials Registry Platform (ICTRP) Search Strategy

Interface: Advanced search portal aggregating multiple registries including ISRCTN, CTRI, ANZCTR, others
Filtered by: Trials with at least one study site in Sub-Saharan Africa
Date range: All registered trials through November 30, 2024
Results: 31 trials screened → 1 trial included after manual verification (not previously identified in ClinicalTrials.gov or PACTR)

Supplementary PubMed Search for Published Studies

Purpose: Identify AI diagnostic studies published from Sub-Saharan Africa that may not appear in trial registries (e.g., pilot studies later formalized as trials, studies with late registration)
Cross-referencing procedure:
  • Published studies checked for clinical trial registry numbers in methods sections
  • Unregistered studies excluded from network analysis (focus on formal trial infrastructure)
  • Studies with registry numbers verified against trial registry databases
Results: 156 publications screened → 11 linked to registered trials in primary databases (no additional trials identified through PubMed that were not already captured)

Platform Access

The Living Systematic Review Platform is publicly accessible, enabling community verification, independent replication, and ongoing monitoring of this nascent field. The platform includes:

  • Complete dataset with version history
  • Search strategies and screening protocols
  • Data extraction forms and decision rules
  • Analysis code for network construction and metrics calculation
  • Interactive visualizations of network evolution over time
GitHub Repository

Contact Information

JF

Jamie Forrest, PhD

Postdoctoral Research Fellow, University of British Columbia

james.forrest@ubc.ca

For more information, or if you have data you wish to contribute to the project, please contact Jamie Forrest.

Technology

Built with modern web technologies for optimal performance and user experience:

Next.js 15React 19D3.jsTypeScriptTailwind CSSPython 3.12