Raquel Cerqueira | Hub LAC Communications
Introduction
As part of our Stories of Change series, we spoke with Luis Ortiz, Coordinator of the Evidencia UC team at the School of Medicine of the Pontifical Catholic University of Chile. Evidencia UC is closely linked to Cochrane Chile in relation to guidelines for conducting evidence syntheses. Throughout its trajectory, the team has supported a range of stakeholders in providing evidence to inform decision-making, including the World Health Organization (WHO) and the Pan American Health Organization (PAHO). In addition, the team supports various university initiatives, providing evidence for research projects and clinical decision-making within the university hospital.
Key takeaways:
- The trajectory of Evidencia UC in using artificial intelligence for evidence production.
- A practical case of AI application in an evidence synthesis project for public policy.
- Changes generated in evidence searching, screening, and synthesis processes.
- Reflections on human oversight, transparency, and the responsible use of AI.
Access the full interview on our YouTube playlist
Context and challenges
The Evidencia UC team has been exploring the use of artificial intelligence tools for several years, even before these technologies became widely popularized. Their close connection with the Epistemonikos platform, which originated within the team when it was known as ProSABE, further strengthened their experience in applying AI to evidence searching.
Over time, the team has incorporated different AI tools into various stages of the evidence synthesis process, always under human supervision, particularly during the early years when the capabilities of these technologies were still below human performance standards.
Evidence Synthesis with Artificial Intelligence: a pilot project and its collective learnings
In 2025, a project led by the UC Family Medicine team was selected through a call from the UC Public Policy Center to develop proposals aimed at strengthening preventive actions in primary healthcare in Chile. The study sought to address the gap between screening processes and the follow-up of identified patients, combining fieldwork with a review of the available evidence.
To support this second stage, the Evidencia UC team joined the project. Given the limited timeframe and constraints in human and financial resources, the team opted to develop an evidence synthesis document heavily supported by artificial intelligence. The methodological approach was not intended to replace people, but rather to maintain human oversight throughout the process while redesigning the traditional synthesis workflow to improve efficiency.
Luis describes the synthesis process in seven stages using AI tools:
Step 1 — Formulating the question and eligibility criteria
Step 2 — Searching for “seed papers”
Step 3 — Citation expansion
Step 4 — Developing the search strategy for electronic databases
Step 5 — AI-assisted screening
Step 6 — Human screening
Step 7 — Data extraction and synthesis
To validate the process, the team developed a working checklist called VERIFICA, in which each letter represents a verification step for reviewing AI-generated outputs. The first step is to verify that the references are real—something as simple as clicking on the cited reference and confirming that it leads to the correct article. The next step is to check whether a more recent version of the article exists, since some AI tools may have been trained on older knowledge bases.
“Classifying a tool as better or worse is very difficult because a tool with a limited but academic body of knowledge may perform better than one with a much broader but non-academic knowledge base. It always depends on the databases being analyzed,” explains Luis.
Main changes and advances resulting from the experience
According to Luis, one of the main benefits of this approach has been the significant reduction in production time. While a policy brief typically takes between three and six months to complete, this AI-supported exploratory evidence synthesis was completed in just four weeks.
The experience also enabled the incorporation of new tools into subsequent projects. In a review commissioned by WHO, for example, artificial intelligence was used to expand the evidence search on a particularly complex topic for which developing a search strategy was extremely challenging, as manuscripts were indexed differently across databases.
Although the international standard for evidence synthesis remains traditional reviews with dual human assessment, Luis highlights that these contextual AI tools have proven especially useful for expanding searches and identifying studies that might otherwise remain difficult to locate.
Ethical considerations and human oversight
Luis emphasizes that one of the key safeguards adopted was to complement the use of artificial intelligence with traditional search and screening processes. This strategy made it possible to identify articles that AI tools had failed to retrieve and increased confidence that no relevant documents were being overlooked in decision-making processes.
Evidence syntheses conducted exclusively with artificial intelligence may still miss important studies. For this reason, the approach combined the strengths of AI with human oversight at every stage of the process, incorporating traditional search and review cycles to minimize the risk of excluding evidence that could influence decisions, particularly in the context of public policymaking.
“The most basic principle is to disclose the use of AI. In all our meetings, all our deliverables to different stakeholders, and all prioritization discussions, we explained the process and clearly stated that it was heavily driven by artificial intelligence. We also made it clear that human oversight was maintained throughout every stage.”
Shifts in mindsets and the broader ecosystem
For Luis, the experience represented an important shift for the team, helping to reduce resistance to the use of artificial intelligence in evidence synthesis. The project demonstrated that these tools can be used safely under human supervision, accelerating processes without compromising quality.
He also notes that this experience aligns with broader changes taking place in organizations such as Cochrane, which has already launched initiatives to evaluate the use of artificial intelligence in updating systematic reviews. Some teams are now able to conduct reviews using both traditional methods and AI-supported approaches, allowing comparisons between results and generating evidence on the potential and limitations of these tools.
Luis further highlights that the institutional environment has facilitated this adoption, as organizations such as the Chilean Ministry of Health, WHO, PAHO, and Cochrane are increasingly exploring the use of AI in the production and updating of evidence. Within this context, the team now feels more comfortable incorporating these tools into traditional projects, leveraging their strengths to complement—rather than replace—conventional evidence synthesis methods.
As Luis concludes:
“Although this initiative originated within the Evidencia UC team, the environment has been supportive, and we are grateful for that—for the willingness to innovate, explore, make mistakes, and try again. At this stage, when everything is still emerging and there are many uncertainties, if we do not start testing these tools, we will continue to have the same doubts. We need to address them in real-world settings.”
External resources
Access the supporting materials for this Stories of Change:
