Background: Evidence-informed decision-making (EIDM) is critical in addressing complex societal problems. In Uganda, we heavily employ the Rapid Response Service (RRS) as one of the EIDM approaches. The RRS employs conventional methods of evidence synthesis similar to systematic literature review (SLR) processes. These methods, however, often fail to keep pace with the rapid evolution of knowledge and the immediate evidence needs of decision-makers, resulting in delayed decision-making and potentially suboptimal outcomes. Digital tools, such as large language models (LLMs)—eg, ChatGPT—offer a solution by generating answers to decision-makers' questions in real time, improving timeliness compared with conventional methods. However, these digital tools have major limitations, such as outdated data sets, lack of precision, and difficulties in assessing the quality of generated outputs.
Approach: We are designing a digital tool that leverages SLR processes using scholarly evidence sources to respond to policymakers' requests autonomously and promptly for research evidence. Our goal is to adapt SLR processes to enable the tool to autonomously and promptly respond to policymakers' requests for research evidence, thereby bridging the gap between the instantaneous nature of digital tools and the rigorous standards of evidence synthesis. Our tool will address the limitations of current digital tools and also meet the increasing demand for evidence by decision-makers, who increasingly consult publicly available digital tools rather than traditional sources of evidence. The tool promises to set a new standard for prompt, autonomous evidence synthesis to support decision-makers in the realm of evidence-informed policy and practice.
Developed through a rigorous iterative process, our tool will automatically identify key terms from requests for evidence according to the PICOS framework, generate their relevant acronyms, select appropriate databases, create search strings compatible with database architecture, export references, screen and extract relevant information from relevant papers, and come up with a contextualized evidence product.
Expected results: By the conference dates, we expect to have a functional prototype, tested its performance through case studies, and gathered feedback from end users. We anticipate that the tool will demonstrate improved efficiency in responding to policymakers' evidence needs, thus supporting EIDM in various domains, including gender, health care, economics, and social welfare.