AI Tool DistillerSR Cuts Literature Review Time for Healthcare Researchers

As the volume of published research continues to grow, manual screening becomes nearly impossible, as it is either very time-consuming or risks lacking thoroughness. DistillerSR helps address this challenge by reducing manual workload while maintaining review quality. More information about DistillerSR and licenses available for Faculty of Medicine and affiliated institutes can be found in the OMARI blog article, DistillerSR to accelerate and improve systematic reviews.

Article Screening

The review was performed across literature using three databases: PubMed, IEEE Xplore, and ACM Digital Library. At the initial step, literature was imported into DistillerSR. PubMed is integrated with DistillerSR, making it possible to use search queries to retrieve relevant literature directly within the tool. For IEEE Xplore and ACM Digital Library, relevant literature was retrieved from the databases and uploaded into DistillerSR, as many common formats are supported. The tool also automatically identifies and removes duplications that occur across databases. At this stage, centralized data management reduces manual effort and helps avoid missed or repeated records.

Before the Review Process

Before the review process, a workflow must be established. Research questions can be set using Forms and assigned to Levels. DistillerSR proposes preset questions, such as whether a paper is relevant to the study, which can be used as a starting point. Once at least the first-level question is set, the review process can begin. For our research, before using the DistillerSR screening functionality, two independent reviewers reviewed papers to prepare a training dataset. The tool requires 25 references or 2% of total references (whichever is greater) to be reviewed manually before the process can be automated. A larger training dataset can provide more accurate recommendations. DistillerSR then provides recommendations on whether a paper is relevant to the study or not. “0” indicates high confidence of exclusion and “1” indicates high confidence of inclusion. The closer a score is to 0.5, the less confident the prediction. This scoring allows researchers to focus on uncertain papers, reducing time spent on clearly irrelevant studies.

Analyzing Papers and Answering Research Questions

As the next step, DistillerSR helps to analyze selected papers. Managing attachments includes uploading PDF files and preparing them for Smart Evidence Extraction (SEE)™. The technology uses large language models (LLMs) to help generate answers to research questions. Questions can be defined in different formats: single choice, multiple choice, open-ended questions requiring text extraction and summarized responses.

The quality of questions and their compatibility with AI are important. Before running the review process, the tool provides an AI check to evaluate whether questions are AI-compatible, allowing researchers to adjust if needed. 

For example, the question “Does this article describe Artificial Intelligence medical device regulations?” is marked as AI-compatible by DistillerSR because it is clear, non-ambiguous, straightforward, and retrieves information from the article. Another example, the question “Why should this article be excluded?” marked as non-compatible, as it relates to the review workflow rather than the article’s content, and the information cannot be directly extracted from the paper.

To review articles, DistillerSR provides functionality where a selected article (title, authors, publication journal, date) is displayed alongside the questions. Users can switch between the abstract tab and the attachment tab, which displays article’s PDF files. To achieve the best results, recommendations include avoiding overly complex or vague questions, avoiding using abbreviations, breaking down complex open-ended questions into smaller ones, and using the AI check feature before running large batches. Additionally, to AI check for questions, PDF files are  preprocessed before SEE use. Ready PDF files are marked with lightbulb icons. At levels where SEE is used to assist with questions, AI can retrieve related information from prepared files and generate suggested answers to research questions. Moreover, users can see from which part of the article the data was retrieved, and they can review and edit the suggested answers within the system.

Based on our post analysis, the answers provided are generally accurate and helpful. However, it is important to note that SEE is intended to guide researchers in analyzing results and highlighting relevant content in the paper, not to substitute researcher analysis.

Finalize Results

DistillerSR assists in reporting results. The search strategy and screening process can be exported as a PRISMA diagram, which simplifies reporting and supports transparency in systematic reviews. Datarama provides users with a comprehensive overview of the literature screening process and how research questions were answered throughout the review. It also indicates whether decisions were made by the user or generated through DistillerSR’s automated review functionality. The data can be reviewed by users and edited if necessary.

Future Development

DistillerSR is continuously being developed and improved based on user needs. The tool provides comprehensive guidelines that help users become familiar with its functionality, address common questions, and get started efficiently.

During our review, we also contacted the DistillerSR team to ensure that we were using the tool correctly, clarify specific questions, and learn how to use it most effectively. We received valuable, practical feedback tailored specifically to our study.

The latest release of SEE introduces several significant enhancements. Full Form Automation enables SEE to complete and submit entire extraction forms automatically, similarly to a human reviewer. Simplified prompting allows SEE to process all questions in a form simultaneously, giving it full context across the extraction form for improved accuracy and consistency between variables. Every suggested answer includes a plain-language explanation to support explainability, regulatory compliance, and auditability. All extracted data remains fully hosted within DistillerSR’s secure platform, which is important from a confidentiality perspective.

Summary and Learning Points

DistillerSR provides good quality results and helps reduce the time required for literature screening and analysis. It is designed to support the review process, not replace the reviewer. Human oversight remains essential for accurate interpretation. Once the steps to using the tool are understood, it becomes significantly more effective and easier to use. DistillerSR delivers tangible benefits by reducing manual workload, improving consistency, and supporting large-scale reviews when combined with well-defined questions and active researcher involvement. 

I would encourage researchers to explore DistillerSR’s functionality and experience firsthand how the tool can enhance research efficiency and consistency, while saving time in conducting literature reviews. To request access to a license and start with DistillerSR contact [email protected].