ROBINS-I
Risk Of Bias In Non-randomized Studies — of Interventions
CoreSR implements the ROBINS-I tool with structured GLASS assessment. The model answers signaling questions across seven bias domains for non-randomized studies of interventions, and the tool keeps transparent rationale for every judgement.
AI Assessment Is a Starting Point
AI-generated assessments should always be reviewed and verified by a human assessor with domain expertise. The structured AI output highlights uncertainty and source-evidence gaps, but the final judgement is yours. For published systematic reviews, all RoB assessments should be conducted by at least two independent human reviewers.
How It Works
Five steps from PDF upload to a complete domain-level risk of bias assessment for non-randomized intervention studies.
Step-by-Step Guide
Each step builds on the previous one. AI assessment runs automatically after you provide the study details.
Upload Study PDFs
Upload one or more full-text PDF files for the non-randomized study you want to assess. The AI extracts text from the PDFs to inform its domain-level judgements.
Enter Study Details
Provide the study title, authors, publication year, intervention of interest, and comparator. Specify the target trial design that the non-randomized study is emulating.
Run AI Assessment
Click "Evaluate with AI" to run a structured GLASS assessment. The model answers ROBINS-I signaling questions for each domain, and the tool derives domain and overall judgements from those structured answers.
Review and Override
Review each domain's signaling questions, AI justifications, and judgements. You can override any answer or domain judgement. Pay special attention to the confounding domain, as it often requires expert knowledge of the clinical area.
The Seven Domains
ROBINS-I evaluates risk of bias across seven domains, each assessed through a series of signaling questions.
Bias due to confounding
Is there potential for confounding of the effect of the intervention? Did the authors use an appropriate analysis method that controlled for all the important confounding domains? Were confounding domains measured validly and reliably?
Bias in selection of participants into the study
Was selection of participants into the study (or into the analysis) based on participant characteristics observed after the start of intervention? Were the post-intervention variables that influenced selection likely affected by the intervention?
Bias in classification of interventions
Were intervention groups clearly defined? Was the information used to define intervention groups recorded at the start of the intervention? Could classification of intervention status have been affected by knowledge of the outcome or risk of the outcome?
Bias due to deviations from intended interventions
Were there deviations from the intended intervention beyond what would be expected in usual practice? Were important co-interventions balanced across intervention groups? Was the intervention implemented successfully for most participants?
Bias due to missing data
Were outcome data available for all, or nearly all, participants? Were participants excluded due to missing data on intervention status? Were participants excluded due to missing data on other variables needed for the analysis?
Bias in measurement of outcomes
Could the outcome measure have been influenced by knowledge of the intervention received? Were outcome assessors aware of the intervention received by study participants? Were the methods of outcome assessment comparable across intervention groups?
Bias in selection of the reported result
Is the reported effect estimate likely to be selected, on the basis of the results, from multiple outcome measurements, multiple analyses, or different subgroups?
Structured AI Review Process
GLASS answers structured signaling questions, then the tool stores domain judgements with source-grounded rationale for human review.
Structured Answers
The assessment is built from signaling-question answers rather than an unstructured overall guess. Each answer includes rationale and confidence so the basis for the domain judgement is inspectable.
Uncertainty Flags
Low-confidence answers, no-information answers, and sparse source evidence remain visible in the domain details so reviewers can focus attention where the automated assessment is least certain.
Human Override
You can override any signaling question answer or domain judgement. Your decision always takes precedence over the AI. Overrides are tracked and clearly marked in exports for audit trails.
Judgement Scale
Comparable to a well-performed RCT for this domain.
Sound for a non-randomized study but not equivalent to a well-performed RCT.
The study has a serious problem in this domain that substantially lowers confidence.
The study is too problematic in this domain to provide useful evidence.
Insufficient information reported to assess risk of bias for this domain.
Frequently Asked Questions
Common questions about ROBINS-I assessments in CoreSR.
What is ROBINS-I?
When should I use ROBINS-I instead of RoB 2.0?
When should I use ROBINS-I vs ROBINS-E?
How does confounding assessment work?
What is the "target trial" concept?
How does the judgement scale differ from RoB 2.0?
How does the AI review work?
Can I export the results?
Ready to assess?
Upload a study PDF and let AI provide a first-pass ROBINS-I risk of bias assessment.
Go to ROBINS-IReferences
1. Sterne JAC, Hernán MA, Reeves BC, et al. ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions. BMJ. 2016;355:i4919. doi:10.1136/bmj.i4919
2. ROBINS-I V2 development group. The Risk Of Bias In Non-randomized Studies — of Interventions, Version 2 (ROBINS-I V2) assessment tool. November 2025.
3. Schünemann HJ, Cuello C, Akl EA, et al. GRADE guidelines 18: How ROBINS-I and other tools to assess risk of bias in nonrandomized studies should be used. J Clin Epidemiol. 2019;111:105-114. doi:10.1016/j.jclinepi.2018.01.012