person
Manual
Both title/abstracts based first-pass and full-text screening performed by human.
On this page
1. End-to-End Workflow
AI-enabled evidence synthesis with reviewer control
SYMPRO AI supports systematic reviews, targeted reviews, data extraction, risk of bias assessment, and report-writing workflows. Automation accelerates repetitive work, but researchers remain involved in review, validation, and final interpretation.
| Workflow stage | Platform support | Researcher responsibility | Traceable output |
|---|---|---|---|
| Question and protocol | Structured project and PICOS configuration | Define and approve the review method | Documented project criteria |
| Search | Search strategy development and database translation | Validate databases, terms and final strategies | Search strings and retrieved citations |
| Screening | Manual and configurable AI-assisted workflows | Review decisions and resolve disagreements | Eligibility decisions and reasons |
| Critical appraisal | Assessment-tool recommendations and structured processing | Select the appropriate tool and confirm judgements | Reviewable risk-of-bias assessments |
| Extraction | Template-led extraction from selected documents | Verify extracted evidence against sources | Structured evidence tables |
| Reporting | Evidence-linked drafting and revision support | Interpret findings and approve final text | Editable report and evidence trail |
AI technology
SYMPRO AI uses large language models through several LLM API providers and applies retrieval-augmented generation for PDF-grounded screening and extraction tasks. Project data is not used to train SYMPRO AI models, and human verification remains required.
2. AI Integration Levels
AccuScreener for literature screening
Teams can choose how much AI participates in screening, from reviewer-led workflows to fully automated screening with targeted human review. Each mode keeps the decision path visible so reviewers understand how work was completed.
A. Level of AI integration
Screening can be performed using artificial intelligence (AI) with configurable levels of automation:
assist_walker
AI Assistant
AI provides suggestions during title-abstract and full-text screening by applying predefined criteria such as PICOS & hierarchy. In this mode, AI decisions are available for human review.
groups
AI Screener - Double Blind with Analysts
AI screens studies in parallel with human analysts in a double-blind workflow, enabling side-by-side comparison without exposing either decision stream upfront.
smart_toy
AI Screener - AI as Sole Screener
AI is the only screener. Human reviewers will only review the screening decisions without conducting initial screening.
B. Blinding modes
Manual and AI-assisted screenings can be configured in either review mode, depending on the protocol and the level of reviewer independence required.
visibility
Single-blind
One decision stream is hidden during screening to reduce reviewer influence while maintaining structured adjudication.
visibility_off
Double-blind
Independent screening paths stay hidden until comparison, supporting unbiased discrepancy detection and manual resolution.
C. Discrepancy review
rule
In double-blind workflows, disagreements are surfaced automatically so the reviewer can move straight to manual resolution.
One-click discrepancy identification for human review
Other modules
All other modules (Accusearch for searches, Accusynthesis for data extractions, Accuscripter for study linkage and report generation) are powered by AI with human-in-loop review.
Human oversight and controls
rule_settings
Admin/Reviewer oversight
Project admin or human reviewer can sample, review, and override any decision - whether made by analysts or AI.
tune
Usage controls
Developers can set AI usage limits at the organization level, permitting AI operations only within approved governance boundaries.
history_toggle_off
Audit & traceability
All decisions or changes made are recorded with the user initials, indicating whether it was generated by AI, analysts or reviewer.
Core working principles
visibility
Evidence traceability
Each AI-generated evidence can be traced to specific article, with supporting text visibly highlighted in the source PDF.
psychology
Reasons for decisions
For screening decisions and extracted fields, the AI's reasoning is displayed so reviewers can verify and validate each outcome, instead of being forced to trust the system blindly.
verified_user
Human-in-loop review
All AI-driven decisions can be exported to support external human review. Human experts remain fully accountable for protocols, final judgments, and interpretation of results.
3. Human-Controlled Decisions
Users remain accountable for every decision
SymproAI is designed with the core principle that users remain fully responsible for every screening decision. Here's how that works in practice:
settings_suggest
Configured by project admins
Project admins configure protocols, AI modes, and decision rules before reviewers begin work.
visibility
Decision ownership is visible
Reviewers can see who made each decision, including decisions generated by AI.
lightbulb
AI supports, users decide
AI outputs are presented as suggestions, not as final decisions.
credit_score
Governance through credit limits
Credit limits help governance teams control how much AI is used across the organisation.
published_with_changes
Reviewers can override anytime
A reviewer can override any AI suggestion at any time.
4. Publications and Product Evidence
Published work and platform evaluation
The selected publications below cover AccuScript evidence-synthesis work and evaluations involving SYMPRO. They are presented as research and methodology evidence, not as customer testimonials or adoption claims.
Published evaluations
Review articles
- Real-world evidence corroborates the clinical trial results for management of HER2-positive metastatic breast cancer with T-DM1: a systematic review. J Breast Cancer Research (ACCEPTED).
- Impact of sleep disturbances on health-related quality of life in postmenopausal women: a systematic review. 2026. Menopause 33(1): p 118-128, DOI: 10.1097/GME.0000000000002633.
Conference abstracts
- MSR82 Efficacy and Safety of Modern Biologics Compared to Conventional Therapies in AQP-4 Positive Neuromyelitis Optica Spectrum Disorder: A Fully AI-Automated Targeted Literature Review. Value in Health, 28 (22 Suppl) December 2025.
- MSR88 Validation of a programmed probability score using the screening tool, SYMPRO to assist selection of studies in systematic literature reviews. Value in Health, 26 (12 Suppl) December 2023.
What this indicates
Published evaluations can help teams understand how AI-assisted workflows perform in specific settings. Results may vary by topic, protocol, dataset, and reviewer configuration.
We recommend:
- Starting with a pilot project before large-scale adoption.
- Maintaining human decision at every step.
Additional verified evaluations can be added here as their methods and publication details become available.
5. Data Privacy, Protection and Quality Management
Security-conscious handling of project data
SYMPRO AI combines role-based access and company-level data separation with verified GDPR and ISO 9001 credentials. Certificate scope and supporting information are available from AccuScript.
GDPR Compliant
Ensures strong data privacy and protection practices.
ISO 9001 Certified
Follows internationally recognised quality management standards.
GDPR Compliant
Data privacy and protection practices are structured around GDPR-aligned controls.
ISO 9001 Certified
Quality-management processes follow recognized international standards.
No Training Use
Project data is not used to train or improve SYMPRO AI models.
Access Control
Role-based permissions restrict visibility to authorized users and configured admins.
Operational commitments
- User rights: You maintain full ownership of all content and outputs you upload or generate.
- No training use: Your data is never used to train or improve SymproAI models.
- LLM provider controls: We use provider data controls and contractual terms intended to prevent submitted content being used for provider training.
- Role-based access: Access controls restrict access to authorised users, typically the project team; organisation administrators have configurable permissions.
- Regional deployment: Organisations requiring EU or US data residency can discuss separate deployment regions.
- DPAs available: Data Processing Agreements can be provided to organisations where required.
Full details are available in our Privacy Policy and Terms of Service. Organisations can contact AccuScript to request certificate and security documentation relevant to their evaluation.
6. Limitations
Responsible deployment boundaries
SymproAI does not replace human expertise. Researchers should:
Limitations
- Evaluate performance in pilot runs before full deployment.
- Set clear protocols for how AI will be used at each stage.
- Maintain human review for high-stakes decisions such as eligibility criteria, quality assessment, and final conclusions.
7. Alignment with AI Principles
Responsible AI in evidence synthesis
supervisor_account
Core Principles
Human-in-the-loop controls that keep admins/reviewers accountable.
bolt
Application
Accelerates Evidence Synthesis (ES) by identifying studies, extracting data, assessing risk of bias, and summarizing findings.
verified
Trustworthy AI
Documented methods, human review checkpoints, and published evaluations provide evidence for responsible use while further validation continues.
update
Rapid Evidence
Structured workflows support repeat review updates; AccuScripter supports cumulative report updates from Version 1 through Version N.
visibility
Transparent AI assistance
Transparent AI assistance with visible rationale and decision flags.
tune
Configurable workflows
Configurable workflows to suit different review standards.
shield_lock
Security-focused data handling
Role-based access, company-level separation, and documented data-handling controls are used to protect project information.
8. Resources
Supporting material for teams evaluating SYMPRO AI
Teams evaluating or using SYMPRO AI can access:
9. Contact
Get in touch
Contact
For further information or questions, please reach out to contact@accuscript.org
Full details are also available in our Privacy Policy, Terms of Service and Trust & Security page.