Syneos Health
Syneos Health runs more than 500 clinical trials simultaneously, and test sites require manual collection and comparison of a large number of sites, subjects and historical data, often for months; the team also handles forecasts, files and client landscape discussions.
The company builds a unified analysis platform on data assets that have been relocated to Azure, using Data Lake, Databricks, machine learning, data quality capabilities and Azure OpenAI to develop generative AI applications; the system masters serve as a list of candidate sites, then support a gradual screening process, and test the extension risk through predictive models, allowing staff to demonstrate different options in real time.
AI generates candidate sites, predicts risks and supports scenario analysis; clinical operations and medical specialists review site qualifications, risks and compliance requirements, and final test design and patient safety responsibilities remain with professional teams.
- Modern data platforms before entering the generative AI scene
- Select the first end-to-end process with the site
- Connect prediction models to generate interactions
- Bringing clinical teams into real-time ICP in client meetings
- Validate results using candidate list duration and active cycle
- Generating AI should reuse existing data and predictive capabilities of the enterprise
- Clinical scenes need to separate candidate generation from final eligibility decisions
- Project cycle, duration of local processing and final business cycle to be measured separately
- Risk alerts are valuable only if they enter the team disposal process.
The Microsoft case states that the platform was deployed over a nine-month period and that the initial site list could be generated from 24 to 48 hours, with subsequent screening reduced from several months to several weeks; and the 2024 site activation cycle reduced by approximately 10 per cent.
- The 24-48 hours, weeks of screening and 10 per cent reduction in the activation cycle of the site list are indicators at different process levels.
- The public page does not disclose test types, samples and comparison methods.