Data and Artificial Intelligence Services
Pubrica provides statistical analysis, bioinformatics, predictive modelling and real-world data analysis for researchers, life-science companies and publishers. Our statisticians and data scientists establish what your data can support, run and validate the analysis, and report it to the standard your destination requires. You receive the code, the log and the working files so your own team can rerun the analysis. Every other Pubrica service group uses this team, which is why it is presented as a separate service group.
- Feasibility assessed first: many clinical datasets cannot support the model requested, and we tell you this at the start
- Validation method stated: internal, temporal or external, because the type of validation determines what the result can claim
- Code, environment and seeds handed over, so the result can be regenerated when a reviewer requests a sensitivity analysis
- Reported to the standard the destination uses: TRIPOD+AI for prediction models and CONSORT-AI where the intervention is an algorithm
- Analysis within your own environment, which is usually the right arrangement for identifiable data because only the output leaves your systems
Research servicesPooling and modelling
Market accessEconomic models and RWE
Medical device & IVDPerformance analysis
Editorial servicesStatistical review
A result decided in advanceNot offered
Data → what it supports → analysis → validation → a defensible claim
- What the data supports, established before modelling
- Validation strategy fixed in advance
- Calibration reported, not only discrimination
- Code, environment and seeds handed over
- Analysis can run inside your own environment
Our Services
Data and AI Services Across Pubrica
You can contact this team directly or through any other Pubrica service you already use. The same team carries out the analysis in either case.
Nine servicesA model that performs well only on its development data has not yet been validated.
Used by almost every other Pubrica teamResults that cannot be regenerated cannot be defended at review.
Pipelines, versions, parametersTool versions affect results, and a methods section that omits them cannot be reproduced.
Before analysis beginsMany analysis problems originate in how the data was collected.
Presenting results clearlyOmitting a confidence band from a chart implies a precision the analysis does not support.
Feeds market accessThis service provides the analysis. The dossiers and economic models it supports are prepared by our market access team.
Samples are illustrative and anonymised. Hover any one of them to play it again.
Compliance and Standards
Reporting Standards for Data Analysis and AI
Health data analysis is assessed against reporting standards as well as model performance.
TRIPOD+AI (for clinical prediction models)
The reporting standard for developing and validating prediction models, including the machine-learning extension. It covers the items reviewers ask for.
CONSORT-AI and SPIRIT-AI (for trials of AI interventions)
Where the intervention is an algorithm, the trial requires its own reporting extension in addition to the core checklist.
FAIR data (for the data behind it)
Data that is findable, accessible, interoperable and reusable, which funders increasingly require and journals ask about at submission.
Quality management (cert. H3CR9TCU20251212INDQM10N)
How the work is run, recorded and released.
Our Process
How Our Data Analysis Process Works
The first step most often determines whether the project should continue, and that is its purpose.
Assess the data
We assess size, events per variable, missingness, leakage and what the record captures. Sometimes the conclusion is that the dataset cannot answer the question.
Pre-specify the analysis
The outcome, predictors, validation strategy and performance measures are written down before the analysis starts.
Analyse and validate
The validation set is held aside from the beginning and the experiment is tracked, so the reported configuration is the one that was run.
Report for verification
Results are reported to the standard your destination uses, with code, environment and seeds handed over.
FAQs
Data and AI Services FAQs
If your question is not answered here, ask an expert — no quote is required.
Every other Pubrica service group uses this team, and keeping the work in one place keeps our methods consistent. A pooled estimate for a systematic review and one for an HTA dossier should be calculated in the same way.
Yes, although a small dataset limits what can be built. It supports a simple model with appropriate internal validation; it does not support a deep learning model with a claimed external performance.
Yes, and for identifiable data this is usually the right approach. The analysis runs under your governance, and only the output leaves your environment.
Sometimes for code and drafting, always with a named specialist accountable for the output. For anything entering a manuscript, disclosure follows the journal’s policy and we tell you what to declare.
Related services
Related Pubrica Service Groups
Most clients reach this team through one of these service groups.
1PUBLISHING IN JOURNALSfive service areas
Publishing in journalsJournal publication support. Research servicesDesign, synthesis and statistics.2SUBMITTING TO REGULATORSsix service areas
Submitting to regulatorsRegulatory documentation. Market access and HEORPayers, HTA and price.3EDITORIAL SUPPORT FOR PUBLISHERSthree service areas
Editorial support for publishersEditorial work for journals.Request a Free Assessment of Your Dataset
Our first assessment sometimes shows that the data supports a narrower claim than intended, and it is more economical to establish this before the analysis begins.