Generating Data Tables

Creating Tables in Analysis Chat

The Analysis Chat can generate structured data tables by reading your included studies and extracting information into organized rows and columns. Tables appear as interactive cards in the chat, with each cell optionally linked back to its source in the original PDF.

Types of Tables

Characteristics of Included Studies

The most common extraction table. Summarizes key details from each study:

  • Author, year, country
  • Study design (RCT, cohort, etc.)
  • Sample size and demographics
  • Intervention and control details
  • Primary and secondary outcomes
  • Key findings

Example prompt: 'Create a characteristics table including author, year, study design, sample size, population, intervention, comparator, and primary outcomes for all included studies.'

Outcome Data Tables

Tables focused on extracting specific outcome data for meta-analysis:

  • Means, standard deviations, and sample sizes per group
  • Event counts for binary outcomes
  • Effect sizes and confidence intervals
  • Timepoints and measurement tools used

Example prompt: 'Extract the mean, SD, and sample size for depression scores (any validated scale) at post-treatment for both the intervention and control groups from each study.'

Risk of Bias / Quality Assessment

Tables assessing methodological quality:

  • Cochrane Risk of Bias domains (randomization, blinding, attrition, etc.)
  • GRADE certainty ratings
  • Newcastle-Ottawa Scale ratings for observational studies
  • Any custom quality criteria

Example prompt: 'Assess risk of bias for each RCT using the Cochrane RoB 2 tool domains. Create a summary table with ratings and supporting quotes.'

Custom Tables

You can request any table structure. Simply describe the columns you want and the AI will extract the data:

  • 'Create a table comparing adverse events across all studies'
  • 'Extract the inclusion criteria, exclusion criteria, and recruitment method from each study'
  • 'Build a table showing follow-up timepoints and attrition rates'
  • 'Summarize the funding sources and conflict of interest declarations'

Working with Generated Tables

Source Citations

Each cell in a generated table can include a source citation. Hover over a cell to see a link icon that opens the original study where the data was found. This makes it easy to verify extracted data without manually searching PDFs.

Refining Tables

If the initial table needs adjustments, simply ask in the chat:

  • 'Add a column for follow-up duration'
  • 'Remove the funding source column'
  • 'Split the outcomes column into primary and secondary'
  • 'Re-extract the sample sizes — some look incorrect'
  • 'Add the studies published after 2020 that were missed'

Exporting Tables

Each table card includes export options:

  • Excel (.xlsx) — formatted and ready for further analysis
  • CSV (.csv) — for statistical software or scripts
  • Copy to clipboard — paste directly into documents
  • Add to draft — send the table into your review draft editor

Tips for Better Tables

  • Be explicit about which columns you want — the more specific your prompt, the better the result
  • Request one table at a time for best results
  • If a table is too wide, ask the AI to split it into multiple focused tables
  • For outcome data, specify the exact outcome, timepoint, and statistical measure you need
  • Always verify numerical data against the source PDFs before using in meta-analysis
  • Use follow-up messages to iteratively improve the table rather than starting over

Handling Missing Data

When a study doesn't report a requested data point, the AI will indicate this in the table (e.g., 'NR' for not reported). You can follow up by asking:

  • 'Can you calculate the missing SD from the reported confidence interval in [study]?'
  • 'Check if [study] reports this data in a supplementary table'
  • 'What alternative measures are available in the studies with missing data?'

Tables generated in the Analysis Chat are flexible and iterative. Unlike traditional extraction forms that must be configured upfront, you can request any table at any time and refine it through conversation. This makes it easy to adapt your extraction as your understanding of the data evolves.