Resources

Only aggregate summary statistics — gradient vectors and Hessian matrices — are transmitted. No individual patient records, no row-level data. Each site retains full control of its patient data at all times.

PDA-OTA converges in a single communication round. The one-shot approach transmits local gradients and Hessians once, allowing the coordinating center to compute a pooled-equivalent estimate immediately.

Unlike traditional meta-analysis which aggregates study-level estimates with inherent information loss, PDA aggregates patient-level sufficient statistics. This yields estimates asymptotically equivalent to pooling all raw data — a fundamentally more powerful approach.

PDA currently supports linear regression (ODAL), logistic regression (ODAL), Cox proportional hazards (ODAC), Poisson regression (ODAP), self-controlled case series (SCCS), and quantile regression. Custom models can be implemented via the extensible API.

IRB requirements vary by institution and study design. We recommend consulting with your local IRB. In many cases, the non-sharing of patient data simplifies the approval process compared to traditional multi-site data sharing agreements.

The coordinating center runs the aggregation step using the PDA R package. Detailed setup instructions are available in our implementation guideline. The PDA server can also be deployed via the hosted PDA-OTA platform for streamlined coordination.