Objective

Enabling the use of distributed statistics across federated data models

Results

Using real world patient haematology data we were able to demonstrate identical results from the two analyses performed using the pooled dataset versus distributed dataset(s) as shown in Tables 1 and 2, respectively. The hazard ratios, 95% confidence intervals and p-values are identical between the two models for all for levels of stage compared to reference, adjusted for age and gender indicating that the distributed model has generated precisely the same proportional hazard estimate as well as variance estimates for variables within the model.

Read the full abstract