hyperion.tables

Overview

Building a parameter table for a NONMEM model by hand means copying each estimate and standard error from the run output, and every value copied is a chance for error. You also have to remember which parameters were estimated on the log scale, back-transform them, compute CV% and confidence intervals, and arrange it all for the report. With hyperion.tables, you describe the table and the model supplies the parameters.

A TableSpec is that description: the title, sections and columns, which values are back-transformed, how confidence intervals appear, and how many significant figures to keep. Each setting is its own step in a pipe. Reuse one spec, and every parameter table in the project lays out the same way, including comparison tables that put several models side by side with the percent change in each parameter. Tables render through gt or flextable to Word or an image, or come back as formatted data for any other table tool. The parameters are already in the model. Estimates and standard errors are read from the run. Each parameter's name, unit, and transform come from its comment in the control stream, and a lookup file fills any gaps. From these, hyperion.tables back-transforms the estimates and computes CV%, RSE and confidence intervals. Summary tables follow each run in the project back to the model it was based on and report its OFV and, where the runs are comparable, its change in OFV and likelihood-ratio p-value against that parent. You decide how the table is laid out. What goes in it comes from the model, not from copying by hand.