Study Designs

A bioequivalence design is characterized by its subjects (each assigned to a sequence), its formulations (reference and one or more tests), its periods (the dosing occasions), and its sequences (which formulation is given at each period). The design is detected automatically from the sequence column; see detect_design and study_paradigm.

Supported designs

A design is replicated when at least one formulation is administered more than once to a subject; this is what allows the within-subject variability of a formulation to be estimated (see Within-subject variability). Replication is fully replicated (every formulation repeated), partially replicated (some repeated), or non-replicated. In the supported partial-replicate designs only the reference is replicated, so only its within-subject variability σwᵣ is estimable.

• EMA ABEL is available on any replicated design (partial or full).
• FDA RSABE requires either the partially-replicated RRT|RTR|TRR design or one of the four-period fully-replicated designs RTRT|TRTR, RTTR|TRRT, or RRTT|TTRR. FDA RSABE for NTID additionally requires a fully replicated design (so that the test within-subject variability can also be estimated).

Paradigms

A run of pumas_be (or run_be), not a dataset on its own, is classified into one paradigm, which fixes the family of models available (see Models) and the criteria that can be assessed. The paradigm does not by itself pin down a single model: on the crossover paradigms custom_model (with homogeneity) selects between the fixed-effects and mixed-effects fits within the paradigm, whereas on the parallel and paired paradigms the model is fixed and custom_model is rejected. The nonparametric paradigm picks its rank procedure from the design, and accepts custom_model = SignedRankTest to pool the within-subject differences instead.

The paradigm is determined by the design together with options such as nonparametric; the chosen criterion does not change the paradigm but sets the inference options (including, for some criteria, custom_model) and drives the assessment within it. For example, nonparametric = true selects the nonparametric paradigm whatever the design (and the endpoint name can trigger it by default, e.g. Tmax), though not every design is then accepted; see Nonparametric paradigm.

The first column is the value study_paradigm returns, shown as it prints at the REPL. These names are not exported, so they are what you see rather than something you type: obtain the value by calling study_paradigm, not by referring to the name directly.