Table 3 Parametric bootstrap algorithm

Step 1

Denote by n₁ and n₂ the number of dogs in the control and treated groups in a study

Step 2

Use Information criteria to identify a statistical model for control data. The possible choices include negative binomial model, Poisson model, and their variants, which take into account unobserved variability using latent effect models.

Step 3

Using the fitted model obtain an estimate of the establishment rate and an estimate of the mean number of worms in the treated group.

Step 4

Using the above fitted model, and including the estimate of the establishment rate, and the mean number of worms remaining in the treated group, simulate the number of worms “available for treatment” and the number of worms remaining after treatment for each dog in the study.

Step 5

Using the result from Step 4, estimate the efficacy using the formula (1).

Step 6

Repeat Step 3 through Step 5 M times to obtain efficacies from M studies. This is typically done to yield results from 1000 or more studies.

Step 7

Order the efficacies from M studies from smallest to largest and obtain the confidence interval by taking the empirical quantile of levels α/2 and (1-α/2) as the lower and upper end-point of the confidence interval. Typical choices for α are 0.95 or 0.90