Predict log hazard ratio

Predict the true log hazard ratio in a new single-arm study by adjusting for additional bias and variability caused by the non-randomized design. This function first draws from the distribution of the true log hazard ratio comparing the treatment to the external control (trt_ec). It then draws the true log hazard ratio comparing the internal control to the external control (ic_ec) from its predictive distribution. Lastly, trt_ec and ic_ec are combined to produce simulated draws of the true log hazard ratio in a comparison of the treatment to the internal control (trt_ic).

# S3 method for ecmeta_jags
predict(
  object,
  newdata,
  n_burnin = 0,
  thin = 1,
  n_adapt = 1000,
  quiet = FALSE,
  ...
)

# S3 method for ecmeta_ml
predict(object, newdata, n_sims = 1000, ...)

Arguments

object	An object of the appropriate class.
newdata	A loghr_data object that stores estimate of the log hazard ratio for a comparison of the treatment arm to the external control arm in the new single-arm study.
n_burnin	Number of
thin	Thinning interval for monitors. Passed to rjags::coda.samples().
n_adapt	The number of iterations for adaptation. Passed to rjags::jags.model().
quiet	If TRUE then messages generated during compilation will be suppressed, as well as the progress bar during adaptation. Passed to rjags::jags.model().
...	Currently unused.
n_sims	Number of simulations to use. Only relevant when using a maximum likelihood based approach.

Value

A list that may contain the following elements:

loghr

A matrix with three columns containing draws of the true log hazard ratios. The columns are: trt_ic (a comparison of the treatment to the internal control), trt_ec (a comparison of the treatment to the external control) and ic_ec (a comparison of the internal control to the external control).

mcmc

When using a Bayesian approach, a coda::mcmc.list object is also included that contains posterior samples of the true log hazard ratio comparing the treatment to the external control. This is not included when using a maximum likelihood approach.

Details

The implementation differs slightly between the Bayesian and maximum likelihood approaches. First, when using a Bayesian approach, MCMC is use to sample trt_ec whereas trt_ec is sampled from a normal distribution based on the log hazard ratio estimates and standard errors in the new study. In practice, these approaches will produce very similar results.

Second, when using the Bayesian approach, the posterior predictive distribution of ic_ec is drawn using the posterior samples of mu and sigma stored in object. In the maximum likelihood approach, ic_ec is simulated by using the point estimates of mu and sigma and drawing from a t distribution, which is the predictive distribution for a future observation.

See also

See vignette("methodology") for a description of the method.