Package index • cpaic

Network setup

Build a (possibly disconnected) component network and code components.

cpaic_network(): Set up a (possibly disconnected) component network for cpaic
build_C_matrix(): Build a component-coded treatment-by-component matrix

Connectivity and estimability

Detect disconnection, and check which relative effects the component design can actually identify. Reconnecting a network does not guarantee that the effects you want are estimable, and both engines will otherwise return a confident-looking number for a contrast that carries no information.

cpaic_connectivity(): Assess connectivity and component-bridge identifiability of a network
estimable_effects(): Which relative effects of a component network are uniquely estimable?
estimable_effects_at(): Which population-adjusted contrasts are estimable at a target population?

Connection layer

Reconnect a disconnected network through shared components.

cnma_bridge(): Reconnect a network through its additive component structure
additivity_test(): Fit statistics for the additive component model

Population adjustment

Anchored, population-adjusted indirect comparison across the network.

cmaic(): Component matching-adjusted indirect comparison (cMAIC)
cstc(): Component simulated treatment comparison (cSTC)
cmlnmr(): Component-additive multilevel network meta-regression (ML-NMR)
effective_sample_size(): Effective sample sizes from a cMAIC fit
edge_influence(): Does the individual patient data actually inform this contrast?

Hierarchies

Rank treatments or components IN A TARGET POPULATION. Because component effects are population-specific under population adjustment, so are the rankings: a component can lead in one population and trail in another.

cpaic_ranks(): Population-adjusted treatment and component hierarchies
rank_curve(): How a hierarchy changes across target populations
rank_probs(): Posterior rank probabilities in a target population

Reporting

relative_effects(): Relative treatment effects from a cpaic fit
league_table(): League table of all pairwise relative effects
component_effects(): Component effects from a cpaic fit

Plots

Every plot returns a ggplot object, so it can be modified with the usual ggplot2 verbs. The network, forest, rankogram, deviance, prior-posterior, integration-error, MCMC, and survival plots are ported from multinma (Phillippo et al. 2020). The rank curve, the estimability map, and the edge-influence plot are specific to cpaic: under population adjustment the hierarchy, and the estimable set itself, are functions of the target population.

plot(<cpaic_network>): Plot the component network
forest() plot(<cpaic_effects>) plot(<cpaic_bridge>) plot(<cpaic_fit>): Forest plot of relative or component effects
plot_rank_curve(): How the hierarchy changes across target populations
plot_estimability(): Map which contrasts are estimable, and on what evidence, across populations
plot_edge_influence(): Plot how much each edge informs a chosen contrast
plot(<cpaic_rank_probs>): Rankogram and cumulative rank plot
plot(<cpaic_ranks>): Plot a population-adjusted hierarchy
plot(<cpaic_dic>): Deviance and dev-dev plots
plot_leverage(): Leverage plot
plot_prior_posterior(): Prior versus posterior
plot_integration_error(): Numerical integration error against the number of integration points
plot(<cpaic_mlnmr>): MCMC diagnostics for a cML-NMR fit
plot_survival(): Fitted survival curves from a cML-NMR fit
geom_km(): Kaplan-Meier curves from the survival data behind a cML-NMR fit

Bayesian diagnostics

dic(): Deviance information criterion
loo(<cpaic_mlnmr>): Pareto-smoothed importance sampling leave-one-out cross-validation
waic(<cpaic_mlnmr>): Widely applicable information criterion
prior_sensitivity(): Refit cML-NMR under tighter and looser priors
prior_predictive_check(): Summarize a prior-predictive cML-NMR fit

Data

cpaic_bin_agd: Example disconnected component network: aggregate contrasts
cpaic_bin_ipd: Example disconnected component network: individual patient data