bayprior Cheat Sheet

✎ Prior Elicitation -- Fit parametric distributions from expert beliefs

elicit_beta(mean, sd, method, ...)

Fit a Beta prior for proportions / response rates.

meansdmethod = "moments" | "quantile"labelexpert_id

p <- elicit_beta(mean=0.30, sd=0.10, method="moments", label="ORR")

Expert belief
mean=0.30
sd=0.10

→

Beta(alpha,beta)
bayprior object

elicit_normal(mean, sd, method, ...)

Normal prior for unbounded quantities (log OR, mean differences).

p <- elicit_normal(mean=0, sd=0.5, method="moments", label="Log OR")

elicit_gamma(mean, sd, method, ...)

Gamma prior for positive quantities (rates, variances).

p <- elicit_gamma(mean=5, sd=2, method="moments", label="Event rate")

elicit_lognormal(mean, sd, method, ...)

Log-Normal prior for hazard ratios and PK parameters.

p <- elicit_lognormal( quantiles = c("0.05"=0.4,"0.50"=0.7,"0.95"=1.2))

elicit_roulette(chips, breaks, family, ...)

SHELF roulette method: expert places chips across histogram bins.

chipsbreaksfamily

chips <- c(0L,1L,3L,7L,9L,7L,4L,2L,1L,1L) breaks <- seq(0, 1, by=0.1) p <- elicit_roulette(chips, breaks, "beta")

elicit_mixture(priors, weights, ...)

Build an explicit mixture prior from component bayprior objects.

m <- elicit_mixture(list(p1, p2), weights = c(0.6, 0.4))

plot(prior)

Density plot with 95% CrI shading and mean dashed line.

print(prior)

Summary: distribution, parameters, mean, SD, 95% CrI.

▲ Expert Pooling -- Aggregate beliefs from multiple experts

aggregate_experts(priors, weights, method)

Linear or logarithmic opinion pooling with Bhattacharyya agreement diagnostics.

priors named listweights must sum to 1method = "linear" | "logarithmic"

e1 <- elicit_beta(0.25, 0.08, "moments", expert_id="E1") e2 <- elicit_beta(0.35, 0.10, "moments", expert_id="E2") con <- aggregate_experts( priors = list(E1=e1, E2=e2), weights = c(0.6, 0.4), method = "linear")

E1: Beta(a1,b1)
w=0.60

⇒

E2: Beta(a2,b2)
w=0.40

→

Mixture
consensus

Bhattacharyya coefficient in con$aggregation$disagreement: 1 = identical experts, 0 = no overlap. Values <0.5 warrant review.

◆ Conflict Diagnostics -- Detect prior-data conflict

prior_conflict(prior, data_summary, alpha)

Box p-value, surprise index, KL divergence, Bhattacharyya overlap.

Binary: type="binary"xn
Continuous: type="continuous"xsdn

cd <- prior_conflict( prior = p, data_summary = list(type="binary", x=18, n=40), alpha = 0.05)

Severity classification:

NONE Box p > 0.10 MILD p < 0.10 SEVERE p < alpha

Field	Description
box_pvalue	Prior predictive p-value (Box 1980)
surprise_index	Standardised distance from prior mean
kl_prior_likelihood	KL divergence prior \|\| likelihood
overlap	Bhattacharyya overlap coefficient
conflict_severity	"none" / "mild" / "severe"
recommendation	Automatic actionable guidance

plot_prior_likelihood(prior, data_summary, show_posterior)

Prior / likelihood / posterior density overlay plot.

gp <- plot_prior_likelihood(p, data_obs, show_posterior = TRUE)

conflict_mahalanobis(prior_means, prior_cov, obs_means, obs_cov, labels)

Multivariate omnibus test for co-primary endpoints. Returns chi-squared p-value and per-parameter z-scores.

mv <- conflict_mahalanobis( prior_means = c(0.35, 0.60), prior_cov = matrix(c(0.01, 0.003, 0.003, 0.015), 2, 2), obs_means = c(0.55, 0.58), obs_cov = matrix(c(0.008,0.002,0.002,0.010),2,2)/50, labels = c("ORR", "OS"))

■ Sensitivity Analysis -- Posterior robustness to hyperparameter choice

sensitivity_grid(prior, data_summary, param_grid, target, threshold)

Evaluate posterior quantities over a hyperparameter grid.

target: "posterior_mean" "posterior_sd" "prob_efficacy"

sa <- sensitivity_grid( prior = p, data_summary = list(type="binary", x=14, n=40), param_grid = list(alpha=seq(1,8,0.5), beta=seq(2,20,1)), target = c("posterior_mean","prob_efficacy"), threshold = 0.30)

sensitivity_cri(prior, data_summary, param_grid, cri_level)

Track credible interval width and bounds across the grid. Key for regulatory submissions.

cri <- sensitivity_cri(p, data_obs, param_grid = list(alpha=seq(1,8,0.5), beta=seq(2,20,1)), cri_level = 0.95)

Grid
alpha x beta

→

cri_lower
cri_upper
cri_width

plot_tornado(sa)

Bar chart: influence range ordered from largest to smallest.

plot_sensitivity(sa, target)

Two-dimensional heatmap of posterior quantity over the parameter grid.

plot_sensitivity(sa, target="posterior_mean") plot_sensitivity(cri, target="cri_width")

▲ Robust Priors -- Sceptical, mixture, and power priors

robust_prior(prior, vague_weight, ...)

Schmidli et al. (2014) robust mixture: informative + vague Normal component.

priorvague_weight 0-1, e.g. 0.20 = 20% vague

rob <- robust_prior(p, vague_weight=0.20) plot(rob) # density overlay

Informative
Beta(a,b) 80%

Vague
Normal 20%

→

Robust
mixture

sceptical_prior(null_value, family, strength)

Spiegelhalter & Freedman (1994) prior centred at the null. Calibrated to weak / moderate / strong scepticism.

family: "normal" | "beta" | "lognormal"
strength: "weak" | "moderate" | "strong"

sc <- sceptical_prior( null_value = 0.20, family = "beta", strength = "moderate")

For Beta priors, null_value must be in (0, 1).

calibrate_power_prior(historical_data, current_data, base_prior, target_bf, delta_grid, method)

Ibrahim & Chen (2000) power prior: down-weights historical data by delta, calibrated to a target Bayes Factor.

method: "bayes_factor" | "compatibility"

calib <- calibrate_power_prior( historical_data = list(type="binary", x=12, n=40), current_data = list(type="binary", x=18, n=50), base_prior = elicit_beta(0.5,0.2,"moments"), target_bf = 3, method = "bayes_factor") plot(calib) # calibration curves + optimal delta

■ Regulatory Reporting -- Quarto HTML / PDF / Word reports

prior_report(prior, conflict, sensitivity, ...)

Render a self-contained regulatory justification report via Quarto.

prior_report( prior = p, conflict = cd, sensitivity = sa, robust_prior = rob, # optional sceptical_prior = sc, # optional output_format = "pdf", # "html" | "pdf" | "docx" output_file = "prior_justification", trial_name = "TRIAL-001", sponsor = "BioPharma Ltd", author = "J. Smith, Biostatistician", notes = "Prior pre-specified in SAP v2.1.")

Requires devtools::install() -- not load_all(). Quarto spawns a fresh R session.

Report sections: Executive summary • Trial information • Prior specification (density + parameters) • Conflict diagnostics (Box p-value, overlay) • Sensitivity (tornado, heatmap) • Robust/sceptical/power priors • FDA/EMA compliance checklist • Session info

▶ Typical Workflow

Elicit prior: elicit_beta() / elicit_roulette()
Pool experts: aggregate_experts()
Check conflict: prior_conflict()
Assess sensitivity: sensitivity_grid() or sensitivity_cri()
Build robust prior: robust_prior() / sceptical_prior()
Generate report: prior_report(..., output_format="pdf")

run_app()

Launch the full Shiny app with dark/light mode toggle. All panels feed directly into the downloaded report.

⇓ Install & Docs

pak::pak("ndohpenngit/bayprior") # or devtools::install_github("ndohpenngit/bayprior", build_vignettes = TRUE)

Docs: ndohpenngit.github.io/bayprior
Vignettes: browseVignettes("bayprior")
Citation: citation("bayprior")
Issues: github.com/ndohpenngit/bayprior/issues