Getting Started with outstandR: A Practical Demonstration
Source:vignettes/practical_demonstration.Rmd
practical_demonstration.RmdIntroduction
The outstandR package provides a unified framework for Population-Adjusted Indirect Comparisons (PAIC). It supports Matching-Adjusted Indirect Comparison (MAIC), G-computation via Maximum Likelihood Estimation (ML) or Bayesian MCMC inference, and Multiple Imputation Marginalization (MIM).
This vignette walks through a practical demonstration of how to load data, specify strategies, run analyses, and compare the results across different outcome types: binary, continuous, and count.
1. Binary Outcome Data
We load the synthetic binary outcome data and continuous covariates.
For regression-based adjustments (G-computation and MIM), we define the formula relating covariates, treatments, and the outcome. For weight-based adjustments (MAIC), we specify a balancing formula of covariates to match.
# Outcome model formula (for G-computation and MIM)
lin_form_bin <- as.formula("y ~ PF_cont_1 + PF_cont_2 + trt + trt:EM_cont_1 + trt:EM_cont_2")
# Balancing model formula (for MAIC)
bal_form_bin <- as.formula("~ EM_cont_1 + EM_cont_2")Running the Analysis
We can run all four strategy classes. For Bayesian methods, we specify fewer iterations and a single chain for speed in this demonstration.
# 1. Matching-Adjusted Indirect Comparison (MAIC)
outstandR_maic_bin <- outstandR(
ipd_trial = AC_IPD_binY_contX,
ald_trial = BC_ALD_binY_contX,
strategy = strategy_maic(
formula = list(outcome_model = formula("y ~ trt"),
balance_model = bal_form_bin),
family = binomial(link = "logit")),
seed = 12345
)
# 2. Maximum Likelihood G-computation (ML G-comp)
outstandR_gcomp_ml_bin <- outstandR(
ipd_trial = AC_IPD_binY_contX,
ald_trial = BC_ALD_binY_contX,
strategy = strategy_gcomp_ml(
formula = list(outcome_model = lin_form_bin),
family = binomial(link = "logit")),
seed = 12345
)
# 3. Bayesian G-computation
outstandR_gcomp_bayes_bin <- outstandR(
ipd_trial = AC_IPD_binY_contX,
ald_trial = BC_ALD_binY_contX,
strategy = strategy_gcomp_bayes(
formula = list(outcome_model = lin_form_bin),
family = binomial(link = "logit")),
seed = 12345,
iter = 250, # Small iterations for speed
chains = 1, # Single chain for demonstration
refresh = 0
)
# 4. Multiple Imputation Marginalization (MIM)
outstandR_mim_bin <- outstandR(
ipd_trial = AC_IPD_binY_contX,
ald_trial = BC_ALD_binY_contX,
strategy = strategy_mim(
formula = list(outcome_model = lin_form_bin),
family = binomial(link = "logit")),
seed = 12345,
iter = 250,
chains = 1,
refresh = 0
)2. Continuous Outcome Data
For continuous data, we load the corresponding datasets and utilize the gaussian(link = "identity") family.
data(AC_IPD_contY_mixedX)
data(BC_ALD_contY_mixedX)
# Define models
lin_form_cont <- as.formula("y ~ X1 + X2 + X3 + trt + trt:(X1 + X2 + X4)")
bal_form_cont <- as.formula("~ X1 + X2 + X4")Running the Analysis
We fit the strategies similarly, changing the outcome family to gaussian.
# 1. MAIC
outstandR_maic_cont <- outstandR(
ipd_trial = AC_IPD_contY_mixedX,
ald_trial = BC_ALD_contY_mixedX,
strategy = strategy_maic(
formula = list(outcome_model = formula("y ~ trt"),
balance_model = bal_form_cont),
family = gaussian(link = "identity")),
seed = 12345
)
# 2. ML G-comp
outstandR_gcomp_ml_cont <- outstandR(
ipd_trial = AC_IPD_contY_mixedX,
ald_trial = BC_ALD_contY_mixedX,
strategy = strategy_gcomp_ml(
formula = list(outcome_model = lin_form_cont),
family = gaussian(link = "identity")),
seed = 12345
)
# 3. Bayesian G-comp
outstandR_gcomp_bayes_cont <- outstandR(
ipd_trial = AC_IPD_contY_mixedX,
ald_trial = BC_ALD_contY_mixedX,
strategy = strategy_gcomp_bayes(
formula = list(outcome_model = lin_form_cont),
family = gaussian(link = "identity")),
seed = 12345,
iter = 250,
chains = 1,
refresh = 0
)
# 4. MIM
outstandR_mim_cont <- outstandR(
ipd_trial = AC_IPD_contY_mixedX,
ald_trial = BC_ALD_contY_mixedX,
strategy = strategy_mim(
formula = list(outcome_model = lin_form_cont),
family = gaussian(link = "identity")),
seed = 12345,
iter = 250,
chains = 1,
refresh = 0
)3. Count Outcome Data
For count data, we use the poisson(link = "log") family with Poisson-distributed outcomes.
data(AC_IPD_countY_contX)
data(BC_ALD_countY_contX)
# Define models
lin_form_count <- as.formula("y ~ PF_cont_1 + PF_cont_2 + trt + trt:EM_cont_1 + trt:EM_cont_2")
bal_form_count <- as.formula("~ EM_cont_1 + EM_cont_2")Running the Analysis
# 1. MAIC
outstandR_maic_count <- outstandR(
ipd_trial = AC_IPD_countY_contX,
ald_trial = BC_ALD_countY_contX,
strategy = strategy_maic(
formula = list(outcome_model = formula("y ~ trt"),
balance_model = bal_form_count),
family = poisson(link = "log")),
seed = 12345
)
# 2. ML G-comp
outstandR_gcomp_ml_count <- outstandR(
ipd_trial = AC_IPD_countY_contX,
ald_trial = BC_ALD_countY_contX,
strategy = strategy_gcomp_ml(
formula = list(outcome_model = lin_form_count),
family = poisson(link = "log")),
seed = 12345
)
# 3. Bayesian G-comp
outstandR_gcomp_bayes_count <- outstandR(
ipd_trial = AC_IPD_countY_contX,
ald_trial = BC_ALD_countY_contX,
strategy = strategy_gcomp_bayes(
formula = list(outcome_model = lin_form_count),
family = poisson(link = "log")),
seed = 12345,
iter = 250,
chains = 1,
refresh = 0
)
# 4. MIM
outstandR_mim_count <- outstandR(
ipd_trial = AC_IPD_countY_contX,
ald_trial = BC_ALD_countY_contX,
strategy = strategy_mim(
formula = list(outcome_model = lin_form_count),
family = poisson(link = "log")),
seed = 12345,
iter = 250,
chains = 1,
refresh = 0
)4. Comparing and Visualizing Results
We can display the structured outstandR results using standard print methods or inspect individual estimates.
print(outstandR_maic_bin)
#> Object of class 'outstandR'
#> ITC algorithm: MAIC
#> Model: binomial
#> Scale: log_odds
#> Common treatment: C
#> Individual patient data study: A vs C
#> Aggregate level data study: B vs C
#> Confidence interval level: 0.95
#>
#> Contrasts:
#>
#> # A tibble: 3 × 5
#> Treatments Estimate Std.Error lower.0.95 upper.0.95
#> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 AB 0.620 0.254 -0.368 1.61
#> 2 AC -0.824 0.152 -1.59 -0.0607
#> 3 BC -1.44 0.102 -2.07 -0.817
#>
#> Absolute:
#>
#> # A tibble: 3 × 5
#> Treatments Estimate Std.Error lower.0.95 upper.0.95
#> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 A 0.241 0.00208 0.151 0.330
#> 2 B -1.03 0.108 -1.67 -0.384
#> 3 C 0.417 0.00534 0.274 0.561The package provides a built-in plot() method to visualize contrasts and absolute estimates side by side. By passing multiple fitted objects, we can directly compare strategies:
plot(outstandR_maic_bin, outstandR_gcomp_ml_bin, outstandR_gcomp_bayes_bin, outstandR_mim_bin)
5. Alternative Outcome Scales
By default, the package reports effects on the link function’s scale. You can customize the scale of interest using the scale argument.
For instance, we can specify a Risk Difference instead of Log-Odds for the binary outcome analysis:
outstandR_maic_rd <- outstandR(
ipd_trial = AC_IPD_binY_contX,
ald_trial = BC_ALD_binY_contX,
strategy = strategy_maic(
formula = list(outcome_model = formula("y ~ trt"),
balance_model = bal_form_bin),
family = binomial(link = "logit")),
scale = "risk_difference",
seed = 12345
)
print(outstandR_maic_rd)
#> Object of class 'outstandR'
#> ITC algorithm: MAIC
#> Model: binomial
#> Scale: risk_difference
#> Common treatment: C
#> Individual patient data study: A vs C
#> Aggregate level data study: B vs C
#> Confidence interval level: 0.95
#>
#> Contrasts:
#>
#> # A tibble: 3 × 5
#> Treatments Estimate Std.Error lower.0.95 upper.0.95
#> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 AB 0.165 0.443 -1.14 1.47
#> 2 AC -0.177 0.00704 -0.341 -0.0121
#> 3 BC -0.341 0.436 -1.63 0.952
#>
#> Absolute:
#>
#> # A tibble: 3 × 5
#> Treatments Estimate Std.Error lower.0.95 upper.0.95
#> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 A 0.241 0.00208 0.151 0.330
#> 2 B 0.0760 0.441 -1.23 1.38
#> 3 C 0.417 0.00534 0.274 0.561