Repeated-effects network meta-analysis from arm-level continuous data

Fits a treatment-design network meta-analysis to arm-level continuous data while constructing multi-arm, outcome, and repeated-time sampling covariance internally. This function is independent of network_meta() and does not alter its contrast-level interface.

Usage

network_meta_repeated(
  data,
  study_id,
  treatment,
  mean,
  sd,
  n,
  outcome = NULL,
  time = NULL,
  reference = NULL,
  treatment_effects = c("common", "by_outcome", "by_time", "by_outcome_time"),
  outcome_correlation = c("independence", "exchangeable"),
  time_correlation = c("independence", "exchangeable", "ar1"),
  rho_outcome = 0,
  rho_time = 0,
  heterogeneity = c("none", "common"),
  estimation_method = c("REML", "ML"),
  ci_level = 0.95
)

Arguments

data

Data frame with one row per study, treatment arm, outcome, and optional time combination.

study_id, treatment, mean, sd, n

Character strings naming study, treatment, arm mean, arm standard deviation, and arm sample-size columns.

outcome, time

Optional character strings identifying outcomes and follow-up occasions. Omit either when it is not present.

reference

Optional reference treatment. Defaults to the first sorted treatment.

treatment_effects

Whether treatment effects are common, outcome specific, time specific, or outcome-by-time specific.

outcome_correlation, time_correlation

Correlation structures used to construct covariance among means from the same arm. "independence" and "exchangeable" are available for outcomes; "independence", "exchangeable", and "ar1" are available for time.

rho_outcome, rho_time

Correlations for exchangeable or AR(1) structures. Values must be between minus one and one.

heterogeneity

Either "none" for generalized least squares or "common" for a common contrast-level random-effects variance estimated by ML or REML.

estimation_method

Either "REML" or "ML" when common heterogeneity is estimated.

ci_level

Confidence level for coefficient intervals.

Value

An object of class nma_repeated_mars containing the derived contrast data, internally constructed study covariance blocks, coefficient summary, fitted covariance, and heterogeneity estimate.

Details

The function currently supports arm-level continuous mean differences. Multi-arm covariance is derived exactly under independent arms. Outcome and time covariance are derived from the chosen correlation assumptions; these assumptions are retained in the fitted object. For time correlation "ar1", equally spaced ordered occasions are assumed unless time is numeric, in which case correlation is rho_time raised to the absolute time distance. This is a treatment-design GLS/REML model, not a wrapper around network_meta().

Examples

dat <- data.frame(
  study = rep(c("S1", "S2", "S3"), each = 2),
  trt = rep(c("A", "B"), 3), mean = c(10, 12, 9, 11, 11, 14),
  sd = 2, n = 30
)
network_meta_repeated(dat, "study", "trt", "mean", "sd", "n")
#> $call
#> network_meta_repeated(data = dat, study_id = "study", treatment = "trt", 
#>     mean = "mean", sd = "sd", n = "n")
#> 
#> $reference
#> [1] "A"
#> 
#> $treatments
#> [1] "A" "B"
#> 
#> $treatment_effects
#> [1] "common"
#> 
#> $contrast_data
#>    study_id treatment_1 treatment_2 outcome_id time_id effect  variance
#> S1       S1           A           B    overall overall      2 0.2666667
#> S2       S2           A           B    overall overall      2 0.2666667
#> S3       S3           A           B    overall overall      3 0.2666667
#> 
#> $sampling_covariance
#> $sampling_covariance$S1
#>           [,1]
#> [1,] 0.2666667
#> 
#> $sampling_covariance$S2
#>           [,1]
#> [1,] 0.2666667
#> 
#> $sampling_covariance$S3
#>           [,1]
#> [1,] 0.2666667
#> 
#> 
#> $design_matrix
#>      B::common
#> [1,]         1
#> [2,]         1
#> [3,]         1
#> 
#> $coefficients
#>                term estimate std_error ci_lower ci_upper
#> B::common B::common 2.333333 0.2981424 1.748985 2.917682
#> 
#> $varcov
#>            B::common
#> B::common 0.08888889
#> 
#> $fitted_values
#> [1] 2.333333 2.333333 2.333333
#> 
#> $residuals
#> [1] -0.3333333 -0.3333333  0.6666667
#> 
#> $tau2
#> [1] 0
#> 
#> $heterogeneity
#> [1] "none"
#> 
#> $estimation_method
#> [1] "REML"
#> 
#> $assumptions
#> $assumptions$outcome_correlation
#> [1] "independence"
#> 
#> $assumptions$time_correlation
#> [1] "independence"
#> 
#> $assumptions$rho_outcome
#> [1] 0
#> 
#> $assumptions$rho_time
#> [1] 0
#> 
#> $assumptions$arms_independent
#> [1] TRUE
#> 
#> 
#> attr(,"class")
#> [1] "nma_repeated_mars"