Calculates ranking probabilities and SUCRA (Surface Under Cumulative Ranking curve) or P-scores for treatments in network meta-analysis.

Usage

calculate_sucra(nma_result, lower_better = NULL, n_sim = 1000, seed = 42)

Arguments

nma_result: Result from network_meta()
lower_better: Logical. Is lower estimate better? Default: TRUE for ratios
n_sim: Integer. Number of simulations for ranking. Default: 1000
seed: Integer or NULL. Random seed for reproducibility. Default: 42. Set to NULL for non-deterministic results.

Value

List with rankings, SUCRA/P-scores, and rankogram data. When n_treatments == 1, the returned sucra value is set to 1.0 (the single treatment is trivially best).

Details

The reference treatment is treated as the baseline in the output and is assigned an SE of 0. For any non-reference treatments with missing/invalid standard errors, the SE is imputed using the median of valid SE values. This imputation can affect ranking probabilities and SUCRA values.

Note

RNG state is isolated via withr::with_seed during simulation, so calling this function does not alter the global RNG state.

Examples

# Calculate SUCRA rankings
nma_data <- data.frame(
  study = c("S1", "S2", "S3"),
  treat1 = c("A", "B", "A"),
  treat2 = c("B", "C", "C"),
  effect = log(c(0.75, 0.90, 0.80)),
  se = c(0.12, 0.15, 0.18)
)
nma_result <- network_meta(nma_data, effect_measure = "hr")
sucra <- calculate_sucra(nma_result)
sucra$ranking
#>   treatment mean_rank sucra prob_best prob_worst final_rank
#> B         B     1.432 78.40     0.585      0.017          1
#> C         C     1.717 64.15     0.413      0.130          2
#> A         A     2.851  7.45     0.002      0.853          3
sucra$interpretation
#> [1] "Treatment ranking by lower is better (SUCRA, %). Best: B (78.4%), Worst: A (7.4%)"