
Direct versus indirect network incoherence
network_incoherence.RdIntended question: do direct and independent indirect estimates disagree for
a treatment comparison? This interface uses network_nodesplit().
Usage
network_incoherence(
object,
method = c("conditional", "ML", "wald"),
adjust = c("none", "holm", "bonferroni"),
comparisons = NULL,
level = 0.95
)
# S3 method for class 'nma_incoherence'
print(
x,
...,
estimable_only = TRUE,
row.names = FALSE,
digits = getOption("digits"),
detailed = FALSE
)
# S3 method for class 'nma_incoherence'
summary(
object,
...,
digits = getOption("digits"),
estimable_only = TRUE,
detailed = FALSE
)Arguments
- object
A fitted joint network model. For the summary method, the data frame returned by
network_incoherence().- method
conditional(default) holds fitted heterogeneity fixed;MLestimates heterogeneity jointly in each split model. Both use normal Wald inference for direct-minus-indirect differences. Thewaldvalue warns and callsnetwork_design_incoherence()for design-level tests.- adjust
none,holm, orbonferronifor the requested node-split pair/stratum family, retaining nonestimable hypotheses in its size. Output addsp_adjustedand anadjustmentattribute. Intervals are pointwise.- comparisons
Optional data frame with
treatment_1andtreatment_2columns selecting unordered pairs. Selection includes all fitted mean strata; reversing a requested pair does not reverse the returned contrast direction.- level
Coverage of pointwise normal-approximation intervals for direct effects, indirect effects and their difference.
- x
The data frame returned by
network_incoherence().- ...
Additional print arguments forwarded to
print.data.frame(); reserved additional arguments for summary.- estimable_only
Logical; print only estimable rows (default
TRUE). Set toFALSEto print all requested rows and nonestimability reasons.- row.names
Logical; print row names. Defaults to
FALSE.- digits
Number of decimal places to display, an integer from 1 to 22; defaults to
getOption("digits"). P-values below 0.0001 print as<0.0001; other p-values use at least four decimal places. Full numerical precision is retained in the result.- detailed
Logical; show standard errors, intervals and all diagnostic fields rather than the compact default (
FALSE).
Value
A data frame of class nma_incoherence, with the same columns and
attributes as network_nodesplit(): direct and indirect estimates, SEs and
intervals; their difference, SE, interval, raw and adjusted p-values; study
counts, estimability, reason and method. There is no global p-value.
Its print method invisibly returns the full input object. A stratum_cols
attribute records outcome/time identifiers when present so numeric labels
are preserved during printing.
Explicit method="wald" instead returns the design-test list with
a warning; new code should call network_design_incoherence() directly.
Details
Whole studies supplying the target direct contrast are excluded from indirect
evidence, including their other multi-arm contrasts. A multi-arm study cannot
supply its own independent indirect evidence. The contrast direction is
treatment_2 minus treatment_1; difference is direct minus indirect.
The printout states the intended question: do direct and independent indirect
estimates disagree for a treatment contrast? It then states the null hypothesis
for each contrast: direct and independent indirect effects are equal
(difference=0). The z test and p-value test this null, not the network effect
or the direct evidence proportion.
Printed comparisons use an explicit contrast label, for example
Metformin - Acarbose, matching the direction of direct and indirect effects.
Stored treatment_1 and treatment_2 columns are unchanged.
Nonestimable splits retain a reason and missing difference inference.
Printed output shows only estimable comparisons by default. The compact display
includes direct-study count (k), direct evidence proportion (prop), coherent
network estimate, direct/indirect estimates, difference, signed z, and raw and
adjusted p-values. Use detailed=TRUE for SEs, intervals and diagnostic fields.
prop is the ratio of primary model-based network variance to direct-only
variance, rather than a study or patient fraction. It does not use the variance
from the whole-study-exclusion indirect fit.
The returned data frame retains all requested rows, including nonestimable comparisons and
their reasons. Use print(result, estimable_only=FALSE) to display every row.
A shared method is printed once above the results. The estimable column is
omitted when every displayed row is estimable; empty reason columns are
omitted. Mixed methods and nonestimability information remain visible.
summary(result, digits=3) records display precision and returns the complete
result; it does not compute column-wise descriptive statistics. Both summary
and print accept estimable_only, and explicit print arguments override the
choices recorded by summary.
Display filtering does not recalculate multiplicity-adjusted p-values or
change the hypothesis family. If no split is estimable, printing reports that
no estimable direct/indirect comparisons are available.
This is a pair-specific assessment, not an omnibus test. Split tests overlap;
neither their statistics nor their degrees of freedom form an additive
heterogeneity decomposition. Use network_design_incoherence() for Higgins'
design-by-treatment global test and network_heterogeneity() for within-design
variation. Use network_q_decomposition() for total/within/between residual Q.
Migration: earlier development versions returned a global/local design-test
list. Move those calls, including method="conditional" and method="ML", to
network_design_incoherence(). Accessing $global or $local on the new split
table raises a migration error rather than silently returning NULL.
Examples
d <- data.frame(s=1:9,a=rep(c("A","A","B"),3),
b=rep(c("B","C","C"),3),g=rep(c(.2,.6,.2),3),v=.04)
f <- network_meta(d,"s","a","b","g","v")
network_incoherence(f,adjust="holm")
#> Direct versus indirect incoherence
#> Question: Do direct and independent indirect estimates disagree for a treatment contrast?
#> H0 (each contrast): Direct and independent indirect effects are equal (difference = 0).
#> Contrast = treatment_2 - treatment_1; difference = direct - indirect.
#>
#> Method: conditional
#>
#> k = direct studies; prop = network variance / direct-only variance.
#> prop uses primary fitted model-based covariance; z = difference / SE.
#> P-value adjustment: holm; family size: 3.
#> nma = coherent estimate; diff = direct - indirect; p = raw; p_adj = adjusted.
#>
#> contrast k prop nma direct indirect diff z
#> B - A 3 0.6666667 0.2666667 0.2000000 0.4000000 -0.2000000 -1.0000000
#> C - A 3 0.6666667 0.5333333 0.6000000 0.4000000 0.2000000 1.0000000
#> C - B 3 0.6666667 0.2666667 0.2000000 0.4000000 -0.2000000 -1.0000000
#> p p_adj
#> 0.3173105 0.9519315
#> 0.3173105 0.9519315
#> 0.3173105 0.9519315
#>
#> Use detailed=TRUE for standard errors, intervals and diagnostic fields.
network_design_incoherence(f)$global
#> statistic df p_value estimable null_heterogeneity alternative_heterogeneity
#> 1 1 1 0.3173105 TRUE study 0 study 0
#> null_convergence alternative_convergence reason
#> 1 0 0
network_heterogeneity(f)$overall
#> component statistic df p_value covariance I2_percent
#> 2 within_designs 0 6 1 sampling covariance only 0