R: Decision Inconsistency and Across-Studies Inconsistency index

inc {metainc}

R Documentation

Decision Inconsistency and Across-Studies Inconsistency index

Description

Calculates the Decision Inconsistency (DI) and Across-Studies Inconsistency (ASI) indices.

Usage

inc(
  x,
  dt1,
  dt2 = NULL,
  dt3 = NULL,
  sm,
  br = NULL,
  utility = NULL,
  scale = 1000,
  transf = TRUE,
  transf.dt = FALSE
)

## S3 method for class 'inc'
print(x, digits = 1, ...)

## S3 method for class 'inc'
summary(object, ...)

## S3 method for class 'summary.inc'
print(x, digits = 1, ...)

Arguments

`x`	An R object created with `getsamples` or a matrix containing the sampled effect sizes of primary studies.
`dt1`	A single numeric defining the decision threshold to distinguish (i) meaningful from trivial effects, if arguments `dt2` and `dt3` are not provided, (ii) negative / harmful from trivial effects, if only argument `dt2` is also provided, or (iii) small from trivial effects if arguments `dt2` and `dt3` are provided.
`dt2`	A single numeric defining the decision threshold to distinguish (i) positive / beneficial from trivial effects if argument `dt3` is not provided, or (ii) moderate from small effects if argument `dt3` is provided.
`dt3`	A single numeric defining the decision threshold to distinguish large from moderate effects.
`sm`	A character string indicating the summary measure used in primary studies (see Details).
`br`	Baseline risk (only considered for odds, risk or hazard ratio).
`utility`	Utility value.
`scale`	The number of people per which absolute decision thresholds are provided (default: 1000, i.e., absolute decision threshold values are defined per 1000 persons). Only considered if `br` is not missing.
`transf`	A logical indicating whether the values of an effect size matrix (argument `x`) are to be transformed. By default `transf = TRUE`, it is assumed that the matrix contains, e.g., log odds ratios instead of odds ratios.
`transf.dt`	A logical indicating whether relative decision thresholds are transformed or on the original scale. If `transf.dt = FALSE` (default), relative decision thresholds are expected to be on the natural scale (e.g., odds ratios instead of log odds ratios for `sm = "OR"`). Note, the GRADE working group recommends to use absolute instead of relative decision thresholds.
`digits`	Minimal number of significant digits to print percentages, see `print.default`.
`...`	Additional arguments (ignored)
`object`	R object of class `inc`.

Details

Calculates the Decision Inconsistency index (DI) and the Across-Studies Inconsistency index (ASI) for a meta-analysis. The following possibilities are considered depending on the type of effect size measures:

Effect size measure corresponding to a ratio (sm = "OR", "RR" or "HR") with the DI and the ASI being calculated based on absolute effects: This requires the specification of a baseline risk (i.e., br must be defined). The decision threshold values (dt1, dt2 and dt3 must be provided as absolute effects (i.e., number of additional or diminished events per N people. By default, it is assumed that these threshold values are provided per 1000 people. However, this can be changed using the scale argument).
Effect size measure corresponding to a ratio (sm = "OR", "RR", "HR" or "GEN_ratio") with the DI and the ASI being calculated based on relative effect size measures: The sampled effect sizes of primary studies are directly compared with decision thresholds (dt1, dt2, dt3) also expressed as relative effect sizes. This is the adopted approach when no information is provided on the baseline risk (br).
Effect size measure corresponding to a difference (sm = "MD", "SMD", "RD" or "GEN_diff"): The sampled effect sizes of primary studies are directly compared with decision thresholds (dt1, dt2, dt3) also expressed as differences.

Of note, when dealing with relative effect size measures, judgements based on absolute effects tend to be considered more important for decision making. The formulae for calculating absolute effects based on relative effect size measures are those used by the GRADE approach (see references below).

Ideally, arguments dt1, dt2 and dt3 should be provided. If only one decision threshold is available, it is either possible to provide (i) only dt1, or (ii) both dt1 and dt2 (if the threshold distinguishing clinically relevant benefits vs trivial effects is different from that distinguishing clinically relevant harms vs trivial effects).

Argument sm must be "OR" (odds ratio), "RR" (risk ratio), "HR" (hazard ratio), "MD" (mean difference), "SMD" (standardised mean difference), "RD" (risk difference), "GEN_diff" (generic difference), or "GEN_ratio" (generic ratio).

The baseline risk (br) must be a numeric value between 0 and 1. It can be provided when sm = "OR", "RR" or '"HR". The baseline risk is also known as assumed comparator risk (i.e., the risk that the outcome of interest occurs in the comparison intervention).

Value

An object of class inc, for which some standard methods are available, see metainc-package. Some of the components include:

`DI`	A percentage corresponding to the Decision Inconsistency index. The higher / closer to 100% the value, the higher the inconsistency.
`ASI`	A percentage corresponding to the Across-Studies Inconsistency index. The higher / closer to 100% the value, the higher the across-studies inconsistency.
`class_distribution`	A data frame containing the proportion of samples indicating (if three decision thresholds had been provided): Large positive effects (effect sizes higher than `dt3`): "large (higher)" row; Moderate positive effects (efect sizes between `dt2` and `dt3`): "moderate (higher)" row; Small positive effects (effect sizes between `dt1` and `dt2`): "small (higher)" row; Non meaningful effects (effect sizes between `-dt1` and `dt1`): "not meaningful" row; Small negative effects (effect sizes between `-dt1` and `-dt2`): "small (lower)" row; Moderate negative effects (effect sizes between `-dt2` and `-dt3`): "moderate (lower)" row; Large negative effects (effect sizes lower than `-dt3`): "large (lower)" row.
`prop_over_null`	A numeric value indicating the proportion of samples with a value higher than the value representing no difference between the groups.

Author(s)

Bernardo Sousa-Pinto bernardo@med.up.pt, Guido Schwarzer guido.schwarzer@uniklinik-freiburg.de

References

Cohen J. (1998). “Statistical Power Analysis in the Behavioral Sciences”, 2nd edition ed. Hillsdale (NJ): Lawrence Erlbaum Associates, Inc.

Schunemann HJ, Higgins JPT, Vist GE, et al. (2019). “Completing ‘Summary of findings’ tables and grading the certainty of the evidence.” Cochrane Handbook for Systematic Reviews of Interventions, 375–402.

Schunemann HJ, Vist GE, Higgins JPT, et al. (2019). “Interpreting results and drawing conclusions.” Cochrane Handbook for Systematic Reviews of Interventions, 403–431.

Skoetz N, Goldkuhle M, van Dalen EC, et al. (2020). “GRADE guidelines 27: how to calculate absolute effects for time-to-event outcomes in summary of findings tables and Evidence Profiles.” Journal of Clinical Epidemiology, 118, 124–131.

Examples


# Example with effect sizes measures expressed as ratios and with
# calculation of the Decision Inconsistency index and the Across-Studies
# Inconsistency index based on absolute effects:

data(anticoagulation)
inc_anticoagulation <-
  inc(anticoagulation, dt1 = 16, dt2 = 31, dt3 = 60, br = 0.5, sm = "OR",
      transf = FALSE)
inc_anticoagulation

# Same result
inc_anticoagulation <-
  inc(log(anticoagulation), dt1 = 16, dt2 = 31, dt3 = 60,
    br = 0.5, sm = "OR")
inc_anticoagulation

# Example with calculation of the Decision Inconsistency index and the 
# Across-Studies Inconsistency index based on effect size measures expressed
# as mean differences:

data(montelukast)
inc_montelukast <- inc(montelukast, dt1 = 0.2, dt2 = 0.4, dt3 = 0.6, sm = "md")
inc_montelukast

[Package metainc version 0.2-0 Index]