mixmeta (version 1.2.0)

mixmetaFormula: Formulae in mixmeta Models

Description

Meta-analytical models fitted with mixmeta are defined by specific formulae in its arguments formula and random. The formulae offer compact symbolic expressions with form y ~ x + z, where the response y in the left-hand side of the operator ~ is modelled in terms of meta-predictors x and z in the right-hand side. Terms are separated by +, and additional syntactic operators and existing functions can be used within a formula to specify transformations such as categorization and interactions, among others, as in standard formula expressions (see formula for details). The usage of formulae in mixmeta for the random-effects part follows closely the definition in the the nlme package.

Fixed-effects formula

The argument formula of mixmeta defines the fixed-effects part. Models for meta-analysis with no meta-predictors can be specified using the form y ~ 1, or alternatively including only the term y (in this case, the formula is reconstructed internally). Multivariate models can be defined by using a matrix-type y, with columns as multiple outcomes, or directly in the formula with form cbind(y1 + y2) ~ 1. In meta-regression models, other terms are added in the right-hand side of the formula as y ~ x1 + ... + xp, defining the linear meta-predictor. In multivariate meta-regression, the same linear predictor is specified for each outcome.

Random-effects formula or formulae

The argument random of mixmeta defines the random-effects part. When this is not specified, it is assumed that each row of data is from an independent study and assigned to a different group, as in standard meta-analytical models. If provided, this is usually represented by a one-sided formula whose basic random-intercept form is ~ 1 | g. The term g at the right-hand side of the special operator | is the grouping factor, always required in a single random-effects formula. A more complex random-effects part can be also specified by ~ z1 + ... + zq | g, where the terms in the left-hand side defines a linear meta-predictor, with syntax identical to the usual formulae.

The argument random also accepts a list of one-sided formulae referring to multiple random-effects levels in multilevel meta-analytical models. In this case, levels are assumed to be nested in the order of the list, from the lowest (outer) to the highest (inner), consistently with the grouping factors. These are usually defined by different terms in the right-hand side of the | operator, although in the list form can also be provided as names of the list components. This information is used internally to reconstruct the grouping structure and the random-effects design matrices. Each level can have different linear predictors, but if these are identical across levels the random-effects part can be defined by a single equation ~ z | g1 / g2, where the special operator / separates the grouping factors g2 nested in g1.

References

Sera F, Armstrong B, Blangiardo M, Gasparrini A (2019). An extended mixed-effects framework for meta-analysis.Statistics in Medicine. 2019;38(29):5429-5444. [Freely available here].

See Also

See mixmeta. See formula for standard regression formulae. See mixmeta-package for an overview of this modelling framework.

Examples

# NOT RUN {
# STANDARD RANDOM-EFFECTS META-ANALYSIS (WITH DIFFERENT SYNTAXES)
mixmeta(logor, logorvar, data=bcg)
mixmeta(logor ~ 1, logorvar, data=bcg)

# META-REGRESSION
mixmeta(logor ~ ablat, logorvar, data=bcg)

# MULTIVARIATE MODEL
model <- mixmeta(cbind(PD,AL) ~ pubyear, S=berkey98[5:7], data=berkey98)

# NON-STANDARD MODEL: REPEATED MEASURED WITHING THE SAME GROUPS
mixmeta(effect, var, random= ~ 1|district, data=school)
mixmeta(absrisk, var, random= ~ 1|trial, data=thrombolytic)

# NON-STANDARD MODEL: MORE COMPLEX RANDOM-EFFECTS PREDICTOR
mixmeta(logOR~time, logORvar, random= ~ I(time-15)|study, data=gliomas)

# MULTILEVEL MODEL (WITH DIFFERENT SYNTAXES)
mixmeta(effect, var, random= ~ 1|district/study, data=school)
mixmeta(effect, var, random=list(~ 1|district, ~ 1|study), data=school)
mixmeta(effect, var, random=list(district = ~ 1, study = ~ 1), data=school)
# }