*** netmeta, version 0.9-6, 2017-08-09 ***

 ** Major changes **

  * Prediction intervals can be calculated for treatment estimates
    from a network meta-analysis

  * In netmeta(), Q statistics for heterogeneity and design
    inconsistency are calculated according to Krahn et al. (2013);
    see help page of decomp.design()

  * In printouts and forest plots, the reference treatment can be
    considered as treatment of interest or comparator (default),
    i.e., either comparisons of reference vs other treatments or
    other treatments vs reference are reported

  * Tests for heterogeneity and design inconsistency are shown in
    printouts

  * A biplot can be generated to show partial ordering of treatment
    rankings for more than two outcomes

 ** User-visible changes **

  * netmeta():
    - new arguments 'prediction' and 'level.predict' to calculate
      prediction intervals
    - list elements 'Q.heterogeneity' and 'Q.inconsistency' based on
      Krahn et al. (2013)
    - new list elements 'prediction', 'lower.predict',
      'upper.predict', 'df.Q.heterogeneity',
      'pval.Q.heterogeneity','df.Q.inconsistency', and
      'pval.Q.inconsistency'
    - list element 'df' renamed to 'df.Q'
    - stop with an informative error message if (i) any treatment arm
      variance derived from the treatment comparison variances is
      negative or zero, or (ii) in case of duplicate comparisons or an
      incomplete set of treatment comparisons within a study
    - argument 'details.tol.multiarm' renamed to 'details.chkmultiarm'

  * netmeta(), forest.netmeta(), print.netmeta(),
    print.summary.netmeta(), and summary.netmeta():
    - new argument 'baseline.reference' to print results for
      comparisons between reference and other treatments, or vice versa

  * print.netmeta(), print.summary.netmeta(), and summary.netmeta():
    - new argument 'prediction' to print prediction intervals

  * print.summary.netmeta():
    - print information on tests for overall heterogeneity and
      inconsistency

  * summary.netmeta():
    - arguments 'level' and 'level.comb' removed from R function
      (i.e., one has to re-run the netmeta() command for confidence
      intervals with other coverage levels)

  * plot.netposet():
    - new argument 'plottype' to choose between scatter plot or biplot
    - new arguments to modify layout ('cex.text', 'col.text', pch,
      cex.points, col.points)
    
  * decomp.design() and netmeasures():
    - new argument 'warn' to suppress printing of warnings

 ** Internal changes **

  * New internal function upgradenetmeta() to add missing list
    elements to older netmeta objects

  * R function ci() from R package meta added to namespace

  * chkmultiarm():
    - additional checks for (i) negative and zero variances as well as
      (ii) duplicate treatment comparisons or incomplete sets of
      treatment comparisons within a study


*** netmeta, version 0.9-5, 2017-05-31 ***

 ** Major changes **

  * New function netleague() to print league table with network
    meta-analysis results

  * pairwise():
    - zero events for binary outcomes or incidence rates are handled
      correctly in multi-arm studies by adding an increment to all
      treatment arms (in older versions of netmeta inconsistent
      treatment effects for multi-arm studies were possible as
      increments were considered in individual comparisons instead of
      all comparisons for a multi-arm study)
    - print warning and information on treatment comparisons with
      missing treatment estimate or standard error

  * forest.netmeta():
    - reference group can be omitted from forest plot
    - treatments can be sorted by treatment estimate (TE), standard
      error (seTE), number of studies in direct comparison (k), and
      proportion of direct information (prop.direct)

  * netmeta():
    - additional checks for correct number of comparisons in multi-arm
      studies and more informative error message for uncorrect number
      of comparisons in multi-arm studies due to missing treatment
      effects or standard errors in single comparisons
    - separator used in comparison names to concatenate treatment
      labels can be specified by user (default: ":")

  * In decomp.design(), by default, only print designs contributing to
    design-specific decomposition of within-designs Q statistic

  * Input to netdistance() can be either a netmeta object or a matrix

 ** User-visible changes **
    
  * forest.netmeta():
    - new argument 'drop.reference.group'
    - argument 'sortvar' can be used in the following ways:
      sortvar = TE, sortvar = -TE, sortvar = seTE, sortvar = -seTE,
      sortvar = k, sortvar = -k, sortvar = prop.direct, sortvar = -prop.direct

  * print.decomp.design() and netheat():
    - new argument 'showall' which defaults to FALSE

  * print.summary.netmeta():
    - print number of designs
    - print preset between-study variance and corresponding
      information if argument 'tau.preset' is not NULL in netmeta()

  * pairwise():
    - in multi-arm studies exclude comparisons with missing sample
      size or standard error from calculation of pooled variance for
      standardized mean difference (sm = "SMD")

  * plot.netposet():
    - new default for argument 'arrows', i.e., by default, do not show
      arrows in scatter plot

  * print.netsplit():
    - number of studies providing direct evidence printed

  * netdistance():
    - argument name changed from 'A' to 'x' in order to reflect that
      input of R function can be either a netmeta object or an
      adjacency matrix
    
  * Help pages:
    - examples corrected for dataset dietaryfat
    - do not run all examples in forest.netmeta() as CRAN only allows
      a run time below 10 seconds for examples provided on a help page
    - R code to produce forest plot added to examples in dataset
      Wood2010

 ** Internal changes **

  * netmeta():
    - new list element 'd' with number of designs
    - new list element 'B.matrix' with the edge-vertex incidence
      matrix

  * summary.netmeta():
    - new list element 'd' with number of designs
    - new list element 'tau.preset'

  * netsplit():
    - new list element 'k' with number of studies providing direct
      evidence

  * netconnection():
    - argument checks added
    - better code documentation

  * Internal function decomp.tau():
    - detach all designs (including protuding edges)

  * New internal function createB() to calculate edge-vertex incidence
    matrix

  * netmeta(), netconnection(), multiarm(), and chkmultiarm():
    - use internal function createB() instead of dedicated R code

  * print.summary.netmeta(), nma.ruecker(), and decomp.tau():
    - use command pchisq(..., lower.tail = FALSE) instead of
      1 - pchisq(...)


*** netmeta, version 0.9-4, 2017-04-07 ***

 ** Bug fix release **

  * netsplit() used wrong comparison labels if argument
    'reference.group' was used in netmeta()

  * netmeasures() ignores value of argument 'reference.group' in
    netmeta object


*** netmeta, version 0.9-3, 2017-03-12 ***

 ** Major changes **

  * Calculate indirect treatment estimates based on direct evidence
    proportion

  * Ranking of treatments based on fixed effect model added to
    netrank()

  * New function netsplit() to split direct and indirect evidence

  * New functions netposet(), print.netposet(), and plot.netposet() to
    calculate, print and plot partial ordering of rankings

  * New function hasse() to draw Hasse diagram of partially ordered
    treatment rankings

  * netmeta():
    - can be used with R objects created with pairwise()
    - checks for consistency of treatment effects and variances in
      multi-arm studies

  * Import ginv() from R package MASS (for consistency checks)

  * Suggested packages added (for Hasse diagram):
    - hasseDiagram
    - grid

  * Bug fixes:
    - netmeta() calculates correct direct evidence estimates under
      random effects model (list components 'TE.direct.random',
      'seTE.direct.random', ..., 'pval.direct.random'); so far results
      from fixed effect model have been used
    - netmeta() excludes a treatment from list component 'seq' if all
      comparisons containing the respecitve treatment are excluded due to
      missing values in treatment effect or standard error
    - netmeasures() does not result in an error if no or only one
      study with two treatments is available

 ** User-visible changes **
    
  * New arguments random and tau.preset in netmeasures()

  * New functions netsplit() and print.netsplit()

  * Consider ordering of treatments in netrank() which is defined by
    argument seq in netmeta()

  * For multi-arm studoes, calculate pooled standard deviation in
    pairwise() if means and standard deviations are provided and
    summary measure is equal to "SMD"

 ** Internal changes **

  * netmeta():
    - new list element 'k.direct' with number of studies in
      meta-analyses with direct evidence

  * nma.ruecker():
    - bug fix such that estimates from random effects model are used
      for direct treatment estimates if argument 'tau.direct' is
      larger than zero

  * nma.krahn():
    - bug fix such that use of function does not result in an error if
      either no or only one study with two treatments is available

  * pairwise():
    - data.frame commands use argument stringsAsFactors = FALSE

  * chkmultiarm():
    new internal function to check consistency of treatment effects
    and variances in multi-arm studies; calls ginv() from MASS library

  * lowertri():
    new internal function to extract elements from the lower triangle
    of a matrix


*** netmeta, version 0.9-2, 2016-11-19 ***

 ** Major changes **

  * R package rgl moved from imported to suggested packages as
    - 3-D network plots are not essential for network meta-analysis
    - installation of netmeta package breaks under Mac OS if XQuartz
      is not available
    
 ** User-visible changes **
    
  * Help page of netgraph() updated (information on rgl package)

 ** Internal changes **

  * Use chkclass() from meta package to check for class membership


*** netmeta, version 0.9-1, 2016-10-13 ***

 ** Major changes **

  * Number of studies can be added to network graph

  * Distance matrix can be provided directly to generate network graph
  
  * shadowtext() from TeachingDemos package by Greg Snow added to
    netmeta package

  * P-scores can be printed in forest plot
  
 ** User-visible changes **
    
  * help page with brief overview of netmeta package added

  * netgraph():
    - new arguments to add number of studies to network graph
      (number.of.studies, cex.number.of.studies,
      col.number.of.studies, bg.number.of.studies)     
    - plastic look retained for highlighted comparisons
    - new argument D.matrix to provide treatment distances directly

  * netmeta():
    - function can be used with a single pairwise comparison without
      resulting in an error

  * forest.netmeta():
    - argument sortvar can be equal to Pscore, "Pscore", -Pscore, or
      "-Pscore" to sort treatments according to ranking generated by
      netrank()
    - argument leftcols or rightcols can include "Pscore" to add a
      column with P-Scores to the forest plot
    - new arguments small.values and digits.Pscore for P-Scores

  * print.netmeta():
    - use correct layout for network meta-analysis with a single
      pairwise comparison

  * decomp.design(), netheat(), netmeasures():
    - print a warning and return NULL for network meta-analysis with a
      single design

  * netconnection():
    - print sensible error message if argument treat2 is missing or of
      different length than argument treat 1

  * netdistance():
    - print sensible error message if argument A is not a matrix

  * Help pages updated:
    decomp.design(), print.decomp.design(),
    netgraph(), netheat(), netmeasures()
    
 ** Internal changes **

  * New function:
    - shadowtext() to print number of studies

  * nma.ruecker():
    - keep dimension of matrices W and B.matrix for network
      meta-analysis with a single pairwise comparison

  * nma.krahn():
    - print a warning and return NULL for network meta-analysis with a
      single design

  * decomp.tau(), tau.within():
    - return NULL for network meta-analysis with a single design


2016-04-26, version 0.9-0

New functions:
- netdistance (calculate distance matrix; replacement for internal
  function nodedist)
- netconnection (Get connectivity information for network)
- print.netconnection (corresponding print function)

Internal function nodedist removed (replaced by netdistance function)

Import functions from R package rgl (for 3-D plots)

New dataset Woods2010 (use long format in pairwise function)

Function netmeta:
- check connectivity of network and stop with informative error
  message if network is not fully connected
- new list components:
  'Cov.fixed' (variance-covariance matrix for fixed effect model)
  'Cov.random' (variance-covariance matrix for random effects model)

Function pairwise:
- extension to long data format (see example on help page)

Function netmeta:
- new arguments 'dim', 'eig3', and 'zpos' to generate 3-D network
  plots

Function stress (used internally):
- extension to generate 3-D network plots
- use netdistance function instead of nodedist

Function nma.ruecker (used internally):
- use of netmeta function does not result in an error for networks
  without heterogeneity / inconsistency, i.e. networks with zero
  degrees of freedom (e.g. a star-shaped network with only a single
  study for each comparison; simple example: single comparisons A-B,
  A-C, A-D)
- calculate variance-covariance matrix

Function print.netrank:
- print title of meta-analysis (if available)

Function print.summary.netmeta:
- print "--" instead of "< 0.0001" in networks without heterogeneity
  / inconsistency
- print "0" instead of "< 0.0001" if tau-squared is zero
- print 'p-value' instead of 'p.value'

Function print.decomp.design:
- print 'p-value' instead of 'p.value'

Help page of netmeta function:
- more details on contrast- and arm-based data format
- reference to book "Meta-Analysis with R" and Rücker & Schwarzer (2014) added
- add information that hazard ratio is a possible summary measure
- change error in description of adjustment in random effects model

Help page of netgraph function:
- example for 3-D network plot added

Help page of netrank function:
- reference to Rücker & Schwarzer (2015) updated

Help page of pairwise function:
- description on use of long data format added
- more information on additional arguments for meta-analysis functions

New help pages:
- netconnection, print.netconnection
- netdistance
- Wooks2010 dataset


2015-06-26, version 0.8-0

New functions netrank and print.netrank:
- frequentist method to rank treatments in network

Function netmeta:
- print less irritating warning if treatment comparisons are resorted
  (as this is more a note than a warning)

Function print.netmeta:
- minor change in printout (old: "Data utilised in network
  meta-analysis ..."; new: "Results ...")

Help pages:
- new help page for netrank function
- reference Rücker & Schwarzer (2015) added in help page of netgraph
  function
- link to pairwise function added in help page of netmeta function


2015-02-04, version 0.7-0

Version of R package meta must be larger or equal 4.0-0

Title of R package changed.

New function pairwise:
- transforms data that are given in an arm-based format (e.g. input
  for WinBUGS is of this format) to contrast-based format that can be
  read by function netmeta

New datasets:
- dietaryfat (dataset with incidence rates as outcomes)
- parkinson (continuous outcomes)
- smokingcessation (binary outcomes)

Function netmeta:
- implement a general check for correct number of comparisons for
  multi-arm studies
- use setseq function to check and set value of argument 'seq'
- use setref function to check and set value of argument
  'reference.group'
- use chklevel function from R package meta to check levels of
  confidence intervals
- consider attribute 'sm' from R objects generated with R function
  pairwise
- function can be used for a pairwise meta-analysis (bug fix in
  nma.ruecker function used internally)

Function netgraph:
- check that matrix 'thickness' (if provided) has same row and column
  names as argument 'labels'
- use setseq function to check and set value of argument 'seq'
- stop with an error message if argument 'seq' or 'labels' is NULL

Function netheat:
- no net heat plot produced if (i) the number of designs is equal or
  smaller than 2 or (ii) no between-design heterogeneity exists
- unintentional warnings omitted

Function forest.netmeta:
- print a warning that the first treatment is used as reference if the
  reference group is unspecified instead of producing an error
- use setseq function to check and set value of argument 'seq'
- use setref function to check and set value of argument
  'reference.group'

Function print.summary.netmeta:
- print "." instead of "0" or "1" for diagonal elements of treatment
  effect and confidence interval matrices
- print "." instead of "0" or "1" for reference group (if provided)
- use setref function to check and set value of argument
  'reference.group'
- use is.relative.effect function from R package meta to check if a
  relative effect measure is used (argument 'sm')

Function print.netmeta:
- use setref function to check and set value of argument
  'reference.group'
- use is.relative.effect function from R package meta to check if a
  relative effect measure is used (argument 'sm')

Function summary.netmeta:
- use setref function to check and set value of argument
  'reference.group'

Function decomp.tau and tau.within (used internally):
- bug fix such that no error is produced in decomp.design and netheat
  function for networks without heterogeneity and inconsistency

Function print.decomp.design:
- omit printing of information on between-designs Q statistic after
  detaching of single designs if no between-design heterogeneity
  exists
- use format.tau function from R package meta to print "0" instead of
  "< 0.0001" if tau-squared is zero

New functions (used internally):
- setseq - check and set argument 'seq' (and argument 'sortvar' in
  forest.meta function)
- setref - check and set argument 'reference.group'
- chklist - check for a list

New help pages for function pairwise and datasets dietaryfat,
parkinson, and smokingcessation.


2014-07-29, version 0.6-0

Function netgraph:
- complete rewrite of this function (without changing previous default
  settings substantially)
- list of major new features:
 - additional layouts beside circular presentation (see argument
   'start.layout')
 - implementation of stress majorization algorithm to optimize layout
   (argument 'iterate')
 - additional methods to determine width of lines connecting
   treatments (argument 'thickness')
 - highlight multi-arm studies (arguments 'multiarm' and
   'col.multiarm')
 - possibility to provide a neighborhood matrix to specify
    neighborhood differently than using the adjacency matrix, for
    example content-based (argument 'N.matrix')
 - possibility to provide x- and y-coordinates for network plot
   (arguments 'xpos' and 'ypos')

Function netmeta:
- calculate treatment estimates from all direct pairwise treatment
  comparisons (both fixed effect and random effects model)
- new list components:
   'tau.preset', 'TE.direct.fixed', 'seTE.direct.fixed',
   'lower.direct.fixed', 'upper.direct.fixed', 'zval.direct.fixed',
   'pval.direct.fixed', 'TE.direct.random', 'seTE.direct.random',
   'lower.direct.random', 'upper.direct.random', 'zval.direct.random',
   'pval.direct.random'

Function nma.ruecker (used internally)
- changed accordingly to reflect changes in netmeta function

Function forest.netmeta:
- new argument sortvar (default: sort treatment effect estimates
  according to list component 'seq' of netmeta object)

New functions stress and nodedist (used internally)
- auxiliary functions for netgraph function

Help pages updated accordingly


2014-06-24, version 0.5-0

Functions nma.krahn, netmeasures, netheat, decomp.design, and
print.decomp.design:
- random effects network meta-analysis added

Function netheat:
- new argument 'random' 

Functions nma.krahn, decomp.design, and netheat:
- new argument 'tau.preset'

Function decomp.design:
- correct design-specific decomposition of Q statistic in network
   meta-analysis with multi-arm studies
- list component 'Q.design' renamed to 'Q.het.design'
- list component 'Q.detach' renamed to 'Q.inc.detach'
- list component 'residuals' renamed to 'residuals.inc.detach'
- new list components:
   'Q.inc.random', 'Q.inc.random.preset',
   'Q.inc.design.random.preset', 'residuals.inc.detach.random.preset', 
   'tau.preset'

New functions tau.within and decomp.tau (used internally)

Help pages updated accordingly


2014-05-27, version 0.4-4

Functions netmeta and nma.ruecker:
- modified such that the estimated tau-squared in random effects model
  considers multi-arm studies

Function print.netmeta:
- information on percentage weight not printed as interpretation is
  difficult

Dataset Senn2013:
- use of unpooled standard error for each treatment comparison


2014-04-14, version 0.4-3

Function netmeta:
- numeric values for arguments 'treat1' and 'treat2' not converted to
  character values (only factors converted to characters)
- check whether treatments are different (arguments 'treat1' and
  'treat2')

Function print.summary.netmeta:
- print random effects estimates according to argument 'seq'

Function forest.netmeta:
- sort treatment effect estimates according to argument 'seq'

Function nma.ruecker (used internally):
- changed such that all treatment effects are calculated irregardless
  of treatment order (some treatment effects remained NA depending on
  order of treatments)


2014-03-31, version 0.4-2

Function netmeasures:
- bug fix using correct formula to calculate direct evidence
  proportion (variance instead of standard error)


2014-03-21, version 0.4-1

Function netmeta:
- Argument 'seq' added (see also R function netgraph)

Function netgraph:
- new default for argument 'seq'

Help pages updated accordingly

Some internal code cleaning to improve readability of R functions


2014-03-07, version 0.4-0

New functions added:
- netgraph (network graph)
- netheat (net heat graph)
- netmeasures (measures for network meta-analysis)
- decomp.design (design-based decomposition of Cochran's Q)
- print.decomp.design (corresponding print function)
- p.ci, format.TE, nma.krahn, nma.ruecker (used internally)

Function netmeta:
- Check added whether all pairwise comparisons are provided for
  multi-arm studies

Help pages added for new functions

Help page of function netmeta updated


2013-08-01, version 0.3-1

Functions netmeta and summary.netmeta:
- new list component n (number of treatments)

Function print.summary.netmeta:
- modified such that number of treatments is printed
- modified such that argument 'reference.group' works as expected for
  random effects model


2013-07-24, version 0.3-0

First version released on CRAN
