RTSA {RTSA}

R version of Trial Sequential Analysis. Used for designing and analysing sequential meta-analyses.

Description

R version of Trial Sequential Analysis. Used for designing and analysing sequential meta-analyses.

Usage

RTSA(
  type = "design",
  outcome = NULL,
  side = 2,
  alpha = 0.05,
  beta = 0.1,
  futility = "none",
  es_alpha = "esOF",
  es_beta = NULL,
  timing = NULL,
  data = NULL,
  design = NULL,
  ana_times = NULL,
  fixed = FALSE,
  mc = NULL,
  RRR = NULL,
  sd_mc = NULL,
  pC = NULL,
  weights = "MH",
  re_method = "DL_HKSJ",
  tau_ci_method = "BJ",
  gamma = NULL,
  rho = NULL,
  study = NULL,
  cont_vartype = "equal",
  zero_adj = 0.5,
  tau2 = NULL,
  I2 = NULL,
  D2 = NULL,
  trials = NULL,
  final_analysis = NULL,
  inf_type = "sw",
  conf_level = 0.95,
  random_adj = "tau2",
  power_adj = TRUE,
  ...
)

Arguments

type

Type of RTSA. Options are "design" or "analysis".

outcome

Outcome metric. Options are: RR (risk ratio/relative risk), OR (odds ratio), RD (risk difference) and MD (mean difference).

side

Whether a 1- or 2-sided hypothesis test is used. Options are 1 or 2. Default is 2.

alpha

The level of type I error as a percentage, the default is 0.05 corresponding to 5%.

beta

The level of type II error as a percentage, the default is 0.1 corresponding to 10%.

futility

Futility boundaries added to design. Options are: none, non-binding and binding. Default is "none".

es_alpha

The spending function for alpha-spending. Options are: esOF (Lan & DeMets version of O'Brien-Fleming), esPoc (Lan & DeMets version of Pocock), HSDC (Hwang Sihi and DeCani) and rho (rho family).

es_beta

The spending function for beta-spending. For options see es_alpha.

timing

Expected timings of interim analyses when type = "design". Defaults to NULL.

data

A data.frame containing the study results. The data set must containing a specific set of columns. These are respectively 'eI' (events in intervention group), 'eC' (events in control group), 'nC' (participants intervention group) or 'nI' (participants control group) for discrete data, or, 'mI' (mean intervention group), 'mC' (mean control group), 'sdI' (standard error intervention group), 'sdC' (standard error control group),'nC' (participants intervention group) and 'nI' (participants control group) for continuous outcomes. Preferable also a 'study' column as an indicator of study.

design

RTSA object where type is design.

ana_times

An optional vector of analysis times. Used if the sequential analysis is not done for all studies included in the meta-analysis.

fixed

Should only a fixed-effect meta-analysis be computed. Default is FALSE.

mc

Minimal clinical relevant outcome value

RRR

Relative risk reduction. Used for binary outcomes with outcome metric RR. Argument mc can be used instead. Must be a value between 0 and 1.

sd_mc

The expected standard deviation. Used for sample size calculation for mean differences.

pC

The expected probability of event in the control group. Used for sample size calculation for binary outcomes.

weights

Weighting method options include IV (inverse-variance) and MH (Mantel-Haenszel). Defaults to MH.

re_method

Method for calculating the estimate of heterogeneity, tau^2, and the random-effects meta-analysis variance. Options are "DL" for DerSimonian-Laird and "DL_HKSJ" for the Hartung-Knapp-Sidik-Jonkman adjustment of the DerSimonian-Laird estimator.

tau_ci_method

Method for calculating confidence intervals for the estimated heterogeneity tau^2. Options are "QP" for Q-profiling and "BJ" for Biggelstaff ....

gamma

Parameter for the HSDC error spending function.

rho

Parameter for the rho family error spending function.

study

An optional vector of study names and perhaps year of study. Defaults to NULL.

cont_vartype

For mean difference outcomes, do we expect the variance in the different groups to be "equal" or "non-equal".

zero_adj

Zero adjustment. Options for now is 0.5.

tau2

Heterogeneity estimate. Used for sample and trial size calculation. Defaults to NULL.

I2

Inconsistency estimate. Used for sample and trial size calculation. Defaults to NULL.

D2

Diversity estimate. Used for sample and trial size calculation. Defaults to NULL.

trials

Number of anticipated extra trials. Used for heterogeneity adjustment by tau2.

final_analysis

Whether or not the current analysis is the final analysis.

inf_type

Stopping time confidence interval. Options for now is sw (stage-wise).

conf_level

Confidence level on stopping time confidence interval.

random_adj

The sample size adjustment based on presence of heterogeneity. Options are "D2" (Diversity), "I2" (Inconsistency) and "tau2" (the heterogeneity estimate). Default is "tau2".

power_adj

Whether the sample size should be adjusted by the sequential design. Defaults to TRUE.

...

other arguments

Value

A RTSA object, a list of five elements:

settings

A list containing all of the settings used in the RTSA call. See Arguments.

ris

List containing sample and trial size calculations for a non-sequential meta-analysis. See documentation for ris function.

bounds

List of stopping boundaries, timing of trials and more. See documentation for boundaries function.

results

List of 3 to 7 elements. AIS Achieved information size. RIS Fixed-effect required information size for a non-sequential meta-analysis. SMA_RIS RIS adjusted for sequential analysis. HARIS Heterogeneity adjusted required information size for a non-sequential meta-analysis. SMA_HARIS HARIS adjusted for sequential analysis. results_df a data.frame of inference, see documentation for inference function. seq_inf a list of conditional inference, see documentation for inference function. metaanalysis A metaanalysis object, see documentation for metaanalysis function. design_df a data.frame containing the stopping boundaries and timings from the design.

warnings

List of warnings

Examples

## Not run: 
### Retrospective sequential meta-analysis:
# A RRR of 20% is expected which gives mc = 1 - RRR = 0.8. 
# No futility boundaries
data(perioOxy)
RTSA(type = "analysis", data = perioOxy, outcome = "RR", mc = 0.8, side = 2,
 alpha = 0.05, beta = 0.2, es_alpha = "esOF")

# Set binding futility boundaries
# And use Lan and DeMets' version of Pocock stopping boundaries 
RTSA(type = "analysis", data = perioOxy, outcome = "RR", mc = 0.8, side = 2,
 alpha = 0.05, beta = 0.2, es_alpha = "esOF", futility = "binding",
 es_beta = "esPoc", random_adj = "D2")

# Set non-binding futility boundaries
RTSA(type = "analysis", data = perioOxy, outcome = "RR", mc = 0.8, side = 2,
 alpha = 0.05, beta = 0.2, es_alpha = "esOF", futility = "non-binding",
 es_beta = "esOF")
 
### Design a prospective sequential meta-analysis
# For continuous data without expected heterogeneity
RTSA(type = "design", outcome = "MD", mc = 5, sd_mc = 10, side = 1, 
timing = c(0.33, 0.66, 1), fixed = TRUE,
alpha = 0.025, beta = 0.1, es_alpha = "esOF", futility = "non-binding", 
es_beta = "esPoc")

# For binary outcome
RTSA(type = "design", outcome = "RR", mc = 0.75, side = 1, 
timing = c(0.33, 0.66, 1), pC = 0.1, D2 = 0.1, 
alpha = 0.025, beta = 0.2, es_alpha = "esOF", futility = "non-binding", 
es_beta = "esOF")

# extract sample size calculation
out_rtsa <-  RTSA(type = "design", outcome = "RR", mc = 0.75, side = 1, 
timing = c(0.33, 0.66, 1), pC = 0.1, D2 = 0.1, 
alpha = 0.025, beta = 0.2, es_alpha = "esOF", futility = "non-binding", 
es_beta = "esOF")
out_rtsa$ris

# plot the design
plot(out_rtsa)

# update the design with data as it accumulates (here toy-data)
fake_data <- data.frame(eI = c(10,10), eC = c(13, 11), nI = c(750, 750),
nC = c(750,750))
RTSA(type = "analysis", design = out_rtsa, data = fake_data)

# plot the analysis
an_rtsa <- RTSA(type = "analysis", design = out_rtsa, data = fake_data)
plot(an_rtsa)

## End(Not run)