MCPMod {MCPMod} | R Documentation |
Tests for a dose-response effect with a model-based multiple contrast test and
estimates a target dose with regression techniques. For details see
Bretz et al. (2005) or the enclosed vignette, available via the command vignette("MCPMod")
.
MCPMod(data, models = NULL, contMat = NULL, critV = NULL, resp = "resp", dose = "dose", off = NULL, scal = NULL, alpha = 0.025, twoSide = FALSE, selModel = c("maxT", "AIC", "BIC", "aveAIC", "aveBIC"), doseEst = c("MED2", "MED1", "MED3", "ED"), std = TRUE, start = NULL, uModPars = NULL, addArgs = NULL, dePar = NULL, clinRel = NULL, lenDose = 101, pW = NULL, control = list(maxiter = 100, tol = 1e-06, minFactor = 1/1024), signTtest = 1, pVal = FALSE, testOnly = FALSE, mvtcontrol = mvtnorm.control(), na.action = na.fail, uGrad = NULL)
data |
Data frame containing the dose and the response data. The code assumes
the columns to be named "dose" and "resp". Other names can be handed over via
the dose and resp argument see below. |
models |
A list specifying the candidate models. The names of the list entries should
be equal to the names of the model functions. The list entries should be equal to the guesstimates. See the Examples (ii)
for details on this topic. If the contMat argument is specified, this argument is ignored, see Examples (iv). |
contMat |
Optional matrix containing the optimal contrasts in the columns. The names of the columns should be equal to the underlying model function names. If specified the code does not calculate the optimal contrasts. |
critV |
Optional numeric specifying the critical value to be used in the multiple contrast test. |
resp |
Character string giving the name of the response column for the data frame
specified in data (default: "resp"). |
dose |
Character string giving the name of the dose column for the data frame
specified in data (default: "dose"). |
off |
Fixed offset parameter needed when the linear in log model is used.
See also documentation of the linear in log model: linlog .
When off = NULL by default (maximum dose)*0.1 is
used for off . |
scal |
Fixed scale parameter needed when the beta model is used.
See also documentation of the beta model: betaMod .
When scal = NULL by default (maximum dose)*1.2 is
used for scal . |
alpha |
Level of significance for the multiple contrast test (defaults to 0.025) |
twoSide |
Optional logical value deterimining whether two-sided or one-sided testing should be performed. Defaults to FALSE, so one-sided testing. |
selModel |
Optional character vector specifying the model selection criterion for dose estimation. Possible values are "maxT": Selects the model corresponding to the largest t-statistic (this is the default). "AIC": Selects model with smallest AIC "BIC": Selects model with smallest BIC "aveAIC": Uses a weighted average of the models corresponding to the significant contrasts. The model weights are chosen by the formula: wi = exp(-0.5AICi)/sum(exp(-0.5AICi)). See Buckland et al. (1997) for details. "aveBIC": Same as "aveAIC", but the BIC is used to calculate the model weights. |
doseEst |
Determines which dose to estimate and which dose estimator to use, possible values are "MED2", "MED1", "MED3" and "ED". See Bretz et al. (2005) for the definition of MED1-MED3. If ED is specified, the dose that gives a pre-specified percentage of the maximum effect is returned. |
std |
Optional logical value determining, whether standardized versions should be assumed for calculation of the optimal contrasts. If FALSE all model parameters need to be specified in the models argument (also location and scale parameters). |
start |
List containing starting values for the nls fitting algorithm. The names of the list elements need to be equal to the names of the model functions. The names of the starting vector should equal the corresponding names for the model parameters. For built-in models starting values need to be provided only for the non-linear parameters. In case of a user model (not built in) starting values for the all parameters need to be supplied. (see Examples (iii) for details). |
uModPars |
Optional character vector with names/expressions of user-defined model parameters (names(start) used by default). |
addArgs |
Optional character vector with names of additional arguments (variables) to user-defined model. |
dePar |
Numeric, defining parameter used for dose estimators. For the MED-type estimators dePar determines
the confidence level gamma used in the estimator. The used confidence level is given by
1-2*dePar . The default for dePar for MED-type estimators is 0.1. For ED-type estimators dePar
determines which effective dose is estimated. Specifying 0.95 for example results in an estimate of the ED95.
If the ED estimator is used the default for dePar is 0.5. |
clinRel |
Numeric specifying the clinical relevance threshold. |
lenDose |
Numeric vector specifying the number of points in the dose-range to search for the dose estimate, defaults to 101. |
pW |
Optional vector specifying prior weights for the different models. Should be a named vector with names matching the names of the models list. |
control |
List of parameters to be used in the calls to the nls function. See also
nls.control function. |
signTtest |
Optional numeric vector multiplied with the test statistics. |
pVal |
Optional logical determining whether p-values should be calculated, defaults to FALSE. If the critical value is supplied, p-values will not be calculated. |
testOnly |
Logical value determining, whether only the multiple comparisons test should be performed. See Examples (v) below. |
mvtcontrol |
A list specifying additional control parameters for the qmvt and pmvt calls in the code,
see also mvtnorm.control for details. |
na.action |
A function which indicates what should happen when the data contain NAs. |
uGrad |
Function to return the gradient of a user defined model, see Examples (iii) below. |
This function performs the multiple comparisons and modelling (MCPMod) procedure presented in
Bretz et al. (2005). The method consists of two steps:
(i) MCP step:
The function calculates the optimal contrasts (if not supplied) and
the contrast test statistics. In the calculation of the critical
value and p-values multiplicity is taken into account.
(ii) Modelling step:
If there is at least one significant contrast, one model or a combination
of models is chosen (depending on the selModel
argument) for dose estimation.
In case of non-convergence of certain non-linear models the
remaining significant models are used. Finally the target dose is estimated.
Built in models are the linear, linear in log, emax, sigmoid emax, logistic, exponential, quadratic and beta model (for their definitions see their help files or Bretz et al. (2005), Pinheiro et al. (2006)). Users may hand over their own model functions for details have a look at the Example (iii).
An object of class MCPMod, with the following entries:
signf |
Logical indicating, whether multiple contrast test is significant |
model1 |
Model with largest contrast test statistic |
model2 |
Model(s) used for estimation of target doses |
input |
A list with entries equal to the input parameters for the function: models , resp , dose ,
off , scal , alpha , twoSide , first entry of selModel , doseEst ,
std , dePar , uModArgs , addArgs , start , uGrad , clinRel ,
lenDose , signTtest , pVal , testOnly |
data |
The data set. |
contMat |
The contrast matrix. |
corMat |
The correlation matrix. |
cVal |
The critical value for the multiple contrast test. |
tStat |
The contrast test-statistics. If 'pVal=TRUE' the p-values are also attached. |
fm |
List containing the dose-response model(s) used for dose-estimation. WARNING: The
model fitting is for computational efficiency done based on the group
means and for positive non-linear parameters (e.g. the ED50 parameter in the Emax model) the
estimate in fm is on log scale. The summary.MCPMod method shows the parameters on their original scale.
Hence some care is hence needed when extracting the fitted model objects from fm
(in particular when interest is in standard deviations of predictions, parameter estimates...).
|
tdose |
Estimated target dose, in case of model averaging the dose estimates under the individual models are attached. |
Note: If testOnly=TRUE
, or no model is significant, the object does not contain fm
and tdose
entries
Bornkamp B., Pinheiro J. C., and Bretz, F. (2009). MCPMod: An R Package for the Design and Analysis of Dose-Finding Studies, Journal of Statistical Software, 29(7), 1–23
Bretz, F., Pinheiro, J. C., and Branson, M. (2005), Combining multiple comparisons and modeling techniques in dose-response studies, Biometrics, 61, 738–748
Pinheiro, J. C., Bornkamp, B., and Bretz, F. (2006). Design and analysis of dose finding studies combining multiple comparisons and modeling procedures, Journal of Biopharmaceutical Statistics, 16, 639–656
Pinheiro, J. C., Bretz, F., and Branson, M. (2006). Analysis of dose-response studies - modeling approaches, in N. Ting (ed.). Dose Finding in Drug Development, Springer, New York, pp. 146–171
Bretz, F., Pinheiro, J. C., and Branson, M. (2004), On a hybrid method in dose-finding studies, Methods of Information in Medicine, 43, 457–460
Buckland, S. T., Burnham, K. P. and Augustin, N. H. (1997). Model selection an integral part of inference, Biometrics, 53, 603–618
logistic
, sigEmax
,
linlog
, linear
, quadratic
,
emax
, betaMod
, exponential
,
plot.MCPMod
, mvtnorm.control
# (i) # example from Biometrics paper p. 743 data(biom) models <- list(linear = NULL, linlog = NULL, emax = 0.2, exponential = c(0.279,0.15), quadratic = c(-0.854,-1)) dfe <- MCPMod(biom, models, alpha = 0.05, dePar = 0.05, pVal = TRUE, selModel = "maxT", doseEst = "MED2", clinRel = 0.4, off = 1) # detailed information is available via summary summary(dfe) # plots data with selected model function plot(dfe) # example with IBS data data(IBS) models <- list(emax = 0.2, quadratic = -0.2, linlog = NULL) dfe2 <- MCPMod(IBS, models, alpha = 0.05, pVal = TRUE, selModel = "aveAIC", clinRel = 0.25, off = 1) dfe2 # show more digits in the output print(dfe2, digits = 8) summary(dfe2, digits = 8) plot(dfe2, complData = TRUE, cR = TRUE) plot(dfe2, CI = TRUE) # simulate dose-response data dfData <- genDFdata(model = "emax", argsMod = c(e0 = 0.2, eMax = 1, ed50 = 0.05), doses = c(0,0.05,0.2,0.6,1), n=20, sigma=0.5) models <- list(emax = 0.1, logistic = c(0.2, 0.08), betaMod = c(1, 1)) dfe3 <- MCPMod(dfData, models, clinRel = 0.4, critV = 1.891, scal = 1.5) # (ii) constructing a model list # Contrasts to be included: # Model guesstimate(s) for stand. model parameter(s) (name) # linear - # linear in log - # Emax 0.2 (ED50) # Emax 0.3 # exponential 0.7 (delta) # quadratic -0.85 (delta) # logistic 0.4 0.09 (ED50, delta) # logistic 0.3 0.1 # betaMod 0.3 1.3 (delta1, delta2) # sigmoid Emax 0.5 2 (ED50, h) # The names for the list elements need to be written exactly # as the model functions ("linear", "linlog", "quadratic", "emax", # "exponential", "logistic", "betaMod", "sigEmax"). # For models with no parameter in the standardized model just NULL is # specified as list entry. # For models with one parameter a vector needs to be used with length # equal to the number of contrasts to be used for this model class. # For the models with two parameters in the standardized model a vector # is used to hand over the contrast, if it is desired to use just one # contrast. Otherwise a matrix with the guesstimates in the rows needs to # be used. For the above example the models list has to look like this list(linear = NULL, linlog = NULL, emax = c(0.2, 0.3), exponential = 0.7, quadratic = -0.85, logistic = matrix(c(0.4, 0.3, 0.09, 0.1), nrow = 2), betaMod = c(0.3, 1.3), sigEmax = c(0.5, 2)) # Additional parameters (who will not be estimated) are the "off" # parameter for the linlog model and the "scal" parameter for the # beta model, which are not handed over in the model list. # (iii) example for incorporation of a usermodel # simulate dose response data dats <- genDFdata("sigEmax", c(e0 = 0, eMax = 1, ed50 = 2, h = 2), n = 50, sigma = 1, doses = 0:10) # define usermodel userMod <- function(dose, a, b, d){ a + b*dose/(dose + d) } # define gradient userModGrad <- function(dose, a, b, d) cbind(1, dose/(dose+d), -b*dose/(dose+d)^2) # name starting values for nls start <- list(userMod=c(a=0, b=1, d=2)) models <- list(userMod=c(0, 1, 1), linear = NULL) MM1 <- MCPMod(dats, models, clinRel = 0.4, selModel="AIC", start = start, uGrad = userModGrad) # (iv) Contrast matrix and critical value handed over and not calculated # simulate dose response data dat <- genDFdata(mu = (0:4)/4, n = 20, sigma = 1, doses = (0:4)/4) # construct optimal contrasts and critical value with planMM doses <- (0:4)/4 mods <- list(linear = NULL, quadratic = -0.7) pM <- planMM(mods, doses, 20) MCPMod(dat, models = NULL, clinRel = 0.3, contMat = pM$contMat, critV = pM$critVal) # (v) Using MCPMod for mutiple contrast tests only mu1 <- c(1, 2, 2, 2, 2) mu2 <- c(1, 1, 2, 2, 2) mu3 <- c(1, 1, 1, 2, 2) mMat <- cbind(mu1, mu2, mu3) dimnames(mMat)[[1]] <- doses pM <- planMM(muMat = mMat, doses = doses, n = 20, cV = FALSE) # calculate p-values fit <-MCPMod(dat, models = NULL, clinRel = 0.3, contMat = pM$contMat, pVal = TRUE, testOnly = TRUE) summary(fit)