index.batch: The main function to compute the point estimates and 95% confidence intervals (for a parametric model) of the conditional probabilities $Pr(q(\boldsymbol{Y}_{i,new}) \ge q(\boldsymbol{y}_{i,new})| \boldsymbol{Y}_{i,pre}=\boldsymbol{y}_{i,pre})$ for multiple subjects.

Description

Let $m_i$ be the number of pre-measurements and $n_i$ be the total number of repeated measures. Then the repeated measure of a subject can be divided into a pre-measurement set and a new measurement set as $\boldsymbol{Y}_{i}=(\boldsymbol{Y}_{i,pre},\boldsymbol{Y}_{i,new})$, where $\boldsymbol{Y}_{i,pre}=(Y_{i,1},\cdots,Y_{i,m_i})$ and $\boldsymbol{Y}_{i,new}=(Y_{i,m_i+1},\cdots,Y_{i,n_i})$ . Given an output of fitParaIND, fitParaAR1, fitSemiIND, fitSemiAR1 or lmeNB, this function computes the probability of observing the response counts as large as those new observations of subject $i$, $\boldsymbol{y}_{i,new}$ conditional on the subject's previous observations $\boldsymbol{y}_{i,pre}$ for subject $i$. That is, this function returns a point estimate and its asymptotic 95% confidence interval (for a parametric model) of the conditional probability for each subject: $Pr(q(\boldsymbol{Y}_{i,new}) \ge q(\boldsymbol{y}_{i,new})| \boldsymbol{Y}_{i,pre}=\boldsymbol{y}_{i,pre})$.

When the semiparametric approach is employed, the standard error and 95% confidence intervals are computed using bootstrap samples. A scalar statistic to summarize the new response counts can be either the total count, $q(\boldsymbol{Y}_{i,new})=\sum_{j=m_i+1}^{n_i} Y_{ij}$, or the maximum, $q(\boldsymbol{Y}_{i,new})=\max{ Y_{ij};j=m_i+1,\cdots,n_i }$.See Zhao et al.(2013), for more details.

Usage

index.batch(data,         labelnp,     ID,  Vcode = NULL, olmeNB = NULL, subset = NULL, 
            qfun = "sum", IPRT = TRUE, i.se = TRUE, MC = FALSE, C = FALSE,i.tol=1E-75)

Arguments

data

See lmeNB. This dataset does not have to be the same as the one used in the computations of negative binomial mixed effect regression (fitParaIND,

labelnp

A vector of length the total number of repeated measures (=$\sum_{i=1}^N n_i$), indicating new measures by TRUE and pre-measures by FALSE. For examples, suppose there are three subjects of interest.

The first subject has a $n_1

See lmeNB. The length of ID must be the same as nrow(data).

Vcode

Necessary only if the olmeNB is an output of AR(1) models. See lmeNB.

olmeNB

Output of fitParaIND,fitParaAR1,fitSemiIND,fitSemiAR1<

subset

An optional expression indicating the subset of the subjects of that the index should be computed.

qfun

If qfun="sum", a scalar statistic to summarize the new response counts is the total count. If qfun="max", a scalar statistic to summarize the new response counts is the maximum.

IPRT

print control.

i.se

If i.se=TRUE then the standard errors of the estimator of the conditional probability are returned for the output of fitParaIND or fitParaAR1

Necessary when olmeNB if the AR(1) model outputs. See CP.ar1.se.

See lmeNB. C=TRUE option could make computations of CPI faster for some patients.

i.tol

See lmeNB.

Value

The $N$ by 4 (3, if hide the SE) numeric matrix, containing the point estimate of the conditional probability, and the lower and the upper bounds of the 95

Details

The standard error of the point estimate on the logit scale is constructed using the delta method for the parametric model, where distributional assumption was made for random effects.

References

Detection of unusual increases in MRI lesion counts in individual multiple sclerosis patients. (2013) Zhao, Y., Li, D.K.B., Petkau, A.J., Riddehough, A., Traboulsee, A., Journal of the American Statistical Association.

Examples

Run this code

## See the examples in help files of 
## fitParaIND, fitAR1IND, fitSemiIND, fitSemiAR1 and rNBME.R

Run the code above in your browser using DataLab