Estimate the Pattern of Longitudinal and Survival Data
Function estimate_pattern_long_surv estimate the pattern of longitudinal and survival data from a dataset of n subjects. This is usually the first step of dynamic screening. The risk of a subject to event is quantified by a linear combination of longitudinal data by a Cox model. The risk pattern can be described by mean and variance depending on the estimation method. When the estimated pattern is used for monitoring new subjects, the collected data from new subjects are compared to the estimated pattern for monitoring abnormality.
estimate_pattern_long_surv( data_array, time_matrix, nobs, starttime, survtime, survevent, design_interval, n_time_units, time_unit, estimation_method = "risk", smoothing_method = "local linear", bw_beta, bw_mean, bw_var )
data_array: observed data arranged in a 3d array format.
data_array[i,j,k] is the jth observation of the kth dimension of the ith subject.
time_matrix: observation times arranged in a numeric matrix format.
time_matrix[i,j] is the jth observation time of the ith subject.
data_array[i,j,] is observed at time_matrix[i,j].
nobs: number of observations arranged as an integer vector.
nobs[i] is the number of observations for the ith subject.
starttime: a vector of entry times
starttime[i] is the entry time of the ith subject.
survtime: a vector of survival times
survtime[i] is the survival time of the ith subject.
survevent: a logical vector of survival events
If survevents[i]==TRUE, then a survival event is observed at survtime[i].
If survevents[i]==FALSE, then no survival event is observed at survtime[i].
design_interval: a numeric vector of length two that gives the left- and right- limits of the design interval. By default, design_interval=range(time_matrix,na.rm=TRUE).
n_time_units: an integer value that gives the number of basic time units in the design time interval.
The design interval will be discretized to seq(design_interval[1],design_interval[2],length.out=n_time_units)
time_unit: an optional numeric value of basic time unit. Only used when n_time_units is missing.
The design interval will be discretized to seq(design_interval[1],design_interval[2],by=time_unit)
estimation_method: a string.
If estimation_method="risk", apply the risk monitoring method (c.f., You and Qiu 2020).
(Currently only the method "risk" is available.)
smoothing_method: a string.
If smoothing_method="local constant", apply local constant smoothing
If smoothing_method="local linear", apply local linear smoothing
bw_beta: an integer value.
The bandwidth parameter for estimating the regression coefficients beta in the Cox model.
bw_mean: an integer value.
The bandwidth parameter for estimating mean function.
bw_var: an integer value.
The bandwidth parameter for estimating variance function.
an object that stores the estimated longitudinal pattern and model parameters.
If estimation_method="risk", returns an object of class pattern_long_surv_risk.
$grid: discretized design interval.
$beta_est: Estimated regression coefficients.
$mean_risk_est: Estimated mean function.
$var_risk_est: Estimated variance function.
Estimate the Pattern of Longitudinal and Survival Data
data("data_example_long_surv") result_pattern<-estimate_pattern_long_surv( data_array=data_example_long_surv$data_array_IC, time_matrix=data_example_long_surv$time_matrix_IC, nobs=data_example_long_surv$nobs_IC, starttime=data_example_long_surv$starttime_IC, survtime=data_example_long_surv$survtime_IC, survevent=data_example_long_surv$survevent_IC, design_interval=data_example_long_surv$design_interval, n_time_units=data_example_long_surv$n_time_units, estimation_method="risk", smoothing_method="local linear", bw_beta=0.05, bw_mean=0.1, bw_var=0.1)
You, L. and Qiu, P. (2020). An effective method for online disease risk monitoring. Technometrics, 62(2):249-264.
Useful links