covselec

Description

This function is used for state-wise weighted biomarker selection for both lasso and fused type penalty and is applicable for a multi-state model encompassing finite number of states and transitions.

Usage

covselec(
  data,
  time_cols,
  status_cols,
  covariate_range,
  alphas,
  lambdas,
  method = "lasso",
  p_cut = 0.2,
  verbose = TRUE
)

Arguments

Details

This function deals with biomarker selection for each state of a multi-state model each phase involves selecting significant biomarkers for that state from two sets: one comprising covariates from the preceding state, and another consisting of all covariates from the current state that were previously unselected.If the total number of biomarkers is referred to as the overall count, and each state in a multi-state model has its own number of selected biomarkers, then in a model with four states, the number of selected biomarkers for the fourth state is determined using a weighted combination. Specifically, it depends on the number of biomarkers selected in the third state, a modified set from the second state, another modified set from the first state

Value

list containing selected biomarkers for each state

Author(s)

Atanu Bhattacharjee,Gajendra Kumar Vishwakarma,Abhipsa Tripathy

Examples

##
set.seed(123)
covselec(data= highDmsm,time_cols = c("state1", "state2","state3"),
status_cols = c("status1", "status2","status3"),
covariate_range = 8:107,alphas=c(0.40,0.45,0.60),
lambdas = c(0.1,0.15,0.20),method = "lasso", p_cut = 0.2,verbose = TRUE)
##

flassomsm

Description

Fits a penalized regression model with combined Fusedlasso penalty using hybrid algorithm

Usage

flassomsm(
  msdata,
  X,
  p,
  lambda_lasso,
  lambda_fused,
  tol_outer = 1e-04,
  max_outer = 50,
  rho = 1,
  tol_admm = 1e-04,
  max_admm = 100,
  trace = TRUE,
  MSM_profile = FALSE,
  use_parallel = TRUE
)

Arguments

Details

This is the core function of the package.This function fits a penalized Cox-type regression model within the framework of a multi-state model. It is designed to handle transition-specific covariate effects across multiple states by incorporating a regularization approach that combines both the Lasso penalty and the Fused penalty. The penalization is of the L1 type, meaning it applies to the absolute values of the regression coefficients, encouraging sparsity in the model. Additionally, it penalizes the absolute differences between corresponding coefficients across different transitions, promoting similarity or grouping of effects across transitions when appropriate. This dual-penalty structure enables both variable selection and smoothing of covariate effects across related transitions, which is particularly useful in complex multi-state settings where covariate effects may share underlying patterns but still exhibit transition-specific behaviors. The parameters are estimated using a hybrid algorithm techinique combining PIRLS and ADMM together.

Value

A list with elements like matrix of estimated beta coefficients along with standard error and p value,number of iterations, aic (Akaike Information Criterion) value, gcv (GCV criterion) value and df (degrees of freedom)

Author(s)

Atanu Bhattacharjee,Gajendra Kumar Vishwakarma,Abhipsa Tripathy

Examples

##
set.seed(123)
data(msdata_3state)
covs1 <- msdata_3state[,9:10]
flassomsm(msdata = msdata_3state,X=msdata_3state[,c(11:dim(msdata_3state)[[2]])],
p = length(covs1),lambda_lasso = 0.3,lambda_fused = 0.5,tol_outer = 1e-4,
max_outer = 50, rho = 1, tol_admm = 1e-4, max_admm = 100,trace = TRUE,
MSM_profile = FALSE)

# For 2 covariates and 3 number of transitions


# Simulate msdata_4state instead of loading from disk
msdata_4state <- simdata(seed=123,n=1000,dist="weibull",cdist="exponential",
                 cparams=list(rate = 0.1),lambdas=c(0.1, 0.2, 0.3, 0.4),
                 gammas=c(1.5, 2, 2.5, 2.6),beta_list=list(c(-0.05, 0.01, 0.5, 0.6),
                 c(-0.03, 0.02, 0.07, 0.08),c(-0.04, 0.03, 0.04, -0.03),
                 c(-0.05,0.05,0.6,0.8)),cov_means=c(0,10,2,3),cov_sds=c(1,20,5,1.05),
                 trans_list=list(c(2, 3, 4, 5),c(3, 4, 5),c(4, 5), c(5), c()),
                 state_names=c("Tx", "Rec", "Death", "Reldeath", "srv"))

set.seed(123)
sub_msdata_4state <- msdata_4state[msdata_4state$id %in% sample(unique(msdata_4state$id), 10), ]
covs1 <- sub_msdata_4state[,9:10]
flassomsm(msdata = sub_msdata_4state,X=sub_msdata_4state[,c(13:32)],
         p = length(covs1),lambda_lasso = 0.5,lambda_fused = 0.6,tol_outer = 1e-4,
         max_outer = 50, rho = 1, tol_admm = 1e-4, max_admm = 100,trace = TRUE,MSM_profile = FALSE)

# For 2 covariates and 10 number of transitions
##

flassomsm_admm

Description

Fits a penalized regression model with combined Fusedlasso penalty using ADMM algorithm

Usage

flassomsm_admm(
  msdata,
  X,
  p,
  lambda_lasso,
  lambda_fused,
  tol_admm = 1e-04,
  max_admm = 100,
  rho = 1,
  trace = TRUE,
  MSM_profile = FALSE
)

Arguments

Details

This function fits a penalized Cox type regression model to a multi-state setting, where the penalty is a combination of Lasso penalty and Fused penalty. It applies L1 type penalization with the penalty applied to absolute transition-specific effects and pairwise difference between the corresponding transition effects and alternating direction method of multipliers (ADMM) algorithm.

Value

A list with elements like matrix of estimated beta coefficients along with standard error and p value,number of iterations, aic (Akaike Information Criterion) value, gcv (GCV criterion) value and df (degrees of freedom)

Author(s)

Atanu Bhattacharjee,Gajendra Kumar Vishwakarma,Abhipsa Tripathy

Examples

##
set.seed(123)
data(msdata_3state)
covs1 <- msdata_3state[,9:10]
flassomsm_admm(msdata = msdata_3state,X=msdata_3state[,c(11:dim(msdata_3state)[[2]])],
p = length(covs1),lambda_lasso = 0.3,lambda_fused = 0.5,tol_admm = 1e-4,max_admm = 10,
rho = 1, trace = TRUE, MSM_profile = FALSE)

# For 2 covariates and 3 number of transitions


# Simulate msdata_4state instead of loading from disk
msdata_4state <- simdata(seed=123,n=1000,dist="weibull",cdist="exponential",
                 cparams=list(rate = 0.1),lambdas=c(0.1, 0.2, 0.3, 0.4),
                 gammas=c(1.5, 2, 2.5, 2.6),beta_list=list(c(-0.05, 0.01, 0.5, 0.6),
                 c(-0.03, 0.02, 0.07, 0.08),c(-0.04, 0.03, 0.04, -0.03),
                 c(-0.05, 0.05, 0.6, 0.8)),cov_means=c(0, 10, 2, 3),
                 cov_sds=c(1, 20, 5, 1.05),trans_list=list(c(2, 3, 4, 5),
                 c(3,4,5),c(4,5),c(5),c()),state_names=c("Tx","Rec","Death","Reldeath","srv"))
set.seed(123)
sub_msdata_4state <- msdata_4state[msdata_4state$id %in% sample(unique(msdata_4state$id), 10), ]
covs1 <- sub_msdata_4state[,9:10]
flassomsm_admm(msdata = sub_msdata_4state,X=sub_msdata_4state[,c(13:32)],
              p = length(covs1),lambda_lasso = 0.5,lambda_fused = 0.6,
              tol_admm = 1e-4,max_admm = 10,rho = 1, trace = TRUE,
              MSM_profile = FALSE)

# For 2 covariates and 10 number of transitions
##

flassomsm_pirls

Description

Fits a penalized regression model with combined Fusedlasso penalty using PIRLS algorithm

Usage

flassomsm_pirls(
  msdata,
  X,
  p,
  lambda_lasso,
  lambda_fused,
  tol_lim = 1e-04,
  max_iter = 50,
  trace = TRUE,
  MSM_profile = FALSE,
  use_parallel = TRUE
)

Arguments

Details

This function fits a penalized Cox type regression model to a multi-state model, where the penalty is a combination of Lasso penalty and Fused penalty. It implements L1 type penalization with the penalty applied to absolute transition-specific effects and pairwise difference between the corresponding transition effects.

Value

A list with elements like matrix of estimated beta coefficients along with standard error and p value,number of iterations, aic (Akaike Information Criterion) value, gcv (GCV criterion) value and df (degrees of freedom)

Author(s)

Atanu Bhattacharjee,Gajendra Kumar Vishwakarma,Abhipsa Tripathy

Examples

##
set.seed(123)
data(msdata_3state)
covs1 <- msdata_3state[,9:10]
flassomsm_pirls(msdata = msdata_3state,X=msdata_3state[,c(11:dim(msdata_3state)[[2]])],
p = length(covs1),lambda_lasso = 0.3,lambda_fused = 0.5,tol_lim = 1e-4,
max_iter = 10, trace = TRUE, MSM_profile = FALSE,use_parallel = TRUE)

# For 2 covariates and 3 number of transitions


# Simulate msdata_4state instead of loading from disk
msdata_4state <- simdata(seed=123,n=1000,dist="weibull",cdist="exponential",
                 cparams=list(rate = 0.1),lambdas=c(0.1, 0.2, 0.3, 0.4),
                 gammas=c(1.5, 2, 2.5, 2.6),beta_list=list(c(-0.05,0.01,0.5,0.6),
                 c(-0.03, 0.02, 0.07, 0.08),c(-0.04, 0.03, 0.04, -0.03),
                 c(-0.05,0.05,0.6,0.8)),cov_means=c(0, 10, 2, 3),cov_sds=c(1,20,5,1.05),
                 trans_list=list(c(2, 3, 4, 5),c(3, 4, 5),c(4, 5), c(5), c()),
                 state_names=c("Tx", "Rec", "Death", "Reldeath", "srv"))

set.seed(123)
sub_msdata_4state <- msdata_4state[msdata_4state$id %in% sample(unique(msdata_4state$id), 10), ]
covs1 <- sub_msdata_4state[,9:10]
flassomsm_pirls(msdata = sub_msdata_4state,X=sub_msdata_4state[,c(13:32)],
              p = length(covs1),lambda_lasso = 0.5,lambda_fused = 0.6, tol_lim = 1e-4,max_iter = 10,
              trace = TRUE, MSM_profile = FALSE,use_parallel = TRUE)

# For 2 covariates and 10 number of transitions
##

Multi state data with high dimensional covariates

Description

Multi state data with 3 states, 100 continuous biomarkers

Usage

data(highDmsm)

Format

A dataframe with 500 rows and 107 columns including all the states and covariates

id

ID of subjects

state1

Time in days from transplantation to state-1 or last follow up

status1

Status of state-1

state2

Time in days from transplantation to state-2 or last follow up

status2

Status of state-1

state3,status3

Time in days from transplant to the respective states along with its status

v1,...,v100

The 100 biomarkers attached to the dataset

Examples

data(highDmsm)

Short Multi state data

Description

Simulated multi-state data with 3 states expanded in msdata format

Usage

data(msdata_3state)

Format

A dataframe with 12 rows and 2 variables expanded in wide format for 3 orders of transitions

id

ID of subjects

from

From which state the individual is shifting

to

To which state the individual is shifting

trans

Order of transition of a particular individual at a specific time

Tstart

Starting time of a transition

Tstop

Stop time of a transition

time

Total time duration of a particular transition

status

Indicator of that particular transition

x1,x2

2 continuous covariates originally in the dataset

x1.1,...,x2.3

The 2 covariates expanded in 3 orders of transitions

Examples

data(msdata_3state)

Multi state data

Description

Simulated multi-state data with 4 states expanded in msdata format

Usage

data(msdata_4state)

Format

A dataframe with 9198 rows and 4 variables expanded in wide format for 10 orders of transitions

id

ID of subjects

from

From which state the individual is shifting

to

To which state the individual is shifting

trans

Order of transition of a particular individual at a specific time

Tstart

Starting time of a transition

Tstop

Stop time of a transition

time

Total time duration of a particular transition

status

Indicator of that particular transition

x1,...,x4

4 continuous covariates originally in the dataset

x1.1,...,x4.10

The 4 covariates expanded in 10 orders of transitions

Examples

data(msdata_4state)

prederr

Description

Evaluates the predictive performance of the multi-state model

Usage

prederr(
  msdata,
  X,
  beta_est,
  times,
  state_of_interest,
  trans_matrix,
  test_fraction = 0.2,
  quick = FALSE,
  verbose = FALSE
)

Arguments

Details

This function is designed to evaluate the predictive performance of a multi-state survival model using Brier score using subject specific covatiates and thier estimated coefficients from a penalized regression model. It also incorporates baseline hazards for each transtion from a Cox model using msfit() for more accurate prediction and also computes the predicted state probabilities with probtrans().

Value

Gives a list of objects like brier score at specified time points, Integrated Brier score,predicted probabilities, their true states and the time points

Author(s)

Atanu Bhattacharjee,Gajendra Kumar Vishwakarma,Abhipsa Tripathy

Examples

##
set.seed(123)
data(msdata_3state)
tmat_3state <- mstate::transMat(x = list(c(2, 3), c(3), c()),
                       names = c("State1", "State2", "State3"))
beta_est1 <- c(0.13, -0.16, -0.12, -0.20, -0.15, -0.55)
prederr(msdata = msdata_3state, X = msdata_3state[, 9:10], beta_est = beta_est1,
times = seq(0.5, 1.5, length.out = 5),state_of_interest = 2,
trans_matrix = tmat_3state,test_fraction = 0.2,quick = TRUE,verbose = TRUE)


data(msdata_4state)
set.seed(123)
sub_msdata_4state <- msdata_4state[msdata_4state$id %in% sample(unique(msdata_4state$id), 10), ]
tmat_4state <- mstate::transMat(x = list(c(2, 3, 4,5), c(3, 4, 5), c(4,5), c(5),c()),
                               names = c("Tx", "Lrc","Fp", "Dp", "srv"))
beta_est1 <- as.numeric (c(0.13,-0.16,-0.12,-0.20,-0.15,-0.55,-0.35,-0.28,-0.34,-0.12))
times1 <- seq(0.5, 1.5, length.out = 5)
prederr(msdata = sub_msdata_4state, X = sub_msdata_4state[,9],
       beta_est = beta_est1,times = times1,state_of_interest = 2,
       trans_matrix = tmat_4state,test_fraction = 0.2,quick = TRUE,verbose = TRUE)

##

simdata

Description

Simulation of a multi-state model with n no. of states.

Usage

simdata(
  seed = 123,
  n = 1000,
  dist = "weibull",
  cdist = "exponential",
  cparams = list(rate = 0.1),
  lambdas,
  gammas,
  beta_list,
  cov_means,
  cov_sds,
  trans_list,
  state_names
)

Arguments

Details

This function is used for simulating a multi-state model with n no. of states and status corresponding to each state along with a number of covariates both continuous or categorical.

Value

a multi-state dataframe with given number of states, corresponding status and the covariate vector

Author(s)

Atanu Bhattacharjee,Gajendra Kumar Vishwakarma,Abhipsa Tripathy

Examples

##
msdata_4state <- simdata(seed=123,n=1000,dist="weibull",cdist="exponential",
                 cparams=list(rate = 0.1),lambdas=c(0.1, 0.2, 0.3, 0.4),
                 gammas=c(1.5, 2, 2.5, 2.6),beta_list=list(c(-0.05, 0.01, 0.5, 0.6),
                 c(-0.03, 0.02, 0.07, 0.08),c(-0.04, 0.03, 0.04, -0.03),
                 c(-0.05, 0.05, 0.6, 0.8)),cov_means=c(0, 10, 2, 3),cov_sds=c(1,20,5,1.05),
                 trans_list=list(c(2, 3, 4, 5),c(3, 4, 5),c(4, 5), c(5), c()),
                 state_names=c("Tx", "Rec", "Death", "Reldeath", "srv"))
##