calc_likes {psm3mkv}

Calculate likelihoods for three three-state model structures

Description

Calculate likelihood values and other summary output for the following three state models structures: partitioned survival, clock forward state transition, and clock reset state transition. The function requires appropriately formatted patient-level data, a set of fitted survival regressions, and the time cut-off (if two-piece modeling is used).

Usage

calc_likes(ptdata, dpam, cuttime = 0)

Arguments

ptdata

Dataset of patient level data. Must be a tibble with columns named:

• ptid: patient identifier

• pfs.durn: duration of PFS from baseline

• pfs.flag: event flag for PFS (=1 if progression or death occurred, 0 for censoring)

• os.durn: duration of OS from baseline

• os.flag: event flag for OS (=1 if death occurred, 0 for censoring)

• ttp.durn: duration of TTP from baseline (usually should be equal to pfs.durn)

• ttp.flag: event flag for TTP (=1 if progression occurred, 0 for censoring).

Survival data for all other endpoints (time to progression, pre-progression death, post-progression survival) are derived from PFS and OS.

dpam

List of survival regressions for each endpoint:

• pre-progression death (PPD)

• time to progression (TTP)

• progression-free survival (PFS)

• overall survival (OS)

• post-progression survival clock forward (PPS-CF) and

• post-progression survival clock reset (PPS-CR).

cuttime

Time cutoff - this is nonzero for two-piece models.

Value

A list of three tibbles: all is a tibble of results for all patients:

• methname: the model structure or method.

• npar: is the number of parameters used by that method.

• npts_1 to npts_4 are the number of patients experiencing outcomes 1-4 respectively (see below), and npts_tot the total.

• ll_1 to ll_4 are the log-likelihood values for patients experiencing outcomes 1-4 respectively (see below), and ll_tot the total. valid is a tibble of the same design as all but only in patients with valid likelihoods for all 4 methods sum is a tibble in respect of patients with valid likelihoods for all 4 methods providing:

• npts: number of patients contributing results for this method.

• npar: number of parameters used by that method.

• ll: total log-likelihood

• AIC: Akaike Information Criterion value for this model

• BIC: Bayesian Information Criterion value for this model

The four outcomes are as follows:

• (1) refers to patients who remain alive and progression-free during the follow-up;

• (2) refers to patients who die without prior progression during the follow-up;

• (3) refers to patients who progress and then remain alive for the remaining follow-up, and

• (4) refers to patients who progress and die within the follow-up.

Examples

bosonc <- create_dummydata("flexbosms")
fits <- fit_ends_mods_spl(bosonc)
# Pick out best distribution according to min AIC
params <- list(
  ppd = find_bestfit(fits$ppd, "aic")$fit,
  ttp = find_bestfit(fits$ttp, "aic")$fit,
  pfs = find_bestfit(fits$pfs, "aic")$fit,
  os = find_bestfit(fits$os, "aic")$fit,
  pps_cf = find_bestfit(fits$pps_cf, "aic")$fit,
  pps_cr = find_bestfit(fits$pps_cr, "aic")$fit
  )
calc_likes(bosonc, dpam=params)