PSsurvival 0.2.0

Breaking Changes

• API Unification: Changed estimand parameter to weight_method in surveff() and marCoxph() for consistency across all package functions
  • estimand = "ATE" → weight_method = "IPW"
  • estimand = "overlap" → weight_method = "OW"
  • estimand = "ATT" → weight_method = "ATT" (name unchanged, but parameter name changed)
• Changed trim parameter from character to logical in surveff(), marCoxph(), and weightedKM()
  • trim = "symmetric" → trim = TRUE
  • trim = NULL → trim = FALSE
• Removed alpha parameter and asymmetric trimming support from all functions due to poor statistical performance in practice
  • Only symmetric trimming (Crump extension for multiple treatments) is now supported
• Default delta is now NULL (automatic selection based on number of treatment groups) instead of a fixed value

New Features

• New function: weightedKM() for weighted Kaplan-Meier estimation with propensity score weights
  • Supports IPW, overlap weights, ATT, and classical unweighted Kaplan-Meier
  • Uses classic weighted Greenwood variance formula
  • Handles binary and multiple treatment groups
• New S3 methods for weightedKM objects:
  • plot.weightedKM(): Visualize weighted Kaplan-Meier or cumulative risk curves with confidence intervals
  • summary.weightedKM(): Tabular summaries with confidence intervals
  • print.weightedKM(): Formatted console output
• Risk table support in plot.weightedKM():
  • Display number at risk and cumulative events below survival/risk curves
  • Customizable statistics, breaks, height, and font size
  • Automatic group name positioning and formatting
• Enhanced visualization options:
  • Three confidence interval types: plain, log, log-log (default)
  • Cumulative risk plots (1 - survival)
  • Customizable colors, widths, transparency, and legends

Documentation

• Unified weighting method descriptions across all main functions (surveff(), marCoxph(), weightedKM())
• Added comprehensive @details sections explaining IPW, OW, and ATT weighting methods
• Updated all examples to use new API with weight_method parameter

Internal

• Moved ggplot2 and cowplot from Suggests to Imports (required for plotting functions)
• Parameter mapping layer in surveff() and marCoxph() to transform user-facing API to internal implementation

Migration Guide for Existing Users

If upgrading from version 0.1.0:

• Replace estimand = "ATE" with weight_method = "IPW"
• Replace estimand = "overlap" with weight_method = "OW"
• Replace estimand = "ATT" with weight_method = "ATT"
• Replace trim = "symmetric", delta = 0.1 with trim = TRUE, delta = 0.1
• Replace trim = NULL with trim = FALSE (or simply omit, as FALSE is default)
• Remove alpha parameter entirely (asymmetric trimming no longer supported)
• Set delta = NULL to use automatic defaults (0.1 for binary, 0.067 for 3 groups, 1/(2J) for J ≥ 4)

Example migration:

# Old (v0.1.0)
result <- surveff(
  data = mydata,
  ps_formula = Z ~ X1 + X2,
  censoring_formula = Surv(time, event) ~ X1,
  estimand = "overlap",
  censoring_method = "weibull"
)

# New (v0.2.0)
result <- surveff(
  data = mydata,
  ps_formula = Z ~ X1 + X2,
  censoring_formula = Surv(time, event) ~ X1,
  weight_method = "OW",
  censoring_method = "weibull"
)

PSsurvival 0.1.0

• Initial CRAN release

• Documentation improvements for CRAN resubmission:

  • Added @return tags to print methods
  • Removed examples from internal functions
  • Updated examples to use \donttest{} with executable code
  • Enhanced output documentation for all exported functions

• Implements propensity score weighting methods for estimating counterfactual survival functions and marginal hazard ratios in observational studies with time-to-event outcomes

• Main functions:

  • surveff(): Estimates counterfactual survival curves and survival differences over time
  • marCoxph(): Estimates marginal hazard ratios via weighted Cox regression

• Supports binary and multiple (>2) treatment groups

• Weighting methods:

  • Inverse probability weighting (ATE)
  • Treated/target group weighting (ATT)
  • Overlap weighting

• Propensity score trimming:

  • Symmetric trimming (Crump et al.)
  • Asymmetric trimming (Sturmer et al.)

• Censoring adjustment methods:

  • Weibull accelerated failure time models
  • Cox proportional hazards models

• Variance estimation:

  • Analytical M-estimation (binary treatment with Weibull censoring)
  • Bootstrap (all configurations, with full or stratified resampling)

• Comprehensive documentation including vignette with examples

• Based on methods from Cheng et al. (2022) doi:10.1093/aje/kwac043 and Li & Li (2019) doi:10.1214/19-AOAS1282