CIMEHR 0.2.0 (2026-10-21)

Rewrite of the package. Most functions from 0.1.0 still exist under the same names, but their arguments and return values have changed, so code written for 0.1.0 needs to be updated.

What changed for users

• Covariates are given by column name, not by formula. Every model function now takes the same arguments, such as visit_covs, long_covs, obs_fixed_covs, obs_random_covs, outcome_random_covs and censor_col, instead of formulas such as fixed = Y ~ 1 or formula = ....
• One fitted-model class. Every method returns a cimehr_fit object with the same print(), summary(), coef(), vcov(), confint(), predict() and plot() methods, so results from different methods can be compared directly.
• One entry point. fit_method(method, data, ...) fits any method by its key; cimehr_methods() and available_comparison_methods() list the keys, and method_comparisons() fits several methods to the same data.

New functions

• CIMEHR_HL(): CIMEHR with a higher-order Laplace correction.
• CIMEHR_spline() and CIMEHR_spline_HL(): CIMEHR with a B-spline baseline mean curve.
• LMM(), Liang(), IIRR() and IIRR_balancing(): short names for Linear_mixed_model(), Joint_modeling_visiting_and_longitudinal_Liang(), Inverse_intensity_rate_ratio_weighting() and Inverse_intensity_rate_ratio_balancing() (the long names still work).
• Lin_Ying() and JMVL_G(): two further comparison methods.
• fit_method() and cimehr_methods(): common interface and method list.
• meanfit(), baseline_mean() and population_trajectory(): fitted mean curves, with analytical or participant-bootstrap pointwise confidence intervals and print(), summary(), plot() and as.data.frame() methods.
• coef_stage(): coefficients for one stage (visiting, recording or outcome).

New and extended methods

• Visit-process and marker-process versions of Lin-Ying, Liang, JMVL-G and IIRR (weighting and balancing), selected with the process argument or by method key.
• Observation-count and visit-count adjusted LMMs (keys oa_lmm, va_lmm).
• Multiple imputation (Multiple_imputation_IP()) and linear increments (Linear_increment_IP()) can now be combined with LMM, Liang or IIRR.
• vcov(fit, all = TRUE) and confint() give stacked inference across all three CIMEHR stages.
• Identification and numerical diagnostics are attached to CIMEHR fits and reported as warnings when a design is weakly identified.

Performance

• The visiting, recording and outcome stages run in compiled C++ code (via ‘Rcpp’). The bivariate normal CDF is computed internally with Alan Genz’s routines (see inst/COPYRIGHTS), replacing the ‘pbivnorm’ dependency.

Data, documentation and examples

• New example dataset cimehr_example (1,000 simulated participants), replacing sim_ehr_data. The script that generates it is installed as system.file("examples", "simulate_example.R", package = "CIMEHR").
• New vignette vignette("cimehr-workflow") walking through a full analysis.
• Every exported function has a help page and a short runnable example.

Removed

• CIMEHR_timevarying_ou() and CIMEHR_timevarying_integral().
• bootstrap(): participant-bootstrap intervals for mean curves are now available through meanfit(..., bootstrap = TRUE).
• sim_data_gen() and the sim_ehr_data dataset (see cimehr_example).
• Linear_increment() and Multiple_imputation() (use Linear_increment_IP() and Multiple_imputation_IP()).

Dependencies and infrastructure

• Requires R >= 4.1.0 and a C++17 compiler.
• No longer depends on ‘MASS’, ‘nleqslv’, ‘pbivnorm’, ‘numDeriv’, ‘data.table’ or ‘slim’. ‘mice’ and ‘pan’ are now suggested and only needed for multiple imputation; ‘ggplot2’ is suggested for plotting.
• NAMESPACE and help pages are written by hand (not generated by ‘roxygen2’), and a Collate field in DESCRIPTION sets the order in which the R files load.

CIMEHR 0.1.0 (2026-03-04)

• First version released on GitHub (https://github.com/ysph-dsde/CIMEHR).