Skip to contents

Survival and Statistical Analysis Workflows

Adam Mattsson

2026-04-28

survival_and_statistical_analysis_workflows.Rmd

Introduction

This vignette demonstrates how to perform robust survival and statistical analyses using LundTaxR, focusing on two key functions: get_glm and get_survival. These functions streamline the process of running generalized linear models and Cox proportional hazards models using your molecular subtype and signature score data, integrating seamlessly with your sample metadata. Both functions handle common preprocessing steps internally, such as binning numeric variables and wrangling prediction outputs, so you can focus on interpreting results.

1. Generalized Linear Model Analysis with get_glm

The get_glm function enables you to test associations between molecular signature scores (or other predictors) and categorical variables of interest (such as response, grade, gender, or region). It supports both Mann-Whitney tests and generalized linear regression, and can be customized to focus on specific subtypes or predictor sets.

Example 1: Testing Signature Scores by Gender 

results <- classify_samples(
  this_data = sjodahl_2017,
  include_data = TRUE,
  verbose = FALSE
)

# Run GLM comparing signature scores by gender
glm_results_gender <- get_glm(
  these_predictions = results,
  these_samples_metadata = sjodahl_2017_meta,
  subtype_class = "5_class",
  categorical_factor = "gender"
)
knitr::kable(head(glm_results_gender))
score p_value odds_ratio conf_2.5 conf_97.5 variable subtype
proliferation_score 0.2876215 0.9349577 0.8038649 1.0892350 gender all
progression_score 0.4406753 0.9378858 0.8001656 1.0980705 gender all
stromal141_up 0.5364414 0.9561336 0.8295723 1.1014490 gender all
immune141_up 0.0869924 0.8786143 0.7616828 1.0100169 gender all
b_cells 0.6963593 0.9587575 0.8286159 1.1131523 gender all
t_cells 0.0347652 0.8593952 0.7399484 0.9953762 gender all

Example 2: Restricting to a Specific Subtype (e.g., GU) 

glm_results_gu <- get_glm(
  these_predictions = results,
  these_samples_metadata = sjodahl_2017_meta,
  subtype_class = "5_class",
  this_subtype = "GU",
  categorical_factor = "gender"
)
knitr::kable(head(glm_results_gu))
score p_value odds_ratio conf_2.5 conf_97.5 variable subtype
proliferation_score 0.2100151 1.3911470 0.8355183 2.4605491 gender GU
progression_score 0.6787651 1.1529189 0.6620496 2.0649017 gender GU
stromal141_up 0.1809123 0.7720693 0.5113234 1.1230869 gender GU
immune141_up 0.1157470 0.7355648 0.4722463 1.0831048 gender GU
b_cells 0.0411555 0.7652588 0.5496156 1.0421317 gender GU
t_cells 0.0190872 0.6587721 0.4249717 0.9650018 gender GU

Example 3: Custom Predictor Columns and Scaling 

glm_results_custom <- get_glm(
  these_predictions = results,
  these_samples_metadata = sjodahl_2017_meta,
  subtype_class = "5_class",
  categorical_factor = "gender",
  predictor_columns = c("b_cells", "proliferation_score"),
  scale = 100
)
knitr::kable(head(glm_results_custom))
score p_value odds_ratio conf_2.5 conf_97.5 variable subtype
b_cells 0.6963593 0.9587575 0.8286159 1.113152 gender all
proliferation_score 0.2876215 0.9349577 0.8038649 1.089235 gender all

2. Survival Analysis with get_survival

The get_survival function provides a convenient interface for running Cox proportional hazards models, allowing you to estimate hazard ratios for molecular signatures or subtypes with respect to clinical outcomes. The function expects survival time and event columns in your metadata, and can be customized for different subtypes or predictor sets.

Example 1: Cox Model for Overall Survival 

surv_results_os <- get_survival(
  these_predictions = results,
  these_samples_metadata = sjodahl_2017_meta,
  subtype_class = "5_class",
  surv_time = "surv_os_time",
  surv_event = "surv_os_event"
)
knitr::kable(head(surv_results_os))
score p_value hazard_ratio hazard_conf_2.5 hazard_conf_97.5 subtype
proliferation_score 0.8545504 1.0081948 0.9239491 1.100122 all
progression_score 0.9255769 0.9958385 0.9124072 1.086899 all
stromal141_up 0.0950648 1.0719405 0.9879748 1.163042 all
immune141_up 0.7804632 0.9886373 0.9122971 1.071365 all
b_cells 0.2217137 0.9464709 0.8665334 1.033783 all
t_cells 0.1117570 0.9315574 0.8536170 1.016614 all

Example 2: Subtype-Specific Survival Analysis (e.g., Uro) 

surv_results_uro <- get_survival(
  these_predictions = results,
  these_samples_metadata = sjodahl_2017_meta,
  subtype_class = "5_class",
  this_subtype = "Uro",
  surv_time = "surv_os_time",
  surv_event = "surv_os_event"
)
knitr::kable(head(surv_results_uro))
score p_value hazard_ratio hazard_conf_2.5 hazard_conf_97.5 subtype
proliferation_score 0.8920954 0.9871613 0.8190416 1.189790 Uro
progression_score 0.7033291 1.0365889 0.8615622 1.247172 Uro
stromal141_up 0.1893854 1.1023689 0.9530485 1.275084 Uro
immune141_up 0.8553808 0.9869615 0.8570494 1.136566 Uro
b_cells 0.6284895 0.9635444 0.8289731 1.119961 Uro
t_cells 0.2289869 0.9100363 0.7804668 1.061116 Uro

Example 3: Custom Predictor Columns and Binning 

surv_results_custom <- get_survival(
  these_predictions = results,
  these_samples_metadata = sjodahl_2017_meta,
  subtype_class = "5_class",
  surv_time = "surv_os_time",
  surv_event = "surv_os_event",
  predictor_columns = "proliferation_score",
  n_bins = 5
)
knitr::kable(head(surv_results_custom))
score p_value hazard_ratio hazard_conf_2.5 hazard_conf_97.5 subtype
proliferation_score 0.9871195 1.001409 0.844044 1.188114 all

By leveraging get_glm and get_survival, you can efficiently perform statistical and survival analyses on your LundTaxR classification results, supporting robust clinical and translational research workflows.