Clinical Research Examples with jsurvival

Overview

This vignette provides practical examples of using jsurvival for common clinical research scenarios. Each example demonstrates a complete workflow from data preparation to interpretation of results.

Example 1: Cancer Treatment Comparison Study

Background

A clinical trial comparing two cancer treatments (Treatment A vs. Treatment B) with overall survival as the primary endpoint.

Research Question

Is there a significant difference in overall survival between Treatment A and Treatment B?

Data Structure

# Example data structure
cancer_data <- data.frame(
  PatientID = 1:200,
  Treatment = rep(c("Treatment A", "Treatment B"), each = 100),
  OS_months = c(rnorm(100, 24, 8), rnorm(100, 30, 10)),
  Death = rbinom(200, 1, 0.6),
  Age = rnorm(200, 65, 10),
  Stage = sample(c("I", "II", "III", "IV"), 200, replace = TRUE),
  ECOG = sample(0:2, 200, replace = TRUE)
)

Analysis Steps

Step 1: Single Arm Analysis (Overall Population)

# First, examine overall survival characteristics
singlearm_result <- singlearm(
  data = cancer_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  cutp = "12, 24, 36",
  timetypeoutput = "months"
)

Interpretation: - Median overall survival for the entire cohort - 1, 2, and 3-year survival rates - 95% confidence intervals for all estimates

Step 2: Treatment Comparison

# Compare survival between treatments
survival_result <- survival(
  data = cancer_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  explanatory = Treatment,
  analysistype = "overall",
  ph_cox = TRUE,
  ci95 = TRUE,
  risktable = TRUE
)

Key Outputs: - Log-rank test p-value for treatment difference - Hazard ratio comparing Treatment B vs. Treatment A - Kaplan-Meier curves with risk tables - Median survival times by treatment group

Step 3: Multivariable Analysis

# Adjust for potential confounders
multi_result <- multisurvival(
  data = cancer_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  explanatory = c(Treatment, Age, Stage, ECOG)
)

Interpretation: - Adjusted hazard ratio for treatment effect - Independent prognostic factors - Model fit statistics (concordance index)

Clinical Interpretation

If HR = 0.75 (95% CI: 0.60-0.95), p = 0.02: - Treatment B reduces the risk of death by 25% compared to Treatment A - This result is statistically significant - The confidence interval excludes 1.0, supporting the treatment benefit

Example 2: Biomarker Cut-point Analysis

Background

Investigating whether a continuous biomarker can predict patient outcomes and finding the optimal threshold for clinical decision-making.

Research Question

What is the optimal cut-point for Biomarker X to predict overall survival?

Analysis Workflow

Step 1: Continuous Variable Analysis

# Find optimal cut-point for biomarker
biomarker_result <- survivalcont(
  data = biomarker_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  contexpl = Biomarker_X,
  findcut = TRUE,
  cutp = "12, 36, 60"
)

Outputs: - Optimal cut-point value - Hazard ratio for high vs. low biomarker groups - ROC analysis for cut-point validation - Sensitivity and specificity at optimal cut-point

Step 2: Validation Analysis

# Validate cut-point using dichotomized variable
# (Assuming cut-point = 15.2 was identified)
biomarker_data$Biomarker_High <- ifelse(biomarker_data$Biomarker_X >= 15.2, "High", "Low")

validation_result <- survival(
  data = biomarker_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  explanatory = Biomarker_High,
  ph_cox = TRUE
)

Clinical Translation

If optimal cut-point = 15.2 ng/mL, HR = 2.1 (95% CI: 1.4-3.2): - Patients with biomarker ≥15.2 ng/mL have 2.1 times higher risk of death - This threshold could be used for treatment stratification - Consider validation in independent cohort

Example 3: Time-to-Event Analysis with Competing Risks

Background

Studying cancer-specific mortality while accounting for deaths from other causes.

Research Question

What factors predict cancer-specific death, considering other-cause mortality as a competing risk?

Analysis Approach

# For competing risks, use cause-specific outcomes
# Death_cancer: 1 = cancer death, 0 = alive or other death
# Death_other: 1 = other death, 0 = alive or cancer death

# Cancer-specific survival analysis
cancer_specific <- survival(
  data = competing_data,
  elapsedtime = OS_months,
  outcome = Death_cancer,
  outcomeLevel = "1",
  explanatory = Treatment,
  analysistype = "overall",
  ph_cox = TRUE
)

# Other-cause mortality analysis
other_cause <- survival(
  data = competing_data,
  elapsedtime = OS_months,
  outcome = Death_other,
  outcomeLevel = "1",
  explanatory = Treatment,
  analysistype = "overall",
  ph_cox = TRUE
)

Interpretation: - Separate hazard ratios for cancer-specific and other-cause mortality - Treatment effects may differ between competing outcomes - Consider cumulative incidence functions for formal competing risk analysis

Example 4: Landmark Analysis

Background

Addressing guarantee-time bias when analyzing post-treatment biomarkers or response markers.

Research Question

Does tumor response at 3 months predict long-term survival?

Analysis Strategy

# Create landmark dataset (patients alive at 3 months)
landmark_data <- subset(original_data, OS_months > 3 | (OS_months <= 3 & Death == 0))

# Adjust survival times for landmark point
landmark_data$OS_months_landmark <- landmark_data$OS_months - 3
landmark_data$OS_months_landmark[landmark_data$OS_months_landmark < 0] <- 0

# Analyze survival from landmark time
landmark_result <- survival(
  data = landmark_data,
  elapsedtime = OS_months_landmark,
  outcome = Death,
  outcomeLevel = "1",
  explanatory = Response_3month,
  uselandmark = TRUE,
  landmark = 3
)

Key Points: - Only includes patients who survived to landmark time - Eliminates guarantee-time bias - Useful for post-treatment predictors

Example 5: Dose-Response Analysis

Background

Evaluating survival outcomes across different dose levels of a treatment.

Research Question

Is there a dose-response relationship for treatment efficacy?

Analysis Steps

Step 1: Categorical Dose Analysis

# Treat dose as categorical variable
dose_categorical <- survival(
  data = dose_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  explanatory = Dose_category,  # Low, Medium, High
  analysistype = "pairwise",
  padjustmethod = "holm"
)

Step 2: Trend Analysis

# Test for linear trend across dose levels
# Create ordinal dose variable (1, 2, 3)
dose_data$Dose_ordinal <- as.numeric(factor(dose_data$Dose_category, 
                                           levels = c("Low", "Medium", "High")))

trend_result <- survival(
  data = dose_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  explanatory = Dose_ordinal,
  ph_cox = TRUE
)

Interpretation: - Test for overall dose effect (3-group comparison) - Test for linear trend (ordinal dose variable) - HR per dose level increase

Example 6: Propensity Score Matching

Background

Comparing treatments in observational data while controlling for selection bias.

Research Question

What is the treatment effect after accounting for baseline differences between groups?

Analysis Workflow

# Step 1: Create matched dataset (outside jsurvival)
# matched_data <- created using propensity score matching

# Step 2: Analyze matched cohort
matched_result <- survival(
  data = matched_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  explanatory = Treatment,
  analysistype = "overall",
  ph_cox = TRUE
)

# Step 3: Sensitivity analysis with additional adjustment
sensitivity_result <- multisurvival(
  data = matched_data,
  elapsedtime = OS_months,
  outcome = Death,
  outcomeLevel = "1",
  explanatory = c(Treatment, remaining_confounders)
)

Reporting Guidelines

Essential Elements

Study Design: Clearly describe cohort and follow-up
Sample Size: Total patients, events, median follow-up
Statistical Methods: Analysis approach and software used
Results Presentation:
- Median survival with 95% CI
- Hazard ratios with 95% CI
- P-values and statistical tests used
- Kaplan-Meier curves with risk tables

Example Results Section

“Among 200 patients (Treatment A: n=100, Treatment B: n=100), there were 120 deaths during a median follow-up of 18.5 months (IQR: 12.2-24.8). Median overall survival was 22.1 months (95% CI: 18.5-25.8) for Treatment A versus 28.4 months (95% CI: 24.1-32.7) for Treatment B. Treatment B was associated with a 25% reduction in the risk of death compared to Treatment A (HR=0.75, 95% CI: 0.60-0.95, p=0.02). The 2-year survival rates were 45% (95% CI: 35-55%) and 60% (95% CI: 50-70%) for Treatment A and B, respectively.”

Figure Legends

“Figure 1. Kaplan-Meier survival curves comparing Treatment A versus Treatment B. The table below shows the number of patients at risk at each time point. P-value from log-rank test. HR: hazard ratio, CI: confidence interval.”

Common Pitfalls and Solutions

Pitfall 1: Immortal Time Bias

Problem: Including post-baseline events as baseline predictors Solution: Use landmark analysis or time-dependent covariates

Pitfall 2: Multiple Testing

Problem: Testing multiple endpoints without adjustment Solution: Use appropriate p-value correction methods

Pitfall 3: Proportional Hazards Violation

Problem: HR changes over time Solution: Use stratified Cox models or time-dependent effects

Pitfall 4: Insufficient Follow-up

Problem: High censoring rates affect reliability Solution: Report median follow-up and censoring patterns

Conclusion

These examples demonstrate the versatility of jsurvival for clinical research applications. The key to successful survival analysis is:

Clear research questions with appropriate study design
Careful data preparation and quality control
Appropriate statistical methods for the research context
Comprehensive reporting following established guidelines
Clinical interpretation that considers both statistical and practical significance

For additional support and examples, visit the jsurvival website and ClinicoPath documentation.

ClinicoPath Development Team

2025-06-30

Overview

Example 1: Cancer Treatment Comparison Study

Background

Research Question

Data Structure

Analysis Steps

Step 1: Single Arm Analysis (Overall Population)

Step 2: Treatment Comparison

Step 3: Multivariable Analysis

Clinical Interpretation

Example 2: Biomarker Cut-point Analysis

Background

Research Question

Analysis Workflow

Step 1: Continuous Variable Analysis

Step 2: Validation Analysis

Clinical Translation

Example 3: Time-to-Event Analysis with Competing Risks

Background

Research Question

Analysis Approach

Example 4: Landmark Analysis

Background

Research Question

Analysis Strategy

Example 5: Dose-Response Analysis

Background

Research Question

Analysis Steps

Step 1: Categorical Dose Analysis

Step 2: Trend Analysis

Example 6: Propensity Score Matching

Background

Research Question

Analysis Workflow

Reporting Guidelines

Essential Elements

Example Results Section

Figure Legends

Common Pitfalls and Solutions

Pitfall 1: Immortal Time Bias

Pitfall 2: Multiple Testing

Pitfall 3: Proportional Hazards Violation

Pitfall 4: Insufficient Follow-up

Conclusion