Clinical Trial Safety Analysis

Survival analysis and safety reporting on a public clinical trial dataset using CDISC SDTM/ADaM standards.

Overview

This project analyzes the CDISC Pilot Study (CDISCPILOT01), a publicly released, CDISC-standardized dataset built to mirror real clinical trial data structures. The (fictional) trial randomized 254 patients with mild-to-moderate Alzheimer's disease to Placebo, Xanomeline Low Dose, or Xanomeline High Dose (transdermal patch).

The analysis follows a real pharmaceutical statistical programming/biostatistics workflow: baseline demographics, adverse event summarization, and time-to-event survival analysis, using ADaM analysis-ready datasets derived from SDTM source data.

Data source: CDISC/PHUSE SDTM-ADaM Pilot Project (public, synthetic data, no real patient information)

Key Findings

• Adverse events occurred in 80.2% of Placebo patients, compared with 91.7% (Low Dose) and 94.0% (High Dose), demonstrating a clear dose-response pattern.
• The adverse events driving this pattern were predominantly dermatologic (pruritus, application-site erythema, rash), consistent with a transdermal patch delivery mechanism. Dizziness also remained significant after formal testing, however, suggesting an additional systemic effect.
• Time-to-first-dermatologic-event analysis (Kaplan-Meier, log-rank p < 0.0001) demonstrated significantly higher hazard in both dose arms relative to placebo (HR = 4.15, Low Dose; HR = 5.03, High Dose), with 95% confidence intervals entirely above 1.
• The proportional hazards assumption was satisfied (cox.zph p = 0.59), supporting the validity of the Cox model.
• An age-adjusted sensitivity analysis showed hazard ratios remained essentially unchanged (4.22 / 5.08), and age was not a significant predictor (p = 0.133), indicating that the dose-response signal is not confounded by age.

Repo Structure

├── data/                  # SDTM/ADaM .xpt files
├── pipeline.Rmd           # Full exploratory analysis: every step, check, and sanity test
├── report.Rmd             # Report write-up: narrative and key results
└── README.md

• pipeline.Rmd contains the complete exploratory analysis, including every data check, intermediate step, and model diagnostic, with all code visible. This document provides a full, transparent record of the analytical process.
• report.Rmd presents the formatted write-up: narrative interpretation alongside the key tables, plots, and models, with routine data-wrangling code hidden and analysis code visible.

Methods

Section	Approach
Baseline Demographics	Summary table (N, age, % female, % white) by treatment arm, from ADSL
Adverse Event Summary	Overall AE incidence by arm; top adverse events by MedDRA preferred term (ADAE)
Time-to-Event Analysis	Kaplan-Meier curves, log-rank test, Cox proportional hazards model, proportional hazards diagnostic, age-adjusted sensitivity analysis (ADTTE)

Tools

R · haven · dplyr / tidyr (data wrangling) · survival (Kaplan-Meier, Cox PH) · survminer (survival plots)

Reproducing

1. Download the three ADaM datasets from the CDISC pilot repo (adsl.xpt, adae.xpt, adtte.xpt) into a local data/ folder.
2. Install required packages: install.packages(c("haven", "dplyr", "tidyr", "survival", "survminer"))
3. Knit either .Rmd file in RStudio.

Note on data

All data used is public, synthetic, and released by CDISC specifically for training/demonstration purposes. No real patient data is involved.