1- More Than Just Code

Welcome to the team dear Nouran. Before we write a single function in R or open a dataset, you must understand the weight of your role here.

In our CRO (Contract Research Organization), we do not just process numbers; we produce evidence. The data we analyze determines whether a new drug is safe for a human being to take, or whether a treatment actually works.

If our code is buggy, a dangerous drug might be approved. If our data management is messy, a life-saving treatment might be rejected because the results "look" unreliable.

This lesson covers the "Rules of the Road", the international standards (ICH-GCP) that govern every clinical trial in the world, whether in Egypt, Europe, or Japan.

2. The International Standard: ICH-GCP

The pharmaceutical industry runs on a set of standards called ICH-GCP (International Council for Harmonisation - Good Clinical Practice).

As a biostatistician, you are primarily responsible for adhering to ICH E9: Statistical Principles for Clinical Trials.

The Three Pillars of ICH E9 for Programmers:

A. Bias Control (The "Fairness" Rule)

Bias is any systematic error that makes the drug look better (or worse) than it really is.

The Rule: We must analyze the data exactly as we promised we would in the Protocol and Statistical Analysis Plan (SAP).
In Practice: You cannot try three different statistical tests and pick the one that gives a p-value < 0.05. This is called "P-hacking" and it is unethical. We execute the SAP, regardless of the result.

B. Pre-Specification

Analysis must be defined before we see the unblinded data.

The Rule: The SAP is signed off before Database Lock (DBL), in the GCC, this is rare
In Practice: Once the database is locked and unblinded, you cannot change your code logic to "fix" a result. Any changes after unblinding must be documented as "Post-Hoc" analyses.

C. Analysis Populations

You will encounter three standard populations in almost every study. You must know how to flag them in your code:

Safety Population: Everyone who took at least one dose of the drug. (Used for Adverse Events).
Intent-to-Treat (ITT): Everyone who was randomized, regardless of whether they finished the study. (Used for Efficacy—to preserve randomization).
Per-Protocol (PP): Only those who followed the rules perfectly (no missed visits, no wrong pills).

3. Data Integrity: The ALCOA+ Standard

Regulators and auditors trust our results only if the data follows the ALCOA+ principles. Here is how that applies to your R programming:

Principle	Meaning	How we apply this in R
Attributable	Who did it?	Your script header must have your name. Your Git commit must use your account.
Legible	Can it be read?	Your code must be clean, commented, and indented. "Spaghetti code" is a compliance risk.
Concurrent	When did it happen?	We do not back-date work. If you ran the analysis today, the log file must show today's date.
Original	Is it the source?	We never manually edit Raw Data (CSVs/Excel) to fix errors. We fix errors via script logic or Data Management queries.
Accurate	Is it correct?	This is why we do Peer Review and Double Programming (QC).

4. Ethics & Privacy: Protecting the Patient

We deal with human lives. Even though we only see rows in a dataframe, those rows are people.

4.1 Anonymization

The Rule: You must never see a patient's name, address, or national ID.
The Identifier: We use the USUBJID (Unique Subject Identifier). It usually looks like ABC-101-001 (Protocol - Site - Subject).
Action Item: If you ever receive a file via email that contains a patient's real name, delete it immediately and notify the Project Manager. Do not save it. Do not analyze it.

4.2 The "Blind"

In many studies, neither the doctor nor the patient knows who is getting the real drug and who is getting the Placebo.

The Danger Zone: As a statistician, you might have access to the randomization codes (Treatment A vs Treatment B).
The Golden Rule: You never merge the treatment variable (ARM or TRT01P) into the main dataset until the study is officially "Unblinded" by the Project Manager.
Why? If the team sees that "Treatment A" is working better while the study is still running, they might subconsciously treat those patients differently, ruining the scientific validity of the trial.

5. All in one

We follow the SAP, not our curiosity. Do not change the analysis just to get a "better" p-value.
Code is Evidence. Write your R scripts assuming an auditor is standing behind your shoulder reading them.
Raw Data is Sacred. Never change the source file. All data cleaning must happen transparently within your R script.
Protect the Blind. Do not reveal which patient is on which drug until you are authorized.

Homework Task for Lesson 1

Open our internal server and locate the folder/resources below
Read ICH E9 Section 5 (Data Analysis).
Write a brief email to me at m.w@mars-global.co.uk summarizing the difference between the "Intention-to-Treat" principle and the "Per-Protocol" set in your own words.

End of Lesson Text

Lesson 1.1: The Biostatistician’s Responsibility – Global Standards & Ethics

You must be logged in to access this course

Module 1: Global Standards & Ethics (The "Rules")