Demographics and Baseline Characteristics Tables

Introduction

The demographics and baseline characteristics table—often called Table 1—is one of the most recognizable tables in a clinical trial report. It describes who entered the study and what the participants looked like before treatment effects are evaluated.

Typical variables include age, sex, race, ethnicity, height, weight, disease duration, baseline laboratory measurements, disease severity, prior therapies, concomitant conditions, and other clinically relevant characteristics.

In a randomized clinical trial, these characteristics are generally displayed by treatment group so that the reader can assess the descriptive comparability of the randomized groups. In a single-arm study, the same variables can be summarized for the overall analysis population.

What Is a Demographics and Baseline Characteristics Table?

A demographics and baseline characteristics table is a structured summary of participant characteristics measured or known at baseline. The table typically places characteristics in rows and treatment groups in columns.

Why Table 1 Matters

A clinical trial's conclusions are interpreted in the context of the population that was actually enrolled. A treatment effect estimated in a narrowly selected population may not generalize to patients with substantially different baseline characteristics.

Table 1 therefore serves several purposes:

• Describes the enrolled population.
• Shows the distribution of important baseline characteristics across treatment groups.
• Allows readers to understand the clinical context of later analyses.
• Provides transparency about the population included in the analysis.
• Highlights clinically meaningful imbalances that may affect interpretation.

Defining the Baseline Population

Before creating the table, the statistical programmer must establish which analysis population the table describes. Common choices include the randomized population, full analysis set, intent-to- treat population, treated population, or safety population.

The appropriate population depends on the study design and the table's purpose. A demographics table describing randomized participants may differ from a baseline table restricted to treated participants.

The table title, footnotes, and statistical analysis plan should make the population definition unambiguous.

Categorical Variables

Categorical variables are generally summarized using the number of participants and percentage in each category.

For a category \(j\), the percentage is commonly calculated as:

where \(n_j\) is the number of participants in category \(j\), and \(N\) is the denominator specified for that variable.

Example: Sex

The percentage is normally calculated within treatment group. Thus, Drug A percentages use 100 as the denominator and placebo percentages also use 100.

Example: Race

Whether missing values receive their own row or are described in a footnote should be prespecified and applied consistently.

Continuous Variables

Continuous baseline variables require a choice of summary statistics. The most common choices are mean (SD) for approximately symmetric variables and median (Q1, Q3) when the distribution is skewed or when the median provides a more informative description.

Mean and Standard Deviation

For observations \(x_1,\ldots,x_n\), the sample mean is:

The sample standard deviation is:

A table might therefore display age as: 64.2 (8.7), meaning a mean age of 64.2 years and an SD of 8.7 years.

Median and Quartiles

A skewed variable such as disease duration may be more appropriately presented as: 4.8 (2.1, 9.7), representing the median and first and third quartiles.

Baseline Characteristics Should Be Clinically Relevant

Table 1 should not become a repository for every variable collected in the case report form. The characteristics selected should have a clear role in describing the study population or interpreting the treatment comparison.

Typical domains include:

• Demographics
• Anthropometrics
• Relevant medical history
• Disease characteristics and severity
• Prior medications or procedures
• Baseline laboratory or biomarker measurements
• Relevant prognostic factors

Baseline Versus Post-Baseline Variables

A common programming mistake is accidentally including information that was collected after treatment began. A true baseline characteristic should be defined relative to the protocol's baseline time point.

For example, an adverse event occurring after the first dose is not a baseline characteristic, even if the event is recorded early in the study. Similarly, a laboratory value obtained after treatment initiation should not be silently substituted for the prespecified baseline value.

Defining Baseline for Laboratory Values and Measurements

Baseline may be defined as the last nonmissing measurement obtained before the first dose, the assessment closest to randomization, or another protocol-specified time point.

The precise definition should be consistent with the study's SAP and data standards. If multiple measurements qualify, the selection rule should be deterministic and reproducible.

Should Baseline p-Values Be Reported?

This is one of the most important interpretive issues surrounding Table 1. In randomized trials, baseline characteristics are usually presented descriptively rather than subjected to formal hypothesis tests of equality between treatment groups.

Randomization creates the basis for valid treatment comparisons; it does not require every baseline variable to have a small p-value difference. With enough variables, some apparently unusual differences can occur by chance.

Whether baseline p-values or standardized differences are included should follow the statistical analysis plan, study design, and reporting conventions applicable to the trial.

Why P-Values Can Be Misleading in Table 1

A p-value depends on both the magnitude of the difference and the sample size. A small numerical difference can become statistically significant in a very large trial, while a potentially important difference can fail to reach conventional significance in a small trial.

Consequently, the reader should focus on the magnitude and clinical relevance of observed differences rather than treating a baseline p-value as a pass/fail test of randomization.

Standardized Mean Differences

For a continuous variable, a standardized mean difference can be used to describe the magnitude of a difference relative to pooled variability. A common form is:

where the pooled standard deviation can be defined as:

Standardized differences are particularly useful when the objective is to quantify the size of an imbalance rather than ask whether it is statistically significant.

Handling Missing Baseline Data

Missingness is an important part of Table 1 because percentages can be misleading if the denominator is not clear.

Suppose 100 participants are randomized to Drug A, but baseline BMI is available for only 92. The table should make it clear whether the denominator for BMI is 92 rather than 100.

Percentages and Variable-Specific Denominators

A table may contain different denominators for different variables. This is not necessarily an error. It can occur when some participants have missing assessments.

For example, if 100 participants are randomized and 97 have race recorded, the race percentages may use 97 as the denominator if the analysis convention is based on nonmissing observations.

The important requirement is that the denominator rule be prespecified, consistent, and clearly documented.

Nested Categories

Some baseline variables contain hierarchical categories. Disease stage, geographic region, and prior-treatment classifications are common examples.

The table should use indentation or formatting that makes the hierarchy obvious. For example:

Units Matter

Continuous characteristics should include appropriate units in the row label or table heading. This prevents ambiguity and reduces the risk of interpreting the same numeric value in the wrong measurement scale.

Rounding Rules

Rounding should be prespecified or standardized across the table. For example, age might be displayed to one decimal place, while percentages might be displayed to one decimal place.

Avoid inconsistent rounding such as reporting one group to one decimal place and another to two decimal places without a reason.

A Complete Worked Example

Consider a randomized, parallel-group clinical trial with 200 participants: 100 assigned to Drug A and 100 assigned to placebo. The study collects demographic variables, anthropometric measurements, disease duration, disease severity, and selected baseline laboratory values.

This table provides a compact description of the randomized population while preserving the distinction between categorical and continuous variables.

Step 1: Define the Analysis Population

Start by identifying the population that the table is intended to describe. In the worked example, assume the table uses all randomized participants.

These group sizes become the default denominators for characteristics without missing observations.

Step 2: Classify Each Variable

Each characteristic should be classified as categorical or continuous before selecting its summary method.

Step 3: Calculate Categorical Summaries

Suppose 58 of the 100 Drug A participants are male. The displayed percentage is:

Therefore the table displays: 58 (58.0%).

Step 4: Calculate Continuous Summaries

For age, suppose the 100 Drug A participants have a mean of 64.2 years and an SD of 8.7 years. The table displays:

If disease duration is strongly right-skewed, the median and quartiles may be preferable:

Step 5: Verify Denominators

BMI illustrates why denominator checks matter. If only 92 Drug A participants have a baseline BMI, the summary should identify the available sample size.

The programmer should not accidentally display the group-level denominator of 100 as though all 100 participants contributed BMI data.

Step 6: Check Internal Consistency

A final quality-control pass should verify that category counts and percentages agree, totals reconcile, and no impossible or contradictory values are present.

Common Statistical Programming Mistakes

1. Using the wrong analysis population. A table based on treated participants should not accidentally include randomized-but-untreated participants.
2. Using post-baseline values. A measurement collected after treatment initiation can introduce post-treatment information into a baseline table.
3. Using the wrong denominator. Percentages should follow the prespecified denominator convention.
4. Failing to show missing observations. A reader should be able to understand why the available N differs from the treatment-group N.
5. Overloading the table. Include clinically relevant characteristics rather than every collected variable.
6. Inconsistent rounding. Apply a uniform display standard.
7. Automatically testing every baseline variable. A large collection of baseline p-values can encourage inappropriate interpretation of random variation as a treatment-group finding.
8. Ignoring clinically meaningful imbalance. Statistical nonsignificance does not prove that two groups are clinically identical.

Baseline Imbalance Does Not Automatically Mean Randomization Failed

Suppose age is 64 years in one group and 61 years in another. Whether that difference matters depends on the disease, sample size, endpoint, and clinical context.

Randomization does not produce identical groups. It creates treatment assignments according to a randomized mechanism intended to balance prognostic factors on average.

Stratified Randomization and Table 1

If randomization was stratified by factors such as disease stage, geographic region, or prior treatment, those variables may deserve particular attention in the baseline table.

The table should remain consistent with the randomization and analysis strategy. A stratification factor can be balanced by design without necessarily being identical across treatment groups in the realized sample.

Multi-Arm Trials

In a multi-arm trial, each treatment group generally receives its own column, with an overall column when useful.

Single-Arm Studies

A single-arm study does not have randomized treatment groups to compare. The table can instead summarize the overall study population and, when appropriate, clinically meaningful subgroups.

The same principles still apply: define the analysis population, use appropriate descriptive statistics, identify missing data, and make the denominators clear.

Baseline Characteristics in Observational Studies

In observational research, baseline tables often have a different interpretive role because treatment groups are not randomized. Large baseline differences can be directly relevant to confounding and may motivate adjustment, matching, weighting, stratification, or other methods.

The same descriptive table can therefore support very different inferential goals depending on study design.

R Implementation

A reproducible workflow in R should begin with a clearly defined analysis dataset. The following example illustrates the basic descriptive calculations rather than prescribing a specific reporting package.

# Example structure
trial <- data.frame(
  treatment = c("Drug A", "Drug A", "Placebo", "Placebo"),
  age = c(64, 71, 59, 66),
  sex = c("Male", "Female", "Male", "Female"),
  race = c("White", "Asian", "White", "Black")
)

# Group sizes
table(trial$treatment)

# Categorical summary
prop.table(table(trial$sex, trial$treatment), margin = 2)

# Continuous summary
aggregate(age ~ treatment, data = trial,
          FUN = function(x) c(mean = mean(x, na.rm = TRUE),
                               sd = sd(x, na.rm = TRUE)))

In production clinical-trial programming, the analysis dataset should be generated from validated source data and the table program should apply the protocol/SAP definitions exactly.

R Workflow for a Production Table

A robust programming workflow separates data derivation from table presentation:

SAS-Style Programming Logic

A SAS implementation follows the same conceptual sequence: define the population, derive baseline values, obtain treatment-group summaries, and format the results for the clinical table.

/* Example: categorical summary */
proc freq data=adsl;
  tables trt01an*sex / missing;
run;

/* Example: continuous summary */
proc means data=adsl n mean std median q1 q3 min max;
  class trt01an;
  var age;
run;

The exact programming structure depends on the study's data standards, derivations, treatment variables, and table shell.

Table Shell Design

The table shell should be created before final programming whenever possible. A shell establishes the intended order of variables, treatment columns, summary statistics, footnotes, units, and population definition.

Footnotes Are Part of the Statistical Definition

Footnotes should not be treated as decorative text. They often contain critical information needed to interpret the numbers.

Useful footnotes may specify:

• The analysis population.
• The definition of baseline.
• The denominator used for percentages.
• How missing values were handled.
• The meaning of SD, Q1, Q3, or other abbreviations.
• The treatment exposure definition for a safety population.

Quality Control Checklist

Common Mistakes in Interpretation

1. Calling a baseline difference a treatment effect. Baseline characteristics precede treatment outcomes.
2. Assuming p > 0.05 proves balance. Failure to reject a null hypothesis does not prove equality.
3. Assuming p < 0.05 proves randomization failed. Some differences occur by chance, particularly when many variables are examined.
4. Ignoring clinical magnitude. A statistically significant difference may be clinically trivial.
5. Ignoring missingness. Different variable-specific Ns can affect interpretation.

Practical Table 1 Workflow

What Should Be Specified in the Statistical Analysis Plan?

The SAP should provide enough information for the table to be reproduced without relying on undocumented programming decisions.

• Analysis population
• Definition of baseline
• Variables to be summarized
• Categories for categorical variables
• Summary statistics for continuous variables
• Handling of missing observations
• Denominator conventions
• Rounding and display precision
• Whether inferential comparisons or standardized differences are presented

Worked Example Summary

The Most Important Concept

A demographics and baseline characteristics table is not merely a formatting exercise. It is a statistical description of the population that provides the context for interpreting the clinical trial.

The most important principles are:

• Define the population before calculating summaries.
• Define baseline before extracting values.
• Use summaries appropriate to the variable.
• Make denominators and missing observations transparent.
• Focus on clinical relevance rather than mechanical p-value screening.
• Make the programming and presentation reproducible.

References

ICH E9. Statistical Principles for Clinical Trials.
ICH E3. Structure and Content of Clinical Study Reports.
CONSORT. CONSORT Statement: Updated Guidelines for Reporting Parallel Group Randomised Trials.
Senn, S. Testing for baseline balance in clinical trials.
Austin, P.C. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples.