Baseline in Clinical Trials: Definition, Baseline Characteristics and Why Baseline Data Matter

Where you learn this: the baseline visit, eligibility confirmation and source documentation are a full section of the VIARES Clinical Study Coordinator II course.

What is a baseline in clinical trials? Definition of baseline

The NCI dictionary defines a baseline, in medical terms, as an initial measurement of a condition taken at an early time point and used for comparison over time. Its example: tumor size is measured before treatment and again afterwards to see whether the treatment had an effect.

In clinical research, baseline means the same thing at the level of a whole trial. Every participant has baseline measurements, and every later result is read against them. A baseline blood test (“baseline bloods”) is a common example: lab values taken before the first dose, so later changes can be detected.

When is baseline measured? Screening, baseline visit and run-in

Baseline is a time point set by the trial protocol. In many trials it is the baseline visit, also called Day 1 or the randomization visit. Screening comes first, when the participant signs informed consent and the site checks whether they qualify.

Some trials add a run-in period between screening and randomization. ICH E9 notes that documentation of subjects who entered such a run-in may be useful. Which value counts as “the” baseline is defined in the protocol or the statistical analysis plan.

Types of baseline data

ICH E9 says that at least demographic and baseline data on disease status should be collected whenever possible. Typical baseline data:

• ✓ Demographics: age, sex, ethnicity, height, weight and body mass index.
• ✓ Medical history: diagnoses, surgeries and the history of the condition under study.
• ✓ Concomitant medications: everything the participant takes, which also helps separate new events from pre-existing ones.
• ✓ Vital signs and labs: blood pressure, heart rate, baseline blood tests, ECG.
• ✓ Disease severity: a baseline score on the scale used for the outcome measure, such as a pain or symptom score, or tumor measurements.

Why baseline matters in clinical research

• ✓ Reference for efficacy: outcomes such as mean change from baseline only make sense against a starting value. ICH E9 asks that derived variables like change from baseline be justified in the protocol.
• ✓ Eligibility: baseline findings confirm that the participant meets the eligibility criteria before randomization.
• ✓ Safety reference: ICH E9 describes “treatment-emergent” adverse events as events that emerge or worsen relative to pretreatment baseline.
• ✓ Statistical analysis: prognostic baseline variables can be used to stratify randomization and as covariates, a core biostatistics tool that improves precision.

Baseline characteristics in research: the “Table 1”

Almost every trial report has a table of baseline characteristics, often the first table. CONSORT 2025 (item 25) asks for a table showing baseline demographic and clinical characteristics for each group. It tells readers who was actually enrolled and how relevant the results are to their own patients.

Look at the severity score: a 3.5-point gap. Is that a problem? The next section explains how to judge it.

Baseline comparability between groups

In a randomized controlled trial, the experimental group and the control group should differ only in the treatment. Randomization makes the treatment groups comparable on average, for measured and, because allocation is left to chance, also unmeasured baseline characteristics. That is why it protects against a confounding variable nobody thought of.

On average is not always. The EMA covariate guideline notes that imbalances between groups at baseline are not unusual in practice. ICH E9 adds that stratification has greater potential benefit in small trials.

Why aren’t statistical tests recommended to assess baseline comparability?

The CONSORT explanation paper puts it simply: proper randomization prevents selection bias, so any baseline differences are the result of chance. A significance test asks whether a difference could have occurred by chance, which we already know. Such tests are superfluous and can mislead.

Senn (1994) called tests of baseline homogeneity philosophically unsound and of no practical value. Instead, judge each baseline imbalance by two things: how strongly the variable predicts the outcome and how large the difference is. Altman and Doré found inadequate handling of baseline comparisons in 41% of 80 trial reports they reviewed.

Can anything be done to increase baseline comparability?

• ✓ Stratify the randomization: ICH E9 says stratification by important prognostic factors measured at baseline, such as disease severity, age or sex, may promote balance, especially in small trials. More than two or three factors is rarely necessary.
• ✓ Dynamic allocation: methods that minimize the overall imbalance across several prognostic factors at once; the EMA notes they do not guarantee balance within combinations of factors.
• ✓ Pre-specify covariate adjustment: the EMA and FDA guidance both ask for covariates to be named in advance, in the trial protocol or before any unblinding. Stratification factors should normally be included.

What can disrupt baseline comparability?

Chance, in small samples. Broken allocation concealment, when someone can predict the next assignment. And the choice of analysis set: excluding participants after randomization can create differences that randomization had removed. Blinding protects comparability after assignment.

If a chance imbalance appears in a prognostic variable that was not pre-specified, the EMA says it is not a reason to add it to the primary analysis. Run a sensitivity analysis instead. ICH E9 agrees: show that an analysis accounting for the imbalance reaches essentially the same conclusions.

Baseline analysis: change from baseline vs ANCOVA

Many randomised trials measure a continuous variable, such as pain or blood pressure, at baseline and again at the end of the trial. Vickers and Altman (BMJ 2001) compare three ways to estimate the treatment effect:

• ✓ Follow-up scores only: biased if the groups differ at baseline.
• ✓ Change scores (mean change): do not control for baseline imbalance, because of regression to the mean. Patients with low baseline scores tend to improve more than those with high scores.
• ✓ ANCOVA: a regression of the follow-up score on the baseline score and treatment group. It gives the same answer whether or not there is baseline imbalance and usually has more statistical power.

Their example: with a correlation of 0.6 between baseline and follow-up, a trial needing 85 patients for a follow-up analysis would need 68 for a change-score analysis and 54 with ANCOVA. A smaller sample size for the same power.

Regulators say the same. The EMA guideline says the baseline value of a continuous primary outcome should usually be a covariate, whether the endpoint is the raw outcome or the change from baseline. With that adjustment, both endpoints give identical treatment effect estimates. The FDA notes that adjusting for prognostic baseline covariates generally narrows confidence intervals. Baseline covariates are measured prior to randomization. Both agencies warn against adjusting for variables measured after randomization, because treatment may have affected them.

Example from trial practice: the baseline visit

Hypothetical baseline visit at the site

Hypothetical example. A participant returns for Day 1 of a randomised clinical trial after screening. The clinical research coordinator follows the schedule of events:

• ✓ reviews and updates the medical history and the concomitant medications log;
• ✓ collects vital signs, baseline bloods and the baseline score, recorded first in the source documents;
• ✓ has the investigator confirm the inclusion and exclusion criteria;
• ✓ randomizes the participant and dispenses the investigational product.

If a criterion is not met, the participant is a screen failure and is not randomized.

Hypothetical CRA monitoring visit

Hypothetical example. ICH E6(R3) lists verifying that the investigator enrolls only eligible participants among the first tasks of monitoring clinical trial data. At the next visit, the CRA checks:

• ✓ that the investigator signed off eligibility before randomization;
• ✓ that baseline values in the case report form match the source records;
• ✓ that conditions present at baseline sit in the medical history, not on the AE log.

Any gap is followed up and documented in the visit report.

Where you learn this at VIARES

In the Clinical Study Coordinator II course (80–100 hours), Section 5 “Baseline Visit” covers the schedule of events, source documents, inclusion and exclusion criteria, eligibility, ConMeds and medical history. Learners study when a subject has screen failed, then complete a microcourse, an assignment and a knowledge check based on a sample study protocol. Section 4 covers the screen visit before it.

In the Clinical Research Associate course, Section 6 covers trial design; its assignment asks learners to explain stratified block randomization and its bias-reducing effect. Section 8 covers how to prepare and conduct monitoring visits.

The Oncology CRA course has a RECIST section whose learning elements include baseline assessment and tumor burden.

FAQ

What does baseline mean in medical terms?

A baseline is an initial measurement of a condition, taken at an early time point and used for comparison over time. In a clinical trial, it is the set of values collected before treatment starts.

What is baseline data in a clinical trial?

Baseline data are the information collected from each participant before treatment, such as demographics, body mass index, medical history, concomitant medications, vital signs, lab results and disease severity scores.

What is a baseline example?

A baseline blood test: lab values are measured before the first dose and compared with later results. Another example is a pain score at Day 1, used to calculate the change from baseline at the end of the trial.

Why aren’t statistical tests recommended to assess baseline comparability?

In a properly randomized trial, any baseline differences are due to chance, so testing whether they are due to chance adds nothing. Judge each imbalance by its size and how strongly the variable predicts the outcome.

Can anything be done to increase baseline comparability?

Yes. Trials can stratify randomization by a few important prognostic factors, use dynamic allocation, and pre-specify adjustment for prognostic baseline covariates in the protocol.

Sources

1. National Cancer Institute. NCI Dictionary of Cancer Terms: baseline
2. ICH E9: Statistical Principles for Clinical Trials, sections 2.3.2, 5.2, 5.4, 5.5, 5.7, 6.3, 7.1 and Glossary (treatment emergent)
3. ICH E6(R3): Guideline for Good Clinical Practice (2025), section 3.11.4.5.4
4. EMA/CHMP/295050/2013: Guideline on adjustment for baseline covariates in clinical trials (2015)
5. FDA. Adjusting for Covariates in Randomized Clinical Trials for Drugs and Biological Products. Guidance for Industry (May 2023)
6. Hopewell S, Chan AW, Collins GS, et al. CONSORT 2025 statement: updated guideline for reporting randomised trials. BMJ 2025;389:e081123 (item 25)
7. Moher D, Hopewell S, Schulz KF, et al. CONSORT 2010 Explanation and Elaboration. BMJ 2010;340:c869 (item 15, baseline data)
8. Vickers AJ, Altman DG. Analysing controlled trials with baseline and follow up measurements. BMJ 2001;323:1123–1124
9. Senn S. Testing for baseline balance in clinical trials. Stat Med 1994;13:1715–1726
10. Altman DG, Doré CJ. Randomisation and baseline comparisons in clinical trials. Lancet 1990;335:149–153