Descriptive statistics and Table 1
What question does this help answer?
Who was studied. It is a description of the sample the study has, not a test of whether groups differ, and it is not sized.
Choosing the summary for each variable
- Continuous and roughly symmetric: mean and standard deviation.
- Continuous and skewed, such as length of stay or a laboratory value with a long tail: median and interquartile range.
- Categorical: number and percentage, with the denominator clear.
- A score or index, such as a comorbidity index, is summarised as a number, not as a list of categories.
- The summary follows from what the variable is, never from where it happens to sit in a list.
Columns
Use one column per group compared: exposure groups in a cohort, arms in a trial, cases and controls in a case-control study, with an overall column where it helps. If the analysis is weighted, the table should show the weighted sample as well, because that is the population the estimate describes.
Rows
Include the characteristics a reader needs to judge generalizability and the variables the analysis adjusts for. Terms that belong to the design rather than to the patients, such as calendar period, site or ward, stay in the model and are not described here as patient characteristics.
Missing values
State how many values are missing for each variable rather than letting percentages quietly use different denominators. A variable missing for a third of one group and hardly any of another is itself a finding.
Testing baseline differences
In a randomized trial, reporting guidance advises against significance tests of baseline differences. With random allocation an observed difference is compatible with chance, and a test cannot say whether randomization worked or whether the imbalance matters for the result. In an observational study a p-value reflects the sample size as much as the imbalance; a standardized difference describes the imbalance itself.
Why Mallard may recommend it
Every Mallard plan includes a Table 1 shell. Mallard types each variable from its name and leaves design terms out of the patient description.
What to watch out for
A p-value column in the Table 1 of a randomized trial answers the wrong question: with random allocation any baseline difference is compatible with chance, and a significance test can say neither whether randomization worked nor whether the imbalance matters. Judge imbalance by its size and by how strongly the variable predicts the outcome. The other routine error is the standard error written where the standard deviation belongs: the standard error describes how precisely the mean is known, not how much patients vary, and it makes a varied sample look uniform.
Clinical example
Question: Who was in a cohort comparing two anticoagulants after atrial fibrillation? Table 1: Age as mean and standard deviation, creatinine as median and interquartile range, sex and prior stroke as number and percentage, one column per drug, with the number missing shown for each variable and no p-values.
- Who is in this study, anyway? Guidelines for a useful Table 1 (opens in a new tab) · J Clin Epidemiol · doi 10.1016/j.jclinepi.2019.06.011
- CONSORT 2010 explanation and elaboration: updated guidelines for reporting parallel group randomised trials (opens in a new tab) · BMJ · doi 10.1136/bmj.c869