What Is Cohort Explorer?
Cohort Explorer is a general-purpose workbench for exploring a tabular research dataset — a patient cohort, an animal study, a sample registry, or any dataset with one row per subject. Upload a CSV or paste values, and the tool infers each column's type (continuous, categorical, binary, or text) and unlocks four analysis tabs: descriptive statistics, bivariate testing, an automatically formatted Table 1, and a pairwise correlation heatmap.
It is designed to answer the questions researchers ask first when they open a new dataset: What does each variable look like? Are any two variables related? How do baseline characteristics differ across groups? Which variables cluster together?
Descriptive statistics, per column
The Descriptive tab summarizes one column at a time. Continuous variables get n, mean, SD, median, IQR, missing count, and a histogram; categorical and binary variables get a frequency table with counts and percentages. This is the first stop for spotting skew, outliers, or unexpectedly high missingness before running any inferential test.
Bivariate testing with automatic method selection
The Bivariate tab picks an appropriate test based on the types of the two variables you select:
- Continuous vs. continuous: Pearson's r with a scatter plot and regression line, alongside Spearman's rho as a robustness check.
- Continuous vs. categorical/binary (2 groups): Welch's t-test (does not assume equal variances) with Mann-Whitney U reported as a nonparametric alternative.
- Continuous vs. categorical (3+ groups): one-way ANOVA with Kruskal-Wallis reported alongside it.
- Categorical vs. categorical/binary: chi-square test of independence with a contingency table and Cramer's V; a Fisher's exact test suggestion appears if any expected cell count falls below 5.
Each result includes the test statistic, degrees of freedom (where applicable), p-value, and effect size, so you can move straight from exploration to a reportable statistic.
Table 1: baseline characteristics, generated automatically
Almost every clinical or epidemiological manuscript opens with a Table 1: baseline characteristics of the cohort, stratified by a grouping variable such as treatment arm, exposure status, or diagnosis. Building this table by hand means re-running summary statistics and the correct between-group test for every single row — tedious and error-prone.
The Table 1 tab automates this: choose a grouping variable and the observed variables to include, and it computes mean (SD) or median (IQR) for continuous rows and n (%) for categorical rows, split by group, together with the statistically appropriate test for each row (t-test/ANOVA/Mann-Whitney/Kruskal-Wallis for continuous rows, chi-square/Fisher's for categorical rows).
Correlation heatmap for exploring structure across many variables
The Heatmap tab computes a full pairwise correlation matrix (Spearman by default, or Pearson) across every continuous and binary variable you select. Cells are colored by strength and direction, and you can set a minimum |rho| threshold to mask weak relationships so the matrix highlights only the structure that matters. This is a fast way to spot multicollinearity before regression, identify candidate confounders, or find variable clusters worth a closer look in the Bivariate tab.
A note on sample size and multiple comparisons
Exploring many variable pairs at once inflates the chance of a false positive: testing 10 variables against each other produces 45 pairwise comparisons, so at alpha = 0.05 you expect roughly two "significant" results by chance alone. Treat the heatmap and bivariate tabs as hypothesis-generating tools, not a substitute for a pre-registered primary analysis with correction for multiple comparisons where appropriate.
Frequently Asked Questions
What counts as a "cohort" or dataset here?
Any tabular dataset with one row per subject (patient, animal, sample) and one column per variable — demographics, exposures, lab values, or outcomes. Upload a CSV or paste values directly; the tool auto-detects whether each column is continuous, categorical, binary, or free text.
What is a Table 1?
Table 1 is the standard baseline-characteristics table in clinical and epidemiological papers: it summarizes your cohort's variables (mean/SD or median/IQR for continuous, n/% for categorical) split by a grouping variable such as treatment arm or exposure status, with an appropriate between-group test for each row.
How does the correlation heatmap decide which test to run?
For each variable pair it computes Spearman's rho by default (robust to non-normal and ordinal data) or Pearson's r if you switch methods, using pairwise-complete observations. Cells below your chosen n or |rho| threshold are masked so sparse or negligible relationships don't clutter the matrix.
Does my data leave the browser?
No. Parsing, descriptive statistics, hypothesis tests, Table 1 generation, and the correlation matrix all run client-side in JavaScript. Nothing is uploaded to a server, which makes this suitable for de-identified but still sensitive research data.