A browser-based tool for reproducible covariate-adaptive randomization in preclinical mouse studies. Upload a CSV with animal IDs, sex, body weight, and bone mineral density (BMD), and the app will assign mice to treatment groups while balancing on sex, weight, and BMD.
- Covariate-adaptive minimization — The default algorithm is inspired by Pocock & Simon (1975). Each mouse is sequentially assigned to the group that minimizes imbalance across sex, weight, BMD, and group size, with a biased-coin probability (p = 0.80) to preserve randomness. Continuous covariates (weight, BMD) are balanced directly on their within-sex group means — no binning or discretization — so the algorithm optimizes the same quantity measured by the standardized mean difference (SMD). Imbalance is measured as the sum of squared deviations of group means from the grand mean (SSD), standardized by within-sex SD. This uses information from all groups (not just the two extremes) and produces smoother optimization than a range-based metric, especially with 3+ groups.
- Within-sex covariate balancing — Weight and BMD are balanced separately within males and females. This prevents a scenario where balanced overall means mask within-sex imbalance (e.g., heavy males concentrated in one group, compensated by heavy females in another).
- Stratified block randomization — Also available as a simpler alternative when only sex balance is needed. Mice are stratified by sex and assigned in random permutation blocks.
- Formal balance assessment — After randomization, pairwise standardized mean differences (SMDs) are displayed for weight and BMD (overall, within-males, and within-females), along with the absolute difference in male proportion for sex balance. SMDs are color-coded against 0.1/0.2 thresholds with an overall verdict.
- Within-sex summaries — Group summary cards report mean ± SD for weight and BMD broken out by sex, since sex is a balancing variable and within-sex balance matters for downstream analysis.
- Cage distribution check — If your CSV includes a "Cage" column, the
app flags any cages where all mice ended up in the same treatment group.
If no Cage column is present, the app automatically infers cage numbers
from the standard Animal ID format (
[text][cageNumber].[mouseID], e.g.VQ10.1.1→ cage 10,VQ16.L1→ cage 16), so cage balance is checked by default for data following this convention. Mouse identifiers after the dot can be numeric or alphanumeric. - Reproducibility and audit trail — Optionally specify a seed for exact reproducibility. Download a full randomization report including: timestamp, algorithm parameters, PRNG name (Mulberry32), seed, input file SHA-256 hash, group summaries with within-sex statistics, the complete SMD table, cage warnings, individual assignments, and a suggested methods statement for your manuscript.
- Allocation concealment — Toggle blinded labels (A, B, C…) in the output CSV, with a separate downloadable key file for unblinding. Supports single-blind study workflows.
- Incremental assignment — Upload a CSV with some mice already assigned to groups and leave the "Group" column blank for new mice. The app assigns only the new mice while maintaining overall balance. Warns explicitly if pre-assigned labels don't match configured group labels.
- Flexible group configuration — Choose 2–8 groups with custom labels (defaults to "Vehicle" and "AP").
- Unequal allocation ratios — Assign different weights to each group (e.g., 2:1 or 3:1) to allocate more mice to a control or treatment arm. Both the minimization and stratified block algorithms respect the specified ratio while still balancing covariates. The ratio is documented in the randomization report and suggested methods statement.
- Robust CSV parsing — Uses Papa Parse v5.4.1 (inlined, no CDN dependency) for reliable handling of quoted fields, embedded commas, mixed line endings, and other CSV edge cases. Non-fatal parser diagnostics (e.g., unexpected delimiters, field count mismatches) are surfaced as warnings rather than silently ignored.
- Strict input validation — Sex values must be
Male,M,Female, orF(case-insensitive); unrecognized values produce a clear error with the row number. Duplicate Animal IDs are rejected explicitly. - Fully client-side — No data leaves your browser. Everything runs in a single HTML file with no server or backend dependencies.
Your CSV should include the following columns (naming is flexible):
| Column | Examples of recognized names |
|---|---|
| Animal ID | Animal ID, Mouse ID, ID, Subject |
| Sex | Sex, Gender — accepted values: Male, M, Female, F (case-insensitive) |
| Weight | Start Weight (g), Weight, BW, Mass |
| BMD | Spine BMD, BMD, Bone Mineral Density, DEXA, DXA |
| Group (optional) | Group, Treatment, Assignment |
| Cage (optional) | Cage, Cage ID, Housing |
If no Cage column is present, the app will attempt to infer cage numbers
from the Animal ID column using the format
[text][cageNumber].[mouseID] (e.g., VQ2.1 → cage 2, VQ10.1.1 →
cage 10, VQ16.L1 → cage 16). Mouse identifiers after the dot can be
numeric or alphanumeric.
If you don't have a CSV yet, click Download template CSV inside the app to get a blank one with the correct headers.
- Drag and drop your CSV onto the upload zone, or click to browse.
- Set the number of treatment groups and edit labels as needed.
- Adjust the allocation ratio if you need unequal group sizes (e.g., 3:1 to put three times as many mice in the control group). Leave all weights at 1 for equal allocation.
- Choose your randomization method: Minimization (Pocock–Simon inspired) for multi-covariate balancing, or Stratified block randomization for sex-only stratification.
- Optionally enter a seed for reproducibility, or leave blank for automatic seed generation.
- Toggle Allocation concealment if you need blinded group codes.
- If your CSV includes a "Group" column with some mice already assigned, the app detects them automatically. You'll be warned if pre-assigned labels don't match your configured group labels.
- Click Randomize Groups to run the assignment.
- Review the summary cards showing sample size, sex split, and mean ± SD for weight and BMD (overall, male, and female) per group.
- Check the Balance Assessment table for pairwise SMDs (overall and within-sex) and sex proportion differences.
- If a Cage column is present, review the Cage Distribution Check for potential confounds.
- Click Re-randomize to try a different seed.
- Download CSV — The result CSV matches your original with the "Group" column added or filled in.
- Randomization Report — A plain-text report documenting all parameters, statistics, and a suggested methods statement. Keep this with your study records.
- Blinding Key — If allocation concealment is enabled, download the code-to-treatment mapping separately. Store securely until unblinding.
This implements a covariate-adaptive minimization procedure inspired by the method described by Pocock and Simon (1975). Note that this implementation has not been formally validated against their original formulation.
- Within-sex standard deviations are computed for weight and BMD across all mice (including any pre-assigned). These are used to standardize the continuous imbalance terms so that weight and BMD contribute on a comparable scale regardless of their units.
- Mice are processed in random order. For each mouse, the app computes
the imbalance that would result from assigning it to each candidate
group, across four balancing factors:
- Sex (categorical): range of allocation-normalized counts across groups for the mouse's sex level.
- Weight (continuous, within-sex): sum of squared deviations (SSD) of within-sex group means from the within-sex grand mean, divided by the within-sex variance.
- BMD (continuous, within-sex): same SSD metric as weight, computed within the mouse's sex.
- Group size: range of allocation-normalized total counts.
- For continuous covariates, only the mouse's own sex stratum is evaluated — the other sex's contribution is constant across candidate groups and cancels out. This ensures the algorithm balances weight and BMD within males and within females independently, preventing a scenario where balanced overall means mask within-sex imbalance.
- The SSD metric uses information from all groups (not just the two extremes), producing smoother optimization than the range, especially with 3+ groups. With 2 groups the two metrics are equivalent.
- When unequal allocation ratios are specified, counts are normalized by each group's target proportion before computing the range, so the algorithm targets the specified ratio (e.g., 2:1) rather than equal group sizes.
- The total imbalance for each candidate group is the sum across all factors. The group(s) with the lowest total are preferred. A biased coin (p = 0.80) assigns to a preferred group; with probability 0.20, a group is chosen uniformly at random.
- Pre-assigned mice are accounted for in the running state (sex counts, weight sums, BMD sums per sex per group) before any new assignments.
- Mice are separated by sex into two strata.
- Within each stratum, mice are shuffled randomly, then divided into blocks. With equal allocation, block size equals the number of groups. With unequal allocation (e.g., weights 2:1), each block contains group slots proportional to the weights (e.g., [A, A, B]), so the target ratio is maintained within every complete block.
- Each block receives a random permutation of group assignments.
- Pre-assigned mice are respected; only unassigned mice are processed.
After randomization, the app reports:
- Pairwise SMDs for weight and BMD: the absolute difference in group means divided by the pooled within-group standard deviation. Reported overall and within each sex.
- Absolute difference in male proportion between each pair of groups (not an SMD, since sex is categorical).
- SMD thresholds: < 0.1 (green), < 0.2 (amber), ≥ 0.2 (red). These are conventional heuristics for assessing covariate balance, not formal significance tests. They should be interpreted in the context of your study design and sample size.
This is a single index.html file with Papa Parse inlined (no external
runtime dependencies aside from Google Fonts). To host on GitHub Pages:
- Push
index.htmlto a repository. - Go to Settings → Pages and set the source to your main branch.
- The app will be live at
https://yourusername.github.io/your-repo-name/.
No build step, bundler, or server is required.
- Pocock SJ, Simon R. Sequential treatment assignment with balancing for prognostic factors in the controlled clinical trial. Biometrics. 1975;31(1):103–115.
This software was developed with the assistance of Claude (Anthropic). The code, design, and documentation were generated through an iterative conversation with the AI model and reviewed by the project author. The randomization algorithm and user interface logic were authored by Claude based on requirements provided by the project author.
This project is licensed under the MIT License.