900 rows · 11 columns

Randomised trial results

A synthetic two-arm trial: 900 subjects, baseline and week-12 scores, adverse events, and a real but modest treatment effect that is easy to overstate if you analyse it badly.

Download CSV 51 KB JSON 169 KB Open in Explorer Query it in SQL

What is in it, and what is wrong with it

The true treatment effect is a 4.1-point improvement on a 0–100 scale. Dropout is informative — subjects doing badly are likelier to leave — so a complete-case analysis will overstate the benefit. There are eleven baseline covariates you could slice by, which is exactly enough for a determined analyst to find a spurious "responder subgroup" at p < 0.05. Try it, then read the multiple-comparisons lesson.

Schema

Column	Type	Description
subject_id	string	Subject identifier.
arm	category	treatment or placebo.
site	category	Enrolling site, 1–6.
age	integer	Age at enrolment.
sex	category	Recorded sex, F or M.
bmi	float	Body mass index at baseline.
prior_therapy	boolean	Whether the subject had prior therapy.
baseline_score	float	Symptom score at day 0, 0–100, lower is better.
week12_score	float	Score at week 12. Empty if the subject dropped out.
dropped_out	boolean	Whether the subject left before week 12.
adverse_event	category	none, mild, moderate, or severe.

First twelve rows

subject_id	arm	site	age	sex	bmi	prior_therapy	baseline_score
S1245	placebo	site-1	51	M	24.2	true	82.3
S1603	placebo	site-4	57	F	32	false	63.2
S1845	placebo	site-6	33	M	28.1	true	53.1
S1640	treatment	site-1	51	M	27.5	false	71
S1231	placebo	site-2	45	F	26.1	false	57.6
S1448	treatment	site-6	38	F	20.2	true	64.2
S1047	placebo	site-1	62	F	30.2	false	44.3
S1899	placebo	site-2	46	F	33	true	67.5
S1457	placebo	site-3	50	F	30.2	true	64.4
S1649	placebo	site-6	49	F	16	false	57
S1802	treatment	site-3	57	M	31.8	false	72
S1831	placebo	site-5	49	F	23.1	false	57.9

Showing the first 8 of 11 columns. Open the full table in the Explorer.

Grain

One row per enrolled subject.

Things to try

Run the intention-to-treat analysis and compare it with complete-case.
Check baseline balance across arms — randomisation should have handled it.
Go subgroup hunting, count your tests, and apply a correction.
Estimate the effect with ANCOVA adjusting for baseline; compare precision.

Load it

python

import pandas as pd

df = pd.read_csv(
  "https://datanemics.com/datasets/data/clinical-trial.csv"
)

Licence

Public domain, CC0 1.0. Synthetic data — no attribution required.