📋 My Report Card
⭐ Total Points: 0
New: AI for Clinicians — Build Your First Working System →

Understand the evidence. Question the model. Make better clinical decisions.

Interactive statistics, critical appraisal, and AI/ML education built around the decisions surgeons actually face.

I need help with a paper

Free · Interactive · Surgeon-built · No statistics degree required

Choose your starting point

What are you trying to do?

Start with the problem in front of you. The curriculum connects the concepts as you go.

01 · LEARN

Read clinical evidence

Understand study design, statistics, effect sizes, regression, survival analysis, and power.

 02 · EVALUATE

Question AI and ML

Understand how models work, where they fail, and whether they add clinical value.

 03 · APPLY

Solve today’s problem

Scan a paper, select an analysis, prepare journal club, or build a practical AI workflow.
📚

Clinical Evidence and Statistics

Start here: 6 available modules from paper reading to power

01 Available

Reading a Paper Without Faking It

Study design, hierarchy of evidence, and spotting red flags before you get fooled.

5 lessons · 45 min
02 Available

Choosing the Right Test

Variable types, study design, and the decision tree for test selection.

5 lessons · 45 min
03 Available

Understanding Your Output

P-values, confidence intervals, effect sizes, and how to smell a fragile result.

4 lessons · 35 min
04 Available

Survival Analysis for Surgeons

Kaplan-Meier, Cox regression, competing risks, and why your amputation data needs special handling.

5 lessons · 55 min
05 Available

Regression Without Crying

Linear, logistic, ordinal. When to use each, how to interpret coefficients, and why your 15-variable model is garbage.

6 lessons · 60 min
06 New

Sample Size and Power

Why your retrospective study is underpowered, how to calculate what you need, and when to admit defeat. Includes an interactive power calculator.

5 lessons · 55 min
07 Coming Soon

Show Your Work

How to present your data so people will listen. Tufte, Schwabish, and the figures every clinical paper actually needs.

5 lessons · 50 min
🤖

AI and Machine Learning Judgment

Use AI safely and evaluate whether a model deserves your trust

START HERE Available

AI for Clinicians: Build Your First Working System

Start from the beginning and leave with an AI setup, personal context, a completed administrative task, a literature-update path, and a reusable skill—all working for you.

10 short lessons · About 3 hours · No exam
LEVEL UP Available

Already using AI? Build Your Personal Context

Skip straight to a guided context builder so Claude, ChatGPT, or another assistant can understand who you are, how you work, and what matters to you.

Plain-language guide · Free builder · 30–60 min
01 Available

AI/ML -- What It Actually Is

The hierarchy, the vocabulary, and why it matters for your clinical research.

5 lessons · 50 min
02 New

The Clinical AI Bullshit Detector

A five-question framework for judging discrimination, calibration, validation, leakage, and whether a model would improve an actual clinical decision.

Interactive critical appraisal · 20 min
03 Available

Large Language Models

How the AI you actually use works -- training, how answers are generated, RAG vs. frontier models, and using it safely.

4 lessons · 45 min
04 Coming Soon

NLP and Clinical Text

What NLP can (and can't) extract from operative notes, radiology reports, and clinical documentation.

4 lessons · 40 min
05 Coming Soon

Computer Vision and Image-Based ML

Because wound photos and CT scans are data too.

4 lessons · 40 min
06 Coming Soon

Survival Analysis Meets ML

Bridging traditional survival methods and the ML world. Random survival forests, DeepSurv, and competing risks.

3 lessons · 30 min
📰

Journal Club

Landmark trials, dissected

SMART-REACH2

How to read a prediction model

A guideline-endorsed lifetime risk model for established vascular disease — and a working tour of discrimination vs calibration, competing risks, and turning a trial's average effect into one patient's life-years gained. Eur Heart J, June 2026.

NOBLE 10-Year

When 5-year ≠ 10-year

PCI vs CABG for unprotected left main disease. The composite said PCI lost. A decade later, the mortality curves are superimposable. A masterclass in why follow-up duration changes the story.

CREST-2

Stenting wins, CEA... doesn't?

Stenting beat medical therapy. CEA didn't reach significance. The trial that's reshaping how we think about asymptomatic carotid disease.

SWEDEPAD

One program, two trials, same answer

3,500 patients across two parallel Swedish trials. No amputation benefit in CLTI, no QoL benefit in claudication, and a 5-year mortality signal. The case against routine paclitaxel-coated devices.

AQUATIC

Aspirin + anticoagulation: friend or foe?

Chronic coronary syndrome patients on oral anticoagulation. When standard of care is scary.

🛠️

Tools

Easy AI wins + quick research utilities

New
→

Easy AI Wins for Clinicians

Small, practical workflows you can try today. Watch the demonstration, copy the prompt, and inspect the result yourself.

Try an easy win Featured
★

Verastar

A verifiable evidence digest for clinicians. Set your north-star concepts and active projects; Verastar finds and ranks the recent literature that matters to you, verifies surfaced evidence against the source or flags the gap, and saves what you keep to a plain-text library you own.

Open Verastar
📄

Is This Paper Worth My Time?

Answer 7 quick questions about a study to get a read/skim/skip verdict with specific red flags and strengths.

Scan a paper
🔬

Which Test Do I Need?

Walk through your study design step-by-step to get the right statistical test, assumptions to check, and what to report.

Find your test
🎤

Prep Your Journal Club

Walk through a 10-step framework and leave with a presentation ready to deliver in under 10 minutes.

Start prepping
☕

Build Your Morning Briefing

A free setup prompt for Claude that builds a daily PubMed digest curated to your specialty, deduped against papers you've already read.

Set it up
You are visitor number visitor count since 2026
Back to Modules
Module 01

Reading a Paper Without Faking It

Study design, hierarchy of evidence, and spotting red flags before you get fooled.

0/5 lessons complete
1
Study Design Hierarchy
RCT, cohort, case-control, and more
Foundation for evaluating any study's conclusions
Available
2
Bias and Confounding
Selection bias, information bias, confounders
Why associations aren't always causal
Available
3
Red Flags in Methods
P-hacking, HARKing, cherry-picking
Spot manipulation before you're fooled
Available
4
Reading Results Tables
Table 1, baseline characteristics, outcome tables
Extract what matters without getting lost
Available
5
The 5-Minute Paper Scan
A systematic approach to rapid appraisal
Decide if a paper is worth your time
Available
Back to Modules
Module 03

Understanding Your Output

P-values, confidence intervals, effect sizes, and how to smell a fragile result.

0/4 lessons complete
1
Effect Size vs P-Value
Statistical vs clinical significance
Why p=0.001 can be meaningless and p=0.09 can matter
Available
2
Confidence Intervals
What CIs tell you that p-values don't
Precision, plausible range, and clinical interpretation
Available
3
Effect Sizes That Matter
NNT, absolute vs relative risk, MCID
Translate statistics into clinical decisions
Available
4
Smelling Fragile Results
Underpowered studies, wide CIs, barely significant
Know when a result won't replicate
Available
Back to Modules
Module 04

Survival Analysis for Surgeons

Kaplan-Meier, Cox regression, competing risks, and why your amputation data needs special handling.

0/5 lessons complete
1
Kaplan-Meier Curves
Anatomy of the most common survival figure
Read KM curves critically instead of just glancing at them
Available
2
Censoring
Right, left, interval, informative vs non-informative
The invisible force that shapes every survival curve
Available
3
The Log-Rank Test
Comparing curves formally, not by squinting
When it works, when it fails, and what alternatives exist
Available
4
Cox Proportional Hazards
Hazard ratios, assumptions, events per variable
The workhorse model -- and its most abused assumptions
Available
5
Competing Risks
CIF, Fine-Gray, cause-specific models
Why your limb salvage rate is probably wrong
Available
Back to Modules
Module 05

Regression Without Crying

Linear, logistic, ordinal. When to use each, how to interpret coefficients, and why your 15-variable model is garbage.

0/6 lessons complete
1
What Regression Actually Does
The line, the intercept, and reading a coefficient
Every "adjusted" analysis in a paper is regression
Available
2
Logistic Regression and Odds Ratios
Probability, odds, log-odds, and the OR vs RR trap
The most common multivariable model in surgical papers
Available
3
Crude vs. Adjusted
Confounders, mediators, colliders, and the Table 2 fallacy
What "adjusting for age and diabetes" actually does
Available
4
Ordinal, Counts, and the Right Model
Proportional odds, Poisson, overdispersion, and rate models
The outcome type dictates the model
Available
5
Assumptions and Diagnostics
Linearity, clustering, multicollinearity, influential points
How to catch a model that is lying to you
Available
6
Model Building Without Overfitting
Events per variable, stepwise sins, and validation
Why your 15-variable model is garbage
Available
Back to Modules
Module 06

Sample Size and Power

Why your retrospective study is underpowered, how to calculate what you need, and when to admit defeat.

0/5 lessons complete
1
The Four-Way Trade-Off
Type I vs Type II errors, and the four levers that set power
Most "negative" studies never had a chance to be positive
Available
2
Where Does Delta Come From?
MCID, the pilot study trap, and reverse-engineered effect sizes
The one judgment call that decides the whole calculation
Available
3
Running the Numbers
An interactive power calculator, plus dropout and event math
Build intuition for what sample sizes actually buy
Available
4
The Underpowered Study Problem
Post-hoc power, CI-based interpretation, honest small studies
Your database has 74 patients and it's not getting bigger
Available
5
Capstone: Audit the Power Statement
A full power audit of a realistic manuscript, plus the checklist
Arithmetically flawless papers can still be wrong
Available
Back to Modules
Module 02

Large Language Models

How the AI you actually use works -- training, answer generation, RAG vs. frontier models, and using it safely in clinical work.

0/4 lessons complete
1
Tokens and Training
Next-token prediction, pretraining, and the knowledge cutoff
Why it confidently cites a paper that doesn't exist
Available
2
How a Model Produces an Answer
Sampling, temperature, context window, reasoning models
Why the same question gives different answers
Available
3
RAG vs. Frontier Reasoning
Why OpenEvidence differs from ChatGPT -- and a 2026 study
Learn the mechanism, not the leaderboard
Available
4
Trust, Verify, and Use Safely
Green/red-light uses, verification habits, patient privacy
Use these tools without getting burned
Available
Back to Modules
Module 07

Show Your Work

How to present your data so people will listen. Tufte, Schwabish, and the figures every clinical paper actually needs.

0/5 lessons complete
1
Why Most Clinical Figures Fail
Data-ink, chartjunk, and the 3-second test
If your figure needs explaining, it's already failed
Available
2
Choosing the Right Chart
Variable types map to chart types
Stop using bar charts for everything
Available
3
Tables That Don't Bury the Lede
Table 1, alignment, ordering, significant figures
A good table beats a bad chart every time
Available
4
Survival Curves and Forest Plots
The two figures every vascular paper lives or dies by
Done right, they answer the question without prose
Available
5
Redesign Clinic
Before/after of real published figures
Practice spotting and fixing what's wrong
Available
Back to Modules
Module 01

AI/ML -- What It Actually Is

The hierarchy, the vocabulary, and why it matters for your clinical research.

0/5 lessons complete
1
The AI Landscape for Clinicians
AI > ML > Deep Learning > LLMs
Map the hierarchy so "we used AI" stops fooling you
Available
2
Supervised vs. Unsupervised vs. Reinforcement
Three paradigms, three different data requirements
The paradigm dictates what data you need
Available
3
Classification vs. Regression vs. Clustering
ML task types mapped to stats you already know
Most clinical ML is supervised classification or regression
Available
4
Features, Labels, and the Data Pipeline
Translating ML jargon to clinical research language
If you can build a REDCap database, you're 80% there
Available
5
When ML Adds Value (and When It Doesn't)
ML vs. traditional stats: choosing the right tool
ML is not a magic upgrade. It has different requirements.
Available
Back to Modules
Module 02

Choosing the Right Test

Variable types, study design, and the decision tree for test selection.

0/5 lessons complete
1
Variable Types
Continuous, categorical, ordinal, time-to-event
Test selection depends entirely on this
Available
2
Outcome vs Predictor
Which variable are you trying to explain?
Determines the direction of analysis
Available
3
Paired vs Independent
Same subjects or different subjects?
Wrong choice = invalid test
Available
4
Number of Groups
One, two, or more?
Determines t-test vs ANOVA vs other
Available
5
The Decision Tree
Pull it all together
Interactive: answer questions → get the right test
Available
Effect Size vs P-Value 0 pts Module 2 · Lesson 1
Introduction

Effect Size vs P-Value

Why statistical significance isn't the same as clinical importance

What You Were Taught

p < 0.05 = significant = real finding = publish
p ≥ 0.05 = not significant = no effect = move on

What We're Going to Show You

That framework is incomplete at best, misleading at worst. By the end of this lesson, you'll understand why a p=0.001 result can be meaningless, and why a p=0.09 result might be the one that should change your practice.

Consider These Two Trials

Both tested Drug X for claudication. Look at the results:

Trial A

Patients
10,000
6-Min Walk Improvement
4 meters
P-Value
0.001

Trial B

Patients
80
6-Min Walk Improvement
85 meters
P-Value
0.09

Which drug works better? Which result should change your practice?

Hold that thought. We'll come back to it.

What the P-Value Actually Tells You

1

The definition

The p-value is the probability of seeing this result (or something more extreme) if the null hypothesis were true.

In other words: "If there's actually no difference, how often would we see data like this by chance?"

2

What it does NOT tell you

The p-value does not tell you:

• The probability that your finding is true
• The size of the effect
• Whether the effect matters clinically

3

The question it answers

"Is this likely due to chance?"

That's it. Nothing about importance. Nothing about magnitude.

The p-value answers the wrong question. You want to know if the effect matters. The p-value only tells you if it's probably real.

The Large-N Trap

With enough patients, any difference becomes statistically significant.

Same effect (2 mg/dL LDL reduction), different sample sizes:

N = 50
2 mg/dL drop
p = 0.40
N = 500
2 mg/dL drop
p = 0.08
N = 5,000
2 mg/dL drop
p = 0.003
N = 50,000
2 mg/dL drop
p < 0.001

The effect didn't change. Your certainty about a trivial effect increased.

The p-value is a function of sample size. Effect size is not.

Effect Size Is the Clinical Question

Effect size answers "how much?"—the question you actually care about.

1

Common effect size measures

• Absolute difference (e.g., 4 meters vs 85 meters)
• Relative risk or hazard ratio
• Number needed to treat (NNT)
• Odds ratio

2

Minimum Clinically Important Difference (MCID)

Below this threshold, who cares if it's significant?

For 6-minute walk distance in PAD: the MCID is roughly 30-50 meters.

A 4-meter improvement? Statistically significant noise.

The question to always ask:

"Is this difference big enough to change what I do for my patient?"

If the answer is no, the p-value is irrelevant.

Confidence Intervals Give You Everything

The 95% CI is more informative than the p-value alone.

1

Point estimate = effect size

The middle of the CI is your best guess at the true effect.

2

Width = precision

Narrow CI = large sample, precise estimate.
Wide CI = small sample, uncertain estimate.

3

Crossing the null = not significant

If the 95% CI for a difference includes zero (or 1.0 for ratios), p > 0.05.

If the entire CI falls within a clinically meaningless range, the study is definitively negative—even if p < 0.05.

Example

A study shows: Effect: 2 mg/dL, 95% CI: 1.5 to 2.5, p < 0.001

The entire confidence interval is below any meaningful LDL reduction. This is a confident null—we're certain the effect is too small to matter.

Exercise: Significant or Important?

Classify each scenario into one of four categories.

Significant AND Important
Significant but NOT Important
Not Significant but Potentially Important
Neither
Question 1 of 8

Back to Our Two Trials

Trial A

Patients
10,000
6-Min Walk Improvement
4 meters
P-Value
0.001

Trial B

Patients
80
6-Min Walk Improvement
85 meters
P-Value
0.09

Now you can answer:

Trial A is statistically significant but not clinically important. A 4-meter improvement is far below the MCID of 30-50 meters. With 10,000 patients, you've achieved high confidence in a trivial effect.

Trial B is not statistically significant but potentially important. An 85-meter improvement is clinically meaningful—nearly double the MCID. The p=0.09 reflects an underpowered study, not an absent effect.

Trial B should influence your thinking more. It suggests a meaningful effect that deserves a larger study.

The Bottom Line

0
Total Points Earned
Exercise (0/8 correct) +0 pts
Lesson Completed +100 pts
  • P-value tells you about chance, not importance. A small p-value means you're confident the effect isn't zero—not that it matters.
  • Always ask: "How big is the effect? Is that big enough to matter?" If the effect is below the MCID, significance is irrelevant.
  • A confident null is more informative than a shaky significant result. A tight CI around zero tells you more than a wide CI that barely excludes it.
  • Look at the confidence interval. It gives you effect size, precision, and significance in one measure.

You Now Know

How to look past p-values and evaluate what a study actually found. You won't be fooled by significant-but-trivial results, and you'll recognize potentially important findings that failed to reach significance due to sample size.