Say I want to test a new wax coating that is supposed to keep mangoes fresh longer.

I go to the market and pick up 200 mangoes. But I only set out to test one variety, so the ones that are a different type do not belong in this particular test. I set those aside. That leaves 160. Then I split them into two baskets: 80 get the new coating, 80 get the usual treatment. Two weeks later, I check which basket kept more fruit fresh.

Simple. But notice how many places I could lose track of a mango.

A flow diagram is just the honest count at every one of those places.

The four stops where the count can change

Every trial flow diagram walks through the same four stages. Here they are in mango terms.

1. Enrollment. I looked at 200 mangoes and kept 160. The diagram simply reports the starting number and why the other 40 were set aside, here because they were a different variety than the one being tested.

2. Allocation. The 160 get sorted into two baskets, 80 and 80. This is where the diagram records which group each one went into.

3. Follow-up. Over two weeks, a few mangoes leave the count. My cousin takes some home. A couple roll off the counter. The diagram records each of these instead of leaving them out.

4. Analysis. At the end, how many mangoes did I actually check? Say I only go back and check 100 of them, and 95 are still fresh. I might announce "95 percent stayed fresh." But that number is about the 100 I checked, not the 160 I started with. Sixty mangoes never got counted at all. The flow diagram keeps both numbers, the 160 and the 100, side by side, so the ones left out are not forgotten.

That last step is the whole point. The clearest approach counts everyone you started with, not only the ones still in the basket at the end.

Let’s get into the details

Now swap mangoes for people, and the coating for a cancer drug.

If a trial starts with 400 patients and reports on the 250 who finished, the diagram is where you learn what happened to the other 150. Maybe some had side effects and stopped. Maybe some moved away. None of that is a verdict on anyone. It is just information. And having it lets you read the result against the whole group, not only part of it.

That is what the flow diagram is for. People leave a study at every stage, for all kinds of reasons. The chart records where and how many, so the final number is understood in full context.

So when you read a study, do not skip that little chart. Find the number at the top. Follow it down. Check that the number at the bottom still adds up. If a lot of people left and the reasons are not reported, that is simply something to note as you weigh the result.

Spotlight

Doug Altman, a statistician who helped build these reporting rules, put it simply: readers should never have to guess what was probably done. They should be told, clearly. The flow diagram is that idea drawn as a picture.

The question I am carrying

The diagram shows where people left, but the reasons still have to be reported next to it. I keep wondering how often a big drop-off gets a quick glance instead of a real look. I am training myself to slow down there.

What’s next

Next, I want to unpack two words that decide how those final mangoes get counted: intention to treat, and per protocol. They sound technical. They quietly change what a trial appears to show.

For now: a result is only as complete as the patients a study accounts for.

Protocol & People is a newsletter and animated series making clinical trials understandable for everyone. Watch on YouTube: @ProtocolAndPeople. New here? Subscribe so you don’t miss the next post.

Compare, don’t just observe.

Views are my own. Educational only, not medical advice.

If you’ve been reading Protocol & People here on Substack, you know the goal: making clinical trials make sense. For everyone.

Today, that mission has a visual home.

I’ve been building an animated video series alongside the newsletter. The idea is simple: some concepts in clinical trials are visual. Enrollment flows, comparison groups, outcome trees. They click when you can see them. Written explanations are great, but sometimes a 60-second animation does what a paragraph can’t.

Episode 0: The Language of Clinical Trials

This is your glossary episode. 25 terms that unlock every study you’ll ever read:

• Randomization

• - Blinding

• - Placebo

• - CONSORT

• - Intention-to-treat

• - P-value

• …and 19 more.

Each one explained in plain language with animation. Bookmark it. Come back to it anytime.

Watch Episode 0 on YouTube: https://youtu.be/7KcbSm9l7QE

What’s coming next

Episode 1: Why Clinical Trials Exist; to accompany the corresponding Substack post

Episode 2: Hello CONSORT, the reporting rules that keep trials honest; to accompany the corresponding Substack post

The YouTube channel will be the home for the animated episodes. Substack stays the home for written deep dives, breakdowns, and the community.

I’d love your feedback!

This is early. I want to know what’s clear, what’s confusing, and what topics you want covered. If something in the animation doesn’t land, tell me. If there’s a term you’ve always found confusing, let me know.

Reply to this email or leave a comment below.

Subscribe on YouTube so you don’t miss the next episode: https://www.youtube.com/@ProtocolAndPeople

Thank you for being here,

Dr. Oye

This is the second post in Protocol & People, a series on how clinical trials work and who makes them work. Last time, I wrote about why trials exist. This time: we will focus on how we know trials are reported honestly.

Let’s Start from the Beginning

In the last post, I wrote about why clinical trials exist. But even a well-designed trial can give you the wrong answer if the reporting is off. You need rules about what gets reported and how.

That’s where CONSORT comes in.

In the early 1990s, people started noticing the same problem everywhere: clinical trial reports were missing critical information.

Trials would get published without explaining how randomization (this refers to:) happened. No information showing where patients dropped out. Conflicts of interest were mentioned vaguely. It wasn’t uncommon for primary outcomes to change between protocol and paper.

So where did that leave the reader? You couldn’t tell if a trial was done well or done badly. Couldn’t reproduce it. Couldn’t use it safely.

Two groups developed solutions in 1994. The SORT group (Standards of Reporting Trials) published guidelines. So did the Asilomar Working Group. Both trying to fix the same problem, different approaches. Drummond Rennie from JAMA pushed them to merge.

The two groups came together, and CONSORT was born- Consolidated Standards of Reporting Trials. Published in 1996 in JAMA and BMJ at the same time. Twenty-five items, a flow diagram. Finally, we, the people, could judge if a trial was believable.

Three people kept it going: Doug Altman, Ken Schulz, David Moher. Updated in 2001. Again in 2010. Each version tighter, more specific. Altman died in 2018. He was a statistician at Oxford who had this way of saying things like “Readers should not have to infer what was probably done; they should be told explicitly.” He spent decades building the tools to fix it. CONSORT. PRISMA. STROBE. The EQUATOR Network that houses them all.

CONSORT is endorsed by journals worldwide now. When enforced, clinical trial reporting is much better.

So…what’s CONSORT?

CONSORT is a checklist. Thirty items telling you what has to be in a trial report (not about how to run a trial). It’s all about proving you told the truth about what you did. The new version dropped this month. JAMA, Lancet, BMJ, Nature Medicine, PLOS Medicine all published it simultaneously. First update in fifteen years.

Why do we care?

You can design a perfect trial and still hide what happened in how you report it.

Say two trials test the same drug. Both randomized, double-blind, big enough. Both show benefit. But one tells you how many patients they screened, what happened to dropouts, whether they changed the primary outcome partway through, who funded it. The other doesn’t.

Without those details, you’re guessing. Can’t reproduce it. Can’t use it safely. Can’t trust it.

What Changed in CONSORT 2025?

Seven new items. Three revised. They pulled in stuff from extensions that used to be separate (Harms, Outcomes, Non-pharmacological Treatment).

Process was methodical. Scoping reviews, three-round Delphi survey with 317 people, two-day meeting. Trying to fix the ways trials still get reported badly.

What stood out to me:

Item 4: Data Sharing Plans: Trials have to say upfront whether they’ll share their data and code. To whom. Under what terms.

Item 5b: Conflicts of Interest. Authors have to disclose all conflicts. Financial, personal, intellectual property, anything that might color their interpretation.

Item 8: Patient and Public Involvement. Trials have to report if patients helped design the study. If so, how. Patients know things researchers miss, such as which endpoints actually matter to those living with a disease.

Item 12b: Eligibility Criteria for Sites and Intervention Deliverers. This surprised me. CONSORT 2025 asks for eligibility criteria not just for patients, but for the sites and people delivering treatment. I often wonder: could a hospital amazing infrastructure do this? In the world of Global Oncology and Global health, the answer to this question matters, with regards to how far the results can travel.

Item 15: Assessment of Harms and Unintended Effects. Trials have to describe how they tracked harms, not just whether the drug worked.

Item 21b-c: Analysis Populations and Missing Data. Two connected pieces. Say which populations you analyzed (intention-to-treat, per-protocol, as-treated). Describe how you handled missing data. Assume dropouts did as well as people who stayed? You might overestimate benefit. Assume they did worse? You might underestimate. The method matters.

The Open Science Section. The biggest structural change is this new section pulling together trial registration, protocol access, data sharing, and conflicts of interest. F

Why Patient Involvement Matters

My favorite part of CONSORT 2025 might be the requirement to report patient involvement. Because it shows how investigators think about their work. When patients help design a trial, they ask different questions. They flag outcomes that matter to real life and endpoints that don’t. They spot enrollment barriers protocols miss. When patients are absent, trials can be technically elegant but clinically useless. Beautiful science no one can actually use. The fact that CONSORT now asks “Were patients involved?” says something.

Spotlight: Sally Hopewell

One name kept showing up when I read about CONSORT: Sally Hopewell.

She’s a senior research fellow at Oxford’s Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences. Been involved with CONSORT since 2010, but took the lead on the 2025 update.

Her work is quiet. CONSORT doesn’t make headlines like a breakthrough drug does. Doesn’t save lives directly. But it shapes how every trial gets reported, which shapes how every clinician decides what works.

Best part? CONSORT 2025 is open access everywhere. She and her coauthors could have published in one high-impact journal and collected citations. Instead they published simultaneously in JAMA, The Lancet, BMJ, Nature Medicine, and PLOS Medicine. Free to read. Free to use.

I don’t know her. But she’s definitely someone who makes it work. So, thank you, Sally.

What’s Next

Next post: walking through the CONSORT flow diagram step by step. How to read it, what the numbers mean, and where trials lose patients & credibility.

For now: a trial is only as good as its reporting. And reporting is only as good as the rules we enforce.

Views are my own. Educational only.

Setting Snapshot

The Question That Started This

A few months ago, I watched an attending explain a new drug to a patient. The data looked good. The patient was hopeful. But something nagged at me: how do we actually know this works? Not “works” as in a plausible mechanism or an impressive case report, but works in the way that should change what we do tomorrow, for this patient and the next hundred?

That question led me here. Trials are how medicine moves from stories to answers. This post is about why that machinery exists at all.

Purpose

Medicine is full of persuasive stories. A patient improves after a new treatment; a colleague swears by a particular regimen. But stories can mislead. The patient might have improved anyway. The colleague might be remembering the successes and forgetting the failures.

Trials exist so we do not confuse a good story with a true effect. A trial is a disciplined comparison between defined groups under prespecified rules. The goal is to isolate the effect of the intervention from everything else that can sway outcomes; timing, selection, expectation, chance.

A Short History, in Plain Words

Clinicians have compared treatments for centuries, but the logic of modern trials crystallized in stages.

In 1747, James Lind divided twelve sailors with scurvy into six pairs and gave each pair a different remedy. Citrus worked. It was a tiny experiment, but it held a radical idea: compare, don’t just observe.

Two centuries later, the 1948 Medical Research Council streptomycin trial introduced randomization and concealment to tuberculosis treatment. Patients were assigned by sealed envelopes that even the physicians couldn’t predict. This protected the comparison from human bias; the tendency to give the new drug to patients we think might do better.

That template still holds. The method evolved to protect patients, reduce bias, and make answers reproducible across settings. What surprises me is how recent it all is. Modern trial logic is younger than antibiotics.

The Rails That Keep a Trial Honest

I think of a trial as a train. The intervention is the engine, but it only reaches a valid destination if it stays on the rails. Three rails keep it honest:

Rail one: Comparison. Two or more groups treated differently but observed the same way. Without a comparator, you cannot know if the outcome was caused by the treatment or by time, regression to the mean, or placebo effect.

Rail two: Prespecification. Clear eligibility, a defined primary endpoint, and an analysis plan written before anyone peeks at the data. This prevents the subtle (and sometimes unconscious) temptation to move the goalposts after seeing results.

Rail three: Standardized observation. Uniform follow-up windows, consistent outcome assessment, and a plan for missing data. If one group is watched more closely, differences in detection masquerade as differences in outcomes.

If any rail breaks, the result can drift; and you may not notice until the answer fails in the real world.

The Basics Every Reader Can Check

You don’t need a statistics degree to read a trial critically. Here are the elements I now look for every time:

Eligibility tells you who the answer applies to. Age, disease stage, prior treatments, biomarkers. A narrow rule can protect safety but limit reach. If the trial excluded everyone over 75, what does it tell you about your 82-year-old patient?

Assignment places people into groups. Randomization balances known and unknown factors; not just age and stage, but things we can’t measure, like motivation or microbiome. Concealment prevents staff from guessing the next slot and consciously or unconsciously steering patients.

Blinding hides the assignment from participants, clinicians, or outcome assessors; sometimes all three. When people don’t know which treatment was given, expectations can’t distort measurement. I used to think blinding was about fairness. It’s really about protecting us from ourselves.

Outcome defines what “better” means. Response rate (did the tumor shrink?), disease-free survival (how long before it returns?), overall survival (how long did patients live?), patient-reported outcomes (how did they feel?). The endpoint must match the question. A drug that shrinks tumors but doesn’t extend life answers a different question than one that does both.

Timing sets the windows. When to measure, and how long to follow. A six-month snapshot may miss late relapses or delayed toxicity.

Analysis plan explains who counts and how missing values are handled. Intention-to-treat means everyone randomized is analyzed, even if they never took the drug. Per-protocol means only those who followed the plan. Each answers a different question.

Ethics and oversight ensure informed consent, safety monitoring, and the right to stop early when a clear signal of benefit or harm appears. Trials are experiments on people. The oversight exists because we’ve learned, painfully, what happens without it.

How This Helps Patients

A therapy can feel promising and still fail to help, or cause harm that only emerges with systematic observation.

In 1989, the CAST trial asked a simple question: do antiarrhythmic drugs reduce sudden death after heart attacks? The drugs suppressed irregular rhythms on monitors, which seemed like a good sign. But the trial revealed the opposite, patients on the drugs died more often. Tens of thousands of people may have been harmed before the trial provided clarity.

Trials reduce the risk of adopting treatments that look logical but prove dangerous. They also protect patients from hype, scarcity, and unfair access. When a treatment is proven, the argument for access becomes harder to deny.

How This Helps the Clinic

Trials tell teams what is feasible, not only what is effective.

If an answer requires weekly imaging, 24-hour lab turnaround, or a visit cadence that normal clinics cannot provide, adoption will lag. I’ve started reading methods sections with “clinic eyes”: asking not just “did it work?” but “could we do this here?”

When a landmark trial demands infrastructure your site lacks, the result may be real but not reachable. That gap between efficacy and implementation is where a lot of good intentions go to die.

A Tiny Example You Can Feel

Two treatments for the same cancer stage are compared. The new therapy is given in the morning at a well-staffed academic center. The old therapy is given in the afternoon at a community site with longer travel times and fewer translators.

Before the first dose, the groups already differ. If the new therapy wins, is it the drug—or the setting?

A clean trial neutralizes those differences by randomizing within sites and standardizing follow-up. Outcome differences get tied to treatment, not clinic quirks. That’s the machinery working.

People Who Make It Work

A protocol is a document. A trial is a human system.

Behind every enrollment number is a network of people solving problems the protocol never anticipated. I didn’t appreciate this until I watched a coordinator spend forty minutes on the phone helping a patient arrange transportation for a day-three visit. That visit mattered for the data. The coordinator made it possible.

Principal investigators hold scientific and regulatory accountability, but they rarely consent every patient or enter every data point.

Clinical research coordinators are the connective tissue—scheduling visits, tracking labs, fielding patient questions at odd hours, catching errors before they become protocol deviations.

Sub-investigators see patients, make eligibility judgments, and manage adverse events in real time.

Data managers translate bedside observations into analyzable records, flagging inconsistencies that could unravel months of work.

Regulatory specialists keep the study compliant with evolving rules across jurisdictions.

Pharmacists manage investigational product—storage conditions, dispensing logs, accountability down to the last vial.

Study monitors visit sites to verify that what is recorded matches what happened.

Patients and caregivers carry the largest burden. They travel, wait, tolerate extra biopsies, and trust that the uncertainty is worth it.

When a trial succeeds, credit flows to the therapy. When it fails operationally—slow enrollment, missing data, protocol deviations—the gap is almost always a human one. I want to understand those gaps. Future posts will explore these roles in detail.

Spotlight: Emil Frei III and Emil J Freireich

In 1956, two physicians at the National Cancer Institute had an idea that seemed almost naive: what if cancer researchers at different hospitals stopped working alone and started working together?

Emil Frei III and Emil J Freireich founded what would become the Cancer and Leukemia Group B, one of the first cooperative oncology groups. Their first trial compared two dosing schedules of chemotherapy for acute leukemia. The protocol was eight pages long. It enrolled 168 patients across multiple institutions and was published in 1958.

That trial introduced two ideas that now seem obvious but were radical at the time: combination chemotherapy and randomization in cancer treatment. Their later work established that treating patients with minimal disease, before relapse, could extend survival. This became the foundation for adjuvant chemotherapy.

Before Frei and Freireich, cancer treatment was largely a matter of individual judgment. After them, it became a matter of collective evidence. The cooperative group model they built now enrolls tens of thousands of patients each year and has shaped the standard of care for nearly every major cancer.

Question I’m Carrying Forward

Should trials report feasibility metrics, like days from abnormal result to first dose alongside efficacy? A single time anchor might change whether a result gets adopted in real clinics or just admired from a distance. I don’t know the answer yet. I’m hoping to find out.

What’s Next

The next post will walk through the CONSORT flow diagram: how to read it, what the numbers reveal, and where trials quietly lose their grip on the truth.

Views are my own. Educational only.

Subscribe now