iiinova — where your digital self keeps evolving

Sean Harrison, PhD @sean-h.bsky.social · 21/09/2026

Like many things, this is a sensitivity vs specificity issue. Maximise sensitivity to pick up everyone with a condition/who is eligible. Maximise specificity to NOT pick up people who don't have the condition/who are not eligible.

110

Sean Harrison, PhD @sean-h.bsky.social · 17/09/2026

Ok, so a few things. 1. England has utterly fucked gender clinics, so referrals will be *obviously* be low: if you can't get even a first appointment, you're not going to be referred for surgery. The England referral rates are *substantially* lower than the need.

120

Reposted by Sean Harrison, PhD

Cal Horton (they/them) @fiercemum.bsky.social · 09/09/2026

Anyone at the mass lobby today. You can tell them about my kid. Supported at school for years. Now out of school - driven out by hateful segregationist school policy introduced by Labour. Trans kids are being persecuted by this government

527078

Reposted by Sean Harrison, PhD

Sean Harrison, PhD @sean-h.bsky.social · 19/08/2026

I spent 18 months doing reviews for the past, current, and future uses of AI in public health. They're done and preprinted now, so time for a post. This will be *long*, so, tldr in image. Research funded by NIHR, opinions expressed are mine alone, etc. etc. AI was *not* used to do these reviews.

"There is only limited evidence for the use of AI in public health: the evidence that exists does not strongly support AI-assisted interventions for reducing alcohol drinking, smoking, of physical inactivity, and while there are many proposed uses of AI in public health, there is limited or no evidence for whether those uses will be beneficial, and there are many risks of using AI, some of which are existential in nature."

From: https://www.exeter.ac.uk/research/groups/medicine/esmi/workstreams/nihrpublichealthreviewteam/#a1

134

Sean Harrison, PhD @sean-h.bsky.social · 19/08/2026

134

Sean Harrison, PhD @sean-h.bsky.social · 03/08/2026

That's one take. Mine is that, when I die, I hope that people can't justifiably say that I had made their life, and the lives of their loved ones, substantially worse. When oppressors and abusers fall, the oppressed and abused have no compunction to be polite.

020

Sean Harrison, PhD @sean-h.bsky.social · 25/07/2026

This is actually really fucking problematic. Forgetting the Gemini response, which I listen to only if I know the answer going in an need a reminder, how we can now trust any source on the internet? I've no idea if any of this is true, and now I can't trust any primary internet sources. Fuck.

Gemini response to the question: "are there submarines in the great lakes":

"Yes, there are submarines in the Great Lakes, including historic museum ships, a sunken World War I vessel, and occasional tourist submersibles, but no active military submarines.

Museum Submarines

You can walk through several famous decommissioned military submarines preserved as public floating museums:

USS Silversides: Berthed in Muskegon, Michigan.
USS Cobia: Located at the Wisconsin Maritime Museum in Manitowoc, Wisconsin.
USS Cod: Located in Cleveland, Ohio.
HMCS Ojibwa: Located at the Elgin Military Museum's Port Burwell, Ontario site.
U-505: A captured German U-boat housed indoors at Chicago's Museum of Science and Industry.

Sunken Submarines and Tourism

SM UC-97: A surrendered German WWI U-boat that was scuttled and sank in Lake Michigan off the coast of Chicago in 1921, where the wreck still rests on the bottom.

Passenger Submersibles: Select modern expedition cruise ships, such as the Viking Octantis and Viking Polaris, carry small 6-passenger tourist submersibles that take guests on brief underwater excursions into the depths of the lakes.

Would you like to know more about the history of how these submarines entered the lakes or details on visiting one of the museum ships?"

100

Sean Harrison, PhD @sean-h.bsky.social · 23/07/2026

Apropos of nothing, I had a peer review back recently that was literally red pen on my manuscript, the only changes being: 1. Use passive voice (no, I use active voice to be clearer) 2. Remove the bullet points (no, bullet points are sometimes useful) That was it. bsky.app/profile/sean...

010

Sean Harrison, PhD @sean-h.bsky.social · 23/07/2026

Counterpoint: Visibly (loudly, obnoxiously, whatever) announcing paternity leave is a good thing, as it normalises an experience that is still frowned upon in many businesses, and despite the UK having pitiful statutory paternity leave, is still under-taken (59%). www.gov.uk/government/c...

gov.uk

Parental leave and pay review: summary of existing evidence

100

Sean Harrison, PhD @sean-h.bsky.social · 21/07/2026

Delegating thinking to someone or something else requires that person or thing to be benevolent or regulated to prevent (or mitigate) abuse. Are LLMs and the people who make them benevolent or regulated? bsky.app/profile/sean...

030

Sean Harrison, PhD @sean-h.bsky.social · 11/06/2026

I wrote my first academic article because the author of a different paper wouldn't listen when I said they were wrong.

000

Sean Harrison, PhD @sean-h.bsky.social · 05/06/2026

Hi, I was a genetic epidemiologist for a while. You can edit the genes to remove only a very small proportion of diseases, the ones that affect a single gene in a deterministic way. Anything else is fantasy: genetics are predictive of disease, we have no idea which are causal.

192

Sean Harrison, PhD @sean-h.bsky.social · 04/06/2026

I once helped with finding evidence for the public health consequences of a week-long power shortage in the UK. Among the questions asked were: how many calories do people need to survive for a week? How much water? How infectious are dead bodies?

100

Sean Harrison, PhD @sean-h.bsky.social · 30/05/2026

I signed this! NHS England, or at least the people who commissioned the reviews, may be ideologically opposed to healthcare for trans kids, but they can't hide behind their intentionally awful evidence reviews to justify the ban they imposed on hormone treatments.

074

Sean Harrison, PhD @sean-h.bsky.social · 28/05/2026

How the fuck do you interpret the results without knowing the methods? Who were the population? How were they selected? Which country? When? Where the eligibility criteria justifiable? Does anything about how they were recruited lend itself to selection bias?

180

Sean Harrison, PhD @sean-h.bsky.social · 28/05/2026

Because AI doesn't just replace labour, it replaces thinking. Software to improve efficiency helps create art (and science, technology, etc.). Software to replace thinking helps create slop no one wants.

122

Reposted by Sean Harrison, PhD

Sean Harrison, PhD @sean-h.bsky.social · 26/05/2026

I assessed these reviews when they were first published, and agree with almost all of what Gideon says. The research questions were designed to exclude any evidence that might exist. Biggest issue: they excluded studies where trans kids received GnRH analogues (puberty blockers) before hormones.

111

Sean Harrison, PhD @sean-h.bsky.social · 26/05/2026

111

Sean Harrison, PhD @sean-h.bsky.social · 17/05/2026

For a non-academic article about the history of aspirin, I bought a physical copy of a 1940s German journal from Polish eBay to check a reference that didn't exist online. I transcribed the article from German & Google translated. Then put it online for everyone. press.asimov.com/articles/asp...

I went to some length to find a copy of this article. It didn’t exist online, but I found a physical copy for sale on eBay in Poland. Unfortunately, the seller wouldn’t initially ship to me in the UK, so I used Google Translate to ask them (in German) to allow shipping to the UK, which they very kindly did. Subsequently, DHL managed to lose the parcel, which is a shame as it was one of the few extant copies of this article in existence. However, the British Library is, thankfully, magnificent: they had a copy, and someone very generously managed to dig it out of the archives and photograph it for me. I transcribed the German text, translated it with Google translate, and put both the original German transcription and the English translation on the Internet Archive for posterity. Weeks later, DHL found the article and delivered it, so I scanned the original and uploaded it to the Internet Archive.

030

Sean Harrison, PhD @sean-h.bsky.social · 05/05/2026

If modern life == comfort: Indoor plumbing (pipes, pumping stations, water purification). I just spent 3 days with my stopcock closed to stop my boiler from exploding, and it was awful. I can live comfortably without everything on that list. I cannot live comfortably without indoor plumbing.

010

Sean Harrison, PhD @sean-h.bsky.social · 04/05/2026

Lol, this is the funnel plot from the paper; there was no forest plot, for whatever reason. This is a good example of what a funnel plot should NOT look like (points should be symmetrical around the pooled effect estimate, not skewing massively in one direction).

A funnel plot showing massive right-sided skew.

Interpretation: smaller studies (technically, more precise studies, but on average this will mean larger) massively favoured ChatGPT being good, larger studies did not.

This is collectively termed "small study bias": could be a result of publication bias (smaller studies showing no beneficial effect of ChatGPT not published), or crappy study designs more at risk of bias, or other, more... self-serving reasons...

https://www.nature.com/articles/s41599-025-04787-y/figures/2

260

Sean Harrison, PhD @sean-h.bsky.social · 30/04/2026

My job is evidence synthesis in epidemiology/public health: I have read through tens of thousands of abstracts & thousands of full texts in the past year alone. I've been doing this for over a decade. I go through any studies included in my reviews in detail, including rigorous critical appraisal.

131

Sean Harrison, PhD @sean-h.bsky.social · 26/04/2026

Ok, so: this was met, justifiably, with many scathing responses. But what separates this from a *lot* of these types of things I have seen is that @barried.bsky.social replied to those responses with what I saw as real humility, and it looks like learning happened. And that's just so good to see.

110

Sean Harrison, PhD @sean-h.bsky.social · 05/03/2026

This feels identical to kids using LLMs to do their homework: the point of an analysis plan isn't (entirely) to do an analysis plan, it's to think through your analysis, which informs data collection and interventions, and does a lot of work up-front you'd otherwise have to do later.

110

Sean Harrison, PhD @sean-h.bsky.social · 03/03/2026

Lukewarm take: the potential proportion variance explained in the outcome by the totality of the dataset is the hard ceiling for the predictive capacity of that dataset. Different methods aim to capture as much of that information as possible, but many models could be close to that ceiling.

210

Sean Harrison, PhD @sean-h.bsky.social · 22/02/2026

The problem, I think, is that science is often taught as a noun: "This is *the science*, learn it, memorise it, regurgitate it." I think we'd probably be better off teaching science as a verb: "Today, we're going to learn how to science the shit out of this." Skill-based learning >>> facts

130

Sean Harrison, PhD @sean-h.bsky.social · 19/02/2026

The gender roles that are forced on both women (and men) from before birth are so pervasive and dominant that I'd be astounded if there weren't any difference in measurements of "child rearing" between genders. Society (in the UK and elsewhere) is constructed such that "women raise children".

120

Sean Harrison, PhD @sean-h.bsky.social · 11/02/2026

That's clearly Hugh Bonneville's voice, presumably created by AI by sampling his actual voice. That's fucking *awful*. I despise this whole thing, but taking someone's voice to dub over an AI-generated video of anything you want is ghoulish.

040

Sean Harrison, PhD @sean-h.bsky.social · 09/02/2026

The idea that "novelty" should be appraised in academic papers is detrimental to science as a whole, and should fuck off. Incentives for novelty are incentives for p-hacking, deception, and shitty research. Then again, I also think journals should fuck off, so...

120

Sean Harrison, PhD @sean-h.bsky.social · 26/01/2026

I've probably read over 300 books in the last year! Out loud! With all the voices! To my toddler!

0332

Sean Harrison, PhD @sean-h.bsky.social · 22/01/2026

Ok, let's do some weird things that *might* happen if you were assigned male at birth and use hormone replacement to transition - there's limited research, but these are things I've seen in more than one place, including here: genderdysphoria.fyi/en/ 1. Eyes may become lighter in colour

genderdysphoria.fyi

The Gender Dysphoria Bible

A dive into the multitude of ways that gender dysphoria manifests and what it means to be transgender.

141

Sean Harrison, PhD @sean-h.bsky.social · 20/01/2026

"Belgium, Belgium, Belgium is the best!" www.youtube.com/watch?v=FtbF...

youtube.com

Belgium is the best

YouTube video by piesoe

000

Sean Harrison, PhD @sean-h.bsky.social · 16/01/2026

Unpopular opinion (from someone who works in evidence synthesis): Meta-analyses shouldn't be at the top of the triangle, the should be alongside RCTs. The best RCT will have better information quality than the best meta-analysis (on the same topic), unless the meta-analysis only includes one RCT.

210

Sean Harrison, PhD @sean-h.bsky.social · 09/01/2026

The Secret Diary of The Children's Crusade

153

Sean Harrison, PhD @sean-h.bsky.social · 09/01/2026

For the past year, I've been working on some evidence reviews for the role of AI in public health, and am a little sad I can't include the following quote in any of them:

"It's all a massive stachel full of wank"

010

Sean Harrison, PhD @sean-h.bsky.social · 07/01/2026

Scrubs has to be up there. Ignoring medschool and the new series...

010

Sean Harrison, PhD @sean-h.bsky.social · 16/12/2025

When I read through parts of the Cass review when it came out, I thought the same as Chris here. Big difference between "absence of evidence" and "evidence of absence", and even academics sometimes struggle with the difference. If there's absence of evidence, you need more evidence! Good letter.

000

Sean Harrison, PhD @sean-h.bsky.social · 15/12/2025

I went back in time as far as I could go, and the first post was DELETED. bsky.app/profile/mlob... Damn.

110

Sean Harrison, PhD @sean-h.bsky.social · 15/12/2025

I despise that "novelty" is a driver of anything to do with publication in epidemiology. It is an incentive to p-hack, data mine, fabricate data, conduct analyses seeking specific results etc. If it is robust and adds to the evidence base, then it should be published.

210

Reposted by Sean Harrison, PhD

Kit Yates @kityates.bsky.social · 14/12/2025

King Charles argued on Friday that, thanks to early diagnosis, his treatment will be reduced in the New Year. He urged people to take up cancer screening saying: “Early diagnosis quite simply saves lives.” www.bbc.co.uk/news/a... 1/10

bbc.co.uk

King Charles 'deeply touched' by reaction to cancer TV message, says Buckingham Palace

In a TV broadcast on Friday night, the King said an early diagnosis was key to the "good news" that his treatment is being scaled back.

99034

Reposted by Sean Harrison, PhD

Tom Hollenstein @tomhollenstein.bsky.social · 07/12/2025

So...my undergrad thesis student is doing a quality analysis of studies found in meta-analyses. She identified a few and we contacted the authors to request their effect sizes and other variables for the studies in their papers. Here's what happened: scientiapsychiatrica.com/index.php/Sc...

scientiapsychiatrica.com

The Impact of Social Media on Adolescent Mental Health: A Meta-Analysis | Scientia Psychiatrica

Introduction: The proliferation of social media has raised significant concerns about its potential effects on the mental health of adolescents. This meta-analysis aims to provide a comprehensive asse...

514675

Sean Harrison, PhD @sean-h.bsky.social · 01/12/2025

Going through references is such a mind-numbing, irritating (paywalls, shitty journals, etc.), and thankless task, but absolutely necessary to ensure statements are backed up by evidence. I wrote about this a while ago when trying to find evidence for a claim: seanharrison.blog/2025/02/21/i...

seanharrison.blog

If we don’t reference properly, we end up standing in bullshit

Or: Do 30 million people really take NSAIDs every day? I wanted to know how many people in the world took non-steroidal anti-inflammatory drugs (NSAIDs). You know, aspirin, ibuprofen, diclofenac, n…

130

Sean Harrison, PhD @sean-h.bsky.social · 28/11/2025

All screening causes harm. Some screening causes benefit. You need studies and stats to tell you whether the benefit outweighs the harm.

0113

Sean Harrison, PhD @sean-h.bsky.social · 24/11/2025

IHT at 100%: 50% to the Government as tax, 50% into a fund to provide every kid with a "starting out" grant at 18 years old. Tackle inequality at source and help out people just starting, many of whom will need the money most. "But they'll just set up trusts!" -- k, guess we'll do nothing!

230

Sean Harrison, PhD @sean-h.bsky.social · 24/11/2025

I believe, for policy, the effect size doesn't matter by itself (unless it's 0): it only matters in conjunction with costs and QALYs (or similar), including side-effects. Cost-effectiveness matters most: a smaller effect size but better side-effect profile or cheaper intervention may be better.

110