iiinova — where your digital self keeps evolving

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

A big thanks to all the co-authors, particularly Karin Knudson who led this work, and @YODA for making data like this accessible for scientific inquiry. We welcome any feedback from the scientific community! (11/n)

000

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

This strategy - of using causal machine learning to identify factors associated with improved efficacy, either through reducing placebo response or amplifying treatment response – is applicable to other therapeutic areas, particularly those with strong clinical response to placebo treatments. (10/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

Excluding such patients increased placebo-adjusted efficacy – and accompanying statistical significance – of the overall trials. (9/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

We distilled these complex causal ML models to much simpler decision frameworks to show that small fractions of patients with extremely high disease severity responded well to either active treatment or placebo in AAD clinical trials. (8/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

This was especially true when considering study-level (rather than traditionally-reported patient-level) replication. (7/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

These causal ML models we developed in AAD showed little degradation in accuracy when moving from training to completely independent testing data, a rare feat in machine learning applications to clinical trial datasets. (6/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

It offers many advances over more traditional strategies like only analyzing patients that were randomized to placebo (which uses ~half the available data and might identify factors that impact drug and placebo response equally) (5/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

Causal ML models estimate “CATE” (the conditional average treatment effect) for each patient that considers how they would competitively respond to treatment A versus B. (4/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

We therefore developed causal machine learning (ML) models that accurately predicted clinical response to antipsychotic versus placebo treatment using only baseline/pre-treatment clinical and demographic characteristics in historical trials. (3/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

Alzheimer's disease (AAD) is a highly prevalent and debilitating neuropsychiatric symptom that often precedes cognitive decline - one of the largest barriers to developing new treatments in AAD has been robust placebo responses within clinical trials for new therapies. (2/n)

100

Andrew Jaffe @aejaffe.bsky.social · 11/06/2026

Thrilled to share our preprint around using causal machine learning to reduce placebo response in neuropsychiatric clinical trials – this new direction of scientific research has probably been as far from functional genomics as you can get :) www.medrxiv.org/content/10.6... (1/n)

medrxiv.org

Reducing placebo response in clinical trials of agitation in Alzheimer’s disease

High placebo response is an obstacle in developing drugs to treat agitation in Alzheimer’s disease (AAD), a prevalent and burdensome symptom. However, it has proved challenging to develop actionable m...

110

Andrew Jaffe @aejaffe.bsky.social · 13/03/2025

Abstract: 'Rapid advancements in high-throughput *sequencing technologies* especially the Illumina DNA methylation Beadchip...' not a good start....

020

Andrew Jaffe @aejaffe.bsky.social · 07/02/2025

Thanks for doing this, super useful

000

Andrew Jaffe @aejaffe.bsky.social · 22/11/2024

Yeah, at least the Logitech pc version I have has the 3 profiles as separate keys on the keyboard (1,2,3 keys) and then a single button on the mouse that cycles

010

Andrew Jaffe @aejaffe.bsky.social · 22/11/2024

Eg computer 1 is hdmi, profile 1 on keyboard, profile 1 on mouse and computer 2 is usb-c and profile 2 on keyboard and mouse

110

Andrew Jaffe @aejaffe.bsky.social · 22/11/2024

You don't really need a kvm these days, a lot of higher end keyboards have multiple Bluetooth devices you can toggle through, eg www.logitech.com/en-us/produc... And presumably the monitor has multiple inputs (this is roughly my home setup)

logitech.com

MX Keys S Combo for Mac Wireless Keyboard and Mouse | Logitech

Shop MX Keys S Combo for Mac. A combo optimized for macOS and iPadOS - a keyboard offering superior typing, true Mac layout, improved backlighting, customizable shortcut keys and an iconic mouse with ...

110

Andrew Jaffe @aejaffe.bsky.social · 16/11/2024

Starter packs were a cool idea and made the switch from Twitter way easier.

120

Reposted by Andrew Jaffe

🇲🇽 Leonardo Collado-Torres @lcolladotor.bsky.social · 12/11/2024

I decided to make a starter pack for members/alumni of @lieberinstitute.bsky.social "Data Science Team 1" 💻 Initially led by @aejaffe.bsky.social until 2020/09; then by me. In 2024, the team became part of the Division of Translational Neuroscience led by @martinowk.bsky.social go.bsky.app/53gy4UP

0136

Andrew Jaffe @aejaffe.bsky.social · 08/11/2024

Trying this out after mostly giving up on twitter...

030