Lucy Teece | Stats Enthusiast

Ban on controversial NDAs silencing abuse The Employment Rights Bill will ban employers from using non-disclosure agreements that silence workplace harassment and abuse.

www.gov.uk/government/n...

Mw holding a yellow fluffy chick toy with yellow glasses up to my face

🐥 I found the chick and got a free coffee ☕
🔎 Now to hunt for 40 easter egg letters!

Anyone fancy helping to solve some 10-letter anagrams later?!
#EasterGames #EasterTreats

🏃‍♀️ 540 runners joggers and walkers today ☀️
Thank you for having me as RD, so much fun!
#ParkRun

Fantastic resource - Thanks for sharing!

May be of interest for the @rssdiversity.bsky.social members?! 👀

Email from Rushcliffe Parkrun tanking volunteers and reminding them that this weeks Run Director will be me (Lucy Teece)!

📣It's official - I'm Run Director at #ParkRun this week 😁

🏃🚶🐕

Selection of Easter treats including rabbit shaped chocolates, chick shaped gummy sweets, and a small pen with a rabbit topper

Thank you #EasterBunny for my (very) early gift 🎁

Silly question - do you have multicoloured thread or are you changing colours? It's beautiful 🤩

Sunset on the top of a hill with a tall white structure (known as White Nancy) and a smiling Lucy

Me, White Nancy, and the sunset 🥾🥳

Looks interesting, adding to the pile!

Photo of me in and Alec holding up coat hooks that we made ourselves wearing yellow aprons

Adding blacksmithing to my CV ⚒️

Welcome to the Royal Statistical Society Bluesky and a hello to our new followers 👋

For the latest stats and data insights here is our starter pack of RSS fellows and groups ⤵️

Yay! What ones have you got lined up?!
I need a stats / popsci restock
📕📙📒📗📘

Always happy to boost posts and sessions about #WomenInData #WomenInStats and #WomenInSTEM
But it feels particularly important this year!

👇 This looks like a fantastic event

Please share more with me (what tags and packs should I be following?) 🙏

A great way to find out about upcoming #Stats events is to follow the @royalstatsoc.bsky.social starter pack 👇

go.bsky.app/HbNRCFp

#StatSky #Data

2.2 Elements of Phase I-II Designs One may consider a phase I-II design to be like a Labradoodle, which is ob-tained by cross-breeding a Labrador Retriever with a Poodle. Labradoodles enjoy the desirable quality of heterosis, or hybrid vigor, since they have greater genetic diversity and hence better overall health than either of their parents. Hybrid clinical trial designs, if carefully constructed, enjoy similar advantages over the simpler designs that motivated them. The analogy is not strict, how-ever, since both Labradors and Poodles both are fine canine breeds, whereas the most commonly used phase I or phase II clinical trial designs suffer from logical and ethical flaws, as we demonstrated in Chapter 1

😄 When you find an analogy that perfectly fits, and you know you're going to use it to explain what you do...

"Consider a phase I-II design to be like a Labradoodle" 🐕🐩

a cartoon cat is dancing with the words let 's make plans then written below it . Alt: a cartoon cat is dancing with the words let 's make plans then written below it .

I'm excited by this year's programme of events at @royalstatsoc.bsky.social

Had a sneaky peek of plans from the @rssdiversity.bsky.social, @rssedinburgh.bsky.social, and draft #RSSConf programme - there's a lot of great things coming and many many more groups and sections still to announce

Meme of Hot Fuzz "Greater Good"

How would you communicate that being a "logical robot" 🤖 is better for the "greater good", and that the emotional response to deviate from the plans is not a sensible thing to do?

It's challenging to communicate how to differentiate between decisions made in medical practice compared to scientific experiments

#Trials follow rational, ethical processes to maximise benefit to patients
Hearing "real" patient stories can make it difficult to not let emotion impact decisions

Sandi at a podium presenting a picture of Ada Lovelace

Poster with apple

Sandi and Ada, a great Friday night in Cambridge

Don't miss this unique opportunity to be one of the next RSS William Guy Lecturers- highly recommend the experience! 🤗

🚫 Traditional Phase I expansion cohorts have many problems and poorly serve patients

👎 Fail to adequately protect patient safety
👎 No true experimental design or stopping rules
👎 Poor job optimising dose or schedule
👎 Numerous logical problems

💡Solution: Phase I-II sequentially adaptive designs

Extract from trials book: These examples may be summarized by the following key points for settings where both Efficacy and Toxicity may be observed quickly enough to make adaptive decisions. 1. 3+3 algorithms do a poor job of choosing a MTD. 2. Conventional phase I sample sizes are far too small. 3. It often does not make sense to base dose-finding on Toxicity alone. 4. It does not make sense to design a phase II trial based on Efficacy alone, since safety is an important concern in any clinical trial and the trial must be stopped if excessive Toxicity is observed. 5. The inclusion of a rule based on Efficacy to stop for futility in phase I-II underscores the important point that, at the completion of a phase I-II trial, there is no reason to conduct a conventional phase II trial since it already has been done.

Summary:
🛑 The 3+3 design is inferior
☣️ Phase I designs focus on toxicity
📈 Ignoring efficacy information could be unethical if efficacy-toxicity relationship assumptions incorrect

Trying to convince your team to plan slightly larger phase I-II trial (over phase I alone)...

"A phase I trial is not an obstacle to overcome quickly.

Rather, if possible it should be a Phase I-II trial... Regarded as an experiment to obtain precious information to optimise a new agent."

#StatSky

Happy International Women & Girls in Science day 👩🏻‍🔬🧑🏼‍💻👩🏾‍🔬

Always a great opportunity to get online and learn more about some fantastic historical innovators

#WomenInScoence #WomenInSTEM
@rssdiversity.bsky.social

📖 Starting "Bayesian Design for Phase I-II Clinical Trials" today 🤓

"These first two chapters are written so that they can be understood by nonstatisticians seeking enlightenment or entertainment" 💡
Perfect fit for my February mornings 🌞

doi.org/10.1201/b21600
#StatSky #OncSky #Stats #Bayes

Developments and alternatives include:

🟰 𝗘𝗾𝘂𝗶𝘃𝗮𝗹𝗲𝗻𝘁 𝗧𝗼𝘅𝗶𝗰𝗶𝘁𝘆 𝗦𝗰𝗼𝗿𝗲 (𝗘𝗧𝗦):
which uses modified isotonic regression

🕵️‍♀️ 𝗤𝘂𝗮𝘀𝗶-𝗖𝗥𝗠:
which incorporates toxicity scores using a Bayesian CRM framework

🧩A composite toxicity score was introduced:
𝗧𝗼𝘁𝗮𝗹 𝗧𝗼𝘅𝗶𝗰𝗶𝘁𝘆 𝗕𝘂𝗿𝗱𝗲𝗻 (𝗧𝗧𝗕)

⚖️ Severity weights (representing patient burden) defined for each toxicity, gives toxicity score
☠️ Grade 5 (deaths) not given weights
📈 Models continuous toxicity score as a function of dose

The model based CRM design has been extended to model ordinal outcomes:

✅ Proportional Odds model
✅ Continuation Ratio model

⚠️ Patients toxicity 🟰 highest grade experienced

🔎 Dose-finding model used to estimate probability of toxic response at each toxicity grade

The 𝗔𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗲𝗱 𝗧𝗶𝘁𝗿𝗮𝘁𝗶𝗼𝗻 𝗗𝗲𝘀𝗶𝗴𝗻 was introduced to account for the natural ordering of toxicities

Uses an algorithmic framework 🤖

Now two grade-2 toxicities in a cohort (2️⃣+2️⃣), results in same decision as a single grade-3 (3️⃣+0️⃣)

✂️ Dichotomising this ordinal grading into a binary safety endpoint...

❌ treats non-DLT toxicities as irrelevant
❌ignores harmful combinations of low-grade toxicities