iiinova — where your digital self keeps evolving

Casey Middleton, PhD @caseymiddleton.bsky.social · 27/08/2026

Ten years ago, Beans was put up for adoption because a family member was abusing him. If you ask me, she should’ve divorced the man and kept the dog. Happy National Dog Day, Beanbo Doggins. You’re the coolest little dude I ever met!

150

Casey Middleton, PhD @caseymiddleton.bsky.social · 09/07/2026

The beauty of this setup? No clutter. I can instantly filter my library by a specific project or pathogen when I'm writing. If I need to dig into the weeds, I just open the page to see my detailed notes, without ruining my clean database overview (see example below).

Screenshot of Notion relational database with notes detailing the sampling scheme and modeling methodology for a SARS-CoV-2 viral kinetics study.

120

Casey Middleton, PhD @caseymiddleton.bsky.social · 09/07/2026

Every entry in my database tracking a paper requires: • Title & DOI • Status (skimmed, read, etc.) • Summary (my main takeaway for MY work—not just the abstract!) I also use specific relational tags: • LaTeX citation key • Relevant projects • Pathogen studied and take detailed notes on key papers.

Screenshot of Notion Database with the following entries for each manuscript:
1. Name: a manuscript identifier with last author and year
2. Title
3. Summary
4. DOI
5. Status: Read / Skimmed / Abstract / Methods
6. Relevant projects
7. Pathogen

100

Casey Middleton, PhD @caseymiddleton.bsky.social · 01/06/2026

Last night, we celebrated two birthdays: Cecil Earnest Middleton, born May 23, 1941 Casey Elaine Middleton, born June 4, 1996 Highly recommend celebrating your birthday with your granddad ✨

040

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/05/2026

🤷‍♀️ We have shown that the adjusted parameterization approach improves model accuracy, but does it change model predictions? Well, predicted herd immunity thresholds are drastically lower under adjusted parameterization! *Note that this work is not meant to establish real-world vaccination targets.

SIR-predicted herd immunity thresholds under standard parameterization and adjusted parameterization. HITs are lower under the adjusted parameterization.

140

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/05/2026

We also extend their adjusted parameterization approach to a heterogeneous vaccine effect model. The improvements, or reductions, in model accuracy from this adjusted parametrization depend on how well our model captures the true distribution of individual-level vaccine-derived protections.

Heatmap showing model prediction error when the distribution of vaccine-derived protection is assumed correctly or incorrectly by the model, under the adjusted parameterization. Error is 0 when the model assumes the correct distribution. If the model assumes protection is leaky and it is truly all-or-nothing, or vice-versa, error is high.

130

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/05/2026

✨ In hopes of changing this, we use a simulation study to show that the Shim & Galvani method makes leaky model predictions more accurate. While the adjusted parameterization reduces error across all scenarios, it performs best when solved using parameters that mirror the context of the VE study.

Line plot showing that adjusted parameterization improved model accuracy even when the modeled vaccine coverage is not aligned with the vaccine coverage at the time of the VE study.

130

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/05/2026

Thus, modelers must consider both the assumed mechanism of action and the statistic used to empirically estimate VE when parameterizing models. 🛑 If a manuscript models both all-or-nothing and leaky vaccine assumptions, the same VE source should not be used for both!

Preferred VE measurement depends on vaccine mechanism. All-or-nothing: cumulative attack rate ratio. Leaky: hazard ratio.

130

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/05/2026

Halloran, et al. show that controlling for exposure (e.g., challenge studies) or using hazards ratio (HR) estimators can more accurately recover direct protection for a leaky vax. However, HR estimators may appear to increase over time for an all-or-nothing vax. doi.org/10.1093/oxfo...

Figure showing how various study statistics estimate VE depending on vaccine mechanism. For leaky vaccines, ARR-based estimates underestimate the true effect, while hazards-based and challenge studies are unbiased, For all-or-nothing vaccines, hazards-based estimators overestimate vaccine effect.

130

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/05/2026

What does this mean for modelers? ARR-based VE estimates generally do not align with leaky model parameters. Statistically speaking: risk ratios != hazard ratios when risk is not low. Plugging these VE estimates into models leads to an underestimate of vaccine effects. 🤔 So what can be done?

Schematic showing that using ARR-based VE estimates to inform leaky model parameters will lead to an underestimate of the modeled population-level vaccine effect.

120

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/05/2026

Lewnard, et al. derive this bias for TND studies. However, all ARR-based VE studies will be prone to this bias if multiple exposures occur. This bias also exists when a vaccine provides heterogeneous protection (i.e., varying levels of leaky protection across recipients). doi.org/10.1093/aje/...

ARR-based VE estimates change with the number of exposures for leaky and heterogeneous vaccines, but not all-or-nothing. More exposures leads to lower VE estimates.

120

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/05/2026

What is VE? Empirical studies often define VE as 1 - ARR, where ARR is the relative attack rate in vaxed vs. unvaxed folks. Smith, et al. showed that ARR-based VE estimates may appear to wane over time for a leaky vaccine, even when true protection is constant. doi.org/10.1093/ije/...

VE estimates using attack rate ratios for leaky vaccines appear to wane over time. For all-or-nothing vaccines, VE approaches the true protection level.

120

Casey Middleton, PhD @caseymiddleton.bsky.social · 28/04/2026

I had to go all the way into the Supplement to figure out how they were really defining "false positive" 🤦‍♀️ Ahhhh, so you mean "likely not infectious"

171

Casey Middleton, PhD @caseymiddleton.bsky.social · 26/01/2026

🎯 The scaled logit model, introduced by A. Dunning for case-control studies, provides this flexibility. We demonstrate that the scaled logit model can accurately infer a wide range of titer-protection relationships by fitting only one additional parameter. doi.org/10.1002/sim....

110

Casey Middleton, PhD @caseymiddleton.bsky.social · 26/01/2026

Using simulated studies and mathematical arguments, we show that logistic regression is ill suited to recover biologically plausible titer-protection relationships in TNDs. ⚠️ This finding is unique to TNDs, which rely on exposure odds ratios (ORs). Thus, we need a model with more OR flexibility!

110

Casey Middleton, PhD @caseymiddleton.bsky.social · 25/01/2026

If you live outside of Minneapolis but want to support the community, Lisa and her husband are doing boots on the ground work to protect families! Donate to their work via Venmo @lisa-szabo-1

011

Casey Middleton, PhD @caseymiddleton.bsky.social · 05/01/2026

I have recently taken up watercolor as a relaxing after-work hobby. This piece captures a subtle sunset over the lake at my parents’ house in Tennessee!

Watercolor painting of a subtle sunset over a lake, featuring light a light yellow to blue gradient in the sky reflected on the water.

130

Casey Middleton, PhD @caseymiddleton.bsky.social · 03/12/2025

When they ask you to bring your talk on a flash drive, and you’re a professional, so you use your Tweety Bird flash drive #Epidemics10

130

Casey Middleton, PhD @caseymiddleton.bsky.social · 01/10/2025

Why, in this incredible world of technological advancements, are published inline equations still blurry? 😵‍💫

140

Casey Middleton, PhD @caseymiddleton.bsky.social · 09/04/2025

A glimmer of good news amidst the chaos! On Monday, I successfully defended my PhD focused on infectious disease modeling. And yes, my thesis title is an acronym for my last name. With gratitude to the incredible friends and mentors who got me here… Dr. Casey Middleton, reporting for duty 🫡🦠

190

Casey Middleton, PhD @caseymiddleton.bsky.social · 19/02/2025

Last week, I had the honor of receiving 2nd place at CU Boulder's 3MT competition. This competition challenges grad students to distill their thesis down into a 3 minute talk for a general audience. My talk focused on how viral kinetics and probability theory inform testing policies... in < 3 mins ⏰

Female researcher in a black button-down shirt presenting her research using animated hand gestures, with a powerpoint slide in the background

Researcher in black button-down shirt shaking hands with presenter as she receives a 2nd place certificate.

2101

Casey Middleton, PhD @caseymiddleton.bsky.social · 23/01/2025

When everything is on fire, just visit coolors.co and get lost in the color schemes you may use in your next manuscript

A panel of four colors, including two shades of blue/green, orange, and purple.

060

Casey Middleton, PhD @caseymiddleton.bsky.social · 17/01/2025

TFW your citation rabbit hole lands you at one of the first examples of vaccine clinical trials 🤓 This trial gained approval for public use of a vaccine that was originally developed to protect troops against influenza A during WWII.

0101