iiinova — where your digital self keeps evolving

T.Y. Lim

@tylim.bsky.social

#DrugsAndBugs Postdoc at CCDD HSPH, PhD @mitsloan.bsky.social, formerly @YaleEnvironment. Complex systems & simulation modelling for public health. Standard disclaimers apply. @limtseyang in the other (worse) place.

Posts Replies Media

T.Y. Lim @tylim.bsky.social · 20/11/2024

Addendum - image in third post messed up, here's a clean version!

000

T.Y. Lim @tylim.bsky.social · 20/11/2024

Thanks to our collaborators at MGB, Tufts Medical Center, and @countyofla.bsky.social Dept of Public Health for sharing data for this work! @yhgrad.bsky.social @jameshay.bsky.social End/

120

T.Y. Lim @tylim.bsky.social · 20/11/2024

Overall our work shows analysis of Ct values can complement traditional incidence metrics and surveillance, using already-collected PCR data to help monitor epidemics – extra important if testing resources are limited or absent other public health surveillance 11/

120

T.Y. Lim @tylim.bsky.social · 20/11/2024

It’s worth reiterating that these are extremely simple models – they run in minutes, can be updated frequently, and are well within the computational capabilities and expertise of any local public health agency 10/

110

T.Y. Lim @tylim.bsky.social · 20/11/2024

Crucially, nowcasting models are robust to limited daily sample sizes – an inherent limitation of count-based methods Dropping Ct outliers in each day’s data improves performance, but accounting for population immunity or symptom status does not; all models face clear bias-variance tradeoffs 9/

110

T.Y. Lim @tylim.bsky.social · 20/11/2024

Our simulations showed that several factors confound nowcasting model accuracy – notably, random inter-individual variation in viral kinetics, and non-random sampling delays (from infection to testing) 8/

110

T.Y. Lim @tylim.bsky.social · 20/11/2024

Interestingly, while our two datasets reflect very different ways of detecting infections (hospital admissions testing vs. community testing) and showed different relationships between Ct values and epidemic growth rates, both datasets were useful in predicting epidemic growth rates over time 7/

110

T.Y. Lim @tylim.bsky.social · 20/11/2024

Simple cubic spline regression models using daily reported Ct value distributions can nowcast epidemic [log] growth rates over two-week time horizons to reasonable accuracy (RMSE ~0.04 – for context, growth rates vary from approx. +/- 0.2) 6/

Figure showing model-predicted (black) vs. observed (blue) log incidence growth rates, with 95% CIs (dark) and PrIs (light), and RMSE of predicted vs. observed growth rates for each 2-week nowcasting window.

110

T.Y. Lim @tylim.bsky.social · 20/11/2024

But how well does this work in practice? Here, we did one of the largest tests to date of how well Ct values can nowcast epidemic growth rates (rate of change in cases) under various conditions, using both simulated data and large real-world datasets from Massachusetts and Los Angeles County, USA 5/

110

T.Y. Lim @tylim.bsky.social · 20/11/2024

These orthogonal indicators, which don’t rely on counting cases (e.g., wastewater viral load), are important for building a complete and accurate picture of an outbreak 4/

110

T.Y. Lim @tylim.bsky.social · 20/11/2024

So in theory, viral loads / Ct values from routine or surveillance testing could tell us if outbreaks are growing or shrinking, separate from trends in case counts – providing early warning and an alternative source of information 3/

142

T.Y. Lim @tylim.bsky.social · 20/11/2024

RT-qPCR tests for viral pathogens like SARS-CoV-2 indirectly measure viral loads through cycle threshold (Ct) values. Previous work (doi.org/10.1126/scie...) found that population-level viral load distributions relate to epidemic trajectories 2/

doi.org

Estimating epidemiologic dynamics from cross-sectional viral load distributions

Cycle threshold values from PCR testing depend on epidemic dynamics and can be used to monitor SARS-CoV-2 and other outbreaks.

110

T.Y. Lim @tylim.bsky.social · 20/11/2024

How do you track an ongoing epidemic, especially without large-scale testing? We explored the use of routinely collected RT-qPCR cycle threshold values in our new preprint out now! doi.org/10.1101/2024... 1/

doi.org

Nowcasting epidemic trends using hospital- and community-based virologic test data

Epidemiological surveillance typically relies on reported incidence of cases or hospitalizations, which can suffer significant reporting lags, biases and under-ascertainment. Here, we evaluated the po...

1132