Thread by @PeterWyrtz

How can we personalize protection of COVID-19 risk groups beyond age and existing conditions? New @NgaleHealth study identifies a blood biomarker score predicting risk of severe COVID-19, before contracting the virus
medrxiv.org/content/10.110… #safereturntowork

We measured blood biomarkers by @NgaleHealth metabolomics in >100,000 samples from @uk_biobank. Blood samples were collected 10 years ago, metabolomics data became available late-May. This initiative will lead to many papers - we started with COVID-19. Summary on what we found 👇

Various blood biomarkers have been linked with COVID-19 severity, but most studies use samples from already infected patients. For preventative screening we need pre-COVID-19 samples, lots of samples, to study enough people with severe COVID-19 vs many metabolic biomarkers.

In ~100,000 @uk_biobank samples, there were 195 people with COVID-19 diagnosis in hospital, deemed severe cases. We focused on 37 blood biomarkers with clinical approval in @NgaleHealth metabolomics platform. Here are the associations 👇

Statistical power was limited despite large sample size. This is due to limited number of COVID-19 cases among the 100,000 people, even in hard hit places like the UK. Also, there was 10+ years from blood collection to the pandemic, which might weaken the biomarker associations.

To overcome these challenges, we exploited the shared risk factor relations between COVID-19 and pneumonia, as shown recently medrxiv.org/content/10.110…. Pneumonia is the most common diagnosis in severe COVID-19, and a leading cause of death worldwide even before the pandemic.

Also, risk factors for COVID-19 and pneumonia are similar incl high age and a compromised immune system. We tested blood biomarker associations vs severe pneumonia; there were 2507 hospitalizations or deaths during ~8-year follow-up.

Many lipoprotein lipids, fatty acids, amino acids and other biomarkers were predictive of future risk for severe pneumonia, with strong statistical evidence. The overall pattern of biomarker associations for severe pneumonia was highly concordant with that for severe COVID-19

Because the molecular signature reflective of risk for severe COVID-19 and severe pneumonia were similar, we used the well-powered analyses of pneumonia to develop a multi-biomarker score and tested how that could predict long-term and short-term risk.

Using machine learning in half of the study population, we derived a multi-biomarker infectious disease score, comprised of 25 blood biomarkers incl inflammatory proteins, fatty acids, amino acids and advanced lipid measures (marked with # in odds ratio plot 👆).

The multi-biomarker infectious disease score was strongly predictive of severe pneumonia in the validation half of the study population (n=52,573, 1250 events). It was ~twice as strong as any of the individual biomarkers.

Since all the biomarkers are quantified simultaneously in the @NgaleHealth NMR metabolomics platform (nightingalehealth.com/research/blood…) this is effectively a single biomarker test; clinically certified and highly scalable (and low cost).

The multi-biomarker infectious disease score was equally strong predictor of severe pneumonia for men and women, and across age groups. Results were also similar when analysing only people free of heart and lung diseases at blood sampling, and when adjusting for BMI and smoking.

The multi-biomarker infectious disease score was also predictive of risk for severe COVID-19, with stronger magnitude than any of the individual biomarkers.

The multi-biomarker score was predictive of severe COVID-19 for both men and women. We observed no clear effects of age. Study participants were 49-84 at time of the COVID-19 pandemic.

The multi-biomarker score was attenuated but remained predictive of severe COVID-19 risk for when adjusting for BMI, smoking and prevalent disease. Statistical power was very limited for these sub-analyses.

While the association of the biomarker score with COVID-19 was statistically robust, the magnitude was still somewhat modest. Could this be due to UK Biobank blood samples being collected 10+ years ago, so no longer reflecting a molecular snapshot of participant's health in 2020?

To explore the influence of the long lag from blood sampling to the pandemic, we again turned to pneumonia, where hospitalizations and deaths occurred gradually after blood sampling and timing was recorded in UK nationwide registries. #ThankYouNHS

First, we mimicked the decade-long lag from blood sample collection to the COVID-19 pandemic. The association of the multi-biomarker infectious disease score with severe pneumonia occurring between 7-11 after blood sampling was ~half the magnitude compared to earlier events

So we speculate that also the biomarker associations with risk for severe COVID are attenuated to at least half the magnitude. What if we then interpolate to short-term risk, mimicking a preventative screening scenario carried out today?

We found that the multi-biomarker score predicts severe pneumonia events that occurred during the first 2 yrs after blood sampling twice as strong as for later events. And ~3 times as strong as for events during 7-11 years from blood sampling.

In other words, the odds for severe pneumonia within 2 yrs from blood sampling is 8-fold for those with high biomarker score levels compared to those with low levels. This is among generally healthy volunteers in @uk_biobank. Prospective biomarkers are usually never that strong!

These results were again robust to adjustment for BMI, smoking and existing disease at time of blood sampling. The odds ratio was still ~6 between upper and lower quintiles of the multi-biomarker score.

The full time-resolution of the cumulative risk for severe pneumonia according to quantiles of the multi-biomarker infectious disease score is shown in this Kaplan-Meier plot

The time-resolved plot provides more insights: pneumonia risk remains low for most people, with prominent increases for those in upper tail of the multi-biomarker score. In association testing of highest percentiles vs low risk group (4 out of 5 people), results look like this

That is 5-fold higher risk for severe pneumonia for the 20% in the high-risk group compared to everybody else, and 10-fold risk for those in the highest 5% of the multi-biomarker score compared to the 80% with low risk.

These results were similar even when focusing on those without cardiometabolic and respiratory disease at the time of blood sampling. We again observed a small subgroup with much higher risk than everyone else.

Another way to look at increase in the upper tail of the multi-biomarker score is using risk gradients, inspired from polygenic risk score papers @amitvkhera @skathire @minouye271

Due to mathematical properties underlying the multi-biomarker scores, it may not be surprising to see risk upshot in the upper tail, but earlier metabolomics studies have not had the sample size to demonstrate this. These non-linear effects suit screening for high-risk.

Another neat feature is high measurement repeatability of the biomarker score, and good consistency in repeat samples years apart. This compares favorably to common inflammatory proteins #WhatAboutCRP

In summary, the molecular signature of biomarker associations is similar for COVID-19 and pneumonia. Association magnitudes are similar for long-term risk, AND the biomarker score predicts short-term risk for severe pneumonia VERY strongly, with non-linear effects in upper tail.

By analogy, if a similar elevation in short-term risk extends to severe COVID-19, this can be used for personalized COVID-19 prevention. Such a predictive tool could extend detection of COVID-19 risk beyond the current emphasis on older people and those with adverse conditions.

The scalability and low cost of this blood biomarker test means that it could be applied at population scale to help protect those at highest risk of severe COVID-19 and support safer back to work strategies during the phased easing of the lockdown restrictions across the world.

Moreover, once a vaccine becomes available this test could stratify high risk patients to prioritize those at most need of the vaccination.

Check out the full paper here medrxiv.org/content/10.110…

Many thanks to @uk_biobank for making this study possible. And to @helijulkunen @CichonskaAnna and Eline Slagboom @LUMC_Leiden for rapid execution. We welcome feedback.

This is the first of many papers to come from the Nightingale Health UK Biobank Initiative. youtube.com/watch?v=NHZ9k1…
Similar strong biomarker scores can be derived for numerous chronic diseases, and apparently even infectious diseases.

Oh. And we are announcing today the launch of this preventative screening solution. The risk identification works well even when focusing on a subset of biomarkers in Nightingale’s blood test that can be captured by self-collection.
nightingalehealth.com/news/nightinga…