Long covid illustration

Long-term COVID 19 Research Study - Update 1

I am excited to provide the first update on our Long-term COVID-19 study. Firstly, a big thank you to all participants and the long COVID communities for supporting this study and thereby enabling what I believe to be the current largest gut microbiome study on long COVID.

I should state that these findings have NOT yet been peer reviewed and our study and analysis process is evolving and ongoing. So please consider these draft findings. 

Purpose of the study

Initial aim: To identify patterns of gut microbiome dysbiosis associated with long COVID

Longer term aim: Identify whether treatments aimed at rebalancing the microbiome is effective at alleviating long COVID symptoms

Participant demographics 

These numbers are current as of 18 May 2022 and includes only results already in Biomesight. We are still actively recruiting additional participants.

Total samples: 204

Adults (18+): 145

Infants: 2

Female: 95

Male: 52

Controls: 3355 (All remaining samples within Biomesight)

The demographic totals doesn't match the total participants as all participants did not complete their date of birth or gender on Biomesight. Please do complete your information as comprehensively as possible to assist our analysis, as it's important to distinguish between the developing and adult gut microbiome.

How to access the long COVID dataset

Common Symptoms

Below are all the self reported symptoms within the cohort, where at least 10 participants experience them. The frequency of these symptoms corresponds to the frequency that these are reported within Biomesight generally with the exception of Muscle weakness, Headache, Heart palpitations and Tachycardia (bold) where these symptoms are 2-5 times more prevalent amongst the study cohort.

Fatigue: 78

Brain Fog: 58

Bloat: 55

Heart palpitations: 43

Tachycardia: 40

Anxiety: 40

Headache: 38

Muscle weakness: 37

Joint pain: 36

Tinnitus: 34

Flatulence: 31

Poor memory: 30 

Insomnia: 23

Heartburn: 22

Abdominal pain: 22

Back Pain: 20

Weight gain: 20

Sore throat: 19

Blurred vision: 16

Neck Stiffness: 16

Irritability: 15

Nausea: 15

Confusion: 14

Itching: 13

Sweats: 13

Swollen or painful lymph node: 12

Urinary frequency: 12

Urticaria: 12

Belching: 10

Key Findings

Based on our current study size, the microbiome of long haulers corresponds to patterns of dysbiosis seen within other diseased or symptomatic populations. Some of these imbalances are more pronounced. We have also identified a list of microbiome targets that are indicative of long COVID that are distinct from these typical patterns of dysbiosis.

What are these typical patterns of dysbiosis?

Typical patterns of dysbiosis are those already highlighted in Biomesight - in particular where Probiotics like Bifidobacteria and Faecalibacterium are too low, Commensals like Bacteroides and Prevotella is overgrown & Pathobionts are elevated.

What are the most distinctive differences between the current cohort & controls?

If we consider only differences with a p-value of at most .05 and an effect size of at least a difference of 5% in both the median and mean values between the cohort and the controls, as well as a consistent effect seen across both median and mean, then the table contain the most distinctive differences.

Using a p-value cut off of at most .05 allows us to focus only on those that have a statistically significant difference. The lower the p-value the more significant the finding. The importance column is the importance of the feature (bacteria or substrate) in the machine learning model. It is out of a total of 100%.

The model is constructed using a RandomForest algorithm and currently has an accuracy of 94.3% in predicting whether a microbiome sample is collected from someone with long COVID or from the control set. For those not familiar with model accuracy and expected performance, this is a rather excellent result. Please do bear in mind that the model needs further validation to ensure no mistakes have been made in the analysis process.

Table 1: Long COVID cohort compared to all controls

Feature Importance (%) P-Value Effect Delta Median (%) Delta Mean (%)
Genus Catonella 0.138799543 1.01389E-18 Lower 75 86.88
Genus Coprococcus 0.146289566 9.97787E-13 Lower 62.19 53.51
Genus Roseburia 0.179798319 3.77537E-10 Lower 45.88 51.49
Class Erysipelotrichi 0.128057436 1.78858E-07 Lower 18.84 37.94
Family Peptococcaceae 0.133358536 2.8521E-07 Lower 9.62 26.79
Species Dorea formicigenerans 0.155253254 1.59981E-05 Lower 54.84 52.66
Genus Eubacterium 0.139635208 5.17882E-05 Lower 40 188.61
Order Enterobacteriales 0.126491243 8.30224E-05 Lower 36.71 107.05
stain gram negative 0.174164935 0.000124615 Higher 7.18 8.85
Family Coprobacillaceae 0.161941478 0.000241944 Lower 11.11 24.38
Class Sphingobacteriia 0.291071902 0.000358003 Higher 27.85 26.08
Order Sphingobacteriales 0.154114398 0.000358003 Higher 27.85 26.08
Family Sphingobacteriaceae 0.221335166 0.000625772 Higher 30.86 25.97
Species Parabacteroides merdae 0.149463653 0.000629741 Lower 36.71 40.71
stain gram positive 0.152917916 0.000944992 Lower 6.03 8
Order Rhodospirillales 0.130338082 0.00140801 Higher 82.13 43.59
Family Deinococcaceae 0.158836169 0.00181982 Higher 33.33 44.48
Phylum Bacteroidetes 0.228547701 0.00197242 Higher 5.6 8.72
Family Rhodospirillaceae 0.137123396 0.00211939 Higher 81.66 42.54
Species Pectinatus cerevisiiphilus 0.154196064 0.00343113 Higher 41.94 27.43
Class Alphaproteobacteria 0.159617991 0.00387281 Higher 63.98 41.01
Order Lactobacillales 0.186492142 0.00453052 Lower 25 41.92
Species Planococcus columbae 0.162421935 0.00695067 Higher 33.33 31.38
Genus Pectinatus 0.198843074 0.00869691 Higher 40 24.1
Phylum Firmicutes 0.142823404 0.00925359 Lower 5.83 5.77
Phylum Thermi 0.130212039 0.00971996 Higher 33.33 32.02
Family Thermobaculaceae 0.450398214 0.0102712 Higher 33.33 41.12
Genus Thermobaculum 0.312272971 0.0102712 Higher 33.33 41.12
Order Thermobaculales 0.166716659 0.0102712 Higher 33.33 41.12
Class Thermobacula 0.166301336 0.0102712 Higher 33.33 41.12
Species Thermobaculum terrenum 0.232684123 0.0105845 Higher 33.33 41.07
Phylum Tenericutes 0.153057347 0.014031 Higher 18.42 43.31
SCFA propionate 0.216853968 0.0151306 Higher 9.76 8.07
Species Streptococcus australis 0.15080343 0.0153033 Lower 16.67 26.07
Species Porphyromonas cansulci 0.18618597 0.0154545 Higher 50 37.5
Species Sutterella wadsworthensis 0.239641964 0.0172127 Higher 68.56 31.31
Genus Sphingobacterium 0.140998734 0.0232307 Higher 17.57 21.19
Genus Anaeroplasma 0.15002829 0.0233091 Higher 86 78.92
Class Clostridia 0.15777963 0.0264358 Lower 5.84 5.08
Class Bacteroidia 0.266295252 0.0269231 Higher 6.54 7.26
Order Bacteroidales 0.147678179 0.0269231 Higher 6.54 7.26
SCFA butyrate 0.130186393 0.0278048 Lower 5.18 5.61
Genus Blautia 0.137676753 0.0303805 Lower 11.06 12.57
Family Paenibacillaceae 0.144184858 0.030747 Higher 28.57 22.29
Genus Treponema 0.201481604 0.0342052 Higher 14.29 36.79
Species Novispirillum peregrinum 0.143156794 0.0467886 Higher 60.68 32.57
Genus Novispirillum 0.145654683 0.0498368 Higher 58.9 32.16

Explanation of remaining columns:

Effect

Higher - Cohort median and mean is higher than controls

Lower - Cohort median and mean is lower than controls

Delta Median / Delta Mean (%)

Difference measured in percentage between cohort and control median/mean. Note, this is not a relative abundance difference, e.g. if the median difference is 10% and the effect is higher, it means the control median is 10% higher than the controls, e.g. the controls might be at 10% relative abundance and the cohort at 11%, which is a 10% net difference but a 1% relative abundance difference.

To see a breakdown of the distributions for each of these, it would be best to view the model using the Biomesight cohort analyzer. I have published it as a public model and anyone with a Biomesight account can request access to the Cohort Analyzer.

Next steps

Biomesight's current personalized recommendations are based on restoring balance to a number of key bacteria that are known to influence the overall ecosystem. It does not explicitly address all the therapeutic targets identified above. Is there enough research out there to understand how to modulate these additional targets? Do we even need to specifically target these or can rebalancing the overall ecosystem modulate these automatically?

Specific action points

  • Refine & validate study analysis & findings (target timeline: next 1-3 months)
  • Share study cohort data with MicrobiomePrescription for their own analysis (target timeline: 1-7 days. We will periodically share additional data in batches as we accumulate more samples.)
  • Integrate more closely with MicrobiomePrescription to bring their recommendations back to Biomesight and display it directly in our UI (target timeline: next 2-6 months)
  • Gather feedback from participants on whether their symptoms have improved and if they would still regard themselves as suffering from long COVID
  • Perform cluster analysis on the cohort to identify potential clusters for curated treatments (target timeline: next 1-3 months)

Do long-term COVID patients need a microbiome test?

Progress from different supplement regimes can be very variable (even over time in the same individual). It will be more time consuming without test results to tailor and quantify your progress, you are therefore likely to take longer and spend more money experimenting on supplements if you were to attempt treatment without testing.

How many tests are needed?

The first test is the most meaningful and will give you a starting point for targeting treatment. Multiple tests will confirm if your progress is as expected. In time you will learn to look for other signals (e.g. stool appearance, consistency, frequency, level of constipation as well as symptoms) that indicate your microbiome is developing in the right direction.

Study timeline

We have no fixed end date yet for this study. Study participants can continue to benefit from a subsidized kit for £60 (RRP149). If you would like to join our study please send an email to support@biomesight.com and outline how you qualify for the study. 

DISCLAIMER This service has not been evaluated by the Food and Drug Administration or other healthcare authorities. Our platform and related products and services are not intended to diagnose, treat, cure or prevent any disease. Ranges apply to over 18s only.