Published using Google Docs
Covid19 UK Models and Advice LIVE
Updated automatically every 5 minutes

Covid19 UK & Europe Models, Strategy & Advice LIVE

Dr Gareth (Gruff) Davies 
PhD Medical Physics, Imperial College (alumnus); Codex World’s Top 50 Innovator; Machine Learning & Data Scientist, Software Engineer; Award-winning Serial Tech Entrepreneur; I have more than three decades of complex system modelling experience with a science and medical background. I also worked for 2 years as a consultant at the NHS Modernisation Agency and The NHS Institute for Innovation and Improvement, so I know a little about how the NHS works from the inside.

Public permalink to this document: bit.ly/Covid19LIVE 

Model Drivers & Assumptions and FRAQs (for models prior to 24th March)

Doctor, Scientist or Data Scientist? Want to help? Join thousands of global collaborators:

https://coronavirustechhandbook.com/ 

How to read this document: Note, it was frozen for updates March 26th (save the comment below with link to our paper). I’m keeping it publicly visible for the historical record. It comprises 1 page of summary advice at the top (next page) and rolling updates in descending date order beneath that. In other words, it should be read from the bottom up. All forecasts made in this document were short-term two week forecasts. I included dates further out to show worst-case scenarios but these were not within the model forecast parameters so should not be taken as such retrospectively - especially as lockdown measures and summer weather are known to have strongly dampened the initially exponential growth curve (thankfully - things would have been much worse if they had not).

June 2020 Update: We have now published v3 of our full research in preprint on medRxiv which includes a causal inference analysis verifying a causal link between vitamin D status and COVID-19 outcomes (v1 published preprint on May 6th):

Evidence Supports a Causal Model for Vitamin D in COVID-19 Outcomes

March 26th Update: my research project investigating seasonal mimicry as possible combat routes has now been closed and replaced with our findings. A 2-page summary of this report can be found here for medical professionals only: bit.ly/VitDCovid19SummaryShort 

Update 2020-03-23 - call to medical scientists and ICUs (note: this is not for general public consumption):

Evidence strongly suggests Vitamin D supplements could be effective in preventing Covid-19, and play a key role in treating patients if added to existing treatment plans, especially if this is done early in the disease progression.

NB: advice and strategic summary were written 14th March prior to UK Gov action. I will update these in due course now more is understood about the disease,

Up-to-date Advice for individuals:

The UK needs to be in PROACTIVE lock down NOW.

  1. Stay home unless you absolutely have to leave. Work from home (if you can). [update: the UK Gov now supports this position]
  1. Make sure you still exercise and if it’s sunny, get fresh air and sunshine. Just practise social distancing. Use your garden if you have one. You can go for walks, just stay 2m from other people that you don’t already cohabit with.
  1. Don’t meet in groups larger than 6 (no pubs, restaurants, sports gatherings etc.) [update: Gov announced schools to close Friday - welcome news]
  1. I would make stronger recommendations now to avoid all social contact until further notice. Hopefully only for the next few weeks.
  1. Use virtual conferencing, the internet, and try to find ways to continue life as normal as possible without moving. Find ways to support your local businesses to continue to operate.
  2. Don’t take Ibuprofen or other NSAIDs (non-steroidal anti-inflammatories) - use paracetamol instead. (See why)
  3. Increase your hygiene levels. Wash hands at least every hour, before and after leaving the house when you must. Wipe down anything coming into the house from outside, including post, food and drinks.
  4. Don’t panic. The majority of people are not at risk. The reason for these precautions is to safeguard those that are. Be socially responsible. Fear and panic are not the same thing.
  5. Take a vitamin D3 supplement - now is not a good time to be Vitamin D deficient:
  1. 2020-03-27 update: breaking news from Turin, Italy confirms what we have suspected, “The first preliminary data collected in these days in Turin indicate that patients hospitalized for Covid-19 have a very high prevalence of Hypovitaminosis D” (Google translated English news)
  2. Vitamin D supplements of between 400iu to 1,000iu per day are safe and probably a good idea anyway: “the intake of ordinary doses of vitamin D supplements seems to be associated with decreases in total mortality rates” (And Gandini n.d.)
  3. After a long, dark winter people in the northern hemisphere are very likely to be Vitamin D deficient and may benefit from taking a short-term higher dose (I take 4,000iu per day currently and will reduce this to 2,000iu and then 1,000iu as summer approaches). Anyone with a pre-existing condition should check to see if this is safe for them. If you live in a tropical or southern hemisphere country you are probably okay right now.
  4. For healthy young individuals, NICE recommendations advise “all adults living in the UK, including people at increased risk of vitamin D deficiency, should take a daily supplement containing 400 international units (IU [10 micrograms]) of vitamin D throughout the year, including in the winter months”.
  5. “The upper tolerable limit, defined as the highest level of daily nutrient intake that is likely to pose no risk of adverse health effects to almost all individuals in the general population, for vitamin D3 is 1000 IU/d in infants ages 0–6 months, 1500 IU/d in infants ages 6–12 months; 2500 IU/d in children ages 1–5 years; 3000 IU/d in children ages 4–8 years, and 4000 IU/d in adolescents and adults” (Tebben, Singh, and Kumar 2016)
  6. Young people with known Vitamin D deficiency and those older than 60 and should supplement with 2,000iu/d during winter.  The Scientific Advisory Committee on Nutrition (SACN, 2016) does not recommend routine monitoring of serum vitamin D levels during long-term supplementation with vitamin D. Furthermore, expert opinion in the National Osteoporosis Guideline Group 2017 (NOG 2017) is that routine monitoring of serum 25(OH)D levels is unnecessary for people on long-term maintenance doses of vitamin D (up to 2000 IU a day).
  7. Please note prolonged supplementation at higher levels carries a risk of hypercalcaemia (Contraindications, British National Formulary) and should only be undertaken with professional advice.
  8. If you are able to get exposure to natural sunlight then your body will synthesize Vitamin D and you are less likely to need supplements to maintain healthy levels.

UK Strategic Summary

Note: this is looking a little out of date now. I wrote it on 12 March. We have taken appropriate action since so we need to continue with lockdown.

  1. We need to throw as much water on the fire now when it will make a difference to slow the accelerating spread. We are past containment. Delaying efforts to slow the spread only makes future efforts futile (like waiting to throw water over an already raging bushfire instead of early when it can slow it down at least a little).
  2. Why? This is a RACE between the exponential spread of the disease in the UK and the acceleration of science and tech understanding of the disease (slower).
  3. If we slow down the spread enough - even by a month - we stand a chance of getting new treatments out, vaccines, making the care system more robust and able to cope.

Third-party articles I agree with that may help get the point across:

2020-04-12 - Spain saved by later onset and resulting better weather? Germany by Vitamin D supplementation?

Two weeks ago Spain looked to be on worse trajectory than Italy. They were later - relatively to the onset of their outbreak - to impose quarantine. I wondered then, given known seasonal influence on spread and our Vitamin D hypothesis, if the weather could save them from a worse outcome than Italy. It seems that may have turned out to be the case. Spain’s good fortune here may be down to the fact that their outbreak started 18 days, almost three weeks later than Italy’s which began mid-February. Given the median 14 day lag between catching Covid-19 and deaths, this does appear to explain why they have experienced a markedly faster slow down in their growth rate than Italy.

Germany has had a far less severe outbreak. Their outbreak also started 3 weeks later, but didn’t take hold relative to Spain, France and the UK. Of the five countries, Germany has the lowest prevalence of Vitamin D deficiency thanks to a strong focus on deficiency studies: just 25%-30% in the people aged 65 to 79, compared to 75-80% in Italy. (Rabenberg et al. 2015)

2020-03-29b - Why early action is so important

Exponential growth can be hard for people to grasp, but it’s the reason I began to call for proactive lockdown in the UK on the 11th March, even though we had only recorded 22 deaths at that time. Comparing the outbreaks of Hubei, Italy, Spain, UK and France clearly shows the importance of early action. Hubei went into full lockdown very early when just 17 deaths had been recorded. The UK declared it would not implement even social distancing when we had an equivalent number of deaths and by the time full lockdown was imposed the death toll was already more than 300. That wasted time means that our final total death toll will probably be between 15,000 and 25,000, compared to Hubei’s 3,000. France and Italy appear to be in a similar position. Italy quarantined 10 towns in Lombardy much earlier and this very likely saved it from a much worse outcome. The UK’s soft lockdown and many Brits proactive self-isolating seem to have similarly saved it from a much worse fate. Spain appears not to have taken the situation seriously enough and its outbreak is without doubt going to come with the highest death toll in Europe. Although today they have just half the deaths counted in Italy, their late action means that they will suffer more. Unless better treatments or weather can reduce their growth rate soon, Spain’s final death toll may be between 25,000 and 60,000 deaths.

In order to compare the worst hit European towns with Hubei, I’ve offset their log growth curves so they coincide around the time more than 20 people had died. This shows that France, Spain and Italy are on comparable trajectories, but Spain’s appears worse. It also shows how Hubei’s early action saved them.

2020-03-29 - Rate of change of natural log deaths is the best view of outbreak control

The daily number of deaths can be alarming three+ weeks into an outbreak, but since we are dealing with exponential growth (if not controlled), the best way to view how well the outbreak is being contained and treated is to look at rates of change in the log domain.

Italy and Spain, despite horrific numbers of absolute deaths, appear to have their outbreaks well contained and controlled now. The picture is far less clear for the UK and France. France is par with Spain. UK looks to be about ten days behind both now.

Looking at the absolute numbers of deaths by day makes it clear why we need to look so carefully:

My old exponential projections are no longer valid as forecasts, but they are still useful to calculate so long as they are interpreted with great care.

The numbers are not forecasts but show what would happen if the outbreak were to continue to spread uncontained. They’re useful to see the number of lives we are saving by staying home. They’re also useful as they compare 3-day trends with 7-day trends showing whether things appear to be speeding up or slowing down - however big changes in these figures can happen because of how they are reported so this isn’t very meaningful.

2020-03-27 - Can we use Hubei to model where Italy is today?

Overnight, I pondered what the right model could be to use to forecast where Italy may be in its disease management timeline. Since lockdown is an unprecedented action, we can’t use normal epidemiological curves, such as sigmoids, to predict when case fatality rates will be low enough to consider the outbreak “handled”. We also can’t predict what impact new treatments and better understanding will have in terms of shortening the necessary measures.

Comparing with Hubei seemed like a wise thing to do. This view suggests Italy has at least another two weeks (ceteris paribus) before rates would be low enough for the disease to be under control and possibly much longer. Hubei took a further 40 days to lift the lockdown… I’m confident though that with improved understanding ot the disease, and new treatments, Italy’s lockdown will be lifted sooner than +40 days. (Note: breaking news from Turing confirms high prevalence of hypovitaminosis D in covid-19 cases. This is far from unexpected and further evidence to support my Vitamin D project).

2020-03-26 - Log polynomial curve fitting forecasts

As the situation in Europe has changed substantially, and it’s clear that people aren’t always following the commentary next to my crude previous forecasts I’ve had a go at a slightly more sophisticated model which needs less explanation. The problem now is it is very hard to justify any underlying mechanism for a model, so I’ve simply done a polynomial logistic regression of order 2 and 3 to attempt to fit the slow down curve. This isn’t something that’s I can justify other than “this is what the slowdown curve looks like right now”. It’s also quite hard to do for countries other than Italy, but if these trends continue (i.e. lockdown must continue) then the forecasts are actually very encouraging for Italy:

It suggests that they have the situation under control now and could expect to see peak deaths between 1 Apr and 5 Apr with between about ~10k and 14k deaths.

Note: it’s kinda weird using an odd number for a poly fit even in log domain but so long it’s treated with care, it does fit the current shape better… I wouldn’t expect this to work at all for the other countries right now though.

What does this imply for the UK?

This is a hard question to answer. I would say simply, “it’s too early to know for sure”. That said, ceteris paribus, if we continue to follow a curve 14 days behind Italy (which is a big IF as we didn’t follow their clean lockdown, so we simply don’t know yet), then it’s possible we could expect to see a similar curve peaking two weeks out from 1-5 April, i.e. 15th to 19th April.

Only Italy has shown clear evidence that its infection spread is under control so far. It’s very likely that the position in Spain, France and the UK will become clearer in the next 3 - 4 days, but I’m not going to make any strong predictions until I feel more confident this is apparent in the data.

2020-03-25 Another Observation

I think it's also worth me pointing out that 2 weeks ago when UK Gov said we were 28 days behind Italy, and I said, er... no we're not. At best we are 14 days behind, I was precisely correct. #covid19 #stayhome

2020-03-25 Updated forecasts

Italy started to show signs of a second slowdown 21st March thanks to national quarantine 9 March (12 days earlier).

Spain has seen a small slowdown, similar to Italy’s first signs about 5 days after lockdown. If Italy is a guide, then Spain should start to see further slowdown in three days’ time on 28th March. This is hard to model so I haven’t tried, I’m just writing what I expect to see.

UK has also started to show small signs of slow down 5 days after the Government changed strategy and announce “no non-essential travel or contact”, despite not backing this with police enforcement. This has bought us 1 day of extra time relative to my forecast on 12th March. The police-enforced lock down from 23-March is welcome news. It appears that many people did proactively begin social distancing despite the softer approach. Currently exponential growth rates for the UK, France and Spain appear similar. Italy has substantially reduced theirs. I hope that all three other countries will follow a similar pattern in due course.

Forecasts further out that the 7-14 day range should not be considered serious or reliable forecasts. It’s very likely the situation will have changed again within a +5 day window. New treatments are being trialled and I’m hopeful that my Vitamin D project will also prove helpful. We have support from some world leading doctors and hope clinical trials will be fast-tracked to find out for certain if treatment additions are helpful. In the meantime, the general population should follow NICE guidelines for vitamin D in the UK. Guidelines will be country specific, so if you are not in the UK please consult guidelines for your country and pay attention to season as this changes what recommended daily supplements are.

2020-03-22 Updated “forecasts” - NB special care now needed interpreting these!

See note in 2020-03-21 update re interpretation. The first forecast I made on 14th March has played out so far for the UK. Thankfully, Italy’s actions have slowed spread so even though today’s death figures seem grim, they are 2-3 days better than I forecast (they had yet to impose quarantine), so they have potentially saved 4,500 lives relative to what would have happened if they had not taken their extreme measures.

2020-03-21 Updated forecasts - NB special care now needed interpreting these!

When I started forecasting on 12th March, the UK had taken no preventative actions and Spain and Italy had yet to implement their extreme lockdown and nationwide quarantine measures. The forecasting model I used was appropriate to the situation. Now that measures are in place for all European countries, this simple exponential growth model is no longer appropriate. It is still useful so long as this is understood and care is taken to interpret the forecasts in the context of the measures taken. The most useful way to interpret the forecasts is comparing the recent 3-day trends with the longer 7-day and all-time trends which shows that the measures taken are effective. Since trends are now changing daily (they were linear on a log scale prior to the measures), the forecasts are no longer truly ‘forecasts’. The UK case is hard to interpret since less stringent measures were enacted, but I’m hopefully that social distancing and individual action is sufficient to control the spread. Only time will tell.

I don’t think it’s worthwhile making a more sophisticated model. I’m now devoting most of my free time investigating potential tactics for combatting the disease so we can return to normal life as soon as possible. I’m very optimistic we will have new treatments and interventions within weeks. Scientists, medics and technologists around the globe are working together as never before to tackle the pandemic.

2020-03-20 Updated forecasts

John Hopkins data up-to-date again.

Spain’s forecast has dramatically improved with clear early signs their dramatic measures are working. Italy’s situation continues to improve. Although their forecasts look similar to yesterday, it’s important to remember that we are now +1 day and if growth was still exponential we would not expect to see similar figures but we do. This is excellent news.

UK’s forecast is less encouraging, but it’s early days and the data are still volatile so I’m not reading much into this.

On a more encouraging note, my research into pandemic seasonality has uncovered some striking patterns that give me hope we will be able to use this information to quickly develop new tactics and strategies, both in ICUs and social and occupational environments.

2020-03-19 Updated stats using trusted media and government sources

Latest John Hopkins Data reports incorrect figures so I’ve manually updated some of the latest dates with information from UK Gov website and news reports of overseas cases. Figures marked in yellow in the data table indicate where I’ve done this. The 3-day trend is, unfortunately, very sensitive to errors in the values from misreporting and “stochastic noise”. It is currently too low in my opinion, but the 7-day trend forecast I hope will be too high now the UK Gov appears to be listening and imposing stronger measures.

2020-03-19 Data problem with John Hopkins delaying updates

I couldn’t update the stats yesterday or today as the figures reported by SCCE are not agreeing with news reports for UK and France. Waiting for a response.

2020-03-18

2020-03-17 Updated 3 and 7 day Trend Forecasts

2020-03-16 Updated 3 and 7 day Trend Forecasts

Spain and UK vs Italy

2020-03-15 35 deaths in UK

UK Reports 35 deaths 9am. This is worse than expected. If this rate continues we can expected 20k deaths by the end of the month. Stochastic noise still high at these low numbers so don’t treat this is a forecast, it’s a worst case scenario, hopefully.

2020-Mar-15 - Country Exponential Deaths Rates (Natural Log Scale)

The gradient of the lines in these graphs show how exponential the growth of death rate is for France, Italy, Spain and UK.

Countries with statistically significant numbers for estimated Case Fatality Rates

2020-Mar-15:

NB: We can’t yet estimate Case Fatality Rates, nor mortality rate for covid19. It depends on many factors (including care systems). For up-to-date estimates see: https://www.worldometers.info/coronavirus/coronavirus-death-rate/

2020-March-14

Based on published covid19 data today (2020-Mar-15). Currently, UK mortality growth rate is slightly higher than Italy's. This may prove to be "stochastic noise" because it's very early days for us and random factors can sway the numbers enormously BUT the UK is more population dense and I'm fully expecting this to be the case. I wanted to model what +14 days (for both countries) and +28 days looks like for us if we don't take **IMMEDIATE DRASTIC ACTION** to slow the spread of infection.

14 days is not an arbitrary figure:

  1. It's the current estimate for the median average number of days from infection to death. Hopefully people have been paying attention the last few days and ignoring government advice.
  2. It is also my current estimate for how far behind Italy we are (a coincidence).

We will almost certainly OVERTAKE Italy's deaths today in 14 days' time if this plays out.

In 14 days' time Italy's death toll could rise to 74,000. By comparison, the UK's could be 120,000 in 28 days, IF infection rates are not controlled and we are about 14 days behind them as we very much appear to be.

The UK needs to be in PROACTIVE lock down NOW.

#covid19 #projections #stayhome

2020-Mar-12

Bibliography

And Gandini, Autier P. n.d. “Vitamin D Supplementation and Total Mortality: A Meta-Analysis of Randomized Controlled Trials. - PubMed - NCBI.” Accessed March 28, 2020. https://www.ncbi.nlm.nih.gov/pubmed/17846391.

Rabenberg, Martina, Christa Scheidt-Nave, Markus A. Busch, Nina Rieckmann, Birte Hintzpeter, and Gert B. M. Mensink. 2015. “Vitamin D Status among Adults in Germany – Results from the German Health Interview and Examination Survey for Adults (DEGS1).” BMC Public Health 15 (1): 1–15.

Tebben, Peter J., Ravinder J. Singh, and Rajiv Kumar. 2016. “Vitamin D-Mediated Hypercalcemia: Mechanisms, Diagnosis, and Treatment.” Endocrine Reviews 37 (5): 521.