Don’t shoot the messenger – don’t blame the science!

“We’re following the science!” Hard to find fault with this as a prudent strategy when dealing with a global pandemic. So, if all the decisions and actions taken are perceived by the “protected populations” to have been slow, confusing, contradictory, counterproductive, or punitive, it must have been the science (or particular “scientists”) that got it wrong?

This is typified by an article in today’s Times – “Professor’s model for coronavirus predictions should not have been used!                                                                

https://www.thetimes.co.uk/edition/comment/professors-model-for-coronavirus-predictions-should-not-have-been-used-z7dqrkzzd?utm_source=newsletter

So, we have somebody to blame? – But is a single Professor’s model the Science? No, a single piece of evidence or postulated prediction is rarely acceptable. Science attempts to present the best available interpretation of all the current evidence (until we have better!); and uses this to hypothesise “models” to try and predict how things may develop. The “experts” are expected to realise the quality, or otherwise, of the evidence and the confidence, or otherwise, in the accuracy given the uncertainties involved in such predictions. Unfortunately, these caveats are not always reflected, or heeded in the advice subsequently given, or understood by the decision makers. This is only compounded if the results are presented simplistically, or selectively, to support a particular preferred policy. The temptation is to “pick and mix” advice and data to justify actions, not, as it is presented, to take actions prompted by the evidence. The classic example is the debate on face masks, which have suddenly become to be seen as beneficial, now the PPE shortages seem to have subsided.

But there is a case for the science to answer, not in its validity, but in its selective and perhaps unthinking pandering to academic silos. Thus the (global) risk management of this pandemic has been almost exclusively based on one particular silo of expertise – epidemiology; hence the Times’ attack on the policy of relying on epidemiologist’s spreadsheet models. To paraphrase Churchill – This is not the “science”, but it may be the beginning of the science.

So far then, the COVID 19 crisis has been managed, based supposedly, on the known science of epidemics. Knowledge of how previous pandemics have developed and behaved cannot be discounted and can provide helpful guidance, if we recognise its limitations and don’t just blindly follow its dictums. This knowledge about the behaviours of past pandemics has been helpfully distilled (Stored?) into spreadsheet models constructed and calibrated with generic “hindsight” derived parameters. These then, pre pandemic, represent the best available guess as to how any new pandemic will develop and would be disregarded at our peril.

The predictions, or the numbers produced by these models, are inevitably based on a wide range of observations accumulated from different age groups, cultures, social conditions and infectious agents. They are thus unavoidably guestimated “averages”, assumed to obey classical statistical distributions to derive overall figures for infections, fatalities, rates of spread, etc. They are an all-important preparation for dealing with a developing situation, but only as “priors”, (Bayesian” not Gaussian).

Once into the fray of the real battle, we should expect, like the military, that such initial guesses will not survive close contact with the enemy. We need to track “actual” behaviours to obtain “posterior” estimates to continually update predicted behaviours.

This explains our problems in managing this crisis in the UK. We seem to have almost exclusively relied on just one source of science – the epidemiologist’s spreadsheet; a necessary but not sufficient response. Initially this model was predicting based on the last flu epidemic and hence the results based on bad flu were not particularly concerning and unexpected – we’d just had a rehearsal (2016?) hadn’t we? Just flatten the curve to avoid overstretching the NHS and wait for herd immunity.

When reality dawned –( better data on hospitalisation etc from Italy became available), the models overpredicted and belated , but justified, on the figures now to hand, more stringent lockdown measures to prevent a much more dangerous pandemic taking hold and perhaps killing nearer half a million in the UK.

Having survived, though, we are now seeing the criticism that it was an overreaction and the models on which it was based were flawed. To quote the Times Red Box article

“The response to Covid-19 in the UK, the US and other countries was shaped by the dramatic headlines in mid-March, suggesting 550,000 deaths in the UK and 2.2 million in the US. Faced with widely publicised, alarming figures, as demonstrated by Imperial College’s Professor Neil Ferguson, governments were forced to react with the unprecedented lockdown to suppress Covid-19. No one looked at his ten years of predictions that were wrong.

The results of his previous models produced wildly inaccurate results: the prediction of 200 million deaths worldwide from bird flu in 2005, when just 282 people died between 2003 and 2009, without locking down economies. That model had serious flaws. He used an undocumented, highly complex, 13-year-old computer code for a feared influenza pandemic. Full details of that model have never been revealed, only a cleaned-up, improved, but still almost undocumented version.”

But a legitimate response could be that it was the “best available”, most other countries and the UN, had similar models and predictions; and anyway, what else should the Government have used. - gut feelings?

Whatever the motivation, this did seem to work and the pressure is now on to return, as quickly as possible, to “normal”; and save the economy and help the economic victims as well as the medical casualties. The problem now is to judge how quickly, in what areas and to what extent the total lockdowns could be eased

The dissatisfaction arises, however, from the seeming inability of these models to suggest and justify detailed proactive responses aimed at protecting and controlling the unwanted spread of the virus. Pubs versus schools, 2 metres versus 1 metre, face masks, rule of six, 30 miles versus 300 miles, Advisers and the public.

The problem is that anything based on these epidemiological models is inevitably based on “average” populations, “global” lagging indicators (R?) and are effectively reduced on close examination to “suck it and see”!

The problem is that this is not an “average” virus - again to quote the Times article,

“Standard simplified epidemic models take no account of the difference between individuals in terms of their susceptibility to infection and the numbers of people with whom each individual comes into substantial contact each day.

“But we all know how much human contact varies from one person to another. Many people live on their own with limited social contact. Then there are those annoying people who never catch colds or the flu. Susceptibility varies with the strength of a person’s immune system. Strengthening people’s immune systems is an important method of dealing with a pandemic which has been neglected”.

But with a different mindset, maybe we could and should have done better. How?

This, despite the scale, is a relatively slow-moving disaster. We do have the time to manage it in real time (unlike the Beirut explosion). This means we have the time to try and get in front of the curve with our attempts at corrective actions and constantly predict and correct with our models. Our current reliance on centralised control of average numbers and testing based on number, not effectiveness, or quality has clearly been counterproductive. Is it too late to concentrate on individuals, ages, social conditions, ethnicities etc. through the local professionals?

Is it too difficult to react to trends before they become problems? Is it too late to realise that the “toxicity” of this virus is exponentially biased to the elderly and young people are very unlikely to experience more than flu symptoms at much less risk?

Look at the data; we could have predicted our second spike a month ago. Why has the rise not (yet) been reflected in hospital admissions? Because if you look at age breakdown most are youngsters, and few are elderly – and what was the difference expected from age related effects – look at the graphs. There is now a discernible up tick in the elderly infections, so in 7 – 10 days we will need more hospital beds. Which hospitals? – what does the local data say? So, we don’t (just) need fancy spreadsheets we need an intelligent smarter response based on common sense. This then is a better description of the scientific approach which relies on objective and intelligent analysis of reality, not theory.

 Surely it not too late to start applying this science. Instead of grandstanding national initiatives – focus testing on intelligence led, local conditions and behaviours. Concentrate on protecting the vulnerable from the general, not one size fits all, track and trace responding to local conditions.  Get smarter, get ahead of the curves and spikes – be more aware of both the importance and the limitations of the science. Its not too late, its half time!