Oral evidence to the Health and Social Care committee highlights the now-well-documented limits to UK testing capacity and PPE stocks (see also NERVTAG on PPE). SAGE does not discuss testing capacity much in the beginning, although on 10.3.20 it lists as an action point: ‘Plans for how PHE can move from 1,000 serology tests to 10,000 tests per week’ and by 16.3.20 it describes the urgent need to scale up testing – perhaps with commercial involvement and to test at home (if can ensure accuracy) – and to secure sufficient data to track the epidemic well enough to inform operational decisions. From April, it highlights the need for a ‘national testing strategy’ to cover NHS patients, staff, an epidemiological survey, and the community (2.4.20), and the need for far more testing is a feature of almost every meeting from then.
Limited contact tracing
Initially, SAGE describes a quite-low contact tracing capacity: ‘Currently, PHE can cope with five new cases a week (requiring isolation of 800 contacts). Modelling suggests this capacity could be increased to 50 new cases a week (8,000 contact isolations)’ (18.2.20: 1).
Previously, it had noted that the point would come when transmission was too high to make contact tracing worthwhile, particularly since many (e.g. asymptomatic) cases may already have been missed (20.2.20: 2) and the necessary testing capacity was not in place (16.4.20): ‘PHE to work with SPI-M to develop criteria for when contact tracing is no longer worthwhile. This should include consideration of any limiting factors on testing and alternative methods of identifying epidemic evolution and characteristics’ (11.2.20: 3; see also Testing and contact tracing).
It returned to the feasibility question after the lockdown, with:
- SPI-M (meeting paper 4.20d: 1-3) estimating that effective contact tracing (80% of non-household cases, in 2 days) could reduce the R by 30-60% if you could quarantine many people, multiple times; and,
- SPI-B (meeting paper 4.20a: 1-3) advising on the need to clarify to people how it would work and what they should do, redesign physical spaces, and conduct new qualitative research and stakeholder engagement to ‘help us to understand more clearly the specific drivers, enablers and barriers for new behavioural recommendations’ to address an unprecedented problem in the UK (22.4.20a: 2). SPI-B also describes the trade-offs between app-informed systems (notification based on symptoms would suit people seeking to be precautionary, but could reduce compliance among people who believe the risk to be low) (see meeting papers 29.4.20: 3 and 5.5.20: 1-8)
- SAGE noting ongoing work on clusters and super-spreading events, which necessitate cluster-based contact tracing (11.6.20: 3)
- A more general message that contact tracing will be overwhelmed if lockdown measures are released too soon, raising R well above 1 and causing incidence to rise too quickly (e.g. 14.5.20)
Low capacity to achieve high levels of information necessary for forecasting
This type of discussion exemplifies a general and continuous focus on the lack of data to inform advice:
‘24. Real-time forecasting models rely on deriving information on the epidemic from surveillance. If transmission is established in the UK there will necessarily be a delay before sufficiently accurate forecasts in the UK are available. 25. Decisions being made on whether to modify or lift non-pharmaceutical interventions require accurate understanding of the state of the epidemic. Large-scale serological data would be ideal, especially combined with direct monitoring of contact behaviour. 26. Preliminary forecasts and accurate estimates of epidemiological parameters will likely be available in the order of weeks and not days following widespread outbreaks in the UK (or a similar country). While some estimates may be available before this time their accuracy will be much more limited. 27. The UK hospitalisation rate and CFR will be very important for operational planning and will be estimated over a similar timeframe. They may take longer depending on the availability of data’ (Meeting paper 2.3.20: 3-4).
A limited capacity to reach a relatively cautious consensus?
These limitations to information contributed to the difference between SAGE’s estimate on UK transmission (such as in comparison with Italy) and the UK’s much faster rate of transmission:
‘the UK likely has thousands of cases – as many as 5,000 to 10,000 – which are geographically spread nationally … The UK is considered to be 4-5 weeks behind Italy but on a similar curve (6-8 weeks behind if interventions are applied)’ (10.3.20: 1)
‘Based on limited available evidence, SAGE considers that the UK is 2 to 4 weeks behind Italy in terms of the epidemic curve’ (18.3.20: 1)
Rather, the UK was under 2 weeks behind Italy on the 10th March, suggesting that its lockdown measures were put in place too late.
At the heart of this estimate was the under-estimated doubling time of infection (‘the time it takes for the number of cases to double in size’, Meeting paper 3.2.20a):
- although described as 3-4 days (28.1.20: 1) then 4-6 days (Meeting paper 2.3.20) based on Wuhan, and 3-5 days based on Hubei (Meeting paper 3.2.20a),
- SAGE estimates ‘every 5-6 days’ (16.3.20: 1) and states that ‘Assuming a doubling time of around 5-7 days continues to be reasonable’ (18.3.20: 1).
- Only by meeting 18 does SAGE estimate the doubling time (ICU patients) at 3-4 days (23.3.20). By meeting 19, it describes the doubling time in hospitals as 3.3 days (26.3.20: 1).
Kit Yates suggests that (a) the UK exhibited a 3-day doubling time during this period (Huffington Post), and (b) although many members of SAGE and SPI-M would have preferred to model on the assumption of 3-days:
Having spoken to some of the modellers on SPI-M, not all of them were missing this. Many of the groups had fitted models to data and come up with shorter and more realistic doubling times, maybe around the 3-day mark, but their estimates never found consensus within the group, so some members of SPI-M have communicated their concerns to me that some of the modelling groups had more influence over the consensus decision than others, which meant that some opinions or estimates which might have been valid, didn’t get heard, and consequently weren’t passed on up the line to SAGE, and then further towards the government, so an over-reliance on certain models or modelling groups might have been costly in this situation (interview, Kit Yates, More or Less, 10.6.20: 4m47s-5m27s)
Yates then suggests that the most listened-to model – led by Neil Ferguson, published 16.3.20 – estimates a doubling time of 5-days, based on early data from Wuhan, using estimate of R2.4 (and generation time of 6.5 days), ‘which we now know to be way too low’ when we look at the UK data:
‘If they had just plotted the early trajectory of the epidemics against the current UK data at that point, they would have seen [by 14.3.20] that their model was starting to underestimate the number of cases and then the number of deaths which were occurring in the UK’ (interview, Kit Yates, More or Less, 10.6.20: 7m2s-7m15s)
Yates’ account highlights not only
- the effect of uncertainty and limited capacity to generate more information, but also
- the wider effect of path dependence, in which the (a) written and unwritten rules and norms of organisations, and (b) enduring ways of thinking (in individuals and groups, and political systems) place limits on new action. These limits are often necessary and beneficial, and often unnecessary and harmful.
Compare with Vallance’s oral evidence to the Health and Social Care committee (17.3.20: q96):
‘If you thought SAGE and the way SAGE works was a cosy consensus of agreeing scientists, you would be very mistaken. It is a lively, robust discussion, with multiple inputs. We do not try to get everybody saying exactly the same thing’.