An ethical quandary arises when values or principles clash and difficult decisions have to be made on the basis of critical analysis of possible choices. The pandemic response in the UK offers some very publicly elaborated examples, the most obvious being the scale of action: restrictions affect the spread of the disease and the economic impact but both are uncertain and complex and may affect some groups differently. While most ethical problems concern actions of individuals and personal consequences, public health is about large population groups. However, the big picture has sometimes been obscured even as it is very important, and when we look more deeply some features are novel.
When the UK government strategy was first announced there were clear aims: to reduce the number of deaths; and to protect the most vulnerable. These are complementary actions in a principled approach of classic utilitarianism: finding the best outcome for the greatest number. They are also easy to understand in their value for individuals which made them clear and coherent to communicate, as well as being so uncontroversial as to reassure. But some of the more complex issues in responding to an epidemic, those which require thinking about populations were left out of the initial debates and deserve more attention. Quandaries also arise in making trade offs between priorities, decisions with limited information and timely responses to new information.
Having set aside the early prospect of containment and eradication (this is complicated but the modelling conclusion was unavoidable and has been borne out by later data), reducing deaths meant controlling community transmission. Overwhelming health services with rapid spread of cases was one concern as evidence suggests the additional individual mortality risk, to people who are infected, is very similar to annual mortality, if we use averages including those most vulnerable to this disease. So, scaling up capacity to meet predicted need, restricting infections by reducing contacts, and monitoring locally were indicated responses, while shielding the most vulnerable. At this first stage, the economics was subordinate and suitable assessment of the knock on and indirect impact of the ‘lockdown’ on all cause mortality and longer term health is needed.
Restricting liberty to protect lives is indicated, but all such restrictions ought to be necessary and proportionate i.e. sufficient to ‘control’ the epidemic. There are negative outcomes of restricting liberty, in terms of human self-actualisation and restricting economic activity as well as more collateral impacts on health from reduced exercise and socialisation. The dilemma here is not just about ethics, but human rights, setting the right to life against other aspects of liberty. But predicting impacts of policy is very difficult, and a feature of the UK government strategy was that they would be guided by the best scientific evidence. Countries vary considerably in their policy, processes and experiences, demonstrating that there are choices to be made even in an emergency, which may depend on societal context, political priorities and available information.
Human Rights and Deontology
Although it was not transposed into the European Convention on Human Rights, in 1948 the UN ratified its universal declaration on human rights which includes a right to science. Specifically, it is a right to “share in scientific advancement and its benefits”, which stands alongside arts and cultural pursuits and a second clause about the right to be recognised as the originator of innovation. The most interesting point about this Article (27) is the word ‘share’, in contrast to other individual rights, which is a point to return to later. So, benefiting from the best scientific advice is also a human right, if less often discussed, but participating in advancement also confers responsibilities.
In fact, ‘science-informed’ is a feature of the UK approach, and coupled with maintaining democracy, is identifiable as a different kind of ethics called deontology. This is a Kantian approach to ethics, about right and wrong, and the strategy articulated that it would always be right to follow the science, and it would always be right to preserve the democratic process. Alternatives to democracy in terms of emergency states and decrees are expediting activity in some other countries, however the UK focus on science is more interesting. Indeed this status raises specific questions around politicising the science, as well as the quality of the advice provided.
Scientific advancement is also prominent in the UK strategy, as an ongoing programme for research. This includes clinical trials for treatments and the development of a vaccine, as well as the better elucidation of the varied natural course of the disease and discriminating the categories of risks in population groups. But the research programme is also more dynamic and social in nature, looking at individual behaviours and extensive personal information. This goes far beyond what is offering direct benefits to individuals for their own clinical care: it is about understanding the effectiveness of the policy interventions on controlling the epidemic in the population.
Politicising science goes beyond urgent analysis and more policy-driven research questions using individual-level clinical data. So the tension between the two deontological principles of science and democracy has already meant some changes in UK practice, including open publication of scientific advice. But conventional democratic scrutiny has supported this as parliamentary select committees have heard from various appointed and invited scientific advisors. And of course regular press conferences have reported the practical implications of the scientific advice and the evidence to support the current direction while parliament was in recess. A substantial challenge remains the partial and evolving synthesis of information promoted by a limited group of experts to a limited number of political actors, but this is a defining feature of our government.
The Big Picture
There is a clamour for transparency in politics e.g. full disclosures of participants, yet for science the concern is much more about trustworthiness, and using intelligent openness to support this. So information provided ought to be accessible, assessable, relevant and interpretable – deluging people with complex information overloads them without informing them. One way to do this is to provide structured layers of detail, but another component is to provide explanations and critique, something the Science Media Centre has been delivering. But this does require accurate interpretations and specialist critiques, which social policy and politics journalists cannot realistically provide. Unfortunately, this has opened a space for armchair epidemiology, producing simple analyses without sufficient consideration of the bigger picture.
A major part of the trustworthiness of science normally means sustaining the scrutiny of specialist peers, but in this crisis wide distribution of preprints (working papers may be more descriptive but the key feature is lack of formal peer review) is common. Prioritising democracy entails communication with the public, and this responsibility now falls on all the scientists involved. Science is complex and so explaining how it is progressing to the public is challenging, especially for the scientists who are used to academic audiences, and specific clients for outputs. This is complicated by the fact that some science has several applications and different empirical work can be used in several models. For example testing of suspected cases is useful for surveillance and general strategic planning, as well as identifying clinical needs for treatment.
Communication has multiple functions, extended to marry up the science and accountability. Decisions, reasoning and evidence can be provided and questioned, as well as the data and even the modelling used in simulations. Plans for future steps and a longer term strategy can also be linked to emerging data and scientific research. But the research based approach inevitably requires new responses and changes to plans, as well as emergency moves to use data. John Krebs outlined sensible steps: say what you do know; what you don’t know; the uncertainties; your intentions and rationale. The UK government has tried to distil the bigger public policy picture into diagrams, branding and slogans but this has left most of the science outside the communications.
A Research-led Strategy
The research is focused on accelerating conventional clinical diagnostics, treatments and vaccines, complemented by more complex activities to support policy. Standard and adaptive protocols of clinical trials are in progress, with official enthusiasm to volunteer for studies. Indeed a challenge is to look for the evidence and the contrasts to be drawn, carefully selecting criteria for effectiveness without compromising on ethics. Intensive care (ICU) death rates from SARS 2 disease are very high and happen quickly, so improvements would be obvious (indeed a trial has stopped early with this reasoning), and side effects readily tolerated, at least for initial candidate treatments. But in other settings the stakes will be more balanced and evaluation phased.
The ethics of vaccine development is more complex because it has justice at its heart. Although a vaccine may protect an individual, the strategic purpose of accelerating vaccine development right through to production is to protect populations. Some individuals may not be healthy enough to receive a vaccination, but can be protected because others are vaccinated. Developing vaccines therefore can take a diversified approach of regimes suitable for different groups but the big challenge is testing a vaccine. Human trials are long, as effectiveness must be proven by their vaccine-increased immunity, unless deliberately infected in ‘controlled human infection’ studies.
All treatments need to work effectively, and a vaccine also, but because a vaccine must go to all healthy members of the population, it should also be very safe. Indeed, even rare adverse reactions to vaccines, coupled with their abstract population mechanism (herd immunity against crowd disease) make compulsory vaccination problematic. So very large trials, of tens of thousands of people will be needed, to study the side effects observed, as well as the success in preventing infections. Susceptibility to side effects and adverse reactions may mean that a population vaccine strategy is differentiated by risks, with a milder vaccine used to protect the vulnerable against disease and another more challenging one deployed to prevent epidemic spread by achieving ‘herd immunity’.
Treatments and vaccines are standard aspects of disease research. But there has been a distinctive feature of the data-driven response to the pandemic – not just the intrusive tracking of infected individuals but modelling of disease dynamics to inform policy. And diagnostic tests have been developed to inform these models through surveillance of infections and immunity as much as to diagnose people needing treatment. Indeed SARS 2 disease is so variable in individual presentation that a positive test for the virus is not as informative as specific individual symptoms like (exertional) hypoxia and other organ functions. Left-censoring of observed disease progression, with pre-symptomatic infection and early general symptoms, mean that prospective surveillance is needed.
Mild clinical symptoms are not a feature of the current health strategy as they do not require hospitalisation, although they are recognised to be debilitating at the time. The crudely consequentialist initial focus was on reducing mortality, but ‘mild’ morbidity is problematic for several reasons: it is much more common, extending beyond the 15% of infections who come into contact with the health system; recovery commonly extends over a period of around two months, with much longer durations for those receiving invasive ventilation; long term consequences in terms of impaired lung function are indicated from SARS 1 and susceptibility to other disease and even reinfection are substantial unknowns. The term ‘harvesting’ would be, uncomfortably, applied to the age-structured mortality curve if it really is bringing forward deaths of the most frail – the 1918 ‘flu pandemic saw a decade long reduction of sex differences in, and specifically cardiac, mortality. Yet the ‘mild’ morbidity applies to a much broader and more economically active portion of the population which is not currently being conceptualised, e.g. in patient reported outcomes, let alone studied systematically.
Surveillance and Solidarity
Surveillance is also working in reverse by identifying individuals in high risk groups and proposing safeguards to reduce their risk of infection. Shibboleths about the privacy of health data have been defenestrated, using new legislation or emergency powers. Those in higher risk categories have been identified from medical records and contacted directly (inevitably with some number of errors in either direction). And now relative risks for population groups and existing health conditions have been estimated by technical innovations remotely analysing linked health records securely in one case, and ONS census records to death certifications in another. These analyses have shown care workers are rather high risk, compared to others in the health sector, raising a question of the equity of different professions.
One particular suggestion is that immunity could be certificated, although reliable antibody tests have only emerged recently and will take time to roll out. However, that immunity from reinfection is not well understood at present and extent of antibodies may also be variable among those with different disease experiences. Immunisation certificates are required for tropical diseases like Yellow Fever (vaccination is required of foreign travellers specifically), but SARS 2 immunity would be useful in identifying people safe to have contact with vulnerable groups in the risk categories mentioned above. An unresolved problem discriminates the freedom available to the immune if there continue to be restrictions on gathering and travel, or requirements of quarantine and isolation. But much simpler and more encouraging is the widespread enthusiasm to donate convalescent plasma as part of a trial treatment for those hospitalised.
Other data about community prevalence of symptoms, and linking these to testing for infection and then on to tracing contacts is familiar from SARS 1. But these represent only the first step, before considering contact tracing using bluetooth interactions, an evolution of technology already used in Iceland, Singapore and South Korea. While privacy advocates are defensive, the experience of SARS 1 was a fight over whether to keep people anonymous at all – if the technology knows but society does not this may be an improvement (questions of ongoing governance of use of the data collected for research purposes need resolution if the solution is to be trusted). Exemplar legislation (now enacted in Australia), much more uncertain is whether an app on such terms will make an effective contribution to control of the epidemic which remains unclear around the world. Efficacy evaluation is something the Joint Parliamentary Committee on Human Rights report recommends is set out more clearly but currently feasibility pilots are focused more on technical matters, public accessibility, and general willingness to use it.
An Honest Conversation with the Public
We need to talk about stats, not because there are statistics being used to report on things like test results and deaths each day, but because they allow us to abstract the big picture. The average reproduction number, R(t), is a statistic which represents the gross growth of the epidemic in a region is generating attention for its estimated sub-national variation. Excess deaths, supplementary to historical mortality trends allow us to see the net impact of the epidemic, even as we cannot infer direct causes so much as partition our expectation. Encouragingly, this has been discussed frequently, if not always entirely accurately, in the general news media, with more specialist discussion in reasonably accessible places. More generally, model-based estimation allows us to see what could possibly happen in the future, whether as simulations, projected trends or forecasts, and even these models see a public appetite. Each of these approaches allows the reduction of more complex information to a form suitable for seeing the consequences of the choices now and, to a degree, to compare to what is happening around the world.
But the point of talking about the statistics now is not to educate the public about statistical methods, however worthwhile that may be. Each of the models, inferences and estimates makes assumptions relating to how data was classified and collected, and the general dynamics of the disease. Each also makes more subtle assumptions about what to measure, and therefore what not to measure, what to optimise, and what to ignore for now. These are the issues whose consequences the public should be able to see, understand, discuss, challenge and ultimately decide. Many have criticised the transparency of scientific advisory groups, both for the breadth of the membership, and the opacity of the working practices. Up to now a fairly technocratic accountability has prevailed, with specific simulations commissioned and published, but it is time to review the longer term strategy and the impact of choices for the future.
Armchair epidemiology has made fools of many, but it has also given some insight into aspects of the problems faced, and the public enthusiasm to understand them. ‘Flattening the curve’ became popular as a public justification for lockdown, but obscured some important details e.g. about the invariance (or not) of the area under the curve. The area in question would, roughly, correspond to total deaths even if it was ICU admissions, but another curve has emerged which is total cases which depends much more on how effective shielding is and the extent of testing. More importantly, curve representations simplify the political problem by considering a singular ‘the’ curve, whereas endemic disease transmission will not be suppressed so easily. International travel poses risks of seeding new epidemics and in the UK, for example, there is a very substantial concern about the winter, and how to manage possible overlap of a further peak with winter ‘flu. In general terms, many people will be infected, and understanding the infectiousness of the disease, and immunity of those who have recovered needs much more study.
So part of the solution is definitely data collection, and the public have responded enthusiastically to new survey projects about infection prevalence, as well as giving daily updates about their symptoms. But these are the easy bits of public health surveillance, much more difficult is the responses in terms of doing tests and acting on individual results, whether tracing the contacts of the infected, or certifying immunity. Making use of this information could be effective in controlling the transmission of the disease by isolating contacts and restricting vulnerable people’s contacts to those who will not infect them. On an individual level, each decision can be reasonably justified and agreed, but when policies need everyone in society to adopt them in solidarity, it is not so simple. This epidemic is a ‘crowd disease’ – many people will be infected but for most it is not severe and the risk is that they infect others. Mathematically, policy recommendations for social distancing correspond to breaking chains of infection and reducing the connectivity of networks, reducing transmission parameters, not direct protections.
The Ethics of Modelling
It remains an enormous frustration that even those who can see maybe there are ethical issues related to the models involved do not see those issues as their responsibility. But this is simply a convenient fiction: if no one else has the ethical responsibility, the originator definitely does, and that is easier to see where only the technically educated can understand the substance well enough to know where choices have been made. Indeed guidance about ethics in various different settings is applicable, from machine learning models, to more simple algorithmic implementations in the public sector or for statistical research reusing social data and data science in the public interest. One striking thing about the heavyweight collaborations (e.g. RAMP, coordinated by the Royal Society) focused on making coherent progress on modelling is they make no mention of ethical review processes, in contrast to the medical and even tech innovations.
Of course the view that there is no ethics in mathematical modelling, or that it is someone else’s problem, is not new to this crisis. But it is more complex now, as the modelling is framed as the resolution of decisions where the consequences are difficult to predict otherwise. And so there is a need for mathematical contributors to be working at many different levels of ethical engagement, not just recognising ethics is present but acting. To some extent this has been seen in excoriating criticism of published modelling, even as some of this generates more heat than light – a meticulous solution to a problem may have its merits but it may also be a distraction if it ignores important issues. Demands to know both the experts and the models being used in official decision making may lead to better scrutiny of the sufficiency of expertise deployed and assumptions made. But these demands have also been heated rather than directing the science towards a more logical openness in terms of documenting the work to facilitate scrutiny.
And of course the modellers need to be reflexive when setting out their assumptions and the basis for them, both in the results of empirical or simulation modelling, and the society envisioned. The audience for modelling in this crisis is quite different to the typical group of peer reviewers for a journal and a handful of keen students, conference delegates and specialists reading a preprint. The audience now includes the public, typically mediated through a range of routes, including specialist media networks, and other more policy oriented actors. The potential for misconceptions and over-interpretation is enormous, not least for the highly strained experience of the ‘lockdown’. Exemplary communicators, like David Spiegelhalter, have been caught out, but what makes them exemplary is that they follow up and address the confusion. A commitment to being part of the conversation, as John Krebs envisioned, has been critical, and it has fallen to a newcomer, the Office for Statistics Regulation (OSR), to set this out assuredly.
It is easy enough to pick out a suitable set of ethical guidance to apply, even in rapid model development work, from the resources linked above. And the principles to apply in thinking about what to model are not new: dignity, justice and solidarity address all the issues very well. Engaging with people, how they want to live their lives, how well they are represented by models, and the priorities they may have given the difficult choices presented will allow them to claim their dignity. The groups of behaviours, priorities and vulnerabilities which emerge from the engagement and the empirical evidence, as well as the experience of other nations mean impacts can be understood and justice offered. The key to thinking statistically about the modelling of a crisis is the notion of solidarity: for all our individualism, to the virus be are more or less exchangeable hosts, and to control the epidemic we must work together. This is not a thin conception of looking after people like ourselves, but planning for the future, contributing as and how we can, and supporting those more in need than ourselves.
Solidarity in Practice
Already we can see convalescent plasma donations from those unfortunate enough to be infected early, and great enthusiasm to repurpose production lines for protective equipment, as well as recognition of equity in access to vaccine development. Individual, collective and global actions will all be essential to resolving the pandemic of SARS 2, but data has a specific role in all of the response to the crisis, which leaves a responsibility on those best placed to use it. Surprisingly, professional organisations have been absent from this more political role – even as it would raise the status of their professions to take it on there is a need to convene the mathematicians to consider their role in society. But statistical and other data professions have an ethics to share in the current crisis, in the independence of production of offical statistics, in the robust and rigorous methodology of estimation, and in trustworthy communication and accountability.
None of these ideas are new, dating back to resolutions from the International Statistical Institute in the 1980s, United Nations Statistics Commission in the 1990s and Onora O’Neill’s Reith lectures in 2002, but their time has come. Then we might come back to the more contemporary ethical questions of the privacy of digital data and the legitimacy and governance processes to assure its repurposing for research. All of the initial response happened so rapidly that the public was not consulted, but that phase has passed and the agility of some new institutions may even be greater than the public infrastructure carrying out policy. Indeed it has become clear that the greatest weakness in the whole response has been the resilience of distribution, intelligence and data collection – solidarity across geography, professions, economic sectors and politics. There remains a need to develop open synthesis, ask and address questions, establishing what is happening, systematising what is not known, and where more capability is needed.