Chris Martin and Steve Bale discuss how actuaries modelled the impacts of critical care demand and long COVID during the pandemic
The emergency arising from the COVID-19 pandemic demanded a rapid response on an unprecedented scale. Many sectors of society responded altruistically, notably those in the NHS and social care sectors, who worked long hours in difficult circumstances and at significant personal risk. Others stepped up to keep the negative effects of the pandemic to a minimum by keeping supermarket shelves stocked, public spaces clean and safe, electricity flowing or broadband streaming. We look at one part of the actuarial community’s response to the pandemic, which could serve as a template for future collaborations across industry and the public sector.
Context
Tan Suee Chieh, the IFoA’s president-elect in March 2020, presided over the creation of the organisation’s official COVID-19 Action Taskforce. He also encouraged the establishment of the autonomous COVID-19 Actuaries Response Group (ARG), a multidisciplinary group of actuaries, researchers and epidemiologists that has worked to suppress misinformation, champion unbiased data and information on the pandemic, and provide high-quality analysis for public consumption. Stuart McDonald, one of the group’s leading members, was invited to represent the IFoA on the UK government’s Scientific Advisory Group (SAGE) and present the actuarial science perspective.
In July 2020, SAGE published a document prepared by the Department of Health and Social Care (DHSC), the Government Actuary's Department (GAD) and the Home Office: Direct and Indirect Impacts of COVID-19 on Excess Deaths and Morbidity (DHSC, 2020). This summarised the potential effects of the pandemic beyond the deaths it would cause, and was produced at a time when evidence was accumulating rapidly and there was a need to review and update it in light of emerging evidence. In this context, the DHSC invited Stuart McDonald to co-ordinate an IFoA review of the document.
The document was in four sections and, in the process of the review, it became apparent that two sections would require additional modelling within very short timescales: a model of the impact of demand for critical care exceeding capacity, and another estimating the potential health consequences of persisting symptoms and disability long after the resolution of infection – so-called ‘long COVID’.
A small team of modellers was assembled, consisting of Stuart McDonald, Michiel Luteijn, Steve Bale and Chris Martin. William Letton and Josephine Robertson contributed to long COVID modelling, while Colin Dutciewicz and others supported the demand and capacity model parameterisation. There was also internal actuarial involvement from GAD's actuarial director Matt Gurden.
Demand and capacity modelling
The modelling was kick-started by an extra-ordinary brainstorming session, from which the outline of the model structure was generated. This evolved as we iterated the model components.
A key challenge was a lack of access to some of the data, which was not publicly available. To get around this, the model structure was established using real data where available, plus dummy data that could be substituted by members of the DHSC. We also sensitivity-tested assumptions to understand how the model would react in extreme scenarios.
Some activity data was difficult to source, particularly the distribution of COVID-19 admissions across general hospital or critical care beds and to certain compartments of treatment, such as high-flow oxygen or mechanical ventilation. Industry links with a team of clinicians in Nottingham and a critical care consultant in Kent enabled us to source crucial data and expert opinion to populate these variables.
The principal output was a compartmental model of demand and capacity. This was iterated in consultation with the DHSC, the ONS and GAD to ensure it generated useful outputs and conformed to standard assumptions used in the public sector. The model has been placed in the public domain and can be accessed at github.com/Crystallize/COVID19_ExceedingCapacityModel. To ensure transparency, the work was written up and submitted to one of the pre-publication repositories, MedRxiv, for timely exposure, and then to a peer-reviewed biomedical journal, BMC Medical Informatics and Decision Making (Martin et al., 2021b).
The results suggested that we came close to breaching critical care capacity in England in January 2020, and that there may have been excess deaths as a direct consequence of lack of capacity. This may have resulted from imperfect distribution of demand and capacity, or from the stresses on the healthcare system before capacity was reached. On sensitivity analysis, excess deaths ranged from zero under an optimistic scenario to about 100,000 under a pessimistic one, with a best estimate of 154.
The model was designed as a framework so that it could be updated and populated with data by DHSC end-users. This will have been important in planning for the Omicron wave of December 2021 and January 2022.
Long COVID modelling
The principal challenge in developing the long COVID model was the availability of data that was free of selection and recall biases. Fortunately, during the process of model development, the Office for National Statistics (ONS) published the first data derived from the Coronavirus Infection Survey, which gave population prevalence of COVID-19-related symptoms at six and 12 weeks (ONS, 2020).
The pattern of symptoms and sequelae of COVID-19 are heterogeneous in type and severity. For many, persisting symptoms consist of fatigue and muscular pains, but others may have more severe and enduring problems, such as loss of limb function after a COVID-19-related stroke, heart failure after a heart attack, breathlessness because of lung fibrosis, or post-traumatic stress disorder secondary to intensive care treatment.
We were able to use a health-related quality of life measure, the EuroQoL EQ-5D, to a translate these different experiences to a common metric, the quality-adjusted life-year (QALY). We did this by mining the medical literature for evidence that estimates EQ-5D scores after COVID-19 in non-hospitalised patients and for those who have survived an intensive care unit stay with similar infectious lung diseases.
The principal output of the work was an Excel-based model estimating the potential future burden of persisting post-COVID-19 illness. The model is available at github.com/Crystallize/longCOVID. Again, in the interests of transparency, a paper was published on the MedRxiv preprint server, and submitted for peer-review to the PLOS ONE biomedical journal (Martin et al., 2021a).
The modelling suggested that, in the first year of the pandemic, nearly 300,000 QALYs were lost in England as a direct consequence of persisting symptoms after COVID-19, and that we could expect about half a million lost QALYs in the 10 years after the start of the pandemic. It also suggests that there could be about 90,000 people left permanently injured by COVID-19 in England alone.
Support to the DHSC
Throughout the model development we had regular meetings with the team at the DHSC and the ONS. This was vital to ensure we delivered models that met their needs, and it allowed them to contribute their own modelling insights, which improved the models’ quality and veracity. They were also able to provide continuing review and feedback.
Through these meetings we were also able to act as a ‘critical friend’ to them. For example, when they were challenged on the proportion of the excess deaths that were explainable by lighter mortality in the winter of 2019/20 or by the choice of baseline, we were able to provide useful answers to the question and subsequently arranged for it to be published as an ARG bulletin.
Many people supported this work in addition to those already mentioned; a full list of contributors can be found in the acknowledgements in the articles published.
Chris Martin is managing director of Crystallise Limited and a member of the IFoA Population Health Management Working Party
Steve Bale is head of predictive analytics at Munich Re UK Life Branch and a member of the Continuous Mortality Investigation’s Mortality Projection Committee
References
- Department for Health and Social Care (DHSC). Direct and Indirect Impacts of COVID-19 on Excess Deaths and Morbidity: Executive Summary. London; 2020.
- Martin C, Luteijn M, Letton W, Robertson J and McDonald S. A model framework for projecting the prevalence and impact of Long-COVID in the UK. Chen RJ, editor. PLoS One. 2021 Dec 2;16(12).
- Martin C, McDonald S, Bale S, Luteijn M and Sarkar R. Construction of a demand and capacity model for intensive care and hospital ward beds, and mortality from COVID-19. BMC Med Inform Decis Mak. 2021 Dec 27;21(1):138.
- Office for National Statistics (ONS). The prevalence of long COVID symptoms and COVID-19 complications. 2020.
