key: cord-0992239-umadvqv5
authors: Coggon, D.; Croft, P.; Cullinan, P.; Williams, A.
title: ASSESSMENT OF WORKERS PERSONAL VULNERABILITY TO COVID-19 USING COVID-AGE
date: 2020-05-25
journal: nan
DOI: 10.1101/2020.05.21.20108969
sha: 77c177b0cac8b3fed4ce5266015b0bd991ca8ae7
doc_id: 992239
cord_uid: umadvqv5

Decisions on fitness for employment that entails a risk of contracting Covid-19 require an assessment of the workers personal vulnerability should infection occur. Using recently published UK data, we have developed a risk model that provides estimates of personal vulnerability to Covid-19 according to sex, age, ethnicity, and various comorbidities. Vulnerability from each risk factor is quantified in terms of its equivalence to added years of age. Addition of the impact from each risk factor to an individuals true age generates their Covid-age, a summary measure representing the age of a healthy UK white male with equivalent vulnerability. We discuss important limitations of the model, including current scientific uncertainties and limitations on generalisability beyond the UK setting and its use beyond informing assessments of individual vulnerability in the workplace. As new evidence becomes available, some of these limitations can be addressed. The model does not remove the need for clinical judgement or for other important considerations when managing occupational risks from Covid-19.

ASSESSMENT OF WORKERS' PERSONAL VULNERABILITY TO 56 COVID-19 USING "COVID-AGE" 57 58 59

Background 60 61 As countries adapt to the longer-term challenges of Covid-19, doctors increasingly will be 62 asked to advise on the fitness for work of patients who might be unusually vulnerable to the 63 disease because of their age, ethnicity and/or comorbidities. The risk of contracting Covid-19 64 through work will depend on the potential for close proximity to people who could be carrying Covid-19 4 . Neither document attempted to quantify risks from specific comorbidities, but 77 research is now emerging that allows more detailed and reliable assessment of vulnerability 78 to Covid-19. In these circumstances, we judged it timely to analyse evidence on risk factors 79 for mortality from the disease, and apply the findings in a risk model that could be used to 80 estimate personal vulnerability. The model is intended principally to assist decisions on 81 occupational placement of workers in the UK. Over time it can be updated and refined as 82 relevant new data become available. 83

The online resource is now operational, and we here summarise the methods that we used, 85 the structure of the risk model, and our initial findings. Further detail can be found on the 86 project website 5 . 87 88

Our aim was to assess and compare risks of fatality in people who contract SARSCov-2 90 infection, according to their age, sex, ethnicity, smoking habits, and various comorbidities. In 91 preliminary searches of the published literature, no evidence could be found on risks of 92 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 25, 2020. Because of the urgency to improve on earlier advice, we initially sought reports that would 103 provide the strongest evidence relevant to the UK, and did not attempt systematically to 104 search for, and review, all published evidence that might bear on the risks that we were 105 trying to characterise. In this respect, one paper stood out as particularly suited to our 106 purpose. 107 108 That report, from the OpenSAFELY (OS) collaborative, presented first results from a cohort 109 study of more than 17 million adults registered with English general practices and followed 110 up from 1 February 2020 to the earlier of death or 25 April 2020 6 . It used multivariate Cox 111 regression to estimate mutually adjusted hazard ratios (HRs) with 95% confidence intervals 112 (CIs) for death in hospital with confirmed Covid-19 (ascertained by linkage to a national 113 notification system) in relation to risk factors determined from pseudonymised individual 114 primary care records. Data on other deaths in the cohort (needed for censoring of follow-up) 115 had been obtained by linkage to records held by the Office for National Statistics (ONS). 116

The report contained a secondary analysis, which censored follow-up at 6 April 2020, 117 allowing exploration of the possibility that HRs for some comorbidities were underestimated 118 in the main analysis because, in response to advice from the UK government at the end of 119

March, people with those diseases had selectively shielded themselves from exposure to 120 infection. As well as sex, age, ethnicity, smoking habits and multiple comorbidities, analyses 121 in the paper adjusted for deprivation (using an index graded to five levels) and for the 122 administrative region of the patient's general practice (to allow for varying rates of infection in 123 different parts of the country). 124 125 This study had unique strengths. It included a substantial proportion of the adult population 126 of England, and was based on more than 5000 deaths attributed to Covid-19. Moreover, 127 information about risk factors came from data recorded before the onset of infection, which 128 reduced the possibility that ascertainment would be biased in relation to the outcome. 129 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 25, 2020. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 25, 2020. . https://doi.org/10.1101/2020.05.21.20108969 doi: medRxiv preprint

We abstracted HRs for risk factors of interest from the OS report, and where possible 166 checked their plausibility against data from the other sources mentioned above. The relative 167 risks that we then adopted for our risk model are shown in Table 1 , together with our 168 qualitative assessments of the strength of evidence ("robustness") on which estimates for 169 each risk factor are based. In the main, these estimates were the HRs from the full follow-up 170 period in the OS analysis, since they were statistically the most precise. However, in a few 171 instances, where they were lower than the corresponding HR in the OS study from shorter 172 follow-up, and it seemed probable that this might reflect selective shielding, we made 173 appropriate adjustments. Also, findings from the ISARIC study suggested that the HR for 174 chronic heart disease in the OS study might be too low, and that was therefore adjusted up 175 slightly. Two previously suggested determinants of vulnerability (smoking and hypertension) 176

were omitted from the risk model because after allowance for other factors, they appeared 177 not to carry any material increase in risk. Further details of the calculations, rationale, and 178 qualitative estimation of robustness of the relative risks shown in Table 1 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 25, 2020. . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 25, 2020. . https://doi.org/10.1101/2020.05.21.20108969 doi: medRxiv preprint ‡Includes HIV, conditions inducing permanent immunodeficiency (ever diagnosed), aplastic anaemia, 193 and temporary immunodeficiency recorded within the past year.

Covid-age 196 A notable feature of Covid-19 is that mortality rates, whether or not they are adjusted for 197 comorbidities, increase exponentially with age. Thus, in the OS report, there were adjusted 198 relative risks of approximately 1.0945 and (1.0945) 10 = 2.5 for increases in age of one and 199 10 years respectively 6 . In these circumstances, vulnerability from other risk factors can 200 conveniently be expressed in terms of the added years of age that would give an equivalent 201 increase in risk 14 . Table 1 sets out vulnerabilities associated with demographic variables 202 and comorbidities in our risk model, quantified as age equivalents as well as relative risks. 203

If it is assumed that when risk factors are present in combination, their relative risks multiply 204 (this is the normal default assumption in regression analyses such as those described in the 205 OS report), then combined effects can be estimated by summing the age equivalent for 206 each. Moreover, by adding the summed age equivalents to the person's true age, it is 207 possible to generate a summary measure of personal vulnerability. We have termed this 208 summary measure a person's "Covid-age". It represents the age of a healthy white male 209 with equivalent vulnerability (white males being the largest demographic group in the UK 210 workforce). Here are some examples: 211  A healthy white woman, aged 40, has a Covid-age of (40-8) = 32 years. 212  A white man, aged 45, BMI 36, with COPD has a Covid-age of (45+5+7) = 57 years 213  An Asian woman aged 50 with uncontrolled diabetes has a Covid-age of (50 -8 + 5 214 + 10) = 57 years. 215 216 Absolute risks can be obtained by translating Covid-ages into estimated case-fatality rates. 217

The process is complicated by current uncertainties about the prevalence of asymptomatic 218 infection (for a given fatality rate in diagnosed symptomatic cases, a higher relative 219 prevalence of asymptomatic cases will imply lower overall case-fatality). However, in a 220 report by Ferguson and colleagues 15 , which drew on findings from a study by Verity and 221 colleagues 16 , the case-fatality rate at 40-49 years of age in men and women combined was 222 estimated to be 1.5 per 1000. Assuming a relative risk of 0.5 in women as compared with 223 men (see Table 1 ), this would imply a case fatality rate in men of 1.5*2/1.5 = 2 per thousand 224 in men aged 40-49 years. Given that some men in this age band will have comorbidities and 225 other risk factors that increase their vulnerability, this fatality rate might correspond to an 226 average Covid-age in the region of 47 years. 227

Limitations 229

Our analysis and risk model have important limitations. 230 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 25, 2020. .

Currently, all our assessments of risk are derived from a single study, albeit with checks on 232 plausibility using other sources. The outcome in that investigation was death in hospital from 233

Covid-19, and did not extend to deaths elsewhere. Data on some risk factors were 234 incomplete (although the extent of missing information was generally small). Although the 235 study was large, its findings, in particular for rarer comorbidities, are liable to statistical 236 uncertainties (confidence intervals for HRs from the OS report are presented in the 237 documentation that accompanies the vulnerability assessments 5 ). Risks associated with 238 some comorbidities may have been attenuated by adjustment for deprivation. We have 239 assumed as a first approximation that relative risks from different factors multiply, but that 240 may not always be true. Also, some risk factors may have been associated with differences 241 in exposure to infection, as well as with differences in vulnerability once infection occurred. 242

The scope for such bias should have been reduced by adjustment of HRs for region, and by Although the methods that we have employed may be relevant to development of similar 264 models for other populations, our risk model is designed for application specifically to adults 265 in the UK. It was developed to assist decisions on occupational placement of workers in the 266 UK, and, although some of its findings might be of utility for other purposes, it is not intended 267 to inform decisions in clinical care. 268 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 25, 2020. . https://doi.org/10.1101/2020.05.21.20108969 doi: medRxiv preprint

We expect that some of these limitations can be addressed as further evidence becomes 270 available. Meanwhile, we believe that our assessment of vulnerability offers an improvement 271 on what previously has been available. We caution against simplistic rules for decisions 272 based only on the impacts that it estimates. It does not remove the need for clinical 273 judgement, and there are other important considerations when managing occupational risks 274 from Covid-19for example, the practicability of different possible control measures, the 275 personal value judgements of the individual worker, and prevailing advice from government 276 (which may be driven by a need to control demands on healthcare services as well as 277 individual risk). With these caveats, we hope that it will prove a useful contribution to 278 decisions about fitness for work during the Covid-9 pandemic in the UK adult population. 279 280

Risk reduction framework for NHS staff at risk of 290 COVID-19 infection. Faculty of Occupational Medicine