What is happening to non-covid deaths?BMJ 2020; 369 doi: https://doi.org/10.1136/bmj.m1607 (Published 24 April 2020) Cite this as: BMJ 2020;369:m1607
All rapid responses
Classifying deaths from COVID-19: Why the official statistics will never reflect the true mortality from coronavirus, and how future studies could try to address this
We read with interest Professor Appleby's article on data on deaths from causes other than Covid-19. In our hepatopancreaticobiliary (HPB) unit, the preparation for Covid-19 has had a profound impact on our services and patients. Due to the nature of oncological HPB procedures (e.g. pancreaticoduodenectomy, hemihepatectomy), under normal circumstances many of our patients require a high-dependency unit (HDU) bed post-operatively. Because of the need to reserve and expand critical care facilities for the COVID-19 pandemic, our elective theatre lists have been (justifiably) hugely reduced; HPB surgery is now only proceeding on a case-by-case basis in those patients who if curative-intent resection was not performed imminently, they would likely become inoperable before normal service resumes. Unfortunately, many patients on the waiting list will develop progressive, inoperable disease before normal theatre capacity is reinstated. Whilst many of these patients would have developed postoperative cancer recurrence even under ordinary circumstances, there will be a number of patients who if they had undergone surgery more promptly, they would have been cured of their disease. In addition, access to palliative chemotherapy is now severely restricted for patients who have inoperable/metastatic disease at the time of diagnosis. Whilst these patients were previously expected to have some extension of good quality life from palliative treatment options, their life expectancies will be significantly reduced if they do not have any locoregional or systemic treatments for their cancer.
We have discussed these different groups of patients from the perspective of our practice, but these groups exist in any oncological surgical speciality (colorectal, oesophageal, etc.) and some medical oncological specialities (e.g. haem-oncology). The official mortality statistics for COVID-19 will capture those patients who die as a direct result from coronavirus infection, but not deaths in patients who die as an indirect result of the pandemic. Research is already being conducted investigating the impact of COVID-19 on mortality in surgery. To better capture the true mortality that includes additional and premature deaths as a result of the impact on clinical services from the pandemic, we suggest that COVID-19-related deaths could be classified in the following way:
• Primary death: Patients who died as a direct result of COVID-19 infection.
• Secondary death: Patients with potentially curable life-limiting diseases who did not die from COVID-19 infection, but died due to a lack of access to medical interventions.
• Tertiary death: Patients with incurable life-limiting diseases who did not die from COVID-19 infection, but died earlier than would be expected due to a lack of access to medical interventions.
As Professor Appleby's article suggests, it will be difficult to determine some of these figures. To include secondary and tertiary deaths, researchers investigating all-cause COVID-19 mortality will need to make use of large data sets from primary (e.g. CPRD, QResearch) and secondary care (e.g. hospital episode statistics). Comparisons will need to be made with previous years to calculate how COVID-19 has impacted access to services (e.g. number of patients diagnosed with operable cancer who proceed to curative-intent surgery, number of patients undergoing palliative therapies and overall survival from diagnosis). Finally, access to services and mortality data pre- and post-pandemic will need to be compared to see if there has been a lasting and permanent impact as a result of the reorganisation of healthcare services, even in the absence of COVID-19. Only when more expanded definitions of COVID-19-related mortality are used will the true impact of the pandemic be known.
Competing interests: No competing interests
In the recent data briefing John Appleby focuses on repeat cross-sectional data to illustrate that the number of excess deaths (the total number of deaths in a week minus the average number of deaths seen in the same week for the past five years) exceeds the number of deaths where COVID-19 is mentioned on the death certificate.  The implication is that the COVID-19 pandemic-related disruption in how healthcare is accessed (for example, patients delaying help-seeking for new complaints) or delivered (for example, the postponement of elective surgery) has a tangible impact on population mortality. Whilst this is almost certainly true, quantifying this impact is more complex than implied.
Naively one may consider the difference between the number of excess deaths and the number of COVID-19 related deaths to equal the excess deaths attributable to the lockdown. However, as Appleby also points out, it is almost certainly the case that some of the deaths attributed to COVID-19 would have occurred anyway either in patients dying with COVID-19 rather than from it, or where COVID-19 infection accelerated the end of life of terminally ill patients. As such, the number of true COVID-19 deaths that should be subtracted from the excess deaths is lower than the total number of recorded COVID-19 deaths.
There are also likely to be non-COVID-19 deaths that were averted by the lockdown--for example, because of reduced road traffic or industrial accidents. Such reductions may have contributed to the fall in A&E attendances and emergency admissions that has been observed and therefore the expected deaths should in theory be reduced to take this factor into account. Given these considerations it is possible to describe the excess deaths attributable to reduced non-COVID-19-related healthcare use as follows
Where EH is the number of non-COVID-19 excess deaths attributable to changes in health service use, TD is the total number of deaths, ED is the expected number of deaths based on historical data, DA is the number of deaths averted which are unrelated to COVID-19 but would have occurred had the lockdown not been applied (e.g. from road traffic accidents), CD is the total number of COVID related deaths and CD* is the number of COVID related deaths which would have occurred anyway.
Even the above approach is an oversimplification as it does not take into account the time varying nature of the problem. When we consider only week-by-week figures we implicitly focus on deaths occurring or potentially occurring during that week. The reality is more complex. For terminally ill patients survival may be very short - someone who may die with COVID-19 this week, may have died next week without contracting the virus. However, much of the indirect negative impact of COVID-19-related disruption to healthcare provision may take longer to accrue.
For example, if a cancer patient who until the COVID-19 pandemic would have been treated by ‘standard’ adjuvant chemotherapy to minimise the risk of tumour recurrence is not going to receive such treatment, then their risk of dying from cancer increases, but most resulting deaths will not occur until sometime, often many years, in the future. To fully understand the impact of the disruption to healthcare delivery and utilisation during the COVID-19 pandemic we will need to consider longitudinal data and account for period effects induced by this disruption.
Even in the simplistic weekly view, estimating the magnitude of deaths averted (DA) and the number of COVID-19 related deaths that would have happened anyway (CD*) is not at all straightforward and will require in-depth analysis of the cause and circumstances of death. However, the point remains that changes to the way the health service is accessed and delivered is almost certainly having a bigger impact than the one implied by the simple calculations presented in Appleby’s data briefing. 
1. Appleby J. What is happening to non-covid deaths? BMJ 2020;369:m1607. doi: 10.1136/bmj.m1607
Competing interests: No competing interests
While data on the epidemiology of Covid-19 remains poor, public health authorities need key information on several aspects of the disease, including lethality, to take decision . Case fatality rate (CFR) is the probability of dying being a case . For a given epidemic, CFR is, technically, the ratio of cumulative number of deaths over cumulative number of cases and should not be confused with the excess mortality due to the disease .
CFR is simple to understand and, especially in comparisons with other diseases, helps people perceive the risk at stakes. This simplicity should not blind us to challenges in its estimation, especially during the early course of an epidemic . The major determinant of CFR is which cases - infected (symptomatic or not), symptomatic (confirmed or not) or confirmed - are considered. There is indeed a preferential ascertainment of severe cases , i.e., symptomatic cases requiring healthcare. During an epidemic, in the absence of population-based sero-survey , estimating the number of all infected cases is often not possible. CFR is estimated most often among reported confirmed cases, that is, symptomatic cases with a positive test. Differences in testing strategy explain part of the huge differences in current report of Covid-19 CRFs.
The other major issue is the number of deaths that can be underestimated due to the time delay between the occurrence of the infection and the death (right censoring) [2, 6]. Miss-classification of the cause of death also bias counts and limit comparability between estimates. It can be difficult indeed to establish especially among old multimorbid patients if they die of the infection or merely with the infection . Cases’ age and comorbidities (confounding and selection biases) as well as the healthcare system performance (survivorship bias) have also an effect on CFR estimates .
Hence, by April 25 2020, while the reported crude CFR was 13% in Italy, it was less than 4% in Germany . This difference is due notably to the large number of tests performed in Germany leading to the identification, on average, of milder confirmed cases or asymptomatic infections compared to Italy. It may be also explained by a reporting delay between cases and deaths because the epidemic started later in Germany.
There is some data to provide a best guess estimate for CFR among symptomatic cases, through the triangulation of sources. Hence, modeling studies, accounting for reporting delays and under-reporting of cases, have estimated CFR of 1.4%  and 1.6%  in China. Another estimate, based notably on the long-term follow-up of infected patients in Wuhan, was 1.4% (95% CI: 0.9 to 2.1%) . Out of the 712 confirmed cases from the Diamond Princess boat, 13 have died, i.e., a CRF of 1.8%. Overall, it suggests that CFR among symptomatic could be between 1 and 2%. Because CFR increases strongly with age and comorbidities, the comparability between populations of different age structures is however limited .
Estimating the CFR among infected, even roughly, is more difficult. Assuming that the number of asymptomatic or minimally symptomatic cases is several times as high as the number of reported cases, the infected CFR is certainly be lower than 1%. In China, it was estimated to be 0.7% and strongly dependent on age (<60 years old: 0.1%; ≥60 years old: 3.3%) . Based on a recent sero-survey conducted in Germany, the infected CFR was estimated to be 0.4% .
Comparisons with other epidemic infectious diseases is disputable but a rough ranking is possible. Covid-19 CFR is certainly lower than for Ebola (25% to 90% ), smallpox (up to 30% ), or 2002/03 SARS (up to 17% among symptomatic ) and lower than for influenza during the “Spanish” pandemic (2 to 3% among infected) [13, 14]. It is probably higher than for measles in high-income countries as well as influenza during seasonal epidemics (both <0.1% among symptomatic) [14, 15].
Prof Arnaud Chiolero: Population Health Laboratory (#PopHealthLab), University of Fribourg, Switzerland
Dr Stéphane Cullati: Population Health Laboratory (#PopHealthLab), University of Fribourg, Switzerland
Prof Antoine Flahault: Institute of Global Health, Faculty of Medicine, University of Geneva, Switzerland
1) Wu JT et al. Estimating clinical severity of COVID-19 from the transmission dynamics in Wuhan, China. Nat Med 2020; 26(4):506-10
2) Lipsitch M et al. Potential Biases in Estimating Absolute and Relative Case-Fatality Risks during Outbreaks. PLoS Negl Trop Dis. 2015; 9(7):e0003846.
3) Appleby J. What is happening to non-covid deaths? BMJ 2020; 369: m1607
4) Nishiura H. Case fatality ratio of pandemic influenza. Lancet Infect Dis. 2010;10(7):443-444.
5) Flahault A. Has China faced only a herald wave of SARS-CoV-2? Lancet 2020; 395(10228):947.
6) Althaus C. Estimating case fatality ratio of COVID-19 from observed cases outside China. https://github.com/calthaus/ncov-cfr, accessed March 23, 2020
7) Onder G et al. Case-Fatality Rate and Characteristics of Patients Dying in Relation to COVID-19 in Italy. JAMA 2020; 10.1001/jama.2020.4683.
8) Verity R. Estimates of the severity of coronavirus disease 2019: a model-based analysis. Lancet Infect Dis. 2020; pii: S1473-3099(20)30243-7
9) Centre for Evidence-Based Medicine (CEBM). Global Covid-19 Case Fatality Rates. https://www.cebm.net/global-covid-19-case-fatality-rates/, accessed April 25, 2020
10) Russell TW et al. Using a delay-adjusted case fatality ratio to estimate under-reporting https://cmmid.github.io/topics/covid19/severity/global_cfr_estimates.html, accessed April 28, 2020
11) World Health Organization. Ebola www.who.int/news-room/fact-sheets/detail/ebola-virus-disease, accessed March 30, 2020
12) John Hopkins University Center for Health Security. Smallpox www.centerforhealthsecurity.org/our-work/publications/smallpox-fact-sheet, accessed April 28, 2020
13) Qualls N et al. Community Mitigation Guidelines to Prevent Pandemic Influenza - United States, 2017. MMWR Recomm Rep 2017; 66(No. RR-1):1-34.
14) Bautista E et al. Clinical aspects of pandemic 2009 influenza A (H1N1) virus infection. N Engl J Med 2010;362(18):1708-1719.
15) Cairns KL et al. Challenges in measuring measles case fatality ratios in settings without vital registration. Emerg Themes Epidemiol. 2010;7(1):4.
Competing interests: No competing interests
Covid Blues or Blue skies?
Appleby’s recent paper in the BMJ  elegantly presents data for non Covid deaths detailing a fall in emergency admissions to A&E of 23%. It has been suggested that this is for psychological reasons; people are too anxious or depressed or concerned about being a burden on the NHS (Covid Blues) to come to A&E because of Covid19. Concerns are that such behavioural avoidance may result in people with cardiovascular disease and asthma not receiving treatment promptly.
To our surprise, with lockdown, the authors (MS and MB) independently observed significant falls in our own blood pressure and an improvement in asthma and considered the possible reasons for this. The link between air quality and blood pressure , and asthma , has been extensively documented. With lockdown, the air quality in major cities across the world, including central London, has significantly improved with Blue Skies observed. Thus although Covid Blues may part of the reason for non attendance at A&E, improvements in air quality (Blue Skies) known to be associated with lower blood pressure and improved lung function, and thus reduced morbidity and mortality may offer another explanation. Similar falls in morbidity and mortality were seen after a smoking ban in public places .
We acknowledge this hypothesis remains to be tested. Once more data are available, analyses of A&E attendance before and after the Covid19 outbreak could be conducted against other potential predictors of attendance, including presenting condition (incorporating, cardiovascular disease and asthma in the model), air quality and location (e.g. Atlantic seaside towns and big UK cities). If A&E attendance and reduced health care costs are associated with improvements in Blue Skies rather than Covid Blues, considerable weight would be added to the argument for the need to improve air quality, especially in major conurbations.
Dr Marc A Serfaty, Associate Professor in Psychiatry UCL, and Dr Mike Beary, Consultant Psychiatrist (Retired).
1. Appleby John. What is happening to non-covid deaths? BMJ 2020; 369 :m1607
2. Giorgine P, Di Giosa P, Grassi D, Rubenfire M, Brook RD, Ferri C. Air Pollution Exposure and Blood Pressure: An Updated Review of the Literature. Current Pharmaceutical Design 2016;22(1):28-51. https://www.ingentaconnect.com/content/ben/cpd/2016/00000022/00000001/ar....
3. Guarnieri M, Balmes JR. Outdoor air pollution and asthma. Lancet 2014;383(9928):1581–1592. doi:10.1016/S0140-6736(14)60617-6.
4. Lightwood JM, Glantz SA (2009) Declines in Acute Myocardial Infarction After Smoke-Free Laws and Individual Risk Attributable to Secondhand Smoke. Circulation. 2009;120:1373–1379. https://doi.org/10.1161/CIRCULATIONAHA.109.870691
Competing interests: No competing interests
Is the overall death rate really increasing exponentially and has a balanced assessment been portrayed? The recent synopsis (1) presents a detailed account of the situation in an attempt to redress the balance, but what will happen in the longer term. COVID-19 represents an unprecedent challenge and its impact is extremely difficult to compare with other previous global health medical disorders. In the UK we are informed of startling numbers of deaths due to COVID-19 daily, and we are extremely saddened by each death, especially in those who had so much life left and much to offer society. However what is less clear is the impact that COVID-19 has had and will have on indirect deaths; due to a potentially future overwhelmed NHS (it is this the government are aiming to minimise); the impact on “essential services” as has been discussed in this excellent article, yet patients dying of other conditions receive no mentioned in the press or news and it would seem they are less important. However one hopes the recent attention will restore the balance that any death deserves equal importance and all people are of equal importance and that COVID deaths should not receive the only mention. Without the balance people will continue to avoid the NHS. Will this group of people stay at home until their condition becomes so severe that we lose that so-called golden hour for optimal care?
This might be particularly true for acute kidney injury where much time has been spent identifying and preventing “crash landers” who have a significantly worse prognosis (2). This may also impact the impending crisis we face for dialysis materials rather than ventilators. There are many unintended consequences of COVID-19 but are the daily pessimistic and sometimes frightening messages accurate and indeed helpful? Indeed in the longer term will the impact of social isolation on mental welfare and suicides lead to further deaths?
If one examines the data on deaths in England and Wales from the National Registry from the past 14 years, many Healthcare workers and indeed the public are unaware that there are normally approximately 40,000-50,000 deaths per month in England and Wales, which equates to approximately 1600 deaths per day (3). It appears to have impacted weekly figures but will COVID-19 have an impact on the annual mortality figure? The question is will it actually increase the overall annual mortality or have brought forward deaths or replaced deaths from other medical causes which may have occurred in the next 12 months.
Currently COVID-19 accounts for 4% of all deaths. In the coming year we will see if the mortality curve spikes as a result of COVID-19 on top of other deaths then drops lower in subsequent months as a result of the premature death of those patients who may have died in the coming year.
1. What is happening to non-covid deaths? BMJ 2020; 369 doi: https://doi.org/10.1136/bmj.m1607
2. Schwenger V, Morath C, Hofmann A, Hoffmann O, Zeier M, Ritz E. Late referral—A major cause of poor outcome in the very elderly dialysis patient. Nephrol Dial Transplant. 2006; 21(4): 962–967
Competing interests: No competing interests