The case for counting multiple causes of death in the COVID-19 era

Abstract

From the onset of the COVID-19 pandemic caused by SARS-CoV-2, an unprecedented international mobilization occurred within the scientific community to expand our knowledge and mitigate the harmful effects of this new viral infection—the deadliest of the past two decades.1 The sustained attention towards COVID-19 resulted in reduced priority given to monitoring mortality trends for other causes of death. Although a PubMed database search with the keyword ‘COVID-19’ yielded >376 000 scientific publications as of July 2023, the number of publications concerning the majority of other leading causes of death have plummeted since the emergence of SARS-CoV-2.2 For instance, publications on cardiovascular diseases—the world’s leading cause of death3—fell short by ∼25% in 2020 compared with 2019,2 even though their death toll has exceeded COVID-19 deaths since the beginning of the pandemic.4,5 Yet, early on in this global health crisis, concerns were raised regarding many potential undesirable direct and indirect effects of the pandemic on non-COVID-19 mortality. Direct effects include a range of long-term complications and sequelae arising from COVID-19, such as myocarditis, acute myocardial injury, venous thromboembolism, diabetes mellitus and renal injury, which are likely to increase the risks of death for these specific diseases.6 There are multiple indirect effects stemming from the pandemic that also need to be considered. The surge of COVID-19 patients in hospitals caused a strain on available resources, including beds, equipment and healthcare personnel. The saturation of hospital resources affected both COVID-19 patients and individuals admitted for other medical conditions. In hospitals that dealt with a moderate to high number of severe COVID-19 patients, an increased death toll from other causes was recorded.7 Hospital avoidance due to the fear of contracting COVID-19 also led to indirect deaths. In Italy, for example, emergency department visits decreased by two-thirds in 2020 compared with 2019, coinciding with a significant increase in out-of-hospital mortality.8 Moreover, the limited disease screening and diagnostic activities during the pandemic9 may have impeded the identification of the underlying cause of death (UCoD, i.e. that initiated the morbid process or producing the fatal injury; see next section for formal definition) or contributing causes (i.e. other than underlying). This can lead to a false impression of reduced mortality for these illnesses.10 In the long run, however, simulation models suggest that the consequences of these reductions could result in excess mortality for certain causes, such as cancer.11,12 Strict confinement measures to limit virus transmission also hold potential to indirectly alter mortality trends for various non-COVID-19 causes, in both the short and the long term. Studies found an increase in death rates for accidental poisoning (drug overdose, alcohol) and alcohol-related liver diseases,13 as well as suicide in certain subgroups (e.g. youth and older adults)14,15 due in part to social isolation, negative effects on mental health and limited access to social services. Conversely, sanitary and social distancing measures likely aided mortality decline for communicable diseases, such as influenza,16 whereas teleworking,—by limiting motorized travel—likely played a role in reducing deaths related to road accidents and air pollution.17 Despite concerns that the COVID-19 pandemic could lead to multiple health crises given the above-mentioned impacts on other diseases, mortality trends during the pandemic were mainly studied without paying attention to causes of death, possibly because of the biases in identifying COVID-19 deaths and the delay in obtaining cause-specific mortality data. During the pandemic, scholars often focused on excess mortality18—a widely adopted measure in which all deaths from various causes are pooled to avoid the pitfalls of identifying COVID-19 deaths directly. Although the measure is useful for making more reliable comparisons across regions and over time, it fails to reveal the broader impact of COVID-19 on other causes of death. Among studies that examined cause-specific mortality trends, most focused solely on the UCoD, with only a few exceptions.19–28 To our knowledge, one systematic literature review investigated the impacts of the COVID-19 pandemic on cause-specific mortality trends; however, it did not differentiate between UCoD and contributing causes of death (CCoD).29 In light of the expected upheavals in non-COVID-19 mortality trends and given the insufficient attention paid to CCoD since the onset of the pandemic, there is an urgent need for cause-specific mortality studies that account for all causes listed on death certificates—regardless of whether they appear as the UCoD or CCoD. The disproportionate attention given to the UCoD in the literature can be explained by the fact that it is often seen as the determinant cause to act upon from a public health perspective.30 Death certificates, however, collect additional indispensable information, including all other causes that led to the person’s death. These causes are recorded in two distinct parts of the death certificate, as illustrated in Figure 1.30 In Part 1, the medical certifier must document the causal chain of diseases, injuries or complications that led ‘directly’ to the death, including the UCoD that initiated the morbid process. The chain of events can include up to three or four causes (depending on the region) that are usually listed in reverse chronological order, so the UCoD usually appears last in Part 1. In Part 2, the medical certifier must list, if applicable, all other diseases, injuries or conditions that ‘indirectly’ contributed to the death, meaning those not involved in the direct causal chain. Once gathered, this information is then validated by an automated coding system to select, for each death certificate, the definitive UCoD in accordance with the World Health Organization (WHO) instructions.30 The definitive UCoD may therefore differ from what was reported on the death certificate. International form of medical certificate for cause of death. Reprinted with the permission of the World Health Organization from: International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Volume 2—Instruction Manual, Fifth Edition 2016, WHO, p. 203, Copyright (2011). https://icd.who.int/browse10/Content/statichtml/ICD10Volume2_en_2019.pdf As per convention, the UCoD refers to ‘(a) the disease or injury which initiated the train of morbid events leading directly to death, or (b) the circumstances of the accident or violence which produced the fatal injury’.30 The CCoD are all other conditions listed on the death certificate, which can be either consequences or complications of the UCoD (usually listed in Part 1), risk factors for the UCoD or conditions that may interact with the UCoD (usually listed in Part 2).31 The analysis of multiple causes of death (MCoD) entails the examination of all causes mentioned on the death certificate, encompassing both the UCoD and the CCoD. MCoD analysis is far from novel. It first emerged in the 1940s and its significance has continued to grow in recent years.32 With population ageing, the prevalence of chronic and degenerative diseases has increased and a rising proportion of deaths now result from the complex interaction of more than one disease, trauma or condition. It is therefore not surprising that, in countries such as Australia, ∼80% of death certificates count at least two causes and, on average, between 2.4 and 3.2 causes are recorded for each death in high-income countries.32,33 To date, MCoD analysis has pursued several objectives. Most commonly, these studies investigated associations between the various causes leading to death and reassessed the burden of specific causes on total mortality, especially those frequently listed as CCoD rather than UCoD.32 We strongly advocate expanding this list to include the analysis of cause-specific mortality trends during health crises, where major disruptions are expected to impact the distribution of causes of death. The COVID-19 crisis has several unique characteristics that further emphasize the importance of using MCoD. First, most COVID-19 deaths are the result of a complex interplay between COVID-19 complications and pre-existing comorbidities.24,28,34 In the USA, nearly all (94.5%) death certificates with COVID-19 as the UCoD also listed at least one additional cause of death in 2020 compared with 78.8% for all death certificates in 2019.34 Using the full spectrum of causes enhances comprehension of the intricate interactions among the unusually high prevalence of health conditions involved in COVID-19 deaths. MCoD can also inform the various pathways linking COVID-19 as the UCoD with its fatal complications. Second, accurate identification and quantification of deaths attributable to COVID-19 have been challenging for mortality statistics, as there are varying definitions and screening strategies across regions and over time, as well as limited testing capacities.35 Precise determination of both the UCoD and CCoD is, however, essential for obtaining reliable cause-specific mortality trends. For these reasons, caution is advised when analysing COVID-19 mortality based solely on the number of registered COVID-19 fatalities.18,35 The warning may also indirectly apply to other causes of death, as an UCoD erroneously attributed to COVID-19 can also impact the ‘true’ UCoD, which may be relegated to a CCoD (and vice versa). These misattributions can thus induce artificial variations in causes of death other than COVID-19. Unlike genuine mortality breakpoints, artificial ones do not align with explanatory factors of mortality change, lack consistent patterns across related causes and/or do not persist over time. Instead, they appear as isolated and inconsistent disruptions. MCoD analysis can help in investigating such sudden unexplained shifts in cause-specific mortality by comparing trends based on UCoD alone and on both UCoD and CCoD, for instance. Third, the introduction of new instructions for coding causes of death since the beginning of the pandemic is another potential major source of artificial variations in mortality trends, much like any new coding rules. About 5 weeks after the declaration of the global pandemic, the WHO released international guidelines for coding COVID-19 deaths on death certificates.36 To better track the progression of mortality due to COVID-19, these guidelines favoured the selection of COVID-19 as the UCoD, potentially disrupting trends in other causes of death and inflating COVID-19 as an UCoD compared with CCoD. The document states that for surveillance purposes, ‘COVID-19 should be recorded on the medical certificate of cause of death for ALL decedents where the disease caused, or is assumed to have caused, or contributed to death’.36 According to the WHO’s definition, a death is considered to be due to COVID-19 when it results ‘from a clinically compatible illness, in a probable or confirmed COVID-19 case, unless there is a clear alternative cause of death that cannot be related to COVID disease (e.g. trauma)’.36 Thus, COVID-19 must be recorded as the UCoD even when the virus is only suspected or probable. In March 2020, the WHO introduced two new codes for this purpose: ‘U07.1 COVID-19, virus identified’ and ‘U07.2 COVID-19, virus not identified’. COVID-19 should also be selected as the UCoD if the decedent had other pre-existing conditions, irrespective of their severity, duration or timing of onset. In fact, the WHO states that deaths due to COVID-19 ‘should be counted independently of pre-existing conditions that are suspected of triggering a severe course of COVID-19’.36 Several diseases or comorbidities have been identified as increasing the risks of death from COVID-19, including cancer,37 cardiovascular diseases,38 hypertension39 and diabetes.40 As per the WHO’s guidelines, even if these comorbidities or pre-existing conditions are suspected of having played a role in the development of a severe form of COVID-19 that eventually led to death, these should be strictly recorded as CCoD.41 Although not specific to COVID-19 deaths, this rule has the potential to significantly alter mortality trends for COVID-19-related causes, especially those usually designated as the UCoD, including cancer and cardiovascular diseases.31 By exclusively examining the UCoD, there is a risk of underestimating the burden of these more lethal causes during the pandemic, creating a misleading perception of decreasing severity as they are increasingly identified as CCoD. The rules for determining the causal chain also prioritize COVID-19 as the UCoD. The WHO established that ‘COVID-19 is not considered as due to, or as an obvious consequence of, anything else in analogy to the coding rules applied for INFLUENZA’.36 To comply with this, deaths due to COVID-19 should therefore appear at the beginning of the causal chain leading to death (i.e. UCoD). The WHO specifies that these instructions regarding the causal chain must always be applied ‘whether they can be considered medically correct or not’36 and that it is not possible to ‘modify its coding in any way’36 since these deaths are of public health interest. Thus, the selection of COVID-19 as the UCoD is based not only on medical observations, but also on epidemiological and public health considerations.42,43 For death certificates on which COVID-19 is listed but not retained as the UCoD, the WHO recommends conducting a manual verification. The three above-mentioned specificities of COVID-19 stressing the importance of using the MCoD are further accentuated by the fact that SARS-CoV-2 emerged suddenly and surged, rapidly making COVID-19 one of the three leading causes of death in most high-income countries.4,5,44 With >6 952 000 deaths due to COVID-19 globally as of July 2023,45 the abrupt appearance of such a highly prevalent new cause is expected to have impacted the distribution of other causes of death in the population. Indeed, COVID-19 is not an independent cause of death.46 Individuals whose death is attributed to COVID-19 have escaped the risk of dying from another cause, which mechanically decreases the contribution to mortality of these non-COVID-19 causes. The scant but noteworthy empirical works on MCoD show the importance of this type of analysis and its potential in evaluating the influence of the COVID-19 pandemic on other causes of death. One ongoing strand of research consists of monitoring mortality trends prior to and during the pandemic, often by calculating changes in age-standardized death rates and in ratios of observed vs expected death counts, or by identifying turning points in trends with joinpoint and linear regression models. Studies using these approaches for both the UCoD and MCoD reveal the potential distortions that may arise when solely the UCoD is included.21,22,23,25,26 In Brazil, for instance, whereas cancer and cardiovascular deaths declined during the pandemic as the UCoD (–9.7% and –8.8%, respectively), there was instead a substantial increase when analysed as the CCoD (82.1% and 77.9%).25 Prior to COVID-19, cancers and cardiovascular diseases were often recorded as the UCoD but, in this case, COVID-19 probably took precedence over these, leaving them as CCoD. Likewise, in Italy, the number of deaths for chronic obstructive pulmonary disease declined by 8% before and after the onset of the pandemic from the sole perspective of the UCoD, but increased by 14% when accounting for MCoD.21 Another ongoing area of research involves using MCoD to identify the prevailing complications associated with COVID-19 fatalities, documented in Part 1 of death certificates, as well as the most common comorbidities, typically listed in Part 2. This is achieved by calculating frequencies and percentages of association. Evidence suggests that the most prevalent complications in Italy and the USA were pneumonia (78% and 54%, respectively) and acute respiratory failure (54% and 20%, respectively).24,34 In Brazil, however, sepsis (11%) and severe acute respiratory syndrome (SARS; 10%) emerged as the primary complications of adult COVID-19 deaths.28 Variations in death certification practices between countries might partially account for the discrepancies in the identified complications and their respective prevalence. In contrast, hypertension and diabetes consistently emerged as the two most prevalent comorbidities associated with COVID-19 fatalities in Italy, Brazil and the USA, exhibiting similar frequencies (hypertension: 11–18%; diabetes: 10–16%).24,28,34 The development of other indicators and analytical methods, specifically designed to address the complexities of MCoD,32,47 hold great potential as future directions for monitoring mortality trends during the COVID-19 pandemic. The cause-of-death association indicator is noteworthy, as it is an age-standardized metric that enables direct comparisons between causes of death and regions. It could thus serve as a robust measure for assessing non-COVID-19 causes of death involved with COVID-19 fatalities. Indeed, the cause-of-death association indicator evaluates the strength of association between two causes by determining whether a CCoD is more frequently linked to a specific UCoD than with all UCoD combined.32 Another interesting measure is the standardized ratio of multiple-to-underlying cause. It could be employed to evaluate changes since the onset of the pandemic in terms of the extent to which a given cause is identified as a CCoD relative to the UCoD.32 Lastly, MCoD weighting techniques have been proposed as another option to MCoD age-standardized death rates. These allocate weights to each cause listed on the death certificate in such a way that the total weight sums up to 1.48 Because each death is counted only once, conventional cause-specific mortality measures are applicable (e.g. rates, cause-deleted life tables), which could undoubtedly be useful for COVID-19 studies.47 In their review, Sanmarchi and colleagues29 stressed the importance of establishing an international consensus on methodological approaches to improve our estimates of the COVID-19 pandemic impacts on cause-specific mortality trends. In reaching this goal, researchers should consider the use of MCoD, which profits from all causes listed on the death certificate, whenever these data are available and reliable. As we discussed in this paper, such an approach is crucial due to the high prevalence of comorbidities involved in COVID-19 deaths, because of the many challenges in correctly identifying COVID-19 deaths and given that coding rules tend to prioritize COVID-19 as the UCoD. The analysis of MCoD provides a more comprehensive understanding of mortality shifts between UCoD and CCoD that can be induced by the COVID-19 pandemic, while avoiding biases associated with the selection of the UCoD. Like any cause-specific mortality assessment, MCoD analysis requires accurate and rigorous coding of both UCoD and CCoD. Recommendations were issued to ensure uniform coding practices that facilitate international comparisons.49 Valuable insights into current practices and analytical methods used to study MCoD are also available.32,47 Given the conceptual and computational complexity involved, we strongly encourage the development of innovative methodological approaches that incorporate MCoD. Ethics approval is not needed because no human patients were involved; this opinion paper exclusively builds on published articles and their respective findings, as well as materials that were publicly released by statistical or health agencies without additional calculations. No new data were generated or analysed in support of this opinion paper. M.P.P.: conducted the literature review, conceptualized and wrote the first draft. N.O.: made a significant contribution to manuscript writing and revision. N.O. and R.B.: critically commented on the manuscript. M.P.P., N.O. and R.B.: read and the manuscript. This was by two from the and of and

Topics & Concepts