What can multiple causes of death tell about cardiovascular mortality during COVID-19 pandemic in the United States?

Abstract

Background

The COVID-19 pandemic may have caused an underestimation of cardiovascular disease (CVD) mortality, as COVID-19 was predominantly recorded as the underlying cause of death. This study investigates CVD-related excess mortality and recording of CVD on the death certificates during 2020–2021, considering underlying (underlying causes of death (UCD)), immediate and contributory causes.

Methods

We utilize US Multiple-Cause-of-Death Mortality Data. Excess deaths are assessed by comparing actual 2020–2021 deaths with Seasonal Autoregressive Integrated Moving Average model predictions. To understand changes in cause-of-death recording, we use the standardized ratio of multiple to underlying causes (SRMU).

Results

Excess CVD mortality is most prominent in contributory causes, including hypertensive disease, essential hypertension, and acute myocardial infarction. While excess of contributory CVDs generally decreased in 2021, acute myocardial infarction, pulmonary heart diseases and other circulatory diseases showed a continual increase. Changes in SRMU from 2020 to 2021, compared to 2010–2019, reveal shifts in coding practices, particularly for pulmonary heart, cerebrovascular diseases, non-rheumatic valve disorders and heart failure.

Conclusions

The COVID-19 pandemic has significantly increased CVD-related mortality, which is not fully captured in conventional analyses based solely on the UCD. The trend of coding CVDs as non-underlying causes of death accelerated during 2020–2021. Multiple-causes-of-death should be employed to evaluate mortality when new leading cause of death emerges.

Introduction

During COVID-19 pandemic, excesses in cardiovascular mortality have been reported repeatedly.^1–4 The key reasons behind the increased number of cardiovascular deaths in 2020–2021 are mostly attributed to increased risk of death in patients with preexisting cardiovascular diseases (CVD),^5,6 limited access to health care services,⁷ people’s preference not to seek for medical help due to fear of the infection or the potential misclassification of death from COVID-19.⁸ Not only CVD but also conditions like Alzheimer disease, dementia and diabetes have followed similar trends.^9,10 Research has indicated that quantifying the excess of deaths from these diseases using a single underlying cause of death (UCD), which refers to the medical condition that initiated a train of events leading to death, may result in underestimation.¹⁰ The inclusion of all medical conditions listed on the death certificate in the analysis can help mitigate this issue. The multiple cause of death approach, as it is commonly referred to, has been shown to be more robust when estimating the excess deaths attributable to conditions other than COVID-19.¹¹

So far, a systematic multiple cause mortality analysis of CVD during COVID-19 pandemic is missing, even though CVDs belong to leading causes of death in developed countries.¹² In this study, we use US multiple cause datafiles¹⁵ to calculate the time-varying excess of deaths related to CVD from a multiple cause perspective. Additionally, we employ detailed death certificate data to examine shifts in how CVD were recorded as causes of death on the death certificate during the COVID-19 pandemic. We stratify the analysis by place of death, as previous research has identified this as a significant factor associated with recording of causes of death.^11,13

Aims of the paper are:

• (i) To quantify the excess of multiple cause mortality from CVD grouped by ‘113 list of causes of death’;

• (ii) To investigate whether there were notable changes in the recording of CVD on death certificates during the pandemic;

• (iii) To examine the importance of place of death in coding strategies of CVD during the pandemic.

Materials and Methods

To address the aims, we employ Seasonal Autoregressive Integrated Moving Average (SARIMA) models to analyze time series on deaths resulting from CVD during 2010–2021. These deaths are categorized in accordance with the 113 ICD-10 Cause of Death list into 16 groups, which are specified in Table 1. We further distinguish these cardiovascular disease-related deaths based on their position on the death certificate as immediate, contributory and associated causes. Our approach involves forecasting number of deaths by cause and type based on trends observed in the years 2010–2019. Subsequently, we compare the forecasted number of deaths with the actual observed figures for the pandemic years 2020–2021, resulting in an estimation of excess deaths. Lastly, we compute the SRMU to UCD, which serves as an indicator revealing changes in the recording of causes of death.

Table 1.

Estimates of percentage excess of observed death from SARIMA model predictions by type of multiple cause of death and year

		Underlying cause of death		Immediate cause of death		Contributory cause of death
ICD-10 codes (revision 2019)	Cause of death group	2020	2021	2020	2021	2020	2021
I00–I09	Rheumatic heart diseases	4.08 (−10.31)	2.84 (−13.7)	7.49 (−18.20)	4.77 (−20.99)	0.80 (−16.66)	−3.12 (−28.8)
I10	Essential hypertension	13.20 (1.73)	12.15 (−5.91)	10.62 (2.11)	3.72 (−9.70)	35.07 (25.55)	40.23 (23.22)
I11–I15	Hypertensive disease (heart, kidney and secondary)	9.11 (2.91)	5.59 (−3.61)	5.82 (−0.51)	−0.49 (−9.25)	42.76 (35.48)	27.96 (19.96)
I21–I23	Acute myocardial infarction	7.73 (−0.88)	11.55 (−3.73)	23.33 (10.32)	29.04 (10.18)	38.86 (23.52)	56.36 (31.01)
I20, I24, I25	Other IHD	6.16 (1.51)	3.41 (−1.51)	10.78 (3.09)	9.49 (−0.83)	31.74 (23.57)	30.39 (19.07)
I26–I28	Pulmonary heart diseases	2.41 (−4.76)	3.68 (−6.58)	17.15 (9.56)	30.37 (21.75)	18.25 (11.42)	33.30(23.63)
I34–I38	Non rheumatic valve disorders	−5.89 (−12.46)	−6.74 (−15.48)	1.65 (−7.89)	4.49 (−5.99)	10.02 (3.31)	9.68 (1.87)
I46	Cardiac arrest	6.18 (−5.07)	9.20 (−7.45)	13.16 (5.98)	8.67 (−0.87)	27.12 (1.53)	27.88 (−10.42)
I50	Heart failure	−3.94 (−10.02)	−8.16 (−16.52)	3.55 (−3.07)	4.73 (−5.37)	21.12 (13.13)	19.07 (7.85)
I30–I33, I40–I45, I47–I49, I51	Other heart diseases	5.24 (−4.48)	9.53 (−10.32)	12.56 (4.26)	14.65 (1.47)	17.88 (9.63)	21.07 (6.69)
I60–I62	Intracranial haemorrhage	2.28 (−4.24)	6.55 (−4.11)	1.94 (−10.39)	10.73 (−8.31)	15.48 (−1.26)	14.42 (−8.43)
I63, I65, I66	Cerebral infarction, occlusion, and stenosis	8.28 (−13.9)	13.43 (−25.89)	21.16 (−6.22)	41.83 (−15.14)	21.17 (6.34)	17.54 (−6.64)
G45, I64, I67	Other cerebrovascular diseases	7.35 (−2.30)	7.35 (−8.92)	9.33 (−1.02)	9.68 (−6.78)	27.82 (20.29)	21.27 (10.67)
I69	Sequelae of cerebrovascular disease	5.28 (−2.26)	−0.42 (−11.03)	14.94 (4.67)	8.24 (−2.76)	29.37 (19.05)	22.14 (8.66)
I70–I78	Diseases of arteries, arterioles and capillaries	−0.38 (−6.28)	3.46 (−4.30)	0.99 (−19.18)	−2.92 (−36.14)	20.48 (13.08)	19.01 (8.58)
I80–I99	Other circulatory diseases	15.13 (2.95)	18.55 (4.98)	7.67 (−2.91)	15.4 (−0.37)	14.43 (6.33)	23.95 (10.24)

Note: Numbers in brackets indicate percentage excess of observed death from higher confidence bounds of SARIMA model predictions.

Data

The Multiple Cause-of-Death Mortality Data was obtained from the National Bureau of Economic Research,¹⁴ which compiles microdata on mortality from the National Vital Statistics System of the National Center for Health Statistics.¹⁵ Each record in the microdata is based on information extracted from death certificates. In accordance with the international guidelines set by the World Health Organization (WHO), the U.S. death certificate has two sections for recording causes of death. Part one consists of four lines to record the chain of morbid events leading to death, where the UCD is the condition that initiates this sequence, and the immediate cause of death is the final medical condition preceding the death itself. Part two of the death certificate is designated for conditions that contribute to death but do not have a significant direct relationship to the underlying cause.¹⁶

In the current paper, we utilize Entity Axis Codes, which are detailed records of multiple causes of death. They provide information about the exact position of the cause of death code on the death certificate, aiding in the identification of immediate and contributory conditions. Here, the immediate cause (IMM) of death is defined as the medical condition recorded on the highest line among all filled lines on the death certificate, the position of which does not coincide with the position of UCD. Contributory causes (CC) are defined as codes recorded in the second part of the death certificate. Furthermore, we also incorporate the ‘any mention approach’ (ANY), which involves considering all recorded causes, regardless of their position, including the underlying one. This approach is also sometimes referred to as ‘associated causes’.¹⁷

Methods

SARIMA models are a widely used methodological approach for modeling time series data that display a significant seasonal pattern, such as deaths from cardiovascular disease. The general form of a SARIMA model can be expressed as ARIMA (p, d, q) (P, D, Q)_s. These models adhere to the established Box-Jenkins methodology, which is described in more detail in other sources.^18–22 The specification of SARIMA models can be written as^23,24:

Where is ordinary autoregressive component and is the ordinary moving average component and and are their analogical seasonal terms. Power D is order of seasonal differencing and d is non-seasonal differencing, s is number of seasons per years (here s = 12), B is the backshift operator, Z_t denotes time series at time t and is error term at time t. Further technical details are provided in supplementary file.

The process of estimating the model can be divided into several key steps:

1. Checking for Model Assumptions: Standard ARIMA models rely on the assumption of time series stationarity. To assess this, we employed the Bartlett test. If the original time series proved non-stationary, we applied differencing to the data, with the order of differences equal to the integration parameter in the model (d).

2. Identification of the Model: Determining the values of p, q, P, and Q, which represent the parameters for autoregressive (AR) and moving average (MA) components, resp. seasonal AR and MA.

3. Estimation of Model Parameters: Maximum likelihood method was used to estimate the parameters of the model.

4. Diagnostics: Ensuring that the model does not violate assumptions regarding the residuals, including their randomness, homoscedasticity, autocorrelation, and normality. Various tests are available to assess these assumptions, and we applied the following: Ljung-Box test for autocorrelation of residuals, ARCH test for homoscedasticity of residuals and Jarque–Bera test to rest for normality of distribution of residuals

5. Selection of the Best Model: To select the optimal model, we compared Akaike Information Criterion (AIC) and selected the model with the minimum AIC.

6. Forecasting: Finally, when the best performing model was found, we forecasted death for 2020–2021 based on time series observed in 2010–2019.

Steps 2 through 5 were executed iteratively until identifying the model with the most suitable parameter combinations that did not violate the assumptions. To systematically explore all possibilities, we created a loop that tested various combinations of parameters (p, q, P, Q) ranging from 1 to 5. All of these computations were carried out using R Studio 9.4. The specific parameter sets that were deemed optimal for each of the cause of death groups can be found in the supplementary information.

After modeling the time series, we calculated the excess deaths for 2020–2021 by finding the difference between the observed number of deaths and the number of deaths predicted by the best-performing SARIMA model. To assess the significance of this excess, we compared two trajectories: the observed number of deaths and the estimated upper scenario, which represents the upper confidence bound of the predicted number of deaths. To determine whether the increase in deaths during 2020–2021 was significant, we compared the observed total number of deaths in that period to the highest scenario of the prediction, which corresponds to the upper confidence interval of the SARIMA models. If the observed deaths exceeded this upper prediction limit, we concluded that the increase during 2020–2021 was significant. This frequentist approach enabled us to identify causes of death that experienced substantial fluctuations at the beginning of the pandemic, likely due to the lack of adequate rules for recording COVID-19 deaths. An example of predicted and observed time series for selected cardiovascular causes of death is shown in Fig. 1.

Fig. 1.

Observed and predicted number of deaths from essential hypertension as underlying, immediate and contributory cause of death, USA, 2010–2021.

Next, our focus shifted toward quantifying alterations in the cause of death recording during the pandemic. To achieve this, we computed ratios of multiple to underlying number of deaths for each year up to 2021 and compared these ratios to those from the base year, 2010. Furthermore, we employed an indicator known as the SRMU, the formula of which is defined²⁵:

Where and are deaths from multiple causes and underlying causes at age x, is mid-year population estimate (obtained from²⁶) at age x and is the weight of standard population at age x. We use WHO World Standard Population 2000–2025 obtained from National Cancer Institute.²⁷ The SRMU, or standardized ratio of multiple to underlying cause, is indeed a ratio of age-standardized death rates calculated based on multiple conditions to those calculated from underlying conditions. In the ‘any mention’ approach, which incorporates all cardiovascular conditions listed on the death certificate, regardless of their position (thus, including the underlying condition), a SRMU_ANY exceeding two indicates that the specific cause of death is primarily categorized as an associated cause rather than the underlying cause. For instance, an SMRU equal to two suggests that the cause of death is recorded equally as both the underlying condition and an associated condition. However, when assessing SRMU for contributory causes (SRMU_CC) and immediate causes (SRMU_IMM), the threshold naturally shifts to 1. This shift occurs because the numerator in these indicators includes only the conditions certified either as immediate or contributory. The conditions coded as underlying are not incorporated in the numerators of either SRMU_CC or SRMU_IMM as they are in SRMU_ANY. This implies that SRMU_CC > 1 and SRMU_IMM > 1 indicate that the condition is primarily contributory or immediate, respectively.

Results

In both 2020 and 2021, ~55% of deaths in the United States had at least one cardiovascular condition recorded on their death certificates, accounting for ~1.85 million deaths each year. Among these deaths, around 25% were attributed to underlying CVD (~0.95 million deaths). Notably, the percentage of deaths with contributory CVD increased from 20% in 2020 to 30% in 2021. Immediate CVD were recorded in ~40% of all deaths during both years of the COVID-19 pandemic. For further descriptive statistics see supplementary information.

Table 1 displays the percentage of observed deaths exceeding the predicted values for 16 distinct groups of CVD in the years 2020 and 2021. Diseases demonstrating significant deviations from the predictions, highlighted in yellow, indicate that the observed excess of deaths exceeded the upper confidence bound of the prediction.

When we exclusively consider CVD as the UCD, a significant increase was observed only in essential hypertension and hypertensive diseases, other IHD, and other circulatory diseases. The majority of these excess deaths occurred in 2020, shortly after the onset of the COVID-19 pandemic during the spring season. Notably, there were noteworthy reductions in the observed number of deaths attributed to specific underlying CVD, such as non-rheumatic valve disorders and heart failure, in the years 2020 and 2021.

A more substantial increase in deaths is observed when considering multiple causes of death. Specifically, in the pandemic years, we witnessed a pronounced surge in immediate cardiovascular causes of death. Notably, acute myocardial infarction, pulmonary heart diseases, and other heart diseases experienced the most significant increases in 2020, with ~23, 17 and 13% rises, respectively. This upward trend continued into 2021, especially for pulmonary heart diseases. Several other immediate CVD causes of death also saw remarkable increases, albeit primarily in 2020. For instance, this was the case for cardiac arrest (+6%), sequelae of CVD (+5%) and other IHD (+3%).

The excess mortality was most pronounced when examining contributory causes of death, with nearly all of them showing a significant increase in both years. Exceptions were observed for rheumatic heart diseases, cardiac arrest, intracranial hemorrhage and cerebral infarction. The notable increase of almost 40 percentage points compared to SARIMA predictions can be attributed to hypertensive disease (+42%), acute myocardial infarction (+38%) and essential hypertension (+35%). Furthermore, it is worth noting that, in the second year of the pandemic, there was an even more substantial increase in the observed number of deaths exclusively for essential hypertension, acute myocardial infarction, pulmonary heart diseases and other circulatory diseases.

Next, we examined the cause of death recording schemes of CVD from 2010 to 2021. According to the SRMU, CVD are perceived as predominantly (i) contributory (SRMU_CC > 1), (ii) immediate (SRMU_IMM > 1) or simply and (iii) associated (SRMU_ANY > 2) than underlying. SRMU <1 (or 2) indicates that a disease is mostly coded as the UCD. The classification of CVD into these categories is provided in Table 2. It is evident that, as previously documented, the classification depends on the place of death, as specified in the notes listed in Table 2. This is particularly relevant for diseases that do not clearly fit into either category, where SRMU_IMM or SRMU_CC are <1. Examples of such diseases include hypertensive disease, other IHD and cerebrovascular diseases (I63–I69). In these cases, the classification may vary depending on the place of death, whereas for other diseases, like for instance, essential hypertension, pulmonary heart diseases or cardiac arrest, it is more straightforward.

Table 2.

Classification of causes of death by standardized ratio of multiple to underlying cause of death, USA, 2010–2019

		Is average SRMU in 2010–2019 smaller than 1 (2)?
		SRMUCC	SRMUIMM	SRMUANY
I00–I09	Rheumatic heart diseases	Yes	Yes	See note 1
I10	Essential hypertension	No	No	No
I11–I15	Hypertensive disease (heart, kidney and secondary)	Yes	Yes	See note 2
I21–I23	Acute myocardial infarction	Yes	Yes	Yes
I20, I24, I25	Other IHD	Yes	Yes	See note 1
I26–I28	Pulmonary heart diseases	See note 3	No	No
I34–I38	Non rheumatic valve disorders	Yes	Yes	No
I46	Cardiac arrest	Yes	No	No
I50	Heart failure	See note 4	No	No
I30–I33, I40–I45, I47–I49, I51	Other heart diseases	No	No	No
I60–I62	Intracranial hemorrhage	Yes	Yes	Yes
I63, I65, I66	Cerebral infarction, occlusion, and stenosis	Yes	Yes	See note 5
G45, I64, I67	Other cerebrovascular diseases	Yes	Yes	See note 4
I69	Sequelae of cerebrovascular disease	Yes	Yes	See note 4
I70–I78	Diseases of arteries, arterioles and capillaries	See note 2	See note 3	No
I80–I99	Other circulatory diseases	No	No	No

Note: 1: Only outside hospital death

2: Only until 2017

3: Only in hospital death

4: Only death at home and unknown

5: Only since 2016 death at home and unknown

SRMU_CC indicates SRMU for contributory causes of death, SRMU_IMM indicates SRMU for immediate causes of death, SRMU_ANY indicates SRMU for associated conditions (‘any mention approach’)

Data availability: Supplementary information

Having now obtained an overview of how certain CVD were certified in the pre-pandemic years from 2010 to 2019, we can proceed with the analysis of the changes in these coding schemes. Firstly, we compare the ratio of multiple causes of death to UCD for each of the 16 CVD groups. The trends of these indices, relative to the base value in 2010, are depicted in Fig. 2, separately for the summer and winter, which corresponds to the COVID-19 waves in the USA. Additionally, these figures specify the type of multiple cause of death (contributory or immediate).

Fig. 2.

Ratios of multiple (contributory and immediate) cause of deaths records to underlying cause of deaths records related to baseline 2010, by group of cardiovascular diseases and season, USA, 2010–2021.

We observe a clear and consistent increase in the burden of contributory causes of death during the winter season, a trend that has been noticeable since 2010 but particularly accelerated during 2020–2021 for most of the CVD. The most notable trajectories that follow this pattern can be seen in acute myocardial infarction, intracranial hemorrhage and pulmonary heart diseases (Fig. 2). During the summer season, a similar pattern is seen for contributory causes of death, though it's somewhat weakened, except for pulmonary heart diseases and cardiac arrest. Cardiac arrest exhibits the most striking trend of increasing importance as a contributory disease. Since 2015, cardiac arrest has been coded as a contributory cause of death almost three times more frequently, even though the number of deaths with cardiac arrest started to decline shortly before the onset of the pandemic, which was disrupted by the COVID-19 outbreak.

The pace of change (with 2010 as the baseline year) in the ratio of deaths with immediate CVD did not exhibit a similarly steep increase during the period from 2010 to 2019, except for conditions like intracranial hemorrhage and other IHD. In contrast to contributory causes of death, there were even slight decreases in the importance of immediate CVD in comparison to 2010, particularly in other cerebrovascular diseases. However, this decreasing trend in development reversed since 2020.

Comparing the ratios of the absolute number of deaths is not informative enough due to the mediating role of different age distributions in deaths caused by both the specific cause of death and the year. Therefore, we have extended the analysis with a more suitable measure, the SRMU, which is calculated for each CVD, its type, and the place of death. Figure 3 presents a comparison of the SRMU in 2020 and 2021 to the average ratio during the years 2010–2019 (the baseline). For example, if the bar height, as shown in Fig. 3, is equal to 1.4 for contributory causes of death in 2020, it indicates that SMRU increased by 40% in 2020 compared to its average value in 2010–2019.

Fig. 3.

Ratio of SRMU in 2020 and 2021 to average SMRU in 2010–2019 by cause of death and place of death (horizontal line indicates average in 2010–2019). Note: indicates, that a disease transferred from being primarily UCD to primarily non-UCD during 2020–2021 for a given place of death.

CC = contributory causes of death, IMM = immediate causes of death, ANY = associated causes of death (‘any mention approach’).

Notably, the increases in SRMU are primarily driven by contributory causes of death occurring in hospitals or healthcare facilities such as hospices, nursing homes, and long-term care facilities. However, there is considerable variability among different CVD. For some diseases, like non-rheumatic valve disorders or diseases of arteries, the excess is similar regardless of the place of death. On the other hand, the majority of excess in cerebral infarction and sequelae of cerebrovascular disease arises from deaths in hospitals and, to a lesser extent, in healthcare facilities. In deaths related to cardiac arrest, there have been some increases in deaths with unknown places of death in SRMU.

In the supplementary material, original values of SRMU are provided. Based on these values, diseases that showed the most substantial excess in 2020–2021, resulting in their SRMU surpassing the threshold of 1 (or 2 in the ‘any mention approach’) can be identified. Such shifts, denoted by diseases highlighted with a star in Fig. 3, which signify a transformation from being ‘rather UCD’ to ‘rather non-UCD’, or vice versa. Among contributory causes, diseases that were coded remarkably differently before the pandemic include rheumatic heart diseases, pulmonary heart diseases, non-rheumatic valve disorders, heart failure, and sequelae of cerebrovascular disease. Additionally, the ‘any mention’ approach shows this transition for cerebral infarction and other cerebrovascular diseases. However, this shift from UCD to non-UCD is not universally observed across all places of death. Rheumatic diseases experienced these shifts, particularly in deaths in healthcare facilities, cerebrovascular diseases in hospitals, and heart failure in deaths at home or unknown places.

Discussion

During the COVID-19 pandemic, the most significant excess deaths were observed in contributory diseases. For instance, essential hypertension, hypertensive disease and acute myocardial infarction experienced an increase of more than one-third compared to SARIMA predictions based on trends from 2010 to 2019. The use of single UCD leads to an underestimation of the burden of CVD during the pandemic. There is also variability among diseases in how long their excess of deaths persisted. Moreover, the patterns of recording CVD on death certificates have changed during the pandemic, although the growing importance of contributory CVDs has been observed since 2010. These changes were moderated by place of death and mostly were attributed to death in hospitals and facilities. For instance, changes in recording were attributed to cerebrovascular diseases and rheumatic diseases.

Previously, it was well-established that patients with preexisting CVD conditions were at a higher risk of death from COVID-19. Additionally, it was widely recognized that COVID-19 was predominantly recorded as the UCD in the USA.^1–4,8–10 Some studies provided estimates of excess deaths from underlying CVD, but these estimates exhibited a high degree of variability due to both the methods and data sources employed.^1–4 So far, studies that implemented cause of death data only considered UCD. The reliability of these data, as demonstrated in prior research using examples such as diabetes and dementia, may be compromised, especially during periods when a new leading cause of death emerges. Such occurrences can result in shifts in the recording of causes of death.^10,11

To address the mentioned caveats, our study adopts a multiple-cause perspective to analyze mortality, offering valuable insights into changes in the recording of causes of death. What sets our study apart is its more detailed approach, differentiating between different types of multiple causes of death (immediate, contributory and associated), and its comprehensiveness, not solely focusing on a selected group of CVDs. An added value of our study is that we measure excess deaths separately for 2020 and 2021, which allows us to highlight diseases that experienced a significant increase only during the initial outbreak of the pandemic in the spring of 2020, which could be attributed to the chaos in death certification at that time. In summary, the novelty of the current paper is that it links estimates of excess deaths with changes in the certification of causes of death. Based on our results, we discuss following possible explanation for excess in multiple CVD:

Firstly, it is conceivable that some deaths classified as ‘with’ CVD (mostly as contributing causes) were actually directly caused by them, making them deaths ‘from’ CVD. This is particularly relevant for causes of death that were previously identified as typically UCD before the pandemic. As a result, some of these causes saw a significant rise in their occurrence as multiple causes of death, leading them to be categorized more frequently as non-UCD. Examples of these causes include pulmonary heart diseases, non-rheumatic valve disorders, heart failure, cerebral infraction, sequelae of cerebrovascular disease and other cerebrovascular diseases. When perceiving these causes as underlying, we did not observe a significant excess of mortality.

Another possible explanation for the observed excess mortality associated with CVD is the potential impact of improved causes-of-death reporting during the COVID-19 pandemic, rather than an actual increase in the burden of CVD within the population. While we cannot definitively address this hypothesis based on the current analysis, examining the results stratified by the place of death can provide some insights. Assuming that individuals who died in hospitals, at home, in facilities, or with an unknown place of death are in general comparable in terms of morbidity, any variations in the excess of multiple causes of death based on the place of death can be assigned to differences in the coding practices for multiple causes of death. This ‘false excess’ could potentially be attributed to cerebrovascular diseases, such as cerebral infarction, sequelae of cerebrovascular disease and other cerebrovascular diseases. These conditions experienced a shift from being classified as rather UCD to rather non-UCD primarily due to substantial excesses observed exclusively in hospitals. However, further research is needed to thoroughly explore this hypothesis.

Our study acknowledges several limitations, with one of the most significant being the reliance on the position of causes on death certificates for the classification of multiple causes of death. This approach is dependent on the accuracy and completeness with which physicians report both chains of morbid events leading to death and contributory causes of death. Also, we work with Entity Axis Codes, which are raw cause of death data that might include inconsistencies (redundant codes, diagnoses incompatible with the sex, dual codes, etc.¹⁷). Additionally, two effects need to be considered: (i) consistent underreporting of multiple causes of death prior pandemic, which might artificially inflate the excess death estimates and (ii) replacement coding of COVID-19 as UCD instead of other causes, which might have affected the denominators of SRMU. Further limitation arises from the shortcomings of SARIMA models. Despite selecting the models that demonstrated the best performance, it is important to emphasize that SARIMA models heavily rely on extrapolating observed trends. Several diseases exhibited considerable non-monotonicity in their trend since 2000 (like cardiac arrest as contributory cause of death or cerebral infraction as UCD), which lead, even in the best performing models, to wide confidence intervals suggesting a lot of uncertainty in the models. Therefore, the estimates of excess of death for this particular disease have to be interpreted with caution.

Conclusion

The recording of cardiovascular-related mortality during the COVID-19 pandemic in the USA underwent significant shifts, particularly noticeable in contributory causes of death. However, the various groups of cardiovascular causes of death differed not only in the magnitude of this excess but also in its timing, duration, and specific components affected. Comprehensive analyses, encompassing all causes of death, are essential to fully assess the mortality related to CVD during 2020–2021. Moreover, an explanation for the notable increase in multiple causes of death is essential to comprehend the underlying reasons for the significant rise in cardiovascular-related mortality during COVID-19 in the USA.

Elizaveta Ukolova, PhD Student

Boris Burcin, PhD