Increased emergency cardiovascular events among under-40 population in Israel du...

Cardiovascular adverse outcomes such as blood clotting (e.g., coronary artery thrombosis), acute coronary syndrome, cardiac arrest and myocarditis have been identified as consequences of coronavirus disease 2019 (COVID-19) infection1,2,3,4,5. Similarly, data from regulatory surveillance and self-reporting systems, including the Vaccine Adverse events Reporting System (VAERS) in the United States (US)6, the Yellow Card System in the United Kingdom7 and the EudraVigilance system in Europe8, associate similar cardiovascular side-effects9,10,11,12,13 with a number of COVID-19 vaccines currently in use.

More recently, several studies established probable causal relationship between the messenger RNA (mRNA) vaccines of BNT162b2 and mRNA-127311,14,15,16 as well as adenovirus (ChAdOx1) vaccines17 with myocarditis, primarily in children, young and middle-age adults. The study by the Ministry of Health in Israel, a country with one of the highest vaccination rates in the world, assesses the risk of myocarditis after receiving the 2nd vaccine dose to be between 1 in 3000 to 1 in 6000 in men of age 16–24 and 1 in 120,000 in men under 3011,12,13. A follow up study by the US Center of Disease Control (CDC) based on the VAERS and V-Safe self-reporting systems18 further confirms these findings19. The CDC has recently posted a warning regarding a vaccine-related risk of myocarditis, but still maintained their recommendation to vaccinate young individuals and children over 127. Similar concerns are reflected in the recent Food and Drug Administration approval to the Pfizer vaccine that requires several follow studies on the short and long terms effects of myocarditis in young individuals20.

While the benefits of COVID-19 vaccination are clear, especially for populations at great risk of developing serious and potentially life-threatening illness15,21, it is important to better understand the potential risks to minimize potential harm. However, assessing the connection between myocarditis and other potential cardiovascular conditions, and the COVID-19 vaccines is challenging. First, self-reporting systems22 of adverse events are known to have self-reporting bias and both under and over-reporting problems23,24,25. Even the study from Israel that is based on more proactive data collection mentions that some of the potentially relevant cases were not fully investigated.

Second, myocarditis is a particularly insidious disease with multiple reported manifestations. There is vast literature that highlights asymptomatic cases of myocarditis, which are often underdiagnosed26,27, as well as cases in which myocarditis can possibly be misdiagnosed as acute coronary syndrome (ACS)28,29,30. Moreover, several comprehensive studies demonstrate that myocarditis is a major cause of sudden, unexpected deaths in adults less than 40 years of age, and assess that it is responsible for 12–20% of these deaths26,31,32,33. Thus, it is a plausible concern that increased rates of myocarditis among young people could lead to an increase in other severe cardiovascular adverse events, such as cardiac arrest (CA) and ACS. Anecdotal evidence suggests that this might not be only a theoretical concern16.

Third, myocardial injury and myocarditis is prevalent among patients with COVID-19 infection26,34. As COVID-19 vaccine rollouts often take place with background community COVID-19 infections, it could be challenging to identify whether increased incidence of myocarditis and related cardiovascular conditions, such as CA and ACS, is driven by COVID-19 infections or induced by COVID-19 vaccines. Moreover, such increases may even be caused by other underlying causal mechanisms indirectly related to COVID-19, for example, patients delaying seeking emergent care because of fear of the pandemic and lockdowns35.

This study aims to explore how additional data sources, such as those from emergency medical services (EMS), can complement self-reporting vaccine surveillance systems in identifying COVID-19 related public health trends. More specially, the study examines the association between CA and ACS incidents in the 16–39-year-old population and two potential causal factors: COVID-19 infection rates and COVID-19 vaccine rollout. The study leverages the Israel National EMS (IEMS) data system and analyzes all calls related to CA and ACS events over two and a half years, from January 1st, 2019, throughout June 20th, 2021.

Study design

This retrospective population-based study leverages the IEMS data system and analyzes all calls related to CA and ACS events over two and a half years, from January 1st, 2019, to June 20th, 2021. The IEMS call data are coupled with data on COVID-19 infection rates, as well as the respective vaccination rates over the same period of time.

The study’s time period spans 14 months of a ‘normal period’ prior to the COVID-19 pandemic and vaccine rollout (1/2019–2/2020), about 10 months of a ‘pandemic period’ with two waves of the pandemic (3/2020–12/2020), and about 6 months of a ‘pandemic and vaccination period’ (1/2021–6/2021), during which Israel launched its vaccination rollout parallel to a third wave of the COVID-19 pandemic. Thus, it allows to study how CA and ACS call counts change over time with different background conditions and potentially highlight factors that are associated with the observed temporal changes.

This study was deemed exempt from review by the Massachusetts Institute of Technology Institutional Review Board (E-3300). The study was also approved by the research committee of the IEMS.

Data sources and study population

CA and ACS call data

The IEMS data system includes records of all the calls received through Israel’s national emergency telephone number (1-0-1). Note that the IEMS is a national organization that manages all EMS calls in Israel. Each record contains multiple fields of information, including the retrospective verified call-type as determined by the EMS team (as opposed to the initial call classification), date, relevant response characteristics (e.g., death on scene and whether resuscitation was required during the response), and the patient’s age and gender.

The study’s dataset includes all non-cancelled calls with reported patient age and a verified call-type of either CA or ACS. CA calls were defined as a sudden electrical malfunction of the heart of presumed cardiac or medical etiology, resulting in collapse of a patient, excluding CAs related to trauma, drug overdose, or suicide. ACS calls were defined as conditions where the patients experience a reduction in blood flow to the heart that is associated with myocardial infarction.

The call codes used to identify CA and ACS calls are determined by the EMS teams based on defined protocols of the IEMS. CA diagnosis was made based on the circumstances of collapse as described by the caller to the dispatch team, the CA victim’s electrocardiogram (ECG) as obtained through an automated external defibrillator, and common indicators of CA as observed by the responding paramedics (e.g., patient unresponsiveness, agonal breathing). CAs due or obviously related to trauma, drug overdose, or suicide were excluded in this call code and from the study. ACS diagnosis was made based on the patient's 12-lead ECG (a 12-lead ECG was performed on all patients suspected of ACS to confirm the diagnosis), symptoms (e.g., chest pain, shortness of breath), medical history, and physical examination, as obtained by the responding paramedics. Importantly, these protocol and diagnoses were the same throughout the entire study period (2019–2021), allowing for a consistent comparison between the call counts during the baseline, pandemic, and vaccination periods.

The Supplemental Methods describe the IEMS call data fields and call type codes in further detail.

Vaccination and COVID-19 infection cases

Data on the vaccinations and COVID-19 cases were obtained from the online Israel Government Database Portal (https://info.data.gov.il/datagov/home/). These data include the number of daily administered 1st and 2nd vaccination doses by age group36, as well as the weekly number of new confirmed COVID-19 cases by age group, across all of Israel37. The age groups consist of bins of 20 years starting with 0–19. Population counts by similar age groups were also collected from publicly accessible data used to complement these datasets38. Note that Israel administered only BNT162b2 vaccines for which the lag between the 1st and 2nd dose is three weeks, and that during January–May 2021, the vaccines were administered to individuals of age 16 and over.

Data and statistical analyses

Trends in CA and ACS calls

For each pair of a diagnosis (CA or ACS), age group (16–39, over-40 or all-ages), and gender (male, female, or both genders) the year-to-year absolute and relative changes in calls were calculated. The respective statistical significance of these changes were based on the two-tailed Poisson E-test39. These changes were calculated separately with respect to the full calendar year (2019–2020) and from January 1st to May 31st (2019–2021). January–May time period was used for comparison as it corresponds with the administration of vaccinations among the 16–39 age group in 202136. The full calendar year comparisons were calculated to examine the changes in calls when COVID-19 infections were prevalent, but no vaccinations were administered among the 16–39 age group. Additional analyses describing the percent of CA calls where the patient died on scene (i.e., death declared prior to hospital arrival) and received resuscitation (i.e., patient received defibrillation or cardiopulmonary resuscitation) are outlined in the Appendix.

To visualize the temporal trends of CA and ACS call volume and potential relationship to COVID-19 infection rates and vaccination rates for the 16–39 age group, graphs were created for CA and ACS calls, respectively. Each graph overlays several moving-average time-series over the study period. These include the five-week centered moving-average of the respective weekly EMS call counts, as well as the three-week centered moving-average counts of new COVID-19 infection cases, administered 1st vaccine doses, and administered 2nd vaccine doses. The graphs also indicate the periods of the three national COVID-19-related public health lockdown advisories in Israel40.

To improve the understanding these trends during the third pandemic wave and vaccination rollout, ‘zoom-in’ graphs were similarly created for the time-period October 18th, 2020, through June 20th, 2021. The zoom-in graphs also highlight estimates of the number of individuals who only received one vaccination dose during this time. This was done by plotting an additional time-series of the three-week moving-average of the administered 2nd vaccine doses shifted backwards in time by three weeks. More precisely, the difference between the number of 1st vaccine doses and the number of 2nd vaccine doses shifted backwards in time by three weeks shows the estimated number of patients that only received their 1st dose following Pzifer’s vaccination administration recommendations (i.e., the estimated number of patients who did not received a 2nd vaccine dose after a 3-week period following 1st vaccine dose administration). This difference is also used to estimated number of single doses administered to individuals who had recovered from COVID-19 infections, which was plotted from April 1st, 2021, onwards (April 1st 2021 was shortly after the Israel Ministry of Health approved vaccination for this population41).

Graphs for the above-40 and all-ages groups are shown in Supplemental Figs. 1–4.

Time-series data processing for CA and ACS call, vaccination administration, and COVID-19 infection counts

To check whether the observed year-to-year trends in weekly counts of CA and ACS calls among the 16–39 age group are associated with either COVID-19 infections or vaccine administration, the following weekly time-series were calculated and considered over the entire study period: CA weekly call counts, respectively, for patients in age groups 16–39 and over-40; ACS weekly call counts of patients in age group 16–39; bi-weekly (current and prior week) cumulative counts of 1st and 2nd vaccine doses administered, respectively, in age groups 16–39 and over-40; and cumulative three-week (current and prior two weeks) new COVID-19 infection counts in age groups 16–39 (approximated by age group 0–39) and over-40, respectively. Note that the COVID-19 infection dataset37 only includes aggregated data for the age grouping 0–39 and thus overestimates the number of COVID-19 infections for the age group 16–39.

The choice of bi-weekly counts of 1st and 2nd vaccine doses is motivated by studies that suggest myocarditis typically appears within two weeks from vaccination19. The choice of three-week cumulative counts of new COVID-19 infections is motivated by the fact that acute symptoms of COVID-19 are typically observed within three weeks of infection onset19. Since the impact of COVID-19 might be variable, some of the analysis described below was conducted also with different COVID-19 new infection counts varying the counting period from one to six weeks (i.e., cumulative counts between one, two, three, four, five and six weeks).

Association of year-to-year call count trends with COVID-19 infections and vaccine administration

The Spearman rank correlation was calculated between the time-series of CA weekly call counts for the age group 16–39 and the time-series of the bi-weekly (current and prior week) cumulative counts of 1st and 2nd vaccine doses administered for the same age group. Similarly, the rank correlation was calculated between the time-series of the CA weekly call counts and the time series of the cumulative three-week (current and prior two weeks) new COVID-19 infection counts. The same was calculated for the sum of the time-series of CA and ACS weekly call counts for the 16–39 age group (i.e., correlation with the respective time-series of vaccine dose and new COVID-19 infection counts). As mentioned previously, the bi-weekly and three-week cumulative counts for the vaccinations and COVID-19 infections, respectively, were determined based on prior literature suggesting adverse events occur within those respective durations of time19. A post hoc power analysis was also performed using G*Power (version 3.1.9.7)42 to determine the statistical power (i.e., the probability of rejecting the null hypothesis, concluding an effect is found, and avoiding a Type II error, when an effect truly exists) of the correlation analyses. Finally, since the impact of COVID-19 might occur across a variable period of time, the same analysis was repeated with respect to the time-series of new COVID-19 infections count but varying the cumulative count period from the original three-weeks to a range between one to six weeks.

To further study the potential association between weekly CA and ACS counts, vaccine administration and COVID-19 infections, and control for cross interactions and other factors, two Negative Binomial regression models43 were developed. Negative binomial regression models are commonly used to model count data and allows for the analysis of cases where the outcome variable counts are over-dispersed (variance of the count data is larger than the mean)43,44. Such models can also be designed to use cumulative historical count data as features to estimate outcome counts during a given current time period35,45,46.

The first model, hereinafter referred to as Model 1, regresses the respective time-series of the CA weekly call counts and the ACS weekly call counts in the age group 16–39 (the dependent variable), against the time-series of the bi-weekly cumulative vaccine dose counts and three-week cumulative new COVID-19 infection counts, both in age group 16–39 normalized by the respective population size (independent variables). The model also controls for the different diagnoses (CA versus ACS), for weeks included in periods of national public health lockdown, as well as year-to-year (2019–2020) variations (e.g., due to population growth) in calls through respective dummy variables.

Similarly, the second model, hereinafter referred to as Model 2, regresses the respective time-series of CA weekly counts of age groups 16–39 and over-40 (the dependent variable) against the time-series of the bi-weekly cumulative vaccine dose counts and three-week cumulative new COVID-19 infection counts in the respective age groups, again normalized by the respective population size (independent variables). Additionally, instead of the dummy variable used in Model 1 above to capture the different diagnosis groups, Model 2 introduces a dummy variable to capture the different age groups (16–39 and over-40).

To identify the most statistically significant predictors, the models use bidirectional stepwise feature selection based on the model’s Bayesian information criterion (BIC). The BIC metric summarizes the model’s goodness of fit while penalizing the number of variables selected to avoid overfitting47. During each step of the selection algorithm, features are tested to be added or removed to minimize the model’s BIC. The adjusted incidence rate ratios (IRR) and 95% confidence intervals (CI), representing the estimated change in weekly calls per unit change of each predictor variable, were reported both for the final model after stepwise BIC selection and the full model without variable selection. Model development was performed using R version 4.0.2.

Sensitivity analysis

As robustness check of the associations determined by Models 1 and 2, the analysis was repeated while considering the one to six-week count time-series of new COVID-19 infections in the respective age groups.

Patient and public involvement

The formal involvement of the public and patients was not feasible under the time and resources constraints of this research project. However, this work has been informed by dialogue with those working in healthcare systems and public policy.

Ethical approval

This study was deemed exempt from review by the Massachusetts Institute of Technology Institutional Review Board(E-3300). The study was also approved by the research committee of the IEMS.