Propensity-Score-Matched Evaluation of Adverse Events Affecting Recovery after COVID-19 Vaccination: On Adenovirus and mRNA Vaccines

This study aimed to observe adverse events following immunisation (AEFIs) that affected recovery within two weeks after COVID-19 vaccination and investigate their risks in propensity-score-matched populations. Data were collected from 447,346 reports from the VAERS between 1 January 2021 and 31 July 2021. Propensity-score-matched populations were constructed by adjusting for demographic characteristics and 11 underlying diseases in eligible subjects who received 1 of 3 COVID-19 vaccines: 19,462 Ad26.COV2.S, 120,580 mRNA-1273, and 100,752 BNT162b2. We observed that 88 suspected AEFIs (22 in Ad26.COV2.S, 62 in mRNA-1273, and 54 in BNT162b2) were associated with an increased risk of delayed recovery within 2 weeks after COVID-19 vaccinations. Nervous system, musculoskeletal and connective tissue, gastrointestinal, skin, and subcutaneous tissue disorders were the most common AEFIs after COVID-19 vaccination. Interestingly, four local and systemic reactions affected recovery in different vaccine recipients during our study period: asthenic conditions and febrile disorders in Ad26.COV2.S and mRNA-1273; general signs and symptoms in mRNA-1273 and BNT162b2; injection site reactions in Ad26.COV2.S and BNT162b2. Although it is necessary to confirm a causal relationship with COVID-19 vaccinations, some symptoms, including paralysis, allergic disorders, breathing abnormalities, and visual impairment, may hinder the recovery of these recipients.


Introduction
At the end of 2019, a novel coronavirus, which is now known as severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), was identified as the cause of a cluster of pneumonia cases in Wuhan, a city in the Hubei Province of China [1]. The rapidly expanding COVID-19 pandemic has impacted all areas of daily life. As of 3 June 2021, COVID-19 outbreaks were reported in 213 countries, with over 171 million confirmed cases and more than 3.68 million deaths [2].
After COVID-19 vaccination, local reactions, such as pain or redness at the injection site, were mild. Moderate systemic reactions, such as fatigue, myalgia, arthralgia, and headache, were observed in less than 50% of mRNA vaccine recipients, starting about 15 h As illustrated in Figure 1, the three files used a common VAERS ID as the primary key. To extract all cases of COVID-19 vaccination, the VAERS IDs were initially screened and used to extract some medical information, such as demographics, prediagnosed illness or medical histories, and AEs for individuals who were vaccinated against COVID- 19. In particular, AEs were defined using standard medical terms in The Medical Dictionary for Regulatory Activities (MedDRA) [32]. Each case had a maximum of five terms, with each term listed for the corresponding VAERS ID.

Standard Medical Terms for Vaccine Adverse Events
MedDRA is organised in a five-level hierarchy. The highest or broadest level is the system organ class (SOC), which is further divided into high-level group terms (HLGT), high-level terms (HLT), preferred terms (PT), and the most granular lowest-level terms (LLT); the hierarchy includes 27, 337, 1737, 24,820, and 83,291 terms in release version 24.0 (March 2021) [33], respectively.
In cases reported to the VAERS database, various AEFIs were provided at the PT level. However, PTs are highly fragmented into signs, symptoms, diagnosis, investigation, or medical procedures, which might lead to failures in identifying differences in the AE incidence [34,35]. Accordingly, we converted the standard terms associated with all AEFIs (expressed as PTs) into HLTs. We collected 1122 COVID-19-vaccine-induced AEs and analysed their relationships at the SOC level. Each AE encoded a binary value indicating whether the COVID-19 vaccine recipient had certain adverse symptoms, where 1 indicated YES and 0 indicated NO.

Standard Medical Terms for Vaccine Adverse Events
MedDRA is organised in a five-level hierarchy. The highest or broadest level is the system organ class (SOC), which is further divided into high-level group terms (HLGT), high-level terms (HLT), preferred terms (PT), and the most granular lowest-level terms (LLT); the hierarchy includes 27, 337, 1737, 24,820, and 83,291 terms in release version 24.0 (March 2021) [33], respectively.
In cases reported to the VAERS database, various AEFIs were provided at the PT level. However, PTs are highly fragmented into signs, symptoms, diagnosis, investigation, or medical procedures, which might lead to failures in identifying differences in the AE incidence [34,35]. Accordingly, we converted the standard terms associated with all AEFIs (expressed as PTs) into HLTs. We collected 1122 COVID-19-vaccine-induced AEs and analysed their relationships at the SOC level. Each AE encoded a binary value indicating whether the COVID-19 vaccine recipient had certain adverse symptoms, where 1 indicated YES and 0 indicated NO.

Study Design
The study population included 447,346 cases: 45,848 for Janssen vaccination, 197,006 for Moderna vaccination, 203,478 for Pfizer-BioNTech vaccination, and 1014 for unknown. After excluding unknown COVID-19 vaccinations, 431,341 individuals with a unique VAERS ID were screened. The primary endpoint was all adverse symptoms reported within two weeks after COVID-19 vaccination. Accordingly, we considered 256,994 individuals as eligible subjects according to the following exclusion criteria: 2 or more different COVID-19 vaccinations; missing age data or under 20 years of age; missing data on the number of days associated with vaccine-related adverse events; cases reported after two weeks; or unreported recovery from vaccine-induced AEs. Study subjects were categorised into 7 and 3 groups according to age and gender, respectively: (1) 20-29, 30-39, 40-49, 50-59, 60-69,  70-79, and ≥80 years; (2) unknown, male, and female. After propensity score matching, 19,462, 120,580, and 100,752 individuals were included in the analysis, as illustrated in Figure 2. This workflow shows the numbers of individuals (N) excluded at different stages and the number of cases before/after propensity score matching.
The study population included 447,346 cases: 45,848 for Janssen vaccination, 197,006 for Moderna vaccination, 203,478 for Pfizer-BioNTech vaccination, and 1014 for unknown. After excluding unknown COVID-19 vaccinations, 431,341 individuals with a unique VAERS ID were screened. The primary endpoint was all adverse symptoms reported within two weeks after COVID-19 vaccination. Accordingly, we considered 256,994 individuals as eligible subjects according to the following exclusion criteria: 2 or more different COVID-19 vaccinations; missing age data or under 20 years of age; missing data on the number of days associated with vaccine-related adverse events; cases reported after two weeks; or unreported recovery from vaccine-induced AEs. Study subjects were categorised into 7 and 3 groups according to age and gender, respectively: (1) 20-29, 30-39,  40-49, 50-59, 60-69, 70-79, and 80 years; (2) unknown, male, and female. After propensity score matching, 19,462, 120,580, and 100,752 individuals were included in the analysis, as illustrated in Figure 2. This workflow shows the numbers of individuals (N) excluded at different stages and the number of cases before/after propensity score matching.

Propensity Score Matching
Data pertaining to our study population collected from the VAERS database includes various age groups, genders, and various underlying diseases for each individual at the time of COVID-19 vaccination. The type of vaccine used might depend on individual characteristics, including age, sex, and comorbidities, and this choice could affect the vaccineassociated AEs and recovery from them. Therefore, we used propensity score matching (PSM) using the nearest neighbour method with a 1:1 matching ratio without replacement, to reduce the bias due to these confounding variables between two groups (i.e., recovery and no recovery from COVID-19-vaccine-induced AEs). The propensity score, specifically the conditional probability of not recovering from COVID-19-vaccine-induced AEs, given the observed covariates (i.e., in our study-age, sex, and 11 underlying diseases), was estimated through binomial logistic regression analysis. To rigorously control for confounding variables, the calliper width (i.e., considered only if the difference in the propensity score between paired subjects was within a prescribed range) was used during the PSM. The calliper threshold was examined between 0.25 and 0.1 of the standard deviation of the propensity score. A chi-square test was used as a criterion to determine the best calliper threshold, at which all covariates had no statistically significant difference (p > 0.05). The absolute standardised difference (ASD) [45] was set below 0.1 to confirm whether the two groups were well balanced.  Data pertaining to our study population collected from the VAERS database includes various age groups, genders, and various underlying diseases for each individual at the time of COVID-19 vaccination. The type of vaccine used might depend on individual characteristics, including age, sex, and comorbidities, and this choice could affect the vaccine-associated AEs and recovery from them. Therefore, we used propensity score matching (PSM) using the nearest neighbour method with a 1:1 matching ratio without replacement, to reduce the bias due to these confounding variables between two groups (i.e., recovery and no recovery from COVID-19-vaccine-induced AEs). The propensity score, specifically the conditional probability of not recovering from COVID-19-vaccineinduced AEs, given the observed covariates (i.e., in our study-age, sex, and 11 underlying diseases), was estimated through binomial logistic regression analysis. To rigorously control for confounding variables, the calliper width (i.e., considered only if the difference in the propensity score between paired subjects was within a prescribed range) was used during the PSM. The calliper threshold was examined between 0.25 and 0.1 of the standard deviation of the propensity score. A chi-square test was used as a criterion to determine the best calliper threshold, at which all covariates had no statistically significant difference (p > 0.05). The absolute standardised difference (ASD) [45] was set below 0.1 to confirm whether the two groups were well balanced.

Cox Proportional Hazards Regression Analysis
We considered 1122 COVID-19-vaccine-associated AEs (i.e., 753, 1019, and 1000 Med-DRA HLTs for Ad26.COV2.S, mRNA-1273, and BNT 162b2, respectively) as independent variables, reported within 2 weeks from individuals being vaccinated with 1 type of the 3 vaccines. We examined the association between those AEs and recovery using a Cox proportional hazards regression analysis. The event of interest was defined as individuals who did not recover from the vaccine-induced AEs during our study period. We removed some AEs with a small sample size to resolve any monotonic likelihood issues that could occur in Cox proportional hazards regression analysis. These issues are observed when fitting a Cox model when at least one covariate estimate diverges to negative or positive infinity [46]. Accordingly, we selected AEs including the minimum number of cases with more than five adverse symptoms within the event (i.e., no recovery group from vaccineinduced AEs). Based on this criterion, we performed univariate Cox proportional hazards regression analysis to screen for possible AEs that delayed recovery after COVID-19 vaccination. Statistically significant AEs were included in the multivariate Cox regression model for adjustment. The results of the Cox regression model are presented as HRs with 95% confidence intervals. Statistical significance was set at p < 0.05. The concordance index or C-index was used to identify the discrimination power of the multivariate Cox regression model.

Experimental Environment and Implementation
All experiments were implemented and evaluated using the following hardware and software: AMD Ryzen 9 3900X 12-Core Processor
In the propensity-score-matched population, 91 adverse symptoms were associated with the risk of developing an AE estimated with a univariate Cox proportional hazard regression model. The concordance index of a multivariate Cox regression model was then 0.6518. Overall, 36 AEs were associated with the Janssen COVID-19 vaccination (see Figure S1). Of these, 22 were major vaccine-induced AEs with extended recovery times. The adjusted HRs were distributed from 1.06 to 2.6 and grouped into 11 SOC terms ( Table  2).
In the propensity-score-matched population, 91 adverse symptoms were associated with the risk of developing an AE estimated with a univariate Cox proportional hazard regression model. The concordance index of a multivariate Cox regression model was then 0.6518. Overall, 36 AEs were associated with the Janssen COVID-19 vaccination (see Figure S1). Of these, 22 were major vaccine-induced AEs with extended recovery times. The adjusted HRs were distributed from 1.06 to 2.6 and grouped into 11 SOC terms ( Table 2).
Among the SOCs, five were common AEFIs, such as nervous system disorders (38.73%), general disorders and administration site conditions (29.05%), musculoskeletal and connective tissue disorders (23.71%), gastrointestinal disorders (20.18%), and skin and subcutaneous tissue disorders (15.26%). In particular, some AEs-headaches (  Note: Non-Recovered Group-includes individuals who had not recovered from Janssen COVID-19-vaccineinduced AEs. For 2 weeks, no information on recovery was an exclusion criterion. NEC-not elsewhere classified-a standard abbreviation used to denote miscellaneous terms that do not readily fit into other hierarchical classifications for a particular SOC. The NEC designation is used only with HLTs and HLGTs for grouping purposes. Total-denotes the number of cases for one or more adverse symptoms belonging to each SOC. HR-hazard ratios. Table S2 shows the distribution of demographic characteristics and 11 underlying disabilities between individuals vaccinated with the Moderna vaccine in the non-recovered and recovered groups before and after PSM. Before PSM, there were significant differences (p < 0.05) in age and sex. Nine underlying diseases, namely, hypertension (10.16%), diabetes (5.79%), asthma (5.38%), allergies (2.81%), obesity (1.77%), cancers (1.47%), pulmonary disease (1.23%), heart failure (0.95%), and stroke (0.25%), were relatively high in the nonrecovered group, with statistically significant differences compared with the recovered group (p < 0.05). After PSM, two groups were paired with a 1:1 matching ratio, the calliper width was adjusted to 0.15, and the differences between these groups disappeared (ASD < 0.1, p > 0.05).

Pfizer-BioNTech COVID-19 Vaccine
In the propensity-score-matched population, we observed 208 adverse symptoms associated with the risk of developing an AE estimated with a univariate Cox proportional hazard regression model. The concordance index of a multivariate Cox regression model was 0.6237. Overall, 92 AEs were associated with Pfizer-BioNTech COVID-19 vaccination (see Figure S3). Of these, 54 were the major vaccine-induced AEs; their hazard ratios ranged from 1.04 to 3.37 and were grouped into 16 SOC terms (Table 4).  Note: Non-Recovered Group-includes individuals who had not recovered from Pfizer-BioNTech COVID-19-vaccine-induced AEs. For 2 weeks, no information on recovery was an exclusion criterion. NEC-not elsewhere classified-a standard abbreviation used to denote miscellaneous terms that do not readily fit into other hierarchical classifications for a particular SOC. The NEC designation is used only with HLTs and HLGTs for grouping purposes. Total-denotes the number of cases for one or more adverse symptoms belonging to each SOC. HR-hazard ratios.

Discussion
Our study focused on analysing major AEFIs and whether their risks affect the recovery after COVID-19 vaccination. We investigated all possible AEs of three COVID-19 vaccines based on data reported to the VAERS database from 1 January 2021 to 31 July 2021 and analysed their relationships for different SOCs.
Rare neurological findings, especially facial paralysis, were noted in several previous studies [7,[13][14][15]. A total of 7 cases suspected of facial paralysis were reported among 36,930 recipients during clinical phase 3 trials for mRNA vaccines [6,7]. A recent study demonstrated that neurological AEs after COVID-19 vaccination may be more than chance events, while reporting 18 cases of facial paralysis and other neurological events, such as syncope (37 cases) and seizure (12 cases), to the VAERS database in December 2020 [15]. Similar to the previous study, we identified that 200 recipients with suspected paralysis and paresis (excluding cranial nerve) did not recover after mRNA-1273 vaccination, even though they did not develop facial paralysis or Bell's palsy. However, they showed a 1.17-fold increased risk compared with the recovered group. Seizure-related symptoms were observed in 169 and 222 non-recovered cases among recipients who received mRNA-1273 or BNT162b2, respectively; however, their risk degree was lower than that in the recovered group (See Figures S2 and S3).
Severe allergic reactions, including anaphylaxis, were very rarely reported as a crucial issue [30,31]. Anaphylaxis is diagnosed primarily based on clinical symptoms, signs, and a detailed description of the acute episode, including antecedent activities and events occurring within the preceding minutes-hours [49]. We did not find adverse cases and the risk of anaphylactic and anaphylactoid responses in the propensity-score-matched populations associated with mRNA vaccination. However, allergic conditions related to immune system disorders, including hypersensitivity reactions (type I, III, or IV), were adverse symptoms associated with adenovirus vector and mRNA vaccines. We identified that during our study period, 207, 751, and 716 cases were non-recovered after Ad26.COV2.S, mRNA-1273, and BNT162b2 vaccinations, respectively. Other adverse symptoms, such as angio-oedemas, yielded 1364 (2.26%) non-recovered cases after mRNA-1273 vaccination.
Our study has several limitations. First, unlike clinical trials that report adverse reactions using standardised data collection procedures, the VAERS database includes various adverse cases [15] that may include reporting bias, over-reporting, or underreporting of AEs [17], because clinicians are encouraged to report all AEFIs. Therefore, the actual mild-moderate or severe AEs may be rarer than our findings. Second, our study data may include individuals who received the SARS-CoV-2 test before our analytic period and exhibited some symptoms of SARS-CoV-2 infection. To reduce such potential bias, we attempted to demonstrate possible major AEFIs concerning each COVID-19 vaccine in the propensity-score-matched populations adjusted for age, sex, and 11 underlying diseases. However, unmeasured and residual confounding may have biased our estimates. For example, we did not know the individual occupation characteristics (e.g., virus exposure degree and protective equipment use) [54], race, and residence. Consequently, our study findings may involve unknown or unverified AEs, since the VAERS database does not provide medically confirmed or valid diagnostic evidence. Third, to investigate AEFIs related to COVID-19 vaccines, we used only limited covariates that satisfied a user-defined sample threshold in propensity-score-matched target outcomes. Even though it is possible to reduce the monotonic likelihood issue (i.e., the bias of maximum likelihood estimates) in the Cox regression model, further investigations will be necessary.

Conclusions
We found that an increased risk of local and systemic reactions (e.g., headache, jointrelated symptoms, muscle pain, musculoskeletal and connective tissue pain, nausea or vomiting symptoms, dermal and epidermal conditions, and febrile disorders) was associated with delayed recovery in non-recovered cases of different propensity-score-matched populations when compared with the group that recovered after COVID-19 vaccination. Furthermore, we observed that some notable AEs, such as paralysis (excluding cranial nerve), allergic disorder, breathing abnormality, and visual impairment, might hinder the recovery of non-recovery cases; although, it is necessary to confirm a causal relationship with COVID-19 vaccination. In the future, we aim to perform an observational study on the effectiveness and side effects of heterologous prime-boost, particularly in preventing severe disease and infection with Delta (B.1.617.2) and Omicron (B.1.1.529) variants.
Supplementary Materials: The following supporting information can be downloaded at: https://www. mdpi.com/article/10.3390/vaccines10020284/s1. Table S1: Baseline characteristics of individuals vaccinated with the Janssen COVID-19 vaccine before and after PSM, Table S2: Baseline characteristics of individuals vaccinated with the Moderna COVID-19 vaccine before and after PSM, Table S3: Baseline characteristics of individuals vaccinated with the Pfizer-BioNTech COVID-19 vaccine before and after PSM, Figure S1: AE hazard ratios after Janssen COVID-19 vaccination estimated from a multivariate Cox model, Figure S2: AE hazard ratios after Moderna COVID-19 vaccination estimated from a multivariate Cox model, Figure S3: AE hazard ratios after Pfizer-BioNTech COVID-19 vaccination estimated from a multivariate Cox model.

Data Availability Statement:
The raw data used in the current study are available in the VAERS (https://vaers.hhs.gov/data/datasets.html, accessed on: 6 August 2021). All data generated in this study are available on request from the corresponding author.

Conflicts of Interest:
The authors declare no conflict of interest.