Research on the Correlation Between Receiving Pfizer-Biontech

COVID-19 Vaccine and Getting Facial Paralysis

Ziyuan Zhang

The College of Letters & Science, University of California, Berkeley, Berkeley, 94720, U.S.A.

Keywords:

COVID-19, Facial Paralysis, Vaccine, Adverse Events, Predictive Analysis.

Abstract: COVID-19 vaccines are widely administered to reduce the spread of COVID-19 and prevent serious illness

and death after getting infected. Pfizer/BioNTech COVID-19 vaccines are administered the most within the

United States, and facial paralysis is one of the adverse events found after receiving the vaccines. The primary

aim of the paper is to develop a predictive model which uses the features of vaccine receivers to predict

whether facial paralysis would emerge as an adverse effect. Four predictive modes trained by undersampled

and oversampled data from the Vaccine Adverse Event Reporting System (VAERS) are developed using

logistic regression and decision tree from the last quarter of 2021, and features are selected using chi-square

statistics. The four performance metrics (accuracy, recall, precision, and F1 score) indicate the incapability of

models in making decisions, which also implies the irrelevance of selected features. However, the associations

between selected features and the high-level anxiety of the general public have in receiving vaccines under

the pandemic are worth further research and physicians to study and explore for a more comprehensive

understanding of the adverse events of COVID-19 vaccines, manufactured by Pfizer/BioNTech specifically.

This paper finds that four performance metrics indicate these models are not capable of making sensitive

predictions, which implies the irrelevance of the selected features and getting facial paralysis after receiving

Pfizer/BioNTech COVID-19 vaccines. However, the connections between selected features and the huge

amount of cases reported within a year reveal the high level of anxiety that the general public has in receiving

vaccines under the pandemic.

1 INTRODUCTION

Since the December of 2019, the global pandemic

caused by the SARS-CoV-2 virus has brought

devastating impacts to the whole society and

economy. According to the World Health

Organization (WHO), as of 6 January 2022, the

SARS-CoV-2 virus has infected 296,496,809 people

worldwide and caused 5,462,631 deaths (Who

coronavirus (COVID-19) dashboard, 2022). To

protect the general public from getting COVID-19, to

reduce the spread of COVID-19, and to decrease the

severity of sickness after getting infected, CDC

recommends people who are 5 years and older get

vaccinated and remain up to date with their vaccines

(Benefits of getting a COVID-19 vaccine, 2022).

Further, a high level of protection against

Multisystem inflammatory syndrome in persons aged

16-18 years after receiving 2 doses of COVID-19

vaccines, manufactured by Pfizer/BioNTech

specifically (Li, 2020). As of 12 January 2022, there

are 9 COVID-19 vaccines validated for use in

Emergency Use Listing by WHO, including the

Pfizer/BioNTech Comirnaty, the Moderna COVID-

19 vaccine, Sinovac-CoronaVac, etc. (Coronavirus

disease (covid-19): Vaccines, 2022).

Among the approved vaccines, approximately 284

million Pfizer/BioNTech are administered as of

December 15, 2021, which is the most COVID-19

vaccine administered in the United States (Mikulic,

2021). CDC has suggested several possible side

effects of getting the vaccine, including tiredness,

headache, fever, etc. (Pfizer-biontech COVID-19

vaccine overview and Safety, 2022). Facial paralysis

is one of the rare and serious adverse events found

after getting vaccinated according to the Vaccine

Adverse Event Reporting System (VAERS).

Although a huge amount of self-reported adverse

effects can be found in VARES, few predictive

models are created to predict the relationship between

specific features of the patients, such as the current

370

Zhang, Z.

Research on the Correlation between Receiving Pﬁzer-Biontech COVID-19 Vaccine and Getting Facial Paralysis.

DOI: 10.5220/0012021200003633

In Proceedings of the 4th International Conference on Biotechnology and Biomedicine (ICBB 2022), pages 370-376

ISBN: 978-989-758-637-8

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

medication and illness, and the adverse events that

may take place.

In this paper, the relationship between specific

features of the patient who got Pfizer/BioNTech

vaccine and facial paralysis are studied by using the

data from VARES during the last quarter of 2021,

September 1st to December 10th specifically. To help

determine the relationship, I first used a chi-squared

test to select features and then applied logistic

regression and decision tree to train the model

separately, and finally check the accuracy, precision,

recall rates, and F1 score of the model. The result can

be used as a reference for people who intend to get

Pfizer/BioNTech COVID-19 vaccines but are afraid

of experiencing facial paralysis as a specific adverse

event.

2 METHOD

2.1 Data Source

All of the data used in the paper are obtained from the

Vaccine Adverse Event Reporting System (VAERS).

VAERS was created by the U.S. Department of

Health and Human Services (DHHS), the Food and

Drug Administration (FDA), and CDC to receive

reports about adverse events which might be

associated with vaccines, among all age groups, after

the administration of any licensed vaccine in the

United States (Vaccine Adverse Event Reporting

System, 2021).

2.2 Data Collection

Due to the large number of cases associated with

Pfizer/BioNTech COVID-19 vaccines reported,

nearly four hundred and fifty thousand until

December 10th, 2021, and due to technological

limitations and a lack of researchers, data from

September 1st to December 10th are chosen for

analysis.

The three tables provided in VAERS are merged

into one table by using the ID of each case. After

filtering other vaccine names except for “COVID19

(COVID19 (PFIZER-BIONTECH))” and keeping the

cases that have facial paralysis as one of the

symptoms, four columns are kept for features

selection, including “OTHER_MEDS,” “CUR_ILL,”

“HISTORY,” and “SEX.”

“OTHER_MEDS” contains information about

any prescription or non-prescription drugs the

vaccine recipient was taking at the time of

vaccination; “CUR_ILL” contains information about

any current illnesses the vaccine recipient had at the

time of vaccination; and “HISTORY” contains

information about any pre-existing physician-

diagnosed birth defects or medical condition that

existed at the time of vaccination (Vaccine Adverse

Event Reporting System, 2021).

The characteristics of patients are shown in table

Table 1: Characteristics of Patients.

Total (n = 113363)

Age (years), mean (SD) 48.8 (20.8)

Sex, n (%)

Female 74638 (65.8)

Male 37889 (33.4)

Unknown 836 (0.7)

Facial Paralysis, n (%) 237 (0.2)

2.3 Data Pre-Processing

2.3.1 Data Variation

Because most of the cases are self-reported, a large

number of variations in reporting the adverse events

have appeared, including missing information,

misunderstanding of requests, and different formats

of writing. The first five rows of the four selected

columns before feature extraction are presented in

table 2:

Table 2: Selected Columns.

SEX OTHER_MEDS CUR_ILL HISTORY

F pre-natal vitamin, vitamine D supplément, Biotin, Vitamin C

lement.

none. none.

F Unknown None listed none listed

F NaN NaN NaN

F Vitamin D, C, and zinc None NaN

F Multivitamin, cymbalta None Depression

and anxiety

Research on the Correlation between Receiving Pﬁzer-Biontech COVID-19 Vaccine and Getting Facial Paralysis

371

There are two ways in reporting adverse events:

report online and through a PDF form, and the

missing information may be due to the second method,

as the PDF used by VAERS does not have the

function of making people fill in essential blanks,

such as age and gender. Misunderstanding of requests

may because of the relatively complicated reporting

system, which makes unwanted information filled in

the form. For example, the type of tests done by the

person, such as TB tests, are filled in blanks for the

current illness. Different formats of writing generate

the greatest barrier in pre-processing the data, as the

long sentences provided by the vaccine recipients

include different capitalizations between words,

different punctuations or whitespaces in splitting the

items listed, and different names of the same

medicines or symptoms.

2.3.2 Features Selection

To select features from the four columns

“OTHER_MEDS,” “CUR_ILL,” “HISTORY,” and

“SEX,” I replaced all the punctuations with

whitespaces and split the substrings by whitespaces.

Then, the frequency of each substring is collected,

and I picked 26 most frequent features from the four

columns. For each feature selected, a chi-squared

statistic is assigned. The chi-squared statistic is a

commonly used method for feature selection, with the

initial hypothesis H0, which assumes the features and

the class label are unrelated. The greater the value of

chi-squared statistic, the greater evidence against the

hypothesis H0, which means the features and class

label are more related (Chawla, 2002). The chi-

squared formula is shown in equation 1:

𝜒



= 𝛴





(



 



)







(1)

In equation 1, k represents the number of features,

represents the observed count for the group, and

represents the expected count for the group.

The chi-squared statistics for the 26 features with

respect to the target “Facial_Paralysis” is shown in

table 3. Features with the ten highest chi-square

statistics are selected for model training afterwards,

including “Metformin,” “Metoprolol,”

“Atherosclerosis,” “Stroke,” “Hypoglycemia,”

“Hypertension,” “Diabetes,” “Amlodipine,”

“Losartan,” and “Gerd.”

Table 3: Chi-Squared Statistics of Features.

Features Score Features Score Features Score Features Score

Metformin 31.79 Amlodipine 13.80 Migraines 2.08 Sex 0.42

Metoprolol 30.34 Losartan 13.41 Atorvastatin 2.00 Levothyroxine 0.15

Atherosclerosis 29.98 Gerd 10.21 Osteoarthritis 1.22 Depression 0.01

Stroke 25.18 Apnea 9.61 Omeprazole 1.20 Asthma 0.01

Hypoglycemia 24.75 Lisinopril 8.07 Arthritis 1.05 Hypothyroidism 0.01

Hypertension 16.00 Aspirin 3.17 COVID 0.99

Diabetes 15.83 Hyperlipide

mia

2.30 Albuterol 0.60

The ten features were further converted to

numerical data by using one hot encoding, which is a

process in converting categorical features into

indicative variable for better model training.

2.3.3 Oversampling and Undersampling

The total number of cases in the quarter data is

113,363, but only 237 of the cases have facial

paralysis, which makes the data in cohort imbalanced.

Classifiers that are trained on extremely imbalanced

data would be biased towards the majority class,

which is not having facial paralysis in this case. In the

experimenting process, the Logistic Regression

model trained with original data, which was split into

67% training set and 33% testing set, had achieved

nearly 100% accuracy, but the problem was that the

predictive model tried to predict everyone in the

testing set without facial paralysis. For better model

training, both oversampling and undersampling

methods are used separately for training models.

For oversampling, the synthetic minority

oversampling technique (SMOTE) is used. The basic

idea of SMOTE is oversampling the minority class by

ICBB 2022 - International Conference on Biotechnology and Biomedicine

372

taking every minority class sample and applying

synthetic examples along the line segments that join

any or all of the k minority class nearest neighbors

(Bidgoli, 2012). The optimal SMOTE ratios for

different models are found through hyperparameter

tuning, using an Exhaustive Grid Search from Scikit

learn with a three-fold cross validation and F1 score

as the scoring method.

For undersampling, random undersampling is

used. The basic idea of random sampling is randomly

deleting examples in the majority class. I randomly

deleted the majority class till the imbalanced cohort

has 50:50 class ratio, because when random

undersampling the negative class to half of the ratio,

similarities can be found in the average performance

when comparing to the model using the entire big

dataset (Hasanin, 2018).

2.4 Model Training

The datasets after oversampling and undersampling

are split into 67% training set and 33% testing set.

Two machine learning algorithms are selected to train

the predictive model, including logistic regression

(Hosmer, 2013) and decision tree classifier (Song,

2015), using the features with 10 highest chi-squared

statistics as independent variables and facial paralysis

as the dependent target. Therefore, there are four

models in total to predict whether the patient with

given features will have facial paralysis after getting

Pfizer/BioNTech COVID-19 vaccine, including

logistic regression classifier with undersampled data,

logistic regression classifier with oversampled data,

decision tree classifier with undersampled data, and

decision tree classifier with oversampled data.

The criterion for choosing the optimum split in the

decision tree classifier is entropy. Entropy is the

measure of the disorder level of the features selected

with the target, and in this case, entropy is the disorder

level of the ten features with facial paralysis. The

higher the entropy of features, the higher level of

disorder, which makes the optimum split chosen by

the least entropy. The entropy formula is shown in

equation 2:

𝑆= −𝛴



𝑝



𝑙𝑜𝑔



𝑝



(2)

In equation 2, is the proportion of data points in a

node with label C.

Pipelines are created for models with the

oversampling method in the training process. All the

pipelines include oversampling approach, SMOTE,

with sampling strategy obtained by hyperparameter

tuning and the machine learning algorithms used.

2.5 Model Evaluation

The performances of the four predictive models are

evaluated based on the four scores obtained from 10-

fold cross validation, including accuracy, recall,

precision, and F1 score. All four scores are calculated

by using True Positive (TP), True Negative (TN),

False Positive (FP), and False Negative (FN).

Equations of the four metrics are shown:

𝑎𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =





(3)

Equation 3 shows the formula of accuracy, which

represents the proportion of correctly predicted

observations in total observations, which is important

when the dataset has a nearly equal number of False

Positives and False Negatives.

𝑟𝑒𝑐𝑎𝑙𝑙 =





(4)

Equation 4 shows the formula of recall, which

represents the proportion of correctly predicted

positive observations in total actual positive

observations, which is important when the cost of

False Negatives is high.

𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =



 

(5)

Equation 5 shows the formula of precision, which

represents the proportion of correctly predicted

positive observations in total predicted positive

observations, which is important when the cost of

False Positives is high.

𝐹1 𝑠𝑐𝑜𝑟𝑒 =





(6)

Equation 6 shows the formula of F1 score, which

represents the harmonic mean of precision and recall,

which is important when a balance in precision and

recall is needed and the dataset is imbalanced.

Although these four metrics are useful in

evaluating the performance of a model, they focus on

different perspectives as mentioned above. In our

case where the dataset is highly imbalanced, with

only 0.2% positive cases, precision, recall, and F1

score are more preferred than accuracy. Despite False

Positives and False Negatives are both undesirable

outcomes, the consequence brought by False

Negatives in medical prediction is worse because

patients would get the incorrect prediction and be

exposed to the risk in getting facial paralysis after

receiving Pfizer/BioNTech COVID-19 vaccine.

Therefore, the importance rank of the four metrics

would be recall, precision, F1 score, and accuracy.

Further, to avoid an optimistic estimate of the

model performance, the 10-fold cross validation is

Research on the Correlation between Receiving Pﬁzer-Biontech COVID-19 Vaccine and Getting Facial Paralysis

373

performed on the original training and testing sets

(before SMOTE and random undersampling) with the

model trained on resampled data (after SMOTE and

random undersampling). Table (4) provides a

summary of the performance of the four models:

Table 4: Model Performance.

Logistic Regression Decision Tree

SMOTE Random

Undersampling

SMOTE Random

Undersampling

Recall 1.84% 28.00% 3.09% 20.00%

Precision 1.30% 0.44% 2.16% 0.50%

F1 Score 1.36% 0.87% 2.73% 0.98%

Accuracy 99.63% 87.24% 99.46% 91.90%

Comparing between the metrics of the four

models, the Logistic Regression model with data after

undersampling is the best-performing model, with

28.00% for recall, 0.44% for precision, 0.87% for F1

score, and 87.24% for accuracy. However, the

Logistic Regression model trained with data after

oversampling using SMOTE has the worst model

performance, with 1.84% for recall, 1.30% for

precision, 1.36% for F1 score, and 99.63% for

accuracy.

Although the Logistic Regression model trained

with data after random undersampling has the best

performance among the four models, it is still

incapable for predicting whether the person who

received Pfizer/BioNTech COVID-19 vaccines with

given features would get facial paralysis. Apply the

metrics in the best-performing model in context,

among those patients who had facial paralysis, only

28% of them are successfully predicted, and among

those who are predicted to have facial paralysis, only

0.44% of them really suffer from it. Therefore, all of

the four models that are trained by using the ten

highest chi-squared statistics features are incapable in

making predictions, which further reveals the fact that

the features selected are relatively irrelevant to the

target class facial paralysis under the condition of

receiving Pfizer/BioNTech COVID-19 vaccines.

3 DISCUSSION

Given the broadly administration of COVID-19

vaccines to combat with the global pandemic

COVID-19, a notable number of adverse events has

emerged. Facial paralysis is one of the adverse events

occurred after receiving Pfizer/BioNTech COVID-19

vaccines according to the Vaccine Adverse Event

Reporting System (VAERS). Using Logistic

Regression and Decision Tree Classifier, I developed

four models to predict whether the person might get

facial paralysis after receiving Pfizer/BioNTech

COVID-19 vaccines given the ten features, including

“Metformin,” “Metoprolol,” “Atherosclerosis,”

“Stroke,” “Hypoglycemia,” “Hypertension,”

“Diabetes,” “Amlodipine,” “Losartan,” and “Gerd.”

The 26 features are selected based on the appearing

frequencies in the dataset among four columns

(“OTHER_MEDS,” “CUR_ILL,” “HISTORY,” and

“SEX”) in the merging dataset tables at first. Further,

a chi-squared statistic is assigned to each feature, and

the ten features with the highest chi-square statistics

are kept for final model training. Due to the imbalance

proportion of positive and negative cases in the

original dataset, random undersampling and synthetic

minority oversampling technique are applied to the

dataset in order to train the model unbiasedly and

improve the overall performance. The performance of

models is evaluated through a 10-fold cross validation

on four metrics, including recall, precision, F1 score,

and accuracy.

The Logistic Regression model trained with

random undersampling data has the best performance

among the four models built, according to the

performance metrics. Recall is the most determining

performance metric as the cost of False Positive is

high in medical cases, whereas the best model only

achieves 28% for recall, which indicates all the

models are not sensitive enough do the prediction.

This also implies the features selected are not relevant

enough to getting facial paralysis as the result of

receiving Pfizer/BioNTech COVID-19 vaccines.

Although the 10 features are not predictive or

causal to facial paralysis as one of the adverse events

after receiving Pfizer/BioNTech COVID-19 vaccines,

the relations between features might worth to be

researched by further studies or considered by

physicians. Within the ten features, 6 of them are

illnesses and the rest 4 are medications, where strong

relationships could be found.

ICBB 2022 - International Conference on Biotechnology and Biomedicine

374

Feature “Stroke” could cause oro-facial

impairment directly, which can be described as facial

paralysis (Schimmel, 2017). Further, features

“Diabetes” and “Hypertension” are the two main

systemic comorbidities associated with Bell’s Palsy,

another form of facial paralysis, and hypertension is

also a major modifiable contributor to stroke

(Mancini, 2019; Buonacera, 2019). Feature

“Atherosclerosis” has an ischemic stroke as one of its

major clinical manifestations (Herrington, 2016).

Feature “Hypoglycemia” is one of the essential issues

for diabetic patients, and repeated hypoglycemia

could rise the risk of cardiovascular diseases

(Tourkmani, 2018).

From the medication features’ perspectives,

feature “Metformin” is widely used to treat type 2

diabetes at early stages (Lv, 2020). Features

“Metoprolol,” “Amlodipine,” and “Losartan” are

used to reduce morality in patients with hypertension

and coronary heart diseases, which might further

reduce the risk of hypertensive stroke (Kwon, 2013;

Pareek, 2010).

The relations between the ten selected features are

presented graphically in Figure 1:

Figure 1: Relations of Features.

Therefore, the weak-performing models and the

direct relations between selected features and facial

paralysis indicate this specific adverse event reported

in VAERS might because of the pre-existing

conditions and current medications intake of patients

but not because of receiving Pfizer/BioNTech

COVID-19 vaccines. However, further studies with

sufficient time and funds are needed to utilize the

whole dataset in order to reach a more comprehensive

conclusion.

Furthermore, nearly 450 thousand adverse events

associated with Pfizer/BioNTech COVID-19

vaccines were reported within a year, beginning

December 15th, 2020, and ending December 10th,

2021. Such a huge amount of reported cases for a

single brand of vaccine in a relatively short period of

time could imply the anxiety of people after receiving

vaccines during a pandemic. It is worth noting that

there is a bidirectional temporal association between

facial paralysis and anxiety (Tseng, 2017). Therefore,

the prevalent anxiety could be another risk factor for

the facial paralysis cases, and further studies need to

explore the correlation between anxiety and facial

paralysis under the condition of receiving

Pfizer/BioNTech COVID-19 vaccines.

4 CONCLUSION

The primary goal of this study is to develop a model

to predict whether the Pfizer/BioNTech COVID-19

vaccines receiver would get facial paralysis as an

adverse event based on the data from VAERS. The

self-report system of VAERS where everyone is able

to submit an adverse event case makes the data

incomprehensive and easily biased, and the feature

extraction process during data analysis might miss

information by breaking the completeness of

sentences. Another major limitation is the partial data

involved, as only the last quarter of the records are

used due to huge amount of data involved and limited

research resources. Further explorations of the

association between facial paralysis and receiving

Pfizer/BioNTech COVID-19 vaccines with more

comprehensive data and more cost-sensitive models

are expected. Although four predictive models are

developed using features with the highest chi-squared

statistics, four performance metrics indicate these

models are not capable in making sensitive

predictions, which implies the irrelevance of the

selected features and getting facial paralysis after

Research on the Correlation between Receiving Pﬁzer-Biontech COVID-19 Vaccine and Getting Facial Paralysis

375

receiving Pfizer/BioNTech COVID-19 vaccines.

However, the connections between selected features

and the huge amount of cases reported within a year

reveal the high level of anxiety that general public has

in receiving vaccines under the pandemic. Further

studies are needed to investigate the association

between anxiety and other diseases before and after

receiving COVID-19 vaccines. Because getting

vaccinated could reduce the spread of COVID-19 and

prevent serious illness and death after getting infected

(Stay up to date with your vaccines, 2022), the

general public should be positive towards COVID-19

vaccines and be confident in the fight against

COVID-19.

REFERENCES

Bidgoli, A. M., & Parsa, M. N. (2012) A hybrid feature

selection by resampling, chi squared and consistency

evaluation techniques. World Academy of Science,

Engineering and Technology, 68: 276-285.

Buonacera, A., Stancanelli, B., & Malatino, L. (2019)

Stroke and hypertension: An appraisal from

pathophysiology to clinical practice. Current Vascular

Pharmacology, 17: 72–84.

Centers for Disease Control and Prevention. (2022) Stay up

to date with your vaccines.

https://www.cdc.gov/coronavirus/2019-

ncov/vaccines/stay-up-to-date.html.

Centers for Disease Control and Prevention. (2022) Pfizer-

biontech COVID-19 vaccine overview and Safety.

https://www.cdc.gov/coronavirus/2019-

ncov/vaccines/different-vaccines/Pfizer-

BioNTech.html.

Centers for Disease Control and Prevention. (2022)

Benefits of getting a COVID-19 vaccine.

https://www.cdc.gov/coronavirus/2019-

ncov/vaccines/vaccine-benefits.html.

Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer,

W. P. (2002) Smote: Synthetic minority over-sampling

technique. Journal of Artificial Intelligence Research,

16: 321–357.

Hasanin, T., & Khoshgoftaar, T. (2018) The effects of

random undersampling with simulated class imbalance

for Big Data. 2018 IEEE International Conference on

Information Reuse and Integration (IRI), 70–79.

Hosmer, D. W., Lemeshow, S., & Sturdivant, R. X. (2013)

Applied Logistic Regression. Wiley-Blackwell,

Hoboken.

Herrington, W., Lacey, B., Sherliker, P., Armitage, J., &

Lewington, S. (2016) Epidemiology of atherosclerosis

and the potential to reduce the global burden of

atherothrombotic disease. Circulation Research, 118:

535–546.

Kwon, H.-M., Shin, J.-W., Lim, J.-S., Hong, Y.-H., Lee, Y.-

S., & Nam, H. (2013) Comparison of the effects of

Amlodipine and losartan on blood pressure and diurnal

variation in hypertensive stroke patients: A prospective,

randomized, double-blind, comparative parallel study.

Clinical Therapeutics, 35: 1975–1982.

Li, G. Y., Zhang, R. Y., Pang, M. F., Liang, Z. R., Yang, X.

P., Wu, J. W., Li, Z. J., Liu, G., Song, R., Ding, J., Wang,

Q., Qi, X. P., & Qian, S. Y. (2020) Multisystem

inflammatory syndrome in children: its current

situation and potential direction in prevention and

treatment. Zhonghua er ke za zhi = Chinese journal of

pediatrics, 58: 780–783.

Lv, Z., & Guo, Y. (2020) Metformin and its benefits for

various diseases. Frontiers in Endocrinology, 11.

Mancini, P., Bottaro, V., Capitani, F., De Soccio, G.,

Prosperini, L., Restaino, P., De Vincentiis, M., Greco,

A., Bertoli, G. A., & De Seta, D. (2019). Recurrent

bell’s palsy: Outcomes and correlation with clinical

comorbidities. Acta Otorhinolaryngologica Italica, 39:

316–321.

Mikulic, M. (2021) Covid-19 vaccinations administered

number by manufacturer U.S. 2021.

https://www.statista.com/statistics/1198516/covid-19-

vaccinations-administered-us-by-company/.

Pareek, A., Chandurkar, N. B., Sharma, R., Tiwari, D., &

Gupta, B. S. (2010) Efficacy and tolerability of a fixed-

dose combination of metoprolol extended

release/amlodipine in patients with mild-to-moderate

hypertension. Clinical Drug Investigation, 30: 123–131.

Song, Y. Y., & Lu, Y. (2015) Decision tree methods:

applications for classification and prediction. Shanghai

archives of psychiatry, 27: 130–135.

Schimmel, M., Ono, T., Lam, O. L., & Müller, F. (2017)

Oro-facial impairment in stroke patients. Journal of

Oral Rehabilitation, 44: 313–326.

Tourkmani, A. M., Alharbi, T. J., Rsheed, A. M.,

AlRasheed, A. N., AlBattal, S. M., Abdelhay, O.,

Hassali, M. A., Alrasheedy, A. A., Al Harbi, N. G., &

Alqahtani, A. (2018) Hypoglycemia in type 2 diabetes

mellitus patients: A review article. Diabetes &

Metabolic Syndrome: Clinical Research & Reviews, 12:

791–794.

Tseng, C.-C., Hu, L.-Y., Liu, M.-E., Yang, A. C., Shen, C.-

C., & Tsai, S.-J. (2017) Bidirectional association

between Bell's Palsy and anxiety disorders: A

nationwide population-based retrospective cohort study.

Journal of Affective Disorders, 215: 269–273.

Vaccine Adverse Event Reporting System. (2021) VAERS

Data Use Guide.

https://vaers.hhs.gov/docs/VAERSDataUseGuide_en_

September2021.pdf.

World Health Organization. (2022) Who coronavirus

(COVID-19) dashboard. https://covid19.who.int/.

World Health Organization. (2022) Coronavirus disease

(covid-19): Vaccines. https://www.who.int/news-

room/questions-and-answers/item/coronavirus-

disease-(covid-19)-

vaccines?gclid=CjwKCAiA0KmPBhBqEiwAJqKK43

wip-_DlRhk3b

WvvO3vuUVwrK4Qv1vufe2STXNnGTwE62MaISm

BoCydkQAvD_BwE& topicsurvey=v8kj13%29.

ICBB 2022 - International Conference on Biotechnology and Biomedicine

376