Rationale and Background

On April 6, 2022, members of the United States (US) Vaccines and Related Biological
Products Advisory Committee (VRBPAC) of the Food and Drug Administration (FDA) voted for
the first time to update COVID-19 vaccine strain compositions from the original wildtype
formulation to a formulation that better matched currently circulating variants. In a
commentary in the Journal of the American Medical Association (JAMA) Network Open one
month later, Dr Peter Marks (director of the Center for Biologics Evaluation and
Research), Dr Janet Woodcock (principal deputy FDA commissioner) and Dr Robert Califf
(FDA commissioner) argued

"Better alignment between the variant(s) covered by the vaccine and circulating
variant(s) of SARS-CoV-2 might be expected to prevent a broader spectrum of disease,
potentially for a longer time".

The US Centers for Disease Control and Prevention (CDC)'s Advisory Committee on
Immunization Practices (ACIP) voted in each of 2022, 2023 and 2024 to recommend updated
COVID-19 vaccines as authorized or approved by FDA in persons ≥ 6 months. There is
therefore a need to evaluate effectiveness of these updated products on an annual basis,
against a variety of health outcomes.

Research Question and Objectives

The research question for this study is "What is the real-world effectiveness of BNT162b2
formulations?"

Primary objective:

1. To evaluate vaccine effectiveness (VE) of BNT162b2 formulations in
non-immunocompromised adults (age 18+) against COVID-19 related endpoints, by age,
comorbid conditions of interest and adapted vaccine formulation.

Secondary objective:

2. To measure COVID-19 VE of BNT162b2 formulations in pediatrics (age < 18) against
COVID-19 related endpoints, by age and adapted vaccine formulation.

Research Methods

Study Design

For all aims of this study, vaccination status will be measured as a time-varying
exposure. Exposed person-time begins 14 days after receipt of a BNT162b2
formulation. Unexposed person-time is unvaccinated person-time, as well as 0-13 days
after vaccination. The unexposed group will be further stratified by previous
status.

Data from one year prior to updated formulation availability will be used as the
look-back period to define patient's characteristics, clinical history, risk
factors, and healthcare utilization. All available data prior to this period may be
used as well.

Setting

The youngest possible age for any person in any aim of the study is 6 months old.

This protocol will be conducted among persons who have all of the following:

- At least one year of continuous medical enrollment prior to index date in an
insurance plan that contributes to the HealthVerity claims database in the
United States. Medical claims are used to measure inpatient, outpatient,
emergency department and urgent care encounters. A gap of up to 30 days will be
allowed.

- At least one year of continuous pharmacy enrollment prior to index date in an
insurance plan that contributes to the HealthVerity claims database in the
United States. Pharmacy claims are used to measure outpatient dispensations of
drugs, vaccines, and other biological products. A gap of up to 30 days will be
allowed.

- At least one year of continuous state residency in a state with mandatory
COVID-19 vaccination reporting that contributes the entire state vaccine
registry to HealthVerity.

- Does not have discrepancies in sex and/or year of birth between any of the
available datasets. These are critical covariates in models as they are amongst
the most important risk factors for severe disease.

- Did not have documented episode of COVID-19 and/or receipt of any COVID-19
vaccine during the 90 days prior to index date. These persons likely have lower
risk of vaccination (the study's exposure of interest) as well as temporarily
lower risk of COVID-19 (the study's outcome of interest), which would introduce
heterogeneity in estimates were they not excluded.

Variables

Primary and secondary objectives, exposure: BNT162b2 mRNA vaccine, enumerated using
Clinical Vaccines Administered (CVX), National Drug Code (NDC) and Current
Procedural Terminology (CPT) codes.

Primary and secondary objectives, outcomes: COVID-19 related endpoints:

- Outpatient encounters with an International Classification of Diseases, Tenth
Revision, Clinical Modification (ICD-10-CM) code U07.1 "COVID-19"

- Incidence, as the number of people experiencing an event divided by the
number of people at risk of an event (i.e. proportions)

- Incidence rate, as the number of people experiencing an event divided by
the total person time at risk of an event (i.e. X events per 100,000
person-months)

- Adjusted hazard ratios

- Cost associated with these encounters, enumerated as mean (standard
deviation) and median (first quartile - third quartile) among all non-zero
costs.

- Emergency department encounters with an ICD-10-CM code U07.1

- Incidence, as the number of people experiencing an event divided by the
number of people at risk of an event (i.e. proportions)

- Incidence rate, as the number of people experiencing an event divided by
the total person time at risk of an event (i.e. X events per 100,000
person-months)

- Adjusted hazard ratios

- Cost associated with these encounters, enumerated as mean (standard
deviation) and median (first quartile - third quartile) among all non-zero
costs.

- Urgent care encounters with an ICD-10-CM code U07.1

- Incidence, as the number of people experiencing an event divided by the
number of people at risk of an event (i.e. proportions)

- Incidence rate, as the number of people experiencing an event divided by
the total person time at risk of an event (i.e. X events per 100,000
person-months)

- Adjusted hazard ratios

- Cost associated with these encounters, enumerated as mean (standard
deviation) and median (first quartile - third quartile) among all non-zero
costs.

- Inpatient encounters with an ICD-10-CM code U07.1. Of note, potentially
incidental findings of COVID-19, such as those during the course of a major
psychiatric or physical trauma associated admission, will be excluded using
previously described methodology.

- Length of stay, as the number of days from admission to discharge
expressed as mean (standard deviation) and median (first quartile - third
quartile)

- Incidence, as the number of people experiencing an event divided by the
number of people at risk of an event (i.e. proportions)

- Incidence rate, as the number of people experiencing an event divided by
the total person time at risk of an event (i.e. X events per 100,000
person-months)

- Adjusted hazard ratios

- Cost associated with these encounters, enumerated as mean (standard
deviation) and median (first quartile - third quartile) among all non-zero
costs.

- If sample size permits, the above outcomes will be further stratified into
cases that did and did not require intensive care unit (ICU) admission,
invasive mechanical ventilation (IMV) or resulted in death.

Potential confounders: a diverse set of variables will be used to balance baseline
characteristics between those who are vs are not vaccinated. In brief, these will
include sociodemographic and clinical features known to be associated with the risk
of exposure and outcome and may need to be tailored based on the structure of
confounding assumed to be present.5,6 The balanced set of covariates will contain a
sample where standardized mean differences (SMD) between measured confounders are ≤
0.1.7.

Data Sources

Medical and pharmacy claims

Insurers contributing to closed claims include a mix of commercial payers, Medicare
Advantage/Part C plans, and Medicaid Managed Care plans. Data elements in medical
claims include patient demographic information, inpatient/outpatient visit-level
information such as diagnoses, procedures, and length of stay, hospital
characteristics, and medication information. Owing to the nature of claims, the data
represent the final set of diagnoses over the course of the hospitalization sent to
the patient's insurer for reimbursement, with diagnosis prioritization assigned by
clinicians or hospital staff. Data elements in pharmacy claims include patient
demographic information, NDC, date of service, dispensed quantity, and days' supply.

State vaccine registries

Since December 2020, vaccinators in many jurisdictions and states have been mandated
to report COVID-19 vaccine administered to their respective state health
authorities.

State registry files have unique person identifiers, vaccination event date, and CVX
codes, which are unique to each vaccine.

Linkage methods between data sources

The administrative insurance claims data and the state vaccine registry data were
independently tokenized using HealthVerity's tokenization software, meaning patient
identifiers are passed through software to create a unique de-identified
identification number for each person. Identifying information is removed by the
vaccine registries before sending the data to HealthVerity for data aggregation.
This process ensures Health Insurance Portability and Accountability Act (HIPAA)
Privacy Rule compliance via expert determination. The insurance claims data were
then linked to the state vaccination registries via the tokenized identification
number.

Study Size

This study uses retrospective deidentified data without ability to recruit to a
target, sample size calculations are not applicable. The expected number of people
eligible for the study is expected to be approximately 30% of a state's population,
given HealthVerity's marketplace coverage. In order to maintain confidentiality as
well as causal inference principles of consistency, positivity and exchangeability,
outcomes, covariates and subgroups will have at least 10 people in each category for
inclusion in outputs. Throughout the protocol, this is referred to as "If sample
size allows".

Data Management

Claims and vaccine registry data refreshes are delivered by HealthVerity monthly.
All HealthVerity structured data described above are stored within Pfizer's data
infrastructure. Data are queried and analyzed using Statistical Analysis System
(SAS) and/or R. The HealthVerity claims, mortality, and vaccine registries are
stored as separate databases and, owing to the nature of the data use agreement, on
a different server than all other enterprise-wide data. The data are only accessible
to Pfizer colleagues who have been trained and approved to access Pfizer's data
warehouse.

In brief, all data in the year prior to index through 6 months after index for
persons who meet inclusion and exclusion criteria detailed in section 9.1 will be
extracted.

Data Analysis

Data are extracted using standardized macro scripts by two experienced programmers,
with all discrepancies resolved before final results delivery. Descriptive analyses
will compare the distribution of sociodemographic and clinical characteristics in
final cohorts.

Time-dependent Cox proportional hazards models will be used to estimate the hazard
of each outcome of interest with time-varying exposure and time-fixed covariates,
resulting in adjusted hazard ratios (aHR) and 95% confidence intervals (95% CI).

Detailed methodology for summary and statistical analyses of data collected in this
study will be documented in a SAP, which will be dated, filed, and maintained by the
sponsor. The SAP may modify the plans outlined in the protocol; any major
modifications of primary endpoint definitions or their analyses would be reflected
in a protocol amendment.

Quality Control

Data in HealthVerity's database are provided monthly in an electronic format.
HealthVerity employs its tokenization product to match patients between different
data sources with high accuracy. Each monthly delivery of data contains a quality
control report, which was co-created between HealthVerity and Pfizer subject matter
experts to ensure continued quality. All analyses will be performed internally and
according to Pfizer analytic standards, including double programming to ensure
quality control.

Limitations of the Research Methods

- Some, but not all, states have COVID-19 vaccine reporting mandates, which are
required for ascertainment of unvaccinated status. Therefore, the results of
this study using states with mandatory reporting jurisdictions may not
generalize to other areas of the country. However, a variety of geographic,
sociodemographic, and health characteristics are included amongst the included
populations.

- There is no laboratory confirmation of the COVID-19 diagnosis. It is difficult
to know the direction of bias that this limitation may create. It is possible
that some patients are being diagnosed with symptoms alone (over-ascertainment
of outcomes of interest) or that we are missing positive cases due to at-home
testing (under-ascertainment of outcomes of interest).

- The results from this study may differ from those of VE studies using other
data sources, such as Kaiser Permanente of Southern California or the Veterans
Administration.

- Attempts to address confounding may not be able to fully account for
confounders.