Public Health Impact of the Pfizer-BioNTech COVID-19 vaccine (BNT162b2) in the first year of rollout in the United States

Study Design

A combined Markov and decision tree model with transition probabilities based on COVID-19 literature, was developed to estimate the one-year public health impact of Pfizer-BioNTech COVID-19 Vaccine (BNT162b2) compared to no vaccination. The base case analysis was conducted in the US target population aged >12 years, from a societal perspective. Age-stratified epidemiological, clinical, economic, and humanistic parameters were derived from published literature.

The expected public health impact was estimated as the number of COVID-19 deaths and symptomatic cases averted, with the latter differentiated among those requiring outpatient versus inpatient care. The expected economic impact was estimated as the amount of avoided healthcare and societal costs corresponding to the health outcomes averted. Productivity and QALY loss associated with early death was accounted for on a lifetime horizon.

The model was programmed in Microsoft Excel® 2021. The manuscript was developed in alignment with the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) 2022 guidelines (Table S1) [38].

Model Structure

A Markov model (Figure A) was constructed with eight mutually exclusive health states based on: i) Susceptible-Infected-Recovered (SIR) structure which has been used in monitoring and predicting the spread of COVID-19 [39], and ii) Vaccinated states of partial (dose 1) or full vaccination (dose 2) depending on the vaccine coverage and breakthrough infections. The ‘infected states’ (infected, infected post dose 1, infected post dose 2) of the Markov model were linked with a Decision tree (Figure B) for transition probabilities of symptomatic and asymptomatic infections, and sequelae such as hospitalization, receiving intensive care (ICU) and/or invasive mechanical ventilation (IMV), and post-acute sequelae of COVID-19 (PASC).

• Download figure
• Open in new tab

Figure A: Structure of the Markov model

• Download figure
• Open in new tab

Figure B: Decision tree for the probabilities of COVID-19 symptoms and sequelae

Individuals entered the Markov model (Figure A; Table 1) in the ‘susceptible’ or ‘recovered’ health states with weekly cycles for one full year. At each weekly cycle, individuals could transition to other health states or remain in the current state. From the susceptible state, individuals could become infected or receive the vaccine (dose 1 with partial vaccination followed by dose 2 of full vaccination). An age-dependent yearly attack rate in the susceptible population defined the share of the model cohort that would become infected with COVID-19 and vaccination was determined by age-dependent vaccine coverage (further explained in model parameters). From the vaccinated states, patients could experience breakthrough infections or become susceptible again due to waning of vaccine protection. The former was governed by the vaccine efficacy, which could vary by dose, age group and the circulating variants of COVID-19. Waning of the efficacy of the vaccine and of infection-induced immunity was captured through their corresponding duration of protection. In each cycle of the Markov model, a fraction of the cohort was moved from the ‘vaccinated’ (1 or 2 doses) and ‘recovered’ health states to the ‘susceptible’ health state, representing the gradual decrease of protection level in the population due to waning.

Once individuals got infected in ‘infected states’ (infected, infected post dose 1, infected post dose 2), they moved to the decision tree (Figure B; Table 1) for symptomatic or asymptomatic infection. We assumed that individuals with asymptomatic infection did not die or require any inpatient or outpatient treatment, but could experience PASC complications. The term PASC, also known as ‘long COVID’, is used to describe the post-acute sequelae and long-term symptoms that can be experienced from weeks to months by persons recovering from COVID-19 [40]. Symptomatic cases were moved further into the decision tree according to outcome probabilities informed by the individual’s health state specific efficacy against symptomatic disease and against hospitalization. The probability that infected individuals experienced symptoms was based on peer-reviewed literature [41]. Symptomatic cases were assumed to be managed in the outpatient or hospitalized setting, and incurred healthcare costs for clinical care including, respectively, visits, testing, medication, and hospitalization treatment. Hospitalized patients remained in the general ward or were admitted to the ICU and could receive IMV in either setting. Further, individuals that survived infection were then subject to a probability of developing PASC and incurred the costs of managing these symptoms. The PASC probabilities were sourced from published literature [42, 43]. We also assumed that individuals who got infected were immune to reinfection for an average of nine months after the infection [44].

At the end of the decision tree, individuals returned to the Markov model component to either the ‘recovered’ or ‘death’ state, based on the outcome of the decision tree. Death following infection in the decision-tree was due to COVID-19, which reflected the case fatality rates (Table 1, Table S9), while death from any other state was due to all other cause mortality. The ‘recovered state’ represented individuals that had recovered from infection and had infection-induced immunity to disease.

Model Parameters

The model included a multitude of inputs derived from published literature and supported by medical review. The main inputs are reported in Table 1, further details are included in the electronic supplementary material (Tables S2-14).

Analyses

The base case predicted the health outcomes of the Pfizer-BioNTech COVID-19 Vaccine (BNT162b2) in terms of the number of COVID-19 symptomatic cases, hospitalizations, and deaths averted over a year, alongside the corresponding direct healthcare and productivity costs saved, and QALYs gained, compared with no vaccination, from a societal perspective.

Two scenario analyses were performed to assess the impact of the parameter uncertainty and to evaluate the robustness of the base case findings. The scenarios used extreme but plausible ranges of parameters, which were informed by the lower and the upper bounds of confidence intervals (CIs) of the values used in the base case, or, in their absence, by medical opinion. The lower bounds were used for a “Low” scenario, and the upper bounds were used for a “High” scenario for all parameters except those related to the proportion of individuals with infection-induced immunity at model start, and the duration of protection from infection-induced immunity; for both the parameters, their upper bound values were used in the “Low” scenario. The parameters tested were those carrying higher uncertainty due to wider CIs, or limited evidence. These included the inputs determining the clinical course of COVID-19 in the absence of vaccination (i.e., attack rate) (Table S4), the proportion of subjects with infection-induced immunity (Table S3), the duration of infection-induced immunity (Table 1), the probability of symptoms (Table S9), the hospitalization rate for new admissions (Table S9) and re-admissions (Table 1), and the probability of outpatient deaths (Table S9). The inputs related to the clinical profile of the vaccine were varied as well: the variant and dose-specific VE against infection, VE against symptomatic disease, VE against severe disease (Table S8), as well as dose-specific duration of vaccine-induced protection (Table 1). Rapid literature review efforts highlighted that a rich body of evidence exists on the burden of disease and the clinical effectiveness of the vaccine, and data on healthcare resource utilization (HCRU) and costs is quickly emerging. As additional evidence was reviewed, alternative plausible ranges for model inputs were derived and used in sensitivity analyses to identify the most sensitive parameters and to characterize the uncertainty. As such, extensive sensitivity analyses were conducted on over 15 model parameters to independently evaluate the impact of individual parameters on the baseline results, using values from alternative sources. One or multiple parameters within the same category were changed at a time using different sets of input values (Tables 4, Tables S2-12).

On an exploratory basis, the sensitivity analyses also examined the impact of the year-end rollout in children aged 5-11, building on the growing clinical and economic evidence in this age group (Tables S2-14).

Implementation costs were also considered on an exploratory basis. This study focused on the assessment of the preventable costs associated with vaccine-preventable disease; it did not consider implementation costs, such as those related to infrastructure, supply, training and human resource management, outreach, or ancillary care management. These costs have been considered largely unknown, and not a primary driver for decision-making on the authorization of COVID-19 vaccines in the context of the pandemic [52]. The cost of vaccine administration for the US payer system has however been pointed out [52], and adverse events have been reported in the RCTs [53, 54]. Sensitivity analyses explored the effects of these two additional cost categories on the overall cost saving estimates (Tables 4, Table S12).

Finally, sensitivity analyses were run to assess the impact of alternative utility decrements and duration of symptoms input values on the QALY loss. Given that literature on COVID-19 specific utilities is scarce, arbitrary values (± 10%) informed the sensitivity analyses on utility decrements. A longer duration of symptoms was tested: 14 days [15], versus 5 days used in the base case (Table S14).

Burden of disease in the absence of vaccination

Table 2 shows the estimated burden of disease in the absence of COVID-19 vaccination in the base case, “Low” and “High” scenarios.

It was estimated that, by the beginning of the vaccination campaign, about 33 million (“Low”: 31 million, “High”: 35 million) US individuals aged ≥12 and older had been infected with SARS CoV-2 and recovered, representing ∼12% of the eligible population. This approximation was based on a seroprevalence study [31], and was conservative, although consistent with prior literature (11.9% by 30^th October according to Sullivan et al., [55], and 14% by mid-November according to Angulo et al., [56]).

Leveraging nationwide COVID-19 burden of disease data reported prior to the introduction of COVID-19 vaccination [31], the base case and scenarios estimated that, of the eligible and susceptible population, 24.2% (“Low”: 17%, “High”: 31.8%) would have become infected with COVID-19 during 2021 in the absence of vaccination. These estimates were consistent with CDC burden of disease data for 2021 including the vaccination period (21.7% based upon [2]) as well as seroprevalence data [57].

Based upon a review of genome sequencing data [49], it was calculated that about 22% of all infections were of the original strain, 17% were Alpha, ∼0% were Beta, 2% were Gamma, and almost 60% were Delta. This distribution was found to be consistent with additional literature [58, 59]. Using age-stratified probabilities of symptoms and hospitalization rates reported by Reese et al., [41] (Table 1), about 3.5 million (“Low”: 2.3 million, “High”: 5.6 million) hospitalizations were estimated to have occurred in 2021 in the absence of vaccination. This amount corresponds to a 5.8% (“Low”: 5.5%, “High”: 6.2%) overall symptomatic hospitalization rate across age groups, which aligns with prior literature (3.4% according to Angulo et al., [56]) and is consistent with a total of over 4.4 million hospitalizations since the start of the pandemic, per the CDC Tracker (accessed on 2/17/2022) [2].

Using age-stratified mortality rates reported in real-world retrospective studies [31, 33], we estimated that almost 500,000 (“Low”: 323,000, “High”: 810,000) deaths would have occurred in 2021 in the absence of vaccination. This estimate is consistent with a total of over 920,000 deaths since the start of the pandemic, per the CDC COVID Data Tracker (accessed on 2/17/2022) [2], and with published estimates of excess deaths among unvaccinated [60].

Base Case results

Based upon the clinical and economic inputs set for the base case (Table 1), it was estimated that, compared to no vaccination, the two-dose series of the Pfizer-BioNTech COVID-19 Vaccine (BNT162b2) prevented almost 8.7 million symptomatic cases, averted approximately 690,000 hospitalizations, and saved over 110,000 lives (Table 3).

Up to, respectively, 17.8%, 54.8% and 77.3% of the vaccine-prevented symptomatic cases, hospitalizations, and deaths were estimated to have occurred among individuals aged ≥65 years.

The total hospitalizations prevented were estimated to be composed mainly of stays in a normal ward without mechanical ventilation (501,191 [72.7%]), although the vaccine was also estimated to prevent more severe hospitalizations (32,104 [4.7%] stays in a normal ward with mechanical ventilation, 65,279 [9.5%] ICU stays without mechanical ventilation, and 91,223 [13.2%] ICU stays with mechanical ventilation). The reduction in COVID-19 burden led to an estimated total direct cost savings of $30.4 billion, driven by inpatient and PASC costs.

Overall, about 1.1 million discounted life-years (LY) and QALYs were gained. Around 85.5% of this QALY gain stemmed from avoiding COVID-19 mortality, while quality of life losses corresponding to morbidity from infection symptoms and long-term effects of mechanical ventilation had a moderate effect.

The observed reductions in COVID-19 symptomatic cases were also associated with productivity gains of $43.7 billion. Around 35.7% of this gain was related to the productivity loss due to early death and 64.3% to lost workdays during illness.

Scenario Analyses

Table 3 shows the outcomes averted in the two alternative scenarios (“Low”, and “High”). Using the most conservative estimates of burden of disease and clinical effectiveness, the “Low” scenario estimated that the Pfizer-BioNTech COVID-19 Vaccine (BNT162b2) prevented over 5 million symptomatic cases, averted over 260,000 hospitalizations, and saved over 45,000 lives.

The “High” scenario estimated that the Pfizer-BioNTech COVID-19 Vaccine (BNT162b2) prevented over 13.8 million symptomatic cases, averted over 1.2 million hospitalizations, and saved over 200,000 lives.

Sensitivity Analyses

Sensitivity analyses results are presented in Table 4. The model was most sensitive to attack rates and hospitalization rates, whereas, probability of symptoms, clinical effectiveness, HCRU, and cost had moderate impacts on the outcomes of interest.

As expected, attack rates defined based upon the CDC reported cases in 2020 or 2021 without adjusting for under-reporting led to significantly lower burden averted. The CDC burden of disease estimates adjusted for under-reporting were in line with the seroprevalence data used in the base case and had only a marginal impact. As under-detection and under-reporting especially in the early phase of pandemic have been widely documented [61], estimates of attack rates based on the reported cases are arguably not realistic and too conservative. As additional literature offered hospitalization rates across different time periods, age and risk-based populations, the hospitalization rates had a significant impact especially on the number of deaths and hospitalizations averted, which has led to a moderate to high effect on the direct costs and productivity loss. Alternative values for VE and vaccine-induced duration of protection moderately affected all the outcomes of interest, whereas, infection-induced duration of protection (i.e., +/- 3 months of base case at 9 months) had a marginal impact. The HCRU and costs had a moderate impact on direct care costs. Sensitivity analyses with 10% higher and lower utility decrement showed a marginal effect on the QALYs saved of ±0.2%. Increasing the duration of symptoms from 5 days (base case) to 14 days [15] led to a marginal increase of 3.2% in the QALYs saved.

The inclusion of the year-end rollout in children 5-11 slightly increased the outcomes averted, however, to a very low extent, driven by low vaccine coverage in the earliest phase of rollout in November and December 2021, as well as lower burden of disease than adults. Finally, the inclusion of additional care management costs (adverse events management and vaccine administration costs) had a low to moderate impact on the direct costs; those costs were offset by the total cost savings associated with vaccine-preventable disease.