Since its emergence in December 2019, SARS-CoV-2, which is the virus that causes COVID-19, has been continuously spreading worldwide [1]. As of November 12, 2021, this pandemic has caused more than 250 million cases and more than 5 million deaths over the previous 2 years [2]. Vaccination programs are one of the most successful public health interventions that primarily aim to develop herd immunity and protection against pathogens. To date, the World Health Organization has approved 8 COVID-19 vaccines that are efficacious in preventing infection, reducing the level of severity and the number of deaths due to SARS-CoV-2 infection [3]. However, most data were generated from clinical trial studies that were different from a real-world setting due to their well-controlled nature [4]. Furthermore, real-world vaccine effectiveness (VE) studies were mostly conducted in high-income nations, and few studies came from low- and middle-income countries, even though these countries might have unique social and economic characteristics that not only affect vaccination acceptance but also disease outcomes [5]. A previous study has reported the association of living in an impoverished neighborhood with an increased risk for mortality even after COVID-19 vaccination [6]. Therefore, studies investigating how these COVID-19 vaccines might perform in real-world settings where social inequalities exist are still essential for informing policies and strategies to improve disease outcomes in future pandemics, especially in societies with health care system constraints [7]. This study aims to evaluate COVID-19 VE against symptomatic and fatal SARS-CoV-2 infection during the Delta and Omicron waves in Jakarta, where income inequality and socioeconomic segregation have become a common phenomenon [8].

The primary objective is to study real-world VE against symptomatic COVID-19 (mild, moderate, and severe).

The following secondary objectives were also explored:

This population-based study adopts a test-negative case-control design to analyze COVID-19 VE against symptomatic and fatal SARS-CoV-2 infection from available health care datasets in the Special Capital Region of Jakarta, one of the provinces with the highest population density (15,900 people per km² in 2020) and case rate per population relative to other regions in Indonesia [9]. Despite its robust development and economic growth, the contrast between rich and poor neighborhoods was highly prevalent in many parts of Jakarta’s districts [8,10].

This study will estimate VE against COVID-19 during the period of interest, which were the COVID-19 waves associated with the Delta and Omicron variants, from June 1 to August 31, 2021, and January 1 to April 2, 2022, respectively. Within each of these periods, a cycle of sustained upward and downward trends of the test positivity rate was observed [11-15] (Figure 1). Furthermore, the Delta and Omicron variants were estimated to be responsible for around 90% of SARS-CoV-2 infections in Jakarta during these two periods [14,16].

During the Delta and Omicron periods, several brands of COVID-19 vaccines were introduced in Jakarta for vaccine rollout, such as AstraZeneca, SinoVac-CoronaVac Biofarma, Covovax-Novovax, Indovax, Johnson & Johnson, Moderna, Sinopharm BBIBP, and Pfizer (Table S1 in Multimedia Appendix 1) [17]. Among these COVID-19 vaccines, Coronavac-Sinovac was the earliest COVID-19 vaccine to be introduced in January 2021 and used for mass vaccinations of doses 1 and 2, followed by AstraZeneca, Sinopharm, Pfizer, and Moderna in August 2021 [18,19]. Booster vaccinations 6 months after the 2-dose primary vaccination with Moderna and Pfizer vaccines were initially introduced to health care workers in July 2021, followed by the approval of the AstraZeneca vaccine as a booster in the national vaccination program in January 2022 [20,21]. To expedite the booster vaccination coverage, the timeline for booster administration was shortened to 3 months after the 2-dose primary vaccination, and additional inactivated COVID-19 vaccines, such as SinoVac and Sinopharm, were included in the national booster program [22]. Different from Pfizer, Moderna, and AstraZeneca, these inactivated COVID-19 vaccines were only approved as boosters for patients who received similar vaccine platform for their primary vaccination regimen [23]. Until the end of the Omicron period, all of the distributed COVID-19 vaccines in Jakarta were developed from the ancestral strain of SARS-CoV-2.

The study will include people 18 years or older who domiciled in Jakarta and underwent an RT-PCR test for SARS-CoV-2 infection due to symptoms of influenza-like illnesses within 10 days of the test date. Although people may have more than one RT-PCR result during the period of interest, this study will only account for the first positive or negative result per person as the index test date for each of the COVID-19 waves. People who had at least one sample with a positive SARS-CoV-2 RT-PCR result during the period of interest will be classified as cases, while people with a negative SARS-CoV-2 RT-PCR without any positive results over the period of interest will be classified as controls.

People with a positive RT-PCR test result within 90 days of the preceding index test date, with inconclusive RT-PCR results, who received different vaccines for dose 1 and dose 2 or a heterologous primary vaccine series, or who had a time interval between dose 1 and dose 2 that was less than the government recommendation (<21 days for Pfizer and Sinopharm, <28 days for CoronaVac and Moderna, and <12 weeks AstraZeneca) will be excluded from the analysis. We will also exclude people with incomplete vaccination records, such as people who declared receiving 2 vaccine doses without being able to verify the previous dose.

The primary outcome in this study is symptomatic SARS-CoV-2 infection, defined as a positive RT-PCR test within 10 days from the onset of influenza-like illness. The secondary outcome in this study is fatal SARS-CoV-2 infection, defined as death within 30 days after a positive RT-PCR test result [24]. Meanwhile, the primary exposure in this study was vaccination status, which can be classified as unvaccinated, dose 1 or partial vaccination, dose 2 or primary vaccination, and booster dose. People without any vaccination entries after data linkage with the PCARE Vaksin dataset at the end of each period of interest will be classified as unvaccinated persons. For vaccinated people, we will only consider the last vaccination data entry that happened before the patients’ index RT-PCR test date.

Several covariates should be assessed as potential confounders in the study, such as age in years at the index test date, gender, calendar week of the RT-PCR collection during the period of interest, living in impoverished neighborhoods, presence of any comorbidities, occupation, and reinfection status. The calendar week of the RT-PCR collection during the period of interest will be presented as a whole number starting from week 1 as the start of the period of interest. Living in an impoverished neighborhood will be presented as a dichotomous variable. People will be classified as living in an impoverished neighborhood if their neighborhood number at the subdistrict level is listed as one of the impoverished neighborhoods in the 2018 Jakarta Governor’s regulation concerning improving the quality of settlements in residential areas. Occupation will be presented as categorical variables that consist of health care worker, public or government official, and civilian. Reinfection or people with previous SARS-CoV-2 infections will be presented as a dichotomous variable. People with a previous positive RT-PCR SARS-CoV-2 result that occurred more than 90 days after the current case will be classified as people with a previous SARS-CoV-2 infection [25].

To control for unmeasured confounders [26], such as societal preventive measures or the changing dynamic of the viral transmission within the community, this study will match each case with a control using a ratio of 1:2 by matching for age within 10 years of the case’s age, gender, and the calendar week of the RT-PCR collection during the period of interest. The matching procedure will be assessed with standardized mean differences, with values less than 0.1 indicating sufficient matching [27,28].

This study will analyze the final datasets derived from the integration of several datasets provided by the Jakarta Provincial Health Office, Ministry of Health, and Social Security Agency on Health after April 2022 (Table 1). Variables such as ID number and date of birth will be used for the dataset linkage (Figure 2). From the New All Records (NAR) and Suspect and Probable (SUSPROB) datasets, symptomatic people with a positive or negative first RT-PCR test result within each of the periods of interest will be identified as cases or controls. Furthermore, people identified as cases will be linked to the entries in the REV POS dataset to extract the outcome of their disease course (Figure 2). People with incomplete data for ID number, date of birth, gender, swab collection date, PCR test result, domicile address, outcome, and date of death (if the case ends in death) will be excluded from further data linkage.

People with data linked to the entries within the Primary Care (PCARE) Vaksin dataset will be classified as vaccinated people (Figure 2). To avoid misclassification of foreigners as unvaccinated people, the study will only use entries with national identity card numbers as ID numbers from each of the study datasets. Vaccinated people with incomplete data on vaccination date and vaccine name will be excluded from the final dataset (Table S2 in Multimedia Appendix 1). From the final dataset, a subset of data containing the matched cases and controls will be created (Table S3 in Multimedia Appendix 1).

As genomic testing was not routinely performed in Indonesia, no data linkage will be performed on the variant dataset. In contrast, the variant dataset, which consisted of the genomic test result of the randomly sampled specimen from the population, will be used to confirm the dominant SARS-CoV-2 variant in Jakarta during each period of interest.

The calculation formula for the sample size used in this study will follow the methodology described by O’Neill [29]. Assuming vaccination coverage between 20%-90%, a precision of ±5%, a type 1 error rate of 0.05, and an anticipated VE of 50%-90%, the minimum sample size needed for each specific COVID-19 vaccine brand and dose VE analysis is roughly around 15,000, with 5000 cases and 10,000 controls [24,29] (Table S4 in Multimedia Appendix 1).

We expect to analyze 400,000-900,000 people after the aggregation of all datasets. Therefore, we should have enough data to conduct the VE analyses for each specific COVID-19 brand and regimen in our final dataset. We will compare the proportion of COVID-19 vaccines received by eligible people in our dataset during each period of interest to those reported by the Ministry of Health [17] (Table S1 in Multimedia Appendix 1). Furthermore, we will exclude people who received COVID-19 vaccine brands with vaccination coverage of less than 20%, as these vaccines are usually distributed briefly in the population or only available to specific subgroups within the community. Moreover, we might consider combining different COVID-19 vaccine brands with the same platform into one group to decrease the required sample size, especially for the subgroup VE analysis [24].

We adopted our VE analysis from the Interim Guidance of Evaluation of COVID-19 Vaccine Effectiveness by the World Health Organization, where the final estimate of absolute VE (aVE) will be determined by calculating the adjusted odds ratio (aOR) for vaccination using the formula aVE = (1 – aOR) × 100% [24]. Unconditional multivariable logistic regression will be used to analyze the aOR of having symptomatic and fatal SARS-CoV-2 infection between the unvaccinated and vaccinated participants. We will perform separate analyses for the Delta- and Omicron-dominant periods. Furthermore, within each period, we will also analyze each vaccinated person separately by their vaccine brand or platform and vaccination doses, such as dose 1, dose 2, or booster. For the main aVE analyses, we will include only vaccinated people with the last dose 1 or dose 2 vaccination date within 14-90 days or with the last booster dose within 7-90 days before their index test date and compare them to the unvaccinated people.

One main logistic model will be assembled by forward inclusion, and only covariates with less than 20% missing data will be considered to be included in the model building. Assessment for confounders and effect modifiers will be performed with the Wald test and by observing the change in the primary exposure coefficient SE. Covariates that act as confounders will be included in the model. If effect modification is present, it is essential to report VE and CIs for each subgroup individually. The linearity between continuous variables and the log odds of having symptomatic SARS-CoV-2 infection will be assessed, and the continuous variables will be modified into categorical variables if necessary. The overall fit of the final multivariate logistic regression model will be assessed with Hosmer and Lemeshow goodness-of-fit test [30], as matching might incur matching-related bias from the cases and controls having similar exposure, thus underestimating the aVE. We will also perform conditional logistic regression as a sensitivity analysis for measuring the aVE estimation against symptomatic SARS-CoV-2 infection or our primary objectives [31,32]. Furthermore, for the analysis of the Omicron period, we will perform an additional analysis of relative VE (rVE) by comparing recipients of one booster dose to recipients of the 2 doses who were eligible to receive the booster dose. The rVE will be calculated using the formula rVE = [(aVE from booster dose – aVE from 2 doses and eligible for booster)/(1 – aVE from 2 doses and eligible for booster)] × 100%. For the rVE analyses, we will include only people whose last booster dose was within 7 to 90 days before their index test date and compare them to people whose last 2-dose vaccination date was ≥90 days before their index test date [24].

This study will also conduct subgroup analyses based on the time between the last vaccination date and test index date to assess for waning of immunity (Table 2). These subgroup analyses will be conducted separately for each wave and vaccine brands or regimens. Similar to the main analysis, each vaccinated subgroup will be compared to unvaccinated people.

A 2-sided P value derived from the multivariable logistic regression analysis will be used to assess VE. Data linkage and data cleaning will be performed in Google Collaboratory with the Python programming language. Furthermore, the Statsmodel library package will be used for statistical analysis [33].

The research protocol received ethics approval from the Ethical Committee of Persahabatan Hospital Jakarta (40.A.1/KEPK-RSUPP/11/2022). The research conducted in accordance with this protocol will not collect consent from people as it involves the use of secondary data. A consent waiver statement has been issued by the Ethical Committee of Persahabatan Hospital Jakarta (DP.04.03/D.XX.10.4/0001/2024).