Immunization has led to a significant decrease in mortality in children under 5 years of age. Several studies have demonstrated that the reduction in mortality and morbidity due to vaccination extends beyond the targeted infections. This seems to result from broader nonspecific effects (also termed heterologous immunity) stemming from the synergistic effects of several live vaccines [1-4]. The vaccines demonstrated to have nonspecific effects include the Bacillus Calmette-Guérin vaccine (BCG), polio vaccine, and measles vaccine [5]. Overall, the nonspecific effects of vaccines vary according to age, sex, time to administration, last vaccine administration, previous and concurrent infections and immunizations, interval between vaccines, genetic factors, nutritional state, season, and co-administration of other immunomodulating agents [1].

BCG is a live attenuated vaccine against tuberculosis (TB), which is administered to newborns ideally at birth or before the seventh day of life. Despite its low efficacy for preventing primary infection or reactivation of latent pulmonary infection and TB infection, BCG is effective against childhood tuberculous meningitis and miliary disease [6]. Hence, BCG at birth is recommended in settings with a high incidence of TB, and it is parsimoniously administered in countries with a low incidence and has been removed from routine vaccination schemes in high-income countries with a low TB incidence [7].

BCG was introduced in the 1920s in Sweden, where it quickly became apparent that the all-cause mortality was lower in BCG-vaccinated children than in those not vaccinated [8]. In England and the United States, the same pattern was observed between the late 1940s and early 1960s, where a reduction in mortality from diseases other than TB was estimated in 25% of vaccinated children (95% CI 6%-41%) [1,9]. Later studies showed the same pattern in infant mortality and morbidity [10], as well as in morbidity in the neonatal period [11]. Furthermore, BCG is used as a standard treatment for carcinoma of the bladder and influences the natural history of infectious and neoplastic diseases [1]. More recently, the effect of BCG was tested in viremia after SARS-CoV-2 exposure [12].

Further studies in children from Denmark and Greenland failed to show a decrease in hospitalization due to infectious diseases in vaccinated children [13,14]. On the contrary, a cohort study of children from 19 different countries revealed that BCG-vaccinated children under 5 years of age had a lower risk of suspected acute lower respiratory infection [15]. In Spain, hospitalization rates due to respiratory infection were lower in BCG-vaccinated children, with an attributable fraction of 32% for children under 1 year of age, and a similar finding was noted for sepsis, with an attributable fraction of 53% [16].

A systematic review concluded that BCG has a beneficial effect on mortality in children, with differential effects according to age (earlier administration of the vaccine is associated with greater effects) [5]. Nevertheless, there has been a call for more research on the nonspecific effects of BCG [5,17]. Currently, studies are seeking to further investigate the nonspecific effects of BCG [18] in terms of the effects on allergy and infection [7], severe morbidity, nonaccidental hospital admission, and all-cause consultation in children under 5 years of age [19].

In Portugal, BCG began to be routinely administered to all newborn children in 1965. Until 2016, a single BCG dose was administered to all newborns, typically before discharge from the maternity ward (most births in Portugal are hospital-based) or as early as possible thereafter [20].

In June 2016, a clinical guideline established that only children under 6 years of age who belong to high-risk groups (those originating from countries with high TB incidence; having contact with active cases or persons under prophylaxis; having HIV-positive mothers; having parents with a alcohol or drug abuse problem and antecedents of TB; having parents who have been in prison in the last 5 years; living in high-risk TB communities; or traveling to high-incidence countries) should be vaccinated as close to birth as possible [21]. Some of these risks were not considered individually but by proxy, with some parishes (freguesias) thus considered high risk for the criteria defined.

In 2016, 42% of children under 1 year of age had received BCG (compared to 98% in the previous year) [20]. Between 2016 and 2018, the incidence of TB increased by 16% in children under 1 year of age (from 7.0 to 8.1 per 100,000 inhabitants) and 192% in children aged 1 to 5 years (from 2.6 to 7.6 per 100,000 inhabitants) [20].

Between the adoption of the high-risk strategy for BCG vaccination in 2016 and 2018, the infant mortality rate (IMR) remained stable at 3.2‰ to 3.3‰, with important variations between the considered years [20]. In 2019, there was a decrease to 2.8‰. The mortality rate in children under 5 years of age decreased from 3.4‰ in 2016 to 3.0‰ in 2019, but not constantly in this period (eg, during 2018 it was 3.5‰) [20]. No death in children under 5 years of age has been attributable to TB since the implementation of the high-risk strategy for BCG [20]. Nevertheless, the number of severe cases of TB in children under 5 years of age decreased in 2020 compared to 2018 (4 cases) and 2019 (7 cases), with just 1 case registered in a child without BCG vaccination [22].

Since the implementation of the high-risk strategy for BCG in 2016, no assessment has been conducted regarding the impact of the BCG policy shift on severe, moderate, or mild morbidity and on mortality in children. It is a common problem that vaccination practices are changed without evaluating the overall health effects of the change and that the nonspecific effects of vaccines are not included in the considerations regarding vaccine policies.

In 2021, the first cohort of children not vaccinated for BCG completed the fifth year of life, and evidence is needed regarding the impact of this strategy, especially given fluctuations in the IMR, a decline in the TB notification rate not accompanied by a decline in the incidence, a higher TB incidence in greater metropolitan areas, and a high median number of days from symptoms to diagnosis [22].

No data have been gathered on overall hospital admissions and morbidity patterns. Similarly, no data have been compiled on the BCG strains used in the country and their differences in terms of efficacy. This evidence will have a bearing on the continuity of the current strategy.

The need for more and better evidence is paramount, especially given that besides deciding on the inclusion of BCG in the country’s immunization plan, arguments on the nonspecific effects of vaccines (and not only of BCG) can help overcome vaccine hesitancy. In high-income countries, including Portugal, vaccine hesitancy has been responsible for a decline in coverage rates and for the re-emergence of severe cases of vaccine-preventable diseases such as measles [23,24].

Given the described effects of BCG on overall mortality, immune and atopic conditions including asthma, and incidence of respiratory tract infections; its nonspecific protection against nonrelated pathogens; and its protective effects that are apparent shortly after immunization and sustained for at least 1 year [2,25], we hypothesize that some of the variations in mortality among children under 5 years of age, in the incidence of TB, and in severe, moderate, and mild morbidity among children under 5 years of age might be partially explained by a reduction in the coverage rate of BCG since 2016. The reduction in the coverage rate of BCG, which resulted from only high-risk children receiving BCG at birth, might have prevented nonvaccinated children from developing trained immunological responses as effectively as BCG-vaccinated children.

We aim to investigate the incidence of the specific and nonspecific effects of BCG by comparing the incidences of TB disease and infection; mild, moderate, and severe morbidity; and mortality in the first 5 years of life among children born in Portugal between 2010 and 2021, according to their BCG status.

This population-based historical (retrospective) birth cohort study will compare the incidence of all-cause mortality (including due to TB disease); TB disease; and severe, moderate, and mild morbidity between BCG-vaccinated and nonvaccinated children born between July 1, 2010, and June 30, 2021.

This study will investigate the specific and nonspecific effects of BCG based on the shift in the BCG vaccination policy that occurred in Portugal in July 2016, when a consistently low incidence of TB led to the revision of the National Immunization Plan. Since then, BCG has been administered only to newborns considered to be at higher risk of TB infection, thus creating an opportunity to compare BCG-vaccinated and nonvaccinated children.

The approach adopted by Portuguese public health authorities to define the high risk for TB was based on geographical risk, with, for instance, all children born in some parishes in the metropolitan areas of Lisbon and Porto being offered BCG independent of socioeconomic status, country of origin, or vulnerability. We believe that this type of risk definition has guaranteed enough variability between BCG-vaccinated and nonvaccinated children, thus allowing comparisons of health outcomes between exposed and nonexposed children.

The study will include children under 5 years of age born and registered in Portugal between July 1, 2010, and June 30, 2021. We have chosen this period because it will allow us to (1) compare the cohorts before and after the change in the BCG immunization strategy (Figure 1); (2) characterize children not vaccinated with BCG and their health outcomes (when universal BCG at birth was recommended); and (3) compare children exposed and not exposed to BCG after the 2016 change in the BCG vaccination strategy.

The study will be based solely on secondary data. The data sources are the National Health Service (NHS) user registry (RNU), Death Certificate Information System (SICO), vaccination registry, communicable diseases surveillance system (Sistema Nacional de Vigilancia Epidemiologica [SINAVE]), surveillance information system for TB (Svig-TB), DRG information system for hospital admissions and visits to the emergency department (Base de Dados de Morbilidade Hospitalar [BDMH]), and primary health care (PHC) information system (SClínico) of children born between July 1, 2010, and June 30, 2021.

The birth registry database (RNU) contains data on gender, age, PHC center where the child is to be followed up, nationality, and place of residence. The SICO database provides data on the cause and date of death. The vaccination registry includes data on the vaccines administered, date of vaccination, and place of residence. The BDMH database gathers data on hospital admissions and emergency department visits to NHS hospitals, including the date of admission and discharge, diagnosis, type of admission, motive for admission, weight at birth, DRG code, major diagnostic category (MDC) code, complications of pregnancy and delivery, and type and place of delivery. Data from admissions at private hospitals are not included. The SClínico database contains data on socioeconomic status and visits (including motive and diagnosis) to NHS PHC units. Svig-TB is a database for notification and follow-up of TB cases, and SINAVE is the general surveillance information system for communicable diseases, including TB, and regarding TB, it contains data on symptoms at presentation, date of diagnosis, type of treatment, other diseases before TB, risk groups, and outcomes of the infection.

Exposure is defined as having received the BCG vaccine during the first year of life and is measured using the vaccination registry. Variables include BCG (yes or no, and number of doses) and time from birth to BCG (in days).

Through data linkage, we will create a single database that links data from the birth registry, SICO, vaccination registry, SINAVE, Svig-TB, BDMH, and PHC database (SClínico) to reconstruct chronological sequences of morbidity and mortality events from birth until the completion of 5 years of life or the end of follow-up (for the 2018 cohort onwards).

In each of the databases, we will identify the data needed and whether a unique identifier (UI) exists and can be provided with the data set. After having access to the requested data and in case a UI is provided (and common to all databases), we will combine information based on that UI.

In case a UI is not provided, we will link data from several data sets using a range of proxy identifiers (eg, date and place of birth, sex, and place of residence) to identify probable matches.

The type of data linkage method to be used will depend on the type and quality of the linkage variables available in the data sets. However, we anticipate having to use a combination of deterministic and probabilistic methods. Before applying linkage methods, we will clean and standardize the data, thus identifying and removing errors and inconsistencies. Given the expected size of the data sets, we will then select sets of block attributes (eg, sex, date of birth, and initials) and compare record pairs with the same matched attributes within blocks. Subsequently, record pairs will be compared for each linkage variable, and an agreement score will be computed. This score will be used to weigh the probability of a record pair belonging to the same child [26,27].

After obtaining the final database, we will compute person-time incidence rates for primary outcomes in BCG-vaccinated and nonvaccinated children. Using the Kaplan-Meier method, we will compute and compare the probability of surviving the first and fifth years of life or of not having a hospitalization, emergency department visit, or mild morbidity episode during the follow-up period according to exposure. Additionally, we will use a log-rank test for assessing differences in survival rates between exposed and nonexposed children and hazard ratios (and corresponding CIs) for quantifying the differences. To explore the effects of several variables on the survival outcomes, we will use proportional hazards regression analysis. If missing data are below 5%, we will use complete case analysis. Otherwise, the frequency and patterns of missing data will be analyzed, and if appropriate, we will use multiple imputation techniques.

Confounders to be measured and controlled for are sex, health unit, completeness and timeliness of vaccination status and scheme, socioeconomic status, and BCG strain (over the years, administered BCG has been provided by different producers, with different strains).

This study has received ethical approval from the Ethics Committee of Instituto de Higiene e Medicina Tropical (IHMT) – Instituto de Tecnologia Química e Biológica António Xavier (ITQB) NOVA – NOVA School of Law (NSL) – Instituto Gulbenkian de Ciência (IGC) (11.23) [28].

The project protocol was submitted to the Ethics Committee of IHMT – ITQB NOVA (11.23), which issued a conditional authorization in July 2022 [29] that became definitive in December 2023.

If the change in the BCG strategy in 2016 is proven to influence the health outcomes of children in their first year of life and during the first 5 years of life, it would further contribute to evidence that BCG primes the immune system against unrelated pathogens and even other health conditions, strengthening the arguments regarding the nonspecific effects of BCG. Additionally, it would provide evidence of the impact of the high-risk strategy for BCG vaccination and inform future decisions regarding BCG vaccination for children in Portugal and other high-income countries. The results will contribute to the accumulating body of knowledge and might have relevance to guide global BCG vaccination policy.

The use of data linkage can also contribute to a swifter mechanism to use available health data to conduct population-based studies and inform policies. The project will demonstrate the feasibility of conducting large-scale epidemiological register-based vaccine studies in Portugal.

The main limitation of this project involves the use of secondary data collected for purposes other than research. The RNU database, which provides the basis for creating a UI, only includes children who have a user number for the NHS. The user number is assigned to each person to identify them when accessing the services of the public health care units of the NHS. People with Portuguese citizenship obtain their user number automatically when they apply for a Citizen’s Card. Foreigners with a residence or stay permit in Portugal have to apply for an NHS user number. Most but not all infants are registered shortly after birth.

In August 2016, the Portuguese government created the program “Nascer Utente” (translated to “Born as a User”) that allowed for immediate registration in the RNU database, assigning the respective user number. This process is conducted by the hospital (public and private) where the child is born, immediately after birth [30]. We therefore expect a bias toward children born after this date in the RNU database.

The electronic registry of vaccines was started in 2017, and all data before this point had to be back introduced in the database, which could have increased the number of typing errors and other issues. We can again have a bias toward the most recent years in terms of the quality of data.

The number of deaths of children under 5 years of age might be underestimated since very young children tend not to have an NHS user number, especially before August 2016. In this case, we can also expect to have a bias in the age of children who died between July 1, 2010, and June 30, 2021, toward older ones.

The BDMH database only pertains to public hospitals. It does not include data from emergency department visits or hospital admissions in the private sector. The same is true for SClínico, which contains data from PHC centers in the NHS. Children using the private sector are not captured by this database. Additionally, the case mix (severity) of admissions in public hospitals is usually greater than that in private hospitals, which might introduce a bias toward higher case fatality rates in hospital admissions.

In 2020, 32% of Portuguese people had voluntary health insurance (in 2010, the proportion was less than 20%), and the proportion is dependent on household income, with those in higher income groups showing a higher probability of having voluntary health insurance. As such, we expect a bias toward lower income groups, especially in the last years of the study period [31].

This is one of the first studies conducted in Portugal linking several databases. During the COVID-19 pandemic, several data linkage studies were conducted [32,33].

To our knowledge, this is the first population-based study that gathers data on mild, moderate, and severe morbidity and mortality among children under 5 years of age and addresses the nonspecific effects of BCG.

In Portugal, this is the first study to assess the impact of the BCG vaccination policy change on the health of children.

The results will contribute to the accumulating body of knowledge and might have relevance to guide global BCG vaccination policy. The use of data linkage can also contribute to a swifter mechanism to use available health data to conduct population-based studies and inform policy decision-making.