This was a single-center, descriptive, prospective, non-interventional study conducted in the Mendoza province in Argentina, between August 22, 2008, and January 12, 2023, with detailed descriptions of earlier assessments available elsewhere [3,4,5,6]. All included participants received the first dose of HAV vaccine (Avaxim® 80U Pediatric) in 2007. There were two groups of participants based on the number of HAV vaccine(s) received prior to inclusion; those who had received one dose before inclusion and those who had received two doses.

The study was undertaken in compliance with the International Conference on Harmonisation Good Clinical Practice guidelines and the principles of the Declaration of Helsinki. Informed consent was obtained from the participants’ parents (or legal guardians) before any study procedures were performed, as was informed assent from participants from the age of 7 years. The protocol and any amendments were approved by the Ministry of Health in the Mendoza province and the independent ethics committee/institutional review board at the central hospital of Mendoza (VV-TMF-562262 AR 001 IRB or IEC Approval 2008-07-03 | VV-TMF-562262 | 1.0, subsequently undated as act numbers 8/2008, 5/2011, 1/2013, and 14/2017), and notified by the provincial and national health authorities. Due to changes in regulations at the provincial and national levels, the extension of the study to year 15 was approved by the ethics committee at the provincial level and registered with the Mendoza Province Health Authority following approval by the hospital’s ethics committee (Act Number 14/2017).

The participants were followed annually through to year 5, and then at years 7 and 10 for medical examinations, identification of suspected HAV cases (if any) among family members, and socioeconomic data. Additional follow-ups were conducted at years 12 and 15 for medical examinations and the identification of suspected HAV cases. To evaluate the persistence of the anti-HAV antibody response, blood samples were collected at study entry (year 1: for example in 2008, 1 year after receipt of the inactivated HAV vaccine as part of the national immunization schedule), followed by additional samples at years 2, 3, 4, 5, 7, 10, 12, and 15. Participants who had received a single dose of the HAV vaccine and had anti-HAV antibody levels below the seroprotective threshold (see next paragraph for assay-specific seroprotective thresholds used) during the study follow-up were offered a booster vaccination, and they comprised the third study group.

Anti-HAV antibody concentrations were measured using microparticle enzyme immunoassay (MEIA) up to year 5 as previously described [4, 6], with seroprotection defined as an anti-HAV antibody concentration of ≥ 10 mIU/ml. After the MEIA kit was discontinued, an electrochemiluminescence immunoassay (ECLIA) was used instead for anti-HAV antibody assessment through to year 10, as previously described, with seroprotection defined as an anti-HAV antibody concentration of ≥ 3 mIU/ml. Blood samples obtained at year 5 were reanalyzed using ECLIA to compare the results with those previously analyzed using MEIA. In the current analysis, anti-HAV antibody concentrations were measured at year 12 using the ECLIA kit, and after it was subsequently discontinued, the Atellica kit was used at the year 15 follow-up. The seroprotection rate was defined as the proportion with anti-HAV antibody concentrations ≥ 10 mIU/ml, as measured with Atellica. Year 12 blood samples were also reanalyzed using Atellica to compare the results with those previously analyzed using ECLIA.

The Atellica® IM aHAVT (ref. 10995446) assay is a fully automated, competitive chemiluminescent immunoassay for the detection of total anti-HAV antibodies and is described in detail elsewhere [8]. Quantitative anti-HAV concentrations were determined by comparison with a serial dilution of the World Health Organization Second International Standard for Anti-Hepatitis A Immunoglobulin (NIBSC, code:97/646). The MEIA and ECLIA assays were described in an earlier publication of this study [5].

Seroprotection rates—the percentage of participants with anti-HAV antibodies ≥ 10 mIU/ml (measured using MEIA from year 1 to year 5, inclusive), ≥ 3 mIU/ml (measured using ECLIA in years 5, 7, 10, and 12), or ≥ 10 mIU/ml (measured using Atellica in years 12 and 15)—were calculated with 95% confidence intervals (CIs) using the Clopper–Pearson exact binomial method, as described by Newcombe [9].

Participants included in this analysis were those with at least two anti-HAV antibody results available, i.e., at year 1 (after first vaccination) and at least one other timepoint. For participants boosted during the study period, only data points prior to the booster vaccination were considered in the analysis. Anti-HAV geometric mean concentrations (GMCs) and the number of participants per visit and group (one- or two-dose group) are presented in Table 1. Over time, GMCs generally decreased, but an increase between some later timepoints, such as between year 5 and year 7, reflects a change in the assay used since the previous timepoint (Supplementary Fig. S1); these variations between tests were taken into account. A high level of consistency was observed between results obtained using the three assays (adjusted R2 > 90%) (Supplementary Fig. S2); nonetheless, rescaling was required to enable comparisons between assays. Detailed information on the rescaling methodology is included in the Supplemental Materials (Section S1: Supplementary methods, Rescaling of antibody titers).

We used a hierarchical model to estimate antibody decay (participant- and group-specific decay) that accounted for differences in the results obtained according to the assay used (MEIA, ECLIA, or Atellica) and the immunological boosting effects due to virus exposure (natural boosting effect). Data for both groups (one or two vaccine doses) were fitted simultaneously using log-transformed titers, and antibody decay was estimated with the most recent assay used to express these outcomes (Atellica) after rescaling all data to MEIA (the assay with the largest number of observed data points).

Four different model specifications were considered for antibody decay: (1) a one-change-timepoint linear model without natural boosting effect; (2) a one-change-timepoint linear model with natural boosting effect; (3) a two-change-timepoint linear model with natural boosting effect; and (4) a log-logistic model with natural boosting effect.

For the piecewise linear models, we considered a first change-timepoint at year 5 and tested a model with a second change-timepoint at year 10 based on findings from the previous analysis. We also evaluated the log-logistic model because it accounted for the switch from rapid to slow antibody decay over time in a non-linear, smooth transition that avoided the abrupt changes characterized by the piecewise linear models. The models were fitted using a Bayesian approach with the R® package rstan [10]. Model fits were obtained using leave-one-out cross-validation (LOO-CV) and were compared using expected log pointwise predictive density (ELPD) methods from the R® package loo (efficient approximate leave-one-out cross-validation for fitted Bayesian models) [11]. For more information, see Models 1–4 in the Supplemental Materials.

The results of these estimations were further used to predict long-term antibody persistence through 30 and 40 years after vaccination. The predicted evolution of GMCs and seroprotection rates was expressed using Atellica titers (≥ 10 mIU/ml), the assay used for the most recent assessments.