Abstract
Introduction:
The trajectory of antibody levels following hepatitis B vaccination (HepB) at various dosages has rarely been explored. This study compares three distinct longitudinal clustering methods to analyse the development of antibodies following HepB to evaluate antibody titres before vaccination, after the first dose, and after the second dose to assess the effectiveness of these clustering techniques.
Methods
The hepatitis B antibody (HBsAb) titres of 312 freshmen at Fujian Medical University in China were analysed to identify clusters in which the antibody level changed over time. Antibody levels were measured at three time points: prevaccination, postfirst dose, and postsecond dose. K-means cluster analysis and latent growth mixture modelling (LGMM) methods were conducted via the R packages kml and lcmm, respectively. Additionally, group-based trajectory modelling (GBTM) was performed with the Stata plugin traj.
Results
K-means clustering and latent growth mixture modelling (LGMM) classified antibody development trajectories into three distinct clusters: high, medium, and low antibody levels. In contrast, the group-based trajectory modelling (GBTM) method identified only two clusters, corresponding to high and low antibody levels. The K-means and LGMM methods demonstrated the highest similarity in cluster shapes and provided a relatively better fit to the data. Conversely, the GBTM method produced more distinct trajectory shapes but did not align as well with the observed data.
Conclusion
After the trajectory analysis packages kml, lcmm, and traj were compared via HBsAb data, it was determined that the kml package offered the most appropriate clustering for antibody data. This finding may help inform strategies to optimize herd immunization.