Contribution of Two-Dose Vaccination Toward the Reduction of COVID-19 Cases, ICU Hospitalizations and Deaths in Chile Assessed Through Explanatory Generalized Additive Models for Location, Scale, and Shape

Abstract

Objectives: To assess the impact of the initial two-dose-schedule mass vaccination campaign in Chile toward reducing adverse epidemiological outcomes due to SARS-CoV-2 infection.

Methods: Publicly available epidemiological data ranging from 3 February 2021 to 30 September 2021 were used to construct GAMLSS models that explain the beneficial effect of up to two doses of vaccination on the following COVID-19-related outcomes: new cases per day, daily active cases, daily occupied ICU beds and daily deaths.

Results: Administered first and second vaccine doses, and the statistical interaction between the two, are strong, statistically significant predictors for COVID-19-related new cases per day (R² = 0.847), daily active cases (R² = 0.903), ICU hospitalizations (R² = 0.767), and deaths (R² = 0.827).

Conclusion: Our models stress the importance of completing vaccination schedules to reduce the adverse outcomes during the pandemic. Future work will continue to assess the influence of vaccines, including booster doses, as the pandemic progresses, and new variants emerge.

Policy implications: This work highlights the importance of attaining full (two-dose) vaccination status and reinforces the notion that a second dose provides increased non-additive protection. The trends we observed may also support the inclusion of booster doses in vaccination plans. These insights could contribute to guiding other countries in their vaccination campaigns.

Keywords: COVID-19; GAMLSS models; ICU hospitalizations; explanatory models; vaccination.

Figure 1

GAMLSS models employ the number of vaccine doses and the statistical interaction term between doses as key predictors to explain epidemiological outcomes of interest. In the graphs, each point corresponds to daily count data per 100,000 inhabitants, curves represents the model fit, and the shaded area is the standard error of the model. Each letter denotes a pair of models for an outcome of interest, where the graph on the left shows our best model, and the graph on the right shows the same model after removal of the interaction term between first and second doses. (A) GAMLSS models for the number of new cases per day. (B) GAMLSS models for the number of daily actives cases. (C) GAMLSS models for the number of occupied ICU beds. (D) Explanatory models for the number of the weekly moving average of deaths. Refer to Supplementary Tables 1A–4A for the predictors used in the best-performing models, to Supplementary Tables 1B–4B for a comparison of model diagnostics between the best models and models with removed predictors, and to Supplementary Tables 1C–4C for model parameters.

Figure 2

GAMLSS models explain the number of occupied ICU beds by age range. In graphs (A–C), the image on the left corresponds to the best constructed model and the image on the right corresponds to a model that incorporates the same predictors, but without the interaction between the first and second doses. In graphs (D,E), the image on the left is the best model and the one on the right is the same model with the interaction term between the first and second doses added. Each point corresponds to ICU hospitalizations per 100,000 inhabitants in a week, the cyan curve is the prediction given by the GAMLSS model, and the shaded area is the standard error of the model. (A) Age group under 39 years old, no significant difference between the two models. (B) Age group between 40 and 49 years old, significant difference between the two models. (C) Age group between 50 and 59 years old, significant difference between the two models. (D) Age group between 60 and 69 years old, no significant difference between the two models. (E) Age group over 70 years old, significant difference between the two models. Refer to Supplementary Tables 5, 6 for the predictors used in the best-performing models, to Supplementary Table 6 for a comparison of model diagnostics between the best models and models without the interaction term, and to Supplementary Tables 7A–E for detailed model parameters.