SARS-CoV-2, a novel RNA virus from the Coronaviridae family, was first identified in Wuhan, China, in December 2019. Closely related to other pandemic-prone viruses such as SARS-CoV and MERS-CoV, it quickly spread worldwide. In March 2020, the World Health Organization (WHO) declared the COVID-19 pandemic a global health emergency. By the end of that year, over 183 million confirmed cases had been reported, underscoring its high transmissibility and substantial impact on public health systems (1, 2).

.Epidemiological and genomic analyses indicate that SARS-CoV-2 originated by zoonotic spillover, with subsequent mutations—particularly in the spike protein—leading to distinct viral variants. By late 2021, strains such as Alpha, Beta, Gamma, and Omicron had demonstrated increased transmissibility and partial immune escape, altered transmission dynamics, and prompted adjustments in surveillance and control strategies (3). Morbidity and mortality varied across regions, influenced by population density, healthcare capacity, and the timing of interventions. Urban areas experienced rapid early spread, with higher mortality in older adults and individuals with comorbidities during periods of limited treatment options and overwhelmed health systems, particularly during the initial phase when therapeutic options were limited, ICU capacity was strained, and clinical protocols were still evolving. As of the cutoff date for the study, 193.31 million confirmed COVID-19 cases and 3.04 million deaths had been reported globally. In Ecuador, between 2020 and 2024, 864,811 confirmed cases and 10,334 deaths were recorded (4–7). Over time, advances in clinical management reduced the severity and lethality of cases. However, excess mortality persisted, especially in low-and middle-income countries where vaccine access was limited, and health systems faced repeated surges (1, 3).

Vaccination played a pivotal role in controlling the COVID-19 pandemic. Accelerated research and emergency use authorizations enabled early access, significantly reducing SARS-CoV-2 transmission and mortality by 57% overall, particularly among high-risk populations. The BioNTech-Pfizer vaccine, the first deployed globally, demonstrated 95% efficacy, followed by AstraZeneca (90%) and Sinovac (79.3%) (8). The full impact became evident after the mass rollout: by the end of 2021, vaccination had prevented an estimated 19.8 million deaths and markedly reduced infections, hospitalizations, and severe cases (9).

Meta-analyses confirmed transmission reductions exceeding 40%, ICU admissions declining by more than 90%, and case fatality rates falling by nearly 70%. Therefore, global mortality from COVID-19 has steadily declined, from 4.3 million deaths in 2021 to 81,753 in 2024—a dramatic downward trend. In Ecuador, the 2021 vaccination campaign resulted in a greater than 40% reduction in cases; by 2022, peak infections had remained below 10,000, and mortality had become negligible. In countries with coverage of 30% or more, case numbers decreased by more than 40%. These outcomes highlight the effectiveness of sustained immunization in reducing disease burden and preventing future public health crises. In Ecuador, the first case of COVID-19 was reported in February 2020; within a few weeks, infections arose from 6–8 cases to over 60 as a consequence, national isolation measures were implemented as a control strategy, reaching 400 cases per month for 2020, 1,000 cases during 2021 and a mortality rate of 300 cases; showing the same pattern as worldwide showing a representative decreasing in it during 2022 and 2023 while in 2024 were 250 reported cases and only five deaths (5, 10–12).

Consequently, in March 2023, the WHO declared that COVID-19 was no longer a pandemic but a public health event of international concern. The organization urged countries worldwide to maintain surveillance and vaccination efforts to control outbreaks or potential epidemics of this disease, as it is not yet able to produce sustainable and long-lasting protection against SARS-CoV-2 infection; hence, the population remains susceptible to COVID-19. Furthermore, the seasonal behavior, the emergence of new SARS-CoV-2 variants, and the co-circulation with other respiratory viruses are also key considerations (13).

Five years into the pandemic, it is now possible to explore the potential seasonality or endemicity of SARS-CoV-2. However, classifying COVID-19 remains a complex and region-dependent process. According to the Centers for Disease Control and Prevention (CDC), such classification is challenging due to the emergence of new variants, fluctuating attack rates, and variable lethality. These factors contribute to an unpredictable transmission pattern, complicating reliable analysis and requiring robust data collection and sustained surveillance. Thus, SARS-CoV-2 has not yet demonstrated apparent seasonality and may continue to cause localized epidemics or even future pandemics (7, 14–16).

Understanding the seasonal dynamics of infectious diseases is essential for guiding vaccine strategies, including the optimal timing of booster dose deployment. It also addresses this public health priority and determines whether its trend would be adjusted in relation to other circulating respiratory viruses. The study aimed to analyze the seasonality of COVID-19 using time series methods, assess its potential transition to an endemic pattern, and develop a case projection model based on the ARIMA (p, d, q)(P, D, Q)[m] method in Ecuador from 2020 to 2024.

During the data collection period from 2020 to 2024, annual increases in reported SARS-CoV-2 cases were observed until 2022, after which infection rates declined. The Kruskal–Wallis test confirmed significant differences in case counts across years (p = 0.0001). The highest case burden occurred in 2022, when weekly reports exceeded 52,000. In contrast, 2024 showed markedly lower case levels, averaging 349.8 per week (90% CI: 240.1–459.6) (Table 1).

Endemic channel analysis across the 5 years revealed consistent patterns. The highest incidence was concentrated in the early epidemiological weeks of each year, while the lowest case counts typically occurred between weeks 40 and 47, followed by a resurgence around week 49. Within the epidemic zone, weekly cases exceeded 50,000, whereas during weeks 28 to 30, they dropped to approximately 2,000. The success zone was narrow, with a maximum of only 450 cases between weeks 34 and 39 (Figure 1).

Seasonality analysis of the raw case data revealed seasonal patterns, with variance equal to zero. After transformation into a stationary time series, the mean and variance approached zero, indicating the presence of seasonality in the adjusted data. An exception occurred in 2022, when a marked increase in reported cases produced values outside the expected range. Autocorrelation analysis of the stationary series revealed a correlation among reported cases that decreased progressively with increasing lag, approaching the characteristics of white noise (Figure 2).

The time series analysis of SARS-CoV-2 revealed a marked decline in infection rates beginning in 2023, stabilizing into a linear trend by late 2024. Since the onset of the pandemic in Ecuador, recurrent peaks have consistently occurred at the start of each year, particularly during epidemiological weeks 1 to 8, followed by a smaller peak between weeks 20 and 28 and a gradual decline toward the end of the year. This secular seasonal pattern remained stable across the study period (Figure 3; Table 2). Model comparisons indicated that ETS produced lower error values than ARIMA (0,2,1)(0,0,1)[52], while STL + ETS further reduced errors but exhibited unusually high MAPE values. The ARIMA (0,2,1)(0,0,1)[52] model achieved in-sample performance comparable to ETS, but external validation revealed substantially higher error metrics (Tables 3, 4; Figures 4–6).

Residual diagnostics revealed significant deviations from normality, both in the original dataset and in the residuals of the fitted model, even after applying a base-10 logarithmic transformation. This was confirmed by multiple normality tests, including the Shapiro–Wilk, Kolmogorov–Smirnov, Anderson–Darling, and Jarque–Bera tests, all of which yielded p-values < 0.001 (Table 3). Despite this, the Ljung–Box tests for residual autocorrelation were non-significant (p = 0.4611), as shown in Table 5, indicating that the residuals behaved like white noise and suggesting the model adequately captured the temporal structure of the series.

The fitted ARIMA (0,2,1)(0,0,1)[52] model successfully replicated the general pattern of the epidemic curve in the training set, capturing moderate fluctuations while underestimating extreme peaks. Forecast outputs preserved the observed cyclical dynamics with broad 80 and 95% confidence intervals. However, out-of-sample validation results (Table 6; Figure 7) revealed higher error metrics, reflecting the challenges in predicting periods of sudden epidemiological change (Table 7).

By the end of 2024, more than 777 million COVID-19 cases had been reported worldwide, with a notable decline in incidence following global peaks in late 2022. Ecuador mirrored this trajectory, reporting its highest national incidence in 2022 with 52,300 cases, followed by a steady reduction through 2024. These national patterns are consistent with WHO reports, which document global declines during the same period (3, 24–29).

Time series analysis of Ecuadorian data between 2020 and 2024 revealed consistent cyclical patterns, characterized by dual seasonal peaks. The ARIMA (0,2,1)(0,0,1)[52] model effectively captured these dynamics, detecting a primary surge from January to March and a secondary rise from May to July. Endemic channel analysis reinforced these findings, demonstrating that transmission is becoming more regular and predictable. The persistence of this dual-peak structure over five consecutive years suggests that SARS-CoV-2 is undergoing seasonal adaptation, likely influenced by the emergence of new variants, increasing vaccine coverage, and progressive development of population-level immunity (27, 30–32).

According to the epidemiology of transmissible diseases, the temporal series would represent seasonal variations as regular fluctuations each year, secular patterns with multiple peaks during the year, or secular trends that persist over a long period of time. In our study of SARS-CoV-2 in Ecuador (2020–2024), even though the ARIMA (0,2,1)(0,0,1)[52] led to assuming a weekly seasonality of 52 weeks per year, a practice nowadays accepted for practical epidemiological analysis, a secular pattern was highlighted by a progressive and maintained decrease in the reported COVID-19 cases during 2021. This mentioned behavior would be explained by the massive vaccination campaigns, the progressive acquired immunity (natural or vaccine-induced), and the appearance of other SARS-CoV-2 strains that showed less virulence and adaptation, transforming the dynamic of transmission through the country. Therefore, the established model effectively captures the secular trend with transient epidemic cycles, in contrast to a strong and marked annual pattern (33–36).

These seasonal dynamics parallel the behavior of influenza viruses, which typically peak in winter in temperate regions and display biannual peaks in tropical and subtropical settings. Influenza A (H3N2) usually peaks earlier in the season, while influenza B often drives secondary waves later in the year (37–39). Although SARS-CoV-2 lacks the rapid antigenic drift characteristic of influenza, its increasingly regular transmission cycles suggest meaningful epidemiological parallels. Recognizing these similarities can strengthen surveillance and guide seasonally timed vaccination strategies (40–44).

Comparison with SARS-CoV-1 underscores differences in long-term dynamics. While both viruses caused sharp surges after their emergence, SARS-CoV-1 outbreaks were contained by 2005 and remained geographically limited. In contrast, SARS-CoV-2 has continued to circulate globally. Its higher transmissibility, supported by spike protein mutations that enhance ACE2 receptor binding and facilitate immune evasion, together with factors such as international connectivity and urban density, contributed to its broader and more persistent spread (24, 26, 45–47).

Endemic channel analysis confirmed that in Ecuador, case counts exceeded the endemic threshold during the first epidemiological weeks of each year, followed by declines into the endemic zone by mid-year. Yet, the subnational assessment was limited by data constraints. Heterogeneity in healthcare access, intervention timing, and population density across provinces may influence local transmission patterns and should be considered in future analyses (27, 29, 48–50). Vaccination campaigns launched in 2021 substantially reduced transmission, but high case numbers persisted into 2022, reflecting delays in vaccine coverage and the emergence of novel variants. The decline observed from 2023 onward highlights the impact of population-level immunity but also suggests that Ecuador transitioned more slowly than other settings (9, 13, 51, 52).

Forecasting results provided additional insights. ARIMA models were accurate in the short term, predicting early 2025 peaks of 240 to 250 cases per week, followed by rapid declines to near-zero levels. However, forecasts beyond 1 year consistently projected zero incidence, underscoring ARIMA’s reduced sensitivity to long-range uncertainty and inability to capture the stochastic nature of epidemics. Similar findings have been reported in Brazil and Romania, where ARIMA models performed well over short horizons but failed at longer ones (16, 31, 53–55). ARIMA validation confirmed stronger long-term performance. Diagnostic tests showed independent residuals consistent with white noise, confirming model adequacy despite the absence of normally distributed errors, which is common in epidemiological series (56–59).

Residual diagnostic tests revealed that the fitted ARIMA (0,2,1)(0,0,1)[52] model did not fully meet the assumption of normality, even after applying a logarithmic transformation to the original series. All normality tests returned statistically significant results (p < 0.001), indicating deviations from a normal distribution in the residuals. However, such deviations are often observed in epidemiological time series of infectious diseases, where residual patterns may be influenced by factors such as reporting lags, outbreak clusters, or sudden epidemic peaks. In contrast, the Ljung-Box test yielded non-significant results (p = 0.377 and p = 0.461), suggesting no autocorrelation and supporting the assumption of independence, which aligns with white noise. The model effectively captured the overall downward trend of SARS-CoV-2 cases and represented the major epidemic waves, although it tended to underestimate extreme peaks.

Out-of-sample validation, using one-third of the data, showed moderate increases in forecast error metrics (MAE, RMSE, MAPE), which is expected given the heightened variability during post-peak epidemic phases. Despite the observed deviations from normality, the independence of the residuals and the model’s ability to replicate seasonal and secular trends support its relevance and validity in epidemiological applications (56–59).

Limitations remain, as case reporting in Ecuador has been affected by variability in testing capacity, evolving diagnostic criteria, and incomplete records, particularly in the early pandemic years. These issues limit the reliability of complex forecasting models. Machine learning approaches, such as LSTM and recurrent neural networks, can capture nonlinearities; however, their effectiveness is constrained by small, noisy datasets and a lack of transparency (27, 30–32, 60). By contrast, ARIMA-based approaches offer greater interpretability, allowing for decomposition into trend and seasonal components, which is essential for epidemiological applications in data-limited contexts (61–63).

Despite reduced incidence across Latin America, SARS-CoV-2 continues to pose challenges due to evolving variants and seasonal surges. The recurrent dual-peak pattern observed in Ecuador supports a gradual shift toward endemicity, though the process remains incomplete. Continued transmission variability highlights the need for caution in defining SARS-CoV-2 as entirely endemic (24, 26, 45–47).

Between 2020 and 2024, SARS-CoV-2 transmission in Ecuador exhibited increasingly regular seasonal peaks, suggesting a progressive transition toward endemicity. Nevertheless, uncertainties related to data quality, viral evolution, and regional disparities require cautious interpretation. Vaccination significantly reduced incidence but is insufficient alone to stabilize transmission. Enhanced surveillance during high-risk seasonal periods is necessary to prevent resurgence. Future research should integrate high-resolution, real-time data and explore hybrid models that combine the transparency of traditional time series with the adaptability of explainable machine learning. Continued monitoring and multidisciplinary approaches will be essential to anticipate transmission dynamics and guide effective public health responses.