Epidemiological characteristics and dynamic transmissions of COVID-19 pandemics in Chinese mainland: A trajectory clustering perspective analysis

Background

The corona virus disease 2019 (COVID-19) pandemic has spread to more than 210 countries and regions around the world, with different characteristics recorded depending on the location. A systematic summarization of COVID-19 outbreaks that occurred during the “dynamic zero-COVID” policy period in Chinese mainland had not been previously conducted. In-depth mining of the big data from the past two years of the COVID-19 pandemics must be performed to clarify their epidemiological characteristics and dynamic transmissions.

Methods

Trajectory clustering was used to group epidemic and time-varying reproduction number (Rt) curves of mass outbreaks into different models and reveal the epidemiological characteristics and dynamic transmissions of COVID-19. For the selected single-peak epidemic curves, we constructed a peak-point judgment model based on the dynamic slope and adopted a single-peak fitting model to identify the key time points and peak parameters. Finally, we developed an extreme gradient boosting-based prediction model for peak infection cases based on the total number of infections on the first 3, 5, and 7 days of the initial average incubation period.

Results

(1) A total of 7 52298 cases, including 587 outbreaks in 251 cities in Chinese mainland between June 11, 2020, and June 29, 2022, were collected, and the first wave of COVID-19 outbreaks was excluded. Excluding the Shanghai outbreak in 2022, the 586 remaining outbreaks resulted in 1 25425 infections, with an infection rate of 4.21 per 1 00000 individuals. The number of outbreaks varied based on location, season, and temperature.

(2) Trajectory clustering analysis showed that 77 epidemic curves were divided into four patterns, which were dominated by two single-peak clustering patterns (63.3%). A total of 77 Rt curves were grouped into seven patterns, with the leading patterns including four downward dynamic transmission patterns (74.03%). These curves revealed that the interval from peak to the point where the Rt value dropped below 1 was approximately 5 days.

(3) The peak-point judgment model achieved a better result in the area under the curve (0.96, 95% confidence interval = 0.90–1.00). The single-peak fitting results on the epidemic curves indicated that the interval from the slow-growth point to the sharp-decline point was approximately 4–6 days in more than 50% of mass outbreaks.

(4) The peak-infection-case prediction model exhibited the superior clustering results of epidemic and Rt curves compared with the findings without grouping.

Conclusion

Overall, our findings suggest the variation in the infection rates during the “dynamic zero-COVID” policy period based on the geographic division, level of economic development, seasonal division, and temperature. Trajectory clustering can be a useful tool for discovering epidemiological characteristics and dynamic transmissions, judging peak points, and predicting peak infection cases using different patterns.