Abstract
We present different data analytic methodologies that have been applied in order to understand the evolution of the first wave of the Coronavirus disease 2019 in the Republic of Cyprus and the effect of different intervention measures that have been taken by the government. Change point detection has been used in order to estimate the number and locations of changes in the behaviour of the collected data. Count time series methods have been employed to provide short term projections and a number of various compartmental models have been fitted to the data providing with long term projections on the pandemic's evolution and allowing for the estimation of the effective reproduction number.
Introduction
The emergence of COVID-19 in early 2020 presented an unprecedented challenge for global health systems. In the Republic of Cyprus, the absence of a dedicated epidemiological modeling unit necessitated the formation of a multidisciplinary task force comprising epidemiologists, statisticians, and data scientists. This paper documents the statistical and mathematical frameworks developed to monitor the pandemic's first wave, evaluate the efficacy of government interventions, and provide actionable intelligence for public health policy.
Methodology
The research utilized an ensemble of three primary methodologies to ensure robust results:
- Change-Point Analysis: The Isolate-Detect (ID) methodology was applied to daily incidence data to identify structural shifts in the spread of the virus. This allowed the team to segment the pandemic into homogeneous phases, mirroring the timing of government lockdowns.
- Count Time Series: Log-linear Poisson models with feedback were used for short-term forecasting. These models were particularly effective at identifying "additive outliers," representing localized clusters that deviated from broader trends.
- Compartmental SEIR Models: Four distinct models were developed, ranging from a single-population SEIR model accounting for asymptomatic cases to a complex 7-state model ($S, P, E, I, Q, R, D$) including a "quarantined" state. Additionally, a metapopulation model incorporated human mobility data between Cyprus's five main districts (Nicosia, Limassol, Larnaca, Paphos, and Ammochostos).
Data and Observations
The primary dataset spanned March 1 to May 31, 2020, provided by the Unit for Surveillance and Control of Communicable Diseases (USCCD) of the Cyprus Ministry of Health. A critical observation was the temporal lag between symptom onset and case reporting.
As shown in the figure above, the peak of symptom onset preceded the peak of reported cases (sampling), highlighting the inherent delay in detection that models must account for. By the end of May, Cyprus had recorded 952 cases with a median age of 45.
Results and Model Performance
The analysis revealed that the effective reproduction number ($R_t$) peaked at approximately 4.47 at the start of the outbreak but plummeted to 0.38 by mid-April following strict lockdown measures. The detected change points (March 23, April 2, and April 17) aligned closely with government decrees on March 24 and 31, providing mathematical validation of the intervention strategies.
The compartmental models demonstrated high predictive accuracy, particularly for recovered cases (Relative Error < 0.4%). While active case predictions initially showed higher variance (approx. 18%), the models "caught up" as more data became available, reducing the error to 0.1% by the end of the wave.
Discussion and Conclusion
The study highlights the importance of integrating mobility data and accounting for unobserved infections in epidemic modeling. While the use of 2011 census data for mobility was a limitation, the consistency of $R_t$ estimates across different models provided a robust basis for decision-making. The researchers suggest that future efforts should focus on incorporating real-time GPS data to refine metapopulation dynamics and better prepare for subsequent waves of infection.
Summary updated