key: cord-0757378-rqoka18m authors: Kakoullis, Loukas; Eliades, Elias; Papachristodoulou, Eleni; Parperis, Konstantinos; Chra, Paraskevi; Constantinidou, Anastasia; Chatzittofis, Andreas; Sampsonas, Fotios; Panos, George title: Response to COVID‐19 in Cyprus: Policy changes and epidemic trends date: 2021-01-03 journal: Int J Clin Pract DOI: 10.1111/ijcp.13944 sha: 0415f059633609f50daa889b4f8ad6d4c08f6cde doc_id: 757378 cord_uid: rqoka18m OBJECTIVES: In late July, Cyprus experienced the second epidemic wave of COVID‐19. We present the steps taken by the government and evaluate their effect on epidemic trends. MATERIALS: Cyprus Press and Information Office data were analysed. Using an R‐based forecasting program, two models were created to predict cases up to 01/09/2020: Model 1, which utilised data up to 09/06/2020, when airports reopened to foreign travelers with COVID‐19 screening; and Model 2, which utilised data until 24/06/2020, when screening for passengers from low‐transmission countries was discontinued. RESULTS: PIO data revealed no significant policy changes between 24/06/2020 and 31/07/2020. Prediction models were robust and accurate (Model 1, R (2) = 0.999, P < .001; Model 2, R (2) = 0.998, P < .001). By August 30th, recorded cases exceeded those predicted by Model 1 by 24.47% and by Model 2 by 20.95%, with P values <.001 for both cases. CONCLUSIONS: The significant difference between recorded cases and those projected by Models 1 and 2 suggests that changes in epidemic trends may have been associated with policy changes after their respective dates. Discontinuation of major restrictions such as airport reopening, can destabilise the control of the epidemic, and may concomitantly necessitate a reevaluation of the current epidemic status. In the face of an evolving situation such as the COVID‐19 pandemic, states are forced to balance the imposing of restrictions against their impact on the economy. of Health to the PIO is any patient with a polymerase chain reaction (PCR) test positive for SARS-CoV-2, regardless of whether they are symptomatic or not. 2 Using these data, daily COVID-19 incidence, tests and hospitalisations were plotted using Microsoft Excel. Furthermore, a timeline of government policy changes was constructed. All data utilised in this study are available from the websites of the Cyprus PIO 2 (https://www.pio.gov.cy/coron aviru s/) and the Cyprus National Opendata Portal 1 (https://www.data.gov.cy/ node/4617?langu age=en). Data were analysed using Prophet, an R-based forecasting model, in order to make predictions for certain dates and evaluate whether changes in government policy impacted outbreak control. Prophet is a time series forecasting model that implements machine learning fitting and time series decomposition, returning a high accuracy time series forecast. The algorithm takes into consideration historical data and their characteristics to predict future observations. Prophet is considered a robust model to outliers, missing data and dramatic changes in time series events. 3 Prophet uses a time series forecasting algorithm, based on an additive model for future prediction, and has been utilised successfully to predict COVID-19 epidemic trends both at a national 4 and international level. 5 In this work, Prophet was implemented in R programming language. In order to provide results, Prophet requires an input of a time series with two features: date ds and value y. Its algorithm chooses a training model according to historical data and their characteristics to find future observation results, which are fitted in yearly, weekly and daily seasonality plus holiday effects. The prophet model has the form y(t) + s(t) +h(t) + ε(t), where g(t) is the trend function, s(t) is the periodic component, h(t) represents holidays which occur irregularly and ε(t) is the error term which is often assumed to be normally distributed. 6 Using the data and algorithms described above, Prophet can execute various functions. The two functions utilised in this study were 'make_future_dataframe', which creates a future date data frame up to a requested date, and 'predict', which makes predictions for each row in the future data frame provided by the 'make_future_ dataframe' function, up to the date chosen. Through this function, Prophet predicts incidence in future dates based on the incidence of previous dates. In this study, a data frame was created in R using the dates and the cumulative incidence of the COVID-19 outbreak in Cyprus. Subsequently, the 'make_future_dataframe' and 'predict' functions were utilised respectively to create dates and predict cases up to September 1st. The effect of government policy changes was evaluated by comparing the projected cumulative incidence from the date of policy changes to the actual cumulative incidence. Utilising Prophet, graphs representing the projected cumulative incidence were plotted by entering non-linear data of confirmed cases up to the date in question. Projective graphs were plotted for June 9th through September 1st (termed Model 1) and June 24th through September 1st (termed Model 2). These graphs were then compared with the actual cumulative incidence until September 1st. Statistical analysis was conducted to evaluate the differences between projected cases predicted by Model 1 and Model 2 to the recorded cases up to September 1st. P values were calculated using Kruskal-Wallis rank-sum test. Statistical analysis was also conducted to compare the proportion of positive COVID-19 tests in August and July, using a two-sample test for binomial proportions and reporting a two-sided P value. Only P values <.05 were considered significant. Data regarding daily new cases were available from March 9th through September 1st. A timeline of government policy changes is depicted in Figure 1 , while daily new cases and tests, as well as What's known • Despite an initial success in containing the local COVID-19 pandemic, Cyprus has been experiencing the 2nd wave of the epidemic since late July. • A de-escalation of measures occurred prior to the second wave. • The short time interval between changes may not have allowed for sufficient evaluation of their impact on the status of the epidemic. • We constructed a timeline identifying dates of important policy changes, which identified that the last major policy changes occurred on June 9 th (when airports reopened to foreign travelers with COVID-19 screening) and June 24 th (when screening for passengers from lowtransmission countries was discontinued). • Using an R-based forecasting program, we have demonstrated that COVID-19 cases in August were 20%-24% higher than predicted by data collected before the deescalation of these specific measures. numbers of patients hospitalised in general wards and ICUs, are depicted in Figure 2 . These data are also presented in a combined form in Figure S1 , which depicts the cumulative incidence of COVID-19 in Cyprus, while highlighting the dates of important changes in policy. The first two cases on the island were identified on 09/03/2020 in returning travelers. On 16/03/2020 (A), entrance to Cyprus was forbidden to non-residents and required a negative COVID-19 test. Non-emergency hospital admissions were cancelled, schools and non-essential businesses were closed, while on 17/03/2020, non-essential personnel in the public sector began working from home. On 20/03/2020, incoming passenger flights were forbidden (B). As a result of rising number of cases, outside movements were limited to 3/day on 23/03/2020 (C) and restricted to 1/day on 30/03/2020 (D). Two days later, the epidemic reached its peak, 23 days since its beginning. Cases entered a steady decline on 17/04/2020, 17 days after the curfew was imposed. On 30/04/2020 (E), following 2 weeks of steadily low cases, the public sector returned to work and outside movements were increased to 3/day. Curfew was entirely lifted on 21/05/2020 (F), 52 days after being imposed and 73 days after the first cases were reported. Concomitantly, most businesses and open-space restaurants reopened with masked employees, while gatherings up to 10 persons were permitted. On 09/06/2020, passengers from low and average transmission countries were allowed to enter Cyprus. Passengers were either tested on arrival or provided a negative COVID-19 test within the previous 72 hours (G). On 24/06/2020, following a streak of four consecutive days without cases, passengers from low-transmission countries were allowed entrance without a test and gatherings exceeding 10 people were permitted (H). On 23/07/2020, 63 days post-end of curfew and 33 days after permitting free entrance to Cyprus, a steady increase in the number of cases was documented in Limassol. On 31/07/2020, 25 new cases were documented, a 107-day high. The government responded by making masks mandatory in closed spaces and, imposed gathering restrictions in Limassol (I). It should be noted that this was the first significant change in government policy since June 24th. A steady increase of cases has been observed ever since. Within 30 days from July 23rd, the start of the second wave, more new cases were identified than within the previous 100 days. Data up to the 9th of June and up to the 24th of June were utilised by the Prophet algorithm to develop two models to predict the cumulative incidence of the virus up to the 1st of September. Graphs depicting the projected cumulative incidence were plotted for June 9th and June 24th, termed respectively Model 1 and Model 2, as depicted in Figure 3 , and compared with the actual cumulative incidence. These dates were considered significant, as each represented a major de-escalation of government restriction measures. Both datasets produced models that could predict future incidence with high accuracy, with R 2 values of 0.999 and 0.998 for the Model 1 and Model 2 respectively. The P values for both models were <.001. As evident from the results depicted in Figure 3 , there is no significant difference between the projected cumulative incidence between the two Models and the actual cumulative incidence until July 31st, when a significant deviation from predicted data is noted. By comparing the recorded cases against the projected cases from Model 2, it becomes evident that recorded cases begin exceeding projected cases on 27/07/2020. The recorded cases exceeded Model 2's cases by 4.04% on 31/07/2020, by 13.34% on 10/08/2020, and by 20.65% on 20/08/2020. The difference peaked on 30/08/2020, with recorded cases being 24.47% higher than those predicted by Model 2. The difference between recorded cases and those pro- island through repatriation flights followed a strict isolation protocol in government designated areas, thereby limiting the community spread of imported cases. On June 9 th , airports were re-opened to foreign travelers, allowing passengers from low and average transmission countries to enter Cyprus, provided they had a negative COVID-19 test within the previous 72 hours; otherwise, they were submitted to a test on arrival and remained isolated pending the results. On June 24 th , only 2 weeks later, testing for passengers from low-transmission countries was no longer necessary. These measures were discontinued in an effort to restart the tourism industry, which accounted for 21.9% of the island's GDP in 2018. 9 Unfortunately, the percentage of imported cases in Cyprus has risen significantly, Based on the results of the comparison of the projected cumulative incidence from Model 1 and Model 2 and the actual cumulative incidence, it is evident that no difference can be detected between the number of cases in these scenarios until July 27th. By July 31st, the difference is significant enough that it can be appreciated as shown in Figure 3 . We speculate that the period between the opening of the airports and the cessation of testing for passengers from low-incidence countries did not provide adequate time for evaluation of the new status. The effects of the discontinuation of a specific measure need time before they become evident and should be studied before another measure is lifted as well. While a combination of factors is more likely responsible for the observed spike in cases, we speculate that the combination of the entrance to Cyprus without a prior negative test and the lifting of Until an effective vaccine against SARS-CoV-2 becomes available, new epidemic waves will continue to be a risk. 12 At this time, the only measure to limit the virus spread is a well-organised and coordinated public health policy. In order to achieve a sustained containment of the pandemic, states should be wary of lifting multiple restriction measures at once. Without adequate time to evaluate the effect of the discontinuation of a specific measure on pandemic trends, lifting of multiple restrictions increases the risk of causing a surge in new cases. Limitations of this study include a possible error in future prediction models made using the machine learning algorithm that may arise from the relatively small size of data inserted in the datasets. In addition, it is not possible to ascertain whether the calculated dif- In conclusion, the lack of policy implementation stamina, while possibly unavoidable, can have significant consequences on the control of the pandemic. The lifting of restrictions, especially of major restrictions such as border, port, airport or-in the case of Cypruscheckpoint closure, can have a major impact in states which have controlled the pandemic. The prognostic data provided in this study could not be used to make interventions at the time of writing but can be used as a reference to guide future studies and management of the pandemic in Cyprus or elsewhere. In the face of an evolving situation such as the COVID-19 pandemic, states are forced to balance the imposing of restrictions against their impact on the economy. This equilibrium is difficult to achieve, as well as maintain for extended periods of time. Considering that a pandemic of magnitude of COVID-19 has not occurred in decades, with states and scientists learning as they go, the intention of this work was to present the effect of the policies applied in Cyprus, to be of use by other nations or territories under similar circumstances. Part of the data presented in this manuscript were previously Cyprus National Opendata Portal. COVID-19 Daily Statistics Forecasting at scale Online forecasting of COVID-19 cases in Nigeria using limited data Prediction of epidemic trends in COVID-19 with logistic model and machine learning technics Estimation procedures for structural time series models European Centre for Disease Prevention and Control. Case Definition for Coronavirus Disease 2019 (COVID-19), as of 29 Announcement regarding the COVID-19 pandemic KNOWEMA. The GDP contribution from tourism in Cyprus COVID-19) in the EU/EEA and the UK-Eleventh Update: Resurgence of Cases Nicosia (Cyprus): Cyprus Ministry of Health Long-term and herd immunity against SARS-CoV-2: implications from current and past knowledge Additional Supporting Information may be found online in the Supporting Information section Response to COVID-19 in Cyprus: Policy changes and epidemic trends All authors declare that they have no conflict of interest.