Introduction
Coronavirus disease 2019 (COVID-19) was declared a global pandemic by the World Health Organization (WHO) on 12 March 2020 [1]. Current and published epidemiological research on COVID-19 has largely focused on China and other high-income countries such as South Korea, Japan, the USA, Italy and Spain [2]. Further research on the distribution and burden of COVID-19 in low- and middle-income countries (LMICs) may give insight on its disease epidemiology in low-resource settings as transmission dynamics are dependent not only on population characteristics [Reference House and Keeling3], but also health system capacity (e.g. access to testing) [Reference Kelly-Cirino4], and the ability to implement mitigation measures (e.g. community-level quarantine, social distancing) [Reference Abrigo5].
The Philippines confirmed local transmission of COVID-19 on 7 March 2020 and implemented enhanced community quarantines (ECQ) from 20 March to 30 April on half the country's population [6]. ECQ entailed very strict community-level social and physical distancing measures such as suspension of public transportation, prohibition of intra-country travel and restriction of people to their homes except for essential activities like food and health care [7].
In this paper, we describe the epidemiological profile and transmission dynamics of the first 8212 confirmed COVID-19 cases in the Philippines.
Methods
Overview of the Philippine COVID-19 surveillance system
The Philippines is an archipelago of three island groups and 17 regions subdivided into 81 provinces covering 146 cities and 1488 municipalities [8]. COVID-19 surveillance, like majority of health service delivery, is decentralised to local government units (LGUs), i.e. provinces, cities and municipalities. Epidemiology and surveillance units (ESUs) exist in every administrative level, namely regional ESUs (RESUs), provincial ESUs (PESUs) and city/municipality ESUs (CESUs/MESUs). Units collect data for their jurisdictions and report to higher level units: CESUs/MESUs cascade daily updates to the PESUs, which cascade those to regional ESUs which then finally submit to the DOH-Epidemiology Bureau (EB) [9, 10]. The 15 cities and one municipality in the National Capital Region (NCR), as well as 37 highly urbanised cities and independent component cities are not overseen by a provincial government, and report directly to the RESUs.
The DOH-EB is the lead national agency for COVID-19 surveillance. It collates data on confirmed and suspected cases nationwide and provides guidance and support to all LGUs. DOH-EB maintains an information system for COVID-19 cases patterned after influenza-like illness (ILI)/severe acute respiratory infection (SARI) surveillance. Confirmed cases are profiled using case investigation forms (CIF), which record patient characteristics, epidemiologic links and select clinical information.
The CIFs include questions on occupation which are used to identify healthcare workers. Only those actively working in a hospital or a health facility were considered as healthcare workers for this analysis. We did not include those who may hold degrees in health professions, but were not in active service. Health care worker status is confirmed and validated by dedicated surveillance teams in LGUs and DOH-EB given that healthcare workers who die or experience severe or critical symptoms receive additional compensation under the ‘Bayanihan to Heal As One' Act passed in April 2020 [11].
Case definition, case detection and laboratory testing
We define COVID-19 cases as patients with positive real-time reverse transcription polymerase chain reaction (RT-PCR) conducted by laboratories accredited by the DOH and Research Institute for Tropical Medicine (RITM) [12, 13]. The RITM is the National Reference Laboratory for Emerging and Re-emerging Diseases and it is the public health authority that accredits laboratories for COVID-19 testing. We included cases recorded up to 29 April 2020 at the time of writing and followed the status of these patients until 22 May 2020 using the latest dataset prior to publication.
Starting 9 April 2020, the DOH limits testing to suspect and probable cases. A suspect case is a person who has any of the following: (1) SARI requiring hospitalisation with no other aetiology that fully explains clinical presentation; (2) ILI with no other aetiology that fully explains clinical presentation AND residence or travel to an area with known local transmission 14 days prior to symptoms OR exposure to confirmed or probable cases during the period 2 days prior to symptoms until they test negative with RT-PCR; and (3) high-risk groups presenting with fever, cough, shortness of breath and other respiratory symptoms, including the elderly 60 years and above, those with comorbidities, women with high-risk pregnancies and health workers [14]. Probable cases are suspect cases (1) referred for RT-PCR testing, (2) with inconclusive RT-PCR results from a DOH-accredited laboratory, or (3) who have a positive RT-PCR result from a non-RITM accredited laboratory. As of 29 April, there have been six versions of COVID-19 case definitions and testing has expanded to include 17 subnational laboratories (see Supplementary Appendix).
Potential cases are detected through multiple avenues. At ports of entry, the DOH – Bureau of Quarantine identifies persons showing symptoms such as fever, shortness of breath and respiratory problems and refers them to health facilities, LGU health offices or sentinel disease reporting units (DRUs). Health providers also use official case definitions to assess potential COVID-19 cases among patients who consult or are hospitalised in their facilities. LGUs, ESUs and DRUs then conduct case investigations and contact tracing for reported confirmed, suspect and probable COVID-19 cases [14].
Death was defined as death due to active COVID-19 and reported as dead by the DOH-EB. An internal team of medical experts at DOH-EB reviews death certificate data, if available, or the data transmitted through the information system by the health facility or ESUs, to validate if the cause of death was related to COVID-19. Recovery was defined as COVID-19 cases reported as recovered by the DOH-EB based on the resolution of symptoms, including discharge from hospital for those admitted or end of home quarantine for those not admitted, and at least one negative RT-PCR test. We allowed 3 weeks of follow-up (29 April to 22 May 2020) to ascertain health outcomes to account for the usual time in which we expect either recoveries or deaths [15].
All data are available as part of the DOH's open data access ‘DataDrop’ initiative, updated daily at https://ncovtracker.doh.gov.ph. Some of the additional variables such as health worker status and known comorbidities were requested separately from DOH-EB and received by the authors as deidentified data.
Statistical analysis
We analysed descriptive statistics of cases, deaths and recoveries by socio-demographics, symptoms at specimen collection and health care worker status to the extent that these data were available. Epidemic curves were constructed using the date of symptom onset. As many dates were missing, they were imputed based on the method by Günther et al. as applied to the COVID-19 outbreak in Bavaria, Germany [Reference Günther16]. A flexible, generalised additive model with a Weibull distribution was fitted for the days between symptom onset and DOH public announcement. Predictors were: (1) day of week of reporting, (2) region of residence, (3) laboratory of testing, (4) sex; and the following smoothed predictors: (5) calendar week of reporting date; and (6) age (Supplementary Appendix). The imputed data set was used for all further analyses.
Among a subset of cases with complete data, we estimated the distributions of six time-to-event variables: (1) serial interval, or time between symptom onset of index and secondary cases; (2) health-seeking behavioural delays, or time between symptom onset and first medical consultation (or specimen collection if missing); (3) diagnostic delays, or time between specimen collection (or first medical consultation if missing) and laboratory confirmation; (4) hospital length of stay for admitted cases, or time between admission and discharge (or death for those who died); (5) length of condition, or time between symptom onset and death or recovery; (6) reporting delays for confirmed cases, or time between laboratory confirmation and public announcement. To calculate the serial interval, clustering analysis was done using known epidemiologic links, exposure history and symptom onset for the first 100 confirmed COVID-19 cases where contact tracing was mandated by DOH guidelines [17]. Distributions that best fit the data were identified using the Akaike information criterion (AIC).
Exponential growth rates, epidemic doubling times and reproductive numbers (basic reproductive number (R0) and effective reproductive number (Re)) were calculated, accounting for the empirical serial interval of our data [Reference Li18, Reference Lauer19]. R0 was estimated during the exponential growth period using the method developed by Wallinga and Lipsitch [Reference Wallinga and Lipsitch20], with the best fit of dates determined by the deviance-based R 2 statistic [Reference Obadia, Haneef and Boëlle21]. Re was estimated using the Wallinga and Teunis method during the period after ECQ was mandated [Reference Wallinga and Teunis22].
Analyses and visualisations were done using Stata 16.1 (College Station, TX, USA: StataCorp LP), Tableau Desktop 2020.1.3 (Mountain View, CA, USA: Tableau Software, Inc) and R 4.0.0 [23] using the EpiEstim [Reference Cori24], epitools [Reference Aragon25], fitdistrplus [Reference Delignette-Muller and Dutang26], gamlss [Reference Rigby and Stasinopoulos27] and MASS [Reference Venables and Ripley28] packages.
Ethics approval
Data came from continuing surveillance efforts by the DOH-EB as part of an ongoing outbreak investigation. Thus, institutional review board approval was deemed unnecessary.
Results
Among the 8212 COVID-19 cases detected in the Philippines up to 29 April 2020, 46.2% were female and 68.8% lived in the NCR (Table 1). Among these, 768 (9.4%) died and 2988 (36.4%) recovered. Median age for cases, deaths and recoveries was 46 years (IQR 32–61), 66 years (IQR 57–74) and 46 years (IQR 32–59). There were 319 (3.9%) cases and 12 deaths among those aged 0–20 years. Health care workers represent 24.7% of detected infections with 35 deaths. Hypertension (17.9%) and diabetes (12.7%) were the most common comorbidities while fever (31.3%) and cough (44.9%) were the most common symptoms at specimen collection. Compared to those who recovered, those who died were more likely to be older, male, presented with difficulty breathing and had comorbidities except for asthma.
Table 1. Characteristics of COVID-19 cases in the Philippines as of 29 April 2020
The epidemic curves based on symptom onset shows that the exponential growth period of the outbreak likely began on 11 February 2020 (Fig. 1). Only 5168 (63.0%) cases had complete dates of symptom onset and the rest were imputed (see Supplementary Appendix for model diagnostics). Several cases (3.5%) visited a country with known local transmission 14 days prior to their symptoms. New cases steadily increased until the week of 15 March 2020 when NCR and Luzon were put under ECQ. Detected cases rose during this period likely due to the expansion of laboratory testing capacity in the last week of March. The decreases in the last weeks leading up to 29 April were likely due to delays between symptom onset and public announcement – and not due to a true decline in new infections.
Fig. 1. Onset of illness among the first 8212 COVID-19 cases in the Philippines, the National Capital Region, and Region 7.
Figure 2 and Supplementary Appendix present the distributions for six key epidemiologic events and surveillance delays from symptom onset to public announcement in the Philippine health system. The serial interval had a mean of 6.90 days (95% CI 5.81–8.41; Weibull) using 55 pairs of index and secondary cases. The mean health-seeking behavioural delay or time between illness onset and first medical consultation was 6.75 days (95% CI 6.70–7.15; γ). The mean diagnostic delay or time between specimen collection to laboratory confirmation was 4.92 days (95% CI 4.82–5.02; lognormal). Among those who were hospitalised, the mean length of stay for recoveries was 16.00 days (95% CI 15.48–16.54; Weibull) while for deaths was 7.27 days (95% CI 6.59–8.24; γ). Among those who had recovered or died, the mean duration of illness with COVID-19 was 26.66 days (95% CI 26.06–27.28; Weibull) and 12.61 days (95% CI 11.88–13.37; γ), respectively. Lastly, mean reporting delays or the time between laboratory confirmation and public announcement was 2.47 (95% CI 2.40–2.51; γ). Notably, these delays were such that of the 541 deaths with complete data on date of death and laboratory confirmation, half of them (276, 51%) received laboratory confirmation after they had died.
Fig. 2. Time-to-Event Distributions for key epidemiologic events and surveillance delays. (a) Serial interval or the time between onset of symptoms for index and secondary cases. (b) Behavioural delay or time between symptom onset to first medical consultation. (c) Diagnostic delay or time between illness onset to laboratory confirmation for COVID-19. (d) Length of stay or time from hospitalisation to discharge (blue) or death (red). (e) Length of condition or time from symptom onset to recovery (blue) or death (red). (f) Reporting delay or time between laboratory confirmation to public announcement by DOH.
Figure 3 presents three of the largest known clusters. Cluster A's likely source of infection was the wake of a cases' mother-in-law; it was also plausible the first generation of cases had multiple points of contact before and after the wake given that they were all friends and family. The second generation cases were household members of first generation cases, save for one who infected a high-level government official. Cluster B and C's likely sources of infection were birthday parties, and similar to Cluster A, infected the rest of their household. Aside from friends and family, other known epidemiological links are between cases and their attending health care providers, such as the one found in Cluster A.
Fig. 3. Some known large Covid-19 outbreak clusters in the Philippines.
In the exponential growth phase pre-ECQ (11 February and 19 March), the epidemic growth rate was 0.143 per day (95% CI 0.137–0.149), and the epidemic doubling time was 4.86 days (95% CI 4.67–5.07). Using the estimated serial interval distribution, the basic reproductive number for this period was 2.41 (95% CI 2.33–2.48). Analysis for the ECQ period was limited to April 15 to account for reporting delays. In the period of 20 March to 15 April, the epidemic growth rate was 0.053 per day (95% CI 0.052–0.055), the epidemic doubling time was 12.97 days (95% CI 12.57–13.39) and the average daily effective reproductive number was 0.89 (95% CI 0.78–1.02).
Discussion
This overview of the first 3 months of the COVID-19 outbreak in the Philippines provides a unique perspective on the transmission of COVID-19 in a health system with limitations in surveillance, testing and service delivery. Such information may aid modelling or data analyses efforts for outbreak response in the Philippines and countries with similar health system constraints as studies of the pandemic outside of high-income countries and in LMICs have been limited.
Our results support findings that COVID-19 disproportionately burdens older age groups, healthcare workers and densely populated urban areas [Reference Wu and McGoogan29–31]. Outbreak epicentres in the Philippines were urban centres, such as NCR and Cebu City, where NCR alone accounted for more than two-thirds of all cases. NCR is the fifth most densely populated metropolis in the world [32]. The first few cases were likely imported into these urban centres, as approximately one-tenth of the population are migrant workers and 8 million tourists visit annually [33].
In the Philippines, one in four confirmed cases were healthcare workers, much higher than the Western Pacific regional average of 2–3% [Reference Esguerra34]. The high proportion of healthcare workers as a share of cases may indicate poor compliance with infection control protocols and shortages of personal protective equipment [35–Reference Tomacruz38]. To note, healthcare workers may also have had better access to testing early in the outbreak and COVID-19 infection in one may trigger testing or surveillance among other healthcare workers to limit hospital outbreaks. Further investigation is necessary to identify exposure sources, gaps in infection control and facilitators of hospital outbreaks.
In the Philippines, surveillance delays were prominent due to the limited healthcare system capacity which affects the timeliness of decisions to suppress the outbreak. For example, on 7 March, the DOH officially reported local transmission upon public announcement of the sixth case [39, Reference Duterte40], but our findings suggest that the exponential growth period likely began more than 3 weeks earlier. Delays also affect rapid isolation of suspected cases, which is critical to reduce the reproductive number to below 1 and suppress the outbreak. To reduce health-seeking behavioural delays, the National Health Insurance Program, of which all Filipinos are members under the newly passed Universal Health Care Law, has made laboratory testing free [41], and covers the majority of community isolation and hospitalisation costs since mid-April [42–44], To reduce diagnostic delays, the Philippines has slowly expanded laboratory capacity from just one laboratory in February. More isolation facilities, such as stadiums, hotels and schools, have been set-up for mild cases to save limited hospital resources for severe and critical cases. Information systems are being strengthened to allow synchronised reporting of cases at all ESU levels in real time.
Our analysis also showed that the prior to the ECQ reproductive number was 2.4, meaning that for every confirmed case, between two and three other cases were infected. This estimate is similar to previous estimates in literature [Reference Park45]. Travel restrictions imposed for China and South Korea in February may not have been sufficient to slow the progression of the outbreak, as the reproductive number remained around 2.4 during the period before ECQs were implemented. While full travel restrictions and the ECQ decreased the reproductive number to 0.9, these interventions merely delayed the progression of the outbreak rather than stopping it completely. Furthermore, the serial interval estimate of 7 days is longer but comparable to previous estimates in other settings [Reference Nishiura, Linton and Akhmetzhanov46].
This study has limitations common in disease outbreak response and health information systems in LMICs. First, there is a large number of missing data that needs to be further validated due to difficulties of retrieving data from a decentralised surveillance system. At the time of this study, data were mostly from paper CIF and not medical records. As such, we did not have detailed clinical information on disease severity or disease progression, and comorbidities may be underestimated. Only 63% of cases reported their date of symptom onset; however, we implemented a previously validated imputation method to correctly specify the epidemic curve and better estimate the reproductive number. Second, the focus of laboratory testing for early phases of the Philippine outbreak was on severe or critical cases, but the DOH has since expanded testing for those with milder symptoms. Since more milder cases were captured by the surveillance system during the latter part of the outbreak, mean surveillance delays, length of stay and duration of illness may have declined as health system capacity improves. Thus, our estimates of the epidemiologic time-to-events and surveillance delays may be overestimated. In addition, the case counts, recoveries and deaths may be artificially undercounted due to underreporting of mild cases, missed asymptomatic cases, backlogs in confirmatory testing, and delays and errors in reporting and validating for recoveries and deaths. While 3 weeks were allotted as follow-up period for deaths and recoveries, some censoring still likely occurred for those with longer length of condition. Thus, we did not calculate the case fatality rate due to these potential biases. Fourth, we could not compute the event attack rates for the clusters illustrated in Figure 3 due to incomplete data on the total number of people in each gathering. Hence, these clusters serve only to illustrate the largest COVID-19 clusters identified by the DOH-EB in the early phases of the outbreak in the Philippines. Finally, there may be additional differential ascertainment in the characteristics of cases due to disparities in access to care; individuals with higher socioeconomic status, health care workers and workers with occupations who are able to stay sick at home may be more likely to test and seek care, even though socially disadvantaged communities may have increased susceptibility to COVID-19 [Reference Chung, Dong and Li47].
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0950268820002137
Open access
