Surveillance objectives
The primary objective of the SC survey is to monitor the occupancy rate of beds and medical devices within Belgian general hospitals by COVID-19 patients. Secondly, additional data and indicators obtained through the SC survey, such as the number of new confirmed and possible COVID-19 hospitalizations, discharges and deaths, are to be used for epidemiological purposes to follow trends over time.
The specific objectives of the Clinical survey are to study: (a) demographic and comorbidity profiles of the COVID-19 patients requiring hospitalization; (b) risk factors for Intensive Care Unit (ICU) admission and mortality among hospitalized patients; (c) lengths of stay in hospital, in ICU and on extracorporeal membrane oxygenation (ECMO); (d) biological status and evolution of hospitalized patients with COVID-19 based on selected laboratory measures; and (e) the use of specific medications for COVID-19.
Surveillance outline
The SC survey has been operational since the 10th of March 2020. A total of 103 general hospitals partake in the survey and their participation is mandatory (Royal Decree of 30.04.2020 [9]). This list includes all general hospitals (including university hospitals) in Belgium, and both hospitals managed by a public authority and privately run are represented. The surveillance does not cover psychiatric hospitals or specialist hospitals (e.g. exclusively providing geriatric, revalidation or palliative care) [10]. Each participating hospital sends their data on a daily basis and before 11 o'clock to Sciensano via a secured online questionnaire. To facilitate participation, reminder emails are sent to hospital contact persons each day at 8 AM. The reminder email contains the link to the survey, and highlights any changes to the survey, as well as a document with Frequently Asked Questions (FAQs).
The collection of individual data of patients hospitalized with confirmed COVID-19 through the Clinical survey was implemented on the 14th of March 2020 using a secured online questionnaire to report all COVID-19 patients hospitalized since the beginning of the epidemic. It comprises two sections: one to fill in after admission of the patient and one to fill in after discharge. Participation to this surveillance is strongly recommended for all hospitals caring for COVID-19 patients, but not mandatory.
Both questionnaires were created using LimeSurvey (GmbH, Hamburg, Germany. URL http://www.limesurvey.org), an open source survey tool. Hospitals can contact Sciensano with questions, feedback, and remarks regarding the surveys.
Study population
The SC survey collects data on both confirmed and possible COVID-19 patients who are hospitalized in one of the 103 general hospitals, with the exclusion of outpatients and day hospitalizations.
The studied population of the Clinical survey includes all hospitalized patient with a COVID-19 confirmed infection. All hospitals were invited to participate. Initially, only patients confirmed by reverse transcriptase–polymerase chain reaction (RT-PCR) tests were reported. As of the 3rd of April, the data collection was extended to cases diagnosed by chest computed tomography (CT) scan and rapid antigen tests to align with modifications in the official case definition [11].
Data collection
The SC survey collects data at an aggregated level per hospital accreditation number. Both prevalence and incidence numbers are obtained through the survey. The following variables are collected as a prevalence number: the number of hospitalized COVID-19 patients in all units of the hospital (including ICU), in the ICU, under invasive ventilation (IV), and under Extra-Corporeal Membrane Oxygenation (ECMO). These numbers reflect the state at the moment of reporting, i.e. a snapshot of the situation. The incidence numbers collected reflect the in- and efflux of patients: the number of new COVID-19 hospital admissions in the last 24h, the number of COVID-19 discharges in the last 24h, and the number of COVID-19 deaths in the last 24h. Both prevalence and incidence numbers are recorded separately for lab-confirmed cases (‘lab-confirmed’ category) on the one hand, and CT-scan-confirmed and possible cases (‘CT-confirmed/suspected’ category) on the other hand.
The SC survey had to be adapted on multiple occasions based on emerging knowledge, management and organizational policies, and modifications of testing criteria [12]. Key changes included the stratification of the number of new COVID-19 hospitalizations according to the origin of the patients, to be able to distinguish the epidemic occurring in nursing homes or other long-term care facilities (LTCF) from the one in the general population. Another major addition was the stratification by SARS-CoV-2 testing indication, in order to differentiate COVID-19-related admissions from other admissions diagnosed with COVID-19 through screening. An overview of the adaptations applied to the SC survey is presented in the Supplementary material [see Additional file 1].
Starting from the 24th of March, the database resulting from the SC survey became the official reference for the epidemiological surveillance of in-hospital deaths resulting from COVID-19. Therefore, individual data collection was added for each confirmed and possible COVID-19 patient who died as a result of a COVID-19 infection, including the date of death, date of birth, sex, postal code of residence, and method of diagnosis.
Individual data collected through the admission questionnaire of the Clinical survey include demographic characteristics (date of birth, sex, ethnicity and postal code), exposure risk, clinical presentation, pre-existing comorbidities, uptake of Angiotensin-Converting Enzyme (ACE) inhibitors and Angiotensin II Receptor Blockers (ARBs), and diagnostic information. Data collected at discharge include severity indicators, selected laboratory data, COVID-19 treatment details, admission in ICU, use of IV and ECMO, and discharge information (alive/deceased). The selection of the variables to be collected at patient’s admission was based on the WHO case report form for COVID-19 [13] supplemented by the input from hospital clinicians. With the evolution of the epidemic, new public health and epidemiological questions arose, triggering adaptation of the variables recorded. In consequence, the questionnaires were modified several times in order to ensure that information remain relevant for clinicians and authorities. For the same purpose, specific questions concerning intensive care were added on the 14th of September. The list of the variables with their date of addition/removal is available in the Supplementary material [see Additional file 2].
Data cleaning, management and analysis
The data from the SC survey is extracted from the LimeSurvey database in csv format. Errors as reported by the hospitals by email or phone are manually corrected in the SC LimeSurvey database. In addition, checks are performed on the data to detect potential inconsistencies, e.g. number of patients in ICU cannot be larger than the number of patients in the hospital. Furthermore, trend analyses per hospital are used to detect erroneous reporting. For the individual mortality data, double entries are identified based on date of birth, date of death, postal code, and reporting hospital. Data cleaning and analyses are performed using Excel and R software (R Studio version 1.0.153 [14]).
Admission and discharge data from the Clinical survey are first screened for identification and removal of duplicate observations. The two datasets are merged using the hospital patient identification number, or alternatively a combination of date of birth, sex and postal code. Extensive consistency checks are then performed with a specific focus on discrepancies between reported dates. Clinical data are extracted from the LimeSurvey and imported in SAS Enterprise Guide 7.1, software in which data management and analyses are performed.
Data protection, ethical and privacy considerations
Sciensano is legally entitled for surveillance activities related to public health in Belgium [7]. To ensure the validity of the collected data, the access to both surveys is limited to hospitals through an access code provided only to participating hospitals from April 11th onwards. All answers filled in the online survey in Limesurvey are saved on the central server of Sciensano.
Collection of aggregated data, as for the SC survey, are allowed by the General Data Protection Regulation (GDPR) [15]. Concerning the individual mortality data, although GDPR does not apply to the personal data of deceased persons, the researchers notified the Belgian Data Protection Authority. Collection of individual data, as for the Clinical survey, was authorized by the Belgian Data Protection Authority, as required by the GDPR. The data collection for the Clinical survey is retrospective in nature as Sciensano retrieves data from records of patients hospitalized with COVID-19. As such, the Clinical survey was approved by the ethical committee of Ghent University Hospital (BC-07507). The privacy authorization was modified on the 1st of September to allow the collection of patient’s pseudonimized national registry number in the Clinical survey.
A letter of information for the patient on the Clinical survey, the storage period, the right to examine and correct own data and the contact details of the person to contact to exercise those rights was sent to all data providers. Hospitals also received a letter sent by the HTSC committee explaining the mandate and the purpose of both surveys.
Data reporting and dissemination of data
Since March 14th, data from the SC survey is reported by Sciensano distinctively to the public and authorities in the epidemiological bulletin published on a daily basis. Key indicators are reported in both and an extra chapter is included in the authorities report on ICU bed occupancy. Since March 26th, an extra and more detailed report is produced on a weekly basis, including amongst others findings from the data collected by both the SC survey and the Clinical survey.
Study outcomes
After an implementation phase of five days, a participation rate of 100% was reached on March 15th for the SC survey, and remained very high during the whole period. The mean participation rate between the 15th of March and the 28th of June was equal to 99.9%.
Between the 15th of March 2020 and the 28th of June 2020, a total of 17,746 new COVID-19-related hospital admissions with a lab-confirmation available at the moment of reporting have been registered through the SC survey. The evolution over time is presented in Figure 1. The estimated total number of lab-confirmed COVID-19 patients treated by the general hospitals in Belgium is equal to 21,416 and is calculated as the sum of lab-confirmed COVID-19 discharges and deaths between the 15th of March and the 28th of June, and the number of lab-confirmed COVID-19 patients in the hospital on the 28th of June. The occupancy rate of beds (Figure 2) and devices (Figure 3) has been daily followed. The ICU occupancy rate is reported based on the total number of confirmed and possible COVID-19 patients in ICU as the nominator and the theoretical number of ICU beds reserved for COVID-19 patients (as stipulated in the HTSC plan) as the denominator.
Up to the 28th of June, detailed individual data have been recorded through the Clinical survey for 17,025 patients by 97 general hospitals which reported more than 99% of the cases and by 19 other – mainly psychiatric – hospitals for less than 1% of the cases. Both admission and discharge questionnaires were filled in for 12,891 patients (76 %). Admission information had not been reported for 1,513 (9%) of the patients and discharge information for 2,621 (15%) of the patients. Patients not yet reported as discharged include the patients still hospitalized and those discharged but not reported yet on the 28th of June. Compared to the patients with both admission and discharge information, those without discharge information were slightly older (mean 67.2 vs. 68.5 years, respectively), but there were no significant differences regarding sex and most comorbidities [see Additional file 3]. Those with no admission information had no significant differences regarding age and sex, but experienced slightly more stays in ICU (13.5% vs. 15.2%, respectively) and a higher proportion of deaths (20.8% vs. 25.5%, respectively). Missingness for key demographic variables (sex: 0.8% and age: 1.2%) in the Clinical survey is very low.
To assess coverage of the Clinical survey, we compared the number of admitted cases recorded with the results of the exhaustive SC survey. In view of comparing similar patients, only laboratory-confirmed patients reported by general hospitals were included, patients transferred from another hospital or readmitted were excluded. The estimated coverage on the period 15th of March - 27th of June was 71% (Table 1). During week 12 and 13, the number of patients reported in the Clinical survey exceeded the number of new lab-confirmed admissions as reported through the SC survey. Since week 18, the coverage of the Clinical survey falls below 50%, most probably due to a delay in reporting.
Comparison of the mortality figures from the Clinical survey with SC survey, restricted to those laboratory-confirmed to compare similar populations, showed no significant difference in sex and age distribution, but a slightly higher proportion of death reported in the Clinical survey (21.0% and 19.8%, respectively; p=0.015) [see Additional file 4].