Protocol for a systematic review and meta-analysis of observational studies examining the impact of COVID-19 safety measures on physical activity patterns in adults

The coronavirus disease 2019 (COVID-19), initiated by the severe acute respiratory syndrome coronavirus 2 (SARS-nCOV-2), has developed into a worldwide pandemic causing substantial physical as well as psychological health issues [1, 2]. To contain the spread of SARS-nCOV-2, authorities have implemented safety measures such as social distancing, the shutdown of public facilities, lockdowns, or home quarantine, largely restricting daily life. The extent of these measures has in most countries been determined by the number of new COVID-19 infections in a certain period and varies accordingly between countries and even regions or cities. As an example, Switzerland, Germany, and Austria enacted strict lockdowns at the beginning of the pandemic [3]. With the number of new infections decreasing in summer, the safety measures have been mitigated, enabling the reopening of shops and public facilities (i.e., gyms, sports clubs, parks) [3]. This shows how public life was and still is impacted by the constantly changing COVID-19 situation.

Already during the first COVID-19 wave (approximately from December 2019 until the middle of 2020), our research group and many others began to draw attention to the arising problem of reduced physical activity (PA) [4‐6]. PA is an important modifiable risk factor for numerous chronic diseases and all-cause mortality [7]. Greater time spent physically active is independent of PA intensity associated with a substantially reduced risk of premature mortality [8]. On the other hand, pronounced physical inactivity is strongly connected to undesirable health outcomes such as chronic non-communicable diseases (including cardiovascular diseases) [7] and therefore an immense economic burden for healthcare systems [9]. According to conservative estimations for the year 2013, physical inactivity was responsible for healthcare costs of 53.8 billion $ worldwide [9]. Besides, it was recently shown that regular PA may act as a protective factor for COVID-19 and especially a severe disease course [5, 10, 11]. It is, therefore, highly relevant to examine potential changes in PA due to COVID-19 safety measures.

Since the onset of the pandemic, numerous studies have been conducted around the globe highlighting changes in PA patterns due to the safety measures in place [12‐15]. Both, studies using self-reported data [12, 15] and trials using objective methods (i.e., wearables or smartphones with accelerometers, pedometers, or global positioning systems) [13, 14] have researched this area. Available studies largely vary in their methodology when it comes to the assessment tool used, study population, country, or period in which the data collection was performed [12‐15]. Such differences make it difficult to draw conclusions. To date, there is one systematic review available that aimed to examine changes in PA and sedentary time from before to during the COVID-19 lockdown (n = 45 studies in healthy adults) [16]. Yet, we feel that since the end of their search in June 2020, further important studies have been published that need to be taken into account. Finally, the discrepancies in safety measures between countries and the fluctuations in their extent since the outbreak of the pandemic make it difficult to transfer study results to real-life.

The planned study will, therefore, synthesize the available evidence from different countries and periods in time, making data more accessible and utilizable for public health policymaking and facilitating the development of adequate countermeasures. We will investigate whether the safety measures led to changes in PA patterns in the adult general population (≥ 18 years of age). Furthermore, this will likely be the first study taking the extent of safety measures into account.

All applied eligibility criteria are listed below according to the PICOS scheme.

The search strategy was developed using the PICOS scheme described above and included database-specific controlled vocabulary used for the indexing of articles (such as medical subject headings in Medline), free-text terms (title and abstract), and subheadings. To search for studies on COVID-19 and safety measures, two search strings developed by the Canadian Agency for Drugs and Technologies in Health were adapted [20]. The search will be initiated in Medline (on Ovid) and later translated for Web of Science (core collection), Scopus, and L^.OVE Coronavirus disease by Epistemonikos. We additionally search for gray literature in ProQuest Dissertations & Theses A&I (on proquest.​com) to reduce publication bias (start of search: 18 April 2021). The full search strategy (available in Additional file 2) was peer-reviewed using the PRESS checklist by an information specialist of the Basel Medical University Library in Basel, Switzerland [23]. We are aware of the recommendation not to include outcomes in the search strategy as they are often poorly described in the title or abstract of articles [24]. Yet, for the planned study, it seems unavoidable. Based on preliminary searches and reviewing relevant studies, we are convinced that the terms used in the search strategy are adequately described in the literature. In addition to the systematic literature search, citation tracking (backwards and forwards) will be used to identify relevant literature. The searches will be re-run prior to final analyses to avoid overlooking any relevant citations.

All citations obtained from the systematic searches of the selected databases will be exported to EndNote X9.3.3 or a later version (Clarivate, Philadelphia, USA), where deduplication will be performed. At this stage, the author’s names, publication year, journal, URL, and abstract will be extracted to a pilot-tested Excel data form that will be compiled based on Tawfik et al. [14].

Before the onset of the screening procedures, reviewers will be trained and evaluated based on Cohen’s kappa of agreement [24]. The reviewers will independently review a set of 50 randomly selected articles [25]. In the case of Cohen’s kappa < 0.8, training will be repeated [25].

All screening procedures will be performed independently by two reviewers (FS and EP) following the aforementioned inclusion and exclusion criteria, blinded for the other reviewer’s decisions. The process will be facilitated using the Rayyan web application for systematic reviews [26] and all decisions will be documented by denoting “yes,” “no,” or “maybe.” Citations that received the denotation “maybe” will be solved by consensus between the reviewers. In the case of no settlement, a third reviewer (JG) will be consulted. During the screening procedures, the deduplication process will continue manually. The process will start with title and abstract screening. Subsequently, the full-text will be obtained and all citations that have not yet been excluded will undergo full-text screening. During this step, all excluded citations will be given an exclusion reason [17]. Given no full-text is available, study authors will be contacted, the request full-text function on ResearchGate will be used, or it will be purchased if available. All decisions and exclusion reasons will be presented in a PRISMA flow diagram. Following the screening procedures, the inter-rater agreement will be documented using Cohen’s kappa of agreement [19].

In this step, all included studies will be divided between the three reviewers (FS, EP, and JG) and the following data will be extracted independently into the Excel data form: study design including the country and period during which the study was conducted, sample size, type of comparison used (e.g., control group or normative data), inclusion and exclusion criteria, subject characteristics, data on the outcomes defined hereinafter, details on the method used to measure PA, and unit of measurement. A data checking step will be included following data extraction to prevent human error. Each reviewer will be assigned one-third of the included citations and compare them to the data collection form.

The primary study outcome will be the standardized mean difference in total PA (TPA) per time unit between pre- and during COVID-19 or between normative data and during COVID-19. The standardized mean difference in sedentary time and the intensity categories of PA (light [LPA], moderate [MPA], and vigorous [VPA] or moderate-to-vigorous [MVPA]) will be secondary outcomes. This will yield important additional information, as especially higher PA intensities are relevant for physical conditioning [22]. Efforts will be taken to transform relevant data into similar units (e.g., min per week to min per day). If no information about TPA is available, LPA, MPA, and VPA will be summed up to yield TPA. For studies investigating the effects of safety measures at several points in time, all relevant data will be extracted.

Supposing that data need to be extracted from graphs, Web plot digitizer [27] will be used to facilitate this process. To calculate missing variables during data extraction, previously reported equations will be used [28, 29]. Results only reported as medians will be transformed to means accordingly [29]. Missing standard deviations will either be calculated based on other available study information [28, 29] or standard deviations of similar studies will be imputed [30]. When missing data cannot be calculated, we will attempt to contact the authors of the publication to obtain the relevant missing data. The procured unpublished information will briefly be described in the manuscript. If studies only reported adjusted outcome values, they will be included according to the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions [24]. To reduce the risk of bias through the inclusion of studies published more than once, we will compare author names, sample sizes, outcomes, and locations where the trials were conducted [19].

Risk of bias assessment will be done by two independent reviewers using the National Heart, Lung, and Blood Institute tool for observational and cross-sectional studies [31]. This tool includes 14 items that are answered with “yes,” “no,” or “not applicable.” These will then be converted into a dichotomous rating (“yes” = 1, “no” and “not applicable” = 0) [17]. Every citation will receive a score by summing up all 14 items [17]. Poor, fair, and good study quality will correspond to a score of 0‑5, 6‑9, and 10‑14, respectively [17]. The scores will be added to the data collection form prior to the analyses. The results of the risk of bias assessment will be presented in a table. Occurring conflicts will be solved by consensus between the two reviewers (FS and EP). In the case of no consensus, a third reviewer (JG) will be consulted.

Following this process, the stringency index reported in the Oxford COVID-19 Government Response Tracker database at the respective periods the studies were conducted will be matched to the citations in the data collection form accordingly [32]. In case of missing data or differences in safety measures depending on the region of a country, we will gather further information by searching publicly available information, i.e., official governmental documents, newspaper articles, etc., and calculate the missing stringency index [32].

The package “metafor” for R (R Foundation for Statistical Computing, Vienna, Austria) will be used to perform the meta-analysis and create meta-analytical forest plots [33]. Relevant information of the included studies (i.e., citation, subject characteristics, country, time of data collection, measurement tool, and study design) will be summarized and presented in a table. The standardized mean difference will be calculated and expressed as Hedges’s adjusted g with 95% confidence interval because PA will likely have been assessed using a variety of methods [34]. On condition of enough relevant studies being available, we will conduct random-effects meta-analyses to assess the overall effect of safety measures on (1) total PA and (2) sedentary time as well as LPA, MPA, and VPA or MVPA. In case there are not enough relevant studies available, a narrative, qualitative summary will be conducted.

The effect of heterogeneity (I²) will be calculated based on Cochrane’s Q, with a value of 0% indicating no heterogeneity and 100% substantial heterogeneity [35]. The degree of heterogeneity will be rated based on the proposed Cochrane categories [24]. Heterogeneity among the included studies will be investigated by a sensitivity analysis that takes the results of the risk of bias assessment into account.

A meta-regression will be performed to investigate the effect of the severity of the safety measures with the stringency index [32] as the independent variable and standardized mean difference in TPA as the dependent variable.

Furthermore, we will run subgroup analyses to examine the impact of self-reported (i.e., using questionnaires, interviews, diaries) vs. objectively assessed PA (i.e., through wearables, mobile applications, global positioning system), differences between age groups (i.e., using age tertiles), and the time passed after the declaration of the pandemic (i.e., with time passed transformed into categorical variables). However, for these analyses, at least three studies per group need to be available. Finally, the confidence in the cumulative body of evidence will be assessed using the GRADE guidelines [36].

To avoid the emerging of an ecological fallacy, no conclusions on individual-level effects will be drawn based on aggregated study-level results [24]. All results of the planned study will be interpreted with caution bearing in mind the issue of an ecological fallacy. Finally, the option to request individual-level data from the authors of relevant studies and subsequently perform an individual data meta-analysis will be reserved. This will depend on the number of studies after full-text screening and temporal resources.