The MOOSE (Meta-analysis Of Observational Studies in Epidemiology) [26] and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 [27] guidelines were followed for reporting this study (Multimedia Appendix 1). The protocol for this study was prospectively registered in the PROSPERO database (CRD42021291684).

A preliminary search was conducted on PubMed, EMBASE, and Web of Science to retrieve studies that reported daily step data both before and during the confinement period of the COVID-19 pandemic in the general adult population published before November 22, 2021. Further, an updated search was performed on February 11, 2023, to add recently published articles. The detailed search strategy is listed in Multimedia Appendix 2 [28-31]. The International Standard Randomized Controlled Trial Number Register (ISRCTN) and the World Health Organization COVID-19 database were also searched for potential gray literature. The reference lists and citing articles of relevant literature on this topic were also checked to find possible studies that could be included in this research. The language of publication was not restricted.

We included observational studies that reported daily steps before and during the confinement period of the COVID-19 pandemic in the general adult population. The reference period was restricted to ≤2 years before the time points when the government implemented the confinement measures. Further eligibility criteria are listed in Textbox 1.

Two independent reviewers (LS and WR) screened all titles and abstracts. The full text of potentially relevant literature was further evaluated. A third reviewer (ZW) arbitrated the decision if a consensus was not reached.

Two authors (LS and WR) extracted the following data independently: study design, country and continent (Africa, Asia, Europe, North America, South America, and Oceania) where the study was conducted, the sample size used for daily step analysis, participant characteristics (ie, mean age and female ratio), monitoring devices used to measure daily steps, the algorithm used to analyze daily steps, data collection duration, and COVID-19 restriction measures by the government at the time of data collection. The outcomes of interest were the total number of daily steps before and during the confinement period of COVID-19 and the percentage change between the 2 periods. If not reported, the mean (SD) of the outcome was estimated based on the methods provided by the Cochrane Handbook [32]. If data needed to be extracted from graphs, the GetData digitizer was used [33]. This tool has been validated and has good accuracy and precision for data extraction; it has been used in meta-analysis by previous studies [34]. Disagreements, if any, were resolved by consulting with a third reviewer (ZW). When any desired information related to daily steps was missing or unclear, the corresponding author was contacted via email to request or confirm the relevant data, as appropriate. If the data could not be obtained, the study was excluded from the meta-analysis.

The modified Newcastle-Ottawa Scale (NOS) for applicable observational studies was used to assess the methodological quality and risk of bias in each study [28-30]. Specifically, the modified NOS consists of 3 domains: selection (maximum 3 stars), comparability (maximum 2 stars), and outcome (maximum 2 stars). Therefore, 1 study can be awarded a maximum of 7 stars. Studies with 6 to 7 stars were rated as high quality, those with 5 stars were rated as moderate quality, and those with 4 or fewer stars were rated as low quality [29]. Two reviewers (LS and ZW) rated the methodological quality of each study independently, and discrepancies, if any, were discussed for consensus or resolved by consulting a third reviewer (YW).

A random effects meta-analysis was conducted using Stata software (version 14.0; StataCorp). The change in daily steps was calculated using the mean difference (MD) with the 95% CI between the mean (SD) of daily steps before and during the confinement period of COVID-19. The Cochran Q statistic of the inconsistency index (I²) was used to test statistical heterogeneity across studies [32,35]. An I² value of more than 75% indicated high heterogeneity [32]. Publication bias was assessed with a funnel plot and further evaluated with the Egger test [36]. To test the robustness of the findings on the primary outcome (ie, the number of daily steps), we performed 3 sensitivity analyses. First, we conducted an analysis that eliminated each study one at a time. Second, we performed an analysis that removed the studies rated as having low methodological quality. Third, we performed an analysis that excluded the studies with a small sample size (less than 140) based on the recommendations of Dunton and colleagues [37]. Finally, we also conducted subgroup analyses according to country, continent, and gender.

A total of 6937 unduplicated records were preliminarily retrieved, and 214 potentially eligible studies were identified after the title and abstract screening. Then, 5 studies were excluded because the full text was unavailable, and full text assessments were made of the remaining 209 articles. Another 189 articles were eliminated. Of these, 151 studies did not report relevant outcomes (ie, daily steps), 14 did not provide necessary data for meta-analysis, 13 focused on irrelevant populations, 5 were not observational studies, 3 were conducted with financial incentives, 2 reported duplicated data, and 1 was a conference abstract. The remaining 20 studies were included in the meta-analysis [19,20,37-54]. Figure 1 shows the PRISMA flow diagram.

The included studies were conducted in Australia [38], Canada [39], China [19,40-43], France [38], Italy [44], Japan [45-48], Norway [49], Singapore [20], South Korea [50], Spain [51], the United Kingdom [38,52], and the United States [37,38,53,54], and all of them investigated the change in daily steps in the early stage of COVID-19 (from January to September 2020). The reference pre–COVID-19 period was from January 2019 to March 2020. The detailed duration of observation for each study is shown in Table 1. Nine studies measured daily steps with smartphones [19,37,40,41,43,45,46,50,54], 8 with wearable activity trackers [20,38,39,47,48,51-53], and 2 with both wearable activity trackers and smartphones [44,49]; 1 study did not specify the measurement device used in the survey [42]. One study provided 4 groups of data from 4 countries separately [38]. Therefore, the 20 studies included a total of 19,253 participants, and 23 sets of data were included in the meta-analysis. Additional study characteristics are summarized in Table 1.

Thirteen studies (65%) were rated as having moderate to high methodological quality (Multimedia Appendix 2, Table S1). Considering the widespread weaknesses of small sample size, imbalanced gender ratio, limited age range of the study sample, and poor nationwide representativeness, only 2 studies were rated as having national representativeness of the exposed cohort (ie, samples during the confinement period of COVID-19). Thus, 10% of the included studies received 3 stars in the selection dimension, 30% received 2 stars in the comparability dimension, and 60% received 2 stars in the outcome dimension.

As shown in Figure 2, the proportion of studies with subjects with optimal daily steps (ie, ≥7000 steps per day) declined from 70% before the pandemic to 25% during the confinement period. The change in daily steps between the 2 periods ranged from –5771 to –683. The pooled MD from the random effects pooled analysis was –2012 (95% CI –2805 to –1218; Figure 3), with significant heterogeneity among studies (P<.001; I²=97.6%). The funnel plot shows a symmetrical horizontal distribution for each study relative to the vertical line, suggesting the absence of reporting bias (Multimedia Appendix 2, Figure S1). The Egger publication bias indicated a lack of publication bias (Multimedia Appendix 2, Figure S2). Sensitivity analysis, performed by omitting each study one by one, revealed that the change in daily steps was not significantly altered by any individual study (Multimedia Appendix 2, Figure S3). Moreover, the results remained robust in sensitivity analyses after removing low-quality studies (MD=–1714, 95% CI –2809 to –618) and after removing small-sample studies (MD=–2024, 95% CI –2836 to –1212). The percentage change in daily steps varied from –69.6% to –10.6% across different studies (Figure 4).

Country-specific analyses (Figure 5) showed that the most significant decline in daily steps occurred in subjects from Spain (MD=–5771, 95% CI –7519 to –4023), followed by Italy (MD=–4990, 95% CI –5819 to –4161) and Singapore (MD=–3813, 95% CI –4077 to –3549). The decline in daily steps was less in subjects from Canada (MD=–1012, 95% CI –1896 to –128), Japan (MD=–1051, 95% CI –1286 to –816), and Norway (MD=–1131, 95% CI –1998 to –264). In terms of continent, European subjects presented the most significant reduction in daily steps but with a wide CI (MD=–2821, 95% CI –4552 to –1089), followed by subjects from Oceania (MD=–2714, 95% CI –3548 to –1879) and Asia (MD=–1931, 95% CI –2940 to –921), while subjects from North America presented the smallest decline in daily steps (MD=–1626, 95% CI –2403 to –850). Of the studies that provided gender-specific data [20,41,43,45-47,52], no apparent difference was observed in the change in daily steps between men (MD=–2105, 95% CI –3770 to –439) and women (MD=–1748, 95% CI –2965 to –531).

Our study quantified the change in daily steps during the confinement period of COVID-19. The proportion of studies with subjects with optimal daily steps (ie, ≥7000 steps/day) declined from 70% before the pandemic to 25% during the confinement period, and the average number of daily steps was reduced by 2000. Our results also suggest a possible inequality in physical activity across countries, whether before or during the pandemic. However, given that most included studies lacked sufficient nationwide representativeness, this difference should be interpreted cautiously. Furthermore, there was no apparent gender difference in the decline in daily steps before and during the confinement period of COVID-19.

The population-level trends in daily steps during the confinement period of COVID-19 may reflect public panic about the virus and adherence to confinement measures (ie, physical distancing and sheltering in place) [55,56]. Our findings are not surprising given that COVID-19 restriction measures deprived the general population of many opportunities for physical activity, such as commuting, shopping, sports, and vacations or trips. In addition, the observed differences in the change in daily steps across studies may reflect regional variations in enforcement and behavioral changes [57,58]. Socioeconomic inequalities might also influence such variations among regions and disparities in the ability to engage in or access recreational physical activity within areas [59].

Daily steps are strongly associated with health outcomes [7-16]. Recent studies indicate that over 7000 steps per day is the critical physical activity standard for minimizing the risk of all-cause mortality [12,13]. Moreover, the risk of cardiovascular events increases by 8% for every 2000 steps per day decrement [14]. We found that daily steps declined by 2000, and half of the included studies reported that average daily steps declined to below 7000 during the confinement period of the COVID-19 pandemic. Such findings imply that the pandemic further exacerbated the ever-increasing prevalence of low-level physical activity [60,61], emphasizing the necessity of adopting appropriate measures to reverse this trend. Furthermore, this meta-analysis focused on observational studies based on monitoring devices. Since monitoring devices may encourage users to engage in physical activities [6], the number of daily steps during the confinement period might have been even lower in the population than was currently observed in the samples. In addition, the impact of COVID-19 on daily steps may persist in the long term. Although some researchers report that daily steps gradually recovered to the prepandemic level after the lifting of restriction measures [49,62], more studies argue that worldwide daily steps remain depressed compared with the prepandemic level (ie, the immediate 2 years before the outbreak of COVID-19) [59,63-66]. Thus, the consequences of long-term physical inactivity, which may be further exacerbated, should be addressed.

Our finding of a reduction in daily steps during the confinement period of the pandemic is consistent with findings from previous studies using different physical activity metrics [67,68]. An earlier meta-analysis found that sedentary time significantly increased, by 126.9 minutes per day in adults and by 46.9 minutes per day in the elderly, during the confinement period of the pandemic [67]. Another systematic review reported that most of the included studies revealed a decrease in physical activity in most participants [68]. All studies showed an increased sedentary time in healthy adults and those with medical conditions during the lockdown period [68]. According to this evidence, public health messages about staying active during the pandemic appear to have had little effect on the population’s engagement in physical activity [69]. In addition, 2 studies explored gender differences in the change in physical activity during the COVID-19 pandemic but had inconsistent conclusions [24,25]. One study found that men had more clearly decreased moderate-vigorous physical activity and increased sedentary behaviors than women during the pandemic [24]. However, another study failed to show significant differences between men and women in self-reported decline in physical activity [25]. Our meta-analysis adopted a primary indicator of physical activity, ie, daily steps, and found that daily steps declined substantially during the COVID-19 confinement period. Such a decline varied by region but not by gender. These results are consistent with previous studies [24,25,67,68].

Based on data collected from 187 countries and districts by a smartphone app, Tison and colleagues [59] found that mean daily steps declined by 5.5% (287 steps) within 10 days of the pandemic declaration by the World Health Organization and by 27.3% (1432 steps) within 30 days. The authors also reported noticeable differences across regions; for example, Italy, France, Brazil, and Iran exhibited a more than 40% maximum decrease, while Sweden exhibited a maximum reduction of daily steps of less than 10% [59]. Furthermore, regional differences were significant. For example, several studies showed that the decline in daily steps during confinement varied considerably among Chinese adults from different provinces [40,41,43]. One nationwide study in Japan found that daily steps declined significantly after the declaration of a state of emergency in urban areas but not rural areas [62].

Several limitations should be acknowledged. First, daily step data are lacking from Africa and South America, limiting the generalizability of our findings. Second, when comparing regional differences across countries, we could not rule out potential sampling bias, because most of the included studies lacked sufficient nationwide representativeness. Therefore, future studies are needed to verify our findings. Third, since long-term data from the period that confinement was lifted are not suitable for inclusion in a meta-analysis, we could not directly assess the long-term impact of COVID-19 on physical activity levels using meta-analysis techniques. As more published data become available [59,63-66], this will be an important direction for future studies. Fourth, we did not examine the impact of reduced daily steps on health outcomes, owing to a lack of relevant data. Future studies are expected to address this crucial issue. Finally, there was considerable study heterogeneity. Potential explanations include discrepancies in confinement measures across countries or districts and in the adherence to policy of residents, as well as unclear or different time frames over which the average daily steps were calculated, the possibility that monitoring devices were changed in the different reporting periods, and different durations of observation.

In summary, we found that daily steps declined substantially during the COVID-19 confinement period. Our findings indicate that COVID-19 further exacerbated the ever-increasing prevalence of low levels of physical activity, emphasizing the necessity of adopting appropriate measures to reverse this trend. Further research is required to monitor the consequence of long-term physical inactivity.