Introduction

In Canada, as within countries across the world, SARS-CoV-2 infections were disproportionately concentrated among people and communities experiencing social and economic marginalization [1]. The response during the first year of the pandemic centered on public health measures to reduce contact rates as a means to halt the spread of SARS-CoV-2. Measures included mandates to close non-essential business alongside limits on indoor gatherings and activities. These restriction policies were met with early concerns about whom these policies could and could not reach [2, 3].

At the start of the pandemic, 60% of working-age adults in Canada were employed in jobs which could not be done remotely [4]. The front-facing jobs were also more likely to be lower-paid, classified as essential services during the pandemic, and included jobs in sales, trades, agriculture, manufacturing, transport, and the food industry [5]. The public health measures were designed as a universal policy, but individuals working in essential services would have to go to, and spend time at, their place of onsite work [1]. Estimating the direction and magnitude of mobility changes to restriction policies, at different pandemic phases, could offer insights into not only what worked–but for whom. In doing so, results could offer insights into future pandemic planning, shaping the implementation of swift non-pharmacological responses to mitigate spread but without amplifying health inequities.

Across countries, emerging data suggest socioeconomic differences in mobility and measures of ability to uptake, or “adhere to”, restriction policies. Huang et al. refer to the phenomena as the “luxury of social distancing” [6]. Among the earliest studies that used area-level mobility metrics from cell-phone data were in the United States [7–9]. These studies found that lower-income areas and counties were associated with smaller reductions in mobility following the policies [7–9]. Findings were similar in the province of Ontario (Canada): higher levels of area economic dependency (a composite measure with age-structure, work-force participation, and dependency on social assistance) was associated with reduced responsiveness to restrictive policies implemented during the first year of the pandemic [10]. The Ontario study used February 2020 as a baseline to calculate changes in mobility, and thus did not have the available data to account for expected differences in baseline mobility across socioeconomic factors [10]. While composite measures of socioeconomic factors are commonly used because they capture clustering or latent features, raw measures such as household income may provide further clarity when drawing inference. For example, pathways by which restriction policies take effect could be masked with a composite exposure measure (e.g. restriction policies would work differently for younger versus older individuals supported by social assistance). For the same reason, raw measures have been commonly used to characterize patterns of SARS-CoV-2 risks [1, 11].

Thus, to build on the previous literature, we leveraged baseline mobility data prior to 2020, and conducted a comparison of mobility response within regions with restrictions and regions without restriction policies–an important comparator to better attribute mobility changes to specific policies [8, 12]. Our aim was to describe and infer the observed mobility response to restrictions in each of two socioeconomic (raw) measures. Specifically, we used area-level cell-phone based mobility and socioeconomic data in five regions in Ontario to: (1) describe variability in the mobility response by income and by occupation in essential services following two restriction policies; and (2) estimate the extent to which these two area-level socioeconomic measures modified the mobility response, accounting for expected differences over time in the absence of restrictions using regions without restrictions as the control group.

Methods

Study design, setting, and population

We conducted a retrospective, observational study in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) recommendations [13] (S1 Checklist). The study population comprises all census tracts in the five public health units (Toronto, Peel, Halton, York, and Durham [14]) in Ontario, Canada’s most populous province and epicenters of the SARS-CoV-2 pandemic. The five public health units make up the Greater Toronto Area (population over 7.1 million [15]) which is the largest metropolitan area in Canada. Census tracts are small, relatively stable geographic units that are only used within metropolitan areas; in Canada, the population size of most census tracts is less than 7,500 persons [16]. S1 Fig shows the map of five public health units within the Greater Toronto Area. The province of Ontario is served by a single-payer, province-wide health care system and COVID-19 related policy measures, including the restrictions, were implemented at the level of the public health unit [17]. Our analyses were conducted at the unit of census tract. We excluded census tracts with missing population size or mobility data (S2 Fig).

The first case of COVID-19 was reported on January 23, 2020 in the Greater Toronto Area [18]. The first province-wide restrictions were implemented on March 17, 2020 and thus, across all five public health units (Fig 1) [19]. Restrictions were eased over the subsequent months. The second restriction was implemented on November 23, 2020 in only two public health units (Toronto, Peel; total population 4.1 million) [20]. S1 Table details the nature of the restriction policies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. COVID-19 epidemic curve and the timing of the restrictions for the Greater Toronto Area between February 23, 2020 and December 13, 2020.

The vertical dashed lines depict the timelines for two restriction policies. The first restriction was enacted March 17, 2020 across all five public health units (Toronto, Peel, Halton, York, and Durham) in the Greater Toronto Area. The second restriction enacted on November 23, 2020 in Toronto and Peel public health units and in the remaining public health units on December 14 or 26, 2020. The shaded areas in blue represent the three-week periods used in the analysis before the restrictions, and the shaded areas in red represent the three-week periods after the respective restriction was enacted. The week in which each restriction policy was enacted was excluded from the analyses, because the week of the policy was enacted cannot be categorized into the pre-restriction or post-restriction period when using weekly data. The weekly cases comprise diagnosed cases, and exclude cases among residents of long-term care homes.

https://doi.org/10.1371/journal.pgph.0002926.g001

Results

The five public health units in the current study include 1,261 census tracts, with population sizes ranging from 484 to 23,401. We excluded 7 census tracts with missing population size data and another 14 census tracts with missing mobility data (S2 Fig). Of 1240 census tracts included, the mean device coverage was lower in neighborhoods with the lower income (S4 Table).

Epidemic curves

Following the first restriction, the epidemic rapidly diverged along socioeconomic quintiles such that neighborhoods with lower-income and neighborhoods with higher proportion essential workers experienced higher number of COVID-19 cases (Fig 2A and 2B). In August 2020, the number of weekly new lab-confirmed cases were small across all socioeconomic quintiles before diverging by mid-September 2020, with the highest number of cases in lowest-income neighborhoods and in neighborhoods with highest proportion of essential workers. The number of cases remained divergent along socioeconomic quintiles prior to and following the second restriction.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Epidemic curves and mobility change stratified by neighborhood-level income and essential worker quintiles in the Greater Toronto Area, Canada (February 23, 2020 to December 13, 2020).

Panel A and B describe weekly new lab-confirmed COVID-19 cases (excluding residents of long-term care homes) by neighborhood-level income quintile (Panel A) and proportion of the working population engaged in essential services (Panel B). Panel C and D depict the weekly crude mobility metric by neighborhood-level income quintile (Panel C) and proportion essential workers (Panel D). The horizontal dashed lines in Panel C and D represent 2019 mobility as pre-pandemic mobility reference, while the solid lines represent 2020 mobility data. The vertical dashed lines depict the two COVID-19 restriction policies under examination: the first restriction enacted March 17, 2020 across all five public health units (Toronto, Peel, Halton, York, and Durham) in the Greater Toronto Area; and the second restriction enacted November 23, 2020 in Toronto and Peel public health units within the Greater Toronto Area. Income reflects the per-person equivalent income in the household. Essential services include: trades, transport, and equipment operation; sales and services; manufacturing and utilities; and resources, agriculture, and production. Neighborhood level is defined at the level of the census tract. Quintiles are weighted by neighborhood-level population.

https://doi.org/10.1371/journal.pgph.0002926.g002

Descriptive analysis: Mobility change following the first restriction

Higher-income neighborhoods consistently demonstrated greater mobility throughout 2019 and 2020 (Fig 2C). In 2019, the crude mobility ranged between 80–90% in the highest-income neighborhoods and 70–75% in the lowest-income neighborhoods. In early 2020 (pre-restriction period), the crude mobility was higher in higher-income neighborhoods (e.g., 80.6% in the highest-income and 73.7% in the lowest-income neighborhoods) (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Mobility metric of pre-restriction^a and post-restriction^b periods for the first restriction in Greater Toronto Area^c stratified by neighborhood-level^d socioeconomic measures.

https://doi.org/10.1371/journal.pgph.0002926.t001

The overall crude mobility declined sharply at the onset of the first restriction from 77.7% to 41.6% (S5 Fig; Table 1; S6A, S6B Fig depict the spatial distribution in the crude mobility before and after the first restriction at the level of the census tract, across the five public health units). All income quintiles experienced a decline in mobility following the restriction; however the higher-income neighborhoods experienced larger reductions than lower-income neighborhoods (Fig 2C). In the post-restriction period, the crude mobility was similar across income quintiles (ranging from 40.1% to 42.6%). After accounting for seasonal fluctuations, the adjusted mobility change (reduction) following the first restriction was 43.3% in the highest-income and 38.4% in the lowest-income neighborhoods with a dose-response pattern (Table 1).

The mobility pattern for proportion essential workers mimicked (inversely) the pattern with income (Fig 2C, 2D). The adjusted mobility reduction was 44.5% and 37.6% in neighborhoods with the lowest and highest proportion essential workers, respectively.

Descriptive analysis: Mobility change following the second restriction

After the large reduction following the first restriction, mobility resumed steadily since early April and reached a plateau in July through November 2020 across the five public health units (S5 Fig). During this plateau, mobility remained the lowest in the lowest-income neighborhoods (Fig 2C).

Following the second restriction in Toronto and Peel, the overall crude mobility experienced a small reduction (-2.8%) (Table 2; S6C, S6D Fig depict the spatial distribution in the crude mobility before and after the second restriction at the level of the census tract in Toronto and Peel). After accounting for seasonal fluctuations, the adjusted mobility reduction in these two public health units were small (-0.8%) (Table 2). The adjusted mobility reduction was larger in the higher-income neighborhoods (-2.7% in the highest-income neighborhoods and -0.1% in the lowest-income neighborhoods). The patterns in mobility change by essential worker quintiles were similar, with the largest reduction in adjusted mobility observed in the neighborhoods with lowest proportion essential workers (-2.3%). However, we did not observe a clear dose-response pattern across other essential worker quintiles (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Mobility metric of pre-restriction^a and post-restriction^b periods for the second restriction in two public health units (Toronto, Peel) stratified by neighborhood-level^c socioeconomic measures.

https://doi.org/10.1371/journal.pgph.0002926.t002

Difference-in-differences analysis: Mobility change following the second restriction

The difference-in-difference analyses in Model 1 (S5 Table) showed that after accounting for seasonal fluctuation, and expected changes over time in the absence of the restriction, the second restriction was associated with a small overall reduction in the adjusted mobility: -0.96% (95% confidence interval (CI): (-1.53; -0.38)).

There was effect modification by income quintiles (p < 0.05 in Model 2A; S6 Table) and by essential worker quintiles (p < 0.05 in Model 2B; S7 Table) on the relationship between the second restriction and adjusted mobility. We observed a non-linear relationship between socioeconomic measures and mobility change: neighborhoods with lower income and higher proportion essential workers dampened the magnitude of adjusted mobility change, consistent with the descriptive findings. However, given the small magnitude of the associations, a consistent dose-response pattern was not observed (Fig 3; S8 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Adjusted mobility change following the second restriction by neighborhood-level socioeconomic measures in the Greater Toronto Area.

Panel A shows the estimates of the adjusted mobility change following the second set of restrictions by income quintiles. Panel B shows the estimates of the adjusted mobility change following the second set of restrictions by essential worker quintiles. The error bars represent the 95% confidence interval. The Greater Toronto Area comprised of five public health units (Toronto, Peel, Halton, York, and Durham). Income reflects the per-person equivalent income in the household. Essential services include: trades, transport, and equipment operation; sales and services; manufacturing and utilities; and resources, agriculture, and production. Neighborhood level is defined at the level of the census tract. Quintiles (Q) are weighted by neighborhood-level population.

https://doi.org/10.1371/journal.pgph.0002926.g003

The sensitivity analysis indicated minimal influence of including the healthcare worker category on effect modification by area-level essential worker quintile, albeit there was less variability in mobility between quintiles 2 and 4 (S4 Text).

Discussion

Using an ecological study of neighborhood-level mobility measures in Ontario, Canada during 2020, we found that the first COVID-19 restriction led to large (30–40%) reductions in crude mobility. Reduction in mobility following the first restriction was largest in the highest-income neighborhoods and neighborhoods with the fewest essential workers. In contrast, a very small (<3%) reduction in crude mobility and adjusted mobility (<1%) could be attributed to the second restriction. However, there was still evidence of effect modification by income and proportion essential workers.

The large effect of the first restriction followed by a much smaller effect of the second restriction is consistent with findings that demonstrate a similar temporal saturation effect of public health measures over subsequent COVID-19 waves [28, 29]. The first restriction occurred during a time of large uncertainty. Then, even as some elements of the restrictions were lifted, there remained an overall–and relatively stable–level of mobility that was a third lower than levels in 2019. This “new normal” in 2020 reflected a shift from in-person to online shopping, education, socializing, and especially working–via occupations amenable to remote work and across businesses that were able to shift to largely remote production [30]. This time-period was also marked by a 4.8% loss of employment in the province [31]. When the second restriction was implemented to try and mitigate rising cases of COVID-19 during Ontario’s second epidemic wave, it had very little effect on mobility in the context of the “new normal” that had been established. For example, workplaces that could transition to remote work had already done so. A similar saturation effect of increasing levels of stringency of COVID-19 measures on cases was observed across multiple provinces in Canada and Colombia, with little effect of stringency measures during the second COVID-19 waves that occurred between August and December 2020 [29, 32]. Taken together, the findings highlight the importance of anticipating how the timing of restrictions and the wider context of what is already in place could influence the impact of the restriction.

The mobility changes to the first and second restriction varied by income and by occupation, and suggest a saturation effect of restrictions by socioeconomic characteristics. The mobility change was lower in lower-income neighborhoods, and in those with higher proportion essential workers. This finding was consistent with prior studies from the United States, Italy, France, Colombia, and Canada that had examined income [7, 10, 33–36], and from survey data in Canada that similarly reported lower levels of ability to “shelter in place” among lower-income households [37]. Our study separated the analyses of income from that of occupation, and confirmed a similar pattern with occupation to that of income. That is, as hypothesized, neighborhoods with higher proportion essential workers experienced a smaller mobility change to the restrictions. This smaller mobility change occurred against a pre-pandemic background wherein neighborhoods with more essential workers, and lower income, already had lower levels of mobility. The 2019 mobility patterns capture non-occupational travel, including travelling outside the home for shopping and recreation and suggests that people living in lower-income have less access to the latter at baseline [38]. This baseline difference means that individuals in lower-income neighborhoods and in neighborhoods with higher proportion essential workers may have already had fewer non-occupational activities to limit when restrictions were put into place. Indeed, mobility data from the US suggest that although people living in lower-income neighborhoods were more likely to work outside the home during the pandemic, they were less likely to have access to and visit recreational venues such as parks [38]. Such findings [38] may help also explain why despite variability in the mobility change by income and occupation, the magnitude of difference between socioeconomic levels was small–especially after the second restriction. That is, in addition to baseline lower levels of mobility (presumably from non-occupational travel), individuals may have reduced mobility in other ways to make up for having to work onsite. Although neighborhood-level income and proportion essential workers are somewhat correlated [11], they represent related but distinct social determinants of mobility. For example, the pathways by which the restriction acts (i.e. its “mechanisms of action”) on mobility across income levels is complex and, in addition to occupation-related travel, can also be related to household size (larger households offer more opportunities for cellular signals of travelling outside the home), caregiver roles (intergenerational household dependents who may require healthcare visits), access to services (e.g. grocery chains for home delivery) and access to household greenspace (backyards). Taken together, baseline (pre-pandemic) variability in mobility combined with differences in one’s ability to transition to remote work by income and by occupation highlight the saturation effect of restrictions by socioeconomic characteristics [8]. This socioeconomic saturation effect stems from the extent to which a policy or intervention is designed to reach and work across subsets of the population.

Our study was not designed to quantify the impact of differential mobility change on differential rates of SARS-CoV-2 cases. However, the small magnitude of difference in mobility reduction by socioeconomic characteristics, especially after the second restriction, suggests that differential mobility may be insufficient to explain the difference in SARS-CoV-2 cases by socioeconomic characteristics in the region. It is important to note that the association between mobility metrics and SARS-CoV-2 has been shown to vary over time [39]. Nonetheless, an implication of our findings would support the presence of residual risks of exposures and of transmission that were not addressed by the restrictions in 2020, including onsite workplace exposures intersecting with high-density and multigenerational households [1, 40]. The findings surrounding extant but small differences in mobility change alongside large disparities in SARS-CoV-2 cases suggests the need for additional, tailored approaches to address residual risks, such as outreach testing and comprehensive isolation support, vaccination, and changes in policies such as paid sick leave [41, 42].

Our study has several limitations. First, our metric only captured one dimension of mobility, and as used in previous studies [7, 8, 23]. However, alternate metrics could lead to different patterns by socioeconomic variables [43], which is particularly important when using mobility metrics to reflect changes in contact rates between people. Data are emerging from self-reported data on contacts between people and their correlation with area-level mobility metrics derived from cell-phones [37]; these correlations, particularly if they differ by socioeconomic levels, offer an opportunity for potential bias-adjustment or correction factors when using cell-phone mobility metrics to evaluate the impact of restriction policies on mobility changes, or downstream outcomes such as SARS-CoV-2 cases [38]. Thus, future work would benefit from examining other mobility measures, such as proportion of time away from home, number of trips, and distance traveled [40] as outlined in previous studies [43]. A second limitation that could underestimate variability in our outcome of interest stems from undersampling of mobility measures in lower-income neighborhoods. Third, the magnitude of change in contact rates (and the mobility metric) that is needed to reduce transmission depends on the underlying transmission potential at the time. That is, how much of a reduction is needed is different at different stages of an epidemic and in different contexts of underlying risks. A smaller reduction in contacts may be needed among high-income households with less crowding and lack of occupational SARS-CoV-2 risks. Fourth, the application of difference-in-differences analyses assume regions with and without restrictions are independent–an assumption supported by a study in Ontario that found regionally targeted interventions or restrictions had a neutral effect on inter-regional mobility [12]. Fifth, we were restricted to neighborhood-level aggregated information. This means that both the socioeconomic variables and the mobility metrics were averaged over a population and may not reflect individual-level patterns. However, individual-level survey data suggest the ecological findings from our study are consistent with self-reported data on income and mobility in response to public health restrictions in Canada [37]. Finally, we did not adjust [44] for other socioeconomic and structural factors that could play a role in the moderating effect of income, or of essential work. Future work to tease apart the anticipated complex pathways, while carefully considering their mechanisms, will contribute to further advancing insights into the effect of socioeconomic factors on restriction policies.

In summary, restrictions used in the COVID-19 public health response in 2020 demonstrated a temporal saturation effect over subsequent waves. Restrictions also demonstrated a saturation effect by income and occupation. However, the magnitude of difference between socioeconomic levels was small–especially after the second restriction. At the same time, there was a consistent and large difference in SARS-CoV-2 cases by socioeconomic characteristics which suggest residual transmission risks along socioeconomic margins. Findings highlight the need for additional approaches to reduce health inequities at the intersections of income and occupation when addressing large epidemics of novel and resurging respiratory pathogens.

Supporting information

Acknowledgments

We thank Dr. Alexander Watts (BlueDot) for insights and support on the mobility data. We thank Samantha Lo (Unity Health Toronto) and Kristy Yiu (Unity Health Toronto) for support with research coordination. SM is supported by Tier 2 Canada Research Chair (CRC) in Mathematical Modeling and Program Science. MMG is supported by a Tier 2 Canada Research Chair in Population Health Modeling.