The public health impact of private fitness expansion

Abstract

The public health, medical, and scientific research communities call for cost-effective interventions that expand access to physical activity infrastructure and services to promote urban health. This study evaluates the public health impact of the private sector expansion of fitness infrastructure in Singapore. Leveraging the phased introduction of a chain of on-demand gym facilities across residential subzones between 2020 and 2022, we find that increased access to on-demand gym facilities is associated with a decrease in monthly doctor visits per capita and overall healthcare costs. We further provide community-level evidence to support a plausible pathway where access to these gym facilities prompts residents to engage in physical activity, ultimately reducing healthcare utilization. First, increased access to on-demand gym facilities is associated with higher use of public exercise amenities, as reflected by increased parking lot utilization near public fitness facilities. This pattern suggests that overall physical activity at the community level rises following the introduction of on-demand gyms. Second, the public health impact of access to on-demand gym facilities is greater in subzones with greater access to green spaces and community fitness venues, suggesting a positive synergy between private and public fitness infrastructure. Collectively, these findings demonstrate that private-sector investments in improving physical activity opportunities can produce public health benefits.

Significance Statement.

Rising healthcare utilization and growing burden of related diseases pose public health challenges worldwide. The public demands cost-effective interventions that encourage physical activity to promote urban health as part of preventive care. This study provides evidence of the public health benefits of expanding access to private, on-demand gyms in urban areas. Leveraging the phased introduction of private gym facilities in Singapore, we find that it reduces doctor visits and may reduce healthcare costs. These findings demonstrate how private-sector investments targeting fitness infrastructure can improve urban health.

Introduction

Ensuring good health and well-being for individuals of all ages is a key sustainable development goal established by the United Nations (1). Yet, nearly 1.8 billion adults worldwide face an increased risk of various diseases due to physical inactivity (2). Rapid urbanization further exacerbates such risk by changing social environment (e.g. socioeconomic conditions), physical environment (e.g. built environment), and access to health and social services (3, 4). In response, policy makers and researchers have developed urban planning interventions to improve urban health (5–10). However, these government-led interventions often face legislative, financial, and logistical constraints, creating not only business opportunities but also social responsibilities for the private sector to address public health challenges (11, 12).

A promising yet underexplored candidate intervention lies in expanding access to private, on-demand gyms in urban areas, which may improve urban health by increasing opportunities for physical activity. The Gym Pod, a privately owned startup, exemplifies this approach: it gradually deployed its facilities across subzones (i.e. precincts) in Singapore between 2020 and 2022. Located in residential areas, shopping malls, and public parks, these on-demand gym facilities offer flexible, affordable, and private spaces for physical activity (see Figs. S1 and S2 in the Supporting Information [SI] Appendix).

Using a difference-in-differences (DiD) design, we exploit the phased introduction of Gym Pod facilities as a natural experiment to evaluate its impact on healthcare utilization. To measure healthcare utilization, we partner with a major health system in Singapore to compile subzone-month panel data from comprehensive nationwide outpatient visit records from January 2020 to December 2022.

We find that on average, increased access to Gym Pod facilities is associated with a 0.66 percentage-point decrease in the number of doctor visits per capita per month (${\beta }_1=-0.0066$, $95\mathrm{\%}$ CI $[-0.0093,-0.0039]$, $P<0.001$). This data pattern holds among diseases commonly associated with physical inactivity such as obesity and type II diabetes. Additional analyses using health expenditure data show that increased access to Gym Pod facilities is associated with a decrease in healthcare costs.

To shed light on potential pathways underlying this effect, we augment our primary data with community-level measures of the built environment and geospatial data on parking lot usage near public exercise facilities. The results emerge. First, Gym Pod access corresponds to higher engagement with nearby public exercise facilities—such as Sport Singapore (SportSG) gyms, which are government-operated community fitness centers—as proxied by increased parking activity around these locations. Although parking is only an indirect measure, this pattern indicates a positive community-level spillover, suggesting that greater access to on-demand gyms encourages broader participation in physical activity. Second, the health impact of Gym Pod access is stronger in subzones with greater availability of parks, green spaces, and community clubs, suggesting a positive synergy between private and public fitness amenities in promoting physical activity.

Materials and methods

Context

The Gym Pod is a commercial fitness startup with a strong social impact orientation. Its mission is to “make exercising more accessible for busy urbanites [sic], affordable for students and young working adults, and provide privacy for people who are shy at gyms” (13), aligning with the public health goals of Singapore by providing convenient and affordable access to fitness facilities (14).

Each Gym Pod facility operates as an on-demand, reservation-based micro-gym built inside a yellow retrofitted shipping container. These units provide users with private access to a fitness space 24 h a day. The company offers two pricing models: a pay-per-use option and a monthly subscription plan. Pay-per-use sessions range from $6 to $12 for a 30-min session, while monthly subscribers receive additional benefits, including 30-day advance reservations, flexible rescheduling, and tiered discounted session rates. Additional details about the context, funding structure, and key stakeholders are provided in the SI Appendix, Intervention background and funding source section.

Site placement followed a structured evaluation of population density, public transport access, parking availability, built environment characteristics, and local housing profiles. Consequently, many Gym Pods were installed near public housing estates, workplaces, shopping malls, and public parks—locations chosen to minimize access barriers and encourage regular physical activity among community members.

Data sources and preparation

We construct our data from a variety of sources. We compiled operating locations of all Gym Pod facilities from the official company website^a and verified opening dates with the management team. Between 2020 and 2022, the Gym Pod gradually introduced its facilities across 26 subzones in Singapore.^b The rollout process is provided in Table S1.

To measure healthcare utilization, we worked with a national health system to obtain 12,642,330 visit records of 1,070,855 adults (about 32% of Singapore’s adult population). The ethics approval was granted by the Domain Specific Review Board (DSRB) National Healthcare Group (approval ref no: 2022/00132). Each observation includes a de-identified patient identification (ID), visit ID, disease classification under the 10th Revision of the International Classification of Diseases (ICD-10), and the patient’s residential subzone. Building on prior medical research (15–26), we identified five disease categories directly associated with physical activity (see Table S2). We excluded nonresidential subzones with populations below 1,000 and only kept completed outpatient visit records, which yielded 3,457,758 medical visit records from 284,755 patients across 199 subzones.

Due to data sparsity at the individual level, we aggregated all medical records to the subzone-by-month level where the outcomes include both overall healthcare utilization and healthcare utilization by each disease category. In addition, our partnership with the national Smart City program enabled us to collect data on transportation, environmental, housing, economic, and demographic characteristics of each subzone over time (see Table S3). Further, we collected the number of coronavirus disease 2019 (COVID-19) cases in each subzone over time from the Singapore Ministry of Health.

Finally, we merged data related to phased introduction of Gym Pod facilities with healthcare utilization to construct subzone-month panel data spanning January 2020 to December 2022. Each observation captures the absence or presence of Gym Pod facilities and the corresponding health outcomes for the same spatial and temporal unit. Details of data screening, aggregation, and processing procedures are summarized in Fig. S3.

Natural experiment evaluation

Because Gym Pod facilities were gradually introduced in different subzones at different times, this phased introduction generates variation in the availability of private fitness infrastructure and services across subzones over time. This Gym Pod-initiated intervention provides a suitable setting for a natural experiment evaluation when randomized controlled trials are not feasible. Next, we outline our approach specified as a two-way fixed effects (TWFE) model and follow recent public health literature on natural experiment evaluations to guide our discussion (27–29).

Our main specification is the TWFE model:

$$Y_{it}={\beta }_0+{\beta }_1{\mathrm{GymPod}}_{it}+{\gamma }_i+{\delta }_t+{ϵ}_{it},$$ (1)

where our dependent variable, $Y_{it}$, is the number of doctor visits per capita (hereafter referred to as the visit rate) in subzone i during year-month t. $Y_{it}$ was computed as the total number of completed outpatient doctor visits by residents of subzone i during year-month t divided by the subzone’s population at the same time. ${\mathrm{GymPod}}_{it}$ is a binary indicator that captures whether a Gym Pod facility was introduced in subzone i during year-month t (1 = Yes; 0 otherwise). When multiple facilities entered a subzone, we focused on the introduction of the first facility to identify treatment timing. ${\gamma }_i$ and ${\delta }_t$ denote subzone fixed effects and year-month fixed effects, respectively; ${ϵ}_{it}$ is the error term. ${\beta }_1$, the TWFE estimate, aims to capture the average treatment effect of increased access to Gym Pod facilities on the visit rate.

An important advantage of the TWFE model is the ability to include subzone fixed effects (${\gamma }_i$) and year-month fixed effects (${\delta }_t$) (30, 31). Subzone fixed effects capture all time-invariant confounders across subzones such as their baseline health, environmental, and socioeconomic conditions (e.g. neighborhood and built environment). Year-month fixed effects control for time-varying shocks that are common across subzones such as nationwide changes, seasonal patterns, or pandemic-related disruptions.

${\beta }_1$ is identified using within-subzone variation in healthcare utilization before and after the introduction of Gym Pod facilities. It is important to note that ${\beta }_1$ is a weighted average of several comparison groups in the context of staggered rollout: (i) treated versus never-treated subzones; (ii) early-treated versus later-treated subzones; and (iii) later-treated versus already-treated subzones. An illustrative figure of these comparisons is shown in Fig. S4. Additional details on empirical strategy are provided in SI Appendix, Model explanations section.

Results

Average treatment effect

Table 1 presents the average treatment effect of access to Gym Pod facilities on visit rate. Results in column (1) show that access to Gym Pod facilities is associated with a 0.69 percentage-point decrease in visit rate (${\beta }_1=-0.0069$, $95\mathrm{\%}$ CI $[-0.0095,-0.0043]$, $P<0.001$). However, it is important to account for potential time-varying subzone-specific confounders due to the COVID-19 pandemic and economic development. Thus, we include in model 1 a rich set of transportation, environment, housing, economic, demographic, and COVID-19 control variables that proxy time-varying subzone-specific unobservables. These control variables are informed by a directed acyclic graph (DAG) framework (SI Appendix, Directed acyclic graph section, Fig. S5). Column (2) shows a consistent and slightly attenuated effect: access to Gym Pod facilities is associated with a 0.66 percentage-point decrease in visit rate (${\beta }_1=-0.0066$, $95\mathrm{\%}$ CI $[-0.0093,-0.0039]$, $P<0.001$).

Table 1.

Effect of Gym Pod access on healthcare utilization.

Dependent variable (DV): visit rate	(1)	(2)
GymPod	$-$ 0.0069a	$-$ 0.0066a
	(0.0013)	(0.0014)
Observations	7,164	7,164
No. of subzones	199	199
Time-varying subzone-level controls	No	Yes
Control for COVID-19	No	Yes
Subzone fixed effects	Yes	Yes
Year-month fixed effects	Yes	Yes
Adjusted $R^2$	0.7949	0.8146

This table shows average treatment effect estimates using a two-way fixed effects regressions that include subzone and time-period fixed effects without time-varying controls (column 1) and with time-varying controls (column 2). Robust standard errors are reported in the table. ${ }^{\mathrm{a}}P<0.001$.

Robustness checks

We conducted a series of robustness checks to strengthen the credibility of our findings. First, the parallel trends assumption must hold for DiD to enable valid causal inference. This assumption requires that in the absence of Gym Pod facilities, the average trends in doctor visit rates for treatment and control subzones would have evolved similarly over time. In our context, this means that subzones where a Gym Pod facility was introduced at different points in time or not introduced at all would have followed similar pretreatment trends in healthcare utilization prior to the introduction. To evaluate this, we estimate an event-study specification and plot the coefficients of relative time indicators around the Gym Pod introduction. As shown in Fig. S6, the preintervention coefficients are close to zero and statistically indistinguishable from zero, indicating no systematic differences in healthcare utilization trends between treatment and control subzones prior to Gym Pod entry.

Second, the phased introduction of these facilities is not random but determined by the Gym Pod team, potentially causing selection bias. To address this, we incorporated the site selection criteria used by The Gym Pod team—population density, public transport availability, parking availability, built environment, and housing characteristics—into our regression model. To further mitigate the concerns about nonrandom treatment assignment, we applied the inverse probability of treatment weighting (IPTW) approach to weight subzones based on the site selection criteria mentioned above and re-estimated the model. The estimates remain consistent (Table S4).

Third, recent literature in econometrics suggests that staggered DiD models may suffer from potentially problematic comparisons associated with staggered rollout (28, 32). Accordingly, we follow the methodological guidance in the literature to use Goodman-Bacon decomposition (33), Callaway and Sant’Anna estimator (34), and stacked event study (35). Results suggest that problematic comparisons associated with staggered rollout are minimal in our context (see Fig. S7 and Tables S5 and S6).

Fourth, we examine the possibility of lagged treatment effects, acknowledging that changes in health behaviors and outcomes may unfold gradually following Gym Pod access. As clarified earlier, the estimated DiD coefficients represent the average monthly treatment effect per subzone over the study period (January 2020–December 2022), rather than short-term effects within a narrow post-treatment window. To assess temporal persistence and delayed responses, we re-estimate the model including treatment indicators lagged by 1 to 3 months. The results, reported in Table S7, remain negative and statistically significant across all specifications.

Fifth, to ensure our findings are not confounded by pandemic-related disruptions, we re-estimate the TWFE model after excluding the lockdown period in Singapore (April–May 2020). The results, reported in Table S8, remain consistent.

Sixth, we compare postintervention visit rates for exercise-related and nonexercise-related diseases (as summarized in Table S9) to robustly assess the potential health impact of Gym Pod facilities. Table S9 reports the effect of Gym Pod access on the difference in healthcare utilization between exercise-related and nonexercise-related diseases. Gym Pod introduction is associated with a 0.35 percentage-point larger decrease in visit rates for exercise-related diseases relative to nonexercise-related diseases (${\beta }_1=-0.0035$, $95\mathrm{\%}$ CI $[-0.0051,-0.0019]$, $P<0.001$).

Effects across disease categories

Next, we examine whether the health impacts of Gym Pod access differ across disease categories. Fig. S8 and Table 2 summarize the estimated heterogeneous effects. Across all five major disease categories: mental and behavioral disorders (abbreviated as mental disease), diseases of the circulatory system (abbreviated as circulatory disease), endocrine, nutritional and metabolic diseases (abbreviated as endocrine disease), diseases of the musculoskeletal system and connective tissue (abbreviated as musculoskeletal disease), and diseases of the respiratory system (abbreviated as respiratory disease), we observe negative and statistically significant effects on doctor visit rates. The largest reduction occurs for circulatory and endocrine disease categories.

Table 2.

Heterogeneous effects of Gym Pod access across disease categories.

	DV: Visit rate across disease categories
	(1) Mental disease	(2) Circulatory disease	(3) Endocrine disease	(4) Musculoskeletal disease	(5) Respiratory disease
GymPod	$-$ 0.0004a	$-$ 0.0042a	$-$ 0.0049a	$-$ 0.0019a	$-$ 0.0013a
	(0.0001)	(0.0010)	(0.0012)	(0.0005)	(0.0002)
Observations	7,164	7,164	7,164	7,164	7,164
No. of subzones	199	199	199	199	199
Subzone-level controls	Yes	Yes	Yes	Yes	Yes
Control for COVID-19	Yes	Yes	Yes	Yes	Yes
Subzone fixed effects	Yes	Yes	Yes	Yes	Yes
Year-month fixed effects	Yes	Yes	Yes	Yes	Yes
Adjusted $R^2$	0.8017	0.7967	0.7923	0.8192	0.8134

This table reports the heterogeneous effects of Gym Pod access across different disease categories. The outcome variable is the monthly visit rate per subzone for various disease categories. Robust standard errors are reported in the table. ${ }^{\mathrm{a}}P<0.001$.

A closer look at specific disease conditions (Fig. S9 and Table 3) reveals that the reductions are particularly pronounced for hypertension and type II diabetes, followed by obesity and ischemic heart disease. These results highlight that the health benefits of Gym Pod access are concentrated in metabolic and cardiovascular diseases that are responsive to improvements in physical activity (19–23).

Table 3.

Heterogeneous effects of Gym Pod access across specific diseases.

Panel A	DV: Visit rate of specific diseases
	(1) Mood disorders	(2) Obesity	(3) Type 2 diabetes	(4) Hypertension
GymPod	$-$ 0.0002a	$-$ 0.0004a	$-$ 0.0025${ }^{\mathrm{b}}$	$-$ 0.0038${ }^{\mathrm{b}}$
	(0.0001)	(0.0002)	(0.0006)	(0.0009)
Observations	7,164	7,164	7,164	7,164
No. of subzones	199	199	199	199
Subzone-level controls	Yes	Yes	Yes	Yes
Control for COVID-19	Yes	Yes	Yes	Yes
Subzone fixed effects	Yes	Yes	Yes	Yes
Year-month fixed effects	Yes	Yes	Yes	Yes
Adjusted $R^2$	0.7983	0.7116	0.7708	0.7779

Panel B	DV: Visit rate of specific diseases
	(5) Ischemic heart diseases	(6) Fatigue or malaise	(7) Sleep disorders
GymPod	$-$ 0.0008${ }^{\mathrm{b}}$	$-$ 0.0001${ }^{\mathrm{c}}$	$-$ 0.0001${ }^{\mathrm{b}}$
	(0.0002)	(0.0000)	(0.0000)
Observations	7,164	7,164	7,164
No. of subzones	199	199	199
Subzone-level controls	Yes	Yes	Yes
Control for COVID-19	Yes	Yes	Yes
Subzone fixed effects	Yes	Yes	Yes
Year-month fixed effects	Yes	Yes	Yes
Adjusted $R^2$	0.7955	0.7205	0.7546

This table reports the heterogeneous effects of Gym Pod access across specific diseases. Panel A and Panel B are estimated using the same dataset and correspond to the upper and lower sections of the table, respectively. The outcome variable is the monthly visit rate per subzone for each disease condition. Robust standard errors are reported in parentheses. ${ }^{\mathrm{a}}P<0.05$, ${ }^{\mathrm{b}}P<0.001$, ${ }^{\mathrm{c}}P<0.01$.

Economic implication

To assess the economic implications of Gym Pod access, we collected an additional dataset containing detailed medical expense records, with each entry capturing the total cost of a doctor visit—including consultation fees, medications, equipment usage, and diagnostic examinations. For each ICD-10 disease category, we observe a range of cost estimates. To ensure robustness across different distributional assumptions, we calculate medical expenses using three statistical measures: the mean, median, and mode. The full methodology is provided in SI Appendix, Materials and methods section. As shown in Table S10, back-of-the-envelope estimates indicate average per capita medical expenses are lower by Singapore dollar (SGD) $3.1154({\beta }_1=-3.1154$, $95\mathrm{\%}$ CI $[-3.8927,-2.3381]$, $P<0.001$), SGD 2.2593 (${\beta }_1=-2.2593$, $95\mathrm{\%}$ CI $[-2.8229,-1.6958]$, $P<0.001$), and SGD 1.6903 (${\beta }_1=-1.6903$, $95\mathrm{\%}$ CI $[-2.1120,-1.2686]$, $P<0.001$) under the mean-, median-, and mode-based estimations, respectively. Each corresponds to an ∼23% reduction in healthcare spending following the introduction of Gym Pod facilities (see Table S11 for results using the log-transformed dependent variable). These results suggest that expanding access to private, on-demand fitness infrastructure may yield tangible economic benefits, as reflected in its consistent association with lower medical expenditures across alternative estimation approaches.

A plausible pathway: Gym Pod access increases physical activity

We next explore a plausible pathway underlying the observed reduction in healthcare utilization, focusing on whether the introduction of Gym Pod facilities stimulates broader community-level physical activity beyond direct users.

First, we examine changes in parking lot utilization near public sports facilities following Gym Pod rollout. Carpark usage around SportSG venues—a nationwide network of government-managed fitness facilities—serves as a behavioral proxy for community physical activity (see SI Appendix, Utilization rate of parking lots around SportSG facilities section for details). As shown in Table S12, Gym Pod access is associated with a 2.81 percentage-point increase (${\beta }_1=0.0281$, $95\mathrm{\%}$ CI $[0.0059,0.0503]$, $P<0.05$) in parking lot utilization near these venues, even after controlling for subzone characteristics.

Second, we find that the association between Gym Pod access and healthcare utilization is stronger in subzones with greater access to parks, green spaces, and community clubs, indicating that supportive built environments moderate the treatment effect of Gym Pod access (see Table S13 and Fig. S10).

Together, these analyses support that Gym Pod access reduces healthcare utilization by promoting sustained increases in physical activity within the surrounding community, especially where complementary public exercise resources are available.

Conclusion

This study provides empirical evidence that private-sector expansion of fitness infrastructure has potential to promote urban health: access to fitness facilities deployed by the Gym Pod is associated with a reduction in monthly doctor visits related to physical inactivity. Furthermore, we shed light on why we observe these effects. First, Gym Pod access is associated with higher usage of nearby public exercise facilities as inferred by the increased parking activity around these sites. Second, the health impact of access to Gym Pod facilities is larger in subzones with greater access to green spaces and community clubs, pointing to a positive synergy between private and public exercise amenities. This supports that visible access to fitness infrastructure may act as a salient nudge to raise health consciousness.

This work has important implications. Despite its commercial nature, the private sector is critical to creating innovative solutions and providing resources to address global health challenges. We highlight that the private-sector investments in expanding fitness infrastructure can serve as a promising lever to promote urban health. Such investments can also complement government-initiated urban planning interventions (e.g. green space, parks, transportation) and campaigns. For example, the Centers for Disease Control and Prevention (CDC) in the United States introduced the “Active People, Healthy Nation” campaign in 2020, aiming to increase physical activity among 27 million people by 2027 (36). Policy makers can build on these findings to ensure equitable and inclusive access to safe exercise venues in residential areas (37).

Our findings also highlight opportunities for collaboration between the public and private sectors. Private-sector innovation can complement public resources by developing compact, technology-powered gym facilities and tailoring services to meet community needs. Governments can incentivize private entities to expand operations into underserved areas through subsidies, tax incentives, or joint projects. These partnerships can bridge gaps in public health infrastructure while fostering sustainable urban development. In Singapore, national strategy leans on public–private collaboration. For example, Singapore’s Healthier SG initiative explicitly engages employers and community partners, a policy signal that the private sector is critical to helping deliver community health outcomes (38).

This study has limitations, which offer valuable opportunities for future research. First, although we utilize a large and comprehensive dataset, it lacks key demographic information such as gender, socioeconomic status, and educational background, factors at individual level that could inform heterogeneity in health improvements across population groups and the underlying mechanisms. Exploring these aspects could provide deeper insights into the actual impact of private gyms on health outcomes and how various usage patterns contribute to these effects. Second, because our analysis is not based on randomized clinical trials, the findings should not be interpreted as causal. Third, potential confounding from unobserved community initiatives may not be fully accounted for. Fourth, the utilization rate of carpark lots near SportSG sports facilities serves only as a proxy for community physical activity, which may also reflect increased car usage. Third, it is worth noting that the transportability of this program evaluation may be limited, as the study is conducted in Singapore, a highly developed city-state whose institutional, economic, and infrastructural context is distinct from that of many low- and middle-income cities or countries. Finally, while the expansion of Gym Pods may encourage physical activity, broader trade-offs exist within the field of public health. Future research could benefit from closer collaboration with the community, relevant government agencies, or the private gym sector to address these gaps and further refine our understanding of the role of residential exercise facilities in public health.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

This work was supported by funds from the Ministry of Education - Singapore (S.T. RT12/21), the AI Singapore (S.T. AISG3-GV-2021-006), the Ministry of National Development, Singapore (K.H.G. COT-CityScan-2020-1), and the Social Science and Research Council, Singapore (K.H.G. SSRC2024-SSHRTG-012).

Author Contributions

M.M.: data curation; software; formal analysis; validation; writing—original draft. S.T.: conceptualization; funding acquisition; project administration; writing—review and editing. Y.C.: conceptualization; writing—review and editing. K.H.G.: resources; funding acquisition; project administration.

Data Availability

Due to the confidentiality clause of the local healthcare system and government requirements under the Human Biomedical Research Act 2015 (Singapore Statutes Online), we are unable to share patient-level and subzone-aggregated health visit data publicly. Requests for access to these data should be directed to the corresponding author, who will coordinate with the relevant authorities to determine the feasibility of sharing such data under appropriate confidentiality agreements. The STATA code are available at https://github.com/MMJ42/PNAS-NEXUS-Gym-Pod-Paper-Stata-Code.