Longitudinal study examining the relationship between physical activity and psychiatric hospitalizations in canadian adolescents and young adults utilizing record linkage

Abstract

Overwhelming evidence suggests that physical activity among youth can prevent mental illness; however, few studies have explored its effects on mental healthcare utilization. This study aimed to examine the longitudinal associations between physical activity among Canadian adolescents and young adults (AYAs; 12–24 years) and incidence of psychiatric hospitalizations. Physical activity was measured in the 2001–2014 Canadian Community Health Survey (CCHS) and was linked to the Discharge Abstract Database. Negative binomial regression analyses were performed on each CCHS cycle to obtain incidence rate ratios (IRRs) for psychiatric hospitalizations by level of physical activity, which were subsequently meta-analyzed to obtain pooled estimates. In total, 96,100 participants were recruited across eleven cycles. Adolescents were more physically active (52%) compared to young adults (39%). The most common cause of hospitalization was mood or anxiety disorders (38%). Fully adjusted models found that moderately active (IRR = 1.30; 95% CI: 1.02–1.66; p = 0.01) and inactive (IRR = 1.33; 95% CI: 1.06–1.66; p = 0.01) participants had higher rates of psychiatric hospitalizations compared to active participants. Our findings suggest that lower levels of physical activity among AYAs were associated with an increased incidence of psychiatric hospitalizations, providing valuable insights for stakeholders and laying the groundwork for future research.

Introduction

The consequences of mental illness are multifaceted; negatively impacting the economy, the healthcare system, and the lives of millions^1–5. Approximately 6.7 million Canadians experience mental illness at a given time, costing the Canadian economy over $50 billion annually^1,5. Mental illness carries a myriad of personal burdens through stigmatization, reduced quality of life, greater risk for morbidity and mortality, and negative familial impacts^6–10. These issues are especially pronounced among adolescents and young adults (AYAs), who experience the highest rates of mental illness and are among the largest users of mental health services in Canada^11–16. The high incidence of mental disorders among AYAs is particularly concerning as psychological problems tend to persist across an individual’s life course, potentially leading to high amounts of future personal and societal burdens^5,17–19. Therefore, it is imperative to invest in early mental health promotion and mental illness prevention strategies^20–23.

Over the past several decades, physical activity has emerged as a promising modifiable risk factor for the prevention of mental disorders in young people^24–28. This rich body of evidence has produced several systematic reviews and meta-analyses demonstrating the many benefits of increased physical activity on quality of life, pro-social behaviours, and cognitive functioning, and decreases in mood and anxiety disorders, psychotic disorders, substance use, and emotional problems^24–27. One review elucidated a causal link between increased adolescent physical activity and higher cognitive functioning, and a partial causal link with decreased depressive symptoms²⁶. Similar benefits are observed in young adults; however, far fewer studies have examined this age group^28–31. Furthermore, studies have found the relationship between physical activity and mental health to be bidirectional^32–35. For instance, an analysis among adolescents found that lower levels of physical activity predicted depressive symptoms, while also revealing that prior depressive symptoms predicted future levels of physical activity³⁵. Likewise, studies of this bidirectional relationship in adults have suggested that depressive symptoms in early adulthood may act as a barrier to future engagement in regular physical activity³⁴. Nonetheless, several studies have indicated that the bidirectional relationship is more pronounced when examining the impact of physical activity on depressive symptoms^32,34,35. Despite the many benefits, only approximately 40% of Canadian children and youth meet recommended levels of physical activity, which poses many concerns to the healthcare system^36,37.

While a plethora of studies have established the mental health benefits of physical activity, far less is known about its ability to prevent the use of costly mental health services^38–43. Thus, the objective of this study was to explore the longitudinal associations between physical activity during adolescence and young adulthood and incidence of psychiatric hospitalizations through the utilization of health survey data linked to administrative data in Canada. Generating evidence to demonstrate the effects of physical activity on mental healthcare utilization can provide decision makers with the policy-relevant information needed for implementing mental illness prevention strategies through physical activity.

Methods

Data sources

The Canadian Community Health Survey (CCHS) is a nationally representative cross-sectional survey reporting on health behaviours among the Canadian household population aged 12 and older in all provinces and territories⁴⁴. CCHS cycles 1.1 (2001), 2.1 (2003), and 3.1 (2005) collected data over 2-year periods and recruited approximately 130,000 participants per cycle. Since 2007, the CCHS has been administered annually, recruiting approximately 65,000 participants per cycle⁴⁵. The sampling procedure follows a multistage stratified cluster design, with data collected through online electronic questionnaires or computer-assisted telephone interviews⁴⁴. Response rates for the CCHS have ranged from 69.8%−78.9%, with approximately 86% of all respondents agreeing to record linkages⁴⁵. Exclusion criteria for the CCHS include the following: persons living on Indigenous reserves, full-time members of the Canadian Armed Forces, institutionalized populations, children living in foster care, and two remote health regions in Quebec; altogether, approximately 96% of Canadians are represented in this target population⁴⁵.

The Discharge Abstract Database (DAD) contains administrative, clinical, and demographic information recorded upon all discharges from acute inpatient care settings in all Canadian provinces and territories except Quebec⁴⁶. Upon discharge from these care settings, patient’s health records are coded and abstracted before being sent to the Canadian Institute for Health Information to be recorded in the DAD⁴⁶. Diagnostic codes found in the DAD are reported using the International Classification of Diseases- Nineth and Tenth revisions (ICD-9 and ICD-10)⁴⁶. The Canadian Vital Statistics Database (CVSD) was used for censoring participants who had died to obtain more reliable estimates of person-time at risk. All data files were linked to the Derived Record Depository (DRD) at Statistics Canada using probabilistic (CCHS and CVSD) or deterministic (DAD) linkage methods; 96.2% of CCHS records, 94.1% of DAD records, and 99.8% of CVSD records were successfully linked to the DRD (Fig. 1)⁴⁵.

Fig. 1.

Venn diagram combined with flow diagram to illustrate the linkage process between data sources, exclusion criteria, and sample sizes.

Study design and setting

This analysis follows a prospective cohort study design, wherein eligible participants from the CCHS were recruited, categorized by physical activity level, and followed for the outcome of psychiatric hospitalizations. Eligible participants in the 2001–2014 CCHS were identified and linked to the 1999/00–2017/18 DAD. The study period began January 1, 1999 and ended March 31, 2018, encompassing all available DAD records. Participants were followed from their CCHS interview date until the first of either the study end date or their date of death. Eleven CCHS cycles were linked to the DAD, independently analyzed, and subsequently meta-analyzed (Fig. 1). Approval for use of microdata linkages was provided by Statistics Canada and accessed through the Prairie Regional Research Data Centre located at the University of Calgary. According to the Tri-Council Policy Statement: Ethical Conduct for Research Involving Humans – TCPS 2 (available at: https://ethics.gc.ca/eng/tcps2-eptc2_2018_chapter2-chapitre2.html#2) governed by the Government of Canada, research involving data that are publicly available through legislation protected by law and where dissemination of results do not generate identifiable information, as in the case for this study, is exempt from Research Ethics Board review. Respondents to the CCHS provided informed consent to participate in the survey and provided additional consent for linkage to administrative data records. Reporting of these findings were completed in accordance with the RECORD Checklist for health administrative data⁴⁷.

Participants

Our sample consisted of respondents to the 2001–2014 CCHS cycles who were aged 12–24 years at the time of their interview and provided consent to record linkage (Fig. 1). Participants who resided in Quebec, were pregnant at time of their interview, or who had missing physical activity data were excluded (Fig. 1). A 2-year washout period was applied, wherein participants with a most responsible psychiatric diagnosis 2-years prior to their interview date, or whose interview date occurred within 2-years of the study start date, were excluded from analysis (Fig. 1).

Variables

The following CCHS variables were used to obtain demographic and covariate data: age, sex, body mass index (BMI), chronic conditions, smoking status, alcohol use, highest household education, household income, and race. Participants aged 12–17 years were categorized as adolescents, while participants aged 18–24 years were categorized as young adults. Adolescents were categorized as underweight, normal weight, overweight, or obese based on age- and sex-standardized Cole Classification cut-offs for children^48,49. Young adults were similarly grouped into underweight (< 18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), or obese (≥ 30 kg/m²) based on the World Health Organization adult BMI cut-offs⁵⁰. Chronic conditions were dichotomized into having at least one or none of the following conditions: diabetes, cardiovascular disease, inflammatory bowel disease, or cancer. Smoking and alcohol use were both categorized as regular use (daily smoker; regular drinker), occasional use (former daily smoker or less than 100 lifetime cigarettes; occasional drinker), or never use (current non-smoker or never smoker; no drink in the past 12-months). Highest household education was categorized as lower than high school, lower than post-secondary, or higher than post-secondary. CCHS cycles 2001 and 2003 used a 5-level income adequacy variable based on total household income and number of people living in the household, while cycles 2005 and onwards used distribution of provincial income (deciles). Race was dichotomized as White or non-White.

Physical activity was measured using the Minnesota Leisure-Time Physical Activity Questionnaire (LTPAQ), which is a self-reported questionnaire used to estimate respondents’ metabolic equivalents (METs) during leisure-time physical activities (LTPA) based on type of activity, frequency, and bout duration over the past 3 months⁵¹. METs are then converted into average energy expenditure (AEE) measured in kilocalories per kilogram per day (KKD). Participants in the CCHS were categorized as inactive (< 1.5 KKD), moderately active (1.5 – 2.99 KKD), or active (≥ 3.0 KKD). Energy expenditure in the Minnesota LTPAQ were found to be highly correlated with values obtained from accelerometry and doubly labelled water^51,52. All variables from the CCHS were measured at a single point-in-time in the year from which a participant was interviewed.

Chapter 5: Mental disorders (290–319) of ICD-9 and Chapter V: Mental and behavioural disorders (F00-F99) of ICD-10 reported as the most responsible diagnosis for each hospital discharge in the DAD were used to measure psychiatric hospitalizations occurring after each participant’s CCHS interview. For descriptive purposes, mental disorders were subclassified into mood and anxiety, substance related, psychosis, personality, organic, and other disorders. A person-time variable counted days between a participant’s CCHS interview date and whichever occurred first between the study end date or a participant’s death date.

Statistical analysis

Sample characteristics for each survey were calculated using sampling and replicate bootstrap weights, which were necessary to address the unequal selection probabilities arising from the complex sampling design in the CCHS. Descriptive estimates from each survey were pooled using random-effects meta-analysis and were represented as pooled percent proportions. Physical activity was analyzed in three levels, with the active participants used as the referent group. Psychiatric diagnoses were represented as count data, and due to over-dispersion, negative binomial regression models with log follow-up time offsets were used. To account for outliers, psychiatric diagnoses were Winsorized at the 99^thpercentile. Cumulative incidence for each survey was calculated as the percent proportion of participants receiving at least one psychiatric hospitalization. Incidence rates for each cycle were calculated using unweighted counts of psychiatric hospitalizations and person-time denominators (psychiatric hospitalizations per 100,000 person-days). Models were built by assessing the Akaike information criterion values of several models that featured different covariates, interaction terms, and other count regression models for each survey. The following models were selected based on goodness-of-fit, parsimony, and consistency throughout all eleven survey cycles: crude models, age- and sex-adjusted models, and full models adjusting for all covariates. For cycles 2001–2005, additional age-stratified analyses of the full model were conducted. Cycles 2007–2014 could not support age-stratification due to their reduced sample sizes and low frequency of events. Regression coefficients and bootstrapped standard errors were used in subsequent random-effects meta-analyses using DerSimonian and Laird weighting. All analyses were performed using sample and bootstrap weights, significance levels of α = 0.05, and included 95% confidence intervals (CIs). Statistical analyses were conducted using Stata 16 and figures were generated using the tidyverse package in RStudio^53,54.

Results

Descriptive analysis

The total sample size for all eleven surveys combined was 96,100. The mean follow-up time among all participants was 9.29-years (95% CI: 7.50–11.08) (Table 1.). This sample included fewer adolescents (43.60%) compared to young adults (56.40%) with similar proportions of males (51.46%) and females (48.53%) (Table 1.). The greatest number of participants were from Ontario (51.59%), British Columbia (16.29%), and Alberta (14.56%), with no other province or territory exceeding a proportion of 5% (Table 1.). A large proportion of this sample’s population were White (74.46%), had high levels of household education (76.51%), had no chronic conditions (97.50%), and were normal weight (65.42%) (Table 1.).

Table 1.

Pooled demographic characteristics for CCHS cycles 2001-2014.

Characteristic		Percent Proportions (95% CI)
Total sample size (n)		96,100
Mean follow-up time in years (95% CI)		9.29 (7.50–11.08)
Age group
Adolescents (12–17 years)		43.60 (42.60–44.60)
Young adults (18–24 years)		56.40 (55.40–57.39)
Sex
Male		51.46 (51.00–51.92)
Female		48.53 (48.07–48.99)
Province
Ontario		51.59 (51.26–51.92)
British Columbia		16.29 (16.03–16.55)
Alberta		14.56 (14.26–14.86)
Manitoba		4.63 (4.51–4.75)
Saskatchewan		4.01 (3.92–4.10)
Nova Scotia		3.38 (3.19–3.57)
New Brunswick		2.63 (2.52–2.73)
Newfoundland & Labrador		1.88 (1.75–2.02)
Prince Edward Island		0.56 (0.54–0.59)
Northwest Territories		0.19 (0.18–0.20)
Nunavut		0.13 (0.11–0.15)
Yukon		0.13 (0.13–0.14)
White/non-White
White		74.46 (72.13–76.79)
Non-White		25.54 (23.21–27.87)
Highest household education
Less than high school		3.23 (2.84–3.63)
Greater than high school, less than post-secondary		20.26 (18.92–21.60)
Post-secondary and higher		76.51 (74.79–78.22)
Household income- deciles a
1		13.57 (12.82–14.31)
2		11.13 (10.68–11.57)
3		10.70 (10.38–11.03)
4		10.33 (9.88–10.79)
5		10.29 (9.76–10.81)
6		10.21 (9.84–10.59)
7		9.18 (8.67–9.69)
8		9.28 (8.85–9.71)
9		8.45 (7.93–8.97)
10		6.86 (6.50–7.22)
Smoking status
Regular		10.94 (9.58–12.29)
Occasional		6.37 (5.93–6.82)
Not at all		82.69 (81.08–84.29)
Alcohol status
Regular drinker		45.97 (44.62–47.32)
Occasional drinker		17.34 (16.48–18.21)
Not at all		36.69 (35.43–37.95)
Chronic conditions
No		97.50 (97.19–97.81)
Yes		2.50 (2.19–2.81)
Body mass index
Underweight		6.85 (6.55–7.14)
Normal weight		65.42 (64.60–66.24)
Overweight		19.20 (18.82–19.59)
Obese		8.54 (7.94–9.13)
Physical activity level
Active		44.92 (43.77–46.06)
Moderately active		22.89 (22.35–23.43)
Inactive		32.20 (31.27–33.13)

^a CCHS cycles 2001–2003 omitted from pooled analyses due to different income measures used.

Overall, most participants were classified as being active (44.92%), followed by inactive (32.2%) and moderately active (22.89%) (Table 1.). By age group, adolescents had a higher proportion of active participants versus young adults (51.99% vs. 39.42%), similar proportions of moderately active participants (23.62% vs. 21.94%) and fewer inactive participants (26.08% vs. 36.97%) (Fig. 2). Comparing the first and last cycles, 40.82% (2001) to 44.47% (2014) of adolescents were active, 23.15% (2001) to 20.02% (2014) were moderately active, and 27.38% (2001) to 27.10% (2014) were inactive (Fig. 3). Among young adults, 33.77% (2001) to 38.63% (2014) were active, 24.59% (2001) to 22.77% (2014) were moderately active, and 41.64% (2001) to 38.59% (2014) were inactive (Fig. 3).

Fig. 2.

Pooled percent proportion of physical activity levels between adolescents and young adults.

Fig. 3.

Percent proportion of physical activity levels among adolescents and young adults across all CCHS cycles.

The cumulative incidence and incidence rates of psychiatric hospitalizations are shown in Fig. 4. Due to longer follow-up periods, earlier cycles had a higher cumulative incidence of psychiatric hospitalizations (2.43%; 2001) compared to later years (0.94%; 2014) (Fig. 4a). Incidence rates showed the opposite, with earlier cycles showing lower rates (1.10 per 100,000 person-days; 2001) compared to later cycles (1.46 per 100,000 person-days; 2014) (Fig. 4b). The most prevalent mental diagnosis reported was for mood and anxiety disorders (38.04%), followed by other (19.31%), substance related (19.19%), psychosis (18.91%), personality (3.98%), and organic (1.95%) disorders (Fig. 5).

Fig. 4.

(a) Cumulative incidence of psychiatric hospitalizations for each cycle of the CCHS. (b) Incidence rate of psychiatric hospitalizations for each cycle of the CCHS.

Fig. 5.

Pooled percent proportion of the most responsible diagnoses received for psychiatric hospitalizations.

Meta-analysis

All models meta-analyzed showed low heterogeneity (I² < 25%), except for the age stratified models (I² = 50.33%). As a result of the large samples sizes in the early CCHS, cycles 2001–2005 were the most strongly weighted in all meta-analyses performed. From the crude models, neither the moderately active (IRR = 1.18; 95% CI: 0.95–1.47; p = 0.14) nor inactive (IRR = 1.21; 95% CI: 0.99–1.47; p = 0.06) groups were significantly different from the active group (Table 2). When adjusted for age and sex, the moderately active group remained nonsignificant (IRR = 1.22; 95% CI: 0.98–1.51; p = 0.07), while the inactive group had a significantly higher IRR = 1.34 (95% CI: 1.07–1.69; p = 0.01) (Table 2). After adjustment for all covariates, both the moderately active group (IRR = 1.30; 95% CI: 1.02–1.66; p = 0.03) and the inactive group (IRR = 1.33; 95% CI: 1.06–1.66; p = 0.01) had significantly higher rates of hospitalizations compared to the active group (Table 2). In the age stratified analyses, adolescents showed no significant differences in either moderately active (IRR = 1.24; 95% CI: 0.69–2.25; p = 0.47) nor inactive (IRR = 1.02; 95% CI: 0.64–1.65; p = 0.93) groups, while in young adults, both moderately active (IRR = 1.83; 95% CI: 1.18–2.83; p = 0.01) and inactive (IRR = 1.63; 95% CI: 1.08–2.47; p = 0.02) groups had significantly higher rates of hospitalizations compared to active young adults (Table 2).

Table 2.

Calculated IRRs for physical activity levels and count of psychiatric diagnoses in CCHS cycles 2001–2014 with pooled overall estimates.

CCHS Cycle	Crude Modelsb	Age- and Sex-Adjusted Models	Full Modelsbc	Full Models- Age Stratified
				Adolescentsbc	Young Adultsbc
Moderately Activea
2001	1.34 (0.76–2.38)	1.37 (0.80–2.36)	1.12 (0.63–1.97)	0.68 (0.37–1.25)	1.93 (0.94–3.99)
2003	1.50 (0.78–2.91)	1.47 (0.78–2.78)	1.95 (1.03–3.68)d	1.91 (1.04–3.52)d	2.10 (1.01–4.37)d
2005	1.01 (0.63–1.61)	0.99 (0.60–1.61)	1.24 (0.74–2.08)	1.47 (0.84–2.56)	1.42 (0.62–3.24)
2007	1.28 (0.63–2.59)	1.26 (0.64–2.52)	1.34 (0.44–4.07)	–	–
2008	1.02 (0.54–1.93)	1.16 (0.59–2.28)	1.38 (0.55–3.46)	–	–
2009	0.95 (0.35–2.63)	1.03 (0.42–2.58)	1.18 (0.43–3.21)	–	–
2010	1.32 (0.48–3.63)	1.47 (0.50–4.36)	1.09 (0.36–3.29)	–	–
2011	0.89 (0.33–2.41)	0.95 (0.34–2.65)	1.20 (0.50–2.87)	–	–
2012	1.20 (0.52–2.77)	1.25 (0.52–3.00)	1.67 (0.63–4.45)	–	–
2013	1.98 (0.78–5.03)	1.92 (0.79–4.65)	2.19 (0.65–7.38)	–	–
2014	0.93 (0.38–2.26)	0.98 (0.41–2.33)	0.59 (0.20–1.73)	–	–
Overall	1.18 (0.95–1.47)	1.22 (0.98–1.51)	1.30 (1.02–1.66)d	1.24 (0.69–2.25)	1.83 (1.18–2.83)d
p-value	0.14	0.07	0.03	0.47	0.01
Inactivea
2001	1.67 (1.05–2.65)	1.85 (1.16–2.94)d	1.42 (0.91–2.21)	1.22 (0.67–2.23)	1.83 (0.89–3.78)
2003	1.13 (0.70–1.84)	1.27 (0.77–2.07)	1.37 (0.72–2.60)	1.68 (0.63–4.54)	1.13 (0.57–2.25)
2005	1.12 (0.70–1.78)	1.36 (0.81–2.31)	1.22 (0.72–2.09)	0.71 (0.42–1.20)	2.23 (1.05–4.71)d
2007	2.91 (1.06–8.00)d	4.36 (1.37–13.83)d	1.92 (0.73–5.08)	–	–
2008	1.38 (0.74–2.57)	1.46 (0.81–2.65)	1.82 (0.81–4.10)	–	–
2009	0.97 (0.34–2.73)	1.11 (0.42–2.91)	1.28 (0.50–3.28)	–	–
2010	0.74 (0.38–1.44)	0.86 (0.43–1.72)	1.17 (0.46–2.94)	–	–
2011	0.64 (0.26–1.57)	0.67 (0.25–1.80)	0.93 (0.34–2.53)	–	–
2012	0.77 (0.33–1.79)	0.71 (0.29–1.74)	1.03 (0.39–2.75)	–	–
2013	1.58 (0.77–3.26)	1.71 (0.80–3.66)	1.83 (0.72–4.66)	–	–
2014	1.43 (0.48–4.21)	1.42 (0.54–3.79)	0.75 (0.26–2.14)	–	–
Overall	1.21 (0.99–1.47)	1.34 (1.07–1.69)d	1.33 (1.06–1.66)d	1.02 (0.64–1.65)	1.63 (1.08–2.47)d
p-value	0.06	0.01	0.01	0.93	0.02

^a Compared to the active referent group.

^b Incidence Rate Ratio (95% CI).

^c Adjusted for age, sex, BMI, chronic conditions, smoking status, alcohol use, highest household education, household income, and race.

^d Statistically significant (p < 0.05).

Discussion

Among this sample of Canadian adolescents and young adults, our fully adjusted models demonstrated a 30% increased rate of psychiatric hospitalizations in moderately active participants, and a 33% increased rate among inactive participants, compared to active participants. Additionally, the age-stratified models showed that these associations were strongest among young adults, with no significant associations found in adolescents. Contrary to our other models, moderately active young adults had higher relative rates of psychiatric hospitalizations (83%) than inactive young adults (63%). Two similar studies in Canada found significant reductions in the incidence of mental health visits by approximately 30–50% among physically active children^38,39. A further two studies from Finland and Norway demonstrated reductions as high as 43% for psychosis-related hospitalizations among physically active adolescents, and another study showed 30–45% reductions in hypnotic drug prescription with increasing physical activity^41–43. To our knowledge, these are the only studies to have examined the link between youth physical activity and any measure of mental healthcare utilization, with all showing some significant reductions.

This is the largest study to examine physical activity and psychiatric hospitalizations among adolescents, and the only to be conducted in young adults. Our study design allowed us to analyze eleven cycles of the same survey with high homogeneity of the sample population, sampling procedures, and measurement instruments. As shown in Table 2, few cycles alone would have been able to detect any significant associations, however, the increased precision gained through meta-analysis allowed us to elucidate this association. The CCHS employs robust sampling procedures, providing us with a highly representative sample of the Canadian population^44,55. The DAD provides coverage in all acute inpatient settings in Canada, excluding Quebec, and has demonstrated high standards of data quality^56,57.

Certain limitations exist within this analysis. Although the Minnesota LTPAQ has been validated for use in both adolescent and adult populations, several domains of physical activity behaviours such as active transportation, occupation, or household activities were not fully captured⁵⁸. Furthermore, the categorization of active participants in the CCHS was defined using an AEE of 3.0 KKD. While this threshold may be appropriate for the adult population, it has been suggested that children and adolescents require at least 8.0 KKD to achieve health benefits⁵⁹. This may partially explain the null finding observed among adolescents, as the “active” category may be over-represented by participants who were not sufficiently active, leading a diminished effect size in this age group. Additionally, adolescents also experience increased exposure to protective factors such as greater inter-personal relationships, intellectual development, sense of belonging, health literacy, social and emotional support, and access to mental health services compared to young adults^18,60. As these added safety nets are no longer accessible to young adults, physical activity may be a larger component cause for protection against mental illness.

Since physical activity is a time-varying exposure that was measured only once at the time of a participant’s interview, movement between physical activity groups likely occurred, especially during the transition to early adulthood. This presents a potential source of non-differential misclassification bias, wherein participants classified as physically active may become less active, resulting in a dilution of the observed associations. Similarly, changes in time-varying confounders or the presence of unmeasured confounders could not be detected. For instance, we were unable to identify participants with early-stage or undiagnosed mental illness, which may confound the relationship between physical activity and psychiatric hospitalizations. As a result, we were unable to fully establish the directionality of this relationship and causal effects cannot be elucidated.

With the onset of mental illness overwhelmingly occurring before the age of 25, early mental illness prevention strategies are critical for fostering healthy development across the life course^17,22,23. There is strong evidence to suggest that increased physical activity during adolescence and young adulthood is effective for preventing mental illness, however, levels of physical activity among Canadian children and youth have remained stable with the vast majority not meeting guidelines^{25,26,28,61,62}. This has prompted several calls to action urging for increased cooperation between policy makers, governments, physical activity stakeholders, and communities to address the issue of persistent inactivity in Canada⁶³. Among the barriers to decision making is a lack of policy-relevant evidence translating to economic benefits⁶⁴. Within the physical activity and mental health literature, there is a disparity between studies measuring mental health through subjective scales and those assessing utilization of costly mental health services. The present study addresses this gap by demonstrating that physical activity is associated with reductions in psychiatric hospitalizations. Although hospitalizations represent a small portion of all mental healthcare utilization, the average daily cost of a mental health stay in Canada is $900, and the average hospital LOS is nearly twice in patients with a mental disorder^11,12,65.

With the paucity of research on this topic, future studies exploring various sources of mental healthcare utilization, such as emergency department visits, outpatient mental health services, or prescription medications are needed. This is made possible through record linkage, allowing researchers to link readily collected data to health administrative data and examine the longitudinal impacts of physical activity on mental health services. Future studies should consider the various domains of physical activity and aim to better understand the mechanisms underlying this link. In addition, future analyses should attempt to examine outcomes by disorder type, which could provide greater insight into the levels of functional impairment experienced by individuals and lead to a greater understanding of the directionality of this relationship. With the importance of health promotion in young adults and lack of existing studies, more studies should incorporate this age group in future analyses. Although this study included a representative sample of the Canadian population, this sample was mostly homogenous with a high proportion of participants being healthy, white, and from high education households.

Overall, we found that lower levels of physical activity were associated with increased rates of psychiatric hospitalizations. The totality of prior literature suggests that physical activity among AYAs can be protective against depression and anxiety^66,67; this study adds to current literature by demonstrating that mental health benefits may translate to reductions in hospitalizations as well. This study offers valuable insights for decision makers and stakeholders, but further investigation is necessary before policy recommendations can be made. Our findings lay important groundwork for future research in this area.

Author contributions

Study concept and design: M.F., S.P., P.R., and M.F.B. Data acquisition, analysis, and interpretations: M.F., S.P., and J.W. Drafting of manuscript: M.F. Revisions: S.P., P.E., and M.F.B.

Data availability

Data for this study are housed within secure facilities managed by Statistics Canada and can be accessed remotely through the Research Data Centres (https://www.statcan.gc.ca/en/microdata/data-centres/data/cphs).

Declarations

Conflict of interest

The authors declare no competing interests.

Ethical approval

This study was carried out in accordance with Statistics Canada Directive on Microdata Linkage and the TCPS-2 Ethical Guidelines. The methods and reporting of this manuscript were completed in accordance with the RECORD guidelines for the reporting of health administrative data.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-81273-6.