Association of hospitalization costs with demographic, socioeconomic, and lifestyle characteristics: Population-based study in Sao Paulo city, Brazil, 2003–2015

Highlights

• •Hospitalization costs were higher in private sector compared to public sector.
• •Physical activity and sedentary habits were associated with hospitalization costs.
• •Adherence to recommended physical activity reduces costs with hospitalizations.
• •Sedentarism and diagnosis of chronic diseases increase costs with hospitalizations.
• •There was low adherence to active lifestyles in Sao Paulo city.

Abstract

Evidence from studies conducted in high-income countries suggests that lifestyle factors, such as leisure-time physical activity, sedentary habits, and obesity, are associated with a significant socioeconomic burden of disease and the attribution of direct costs to healthcare systems. In Brazil, the occurrence of primary care-sensitive hospitalizations is responsible for a relevant socioeconomic burden. However, there is a scarcity of evidence regarding the association of lifestyle factors on the direct costs of the Brazilian healthcare system. In this context, the present study aims to analyze the association between leisure-time physical activity, sedentary habits, and obesity with hospitalization costs in São Paulo city, Brazil. A quantitative analysis of microdata from the São Paulo Health Survey (ISA-Capital), which is representative for the urban population of São Paulo City, and was conducted in 2003, 2008, and 2015, was employed. Multiple two-part regression models (logit and GLM) and marginal effects (ME) were estimated. The study’s findings suggest that meeting the weekly frequency of leisure-time physical activity recommended by the World Health Organization is associated with lower hospitalization costs in the public (logit β = −0.475, p < 0.05; ME = −31.03, p < 0.05) and private sector (logit β = −0.494, p < 0.01; ME = −37.89, p < 0.01). Sedentary habits (logit β = 0.442, p < 0.05; ME = 40.92, p < 0.01), and obesity (GLM β = 0.385, p < 0.05) were associated with higher costs in the private sector. No associations were observed between sedentary habits and obesity for hospitalization costs in the public sector. The evidence from the present study suggests that policies encouraging the adoption of healthy active lifestyles, such as practicing leisure-time physical activity and reducing sedentary habits, as well as policies for obesity prevention, may be important strategies for minimizing hospitalization costs in urban population contexts in the two-tier of the Brazilian healthcare system. Yet, associations identified in the study should be interpreted with caution due to the impossibility of establishment of causal links between lifestyle factors and healthcare expenditures.

1. Introduction

Brazil's two-tiered health system features a complex network of public and private providers that establish cooperative and competitive relations, offering services at primary, secondary, and tertiary levels [1,2]. Its public component, the Unified Health System (SUS), provides universal healthcare free of charge, based on the Beveridge model [3]. Although the SUS has advanced universal coverage by providing comprehensive healthcare [4,5], the private sector accounts for most health expenditures [1,2,6]. Accessing private healthcare requires out-of-pocket or insurance payments, which imposes a financial burden on low-income individuals, increasing their risk of catastrophic health expenditures and impoverishment [1,2,[7], [8], [9]].

Hospitalization costs for conditions amenable to primary care have risen in recent decades [10], while the increasing prevalence of chronic diseases, linked to ongoing demographic, epidemiological, and nutritional transitions, has created significant financial stress on the Brazilian health system [[11], [12], [13], [14], [15], [16]]. Recent evidence of sustained unhealthy behaviors and growing morbimortality from chronic diseases in Brazil [[17], [18], [19], [20], [21]] underscores the potential for evidence-based policies. Such policies could promote healthy lifestyles to reduce costs from preventable conditions and ensure the health system's financial sustainability [22,23].

Yet, literature on lifestyle-related healthcare costs in Latin America is scarce due to a lack of representative population surveys that collects data on lifestyle factors and health services use, beyond demographic and socioeconomic characteristics [24,25]. To address this gap, our study analyzes the association between adherence to World Health Organization (WHO) leisure physical activity (PA) guidelines, sedentary habits, obesity, and the public and private costs of hospitalizations. Using a representative urban sample from São Paulo city, Brazil (2003–2015), our findings might inform strategic health policies for cost reduction and disease prevention in urban centers in Brazil and other low and middle-income countries, especially, in the Latin America region.

2. Materials and methods

2.1. Study design

The study is based on quantitative analysis of observational microdata obtained in three cross-sectional surveys representative of the urban population of São Paulo city, Brazil conducted in 2003, 2008, and 2015.

2.2. Data collection

The Health Survey of São Paulo (ISA-Capital) was based on data collection in a sample of individuals living in urban areas of São Paulo city, selected through a complex probabilistic sampling process in two stages (census tracts and households). Residents of the households selected in the sample were invited to participate in the surveys.

São Paulo city has been the largest urban center in Brazil since the 1960s, and one of the largest in Latin America [26]. The objective of the ISA-Capital is to provide information on the demographic, socioeconomic, lifestyle, and health characteristics of the population, including healthcare utilization, ensuring evidence-based decision-making in health policies. Data collection was based on the application of structured questionnaires applied by trained interviewers.

Additional information on the sampling procedures, including sample size calculation and data collection tools, has been previously published elsewhere [27,28].

2.3. Data sources

The datasets of the ISA-Capital were obtained through data collection and organized into a single database to ensure the selection of variables directly comparable across the three editions of the survey. Therefore, questions with substantial changes in phrasing or response options were excluded, and missing data were excluded from the analysis.

The ISA-Capital sample consists of data from 3357 individuals participating in ISA-Capital 2003; 3271 individuals for the 2008 edition; and 4043 individuals in the 2015 edition. Considering the present study, the sample consists of data from 1930 individuals participating in ISA-2003; 2367 individuals in ISA-2008; and 3069 in ISA-2015, totaling a sample size of 7366 participants considering the three editions. Differences in the sample characteristics can be found in the supplementary material (tables S1 and S2).

The pooling of the three datasets was performed while preserving the survey's complex sample design. This was done by creating unique identifiers for the primary sampling units and strata for each survey year and by rescaling the sample weights, to ensure population-level representativeness for the urban area of São Paulo city. Additionally, the analyses include controls for temporal effects by accounting for the survey editions. This approach made it possible to conduct analyses in the final analytical sample for the entire period (2003–2015).

In addition to individual-level data from ISA-Capital, information on the monetary values assigned to hospitalizations within the SUS and the private health sector were extracted from the website of the Department of Informatics of the SUS (DATASUS) and the reports of the Brazilian Hierarchical Categorization of Medical Procedures (CBHPM), respectively. The information on values of hospitalizations according to diagnosis, type of funding, and year of the survey were used to estimate the direct costs of inpatient days reported by individuals interviewed in the three editions of ISA-Capital.

The DATASUS comprises an administrative information system based on monthly data collection on healthcare procedures (products and services) within the public sector, designed to support the operationalization of the SUS. Information on procedures, hospitalizations, inpatient days, and hospitalization costs is publicly available in anonymized datasets online on the DATASUS website, according to characteristics of patients (sex, and age), diagnosis (codes of the 10th edition of the International Classification of Diseases, ICD-10), location (city of residence of patients), and period (month and year) of healthcare utilization.

The CBHPM encompasses codes, composition, and prices attributable to standardized healthcare procedures in the private sector according to diagnosis, type of intervention, and operational costs, continuously updated and published by the Brazilian Medical Association since 2003 [29]. The information available on prices of standardized healthcare procedures in the private sector allows estimating direct costs per procedure per day in São Paulo city during the reference period of the three editions of the ISA-Capital survey, according to the composition of healthcare procedures required during hospitalizations.

2.4. Variables

2.4.1. Outcome variables

The outcome variables in the present study were the occurrence and direct costs of hospitalizations, based on self-declared information on demand for hospital care, length of stay, and diagnosis linked to hospitalizations in the year before the ISA-Capital survey. Information on the diagnosis was based on health issues declared by individuals in the survey, categorized according to the ICD-10.

The estimation of direct costs of hospitalizations was based on a macro-costing technique using the perspective of payers, following guidelines of the Brazilian Ministry of Health [30]. The information on occurrence, length of stay, and diagnosis linked to hospitalizations among individuals surveyed in ISA-Capital was used to calculate direct costs by using the following equation:

$$C_{\mathit{idt}}=h_{\mathit{idt}}\times n_{\mathit{idt}}\times v_{\mathit{idt}}$$ (1)

where: C_idt = direct hospitalization costs of individual i due to diagnosis d in period t; h_idt = hospitalizations of individual i due to diagnosis d during the period t; n_idt = inpatient days of hospitalization due to diagnosis d for individual i during the period t; v_idt = value per day of hospitalization due to diagnosis d for individual i during the period t.

The value per day of hospitalization (v_idt) according to the patient’s characteristics, diagnosis, and period was estimated using information extracted from DATASUS and CBHPM. The calculation of cost estimates for each diagnosis reported by hospitalized individuals in the ISA-Capital survey was linked using the ICD-10, and the length of stay was based on the number of inpatient nights.

Individuals declaring the absence of hospitalizations in the period of reference were included in the analysis considering inpatient costs equal to zero (0) to allow the identification of characteristics associated with differences in healthcare costs linked to hospitalizations.

Monetary values referring to direct hospitalization costs and household income per capita were updated to December 2015 (period of reference of the last edition of ISA-Capital), using the Broad Consumer Prices Index in São Paulo city from the Brazilian Institute for Geography and Statistics (IPCA-IBGE), and converted into purchase power parity (PPP) to allow international comparison, based on the PPP conversion factor available at the World Bank data bank [31].

The distinction between hospitalizations in the public and private sectors of the Brazilian healthcare system was determined by individuals' self-reported out-of-pocket expenditures for the hospitalization. Individuals who reported paying for the hospitalization in full, or through health insurance, were considered to have received care in the private sector, while those who reported no personal expenditure were considered to have received care in the public sector.

2.4.2. Variables of interest

The variables of interest in the analysis encompass lifestyle characteristics of individuals regarding PA and sedentary habits, and diagnosis of obesity estimated through the Body Mass Index (BMI).

The adherence to recommended levels of PA during leisure was based on the duration (in minutes) of PA during the week, based on information collected from individuals in the ISA-Capital survey through the application of the long form of the International Physical Activity Questionnaire (IPAC), Portuguese version translated and validated for the Brazilian population [32,33]. The categorization of PA level was based on international recommendations for ≥ 150 – 300 min of moderate to vigorous PA per week among adult individuals, and recommendations for ≥ 300 min of moderate to vigorous PA per week among adolescents [34], being categorized into binary variable corresponding to individuals physically inactive (0) or physically active (1).

The occurrence of sedentary habits was based on self-reported information on the duration (in minutes) of sedentary activities (i.e., activities performed in a sitting position) during weekdays and weekends, particularly screen time, including time spent in sitting position during work/study. Time spent in sitting position during transportation was not considered in the calculation. Giving the lack of consensus on potential epidemiological thresholds for sedentary habits, and risks for development and evolution of NCDs, individuals in the upper tertile of time spent in a sitting position and that lack of adherence to the recommended level of PA were considered sedentary [34]. The information on sedentary habits was converted into binary variable presence (1) or absence (0) of sedentary habits.

The diagnosis of obesity was based on the estimation of the BMI, using body weight (in kg) divided by height squared (in m²), and categorized into binary variables indicating the presence (1) or absence (0) of obesity according to specific recommendations of the World Health Organization for adults, elderly [35], and adolescents [36]. Body weight and height were self-reported by the individuals participating in the surveys. Subsequent studies by our research group validated the self-reported information in relation to the anthropometric assessment of the individuals, taking into account characteristics such as sex, age, PA and cardiometabolic risk factors [37].

2.4.3. Control variables

Models estimated in the study included control variables referring to demographic, socioeconomic, lifestyle, and health characteristics of individuals, in addition to variables referring to the edition of the survey (Table 1):

• •Demographic characteristics: sex (male = 0, and female = 1); age (years); self-reported skin color (black/brown/indigenous/yellow = 0, and white = 1).
• •Socioeconomic characteristics: marital status (living alone = 0, and living with company = 1); education (years of formal education); occupational (unemployed = 0, and employed = 1); household income per capita ($PPP); health insurance ownership (no = 0, and yes = 1).
• •Health status and health behaviors: tobacco use (no = 0, and yes = 1); diagnosis of hypertension (no = 0, and yes = 1); diagnosis of type 2 diabetes (no = 0, and yes = 1); diagnosis of cardiovascular diseases (no = 0, and yes = 1).
• •Survey characteristics: year of the survey (binary variables for the editions of the survey).

Table 1.

Characteristics of variables in the study. ISA-Capital 2003–2015, São Paulo city, Brazil.

Variables	Categories	μ	SE	Min	Max	N
Outcome variables
Hospitalization
Public sector	(1 = yes)	0.03	0.00	0	1	7366
Private sector	(1 = yes)	0.04	0.00	0	1	7366
Hospitalization costs
Public sector	($PPP)	2529.32	594.60	35.56	55152.22	7366
Private sector	($PPP)	2516.82	132.02	84.38	12813.65	7366
Independent variables
Leisure-time PA	(1 = yes)	0.23	0.01	0	1	7366
Time spent in sitting position (ref. = lower tertile)	(1 = middle tertile)	0.31	0.01	0	1	7366
	(1 = upper tertile)	0.32	0.01	0	1	7366
Obesity	(1 = yes)	0.15	0.01	0	1	7366
Control variables
Sex	(1 = female)	0.52	0.01	0	1	7366
Age	(years, continuous)	38.93	0.30	12	101	7366
Skin color/ethnicity	(1 = white)	0.59	0.01	0	1	7366
Marital status	(1 = accompanied)	0.51	0.01	0	1	7366
Education	(years, continuous)	9.89	0.10	0	16	7366
Occupational status	(1 = employed)	0.62	0.01	0	1	7366
Income per capita	($PPP)	871.78	39.97	0.91	18294.86	7366
Health insurance	(1 = yes)	0.36	0.01	0	1	7366
Tobacco use	(1 = yes)	0.33	0.01	0	1	7366
Hypertension	(1 = yes)	0.18	0.01	0	1	7366
Diabetes	(1 = yes)	0.06	0.00	0	1	7366
Cardiovascular diseases	(1 = yes)	0.05	0.00	0	1	7366
Year of survey	(2003; 2008; 2015)	2009.11	0.14	2003	2008	7366

μ = weighted mean; SE = standard error; Min = minimum; Max = maximum; N = total observations. $PPP = purchase power parity; PA = leisure time physical activity.

The use of tobacco was based on self-declared information on the daily habit of smoking cigarettes. The variables on the occurrence of chronic diseases referred to self-reported previous medical diagnoses of hypertension, type 2 diabetes (DM), and cardiovascular diseases (CVD). The household income per capita was calculated by summing income earned by the residents in the month previous to the survey and dividing by the number of household residents.

2.5. Statistical analyses

Descriptive analysis of data was based on weighted median and interquartile range for continuous variables due to lack of normal distribution, and weighted frequency and 95 % confidence interval for categorical variables. The trend over the study period was analyzed using simple quantile regression for the survey years among continuous variables, and with the chi-squared test for the survey years (2003–2015) among categorical variables.

Two-part regression models and mean marginal effects (ME) were estimated using two specifications: direct costs of hospitalizations among individuals using the SUS, and individuals using the private sector. Two-part models comprise models with a logit model and a generalized linear model (gamma family and log-link function) to allow the identification of the likelihood of occurrence of values equal to or higher than zero. Variables measuring costs usually present a high proportion of zero values due to the low frequency of healthcare utilization at the population level, whilst simultaneously zero values should be included in the analyses because they represent a lack of demand for healthcare. Therefore, considering the importance of zero values, the adoption of two-part models is ideal to fit data on healthcare costs in population-based studies. In addition, two-part models avoid interference of zero-inflated data on the analyses, reducing potential bias in the estimates compared to other empirical strategies [[38], [39], [40]].

The analyses were performed using the software Stata® (StataCorp., College Station, Texas, EUA), version 14.2, adopting a significance level of 5 % (p < 0.05), including complex sample design to ensure representativeness, with adjustments for potential correlations among subgroups of the sample.

3. Results

Major part of participants were women (52.35 %), individuals declaring white skin color (59.30 %), living accompanied (51.20 %), and employed (61.99 %). Minor proportions of individuals present health insurance ownership (36 %), and diagnosis of hypertension (17.58 %), DM (5.45 %), CVD (5.44 %), or obesity (14.53 %). Regarding lifestyle characteristics, there was low adherence to recommended PA level (23.23 %), presence of sedentary habits (32.23 %), and tobacco use (33.25 %) in the sample. The sample has a median age of 36 years and a median of 11 years of formal education. Most hospitalizations occurred for clinical procedures and diagnosis (56.3 %), being related to diseases of circulatory system (14.3 %), pregnancy and childbirth (13.6 %), consequences of external causes like accidents, poisoning and injuries (9.8 %), and diseases of genitourinary system (8.8 %) (Supplementary Materials, Table S3).

Trends show a significant increase in the median age and median years of formal education of the population, in the proportion of individuals living accompanied, in the prevalence of chronic diseases and cardiometabolic risk factors, including obesity, and in the occurrence of hospitalizations in the public sector throughout the period from 2003 to 2015. Contrarily, individuals declaring white skin color and health insurance ownership presented a significant decline in the period; whereas adherence to recommended PA, sedentary habits, tobacco use, and hospitalizations in the private sector remained stable. (Table 2).

Table 2.

Sample characteristics by survey year. ISA-Capital 2003–2015, São Paulo city, Brazil.

Variables	Categories	2003 (N = 1930)	2008 (N = 2367)	2015 (N = 3069)	2003–2015 (N = 7366)	P
Sex	(1 = female)	51.88 [48.89,54.86]	52.85 [50.64,55.05]	52.27 [50.34,54.19]	52.35 [51.00,53.69]
Age A	(years)	34[27]	36[26]	38[26]	36[27]	*
Skin color/ethnicity	(1 = white)	67.13 [63.01,71]	61.91 [56.86,66.71]	50.68 [47.01,54.34]	59.30 [56.77,61.79]	*
Marital status	(1 = accompanied)	48.41 [45.38,51.45]	50.84 [47.89,53.78]	53.77 [51.40,56.12]	51.20 [49.58,52.82]	*
Education A	(years)	10[7]	12[5]	12[6]	11[5]	*
Occupational status	(1 = employed)	58.35 [54.78,61.83]	63.43 [60.68,66.10]	63.58 [61.22,65.88]	61.99 [60.30,63.65]
Income per capita A	($PPP)	429.57 [701.55]	537.38 [685.97]	555.52 [658.26]	509.99 [689.42]
Health insurance	(1 = yes)	39.09 [34.85,43.50]	36.75 [31.69,42.13]	32.84 [29.37,36.50]	36.00 [33.48,38.60]	*
Leisure-time physical activity	(1 = yes)	23.88 [20.80,27.26]	23.80 [21.86,25.85]	22.20 [20.20,24.34]	23.23 [21.87,24.66]
Time spent in sitting position (ref. = lower tertile)	(1 = middle tertile)	28.8 [24.75,33.22]	33.54 [30.66,36.55]	31.29 [29.37,33.27]	31.32 [29.61,33.08]
	(1 = upper tertile)	33.57 [29.45,37.96]	31.28 [27.27,35.58]	32.04 [29.49,34.69]	32.23 [30.16,34.38]
Obesity	(1 = yes)	10.04 [7.95,12.59]	12.68 [11.01,14.55]	19.81 [18.18,21.53]	14.53 [13.45,15.68]	*
Tobacco use	(1 = yes)	34.90 [32.06,37.86]	35.37 [32.54,38.3]	29.99 [27.91,32.17]	33.25 [31.76,34.78]	*
Hypertension	(1 = yes)	13.18 [11.15,15.51]	18.34 [16.58,20.23]	20.41 [18.80,22.11]	17.58 [16.52,18.70]	*
Diabetes	(1 = yes)	3.71 [2.82,4.86]	5.35 [4.48,6.36]	6.94 [6.02,7.97]	5.45 [4.91,6.04]	*
Cardiovascular diseases	(1 = yes)	2.94 [2.02,4.24]	4.60 [3.67,5.72]	8.22 [7.06,9.52]	5.44 [4.81,6.14]	*
Hospitalization
Public sector	(1 = yes)	3.67 [2.92,4.61]	3.59 [2.91,4.42]	4.62 [3.93,5.42]	4.04 [3.61,4.52]	*
Private sector	(1 = yes)	3.62 [2.87,4.56]	3.54 [2.87,4.37]	3.51 [2.92,4.23]	3.55 [3.15,4.01]
Hospitalization costs A
Public sector	($PPP)	702.17 [1190.84]	632.08 [759.38]	702.06 [716.73]	697.71 [885.36]
Private sector	($PPP)	1919.42 [1724.00]	2085.73 [2299.49]	2039.12 [1719.43]	1946.94 [2090.69]

Data is presented in weighted frequency and 95 % confidence interval in brackets. P-values obtained from the chi-squared test for evolution through survey years (2003–2015). ^A Data presented in weighted medians and interquartile range in brackets. P-values obtained from simple quantile regression for survey years. Statistical significance: * = p < 0.05.

The majority of individuals did not report any hospitalization (n = 6806), approximately 92.7 % of the sample. Individuals reporting hospitalization in the public sector (n = 298), approximately 4.04 %, and hospitalizations in the private sector (n = 262), approximately 3.55 % of the sample (Table 2).

The model for direct costs of hospitalizations in the public sector indicated lower likelihood of hospitalization costs among individuals with health insurance (logit β=-0.891, p < 0.001; ME = −36.45, p < 0.01), individuals with recommended PA levels during leisure (logit β=-0.475, p < 0.05; ME = −31.03, p < 0.05), employed individuals (logit β=-0.373, p < 0.05), and among individuals with higher income level (logit β=-0.328, p < 0.001; ME = −16.91, p < 0.01). Lower costs were observed among women (GLM β =−0.556, p < 0.01; ME = −26.25, p < 0.05) (Table 3).

Table 3.

Two-part model for hospitalization costs of individuals using the public sector. ISA-Capital 2003–2015, São Paulo city, Brazil.

Variables	Categories	Logit			GLM			Marginal Effects
		β	SE	Sig.	β	SE	Sig.	dy/dx	SE	Sig.
Sex	(1 = female)	−0.049	0.151		−0.556	0.175	**	−26.25	11.30	*
Age	(years)	0.001	0.004		0.016	0.004	***	0.79	0.28	**
Skin color/ethnicity	(1 = white)	−0.015	0.158		−0.263	0.193		−12.08	11.16
Marital status	(1 = accompanied)	0.220	0.167		0.157	0.161		16.14	10.13
Education	(years)	0.012	0.018		−0.015	0.024		−0.12	1.31
Occupational status	(1 = employed)	−0.373	0.162	*	0.440	0.182	*	3.36	10.57
Income per capita	($PPP)	−0.328	0.069	***	−0.070	0.115		−16.91	5.63	**
Leisure-time PA	(1 = yes)	−0.475	0.220	*	−0.252	0.253		−31.03	14.68	*
Time spent in sitting position (ref. = lower tertile)	(1 = middle tertile)	−0.080	0.186		0.343	0.285		11.25	15.17
	(1 = upper tertile)	0.054	0.176		0.180	0.176		9.74	10.39
Obesity	(1 = yes)	−0.281	0.313		1.004	0.535		31.78	27.69
Tobacco use	(1 = yes)	0.492	0.161	**	0.304	0.221		33.98	12.95	**
Health insurance	(1 = yes)	−0.891	0.232	***	0.026	0.261		−36.45	14.40	*
Hypertension	(1 = yes)	−0.041	0.235		1.670	0.463	***	70.83	26.56	**
Diabetes	(1 = yes)	1.219	0.407	**	1.574	0.645	*	119.81	36.94	**
CVD		1.136	0.331	**	0.285	0.317		60.30	19.90	**
Year (ref. = 2003)	(1 = 2008)	0.243	0.221		−0.013	0.245		7.14	10.26
	(1 = 2015)	0.611	0.204	**	0.258	0.231		38.09	14.16	**
Constant		−1.503	0.449		6.724	0.612
N		7104			298			7104

GLM = generalized linear model (gamma family with log-link); β = regression coefficient; SE = robust standard error; dy/dx = average marginal effects. Statistical significance: *** = p < 0.001; ** = p < 0.01; * = p < 0.05.

PA = physical activity; CVD = cardiovascular diseases.

In contrast, higher likelihood of hospitalization costs was observed among individuals with a diagnosis of CVD (logit β=1.136p < 0.01; ME = 60.30p < 0.01) and tobacco use (logit β=0.492p < 0.01; ME = 33.98p < 0.01). Individuals with a diagnosis of type 2 diabetes (logit β=1.209, p < 0.01; GLM β=1.574, p < 0.05; ME = 113.87, p < 0.05) presented higher likelihood and costs of hospitalization. Individuals with a diagnosis of hypertension (GLM β=1.670, p < 0.001; ME = 70.83, p < 0.01), employed individuals (GLM β=0.440, p < 0.05), and increased age (GLM β=0.016, p < 0.001; ME = 0.79, p < 0.01) were associated with higher costs (Table 3).

Regarding direct costs of hospitalizations in the private sector, individuals with recommended PA levels during leisure presented lower likelihood of hospitalization costs (logit β=-0.494, p < 0.01; ME = −37.89, p < 0.01), while individuals with sedentary habits presented higher likelihood (logit β=0.442, p < 0.05; ME = 40.92, p < 0.01). Higher likelihood was also observed among women (logit β=0.455, p < 0.05), accompanied individuals (logit β=0.678, p < 0.001; ME = 40.49, p < 0.01), individuals with higher income levels (logit β=0.258, p < 0.01; ME = 11.79, p < 0.05), and individuals with health insurance (logit β=1.355, p < 0.001; ME = 65.10, p < 0.01). Higher costs were observed among individuals with obesity (GLM β=0.385, p < 0.05); individuals with hypertension diagnosis (GLM β=0.318, p < 0.05); and with increased age (GLM β=0.009, p < 0.01; ME = 1.01, p < 0.01) (Table 4).

Table 4.

Two-part model for hospitalization costs of individuals using the private sector. ISA-Capital 2003–2015, São Paulo city, Brazil.

Variables	Categories	Logit			GLM			Marginal Effects
		β	SE	Sig.	β	SE	Sig.	dy/dx	SE	Sig.
Sex	(1 = female)	0.455	0.201	*	−0.098	0.108		18.67	11.63
Age	(years)	0.009	0.005		0.009	0.002	**	1.01	0.32	**
Skin color/ethnicity	(1 = white)	0.108	0.194		0.149	0.136		13.75	12.48
Marital status	(1 = accompanied)	0.678	0.193	***	0.089	0.117		40.49	11.80	**
Education	(years)	0.031	0.019		0.004	0.014		1.88	1.27
Occupational status	(1 = employed)	−0.313	0.189		0.026	0.101		−15.08	11.55
Income per capita	($PPP)	0.258	0.091	**	−0.033	0.061		11.79	5.87	*
Leisure-time PA	(1 = yes)	−0.494	0.225	*	−0.221	0.121		−37.89	13.72	**
Time spent in sitting position (ref. = lower tertile)	(1 = middle tertile)	−0.001	0.199		−0.041	0.129		−1.80	10.11
	(1 = upper tertile)	0.442	0.198	*	0.225	0.121		40.92	15.25	**
Obesity	(1 = yes)	−0.008	0.306		0.385	0.159	*	20.29	18.22
Tobacco use	(1 = yes)	−0.130	0.187		−0.132	0.111		−14.01	11.67
Health insurance	(1 = yes)	1.355	0.191	***	−0.114	0.093		65.10	11.48	**
Hypertension	(1 = yes)	0.130	0.207		0.318	0.132	*	17.87	13.14
Diabetes	(1 = yes)	0.177	0.512		−0.416	0.305		−13.08	31.63
CVD	(1 = yes)	0.681	0.545		0.058	0.306		38.96	33.28
Year (ref. = 2003)	(1 = 2008)	−0.068	0.207		0.089	0.129		1.20	12.83
	(1 = 2015)	0.075	0.218		−0.039	0.118		1.79	13.03
Constant		−7.045	0.602		7.415	0.400
N		7068			262			7068

GLM = generalized linear model (gamma family with log-link); β = regression coefficient; SE = robust standard error; dy/dx = average marginal effects. Statistical significance: *** = p < 0.001; ** = p < 0.01; * = p < 0.05.

PA = physical activity; CVD = cardiovascular diseases.

No significant associations were observed for the likelihood of hospitalization costs, and costs with hospitalizations, in the public, and private sectors of the Brazilian health system, considering the ethnicity/skin color and education level of the individuals (Table 3, Table 4). Considering the evolution during the analyzed period, higher likelihood of hospitalization costs in the public sector were observed in 2015 (logit β=0.611, p < 0.01; ME = 38.09, p < 0.01) (Table 3).

4. Discussion

Our findings suggest that adhering to recommended leisure-time PA levels is associated with a lower likelihood of hospitalization and costs reduction in São Paulo city. This effect, potentially stems from improved well-being, prevention of chronic conditions, and delayed disease progression and mortality, particularly for conditions amenable to primary care [41]. Conversely, systematic reviews show physical inactivity is pathophysiologically linked to 35 chronic diseases, including cancers, representing significant healthcare costs [42,43].

Physical inactivity may be considered a major determinant of healthcare demand, driving chronic diseases and imposing a significant financial burden on individuals and health systems worldwide [44]. This burden includes direct costs from healthcare utilization and indirect costs like productivity loss and absenteeism. Global trends indicate that rising inactivity will increase future morbimortality and financial strain [[45], [46], [47]].

Our findings corroborate previous evidence from the United Kingdom [48], France [49], and Brazil [50]. Prior studies indicated that adherence to PA recommendations was associated with decreased hospitalization costs and inpatient days compared to inactivity. In Brazil, regular PA was linked to lower healthcare costs for individuals with CVD and DM [50]. Consistently, physically inactive individuals present higher healthcare costs, regardless of other demographic, health, or lifestyle risk factors [51].

A recent study by our research group, conducted in São Paulo city, suggests that adherence to recommended leisure-time PA is associated with a lower risk of impoverishment from catastrophic health expenditures, whereas chronic diseases like obesity, CVD, and DM were found to increase this risk [52]. The present study provides evidence that adhering to recommended PA levels is associated with a lower likelihood of hospitalization in São Paulo city, not just lower costs. This finding is consistent with a Swedish study of adults with CVD, which also found that recommended PA levels reduced length of stay, readmission, and mortality [53].

Physical activity is a key mechanism for managing obesity [42,43,54], a condition whose rising prevalence has increased the financial burden on Brazil's health system [55,56]. Although our study found no significant association between obesity and the occurrence of hospitalization, probably because of the analysis controls for other NCDs, it did find a positive association with higher direct costs among those who were hospitalized in the private sector. This result aligns with previous Brazilian research [57] and suggests that obesity prevention is a vital strategy for minimizing hospitalization costs.

Promoting leisure-time PA is a cost-effective strategy for preventing obesity and other chronic conditions, supplementing traditional treatments for DM [58,59]. Similarly, interventions that reduce sedentary time at work are also cost-effective for disease prevention [60]. Given the rising prevalence of inactivity in low and middle-income countries like Brazil [47,61], implementing cost-effective interventions that promote active lifestyles can reduce health system costs and increase population well-being.

However, achieving population-level changes in health behavior is challenging due to the complexity of lifestyle choices [62]. Adherence to PA is often hindered by socioeconomic inequalities (income, education) and demographic factors (sex, gender, age) [63]. Furthermore, health professionals frequently neglect to counsel patients on PA at the primary care level [64]. Yet, it is important to emphasize that promoting recommended levels of leisure-time PA requires a multi-sectoral approach beyond primary care guidance [65]. Investing in urban infrastructure like parks and bike paths is an essential strategy for fostering active lifestyles, particularly for low-income groups [66]. Furthermore, controlling urban violence and improving neighborhood social cohesion also play a relevant role in encouraging PA [67].

Our study also identified differences in direct hospitalization costs between São Paulo's city public and private sectors. We found lower occurrence, but greater cost of hospitalizations in the private sector. This finding is supported by national-level studies showing higher private sector expenditures for CVD treatments in Brazil [68]. This evidence may suggest a financing logic that incentivizes higher costs in Brazil's private sector. While a lack of consensus on quality differences makes it difficult to determine patient preferences, many studies indicate a perception of higher quality in the private sector in low- and middle-income countries. This perception may favor a focus on curative services for complex cases over long-term preventive care [[69], [70], [71]].

Our findings highlight that adherence to WHO’s PA recommendations was associated with a lower likelihood and lower direct cost of hospitalization in both public and private sectors. Therefore, interventions promoting PA and minimizing sedentary habits might benefit all individuals within the Brazilian health system, contributing to comprehensive healthcare [23,72].

The present study has several limitations. Direct hospitalization costs were estimated using macro-costing techniques with secondary data from DATASUS and CBHPM, which had data availability limitations. Therefore, the results represent general population patterns and should be interpreted with caution. However, given the scarcity of surveys on healthcare costs, this approach provides valuable evidence on hospitalization cost drivers. Furthermore, the procedural cost data were sourced from official and trustworthy public and private organizations.

It is important to recognize that PA is a complex health behavior influenced by unmeasured characteristics due to data limitations in the ISA-Capital survey, creating a potential for omitted variable bias. Nevertheless, using the leisure-time domain and WHO's recommended cutoff for PA effectively captures key behavioral dimensions and the preventive potential against NCDs.

Finally, the cross-sectional design of the survey only allows to establish associations given its observational nature, and the findings cannot be extrapolated to rural contexts. Additionally, potential temporality bias must be considered, stemming from the retrospective nature of hospitalizations (measured in the twelve months prior to the interview date) and the current assessment of leisure-time physical activity (measured in the week prior to the interview date). Although major part of hospitalizations was associated with clinical procedures and exams, leisure-time physical activity may have been registered subsequently to the occurrence of hospitalization, thus associations observed in the current study should be interpreted with caution to avoid inferring any causation.

Although there were numerous missing observations, the sample size remained valid for inferential analysis. And considering the scarcity of similar individual-level data for São Paulo's urban population, and other urban centers in the Latin America region, this study provides original contributions for decision making in health policy and guiding future longitudinal research on the associations of lifestyle factors and healthcare utilization costs.

5. Conclusions

The evidence of the present study suggest that the promotion of active lifestyles, with incentives on the adherence to WHOs recommendation of weekly frequency of leisure-time physical activity; minimizing the weekly frequency of time spent in a sitting position; and the prevention of obesity; may comprise important health policy strategies with the potential to reduce societal costs with hospitalizations in São Paulo city, considering both public and private sectors of the Brazilian health system. Findings from the study adds relevant evidence on the association between lifestyle factors and healthcare costs in an urban population context from low- and middle-income country, and may be relevant for other similar countries, specially, in the Latin American region. Nevertheless, the associations observed in the current study should be interpreted with caution, given potential temporal mismatch between the registry of leisure-time PA and the occurrence of hospitalization. Furthermore, additional longitudinal studies on the associations between adherence to recommended leisure PA levels, sedentary habits and obesity, and healthcare costs are required to provide robust evidence for public policy.

Ethical considerations

The present study and the three editions of the ISA-Capital survey were assessed and approved by the Ethics Committee of the School of Public Health at the University of Sao Paulo, Brazil (CAAE 48271721.4.0000.5421; e CAAE 003.0.162.000-08; 32344014.3.0000.5421; 36607614.5.0000.5421, respectively). The study protocols of the three editions of the ISA-Capital survey were also assessed and approved by the Ethics Committee of the Municipal Secretary of Health of Sao Paulo City (CAAE 32344014.3.3001.0086). Individuals participating in the ISA-Capital survey signed written consents containing information on their rights and a description of the objectives of the survey before recruitment, by the ethical principles of the Helsinki Declaration.

8. Contributions

Authors made substantial contributions to the conception and design of the study, acquisition of data, analysis and interpretation of data, drafting the article and revising critically for important intellectual content, and final approval of the version to be submitted.

CRediT authorship contribution statement

Lucas Akio Iza Trindade: Writing – original draft, Project administration, Methodology, Funding acquisition, Formal analysis, Data curation, Conceptualization. Jaqueline Lopes Pereira: Writing – review & editing, Validation, Supervision, Resources, Investigation, Data curation. Marcelo Macedo Rogero: Writing – review & editing, Validation, Resources, Investigation, Funding acquisition. Regina Mara Fisberg: Writing – review & editing, Validation, Resources, Investigation, Funding acquisition. Flavia Mori Sarti: Writing – review & editing, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation.

Funding

The study was financed in part by the São Paulo Research Foundation (FAPESP), Brazil (Processes #2022/11919-4, #2021/05327-4, #2017/05125-7, #2012/22113-9, #2009/15831-0, #2007/51488-2, and #98/14099-7), the São Paulo Municipal Health Department (Grant 2013-0.235.936-0), the National Council for Scientific and Technological Development (Grants 502948/2003-5, 481176/2008-0, 472873/2012-1, 473100/2009-6, 402674/2016-2, and 301597/2017-0), and the Brazilian Ministry of Education − Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), Brazil (Finance Code 001).

Funding sources were not involved in the study design, collection, analysis and interpretation of data, writing of the report and decision to submit the article for publication.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Lucas Akio Iza Trindade reports financial support was provided by São Paulo Research Foundation (FAPESP). Regina Mara Fisberg reports financial support was provided by São Paulo Research Foundation (FAPESP). Flavia Mori Sarti reports financial support was provided by São Paulo Research Foundation (FAPESP). Marcelo Macedo Rogero reports financial support was provided by São Paulo Research Foundation (FAPESP). Jaqueline Lopes Pereira reports financial support was provided by São Paulo Research Foundation (FAPESP). Lucas Akio Iza Trindade reports financial support was provided by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), Brazil. Regina Mara Fisberg reports financial support was provided by São Paulo Municipal Health Department. Flavia Mori Sarti reports financial support was provided by National Council for Scientific and Technological Development. Regina Mara Fisberg reports financial support was provided by National Council for Scientific and Technological Development. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: