Over 80% of adolescents worldwide are insufficiently active, posing massive public health and economic challenges. Declining physical activity (PA) and sex differences in PA consistently accompany transitions from childhood to adulthood in post-industrialized populations and are often attributed to psychosocial and environmental factors. An overarching evolutionary theoretical framework and data from pre-industrialized populations are lacking. This cross-sectional study tests hypotheses from life history theory, that adolescent PA is inversely related to age, but this association is mediated by Tanner stage, reflecting higher and sex-specific energetic demands for growth and reproductive maturation. Detailed measures of PA and pubertal maturation are assessed among Tsimane forager-farmers (age: 7–22 yrs.; 50% female, n=110). Most Tsimane sampled (71%) meet World Health Organization PA guidelines (≥60 minutes/day of moderate-to-vigorous PA). Like post-industrialized populations, sex differences and inverse age-activity associations were observed. Tanner stage significantly mediated age-activity associations. Adolescence presents difficulties to PA engagement that warrant further consideration in PA intervention approaches to improve public health.

Experimental design

Data for this study were collected among Tsimane children and adolescents from four adjacent villages located along the Maniqui River in the summer of 2014 in collaboration with the Tsimane Health and Life History Project. To encompass the transition from childhood to adulthood, a wide age range of participants (ages 7–22) was recruited, in line with a broader conceptualization of adolescence that extends the upper limit to align more closely with brain development. The average age of menarche in the Tsimane is 13.9 years. Tsimane children and younger adolescents play with others in neighboring families, assist in caring for younger siblings and food production, and spend limited time in formal schooling compared to children in post-industrialized populations. Older Tsimane adolescents in this age range are typically married with children, and produce food, but typically reside in family clusters and are still reliant on their parents and grandparents for additional food production and allocare.

Participants were recruited during their annual medical exam with the THLHP medical team by AC and PH. All individuals between the ages of 7–22 who attended their medical exam were invited to participate (n=126, 37% of individuals in this age range were residents in these communities in 2014).

Participation lasted 3–4 days allowing inclusion of all eligible participants from each community during the medical team visit (1–3 weeks). On Day 1, participants and parents provided informed consent/assent, completed a structured interview, and had anthropometric and strength measurements taken (detailed below). Participants wore ActiGraph wGT3X+ accelerometers on their hip for the duration of study participation. On days 2 and 3, participants provided first void urine samples to measure urinary hormones (detailed below). On days 2, 3, and 4 (where possible) participants completed 24-hour PA recall interviews while concurrently examining Actigraph data on a computer screen to validate wear-time and aid in recall. Small gifts were provided to remunerate participants for each day of study participation including pens and notebooks, backpacks, soccer jerseys, sardines, and oil.

All study methods were approved by the Institutional Review Board (IRB) at the University of New Mexico (HRRC # 07–157). Informed consent was established at three levels: (1) the Tsimane governing council; (2) village leadership; and (3) study participants. Informed assent was established with parents/guardians of minors.

Participant Characteristics

A total of 110 Tsimane children and adolescents (mean±SD age = 13.7±3.2 y, 50% female) were sampled from four villages (Supplemental Figure 2). Table 1 summarizes sample characteristics for all variables of interest in the full sample, separately by sex, and within-sex by Tanner stage. Males and females were similar with respect to mean age, weight, height, DHEA, testosterone, and distribution across Tanner stages, Χ² (4, n=110) = 1.34, p=0.85. Males have significantly higher grip strength, non-significantly (ns) higher mean height velocity (4.5 ± 2.2 vs. 3.7 ± 2.9 cm/year, p=0.09); while females had significantly higher mean body fat. The highest mean height velocity across Tanner stages was the same in males and females (6.4 cm/year) and was observed in Tanner 2 females and Tanner 3 males.

Measures

Age and Anthropometry. Age was obtained from census data collected by the Tsimane Health and Life History Project (THLHP) over the past 20 years. Height was measured to the nearest mm with a portable stadiometer (Seca 213) and weight with a digital scale (Tanita BC-1500). Body fat was estimated using the Jackson & Pollack 3-measurement skinfolds taken in triplicate using FatTrack®II Digital Body Fat Calipers (AccuFitness). Grip strength was measured taking the average of three measurements using a hydraulic dynamometer (Baseline, TN).

Free-living PA and sedentary time. ActiGraph wGT3X+ accelerometers were used to estimate PA/sedentary time and intensity. Actigraphs were initialized with a 30hz sampling rate. Participants were fitted with the device on the hip, instructed on the proper positioning and to only remove the device when submerging themselves in water. Daily 24-hour activity recall interviews were completed to assess activity types among participants who lived within a reasonable distance from the mobile medical camp and could return daily. To aid participants’ recall of the previous day and validate Actigraph wear-time in real-time, investigators and participants viewed Actigraph data together on a laptop screen while completing the interview.  Completed files were downloaded using Actilife software with the normal filter into 60-second epochs and converted to CSV tables. Data were processed in R (v4.1). Wear-time was additionally assessed using a criterion of 60 consecutive minutes of zero-counts with an allowance of 1–2 minutes of counts to 100. Epochs were delineated into periods of sedentary, light and MVPA activity-levels with an algorithm developed in Brazilian adolescents using vector magnitude counts. Sixty-second activity data were summed into person-day summaries including separating the 12-hour periods from 7am to 7pm and 7pm to 7am. Person-day summaries were averaged into individual daily estimates of activity. Participants were included in analysis if there were at least 10 hours of valid wear time between 7am and 7pm. Reported daily estimates for light intensity, MVPA, and steps/day were based on a 24-hour day from 7am until the following morning at 7am; while sedentary activity was based on a 12-hour period from 7am-7pm to separate daytime sedentary behavior from sleep. For daytime sedentary behavior, person-day level data were included if there were at least 9 hours of valid wear time between 7am and 7pm. Of the initial 110 enrolled in the accelerometry protocol, four did not meet the criteria of at least 1 valid day daytime sedentary behavior and an additional two participants did not meet the valid wear-time criteria for PA (Figure 4). The remaining participants provided 1.9 days (SD=0.4) of data. Previous work indicates that 1-3 days of accelerometer wear-time is sufficient to capture reliable estimates of habitual PA. We examined the reliability of our estimates using a larger unpublished dataset of wrist-worn Actigraph estimates of steps/day for 1986 person-days among n=299 Tsimane (age 7–22) with ≥ 4 valid-days of PA data (collected 2019–2022). The correlation of steps/day estimated using all valid days with steps/day estimated with a random selection of only 2 days was strong (r=0.76, p < 0.001) with good reliability (ICC = .74, p<0.001), suggesting 2 valid days of accelerometer data is sufficient to estimate habitual PA.

Daily 24-hour activity recall interviews were completed among n=99 participants, covering an average of 3.8 days (SD=0.06), 15.6 hours per day (SD=3.0), and 3601 minutes per participant (SD=962). In total, 62 activities were recorded with mean=37 activities (SD= 9.1) recorded/participant. The 62 activities were coded using previously used Troject time allocation codes. Activities were grouped into 9 bins along two dimensions: intensity (sedentary, light, moderate to high) and type (transportation, habitual/obligatory, and leisure) by AC and HD and modified until consensus was reached among all authors who have completed fieldwork with the Tsimane (Table S2). Sleep was not included in the results and accounted for 43.3% of total recall time. Time periods when participants couldn’t recall what they were doing for a given time-period, and activities involving medical exams or interaction with anthropologists were also excluded, accounting for 8.9% of time captured by the interviews.

Tanner Stage. Tanner staging is considered the gold standard for tracking the development and sequence of secondary sex characteristics throughout puberty. It is typically evaluated on a scale from 1–5 by a physician through clinical examination of secondary sex characteristics such as pubic hair growth and breast/genital development. However, it is challenging to perform in field settings because of the sensitivity of the assessment and the need for privacy – particularly in anthropology field settings where direct examinations, interviews, or pictorial representations of breasts and genitals are not culturally appropriate. Thus, Tanner stage 1–5 was assessed by two authors (AC and PH) independently based on secondary sexual characteristics visible through clothing, including breast development and underarm hair, as well as menarche, lactation status and reproductive history obtained from THLHP medical records. Facial/underarm hair, muscular development, jaw shape, and voice change served as additional maturation cues in males. Inter-rater reliability was high, with discrepancies reflecting only ±1 stage difference that were resolved through discussion between researchers. Tanner stage assessments were validated with measures of growth and reproductive maturation.

Urinary Hormone Assays. First morning void urine specimens were collected by participants on two consecutive days, and frozen in liquid nitrogen within several hours of specimen collection. They were stored for up to one month in liquid nitrogen, before being transferred on dry ice to the US where they were stored at -80C for two years before analyses. After arrival, specimens were thawed, specific gravity was measured with a refractometer (Atago, Inc), and urine specimens were analyzed in duplicate via enzyme immunoassay for DHEA (Enzo Life Sciences, ADI-901-093) and testosterone (R156/7). All specimens were run on the first freeze thaw, and results were specific gravity corrected. See Supplemental Information for a detailed description of DHEA and testosterone during maturation.

Height Velocity. Height velocity in 2014 was computed using height measurements collected 1-4 years prior to 2014, with the majority (70%) taken 2 years prior to the study. Height velocity was calculated as:

Statistical analysis

Analyses were performed in R version 4.1. Sex differences in participant characteristics were assessed using independent sample t-tests for continuous variables and Chi-square tests for distribution across Tanner stages. Spearman correlations were tested to validate Tanner stage with log transformed hormone levels (DHEA and testosterone which were both skewed) and measures of somatic growth and maturation (height velocity, grip strength, and body fat). Non-parametric tests were used because Tanner stage is ordinal, and body fat in males violated normality assumptions. To test P1 and P2, separate linear regression models were run for each activity outcome (sedentary time, light intensity PA, MVPA, steps/day) with age, sex, and the age x sex interaction included as independent variables. To test the mediating role of Tanner stage on age-sedentary/PA associations (P3), structural equation models (SEM) were fit in the lavaan package version 0.6-11 using listwise deletion for participants missing sedentary (n=4) or PA outcome data (n=6). Diagonally weighted least squares estimation was used to account for Tanner stage  being ordinal as opposed to continuous. Age and outcome variables were standardized to facilitate model convergence given the vastly different scales of the variables. Categorization and visual representation of self-reported PA was performed in R 2022.02.0 using the ggplot2 and lmer packages. 

Validity of Tanner Stage with Hormone Levels and Somatic Maturation Markers

Estimates of the within-sex Spearman correlations between Tanner stage assessment and DHEA, testosterone, height velocity, grip strength and body fat are presented in Supplemental Table 1. Tanner stage assessments were significantly correlated with all these variables in both sexes (p’s < 0.01), with coefficients in females ranging from r = 0.46 for testosterone to r = -0.79 for height velocity and in males from r = -0.38 for height velocity to r = 0.92 for grip strength.