Twelve middle schools in Central Texas (six intervention and six control schools) were recruited and pair-matched based on school characteristics at baseline. The matching criteria used were student ethnicity, percent of students who were economically disadvantaged and percent girls in athletics. The study design was a randomized clinical trial with one school from each matched pair randomly assigned via computerized randomization to either the IMPACT program, or the control group (usual health program). Girls from the sixth grade who were enrolled in two semesters of physical education were eligible to participate in the study. At the beginning of the study, baseline measurements were conducted in the Fall of 2000, followed by interim measurements conducted in Spring 2001, and final measurements taken in Spring 2002. The actual intervention duration was from late November 2000 to May 2002, spanning a period of approximately 1 1/2 school years. Written parental consent and student assent were obtained before participation in the measurements and for publication of study results. The study was approved by the Human Subjects Review Committees at the University of Texas-Houston, School of Public Health, Baylor College of Medicine, and school district research committees.

The IMPACT intervention was based on the theoretical foundations of the Social Cognitive Theory and the Trans-Theoretical Model. These theories integrate determinants of behavior (such as outcome expectations, self-efficacy, behavioral capability, and environment) with methods of behavior change. In accordance with the premises of these theories, the IMPACT intervention sought to affect behavior change through promoting active learning in classrooms, as well as through environmental reinforcement. To this end, the intervention consisted of three major components: a health curriculum for grades 6 and 7 which included classroom lessons and behavioral journalism, a physical education program, and a school food service component that emphasized calcium rich food choices. The use of peer-based behavioral journalism involved the use of media such as a school-based newsletter with role model stories to increase adoption of desired behaviors. The over-arching goal of the physical activity component of the IMPACT trial was to improve bone health in the study sample by increasing overall levels of physical activity, specifically focusing on increasing weight bearing physical activity. To this end, the intervention employed a 6^th grade health curriculum which included 16 sessions that were implemented during physical education classes (3 times/week). The lessons in this curriculum were designed to promote increased consumption of calcium-rich foods and increased activity, specifically weight-bearing physical activities, while participating in behaviorally-based and active lessons adapted to the physical education environment. During 7^th grade, a series of science-based lessons were administered during science classes. The physical education component of the program known as IMPACTivities, was implemented in the 6^th and 7^th grades during physical education (PE) and athletics classes. The PE classes focused on an initial 10 minute warm-up (range: 5–15 minutes), which consisted of high impact activities such as rope-jumping, circuit training and box-step activities. This time was included in the outcome measure for WBPA. The overall emphasis of the PE program was to increase the duration of WBPA as well as overall levels of MVPA.

Various measures were used to insure the proper implementation of the physical activity intervention at the school level. These measures included regular observation of physical education and health classes and teacher checklists for completion of designated lessons, as well as observation of specific lessons. During 7^th grade, the number of advisory periods and newsletters, as well as the number of science lessons taught were enumerated for each school.

The main exposure for the presented analysis was the IMPACT intervention. Main outcome variables related to activity included: 1) duration of WBPA, MVPA and VPA, 2) duration of sedentary activities before and after school on weekdays and on weekends, and 3) duration of physical activity outside of the school, i.e. before and after school on weekdays and on weekends. For each outcome of interest, baseline and follow up values were calculated and compared between participating intervention and control schools.

Physical and sedentary activity data were collected using multiple administrations of the Self-Administered Physical Activity Checklist (SAPAC), as well as with the Calcium, Osteoporosis and Physical Activity (COPA) questionnaire.

The SAPAC was the primary instrument to assess total activity, WBPA, MVPA, and VPA. The SAPAC collected data on the intensity, duration and types of physical activity and consisted of 25 physical activities with spaces for listing up to three "other" activities that may have not been included in the main activity list. The SAPAC also had an additional section for sedentary activities such as television and video watching at three time periods during the previous day, i.e. before, during and after school. The self-report version of the SAPAC was previously validated against heart rate monitors (r = 0.57, p = .0001), and interviewer-administered checklists, (r = 0.76, p = .0001) in 125 fifth-grade children from four regions of the United States 26. To improve precision and reliability, the SAPAC was administered to study participants on three separate days, which included one weekend day and two random weekdays, and an average score was calculated for each participant.

Values for each category of physical activity (MVPA, WBPA, VPA, etc.) were calculated using the sum of the minutes of the activity corresponding to these intensity levels for each student. Sedentary activity levels were measured using three variables from the SAPAC: i) mean daily minutes of television-video viewing; ii) mean daily minutes of computer-video game playing; and iii) mean daily minutes of combined sedentary behavior (sum of television-video viewing and computer-video game playing minutes). Sedentary variable values represent mean values from the three SAPAC administrations for each student.

Prior to analysis, items on the SAPAC were coded to provide information on weight bearing physical activities and metabolic equivalent values. Classification of activities as WBPA or NWBPA (non-weight bearing physical activities) was obtained from previously cited literature on the subject 2728. SAPAC items were assigned a metabolic equivalent or MET value based on established guidelines 29. The MET value provides an accurate estimate of the intensity of a physical activity and also provides a reliable measure of energy expenditure during an activity relative to energy expenditure at rest 2930.

The COPA (Calcium, Osteoporosis and Physical Activity Questionnaire) was primarily used to gather data on psychosocial variables related to calcium intake, physical activity and osteoporosis. The COPA was specifically developed for IMPACT and the key constructs measured by this instrument included variables such as calcium knowledge; weight bearing exercise and osteoporosis knowledge; physical activity expectations, etc. The COPA was developed and adapted from previous instruments such as the School Based Nutrition Monitoring (SBNM) student questionnaire 31 and the Health Behavior Survey 32. The COPA was evaluated for reproducibility using a sample of 93 sixth grade girls from the same sample with duplicate administration of the questionnaires 11 days apart. Reproducibility was evaluated using percent agreement or Spearman correlations. The correlations for physical activity items ranged from 0.42 to 0.67. An evaluation of the physical activity self-efficacy scale using the baseline IMPACT sample revealed a Cronbach's alpha value of 0.87.

Seven items on the COPA directly assessed student participation in physical activity and support from friends and family for the same. These items asked students to rate the following: i) how often during the past month their family did physical activity with them; ii) how often during the past month their family encouraged them to do physical activity; iii) how often during the past month their friends did physical activity with them outside of school; iv) how often during the past month their friends encouraged them to do physical activity; v) how safe it was for them to play outdoors in the neighborhood with their friends without adult supervision; vi) how many sports teams they were on during the past 12 months (not including PE classes); and vii) whether they were currently participating in any other organized physical activity such as martial arts, gymnastics or tennis. Answers to these questions from the COPA were used for data on covariates.

The instruments and details on their administration protocols may be viewed at the IMPACT project website at: http://www.sph.uth.tmc.edu/DellHealthyLiving/default.aspx?id=4016

The main endpoints for this study were measures of physical activity (WBPA, MVPA, VPA, total of all activities, sedentary activities, before/after school activities, and weekend activities). The analysis for the above variables was performed using mixed model analysis of covariance (ANCOVA) on the calculated scores for each variable (pre- and post-intervention) to test for differences among groups using specific techniques applicable to the unique challenges associated with the analysis of group-randomized trials. The essence of the unique methodological issue that arises with a group randomized design is that the condition (intervention/control) is assigned at the group (school) level, whereas the data is gathered at the member (student) level. Since students in a cluster (in this case, the school) are more likely to be similar to each other than to students in other schools, an adjustment is required to account for the clustering of values within schools. This is due to the fact that the within-schools correlation in the data, indexed by the intra-class correlation coefficient (ICC) adds an additional component of variance to the variability of the means in the intervention group, over and above the variance attributable either to the individual units (students) or to the intervention itself. It is essential to account for this within-schools correlation component during analysis in order to prevent inflated estimates of statistically significant differences between groups (represented by exaggeratedly lower p-values and lower estimates of standard error) than those that actually exist 33.

A pretest-posttest control group design analysis using the General Linear Mixed Model was utilized for the data obtained for this study. The variables were analyzed with the intervention arm (intervention/control) as the independent variable, and the specific physical activity/sedentary activity endpoints as the dependent variables. The analysis examined change in the dependent variables over time adjusting for baseline differences. The covariates examined were specific to the relevant response variable, but included ethnicity, body mass index, menarcheal status (yes/no), participation in PE/sports teams, neighborhood safety, and familial/friend support. Posttest data were analyzed with regression adjustment for covariates measured at baseline, thereby including time-related information without modeling time explicitly in the analysis 33. This option was particularly suitable to our study as data was considered from only two time intervals, thus making it possible for time to be reflected only indirectly in the analysis of posttest data that made a regression adjustment for baseline values 34. Time and treatment condition were modeled as fixed effects, with the individual student modeled as a random effect, and the school unit modeled as a nested random effect. This model thus accounted for variation effects attributable to the schools, to residual error, and to the heterogeneity among the school-specific slopes.

All statistical analyses were conducted using STATA version 9.0 program for longitudinal and mixed modeling, a mixed model regression program specifically suited for analysis of data from group-randomized trials 35.

Study participants were mostly white (72%) with 12% Hispanic and 5% African-American (Table 1). Mean age at baseline was 11.6 years (± 0.4 years) compared to 13.2 years (± 0.4 years) at the conclusion of the study. Most of the girls at baseline were pre-menstrual (20% menstruating at baseline) relative to 67% menstruating at the follow-up phase of the study.

In each category of the physical activity variables (WBPA, MVPA, VPA, MET-weighted WBPA and MVPA), increases in activity levels were observed in the expected direction in intervention schools relative to the control schools, but results were only significant for VPA. Relative to the students in the control schools, students in the intervention schools engaged in more daily minutes of WBPA (difference = 5.12 minutes, 95% CI = 4.82–5.42, p = 0.33) at follow up compared to baseline (Table 2). Although not statistically significant, this represents an increase of approximately 6.0% in daily WBPA from baseline for students in the intervention schools, while WBPA for students in the control schools increased by 1.6%. Relative to students in the control schools, students in the intervention schools had higher overall total daily minutes of physical activity and daily MVPA minutes, although neither change was statistically significant. In contrast, the total daily minutes of vigorous physical activity (VPA) were significantly higher at follow-up for girls enrolled in intervention schools (difference = 6 minutes, 95% CI = 5.82–6.18, p = 0.05) compared to girls in control schools. This represents an increase of 45.4% in VPA minutes from baseline for students in intervention schools, while VPA minutes at follow-up relative to baseline decreased for students in control schools by 4.1%. There were no statistically significant changes in MET weighted minutes for both WBPA and MVPA (Table 2).

Students in intervention schools also spent approximately 9 more minutes per day (95% CI = 8.69–9.21, p = 0.04) than those in control schools in daily after school physical activity at follow-up, suggesting a statistically significant difference in the follow-up means of the two groups (Table 3). This represents an increase of 13.3% in daily after school activity minutes from baseline for students in intervention schools, while those in control schools increased by 1.7% for the same variable. Students in intervention schools spent significantly higher number of minutes in daily weekend physical activity (difference = 19 minutes per day, 95% CI = 18.40–19.60, p = 0.05) relative to those in control schools at follow up. There was no statistically significant difference in the follow-up means of the intervention and control groups with respect to daily before school activity minutes.

Relative to girls in the control schools, daily TV and video minutes were lower for students in the intervention schools at follow-up (↓12.11 minutes, 95% CI = 11.74–12.48, p = 0.05) suggesting a decrease in TV/video watching from baseline for students in intervention schools of 16.7%, while TV/video watching in students in the control schools increased by 17.9% (Table 3). Total daily minutes of sedentary activity were significantly lower for students in intervention schools relative to those in control schools at follow-up (↓17 minutes, 95% CI = 16.49–17.50, p = 0.04). This represents a decrease in sedentary activity minutes from baseline for students in intervention schools of 1.7%, while sedentary activity minutes in students in control schools increased by 33.5%.

The main strengths of our study include its strong study design, high quality measurement of outcomes, and high reliability and validity of measurement instruments. In particular, one of the strengths of the study is that the physical activity data collection instrument, the SAPAC was administered three times within a single week at baseline (two weekdays, and one weekend day), reducing variability of this measure from 40 to 25% (data not shown). The majority of PA variables were calculated from the averages obtained from the reported PA levels for three days. This provided us with a more reliable estimate of the subjects' daily activity patterns during the course of both weekdays and weekends.

One of the potential weaknesses of our study includes the evaluation of data at two distinct time periods only, i.e. at baseline and follow-up. Data obtained through process evaluation and for other measures (e.g. the dietary intakes) suggest that the intervention was most effectively implemented in the first year (6^th grade). For several key behavioral variables (e.g. calcium intake), optimal (peak) effects were obtained at the interim measurement phase of the intervention, followed by a moderate decline in the variable values at the final measurement phase of the study. While data were collected on nutrition-related variables at the interim phase of the study, data were not collected for the physical activity variables at this stage of the study, thus precluding the possibility of analyzing trends in physical activity patterns over the entire duration of the intervention. In light of the study implementation being close to optimal in the first year of the intervention, it is conceivable that more significant effects in WBPA and MVPA may have been obtained at the interim phase of the study.

Another limitation of the study is the relatively few number of groups randomized to treatment conditions. This feature of our study is quite typical of group-randomized trials due to the costs involved in recruiting and administering the intervention to the groups. Although our study utilized a strong analytical approach and had sufficient power to detect differences between the groups, a higher number of groups would yield greater statistical power and accuracy in determining intervention effects.

Additional potential study limitations include reliance on self-reported measures of physical activity (generally leading to an over-reporting of PA levels) and limited information on additional possible covariates such as parental involvement. Data from other PA studies suggest that parental involvement in school or community based interventions may improve compliance and participation in intervention activities 19.