The Sun Protection of Florida's Children project (ClinicalTrials.gov
number, NCT 00346021) is a cluster randomized trial, in which participating schools were randomly assigned to either control or intervention status (). Block randomization was used by the project's biostatistician to ensure the number of schools assigned to each intervention group was balanced; however, there was no matching or stratification of schools. Participating schools were chosen from Hillsborough County, Florida, which includes Tampa and its surrounding suburbs, as well as largely rural areas in the southern and eastern areas of the county. All 140 kindergartens through grade 5 schools in the School District of Hillsborough County were eligible to participate.
Comparison of intervention and control schools in the Sun Protection of Florida's Children project.
On the basis of sample size requirements of the study, 20 schools were needed to assess a 10% absolute difference in hat use among intervention and control schools with 80% power. We enrolled 24 schools, however, to allow for possible attrition. We extended an invitation to all 140 schools to participate and closed accrual after 24 schools agreed to participate (all of which met study eligibility criteria). Of the 140 schools invited, 42 schools declined to enroll through fax or phone communication. For each school, we determined the overall enrollment of fourth-grade students and the school uniform policy (required, voluntary, or none) from school officials. We also examined socioeconomic characteristics of the school's surrounding neighborhood by use of census-derived data (urban vs nonurban, median household income, and percentage of population with less than high school education).
Two schools (one in the intervention group and one in the control group) withdrew from the study after consenting but before any data were collected or intervention activities took place. The control school withdrew because of a change in leadership, and the intervention school withdrew because it determined that it could not comply with all study requirements. A third school withdrew from the intervention arm after baseline data were collected but before receiving the intervention. Missing data because of the withdrawal of schools were ignored in the analysis; missing data were not imputed. The number of students who received the allocated assignment was 1115 for the intervention schools and 1376 for the control schools.
This study was approved by the University of South Florida Institutional Review Board. Activity involving human participants did not commence until after approval from the Institutional Review Board. Written informed assent was obtained from each student before the start of the intervention.
To be eligible for the study, schools were required to meet the following eligibility criteria: 1) the school had at least one fourth-grade class (with a minimum of 20 students per class), 2) the principal and all fourth-grade teachers agreed to participate, 3) the school provided a minimum amount of outdoor student activities defined as at least one outdoor session (physical education, recess, and/or lunch) lasting at least 20 minutes and occurring a minimum of three times per week, 4) outdoor activities occurred in an area that was exposed to sunlight and could be observed by data monitors, 5) schools had the ability to accommodate the major intervention components, 6) school environment was expected to be stable over the intervention time period (ie, no anticipated changes in leadership or student population served), and 7) if randomly assigned to the control group, schools were willing to delay other sun protection initiatives until the project was completed. It is also important to note that all schools allowed children to wear hats while outdoors at school; this behavior was not specifically prohibited or restricted at any school.
The intervention was based on the Theories of “Reasoned Action” (38
) and “Planned Behavior” (39
). In brief, the Theory of Reasoned Action posits that intentions to engage in healthy behaviors directly predict actual behaviors and are in part influenced by individual attitudes and the attitudes of peers regarding the behavior (subjective norms). The Theory of Planned Behavior extends this theoretical model further by theorizing that perceptions of behavioral control influence behaviors directly, and also indirectly, by affecting intentions (40
). Perceived behavioral control is a person's perception of his or her ability to enact a healthy behavior.
The intervention was therefore structured to accomplish several goals. First, the intervention sessions were designed to increase sun protection knowledge of participating children and foster more positive attitudes in regard to wearing hats. In addition, the intervention was designed to change the subjective norm of wearing hats at school by instilling the belief that peers would also have positive attitudes regarding hats and would be wearing hats when outside. The intervention lasted 23 months from August 2006 to June 2008.
To accomplish these goals, the intervention included the following components that occurred over the school year: 1) Introductory meetings with principals, teachers, and physical education instructors were conducted to explain the project and obtain their informed consent. 2) A package of materials was sent to parents at the start of the school year that explained the project and obtained consent to participate. 3) An initial introductory session with students was conducted to explain the project and obtain assent to participate. 4) A 45-minute comprehensive sun protection educational session for fourth-grade students was carried out in classrooms by a community health education organization (MOREHEALTH). 5) Three 60-minute follow-up sessions addressed the benefits of sun protection (with emphasis on hat use), promoted favorable attitudes about sun protection, and made clear that fourth-grade students were both allowed to wear hats at school and should be wearing hats while outside at school. 6) Each participating child was provided two free wide-brimmed hats (one to use at school and one to use at home). 7) Three brief follow-up sessions were scheduled (one that addressed logistic issues, such as where hats would be stored, how they would be identified for each student, and how they would be distributed before outdoor activities; one that addressed any problems that classes were having; and one that was reserved for hat collection at the end of the school year). At the end of the school year, hats were cleaned and stored over the summer and were to be redistributed to the same students at the start of their fifth-grade school year.
At control schools, similar introductory meetings were held with principals, teachers, and students to explain the project and to obtain informed consent of parents and assent of children. Students at control schools received three to five 60-minute educational sessions on topics in science unrelated to sun protection. The topic of the lessons included learning about the sun as a star, the nature of light and heat, and the reflectance and absorbance of heat energy.
Twenty-two schools were enrolled at the start of the 2006–2007 school year. Data were systematically collected from all participants during the following three time periods: baseline (from August 8, 2006, through September 29, 2006) before intervention activities took place at the specific school, after the intervention was implemented at the school (November 30, 2006, through March 15, 2007), and again in the spring (March 27, 2007, through May 22, 2007).
Child's Sun Protection Behavior Survey
The Child's Sun Protection Behavior Survey was a 13-item self-administered instrument designed to assess the frequency with which children used various sun protection methods at school and at home. The questions asked about use of long-sleeve shirts, hats, sunscreen, and sunglasses and clarified the type of hat worn (wide-brimmed vs other types) if hat use was reported. The main outcome assessed with this survey was the child's use of wide-brimmed hats when not at school. This question was worded, “When you were outside playing this weekend or after school, how often did you wear a wide-brimmed hat?” Possible answers were as follows: 1 = “Never”; 2 = “Rarely”; 3 = “Sometimes”; 4 = “Often”; and 5 = “Always.” This question has been previously validated as a measure of children's hat use (21
). In addition to the sun protection measure, each student completed a brief demographic survey.
Direct Observation of Hat Use
The primary outcome of the study, hat use at school, was measured by direct observation. Four research assistants underwent standardized training in direct observation of hat use at school. Research assistants first underwent a 60-minute training session in which wide-brimmed hats were defined (ie, a hat with at least a 2-inch brim covering the circumference of the hat) and differentiated from other hats and other forms of head coverings. Each research assistant was then trained in a standardized method of assessing the number of students who were observed in an outdoor school setting and the number observed who were wearing wide-brimmed and other hats. Research assistants used a handheld counter to facilitate measurement. To assess reliability of measurement, all four research assistants conducted measurements simultaneously with the project director. There was perfect agreement between measures taken by the project director and all four research assistants in regard to observed hat use.
Assessment of hat use at schools was based on the principles of valid measurement of sun protection behavior, as described by Milne et al. (41
). Research assistants used listings of outdoor scheduled activities to schedule their observation sessions to measure hat use among fourth-grade students from each class and school during each of the three data collection periods. For each school, the data collection sessions were at least 8 weeks apart. Research assistants chose inconspicuous vantage points to ensure that student and teacher behaviors were not influenced by the measurement process. Research assistants recorded the following data at each observation session: 1) the total number of students observed, 2) the number of students observed wearing any type of hat, and 3) the number of students wearing wide-brimmed hats. In addition, ambient conditions during the outdoor activity (time of day, temperature, cloud cover, wind speed, and amount of shading) were assessed by each researcher as described previously (21
We hypothesized that use of hats at school would result in less pronounced sun-induced physiological changes in children's skin. In an exploratory fashion, we conducted an interim analysis of the impact of the intervention on skin pigmentation among a convenience sample of students who agreed to undergo additional measurements (skin pigment assessment and nevi counts) for a 2-year period. We recruited a minimum of 15 students at each school by asking for volunteers (no incentives were used) who were willing to participate in this portion of the study, and a total of 378 students (178 in the intervention group and 200 in the control group) agreed to participate, with 280 (74.1%) completing all three measurements in the first year. For each student, measurements of skin pigmentation were assessed at baseline (August 8, 2006, through September 29, 2006), midyear (November 30, 2006, through March 15, 2007), and again in spring (March 27, 2007, through May 22, 2007). We compared changes in each subject's skin pigmentation over the course of the study, so that each subject served as his or her own control.
We fitted the generalized linear mixed models (to account for clustered data) separately for each intervention group to determine whether or not demographic characteristics differed among students who participated in this part of the study compared with those who did not. Students who volunteered were found to be similar to those who did not volunteer with regard to age (P = .5 for both the intervention and control groups), sex (P = .9 for the intervention group and P = .1 for the control group), and race (P = .6 for the intervention group and P = .98 for the control group).
All measures of skin pigmentation were made with a DermaSpectrometer (Cortex Technology, Hadsund, Denmark). The DermaSpectrometer measures light reflectance at two narrow wavelength bands corresponding to skin erythema (568 nm) and melanin content (655 nm). The DermaSpectrometer then calculates an erythema and melanin index, each ranging from 0% to 100%. DermaSpectrometers have been compared with colorimeters and both provide similar estimates of skin melanin pigmentation (44
We assessed skin pigmentation at a child's forehead as a representative anatomical site that would be protected by hat use. We initially used a pilot test to measure skin pigmentation on 49 fourth-grade students. In the pilot study, we first assessed skin pigmentation at the center of the forehead (just above the glabella) and to the right and left of center. Measures at these three sites were highly correlated (correlation coefficients center-right = .96, center-left = .97, and left-right = .95). Consequently, we selected the center forehead reading as the single standard measurement site. We next confirmed the test–retest reliability of skin pigment measurement by repeating measurements 1 week apart and calculating the interreliability coefficient (correlation coefficient = .86), as described previously by Shrout and Fleiss (46
). We also confirmed the interrater reliability of our methods by comparing measurements taken simultaneously by the project director and research assistants (correlation coefficient = .98).
To determine if participating schools were representative of all Hillsborough County schools, we compared characteristics of schools that participated in the study with schools that did not participate in the study. To ensure comparability of intervention and control arms, we compared baseline characteristics between the intervention schools and control schools. Comparisons were made by use of the Fisher exact test for categorical data, the Wilcoxon rank sum test for continuous data at the school level, and the general or generalized linear mixed model for comparisons at the student level by taking into account the cluster randomized trial design. The rate of hat use and its 95% confidence interval were estimated for each intervention and assessment period by use of the exact binomial distribution.
We examined two endpoints regarding hat use: directly observed hat use at schools and self-reported hat use outside of schools. Both outcomes were measured in each of the following three periods: baseline (late August to early September), late fall to early winter, and early spring. The clustered design in which schools rather than students were randomly assigned to an intervention was accounted for in the generalized linear mixed model. Allocation arm (intervention vs control), date of data collection (represented by a linear or quadratic term), and their interaction were included as fixed effects in the generalized linear mixed model. Multiple random effects were also included in the model: the school-specific mean intercepts and slopes (schools were nested within the intervention arm) with random variation among the students.
The random effects were assumed to be independent and distributed as normal distributions. For self-reported hat use, an additional random effect was included to allow for correlation among repeated observations taken on the same student over data collection rounds, with compound symmetry structure. Several fixed covariates were tested in multivariable models as confounding factors (eg, age, sex, race, and school uniform policy). After checking the distribution of the normality assumption, the pigmentation changes were analyzed by use of a generalized linear mixed model for a subset of the students. In addition, pigmentation changes were analyzed by subgroups defined a priori (lighter-skinned vs darker-skinned children).
The intraclass correlation coefficient (ICC) was estimated by use of the covariance parameters that were estimated from the generalized linear mixed models. For binary data, the estimates were unscaled by use of the average value of the response variable because the estimates of variance were in the logit scale. The statistical analysis was conducted with SAS version 9.1 (SAS Inc, Cary, NC). We used the MIXED procedure for normally distributed continuous data and the GLIMMIX procedure with the logit link function for binary data. To adjust the small number of degrees of freedom in mixed models, the method of Kenward and Roger (47
) was used. All P
values were from two-sided statistical tests and were claimed as being statistically significant at a level of .05.
The required sample size was determined with a method that takes into account the intraclass correlation coefficient, the expected effect size, and the power of the study. We assumed an intraclass correlation coefficient of .02 and a minimum difference of wearing a hat outdoors (self-report) between the two arms of 10% (control arm = 10% and intervention arm = 20%). Under these assumptions, we anticipated a power of 80% to detect the difference of 10% between the two arms at a statistical significance level of .025 with 10 schools (the average number of fourth-grade students per school was assumed to be 50) for each intervention arm, so that the total included 1000 students. Although the sample justification was based on the self-reported hat use outside of school, we expected that the statistical power for the observed hat use would exceed 80% power, given the resultant sample size.