Evaluation of the Be the Exception Sixth-Grade Program in Rural Communities to Delay the Onset of Sexual Behavior

Abstract

Objectives. To investigate the impact of Be the Exception, a newly developed program to delay onset of sexual behaviors, in a White, rural population.

Methods. A cluster randomized controlled trial in northwestern Indiana (14 schools, 1776 students, 2011–2015) compared an intervention (5 group sessions and multimedia assembly) with a no-intervention group; both continued usual standard health education. Multilevel mixed-effects logistic regressions with 1455 students measured long-term outcomes 12 months after baseline questionnaire.

Results. Intervention group students reported ever having had sexual intercourse and sexual intercourse in past 3 months significantly less often than did the comparison group (1.91% vs 6.29% and 1.09% vs 4.26%, respectively). No statistical differences were observed for reported sexual intercourse in past 3 months with risky behavior (1.23% vs 2.24%), without condom use (1.04% vs 1.73%), or without birth control (1.00% vs 1.53%). Cumulatively, intervention group students significantly more often reported no activity, holding hands, hugging and kissing and less often reported touching above and below the waist, other sex, or sexual intercourse.

Conclusions. Be the Exception is effective in delaying the onset of sexual behavior among rural middle school students.

Adolescent parenthood is more widespread in the United States than in any other developed country, and it is most common in the rural United States.^1,2 Adolescent birthrates in rural counties are nearly one third higher than those in the rest of the country.² Although the birthrate among adolescents was cut in half in major urban centers and by 40% in suburban counties between 1990 and 2010, the decline in rural counties was only 31%.^2,3 Youth Risk Behavior Survey middle school data have shown that 5% to 20% of sixth graders and 14% to 42% of eighth graders have engaged in sexual intercourse.⁴ Onset of first sexual intercourse in middle school is a public health problem for additional reasons. Research has shown that early onset of sexual behavior is associated with risky health behaviors such as tobacco, alcohol, and other drug use; peer violence; and poor school performance.^5–8

In rural communities that want to implement effective adolescent pregnancy programs in middle school, a limited number of evidence-based programs have demonstrated that they slow the onset of early adolescent sexual behavior.^9–11 Several middle school programs have demonstrated reduction of sexual behavior among young adolescents.^5,9–11 However, community, culture, and student characteristics, such as region, religious beliefs, or race and ethnicity, can moderate and reduce intended positive program impacts.⁹ The evidence for programs that demonstrate a positive impact on reduction of sexual behavior in early adolescence in rural communities with a largely White population is limited.^10,11 Furthermore, nationally representative data from 2006 to 2013 have shown significant declines in receipt of formal school-based sex education instruction, with more rapid declines among adolescents living in nonmetropolitan or rural areas.¹²

We present results from a clustered randomized controlled trial of Be the Exception, a program implemented in grade 6 (Z. H. Piotrowski and M. A. Lee, unpublished data, 2016).¹³ We investigated 4 outcomes: (1) ever having had sex, (2) having had sex in the past 3 months, (3) having engaged in risky behavior (such as alcohol use or not using a condom) while having sex, and (4) having engaged in physical intimacy behaviors ranging along a continuum from presexual behavior such as holding hands and hugging and kissing to sexual behavior such as touching above or below the waist and sexual intercourse. We developed a physical intimacy behavior hierarchical scale as a program impact measure geared to younger adolescents.

METHODS

The Be the Exception program is a developmentally appropriate holistic intervention for adolescents (Z. H. Piotrowski and M. A. Lee, unpublished data, 2016).¹³ It is based on theory of possible selves and social learning–norming theory^14,15 (Figure A, available as a supplement to the online version of this article at http://www.ajph.org). The program supplements the standard health and physical education curriculum that students already receive and was developed for middle school students in predominantly rural communities. It focuses on promoting attitudes, skills, and behaviors that support positive youth development. The emphasis is on a child’s positive possible self and future self, school performance, and goal orientation. Topics include risk behaviors such as alcohol, drugs, and peer violence. Program elements include activities that promote health, student readings, group exercises, demonstrations, and take-home assignments to be worked on with parents.

The program was delivered over 5 classroom sessions of 45 to 50 minutes on consecutive days and 1 multimedia class assembly in May. May was selected as an appropriate time to reinforce student learning from the 5 sessions before the beginning of summer. Instruction was provided by male–female facilitator pairs, specially trained by Positive Approach to Teen Health, who used portable tablets with scripted content. Be the Exception is the first of 3 independent components of the Positive Approach to Teen Health Positive Potential program for grades 6 through 8 (Z. H. Piotrowski and M. A. Lee, unpublished data, 2016).¹³ Analysis of Be the Exception students who went on to participate in programs in grades 7 and 8 was not part of the present study.

Comparison group students attended an unrelated assembly with a speaker on topics not covered in the Be the Exception program (obesity, healthy eating, suicide prevention). Both comparison and intervention students continued participation in standard health education. Be the Exception supplemented health education instruction.

Setting and Recruitment

Twenty-nine schools in 5 northwestern Indiana counties met recruitment eligibility criteria: public school with sixth grade in a nonurban, nonsuburban rural community almost always surrounded by farms. Charter, specialized academy, and special education schools were excluded. Sixteen schools were enrolled for 5 months from 2011 through 2012. Cohort 1 students were enrolled from October 2011 to April 2012; cohort 2, from October 2012 to December 2012. Before randomization to the 5-year study, discussions with school administration and teachers resulted in exclusion of 13 schools because they were unable to ensure compliance with randomized assignment to groups and scheduling of 18 sessions and 7 survey administrations (Figure B, available as a supplement to the online version of this article at http://www.ajph.org). School randomization was performed in pairs of similarly sized sixth grades to reduce a possible imbalance in group size and an unequal number of clusters. School-level attrition was 12.5%. Two of 16 schools in the same pair-block dropped out within 4 months after randomization: 1 because of a change in administration and 1 because of inability to comply with planned class schedules. Students were told of the school’s random assignment status (intervention or control) after consent–assent forms were collected. Parents or students in cohort 2 may have been aware of the school’s intervention or comparison status as a result of cohort 1’s group assignment. This concern is addressed below with equivalence analysis of two cohorts.

Of 2931 recruited students, 1776 consented and assented (60.6%), 1747 completed the baseline questionnaire, and 1622 completed the 12-month follow-up questionnaires. There were 37 to 234 (median = 105) students per school and 4 to 52 (median = 16) students per classroom. The analytic sample consisted of 1455 students with no missing data on any study item; 167 were excluded for the following reasons: 72, missing baseline; 40, sexual intercourse; 40, physical intimacy behavior; and 15, both sexual intercourse and physical intimacy behavior.

Design and Measures

The evaluation was a cluster randomized controlled trial, with 50–50 group probability assignment by the independent evaluator using a table of random numbers. The evaluation team collected the baseline and follow-up data. Fidelity data are fully reported elsewhere (Z. H. Piotrowski and M. A. Lee, unpublished data, 2016).

The baseline questionnaire’s 2-category ethnicity and 6-category race items were combined because of the small sample size and nonestimable regression analyses. For ethnicity, 13.8% responded yes to the question “Are you Hispanic or Latino?” Responses to the question “What is your race?” were as follows: American Indian or Alaska Native, 0.21%; Asian, 0.27%; Black or African American, 2.27%; Native Hawaiian or other Pacific Islander, 0.07%; White, 96.84%; and other, 0.34%. Covariates included gender, age (mean centered to 0, age in years minus 11), and race/ethnicity (3 binary dummy-coded variables: Hispanic = 1, other = 0; non-Hispanic non-White = 1, other = 0; and non-Hispanic White = 1, other = 0 was the reference category). Students reported parents’ belief about sex before marriage as “OK,” “don’t know,” or “should not have sex” (Table A, available as a supplement to the online version of this article at http://www.ajph.org). Dichotomous items were converted to composite measures with a mean range of 0–1.0 for the purpose of analysis: sexual risk avoidance knowledge; ever suspended or expelled; and nonsexual risk behaviors—substance use (alcohol, tobacco, marijuana, other drugs), fighting (physical fight, hurt someone in physical fight), and bullying (cyber, physical). Demographic and baseline characteristics were predictors in all regression analyses (Table 1). They were selected a priori to affirm prior research on associations among sexual behavior and other risk factors (Z. H. P., unpublished data, 2011).^4,8,10,16

TABLE 1—

Baseline Measures of Students in Comparison and Intervention Groups and Analysis of Equivalence: Be the Exception Program, Northwestern Indiana, 2011–2012

Baseline Measures	Comparison Raw Mean (SD) or Percentage	Intervention Raw Mean (SD) or Percentage	Difference (Probability)
Demographic characteristics
Gender: male, %	48.3	48.0	0.3 (.925)
Race/ethnicity, %
Hispanic	13.8	13.8	0
Non-Hispanic non-White	1.9	3.1	−1.2 (.198)
Non-Hispanic White	84.5	83.4	1.1 (.667)
Age, y
At baseline	12.018 (0.437)	11.985 (0.447)	0.033 (.171)
At 12-mo follow-up	12.915 (0.424)	12.917 (0.446)	−0.002 (.815)
Belief and knowledge
Student opinion about parents’ belief about sex before marriage, %
My parents believe it is OK to have sex before marriage	8.5	6.8	1.7 (.171)
I really DON’T KNOW what my parents think	39.6	45.5	−5.9 (.206)
My parents believe I should NOT have sex before marriage	51.9	47.7	4.2 (.451)
Sexual risk avoidance knowledge	0.802 (0.225)	0.793 (0.224)	0.009
Nonsexual risk behaviors, %
Suspended or expelled	9.7	14.4	−4.7 (.002)
Substance use	9.8	8.7	1.1 (.823)
Fighting	32.0	38.0	−6.0 (.004)
Bullying	16.8	16.8	0
Study design measures
Survey time in mo (mean centered to 0 = starting month of first school group survey in cohort)
At baseline	2.414 (1.853)	1.940 (1.543)	0.474 (.005)
At 12-mo follow-up	13.190 (1.370)	13.130 (1.459)	0.060 (.133)
Analytic sample size	671	784

Note. Raw or observed means with standard deviations and percentages are reported. Equivalence was examined by regressing each baseline student covariate on the intervention indicator variable, a series of school pair indicators, and 12-mo postbaseline survey time and cohort indicators, while clustering standard errors at the school level and classroom level for the full analytic sample of 1455 students. P values were adjusted for clustering standard errors. Table A (available as a supplement to the online version of this article at http://www.ajph.org) presents the response option and derivation of composite scale score of baseline measures that were used for regression analyses. The percentage of students who responded yes to any item within the composite scale is reported for the 4 nonsexual risk behaviors. The percentage is reported to describe the occurrence of nonsexual risk behavior among the students instead of the mean and standard deviation as used in the regression analysis. Group-specific percentages are presented. The overall description of the analytic sample is as follows: about 7.6% reported their parents believe sex before marriage is OK, and 42.8% reported they really did not know what their parents believe about sex before marriage. Knowledge about avoiding sexual risk behaviors appeared to be relatively high, with a mean of 79.7%. Nonsexual risk behaviors ever engaged in were as follows: 12.2% suspended or expelled or both; 9.2% substance use (any 1 of the following: alcohol, 5.9%; tobacco, 5.9%; marijuana, 2.1%; other drugs, 0.8%); 16.8% physical or cyberbullying or both; and 35.0% physical fight (physical fight or hurt someone in a fight).

Students responded yes or no to sexual intercourse behavior items 12 months after baseline: ever had sexual intercourse, sexual intercourse in past 3 months, and sexual intercourse in past 3 months with risky behavior (such as alcohol or drug use during sexual intercourse) and sexual intercourse in past 3 months without a condom even once and without effective birth control even once (see the box on page S135).

Outcome Measures at 12-Month Follow-Up: Be the Exception Program, Northwestern Indiana, 2011–2012

Sexual Behavior and Physical Intimacy Outcome Measures	Response Option and Derivation of Score for Analysis
1. Ever had sexual intercourse Sexual intercourse is defined as making love or going all the way. By sexual intercourse we mean a male putting his penis into a female’s vagina. Have you ever had sexual intercourse? If no, skip to question xx on the next page.	Dichotomous item: yes or no If response was missing and ever sexual intercourse in past 3 mo item 2 was a yes, a code of 1 for yes was assigned to item 1.
2. Sexual intercourse in past 3 mo Now please think about the past 3 mo. In the past 3 mo, have you had sexual intercourse, even once? If no, skip to question xx on the next page.	Dichotomous item: yes or no If response was missing and ever sexual intercourse item 1 was not missing, a code of 0 for no was assigned to item 2.
3. Sexual intercourse in past 3 mo with risky behavior a. In the past 3 mo, have you or your partner had sexual intercourse without using a condom, even once? b. In the past 3 mo, have you or your partner had sexual intercourse without using an effective method of birth control, even once? c. Did you or your partner drink alcohol the last time you had sexual intercourse? d. Did you or your partner use drugs the last time you had sexual intercourse?	Dichotomous item: yes or no Composite score was computed from responses to 4 items, a–d. Any yes response was coded 1, indicating risky behavior. No response was coded 0. If all 4 items were missing and the “ever sexual intercourse” item was not missing, a code of 0 was assigned. The a priori composite score was formed because the prevalence of each single risky behavior indicator was expected to be very low with seventh graders and not sufficient to test the difference between groups. The observed percentages of reporting “yes” for items a–d among 1455 students were as follows: 1.51%, 1.24%, 0.62%, and 0.34%. Impact analyses of noncondom and non–birth control use, items a and b, were performed and are seen in Table B, available as a supplement to the online version of this article at http://www.ajph.org.
4. Physical intimacy behavior (a-d items, 0–8 9-stage scale) a. In the PAST, how far have you gone with a boy or girl? (Check one.)  0. No activity  1. Holding hands  2. Kissing and hugging  3. Touching above the waist  4. Touching below the waist  5. Other sex b. 6. Ever sexual intercourse, yes to measure 1 above. c. 7. Sexual intercourse in past 3 mo, yes to measure 2. d. 8. Sexual intercourse in past 3 mo with risky behavior, yes to measure 3.	Stage score, 0–8 The assignment of stage of physical intimacy behavior was based on 4 questionnaire items (a–d). A student was placed into the highest numerical stage on the ordered stage scale 0–8 as follows: checked response to item a, corresponding to stages 0–5; yes response to item b, stage 6; yes response to item c, stage 7; or yes response to item d, stage 8. For example, if a student checked item a, “other sex,” indicating stage 5 and also responded yes to items b, c, and d, the student was assigned to stage 8.

Note. Sexual behavior and physical intimacy behavior responses were collected for the first time at 12-mo follow-up in grade 7 because not all schools provided permission for grade 6 questionnaire. The response option “other sex” in item 4a was not defined because not all schools provided permission to include oral or anal sex behaviors in the questionnaire. Sexual behavior measures 1–3 are from similar items in adolescent pregnancy prevention research.^9–11 Item 4a was adapted from prior research (Z.H.P., unpublished data, 2011).^6,7,17 The 9 stages were collapsed into 6 ordinal stages (0, 1, 2, 3–5, 6, 7–8) to examine intervention versus comparison group differences. The reduction to 6 cumulative ordinal stages was based on logical groupings about risk of pregnancy and sexually transmitted infection and the size of sample at some stages. The reduced 6-stage physical intimacy scale starts with no sexual behavior at stage 0 (same as stage 0 in the 9-stage index) and continues to the contemplative stages 1 and 2 (1 and 2 in the 9-stage index); precursors to sexual intercourse, stage 3 (3–5 in the 9-stage index); ever having had sexual intercourse, stage 4 (6 in the 9-stage index); and sexual intercourse in the past 3 mo, stage 5 (7-8 in the 9-stage index). The ordinal scale resembles a Guttman hierarchical scale in which the participant provides a response to each dichotomous item, for example yes or no to the “how far have you gone” question. We were not able to perform scalogram cumulative analyses because the scale was constructed in a different manner. However, Cronbach α reliability with 4 items was .87. Analysis and use of the scale with grades 7–9 are available from the first author. There were some inconsistent responses about sexual intercourse. A student was logically imputed as “yes” if he or she had a missing “ever had sexual intercourse” item and a “yes” response to “sexual intercourse in past 3 months.” There were 35 students among the 1455 students who provided inconsistent responses: check to option 0 (indicating no sexual intercourse) and yes to item b, ever sexual intercourse. The assignment code was logically imputed to yes, corresponding to stage 6, ever had sexual intercourse. A sensitivity analyses performed with the reduced 1420 student sample yielded similar results (Table C, available as a supplement to the online version of this article at http://www.ajph.org). We also present the mixed-effects regression analysis for the full analytic sample with the 9-stage physical intimacy behavior scale (Table D, available as a supplement to the online version of this article at http://www.ajph.org).

A multicategory 9-stage hierarchical index of physical intimacy behaviors was developed from 4 questionnaire items (Z. H. P., unpublished data, 2011).^{6,7,14,16–18} The ordinal 9 stages of behavior fell on a continuum ranging from no activity to sexual intercourse in the past 3 months with risky behavior. A student was placed into 1 of 9 stages. The 9-stage hierarchy was reduced to a 6-stage hierarchy for analysis.

As Hennessy¹⁷ has noted, a position on an ordinal scale provides an estimate of which behaviors the student has engaged in, which behaviors he or she is currently engaged in, and which behaviors he or she is not currently engaged in. In this study, the continuum was a hierarchical cumulative scale with increasing physical intimacy behaviors. For example, a student at stage 2, hugging and kissing, has passed through the previous stages of no intimacy (stage 0) and holding hands (stage 1). The assumption is that the student has engaged in the behaviors in cumulative stages 0 to 2, is at a threshold, and has not engaged in any of behaviors in cumulative stages 3 to 5, pre–sexual intercourse (touching above and below the waist, other sex [stage 3]) and sexual intercourse (stages 4 and 5).¹⁸

Statistical Methods

Baseline equivalence between intervention and comparison groups was examined by regressing each baseline characteristic on the intervention indicator variable, a series of school pair indicators, a cohort indicator, and survey time at 12-month follow-up, while clustering standard errors at the school and classroom levels. Because parents or students in cohort 2 may have been aware of school assignment and on that basis elected to consent or not consent, additional cohort 1 versus 2 equivalence analyses were performed. Multiple methods with descriptive statistics were used for the implementation analysis that included tracking attendance, student and teacher ratings, and educator team self-assessments.

Three-level hierarchical mixed-effects logistic regression models were used for the 3 sexual intercourse behavior outcome measures.^19–21 We used the Bonferroni multiple-comparison method to adjust the P-value threshold to .017 to determine statistical significance for the 3 measures of sexual intercourse behavior.²¹ Elsewhere, a P value of less than .05 was considered statistically significant. Adjusted odds ratios (AORs) with 95% confidence intervals were calculated.

Mixed-effects proportional odds and nonproportional odds logistic regression models were compared.^18,22,23 The proportional odds model assumes a common odds ratio (OR) for intervention and comparison groups on each threshold. The nonproportional odds model allows statistically significant ORs, indicating interaction between intervention and comparison groups and 1 or more thresholds.

We analyzed intervention and comparison group cumulative probabilities up to significant thresholds, making the range of hierarchy of stages binary at that threshold. A statistically significant OR at a threshold separates the intervention and comparison group cumulative probability of behaviors. The OR indicates the direction and magnitude of the difference between the 2 groups at the particular threshold. With an appropriate analysis approach and software,^18,22,23 we examined the effect of the Be the Exception program on 6 ordered stages and 5 (6 − 1) thresholds.

All outcome regression analyses included the same random school and classroom effects, fixed design covariates, and fixed baseline student characteristic predictors. Classrooms and schools were included as random design indicators to account for any characteristics that may have been specific to either class or school. The only intraclass correlation above .001 was for classrooms for the ever sexual intercourse measure (.01). The fixed design effects were blocked school pairs (n = 6; the seventh was the reference); 2 cohorts coded 0 or 1; and survey time at 12-month follow-up (months since first baseline survey) treated as a continuous measure to track any substantive differences between group and cohort survey administration times.

RESULTS

Analytic sample demographics at baseline for intervention and comparison groups combined were as follows: 48.1% male; 83.9% non-Hispanic White, 13.8% Hispanic, and 2.3% non-Hispanic non-White; and a mean age of 12 years. Implementation data demonstrated high fidelity: Of the students, 94.3% attended at least 80.4% of the sessions (school median = 92.0%; range = 91.3%–100%); more than 85.0% attended the booster assembly; 94.9% of planned activities were implemented; teacher positive ratings of instruction ranged from 83.6% to 97.8%; and students strongly agreed or agreed more than 84.0% of the time with positive indicators of educators, instruction, and activities and positive behavior intentions and attitude changes (Z. H. Piotrowski and M. A. Lee, unpublished data, 2016). Survey item results obtained after the grade 8 program from student reflections on health and physical education instruction in grades 6–7 indicated modest group differences on instruction regarding sexually transmitted infections or HIV and abstinence from sex, and no group difference was found in condom and birth control instruction (results available from Z. H. P.). No demographic differences were found between the intervention and comparison groups. Intervention group students reported a higher mean occurrence of being suspended or expelled and fighting. These differences were controlled for in all outcome regression analyses.

Comparison group baseline administration survey time was a negligible 2 to 3 weeks later in the school year, largely because scheduling occurred after scheduling of the paired intervention school survey administration and instruction dates was completed. Equivalence analyses indicated minor statistically significant baseline differences between cohorts 1 and 2 (Table E, available as a supplement to the online version of this article at http://www.ajph.org). Compared with cohort 1, cohort 2 students reported different parent beliefs about sex before marriage. Within cohort 2, the intervention group had a higher mean occurrence of suspension or being expelled and fighting. Because cohort 1 recruitment was staggered over 5 months and occurred later in the school year, cohort 2 students were 2 months younger and completed the baseline and 12-month follow-up questionnaires 7 weeks earlier in the school year. Regression analyses have taken these differences into account in estimates of outcomes.

The Be the Exception program had a positive impact on sexual intercourse behavior outcomes at the 12-month follow-up (Table 2). (Table F, available as a supplement to the online version of this article at http://www.ajph.org, provides regression results with student baseline predictors.) In the intervention group, 1.9% reported ever having had sexual intercourse; 6.3% of the comparison group did. The comparison group was 3 times more likely to report “Ever had sexual intercourse” (AOR = 3.18; P = .001). The prevalence of those reporting “sexual intercourse in the past 3 months” was 1.1% in the intervention group and 4.3% in the comparison group. The comparison group was almost 4 times as likely to report “Sexual intercourse in the past 3 months” (AOR = 3.82; P = .002). The prevalence of sexual intercourse in the past 3 months with risky behavior was higher in the comparison group than in the intervention group, but the difference was not statistically significant (2.2% vs 1.2%; P = .238).

TABLE 2—

Analysis of Be the Exception Program Impact on Sexual Intercourse Behaviors: Northwestern Indiana, 2011–2012

Comparison Group and Intervention Group	Ever Had Sexual Intercourse	Sexual Intercourse in Past 3 Mo	Sexual Intercourse in Past 3 Mo With Risky Behavior
Comparison group: standard instruction, mean %	6.29	4.26	2.24
Intervention group: Be the Exception, mean %	1.91	1.09	1.23
Difference: Comparison group − intervention group, mean %	4.38	3.17	1.01
Comparison group = 1, intervention group = 0, AOR (95% CI)	3.18 (1.66, 6.08)	3.82 (1.68, 8.68)	1.73 (0.70, 4.27)

Note. AOR = adjusted odds ratio; CI = confidence interval. The comparison and intervention group mean percentages are estimated values and were adjusted to covariates in the mixed effects models. All 3 outcome analyses were adjusted as follows: baseline demographics, gender, ethnicity, race, and age; baseline belief and knowledge measures; suspension or expelled from school, and baseline nonsexual risk behavior measures. The survey time at 12-mo follow-up, cohort 1 vs 2, and 6 school pair blocks were included in the analyses. The error terms were adjusted for nonindependence with school and classroom random effects. All analyses were performed with the full analytic sample of 1455 students who completed the 12-mo follow-up surveys. Supermix software^18,21 was used for the mixed-effects 3-level (student, classroom, school) logistic regression analysis models. Bonferroni correction was performed on the 3 impact measures. Table F (available as a supplement to the online version of this article at http://www.ajph.org) presents the analysis of program impact on sexual intercourse behaviors with student baseline predictor coefficients.

Be the Exception had similar positive impacts on the 3 sexual intercourse behaviors without inclusion and adjustment of student baseline covariate characteristics (Table G, available as a supplement to the online version of this article at http://www.ajph.org). AORs were statistically significant regardless of whether baseline demographic characteristics, belief and knowledge, or nonsexual risk behaviors were included in the regression models.

We also examined the impact of Be the Exception on 2 common outcome measures in adolescent pregnancy prevention programs¹¹: sexual intercourse without using a condom and sexual intercourse without effective birth control. The Be the Exception students reported sexual intercourse in the past 3 months without condom use and without effective birth control, use but the differences were not statistically significant (Table B, available as a supplement to the online version of this article at http://www.ajph.org).

We considered whether Be the Exception had an impact on any of the 5 cumulative thresholds (Table 3, Table H [available as a supplement to the online version of this article at http://www.ajph.org]). The intervention group significantly more often reported being in the no activity, holding hands, and hugging and kissing stages (cumulative logit 3 for stages 0, 1, and 2) than in the presexual intimacy behaviors, touching above and below the waist, other sex and sexual intercourse stages (stages 3, 4, and 5) than the control group (OR = 0.57; P < .003). In addition, the intervention group was less likely to have ever engaged in sexual intercourse or to have engaged in sexual intercourse in the past 3 months (threshold at cumulative logit 4, stages 4–5, OR = 0.35; P ≤ .001). The largest positive impact of Be the Exception was indicated by the lowest AOR of 0.29 (P < .001; threshold at cumulative logit 5, stages 0–4 vs 5, occurrence of sexual intercourse in the past 3 months). Overall, intervention group students were about half as likely to report presexual and sexual physical intimacy behaviors than control group students. Also, the nonproportional odds logistic regression of the physical intimacy behavior 6-stage continuum yielded similar threshold AOR differences (1) with and without demographic or baseline characteristic and (2) within cohort 1 and cohort 2 (results available from Z. H. P.).

TABLE 3—

Analysis of Program Impact on the 6-Stage Physical Intimacy Behavior Index: Be the Exception, Northwestern Indiana, 2011–2012

Physical Intimacy Behavior Stage	Control Group Observed %	Intervention Group Observed %	Interactions of Intervention and Control Groups With Cumulative Logits	Estimated Logit	AOR (95% CI)
0: No physical intimacy	39.79	39.03
1: Holding hands	16.10	18.24	Cumulative logit 1 (5, 4, 3, 2, 1 vs 0)	0.02	1.02 (0.80, 1.31)
2: Hugging and kissing	29.66	32.53	Cumulative logit 2 (5, 4, 3, 2 vs 0,1)	−0.09	0.91 (0.71, 1.16)
3: Touching above or below the waist, or other sex	7.90	7.14	Cumulative logit 3 (5, 4, 3 vs 0, 1, 2)	−0.55	0.57 (0.39, 0.82)
4: Ever engaged in sexual intercourse	2.09	1.40	Cumulative logit 4 (5, 4 vs 0, 1, 2, 3)	−1.03	0.35 (0.20, 0.51)
5: Sexual intercourse in past 3 mo	4.47	1.66	Cumulative logit 5 (5 vs 0, 1, 2, 3, 4)	−1.22	0.29 (0.14, 0.59)

Note. AOR = adjusted odds ratio; CI = confidence interval. A proportional ordinal logistic regression and a nonproportional ordinal logistic regression were performed with intervention (n = 671) and control (n = 784) groups. Their deviance scores were compared to formally test the proportional odds assumption of identical AORs among the 5 threshold or dichotomizations. The proportional model was rejected in favor of the nonproportional model based on a statistically significant χ² of 15.82, 4 degrees of freedom (deviance score of 3721.42 with 27 parameters for the proportional odds regression and deviance score of 3705.60 with 31 parameters for the nonproportional odds regression). All analyses were performed with the full analytic sample of 1455 students who completed the 12-mo follow-up surveys. Supermix software^18,21 was used for the mixed-effects 3-level (student, classroom, school) logistic regression analysis models. Table H (available as a supplement to the online version of this article at http://www.ajph.org) presents the 6-stage analysis baseline predictor coefficients for the 6-stage analysis. Table D (available as a supplement to the online version of this article at http://www.ajph.org) presents the 9-stage analysis with the baseline student predictor coefficients.

Were student demographics and baseline characteristics associated with the occurrence of sexual intercourse and physical intimacy behavior at the 12-month follow-up in grade 7? Most demographic characteristics did not appear to be strongly associated with sexual behaviors. However, there were strong, statistically significant associations between several baseline characteristics and the 3 sexual intercourse measures and the physical intimacy behavior hierarchy thresholds (Tables D, F, and H, available as a supplement to the online version of this article at http://www.ajph.org). Sensitivity analysis results with the physical intimacy behavior scale with the reduced sample were essentially the same as those for the full analytic sample (Table 3, as well as Tables C and H).

DISCUSSION

Be the Exception had a strong positive impact on young adolescents: lower occurrence of ever having engaged in sexual intercourse, lower occurrence of sexual intercourse in the past 3 months, and lower occurrence of presexual physical intimacy behaviors. Be the Exception program students more often remained at the no activity, holding hands, and hugging and kissing cumulative stages threshold and were less often at the touching above and below the waist, other sex, and sexual intercourse cumulative stages. Although adolescents in the Be the Exception program, relative to the comparison group, reported a lower percentage of sexual intercourse in past 3 months (1) with risky behavior, (2) without using a condom, or (3) without using an effective method of birth control, the group differences were not statistically significant.

The physical intimacy continuum, with 6 stages ranging from no activity to presexual behavior to sexual intercourse behavior, may be especially applicable in assessment and evaluation programs with adolescents in elementary school and middle school. The hierarchy can be applied as a process or implementation measure or as a measure of longitudinal program impact. The student can be observed to change and transition to numerically higher stages of physical intimacy behavior (see Coyle et al.’s²⁴ use of ordinal psychosocial measures associated with the occurrence of sexual intercourse).

As others have suggested,^{6,7,17,18,20,23} this type of measure is appropriate for evaluation of intervention outcomes for either abstinence-only or comprehensive sex education programs when communities may decide it is developmentally appropriate to slow the progression of presexual or sexual behaviors. When used with caution, the measure can be informative in defining what is normative for elementary or middle school youths in a particular community.

Although the findings of this study are promising, it has some limitations. Thirteen of 29 schools did not participate. Within 4 months after randomization, 2 schools dropped out (because of a change in school administration and inability to fit the intervention into the school schedule). Also, 61% of students consented or assented to participate. Although this was an efficacy trial, these events reduced the sample size and may have limited the findings.

We were not able to follow up with students who did not complete the baseline or 12-month follow-up surveys (n = 72). Additional students did not respond to all sexual intercourse or sexual behavior questions (n = 95). We did not have a baseline measure of sexual behavior to include as a control. There was a potential for bias because the cohort 2 parents may possibly have learned of random assignment before consent. Some cohort and baseline group differences were noted, although their effect on results was limited with regression-type analyses. Self-report of sexual intercourse behavior and other risk behaviors as well as the choice to skip those items may have been subject to cognitive and social biases. Generalizability to other rural populations is limited, particularly with different baseline risk profiles. The small sample of sexually active students further limited examination of the program’s impact among student subgroups.

This study has several strengths. Use of a randomized controlled trial design with multilevel mixed-effects regression analyses produced more accurate estimates with the adjusted standard errors. Baseline behavior and outcome behavior items were adapted from prior research, thereby increasing the validity of student responses. Analyses indicated minimal nonequivalence between groups and between groups and cohorts. Positive effects of Be the Exception were maintained with and without demographic and baseline covariate adjustments.

Middle school evidence-based sexual behavior prevention programs can have a positive short-term and long-term effect on adolescent and young adult health risk behaviors and ultimately improve quality of life.^10,11,24,25 The Be the Exception program is a promising beginning intervention in middle school that can address the limited number of rural community programs.^10,11 This research affirms the positive impact of the Be the Exception program on middle school adolescents with the benefit of health risk reduction to delay student transition to physical intimacy sexual behavior stages and the benefit of health risk avoidance to delay the onset of sexual intercourse behaviors. This study’s findings suggest that the Be the Exception program can be particularly effective with sixth-grade students in predominantly White rural communities.