Copyright 1999 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.1999
African American adolescents living in high-poverty urban settings are at increased risk for early sexual initiation and sexually transmitted diseases.
To determine whether parental stategies to monitor their children's social behavior and to communicate with them about sexual risks help to reduce the initiation of risky sexual behavior and prevent the resulting adverse health outcomes.
To assess the viability of these strategies, we surveyed a stratified cross-section of African American children aged 9 to 17 years (N=355) living in urban public housing. Talking computers were used to increase the confidentiality and comparability of the interviews across the wide age range.
Children who reported high levels of parental monitoring were less likely to report initiating sex in preadolescence (aged ≤10 years) and reported lower rates of sexual initiation as they aged. Children who reported receiving both greater monitoring and communication concerning sexual risks were also less likely to have engaged in anal sex. Communication was also positively related to the initiation of condom use and consistent condom use. The protective correlates of these parenting strategies were independent of the type of guardian (mother vs other family member).
Interventions with parents and other guardians to increase monitoring and communication about sexual risks seem to be promising health-promotion strategies for adolescents in high-risk settings.
YOUTHS LIVING in high-poverty urban settings are at increased risk for adverse developmental and health outcomes,1,2 and African American adolescents are particularly vulnerable.3,4 One such outcome, early sexual initiation, has potentially severe consequences since it places children at risk not only for early childbearing4 but also for sexually transmitted diseases,5 including infection with human immunodeficiency virus.6,7 One potential intervention strategy to avert these outcomes is to increase adolescent skills for condom use and negotiation.8,9 Another approach is to change the adolescent's social environment so that risk avoidance is continually reinforced. This approach includes training parents to engage in risk-reducing strategies,10,11 such as increased monitoring of their children's social behavior12,13 and greater communication about the risks of sex.14- 16 This research was designed to determine if risk-reducing strategies currently adopted by parents might be effective for children living in high-poverty urban settings and the potential mechanisms that might enable such strategies to be successful.
Recent theories emphasize the important role that parents can play by encouraging less favorable attitudes in their children toward risky behavior and ultimately less socializing in risky peer groups.17,18 Studies of sexual activity12,13 and alcohol and cigarette use19 support this theory. However, these studies were not conducted with adolescents in high-poverty urban settings, where early sexual initiation is more normative than in lower-poverty settings.20 In a high-risk sample of African American youths aged 9 to 15 years, Romer et al21 found that parental monitoring was related only to very early sexual initiation (aged ≤10 years) and not to subsequent initiation of sex or condom use. These findings suggest that increased parental monitoring may be a much less effective strategy in settings where early sexual initiation is common.
Despite the lack of evidence supporting the protective role of parents in high-risk settings, other explanations for the limited effects of monitoring should also be considered. One such explanation is that monitoring alone may be insufficient to encourage risk-reducing behavior. Parents may also need to engage their children in specific communication about sexual risks. There is evidence that such communication is related to greater use of contraception by girls15,22,23 and less sexual activity overall.14- 16 However, parents who monitor their children's social behavior may not counsel them about sexual risks.
To identify their separate influences, we assessed both the level of parental monitoring and the frequency of discussion children experienced with their parents about sexual risks. To test the mediating role of attitudes toward risk behavior and exposure to risky peers, we also assessed children's attitudes toward sex and condom use as well as perceptions of the prevalence of sexual experience and condom use among their friends. Finally, because monitoring tends to decline as children age,21 cross-sectional studies may obscure the true role of monitoring by failing to distinguish between youths who have experienced long-term low levels of monitoring and those whose monitoring has declined more recently. To reduce the effect of this problem, we studied a wide age range (ages 9 to 17 years) so that the effects of parenting and communication could more easily be detected not only for very early onset but also for later sexual behavior.
Participant recruitment was similar to that of Romer et al,21 with the exception that 8 housing developments were surveyed (2 more than the original 6). The housing sites were located in the same large US city and contained approximately 11,500 residents, with approximately 2400 children between the ages of 9 and 17 years. It is estimated that more than 80% of the families are headed by a single parent and that virtually all the families are at or below official poverty levels and are African American.
Children were recruited to participate in the survey with the help of community leaders (tenant council presidents, recreation center directors, and social workers). The leaders distributed information about the survey and parental consent forms to youths and parents in the sites. The interview was described as important for learning more about ways to reduce acquired immunodeficiency syndrome risk behavior, with an emphasis on what youths do with their friends. The privacy and confidentiality of the survey results were also emphasized.
Interviews were conducted in a public setting (eg, recreation room) for 1 week at each of the public housing developments. All age-eligible children who appeared during this time and whose parents had completed the appropriate consent forms were interviewed and given $10 for participating. We recruited and interviewed 315 children in 6 sites, with approximately 50 children per site. Additional interviews with 45 children in 2 other sites were conducted to increase our representation of children underrepresented in the first set of sites (13-year-old girls and children aged 16 and 17 years). Data for a small number of children (n=5) were not included because of computer failure or data loss. This resulted in a total sample of 355 children.
The interviews were administered by "talking" computers that were programmed to deliver the questions through earphones. This procedure provides several benefits. First, it increases the confidentiality of the survey because no one else can see or hear the questions that the respondent receives, and no one can see the answers after they are entered into the computer. Second, by presenting a uniformly accessible interview to all respondents, computer administration reduces differences in cognitive and reading ability that might influence the rate of response across the wide age range. Finally, it permits group testing in a relatively public space, making it more likely that we can attract respondents of all risk levels to participate.24 Considerable research now indicates that children and young adults report higher levels of sexual risk behavior when responding in private either by computer24,25 or tape recorder.26
Instructions given at the beginning of the interview by a member of the research team explained how to use the computer to answer questions and ensured that each respondent knew how to complete the interview. Respondents answered most of the questions by using a mouse to click "yes" or "no" on the screen. Some questions required entering numerical and verbal information via the keyboard. Questions could be repeated if necessary and skipped if the respondent chose not to answer. Although children completed the interview by themselves, members of the research team were present to answer questions. The interview took about 30 minutes to complete.
The interview was similar to the one administered in Romer et al.21 It opened with questions about the respondent's friends, asking for the first names (or nicknames) of up to 9 same-sex and opposite-sex friends. It then moved to questions about the respondent's family, including questions about how often parents or other responsible adults monitored the respondent's social activities (never, sometimes, or always). The items were adapted from questionnaires used by Steinberg et al 27 and by Patterson and Stouthamer-Loeber.28 A factor analysis of the entire set of 9 items indicated that 3 items formed a separate factor from the other 6 and that these items focused more heavily on parental initiation of discussion about the youth's activities (Do your parents: expect you to call them if you are going to be home late? want to know who you are going to be with before you go out? ask you where you're going when you go out?) rather than relying on the youth's disclosure of social activities (eg, Do you talk with your parents about what you are going to do before you go out with friends?). The item scores (1-3) were added together to form a single parental monitoring scale (α=.65).
The interview also asked 11 questions about the frequency with which the respondent and parents discussed risks associated with sexual and other behaviors (never, sometimes, or often). The item format was taken from Freeman and Rickels.23 Factor analysis of the responses to these questions revealed that discussions with parents about sexual risks (ways to avoid contracting sexually transmitted diseases and acquired immunodeficiency syndrome and how to avoid pregnancy) and about sexual maturation were likely to co-occur and yet were distinctive from discussions about drug use, violence, and peer relations. We averaged the responses to the 5 sexual risk questions to form a single scale (α=.77). This scale was correlated with parental monitoring to a moderate degree (r=0.36, P<.001), indicating that it measured a related but distinctive aspect of parental behavior.
Parental monitoring tends to provide protection from adolescent risk behaviors irrespective of the type of parent or guardian engaging in the oversight.27 Nevertheless, to determine if the type of guardian could explain the success of parental risk-reducing behaviors, we asked respondents to select the person at home "who watches you the most." Choices were mother, father, aunt or uncle, grandparent, or someone else.
The interview then moved to the respondent's sexual behavior, including questions about engaging in sexual intercourse and use of condoms. The question assessing sexual intercourse asked if the respondent ever "had a boy put his penis in your vagina (girls)/put your penis in a girl's vagina (boys) which is sometimes called having sex, making love, screwing, doing it, or going all the way?" The respondent was also asked: "In the next year, do you think you will have sex with someone?" To determine if respondents had ever engaged in anal intercourse, we also asked if the respondent ever "had a boy put his penis in your anus or butt (girls)/put your penis in a girl's anus or butt? (boys)." Responses to the vaginal sex question exhibited a high level of test-retest stability (κ=.82) over a 2-week period of retesting.20
For those who had acknowledged having sex, condom use was measured with a question assessing whether the respondent had ever used a condom ("When you have had sex, did you or your partner ever use a condom?"). A score based on this question exhibited high test-retest stability (r=0.83 over a 2-week interval) in a study by Romer et al.21 In addition, we measured consistency of condom use (among those respondents who reported ever using condoms) by asking how often condoms had been used, with possible responses including never (1), sometimes (2), most of the time (3), and always (4). We also asked how often condoms would be expected to be used in the next year (for all respondents who said they expected to engage in sex in the next year).
To measure attitudes toward sex, we asked respondents to rate their feelings about engaging in sex in the next year, with a scale ranging from very bad (1) to very good (5). This item had also had high test-retest reliability (r=0.74) in a study by Romer et al.21 We also asked how respondents would feel about not using a condom, with the same response alternatives. To assess the role of perceived peer prevalence of risk behavior, we included questions about the behavior of the respondent's friends who had been named earlier in the interview. These questions asked if none, some, or most of the respondent's friends had engaged in sex and if at least some had, whether none, some, or most ever use condoms. Test-retest r values (over a 2-week interval) were 0.67 and 0.57, respectively, in a study by Romer et al.21
At the close of the interview, we asked respondents if they had ever participated in a computer interview with our project before. Respondents might have been surveyed in the earlier study21 or in other pilot work done at the sites.
Because test-retest correlations for the critical questions were high in earlier research (ranging from 0.5 to 0.8),20,21 we did not conduct a reliability study in this wave of interviewing. Furthermore, the consistency of responses within the interview for the parental communication and monitoring scales indicated that children were responding in a systematic and reliable manner.
Table 1 contains the distribution of children by age and sex as well as the percentage who had initiated sex. Approximately 45% of the respondents reported engaging in vaginal sex. A smaller proportion of respondents (data not shown) reported engaging in anal sex (19%). Most of these respondents were boys (59 [89%]), nearly all of whom had also engaged in vaginal sex (56 [95%]). All of the girls who reported having engaged in anal sex also reported experience with vaginal sex (n=7).
Also shown in Table 1 are the scores for prevalence perceptions of sexual experience among friends and attitudes toward engaging in sex. Multiple regression analyses of these scores indicated that they both increased with age (t344=7.77, P=.0001, ΔR2=0.145 and t340=8.81, P=.0001, ΔR2=0.163, respectively). In addition, boys were more likely than girls to have favorable attitudes toward engaging in sex in the future (t340=5.84, P=.001, ΔR2=0.072). Both age and sex were related to the rate of both sexual and condom-use initiation (D.R., B.S., J.G., S.F., M.M.B., and X.L., unpublished results, 1994). In addition, there were strong relationships between prevalence perceptions and sexual initiation and between attitudes and behavior. Finally, neither site differences nor previous participation in similar surveys (35%) was related to differences in sexual behavior. Therefore, in this report we focus on the relationship between sexual behavior and potential parental influences.
Table 2 contains the results for the 2 parental behavior scales: parental monitoring and communication regarding sex. For ease of presentation, each of the scale score distributions was trichotomized into groups indicating high (>2.67), moderate (2.0-2.67), and very low (<2.0) rates of parental influence. Approximately 70% of the children reported that their mothers watched them the most at home. Grandparents were the next most frequent guardian (12%), followed by aunt/uncle (6%), father (5%), and someone else (7%). Analyses of differences between these types of guardian revealed that mothers were more likely than other guardians to engage in monitoring (t349=3.13, P=.0001, ΔR2=0.025) and discussion regarding sex (t347=3.07, P=.0001, ΔR2=0.024). These analyses also revealed that older children reported less monitoring than younger children (t349=3.62, P=.001, ΔR2=0.034) but no differences concerning communication about sex (t349=.56, P=.57). In addition, girls reported more parental influence of both types than boys (t349=4.13, P=.0001, ΔR2=0.044, and t348=6.75, P=.0001, ΔR2=0.114, respectively). On average, only 11% of boys reported high levels of communication from their parents regarding sexual risks, whereas more than a third of girls reported such discussions.
Figure 1, left, contains a plot of the proportion of children who reported initiating sex as a function of age and parental monitoring. Data were aggregated by age into 4 groups (ages 9-10, 11-12, 13-14, and 15-17 years) to increase the sample size in each comparison of parental monitoring by age. A multiple logistic regression analysis indicated that parental monitoring was inversely related to sexual initiation (χ21=19.16, P=.0001), holding age, sex, their interaction, and type of guardian constant. This finding indicates that more heavily monitored children were less likely to exhibit early sexual onset (aged ≤10 years), a result consistent with earlier research.21 However, the linear interaction between monitoring and age was also a significant predictor (χ21=5.26, P=.03). As seen in Figure 1, the rate of initiation was lowest in the most heavily monitored group. Children's reported discussion with guardians about sexual risks did not add to the prediction of sexual initiation, either alone (χ21=0.66, P=.42) or in interaction with parental monitoring (χ21=0.35, P=.55).
Left, Proportion of adolescents sexually initiated by age and level of parental monitoring. Right, Proportion of adolescents engaging in anal sex by level of monitoring and communication about sexual risks.
The protective effects of monitoring were also evident for anal sex. The more parental monitoring that children reported, the lower their likelihood of engaging in anal sex (χ21=5.79, P=.02). However, as seen in Figure 1, right, both monitoring and discussion regarding sex were implicated in this risk behavior. When monitoring occurred with some frequency, discussion regarding sex was inversely related to engaging in anal sex (χ21=4.70, P=.04). This interaction was significant when holding constant other predictors, of which only sex was an important contributor (χ21=29.15, P=.0001).
We entered 3 scores into the prediction equations for vaginal sex to see if we could explain the mechanism by which parents influenced their children's behavior (reported communication from parents concerning sexual risks, attitude toward engaging in sex, and perceived prevalence of sexual experience among friends). Virtually none of the relationship between vaginal sex and monitoring was attributable to communication about sexual risks. However, part of the influence of monitoring seemed to be related to children's attitudes toward sex. Attitudes were related to monitoring (r=−0.26, P=.001), and controlling for attitudes reduced the predictive value associated with monitoring by 28%. Monitoring was also related to perceived prevalence of sexual experience among friends (r=−0.19, P=.001). However, controlling for prevalence did not significantly reduce the predictive value associated with monitoring beyond that attributable to attitude.
A similar analysis conducted for anal sex revealed that attitudes toward sex were also implicated in the relationship between monitoring and initiation of anal sex. Controlling for attitudes reduced the predictive value associated with monitoring by 11%. However, attitudes toward sex did not seem to be related to the additional role played by discussion in combination with monitoring. As with vaginal sex, perceived prevalence among friends was unable to account for any additional predictive value associated with monitoring.
Table 3 contains the results for reports of ever having used condoms among sexually experienced respondents. Because of the smaller number of sexually experienced respondents, we report these results in 4 age groups (9-10, 11-12, 13-14, and 15-17 years). Also contained in Table 3 are results for perceived prevalence of condom use by friends and attitudes toward not using condoms. Perceived prevalence increased with the age of the respondent (t136=1.81, P=.08, ΔR2=0.023), but did not differ by sex (t136=0.61, P=.54). However, older children felt better about not using condoms than younger children (t153=2.30, P=.03, ΔR2=0.033), and boys felt better about nonuse than girls (t153=1.91, P=.06, ΔR2=0.023).
Parental monitoring was unrelated to condom use initiation (χ21=1.06, P=.30) when holding constant age, sex, age×sex interaction, and type of guardian. However, as expected, communication about sexual risks was related to the rate of initiation of condom use. To increase the sample size in each cell, we aggregated the data into 2 age groups (ages 9-14 and 15-17 years). As seen in Figure 2, children who reported frequent communication from parents about sexual risks exhibited a faster rate of initiation of condom use than children who reported no communication from their parents (χ21=9.65, P=.001). Intermediate levels of communication (sometimes) also seemed to be protective. The difference in the initiation rate was primarily evident in the later age range of 15 to 17 years. In addition, these patterns occurred for both boys and girls, with no differential effectiveness of discussion across increasing levels of monitoring.
Proportion of adolescents reporting condom use by age and level of parental communication about sexual risks.
Table 4 reports means for 2 additional indicators of condom use: consistency of use in the past (among those who reported using a condom) and expectations for consistent use in the future (among those who reported that they will have sex in the future). Consistency of past use was not significantly related to either age (t131=−0.95, P=.31) or sex (t131=1.25, P=.20). However, expectations for future condom use tended to decline with age (t185=−1.56, P=.12, ΔR2=0.012) and were particularly likely to decline among girls (t185=2.63, P=.01, ΔR2=0.035). To allow sufficient sample sizes in the various cells for these variables, we aggregated the data into 2 age groups (9-14 and 15-17 years) for past consistency and into 3 age groups (9-12, 13-14, and 15-17 years) for future consistency. Children who discussed sexual risks with parents reported using condoms more consistently in the past (t131=2.07, P=.04, ΔR2=0.031) and expected to continue higher rates of condom use in the future (t185=2.47, P<.05, ΔR2=0.039) (Figure 3). Furthermore, the relationship between parental communication and expectations for the future were even stronger among older children (t185=2.45, P=.03, and ΔR2=0.030). The associations observed between communication and consistency occurred for both boys and girls, with no differences as a function of monitoring.
Left, Mean consistency of past condom use (among those who have initiated condom use). Right, Mean expected condom use in the future (among those who say they will have sex in the future) was measured using a 4-point scale, ranging from never (1) to always (4).
Parental communication about sex was related to attitudes toward using condoms (r=−0.13, P=.02) but was unrelated to perceived prevalence among friends (r=−0.02, P=.80). However, in a regression analysis, attitudes toward condoms were too weakly related to parental communication to reduce the observed relationship between condom use and parental communication about sex.
Despite legitimate concerns about the harmful effects of peer influences on adolescent development in high-poverty settings,1,2,29 our results suggest that parental practices can affect the developmental course of sexual risk behavior beyond the preadolescent years. Parental monitoring was related to the rate of sexual initiation throughout the age range studied, with heavily monitored youth exhibiting a slower rate of initiation than less-monitored youth. This finding suggests that children in homes with parents or other guardians who continued to monitor their children's social behavior throughout adolescence were less likely to initiate sex than children whose guardians either never monitored their behavior or discontinued monitoring as they aged.
Although we replicated the finding that monitoring is inversely related to early onset of sexual activity (aged ≤10 years), the results stand in contrast to other findings21 that monitoring was related only to early sexual initiation (as assessed by the simple effect of monitoring) and not to the rate of initiation as adolescents age (the interaction with age). Monitoring continued to be related to sexual initiation after controlling for different starting levels of initiation. The previous failure to find enduring effects of monitoring may have been the result of studying too narrow an age range. With this study's wider age range, the relationship between monitoring and later sexual initiation was more evident. Indeed, when we restricted the analysis to children aged 9 to 15 years, the interaction between age and monitoring was much less evident (χ21=0.03, P=.85).
Both monitoring and discussion regarding sexual risks were related to engaging in anal sex. When discussion was infrequent, monitoring was largely unrelated to the initiation of anal sex. However, monitoring was strongly related to the initiation of anal sex when children reported some discussion with their guardians regarding sexual risks. These results suggest that parental behaviors may be critical for reducing the initiation of this risky behavior.
The importance of parental communication was also evident for condom use. Discussion with guardians was related to condom use for both boys and girls even though parental monitoring was not. Parental communication was particularly powerful among older adolescents. Whether this finding reflected greater effectiveness for parents of older children or increased effectiveness over time is not clear. However, these results suggest that more focused communication from parents concerning sexual risks is critical to encourage condom use and that parental monitoring alone is insufficient to promote safer sexual behavior. Interventions to increase parental monitoring could benefit from also focusing on greater communication with children about sexual risks and ways to avoid them.
Parental communication regarding sexual risks also increased the likelihood that children used condoms consistently in the past and expected to do so in the future. Consistency of condom use tended to decline with age. This finding replicates the findings of other studies with adolescents30- 32 and suggests that although initiation of condom use is high and increases with age, actual consistency of use tends to decline as adolescents age. A similar pattern was observed for attitudes toward condom use. It is encouraging therefore that parental communication might be a mechanism for increasing not only the initiation of condom use but also the consistency of condom use.
Parental practices seemed to influence the initiation of both sex and condom use primarily by affecting attitudes toward behavior rather than by affecting exposure to risky peers. This finding is important because it suggests that even if parental monitoring or communication about risks does not reduce affiliation with risk-prone peers, parental influence could moderate the impact of peer influence by reducing the attractiveness of risky behavior. The initiation of sex diffuses through peer networks more rapidly among youth with favorable attitudes toward the behavior (D.R., B.S., J.G., S.F., M.M.B., and X.L., unpublished results). If parents can influence these attitudes, they can exert some control over their children's sexual risk taking by reducing the influence of risky peers, even if parents cannot directly control the peer networks in which their children socialize.
A less optimistic interpretation of the findings would attribute them to the unique characteristics of families in which parents monitor their children's behavior or communicate with them regarding sex.33 If this were the case, changing parenting practices might not be a promising intervention strategy. Our finding that monitoring and communication remained as predictors of safer behavior even after controlling for type of guardian suggests that it is the parental behavior itself and not the parental figure per se that engages the protective effects of these risk-reducing strategies. Nevertheless, further research is needed to determine if this strategy can be adopted by parents who would not use it spontaneously and if interventions to increase this influence are effective. Some initial work on this intervention strategy suggests that guardians who do not normally monitor their children's social behavior can be trained to do so.34 It remains to be seen if this intervention strategy can produce the same enduring change in adolescents' sexual risk-taking behavior that was associated with the natural adoption of these strategies by some guardians in this study.
The present sample was drawn from housing developments, with appropriate stratification by age and sex. Nevertheless, there is the concern that our recruitment methods, which relied on the assistance of parents and housing staff, might overrepresent adolescents who are less prone to risk. Even if this were true, it need not prevent us from drawing conclusions about the relationship between variation in risk behavior and parental influences. Indeed, the high levels of early sexuality and unprotected sex exhibited by the sample suggest that we have included a sufficient proportion of high-risk youth to enable tests of the major study questions.
We are also unable to rule out the existence of coercive sexual relations, especially among the younger youth in this sample. However, the finding that even moderately monitored children were less likely to report very early sexual experience is consistent with the hypothesis that parental monitoring can reduce the chances of early initiation, whatever the source of this experience.
Analysis of the recent National Longitudinal Study of Adolescent Health,35 a nationally representative sample of school-attending adolescents, indicated that contact between parents and adolescents is a major protective factor for a wide range of risk behaviors, including early sexual initiation. Our results extend these findings to the specific behaviors of parental monitoring and communication in a very-high-risk setting. The use of computer interviewing increases the comparability of the results across ages and provides sensitive measures of the relationship between parental influences and sexual behavior. These features further increase our confidence that the relationships we observed between parenting and youth were not distorted by the sample we obtained.
Interventions with parents and other guardians are a promising strategy for reducing the risks of sexual initiation, because parents may be a more constant protective influence on children during the adolescent years than other influences. Interventions with adolescents designed to change their skills and attitudes may not persist over time. For example, interventions that have successfully increased condom use in adolescents8,9 or reduced early initiation of sex36 have been found to lose their effect within 1 year. However, interventions that influence both parents and adolescents may have better long-term effects than comparable efforts directed at adolescents alone. This prospect deserves to be tested to determine the separate and interacting influences of both parent and adolescent change programs. In addition, practitioners who serve adolescents in high-risk settings could encourage parents or guardians to become more involved with their children's sexual decision making and thereby increase the chances that adolescent sexual initiation will be either delayed or appropriately protected.
Accepted for publication March 1, 1999.
This study was supported by grant R01 MH54983 from the National Institute of Mental Health, Bethesda, Md (Dr Stanton, principal investigator), and by funding from the University of Pennsylvania, Philadelphia (Dr Romer, principal investigator).
Editor's Note: Just in case we needed it, here's more evidence that good parenting usually produces good results in children.—Catherine D. DeAngelis, MD
Corresponding author: Daniel Romer, PhD, Center for Community Partnerships, University of Pennsylvania, 133 S 36th St, Fifth Floor, Philadelphia, PA 19104-3246 (e-mail: email@example.com).
Romer D, Stanton B, Galbraith J, Feigelman S, Black MM, Li X. Parental Influence on Adolescent Sexual Behavior in High-Poverty Settings. Arch Pediatr Adolesc Med. 1999;153(10):1055-1062. doi:10.1001/archpedi.153.10.1055