Prognostic value of parent–adolescent disagreement in a referred sample
- First Online:
- Cite this article as:
- Ferdinand, R.F., van der Ende, J. & Verhulst, F.C. Eur Child Adolesc Psychiatry (2006) 15: 156. doi:10.1007/s00787-005-0518-z
- 281 Views
To investigate whether parent–adolescent disagreement regarding reports on adolescents’ psychopathology indicates a poor prognosis.
A total of 151 11- to 18-year-olds who had been assessed with the Child Behavior Checklist (CBCL) and Youth Self-Report (YSR) at referral to an outpatient psychiatry clinic were followed up. At follow-up, 4.3 years later, signs of poor outcome were assessed.
Discrepancies between CBCL and YSR scores predicted future disciplinary problems at school, police/judicial contacts, and drug use.
To determine the prognosis of psychopathology in adolescents who are referred to mental health services, discrepancies between parents and adolescents may be useful. Given the relatively large number of statistical comparisons (n=16) that was needed to obtain these results, future studies are needed to test if the results are robust across different settings.
It is often recommended that assessment of psychopathology in adolescents should include gathering data from multiple informants . After completing the assessment of an adolescent, the clinician has to weigh and judge all available information. Parents and adolescents themselves often disagree about the presence or severity of psychopathology [6, 10, 26, 31, 33]. Until now, it is unclear how discrepancies should be dealt with, especially because information both from parents and from adolescents themselves can be valid, despite the presence of differences .
Theoretically, successful treatment might be hampered if parents and adolescents disagree about the presence or absence of problems. Furthermore, differences in reports from key informants might warrant a specific treatment approach. As an example, in their book on empirically based assessment, Achenbach and McConaughy  mentioned that aggressive behavior reported by the teacher, and not by other informants, deserves a different treatment strategy than aggression reported by all informants. They did not provide empirical evidence for this hypothesis. Moreover, although it is likely that parent–child disparities have clinical significance, only two studies have assessed the significance of parent–adolescent disagreement regarding the level of psychopathology as a risk factor for poor outcome [18, 24]. Hence, it is largely unknown if clinicians should consider differences in parent versus self-reports as a warning sign. Kendall et al.  reported that such disagreement regarding the presence of anxiety was associated with slower improvement during cognitive behavioral treatment for anxiety disorder in 8- to 14-year-olds. Ferdinand et al.  assessed adolescents from the general population with standardized self-report and parent rating scales and followed them up across a 4-year time-span. They found that discrepancies between parents’ and adolescents’ reports predicted future referral to mental health services, expulsion from school or from a job, police or judicial contacts, drug use, and the feeling of needing help without receiving it. The study by Kendall et al.  was confined to individuals with anxiety disorders, whereas the study by Ferdinand et al.  concerned adolescents from the general population. More knowledge is needed regarding the clinical significance of parent–adolescent disagreement in a referred sample of adolescents with a wide problem range.
Discrepancies between informants are associated with factors that influence the prognosis. This underscores the possible usefulness of parent-child disagreement as indicative of poor prognosis. Parent–child disagreement was found to be related to high family stress and conflict [19, 25]. These factors might change the way parents perceive their child’s behaviors/emotions. Others found that parental factors, such as low child acceptance and parental dysfunction , maternal depression and/or anxiety [9, 40], and depression in mothers  were associated with disagreement.
The aim of the present study was to investigate whether parent–adolescent disagreement is a risk factor for poor outcome. A group of 11- to 18-year-olds who had been referred to an outpatient psychiatry clinic were assessed with standardized parent and self-report rating scales during the initial assessment phase. At follow-up, 4.3 years later, poor outcome variables were assessed.
Each assessment phase of this study was approved by the Committee for Medical Ethics, Sophia Children’s Hospital/Erasmus University Rotterdam. At each phase, informed consent was obtained from all subjects who completed a questionnaire (parents and youths), after the procedure had been explained.
The present study was a part of the follow-up study of referred children and adolescents [20–22]. For this study, all 4- to 18-year-old children and adolescents who had been consecutively referred between June 1982 and January 1995 the outpatient—university—clinic for child and adolescent psychiatry of Erasmus Medical Center Rotterdam/Sophia Children’s Hospital Rotterdam, and for whom CBCL information was available, were followed up between August 1995 and July 1997. Initial exclusion criteria were: (1) never visited the clinic, (2) IQ < 75, (3) referred for problems other than behavioral/emotional problems (evaluation of developmental level or IQ, custody decision), (4) did not receive psychiatric diagnosis after assessment due to absence of problems. At follow-up, parent forms were sent to parental informants. For adolescents, self-report forms were sent along with the parent forms. If no response was received on this initial approach, a reminder was sent, followed by several attempts to reach respondents by phone, followed by another written reminder, and, finally, if necessary, a home visit. In this way, information on 77.8% of the initial sample was obtained. Heijmens Visser et al.  found that responders and non-responders at follow-up did not differ with respect to initial problem levels or socio-economic status. See previous publications for more details [20, 21].
For the present study, individuals who were 11- to 18-year-olds at referral and for whom CBCL and YSR information was available at initial assessment were selected (n = 151; 96 boys, 55 girls; mean age = 13.0 years; sd = 1.8 years). The maximum time-span between completion of the CBCL and YSR was confined to 1 month. CBCLs were completed by the mothers (n = 148; 98.0%) or fathers (n = 3; 2.0%). Re-assessment took place on average 4.3 years later (sd = 2.2 years, range = 1–9 years). The average age at follow-up was 17.2 years (range = 12–26 years; sd = 2.5 years).
Properties of CBCL and YSR scores
% >P90 normative sample CBCL2(boys/girls)
% >P90 normative sample YSR2(boys/girls)
Pearson correlation1 (entire sample/boys/girls)
Mean CBCL score in the study sample
Mean YSR score in the study sample
Effect size difference CBCL-YSR3 (P)
The Child Behavior Checklist—CBCL —is a parent questionnaire for assessing problems in 4- to 18-year-olds. The part of the CBCL that was used in the present study contains 120 items on behavioral or emotional problems during the past 6 months. The response format is: 0 = not true, 1 = somewhat or sometimes true, and 2 = very true or often true. The good reliability and validity of the CBCL  were confirmed for the Dutch translation [23, 37].
The Youth Self-Report—YSR —was modeled on the Child Behavior Checklist, and can be used to derive self-report ratings of behavioral and emotional problems. The YSR can be used for ages 11 and up, and is worded in the first person. Around 89 of the CBCL and YSR problem items are identical. A few CBCL items on which information can better be obtained by others than subjects themselves (i.e. ‘talks during sleep’) were omitted from the YSR by Achenbach. The good reliability and validity of the Dutch YSR were supported by Verhulst et al. .
Achenbach  constructed eight narrow-band “cross-informant syndromes” that were similar for both sexes: Withdrawn, Somatic Complaints, Anxious/Depressed, Social Problems, Thought Problems, Attention Problems, Delinquent Behavior, and Aggressive Behavior. The cross-informant syndromes were empirically derived from parent reports (CBCL), self-reports (YSR), and teacher reports (TRF), in large clinical samples. The TRF—Teacher’s Report Form  is a teacher questionnaire that was based on the CBCL.
In the present study, the few CBCL/YSR syndrome items that are not similar for both informants were left out of syndrome scores. In total 81 items were similar across the two instruments. In this way, we derived CBCL/YSR syndrome scores that comprised the same syndromes, which enabled optimal comparisons.
Poor outcome variables that reflected a wide range of problems were assessed at follow-up with a parent questionnaire that contained items on (1) disciplinary problems at school during follow-up, defined as problems relating to school other than learning problems, e.g. being suspended or expelled from school, truancy, violent behavior, misbehavior, and social problems) (n = 40), (2) disciplinary problems at a job during follow-up (n = 11), (3) self harm, suicidal ideation or suicide attempt during follow-up (n = 18), (4) police/judicial contacts during the follow-up period (n = 16), (5) parents’ wish for professional help regarding problems of the child at Time 2 (in case the child was still under treatment we regarded the parents’ wish for more or alternative treatment) (n = 52), (6) outpatient treatment at follow-up (n = 32), (7) inpatient treatment during follow-up period (n = 16), (8) having had an accident that required professional attention (n = 25), and (9) drug use during the 6 months prior to follow-up assessment (n = 11). Each outcome variable was scored 0 if absent and 1 if present.
Pearson correlations were computed to assess associations between CBCL and YSR syndrome scores. Repeated measures analyses were used to assess differences in average CBCL and YSR syndrome scale scores for each cross-informant syndrome. These differences characterize the sample with respect to differences between judgments of parents versus adolescents and can be used for comparisons with other samples.
To determine which types of behavioral or emotional problems were predictive of poor outcome, we performed forward stepwise logistic regression analyses (likelihood ratio tests were used for adding a candidate predictor to a model), for each poor outcome variable separately. Scores on the eight CBCL syndromes and 8 YSR syndromes were entered simultaneously as predictors, together with age and sex. Similar analyses were performed for each poor outcome variable. CBCL syndrome scores were recoded into categorical variables by dividing subjects with high versus low scores, before entering them into the regression analyses. The 90th percentile (P90) of the cumulative frequency distribution of a syndrome score in 11- to 18-year-olds from a Dutch epidemiological sample was used as a cutpoint. This sample had been used previously to derive Dutch normative data for the CBCL [34, 35], and contained 867 11- to- 18-year-olds for whom both CBCL and YSR information was available. Syndrome scores above the P90 of the general population sample were considered as deviant (and coded as 1), whereas scores equal to or below the P90 were considered as non-deviant (and coded as 0). YSR syndrome scores were recoded in a similar way, by making use of the same epidemiological data set as for the CBCL. See Verhulst et al.  for a description of the YSR data set.
Then, a second set of similar logistic regression analyses was conducted, to investigate if information regarding discrepancies between informants provided a better predictive model of an outcome. Therefore, to derive scores that reflect discrepancies between CBCL and YSR syndrome scores, continuous YSR syndrome scores were subtracted from continuous CBCL syndrome scores. Thus, scores on the YSR syndrome Withdrawn were subtracted from scores on the CBCL syndrome Withdrawn, and so on for the other scales. In this way, discrepancy scores were derived for each scale. We also derived discrepancy scores from the epidemiological data set. After calculating a frequency distribution for the discrepancy scores in this data set, discrepancy scores in the clinical data set were recoded into 0 (equal to or below 10th percentile score in the epidemiological data set), 1 (above 10th percentile, equal to or below 90th percentile), or 2 (above 90th percentile). In this way, individuals were divided into 3 groups: (a) adolescents who reported much more problems than their parents, (b) parents and adolescents with scores in the same range, and (c) parents who reported much more problems than their adolescent children. These discrepancy scores were entered into the logistic regression analyses, together with the recoded CBCL and YSR syndrome scores. In this way, the additive value of discrepancy scores to predict a poor outcome, over and above the predictive value of regular CBCL and YSR syndrome scores, was computed. By computing extra contrast effects—(a) between those who scored 0 and those who scored 1, indicating that youths provided higher scores than their parents, and (b) between those who scored 2 and those who scored 1, indicating that parents provided higher scores than youths, we were able to assess which type of discrepancy was important.
In Table 1, Pearson correlations between CBCL syndromes and their YSR counterparts are presented, for the entire sample, and for each sex separately. These correlations can be used to interpret results of regression analyses. The higher the correlations between CBCL and YSR syndromes, the higher the informant agreement, and the less likely CBCL and YSR syndrome scores will constitute independent predictors of poor outcome at follow-up. Furthermore, if these correlations are high, the probability of informant disagreement, and therefore, the chance that informant disagreement effects will occur, declines.
Repeated measures analyses showed that 5 of the 8 average CBCL syndrome scores were higher than their YSR counterparts. According to Cohen , the effect size (partial eta squared) for Social Problems was large (>13.8%), for Anxious/Depressed, Attention Problems, and Aggressive Behavior moderate (between 5.9% and 13.8%), and for Withdrawn small (between 0.99% and 5.9%).
Results of logistic regression analyses
Predictors from analyses without discrepancy scores
Extra predictors from analyses with discrepancy scores6
Disciplinary Problems at school
P>Y2 Attention Problems
CBCL Delinquent Behavior
Disciplinary Problems at a job
YSR Aggressive Behavior
Self harm/suicidal ideation/attempt
YSR Delinquent Behavior
Wish for help
YSR Somatic Complaints
YSR Aggressive Behavior
Outpatient treatment at follow-up
Inpatient treatment during follow-up
YSR Attention Problems
CBCL Attention Problems
YSR Social Problems
Y>P1 Aggressive Behavior
YSR Aggressive Behavior
It is shown that most outcome parameters were predicted by CBCL or YSR syndrome scores. For instance, future accidents were predicted by scores on the CBCL syndromes Withdrawn and Attention Problems. The odds ratio of 0.22 for Withdrawn indicates that adolescents with scores >P90 on the CBCL scale Withdrawn had a 0.22 times lower chance to have future accidents than those with low scores on Withdrawn. The odds ratio for Attention Problems indicates that, in case of high instead of low scores on the CBCL scale Attention Problems, the probability of having serious accidents as an outcome increased 20.6-fold.
Furthermore, it was found that P>Y or Y>P effects improved the prediction of three outcomes: disciplinary problems at school, police/judicial contacts, and drug use. For instance, Y>P effects for Aggressive Behavior improved the prediction of drug use. The odds ratio of 27.5 indicates that those adolescents who reported far more aggressive behaviors than their parents were more likely to use drugs in the future than those who tended not to disagree with their parents about the presence of aggressive behaviors.
Table 2 also shows that the logistic regression analyses did not reveal sex or age effects.
In the present study, we tested the usefulness of information regarding discrepancies between parents and youths to determine the prognosis of psychopathology in a referred sample of 11- to 18-year-olds. This sample was followed up—on average—4.3 years later. The study found that discrepancies between parents and adolescents themselves may be useful to determine the risk of poor outcome. CBCL-YSR discrepancy scores predicted disciplinary problems at school, police/judicial contacts, and drug use.
Two P>Y-discrepancy effects were found. P>Y Attention Problems scores predicted disciplinary problems at school, whereas P>Y Anxious/Depressed scores predicted police/judicial contacts. Poor prognosis in adolescents with P>Y discrepancies may result from different definitions of problems by parents and adolescents. The finding that P>Y Attention Problems scores predicted disciplinary problems at school suggests that the risk of school problems increases if the adolescent is scored high on Attention Problems by a parent, but not high by him or herself. It is likely that adolescents who do not recognize their problems may be less receptive for therapeutic or educational interventions. Indeed, previous studies have shown that ADHD is associated with a limited self-awareness and related meta-cognitive abilities for self-evaluation arising from the disorder . The fact that we found a similar P>Y effect for Attention Problems regarding the prediction of disciplinary problems at school in a Dutch general population sample  suggests that this effect is not a chance finding.
The P>Y effect for Anxious/Depressed indicated that adolescents who receive much higher anxiety/depression scores from their parents, compared to how they score themselves, may be at risk for future police/judicial contacts. This was not found in our previous study in the general population. Apparently, this finding represents a mechanism that may be specific for a referred sample. Referred adolescents who are regarded as anxious or depressed by others, but not by themselves, may have difficulties with recognizing their emotions, which may make them prone to a worse prognosis. This is somewhat in line with other studies that suggested severe antisocial problems are accompanied by lower emotional responsivity . The present study suggested that the assessment of under recognition of anxiety/depression could be useful in clinical practice to identify those who are at risk for a delinquent outcome.
The Y>P effect that was found for Aggressive Behavior indicated that adolescents who scored themselves much higher on the Aggressive Behavior scale than their parents were at risk for future drug use. Apparently, especially those adolescents with high levels of aggressive behaviors, of which, however, their parents were not aware, were at risk for future drug use. This was in line with previous studies that investigated the importance of parental monitoring with regard to initiation of drug use [27, 32].
In their meta-analysis, Achenbach et al.  found an average correlation of 0.25 between self-reports and parent reports. This is considerably lower than the correlations between CBCL and YSR syndrome scores in the present study. Even though inter-informant correlations tend to be somewhat higher in referred versus non-referred samples , the correlations that were found can still be regarded as considerable. The results of the present study also differed from previous evidence with regard to different inter-informant associations for internalizing versus externalizing problems. First, correlations were similar for internalizing versus externalizing problems, whereas disagreement between parents and referred children has previously been reported to be higher for internalizing than for externalizing problems [12, 16, 30, 39]. Second, Sawyer et al.  found that, whereas adolescents from the general population generally score themselves higher than parents, scores in referred adolescents tend to become higher for internalizing problems, and lower for externalizing problems. In the present study, some of the average CBCL scale scores were higher than YSR scale scores, which, however, was true for internalizing and externalizing problems. The constitution of the sample that, given its constitution has undoubtedly been affected by referral biases [8, 11], may be responsible for these differences. For instance, the fact that, especially in adolescence, parents and children will have to agree to a certain degree on the presence of problems before referral will actually take place, may have inflated correlations between CBCL and YSR scores.
Although some discrepancy effects were found, it is clear from Table 2 that levels of psychopathology ratings by parents and adolescents themselves still were the most powerful predictors of future malfunctioning. Hence, main assessment procedures should remain aimed at assessing levels of behavioral and emotional problems, and the importance of discrepancy effects should not be overestimated. The finding that CBCL scores and YSR scores predicted poor outcome underscored the need for information from different informants in clinical practice.
In the presence of predictive effects of separate CBCL and YSR scores, discrepancy effects added to the predictive models for three of the eight poor outcome variables. This indicated that, for the prediction of these poor outcomes, information from one of the two informants became more relevant when information from the other informant was taken into account. Apparently, if parents and adolescents disagree, assessment and diagnosis of child psychopathology should include more than trying to determine which informant is ‘right’. Instead, it might be helpful to consider information regarding discrepancies between informants as possibly useful, and of clinical significance. If parents and adolescents disagree, the risk of adverse outcome may be higher. This might warrant special attention for communication between parents and adolescents. For instance, the present study indicated that adolescents who reported aggressive behaviors, while their parents did not, were at risk for substance use. For such adolescents, it could be important to pay attention to parent–child communication about the adolescent’s aggressive behaviors, and for getting to a problem definition that is shared by parents and child. Hence, compared to a situation in which parents and adolescent would agree about the presence or absence of problems, a different clinical approach might be needed. Future studies are needed to test this type of hypotheses in treatment outcome research.
Information regarding poor outcome was obtained from parents and not from adolescents themselves. This could be an explanation for P>Y discrepancies predicting poor outcome. Adolescents may merely have a different perception than their parents of the presence or absence of problems. Further, a clinical sample was investigated, which enlarged the probability of referral biases. However, the sample provided the opportunity to investigate factors that influenced the prognosis of psychopathology in adolescents, once they are referred to an outpatient clinic. The application of a relatively large number of statistical tests (n = 16) to assess prediction of poor outcome can be regarded as another disadvantage. To test the importance of the findings future studies would be needed to test the robustness of the findings across settings and cultures.