Design
Factors associated with patient-clinician discordance regarding unmet care needs were investigated using a cross-sectional design.
Setting
The study was conducted at the department of child and adolescent psychiatry at a large specialized mental healthcare institution in the Netherlands. This department had two general outpatient clinics and one youth-Assertive Community Treatment team (ACT). Diagnostic assessments and treatment (e.g., cognitive behavioral therapy, family support, and pharmacological treatment) were provided by three child psychiatrists, seventeen psychologists, five clinical nurse specialists, and two mental health nurses.
Participants
The target population consisted of all 6 to 18-year-olds who had been referred to the department, and their clinicians. Only one child per household was included in the study. A total of 467 patients were eligible for inclusion. The final sample consisted of 244 patients. Figure 1 shows the flow chart of the inclusion process.
Ethical approval
The study was approved by the following: the Medical Ethical Committee at VU University Medical Center Amsterdam (protocol no. 2015.245); the Scientific Committee at the Amsterdam Public Health Research Institute; and the local research committee at the participating mental health institution.
Separately, participating children/adolescents and clinicians received written and oral information on the research project. In accordance with prevailing Dutch legislation, the written consent of parents and/or children/adolescents was obtained on the following basis: (i) if children were younger than 12, only parents were asked for consent; (ii) if children were aged between 12 and 16, parents and children were both asked for consent; and (iii) if adolescents were 16 or older, informed consent was obtained only from the adolescents themselves.
Measurement instruments
Sample descriptives
The Demographic Information Questionnaire (DEMOG) child version was used to establish age, gender, and country of birth [33]. The Neuropsychiatric Interview for Children and Adolescent (MINI-KID) was used to establish the patients’ psychiatric diagnoses [34]. If a child was aged 12 or above, the MINI-KID was administered. If a child was younger, the MINI-KID was administered in the presence of one of the parents. Parents were allowed to clarify questions for their child. For disorders that were not covered by the MINI-KID (personality disorders, autism spectrum disorders), clinical diagnoses were used.
Assessment of care needs
To assess a patient’s unmet care needs, we used the Camberwell Assessment of Need Short Appraisal Schedule (CANSAS) [35], which covers 25 care need items that can be scored on a three-point scale. The response format is 0 = no need, 1 = met need, and 2 = unmet need. The CANSAS was used in the form of an interview. If a child was aged 12 or older, it was administered to the child alone. If the child was younger, it was administered in the presence of one of the parents. At the start of the interview, parents were instructed not to answer for the child, but to clarify the questions in such a way that the child was able to answer the question from her or his own perspective. Simultaneously, the clinician also completed the CANSAS scoring form on the basis of all the clinical information available.
Outcomes
Three dependent variables were studied: discordance between young people and their clinicians for (i) unmet care needs regarding mental health problems, (ii) unmet care needs regarding information on diagnosis and treatment, and (iii) unmet care needs regarding making and/or keeping friends.
To determine the presence or absence of patient-clinician discordance, scores for each of the three CANSAS items were recoded into 0 = no need/met need, and 1 = unmet need. Next, the item score of the clinician was subtracted from the patient’s score: 0 = concordance, and 1 or -1 = discordance. As a explorative investigation, the present study did not focus on the nature of discordance i.e., on whether the clinician reported more care needs than the patient, or vice versa. Hence, all negative scores (= -1) were recoded into positive ones (= 1).
Predictors
Child factors
Candidate predictors at the child level were assessed as follows.
Severity of psychiatric problems. The Strength and Difficulty Questionnaire (SDQ, version parent) was used to assess the severity of mental health problems of the child from the parent’s perspective [36]. SDQ is a questionnaire that scores 33 items on a 3-point scale, in which 0 = not true, 1 = somewhat true, and 2 = certainly true [37].
Severity of internalizing problems and externalizing problems. To measure the severity of internalizing problems and externalizing problems, we used two SDQ (parent version) subscales: “internalizing problems” and “externalizing problems” [38].
Dangerous behavior towards self. To measure whether a patient currently showed dangerous behavior towards themselves, we used the MINI-KID domain “suicidal risk” (no = 0, or yes = 1) [34].
Rule-breaking behavior. The MINI-KID domains “conduct disorder” and “oppositional deviant disorder” were used to estimate rule-breaking behavior (diagnosis absent = 0, diagnosis present = 1) [34].
Age. The age of the child/adolescent was measured using the DEMOG [33].
Parent factors
Candidate predictors at the parent level were assessed as follows:
Degree of parental stress. The Parental Stress Scale was used to measure the degree of parental stress by asking primary caregivers to indicate their degree of parenting stress on a scale ranging from 1 to 10.
Severity of psychiatric problems. The Health of the Nation Outcomes Scale (HoNOS) sum score was used to measure the severity of the parent’s psychiatric problems [39]. The HoNOS consists of 12 items to be scored on a 5-point-Likert scale, ranging from 0 (no problems) to 4 (severe problems).
Family/social-context factors
Candidate predictors at the family/social-context level were assessed as follows:
Family SES. Family SES, expressed as the highest educational achieved by the parents, was measured using the DEMOG-Adult.
Growing up in a single-parent household. The DEMOG-Adult was also used to determine whether a child was growing up in single-parent or two-parent household [33].
Severity of problems with peers. The Kidscreen-27 (parent version) “friends” subscale was used to assess problems with peers as perceived by the parents. This subscale comprises spending time with friends, fun with friends, support from friends, the extent to which a child could trust his/her friends. Originally, higher item scores on Kidscreen-27 reflect better functioning, and range from 0 (= never) to 4 (= always). As we wanted to use an indicator that reflected greater severities of problems with friends as a candidate predictor, we recoded all item scores (0=4, 1=3, 2=2, 3=1, 4=0) before calculating a sum score.
Severity of problems related to school. To measure parents’ view of the severity of their child’s school problems, we used the parent version of the Kidscreen-27 subscale “school and learning,” which taps “had a good time at school,” “it went well at school,” “was able to pay attention in class,” and “quality of contact with teachers.” A sum score was for this scale was calculated similarly as for the scale regarding “problems with peers.”
Severity of child-parent discordance on mental health problems. The child and parent version of the Strength and Difficulty Questionnaire (SDQ) were used to assess the severity of child-parent discordance regarding the presence of mental health problems [36]. Higher item scores of the SQD reflect more difficulties, and range from 0 (= not true) to 2 (= certainly true). The discordance was calculated by first subtracting the parent score from the child score for each item separately, which yielded discrepancy scores for each item. We then recoded all negative scores as positive scores. Finally, we summed all discrepancy scores [37]. A higher sum score thus indicates greater discordance.
Quality of the parent-child relationship. The “parent version” of the Kidscreen-27 “family” subscale was used to assess the quality of the parent-child relationship. The “family” subscale covers 3 items: “support from parents,” “treated fairly by parents,” and “communication with parents.” Items are scored on a 5-point-Likert scale ranging from 0 (= never) to 4 (= always). To calculate the quality of the parent-child relationship, the scores of all 3 items were summed [40].
Data analysis
To analyze background characteristics, we first calculated descriptive statistics of the sample. Next, we conducted a set of univariable binary logistic regression analyses by using (i) concordance/discordance between young patients and their clinicians for each of the three outcomes variables (“mental health problems,” “information regarding diagnosis and/or treatment,” and “making and/or keeping friends”; and (ii) candidate predictors at all three levels of the Bronfenbrenner model (child / parent / family, social context). A separate regression analysis was performed for each candidate predictor (P < 0.10) [41], and yielded information on predictors at the child, parent, and family/social-context levels that predicted discordance between patient and clinician regarding the three outcomes. Our a priori hypothesis was that discordance between clinicians and children/adolescents would be predicted by predictors at child level, parent level, and family/social-context level. Since predictors at different levels may correlate despite being significant in univariable analyses, we then conducted stepwise multivariable logistic regression analyses to identify predictors at each level that were independent of other predictors, either at the same level, or at other levels. Therefore, stepwise multivariable logistic regression analyses were conducted for each of the three outcomes variables. As a first step, all child-level predictors that were significant in the set of univariable analyses were entered as possible predictors. Next, variables at parent level were entered, following by variables at family/social-context level. This step-by-step approach did not violate the statistical rule of 10 events per 1 variable [42, 43]. To test the assumptions of linearity and homoscedasticity, we generated a scatter plot of the standardized residuals [44], and tested assumptions of the logistic regression analyses for indications of multicollinearity by investigating the variance inflation factor (VIF) [41]. To measure the predictive value of models, we used the Hosmer and Lemeshow goodness-of-fit-test. Nagelkerke R2 was used to obtain an indication of the strength of the relationship between the predictor and the outcome variable [41]. All statistical analyses were performed using SPSS version 24.