Modification and psychometric evaluation of the child perceptions questionnaire (CPQ_11–14) in assessing oral health related quality of life among Lithuanian children

The study followed a cross-sectional design and targeted adolescents aged 11 to 14 years. A number of 323 respondents was calculated to be sufficient for assessment of the prevalence of orthodontic anomalies in children, hypothesizing their prevalence to be 30% with 95% confidence interval from 25 to 35% [31]. The hypothesized prevalence of orthodontic anomalies was in agreement with our study in Lithuania among schoolchildren aged 11–15 years [32]. With regard to factorial analysis, this number of respondents is also sufficient, as the sample size should be at least 5 × k, where k is the number of items in factorial analysis (in our study k = 37) [33]. Then, accounting for anticipated non-response and participants who do not meet age requirements (11 to 14 years), the primarily sample size was increased in 50% (480 persons). The study sample was comprised of students from general education governmental schools in Lithuania. A list of schools and number of students by class was obtained from the education management information system of the Lithuanian Centre of Information Technologies in Education.

A two-stage cluster sampling method was used to draw a representative sample of students. In the first stage, 16 schools were randomly selected, ensuring equal presentation of urban and rural administrative regions. In the second stage, classes of students attending grades six to nine were chosen with a probability proportionate to the number of students in school. Thus, the primary sampling unit (cluster) was a class. In total, 25 classes were selected with estimated a required total number of students. Although the primary sampling unit was the class, a clustered sampling design effect was not accounted for either in the sample size calculation, nor in the analyses of data, as it was considered that students within the same class were similar to each other in oral health behavior, though they may not be as similar as in other health behavior patterns (e.g. smoking or bullying [34]).

School authorities were contacted by researchers and informed about all aspects of the study. Class tutors of selected classes were instructed about the process of carrying out the survey among students and their parents. They sent a description of the study, asking for written informed consent that their child be allowed to participate in the study to students’ parents. Of the 463 parents who initially received a request for written informed consent, positive answers were obtained from 393 (85%) of them. Those parents who gave consent were also asked to complete a questionnaire about their child’s oral health and well-being. The number of correctly completed parents’ questionnaires was 315 (68% of parents who were initially contacted). The students’ questionnaires were administrated in school classrooms. Eligible participants could freely choose to participate or not in the survey. Measures of anonymity and confidentiality were ensured. Respondents did not write their name in questionnaires, and upon completing the questions, they sealed the questionnaires in provided envelopes. A total of 381 students correctly completed the questionnaires, but 74 were excluded from the present analysis because of the students’ age criterion.

The final number of students, aged 11 to 14 years, whose questionnaires were used in the present study, was 307 (66% of initial sample size). Maintaining the same fieldwork methods, the data were collected during the 2013/2014 school year (n = 179) and in 2016/2017 school year (n = 128). The number of questionnaires completed both by the child and parents was 255. The gathering of questionnaires was ensured by using codes.

Self-reported data were collected from 307 children aged 11 to 14 years. The mean age of the children was 13.27 years (standard devition (SD) = 1.01), and 128 (41.7%) were boys. The respondents represented all social groups by family affluence levels: 32 (10.7%) of low, 105 (35.1%) of middle and 162 (54.2%) of high (8 missing cases) level; and by residence area: 221 (72.2%) of urban and 86 (27.8%) of rural residence.

In our study, a single item on the pain in teeth, lips, jaws or mouth from the OS domain was substituted by a series of five sub-items identifying a location of oral pain. The distribution of respondents’ answers to these sub-items is presented in Table 2. Responses provided a variety of locations of oral pain, although pain in teeth seemed to be the most frequent complaint. The new item that combined responses to all sub-items showed that 67.3% of respondents have experienced pain in teeth, lips, jaws or mouth at least one or two times in the prior three months. In further analyses, this variable was used as the original item OS1 “Pain in teeth, lips, jaws or mouth”.

The response rate to the items of the CPQ ranged from 97.1 to 100% with the highest rate of blanks (2.9%) in responses to item S10 “Argued with other children or your family because of your teeth or mouth” (SWB domain).

The impacts, that is the items scored from 1 (‘1 or 2 times’) to 4 (‘everyday or almost everyday’), were reported most frequently in the OS domain (“Pain in teeth, lips, jaws or mouth” – 67.3%; “Food stuck in or between teeth” – 59.6%; “Bleeding gums” – 56.8%; “Bad breath” – 47.9%) and in the EWB domain (“Worried that he/she is not as good looking as others” – 42.4%; “Worried that he/she is not as healthy as others” – 32.4%).

Descriptive statistics of the total CPQ and its health domains are presented in Table 3. Sum scores were found to be highly skewed and not normally distributed in all the health domains with a very noticeable floor effect, especially in the SWB domain. Out of the theoretical range of 0–100% of transformed scores, their mean (except OS domain) and median did not exceed 20%.

Assessments of internal consistency reliability of the CPQ and its domains are displayed in Table 4. Cronbach’s alpha for the total CPQ was 0.875. For the domains, the coefficient ranged from 0.645 for OS domain to 0.862 for SWB domain, indicating from an acceptable to good internal consistency reliability. Similar assessments were found for the ICC.

All items of the OS and EWB subscales were found to be significantly inter-item correlated, while this was not achieved for the items of FL and SWB subscales (data not shown). The item-total correlations were significant at a 0.01 level for all domains, as well as for the total CPQ (see Table 1). These results, in combination with Cronbach’s alpha if item deleted and a 1-factor solution from the EFA, ranked the items by their impact to the total variance of the domain and indicate which items have the weakest impact within the domain and, consequently, could be removed from the corresponding domain. The items “Sores in mouth” (O3), “Irritable or frustrated” (E1), “Unsure of himself” (E2), “Difficulty playing a musical instrument such as a recorder, flute, clarinet, trumpet” (S8), and “Teased or called names by other children” (S11) are examples of such kind (see Table 1).

Table 5 displays the correlation between scale sum scores and overall ratings of oral health and well-being, as well as with global life satisfaction. All domains and total CPQ were found to be significantly (p < 0.01) and positively correlated with oral health and oral well-being. The correlations between the global life satisfaction and the domains were all significant (a negative correlation value indicates that higher life satisfaction is related to lower rating of oral problems).

Discriminant validity of the instrument was tested by assessing CPQ scores by gender, family affluence, self-reported caries experience and malocclusion traits (Table 6). A significant gender difference in the EWB domain, as well as in the total CPQ revealed that there was poorer emotional well-being among girls. Across the groups of adolescents by family affluence level, a significant gradient within the family affluence groups was observed overall as well as in EWB and SWB domains. Adolescents who subjectively reported experience with caries in comparison with their ‘healthy’ contemporaries indicated significantly greater scores in the OS, FL and EWB domains but not in the SWB domain. In parallel with this, self-reported malocclusion also indicated a negative impact on the overall OHRQoL, but was significant within the OS and EWB domains only.

It was possible to compare 255 pairs of OS and FL as a sum score rated by children and their parents (Table 7). Positive significant correlations, which obtained value of a moderate level, were observed for sum scores of both domains. The moderate values of kappa and ICC also confirmed agreement between child and parental reports. These results suggest reliability of two subscales of the CPQ in respect of agreement between two different raters.

Table 8 presents the factor structure of each domain of the CPQ obtained from the EFA. The appropriateness of these factor models was evaluated by Bartlett’s test of sphericity (p < 0.001 for all four domains) and KMO measure of sampling adequacy (it ranged from 0.666 in FL domain to 0.862 in EWB domain). The analysis revealed a complex factor structure in all domains of the CPQ. The estimated loadings indicate that the domain of oral symptoms (OS) includes two factors that explained 38.4 and 16.6% of the total variance correspondingly. Factor 1 combined four items (“Bleeding gums” (O2), “Food caught in or between teeth” (O5), “Bad breath” (O4), and “Food stuck to roof of mouth” (O6)), while the factor 2 combined two items (“Mouth sores” (O3), and “Pain in teeth, lips, jaws or mouth” (O1)) (items are listed by loading weights). Analysis of the FL domain showed that at least two factors could be extracted, which explain 31.0 and 16.1% of the total variance correspondingly. Factor 1 combined six items on limitations of the eating function (F5, F8, F7, F9, F4, F2), and factor 2 combined breathing (F1), sleeping (F3) and speaking (F6) disorders. The items of the EWB domain were split explicitly into two factors that explained 45.2 and 20.5% of the total variance correspondingly. Factor 1 combined five items (E4, E5, E9, E8, E6) that describe adolescent emotions, while factor 2 combined the remaining four items (E1, E2, E7,E3) concerned with adolescent personality. Items of the SWB domain showed a three factor structure, explaining 40.2, 14.2 and 8.5% of the total variance by the corresponding factors. Factor 1 combined six items (S10, S7, S12, S9, S13, S6) specifying the adolescent’s isolation from his/her peers; factor 2 combined five items (S2, S3, S8, S1, S4) on adolescent difficulties in school work and out-of-school activity; and factor three combined two items (S11, S5) that measured disorders in adolescent relations with other children due to his/her oral problems.

A complex factor structure of the CPQ domains was also seen from noticeable loadings of several items (F2, F4, E3, S4, S6); therefore, they might be attributed to more than one specified factor. Further, the dimensionality of the CPQ domains was assessed employing CFA.

We performed CFA on each of four domains (OS, FL, EWB and SWB). Table 9 reports the goodness-of-fit statistics for the final models of each domain.

Among four domains, the OS domain model had the best fit estimations, unless its internal consistency reliability (Cronbach’s alpha) was the lowest. Its goodness-of-fit indices showed excellent model fitting to the given data including relatively great p-value, which is not uncommon for such a sample size. As presented in Fig. 1a, the model includes two latent variables (factors). In accordance with the EFA solution, the items “Bleeding gums” (O2), “Food caught in or between teeth” (O5), “Bad breath” (O4) and “Food stuck to roof of mouth” (O6) went to factor one and the items “Pain in teeth, lips, jaws or mouth” (O1) and “Mouth sores” (O3) went to factor two. The item “Pain in teeth, lips, jaws or mouth” (O1) that was modified in our study had a significant positive impact on the factor two. Both factors were significantly correlated (r = 0.68, p < 0.001).

The model of the FL domain had acceptable fitting to the concerned data. It was in accordance with the two factors structure that was revealed in the EFA. Results shown in Fig. 1b illustrate a positive association between selected factors (r = 0.25, p < 0.001), unless they reflect different aspects of life quality related to limitations of oral functions. Both EFA and CFA confirmed low effect of the functional limitation “Taken longer than others to eat a meal” (F2).

The model of the EWB domain had excellent fitting indices and confirmed a two factor structure, as well as being revealed in the EFA (Fig. 1c). There was a positive correlation between factors (r = 0.33, p < 0.001). The item “Shy and embarrassed” (E3) provided unspecified position in factorial structure.

The model of the SWB, which included three latent variables, was checked. As a result, this model had a complex structure and was not well-fitted to the research data (Fig. 1d).

This study analysed data that were collected in cross-sectional population survey of young people samples but not within samples of patients attending dental treatment as in several studies [11, 45, 61, 62]. The children completed their questionnaires at school anonymously without any influence of their parents‘opinion, thus, children could express their own feelings towards their QoL that is an important condition comparing children’s and their parents responses. Employment of a CFA in testing of the CPQ instrument reliability that is unusual by most of the previous cross-cultural validation studies of CPQ_11–14 also can to be considered as a novelty and strength of the study.

There are several limitations in the current study. First and foremost, in the present study the CPQ sum scores were compared with the self-reported caries experience and malocclusion traits/orthodontic treatment need but not with clinical dental examination outcomes. The literature also shows that several studies did not reveal any effect of caries and malocclusion on the OHRQoL [18‐20, 22, 23, 46]. Moreover, although our modification in the studied instrument helped to identify the kind of oral pain but the discriminant validation on the instrument was limited to caries and malocclusion. We believe that these comparisons are worth of greater attention, so their analysis will be an objective for another study. Next, given the aim of the present study, we worked on the “long form” (37 items) of the original CPQ_11–14 together with other scales, including such as eating behaviour and self-esteem. Practical guides to develop measurement scales show that an increase of number of items in the questionnaire may affect respondent’s accuracy, especially for children, providing inaccurate answers, which may, consequently, reduce reliability of the tested scale [30]. Moreover, we added five sub-items for the first OS item, increasing the number of items in the questionnaire. We believe that this type of change would be better applied to the short-form CPQ, and further research should confirm this. Another important consideration is the age limitation (from 11 to 14 years old) of children to whom the instrument is addressed. Jokovic et al. [11] discussed the role of child’s cognitive abilities on self-report health status and suggested to be used age-specific questionnaires for children younger than 11-year-old. Therefore, we raise the hypothesis that the CPQ_11–14 measure in older adolescent group (e.g. aged 16–18 years) is more reliable as it is in 11–14-year-olds, however, to date we haven’t found any studies confirming this hypothesis [63]. Finally, in the present study, test-retest reliability of the CPQ instrument was replaced with an alternative analysis that included comparison of children’s and their parents’ responses to the same questions of the OS and FL domains. This approach is not free from limitations, especially in relation to its accuracy because children and parents may not share the same views about illness and health [11].