Colombian Rasch validation of KIDSCREEN-27 quality of life questionnaire

Research and assessment of Health-related quality of life (HRQoL) on children have experienced substantial progress during the last few years. Some instruments are disease-specific and others are generic. In the latter group, the family of KIDSCREEN questionnaires is the most widely used [1]. Amongst its advantages are their multidimensional construct based on the World Health Organization’s definitions of health and quality of life [2], and their simultaneously development in 13 European countries, which improves their trans-cultural applicability. Previous examinations have showed that KIDSCREEN-27 exhibits good psychometric properties in terms of Rasch measurement, reliability and validity of its scores [3‐8]. KIDSCREEN questionnaires were developed for self-perceived health measures concerning physical, emotional, mental and social well-being in children and adolescents aged 8 to 18 years. There are three versions: KIDSCREEN-52, KIDSCREEN-27, and KIDSCREEN-10, available for children/adolescents self-report and for parents-proxies.

Traditionally, the construct validity of scales has been developed from the Classical Test Theory approach (CTT). However, the availability of new statistic software enables the use of more powerful and discriminative techniques like Rash analysis [9, 10]. CTT and Rasch approaches provide numerous statistical techniques to examine the psychometric quality of the items and to detect weak or biased items, but with different theoretical assumptions. CTT is based on the Spearman linear model which proposes a linear relationship between observed and expected scores and assumes that score’s variations reflect differences in measured traits among different subjects [11]. Item Response Theory (IRT) gathers a set of probabilistic models, among which Rasch analysis is the most widely used and accepted [11]. It does not assume that each item is equally difficult, and therefore, Rasch incorporates the idea that the probability of a correct response to an item is a mathematical function of person and item parameters [12]. This mathematical relationship instead of a linear function is a logistic function of the relative difference between item difficulty and respondent trait level [13].

Some advantages of Rasch analysis are that: 1) allows parametric statistical analysis, which is an advantage over CTT given that facilitates linear regression analysis in different epidemiological studies; 2) provides accurate estimations of individual measurement errors, and identifies biased and inappropriate items [14]; and 3) estimates both item and person statistics independently of the current sample of items and persons respectively [15] allowing more versatile applications of the measurements in different situations and with different populations.

The measurement of the scale properties by Rasch analysis include the assessment of the fitting of data to the model [16], and the measurement of differential item functioning (DIF), which occurs when the likelihood of answering an item correctly by individuals with the same trait level varies according to their age or sociocultural group [17]. Increasingly, Rasch analysis has demonstrated satisfactory psychometric properties of KIDSCREEN-27, as unidimensionality, the fit of person and items, and absence of DIF when comparing children and adolescentes, in both healthy and ill kids [10, 18, 19].

In Colombia, the validation of HRQoL questionnaires for children and adolescents is incipient, only the generic questionnaires KIDSCREEN-27 and recently PedsQL 4.0 (unpublished results) have CTT validations, both showing satisfactory internal consistency, content and construct validity [20, 21]. Our research team validated KIDSCREEN-27 child self-reported version in 2007 [20], and the KIDSCREEN-27 parent-proxy version in 2012 [21]. The aim of this article was to improve the information about the psychometric properties of KIDSCREEN-27 child and parent-proxy versions using the Rasch measurement model, contrasting the results with those obtained from the CTT analysis. Our hypothesis is that KIDSCREEN-27 has appropriated psychometric properties to measure HRQoL in children and adolescents in Colombia.

In this study, we evaluated the psychometric properties of KIDSCREEN-27 by Rasch analysis in two different samples (children/adolescents and parents/proxies). Overall, the results were satisfactory, with high response rates, good reliability, and acceptable item goodness-of fit in 23 of 27 items.

The aim of this study was to complement our previous classical test theory validations with a modern technique such as Rasch analysis. The Rasch analysis is a psychometric approach allowing identification of aspects relative to measurement not easily detected by traditional analysis, such as item bias and problems with the response form [13]. The Rasch model measures the intrinsic difficulty of items independently of the trait level of measured individuals. The combination of both classical and modern methods contributes to having more and better information about the psychometric properties of this scale.

This study found that in KIDSCREEN-27 child self-reported, items “How is in general your health?”, “Have you felt sad?”, “Have you felt so bad that you didn't want to do anything?” and “Have you felt lonely?” overfit the model; while in KIDSCREEN-27 parent-proxy version, only the item “Has your child had fun?” under-fit the model. Despite these misfits, when those items were eliminated no improvements on the characteristics of the scale were observed. This result is consistent with reports of Linacre affirming that statistics of fitting between 0.5-1.5 are productive for measuring, those found between 1.5-2.0 are unproductive when it comes to building a measurement, but do not distort it [26].

The validation of KIDSCREEN-27 with a sample of cancer surviving children, reported that items “Have you felt fit and well?” (0.53), “Have you been able to run?” (1.60), “Have your parents treated fairly?” (1.62) and “Have you been able to rely on your friends?” (1.51), misfit the model [10]. Authors explain that those misfits are a consequence of a small sample size, but in our case it could be related to our global assessment of the scale because the original validations and others studies with dimension analysis have reported well goodnes of-fit for all the items when Rasch analysis has been performed [18, 19].

The unidimensionality assessment showed an explained variance close to 50 %, which is an interesting finding given that in this study the scale was analyzed as a whole, without individualizing dimensions. Other studies have performed the analysis for each dimension and have found that explained variance is greater than 50 % [10, 19]. In particular, Jervaesus reports that only the Autonomy & Parent Relations dimension has an explained variance below 50 % (39.8 %), probably because that dimension is a combination of dimensions of KIDSCREEN-52: Autonomy, Relationship with parents and family life, and Economic Resources [10]. Problems related to Autonomy & Parent Relations dimension was also identified and discussed in our validations by classical test theory [20, 21], in which we found that for both KIDSCREEN-27 versions, exploratory factor analysis yielded seven dimensions, unlike the original validation, which has five. After excluding the item “How is in general your health?”, six dimensions were identified. Of these, three correlated with the original dimensions: Physical well-being, Social Support & Peers, and School Environment. But in the case of Autonomy & Parent Relations dimension, items regarding relationships with parents were separated in a different dimension of those regarding money. Confirmatory factor analysis validated the six dimensions found in the exploratory analysis (indices of the model for KIDSCREEN-27 child self-reported version were: CFI = 0.754; NFI = 0.699, and RFI = 0.662; RMSEA = 0.097, GFI = 0.754 and AGFI = 0.701; for the parent-proxy version values were CFI = 0.891, NFI = 0.867, RFI = 0.846; RMSEA = 0.057, GFI = 0.903, and AGFI = 0.88) [20, 21].

Regarding the response form, we found that five categories displayed disordered thresholds between categories 1 and 2 (1 = never/not at all, 2 = seldom/slightly); when these two categories were collapsed in one, separation, reliability, goodness of-fit, and explained variance of the model improved. This finding is consistent with other KIDSCREEN-27 validations that have suggested combining never and seldom response categories for all items in both versions [10, 19].

One of the strengths of this study was to measure DIF based on age, sex, socioeconomic condition, and health condition. So far there were limited data reporting DIF with KIDSCREEN-27 [10, 22, 33, 34]. DIF was not detected when comparing girls and boys; children (8 to 11 years) and adolescents (12 to 18 years); and ill and healthy kids (except for the item “In general, how would your child rate her/his health?” in parent-proxy version), implying that items of KIDSCREEN-27 are suitable for comparing HRQoL in these Colombian groups. On the other hand, DIF was detected in four items in KIDSCREEN-27 child self-report version and in five items in the parent-proxy version, principally was identified when comparing participants from low and high socioeconomic status. These findings deserve further qualitative research to explain how and why low-income and high-income populations understand these questions differently. DIF would represent a difficulty in the use of KIDSCREEN-27 in Colombia because it means that those items do not measure population independently of their socioeconomic status. However, it is important to consider that performing the whole analysis of the scale, instead of an analysis of dimensions, might taint DIF results since one cause of differential item functioning is the presence of multidimensionality in a test [35].

As this the first research reporting DIF on KIDSCREEN-27, it is important that prospective research assesses DIF so as to establish whether these findings are consistent or specific to the present study.

KIDSCREEN-27 is a reliable and valid scale to measure HRQoL among Colombian children and adolescents. This validation allows its use in different health contexts in Colombia, for identifying children and adolescents who are at risk for health problems, and planing and evaluating interventions to improve children and adolescents health. Comparisons of HRQoL between kids from low and high income families in Colombia should be interpreted carefully due to differential item functioning (DIF) in four items in the child self-report version, and in five items in parent-proxy version. In the case of the items “Have you had enough money as your friends?” and “Have you had enough money for your expenses?” in KIDSCREEN-27 for parents-proxies, there is a risk of measuring a construct different to the one considered in the dimension Autonomy & Parent Relations. Additionally, it does not measure HRQoL in the same way in families of high and low socioeconomic status. For these reasons, we suggest their inclusion as separate dimensions until there is more data about psychometric properties of these items.