The development and cross-national validation of the short health literacy for school-aged children (HLSAC-5) instrument

Health literacy is an asset for and indicator of adolescents’ health and wellbeing, and should therefore be monitored and addressed across countries. This study aimed to develop and validate a shorter version of the original 10-item health literacy for school-aged children instrument in a cross-national context, using data from the health behaviour in school-aged children 2017/18 survey. The data were obtained from 25 425 adolescents (aged 13 and 15 years) from seven European countries. Determination was made of the best item combination to form a shorter version of the health literacy instrument. Thereafter, the structural validity, reliability, measurement invariance, and criterion validity of the new 5-item instrument were examined. Confirmatory factor analysis showed a good model fit to the data across countries and in the total sample, confirming the structural validity (CFI = 0.995, TLI = 0.989, SRMR = 0.011, RMSEA = 0.031). The internal consistency of the instrument was at a good level across countries (α = 0.87–0.98), indicating that the instrument provided reliable scores. Configural and metric invariance was established across genders, ages, and countries. Scalar invariance was achieved for age and gender groups, but not between countries. This indicated that the factor structure of the scale was similar, but that there were differences between the countries in health literacy levels. Regarding criterion validity, structural equation modelling showed a positive association between health literacy and self-rated health in all the participating countries. The new instrument was found to be valid and reliable for the purposes of measuring health literacy among adolescents in a cross-national context.

components.We also provided instructions for classification into low, moderate, and high health literacy levels.We then examined the structural validity, reliability, measurement invariance, and criterion validity of the new instrument, here referred to as HLSAC-5.

Construction of the HLSAC-5 instrument
Firstly, regression analyses were conducted to identify the best set of five items for the short health literacy scale.The 10-item score was predicted via a 5-item score consisting of five health literacy items, constrained in such a way as to have at least one item from each core component of health literacy (Table 1).Regression analyses were conducted separately for each country and for all the countries together; this yielded a total of 32 models for each of these conditions.
Table 2 reports the adjusted amount of explained variance (R 2 ) in the country-specific and combined analyses for the 5-item instrument, in terms of predicting the original HLSAC instrument.The results show that while variation in explanatory power exists between countries, it is relatively limited.Although the best combination of health literacy items in the total sample (items HL1, HL3, HL6, HL7, and HL10) did not rank highest in every country when the countries were analysed separately, it still performed very close to the country-specific best combination.The largest difference between the country-specific best item set (explained variance) and the best total sample item set (explained variance) was 0.0153 (Belgium (Fl.)), and the smallest 0.0007 (Slovakia).Taking into account the minimal variability in explanatory power between the total sample and the best country set of items in five countries and the consistent results in two countries, the combination of HL items HL1, HL3, HL6, HL7, and HL10 was regarded as providing the best combination of items to form the short HLSAC-5 instrument (Tables 1 and 2).
When the instruments were compared, the distribution analysis showed that the short HLSAC-5 performed logically, and in the same way as the original HLSAC instrument (Fig. 1).When measured with the short HLSAC-5 instrument, the number of respondents with low health literacy was systematically slightly lower (range 0.3-2.2percentage points) in all the countries except Slovakia, as compared to the results obtained with the original HLSAC instrument.In addition, when measured by the HLSAC-5 instrument, the number of respondents with moderate health literacy decreased slightly (range 1.8-5.8percentage points), and the number of respondents with high health literacy increased (range 2.3-7.9 percentage points) in all the countries as compared to the results obtained with the original HLSAC measure.Across the total sample, the differences between Table 1.Theoretical components of health literacy and corresponding HLSAC items (10 items in HLSAC, five items in HLSAC-5).The five items selected for HLSAC-5 are marked in bold.

Structural validity, item validity, and reliability of the HLSAC-5 instrument
Means, variances, and correlations for the five health literacy items included in HLSAC-5 are presented in Table 3.
All the health literacy items were moderately correlated (with correlations varying between 0.36 and 0.48).
CFAs were conducted in order to confirm the structural validity of the HLSAC-5 (consisting of five items selected on the basis of the regression analyses described above).In the CFA, the five items were set to load on one factor.The CFA models were run separately for each country.The results showed that in the total sample and in all the countries, a one-factor model had a good model fit (min.CFI and TLI = 0.979 and 0.958; max.RMSEA and SRMR = 0.047 and 0.020) (Table 4).
In the total sample, standardized factor loadings ranged from 0.59 to 0.72.Finland had the highest factor loadings (λ = 0.79-0.87)and Poland the lowest (λ = 0.48-0.64)(Table 5).The lowest standardized factor loading, 0.48, was in Poland, for item HL1 ('I have good information about health').Item-level reliability coefficients (R 2 ) ranged from 0.23 (Poland, Item HL1: 'I have good information about health') to 0.76 (Finland, Item HL7: 'I find health-related information that is easy for me to understand'), suggesting that the latent factor explained

Measurement invariance of the HLSAC-5 instrument across genders, ages, and countries
The analysis of configural and metric invariance showed that constraining the factor loadings to be equal across countries, age groups, and genders did not substantially decrease the model fit (Country invariance: ΔCFI = 0.001, ΔRMSEA = 0.005; Age invariance: ΔCFI = 0.000, ΔRMSEA = 0.002; Gender invariance: ΔCFI = 0.000, ΔRMSEA = 0.001), indicating that the factor structure was comparable across countries, age groups, and genders (Table 6).
In assessing for scalar invariance, the intercepts of the five items were constrained to be equal across all groups.ΔCFI and ΔRMSEA values showed scalar invariance for age and gender, but not perfectly for country.These results indicated that for both age groups and both genders, the factor structure and item levels did not differ.However, there were differences between countries, indicating that while the factor structure was similar, there were health literacy level differences between the countries (Table 6).

Criterion validity
In the total/pooled sample, there was a positive association between health literacy and self-rated health, indicating that higher health literacy was associated with higher self-rated health.In line with the pooled sample, the analysis by country showed a positive association between health literacy and self-rated health in all seven Table 3. Spearman correlation coefficients between the items in the HLSAC-5; also item means, standard deviations, skewnesses, and kurtoses for the total sample.***Spearman's rank correlation significant at the 0.001 level.).Health literacy explained self-rated health to some extent, ranging from 2% in Germany to 10% in Belgium (Fl.) and Finland (Fig. 2).

Discussion
The aim was to construct and validate a brief, theoretically comprehensive, and internationally comparable instrument for measuring children's and adolescents' subjective health literacy.The cross-national development and validation process encompassed seven European countries and was based on the original 10-item HLSAC instrument 21 .The original HLSAC instrument has been validated in several studies 9,[32][33][34][35][36][37][38][39][40][41] and has proven to be an appropriate tool for measuring subjective health literacy.However, for large-scale studies where the purpose is to explore the relationship between health literacy and other phenomena, there is a need for shorter instruments.The research resulted in a 5-item HLSAC-5 instrument containing one item from each of the health literacy theoretical core components.For the total sample, the HLSAC-5 instrument had good predictive properties, predicting 90% of the variance of the original 10-item HLSAC instrument.The HLSAC-5 was constructed as a one-factor model, similar to the original longer HLSAC instrument.All five items were derived from a conceptual framework; they contribute substantially to the underlying construct of health literacy, addressing the recommended attributes for such a measure 42,43 .The essential advantage of the one-factor model is that it does not violate the requirement on additivity, and enables reliable calculation and interpretation of sum scores.It thus differs from multifactorial health literacy scales, which consist of several subscales 44,45 .Although the HLSAC-5 instrument is based on five theoretical core components, it is possible to construct a one-factor model, because background theory and previous studies have shown that the core components are partly overlapping (having cross-correlations) and exist in a somewhat hierarchical relationship with each other 3,21 .The challenge in constructing a short, comprehensive, and generic health literacy instrument lies in the need to adequately take into account the complex and multidimensional nature of health literacy.Interpretative clarity of the model is essential, meaning that one must go beyond modifying every detail in pursuit of statistical adequacy (thus seeking a balance between content validity and structural validity).As there was only a small difference in explanatory power between the country-specific models and the total sample model, the items selected for the final model can be considered to form a sound and content-valid measure of health literacy, regardless of the country in question.
There was some variation between countries regarding the items measuring theoretical knowledge, critical thinking, and citizenship.The items finally selected for the model were 'I have good information about health' , 'I can compare health-related information from different sources' and 'I can judge how my own actions affect the surrounding natural environment' .Note that regarding the items that were not selected for the model, the respondents had to assess their ability to give examples on how to promote health (theoretical knowledge), decide whether health information is right or wrong (critical thinking), and give ideas on how to promote health in the person's immediate surroundings, such as the nearby place or family (citizenship).These items were selected for the country-specific model in some countries, but not in others.One reason for this may be that young people from different countries may perceive their skills in the tasks required to produce differently.Items measuring practical knowledge and self-awareness ('When necessary, I find health-related information that is easy for me to understand' and 'I can judge how my own actions affect the surrounding natural environment') were found appropriate for the model in almost every country surveyed.These items may require less interpretation and be slightly less cognitively challenging than their alternative items, and they could therefore be selected for the model in a relatively unequivocal way.
For descriptive purposes, the health literacy levels were classified into three groups (low, moderate, high) based on cut-off scores.Inspection of the response distributions showed that the short HLSAC-5 instrument performed logically in the classification of health literacy levels, i.e. in the same way as the original HLSAC instrument.However, there were small differences between the two scales; thus, when HLSAC-5 was used the proportion of respondents with low and moderate health literacy was slightly smaller, and the proportion of respondents with high health literacy somewhat greater.This may be because the selected smaller number of items approached health literacy in a more limited way, thus placing greater emphasis on the single competency as expressed in a single item.
Examination of the structural validity provides evidence on the construct validity of the instrument 46,47 .The structural validity of the HLSAC-5 was analysed via CFA, i.e. with the model-fit parameters indicating how the data fitted into the assumed factor structure.This kind of structural validation is a strong method to examine the validity of an instrument when (as was the case in this study) the instrument structure is specified a priori, and when the contextualization is based on evidence from previous research 47 .The HLSAC-5 instrument's overall goodness and sufficiency were at a good level, and the instrument showed an excellent fit with the data in all the participating countries.This was a notable result, especially considering the large amount of data involved 39,48,49 .
The reliability of HLSAC-5 was inspected.Both item reliability and scale reliability were at an adequate level.The factor loadings and reliability coefficients (R 2 ) indicated that the items were reliable measures of a latent variable (health literacy).The internal consistency reliability varied to some extent between countries, but was at a high level in the total sample, exhibiting a Cronbach alpha of 0.80.This means that the items measured the same construct.The high α-value is also noteworthy, bearing in mind that the number of items included in the instrument has an impact on the value of Cronbach's alpha coefficient, with a low number of items reducing the α-value 50 .Despite this, the HLSAC-5 instrument with five items demonstrated a high level of internal consistency.
Regarding measurement invariance, configural and metric invariance was established across genders, ages, and countries.Scalar invariance was achieved for age and gender groups, but not between countries.This indicated that the structure of the instrument was similar across countries and that it properly measured the same phenomenon in each country, but that there were differences in the levels of health literacy between countries.The establishment of configural, metric, and partial scalar invariance, together with the excellent fit of the model to the data, indicates that partial invariance holds, and provides a sufficient condition for comparing mean values between countries 51,52 .This can be regarded as a good result, considering the number of countries involved and the size of the sample.One should note the well-known difficulty of achieving full measurement invariance in most empirical studies [53][54][55] ; in particular, scalar invariance has been described as an unachievable ideal that can only be approximated in practice 53 .There can be several reasons why full measurement invariance is not often met in large cross-national studies such as the present study.The original HLSAC instrument, which was used as a starting point for the study, was in the English language, with subsequent translation into the language of the target countries (translation-back translation).Even if the translation process was carried out to a high standard, minor differences of interpretation may remain, and together with possible cultural differences, the meanings of the response options or the items may involve slightly different connotations.There may also be other explanations: for example, participants' understanding of the content of the items can vary, as may familiarity of the participants with the response format, or the extent to which respondents give socially acceptable answers 56,57 .
There is no gold standard scale for measuring children's and adolescents' health literacy, and hence no absolute criterion for assessing the validity of the HLSAC-5 instrument.Here, criterion validity was assessed in relation to self-rated health, in line with the notion of defining 'the extent to which a construct [here, health literacy] relates to another construct that it should theoretically be related to' 20 .As measured with the HLSAC-5 instrument, health literacy was associated with differences in self-rated health in each participant country.On average, the short instrument explained 7% of the variance in self-rated health, which is similar to the results on the original 10-item HLSAC instrument 22 .Paakkari et al. 22 discuss what coefficient of determination would be sufficient to indicate an important or critical factor for self-rated health.In line with DeSalvo et al. 24 , they emphasize that selfrated health is a robust predictor of mortality, and argue that 'any factor that contributes to a decrease in health disparities is important, including health literacy' .It should also be noted that according to previous studies, health literacy explains more of adolescents' self-rated health than, for example, family affluence, gender, age, school achievement, or educational orientation 8,22 , which are often considered to be valid explanatory factors.
This study was based on a previously-developed HLSAC instrument.This limited the choice of items and precluded the possibility of developing a completely new health literacy instrument.It should also be noted that the conceptual framework behind the HLSAC-5 tool is only one possible way to conceptualize health literacy.In using the instrument, it will be important to know the framework that it is based on, the purposes and contexts appropriate to its use, and the advantages or limitations thus implied.Even a good health literacy instrument can give biased results if used in the wrong context.The HLSAC-5 instrument is by nature a comprehensive and generic tool, based on a relatively broad notion of health literacy.It can provide a good overview of health literacy, but domain-specific instruments may be capable of providing a more focused picture of health literacy in a specific context.Note also that the current HLSAC-5 instrument has been validated for 13-and 15-year-olds; hence, further research is needed on the applicability of the instrument to both younger and older age groups.
Health literacy has been shown to be a relevant indicator of and contributor to adolescents' health and wellbeing, highlighting the need to research health literacy across countries, and to develop appropriate measurement tools for this purpose.The present study indicated that HLSAC-5 is a valid and reliable instrument for use in a cross-national context.Its brevity also makes it applicable as a component of large-scale surveys in which a range of phenomena are examined at the same time.

Health literacy
The HLSAC instrument 21 was used to measure the adolescents' subjective (self-reported, perceived) health literacy.The validated 10-item instrument contains two items from each of five previously-identified core components, namely theoretical knowledge, practical knowledge, critical thinking, self-awareness, citizenship 3 .Respondents evaluated the 10 items starting with 'I am confident that…' on a 4-point scale (1 = not at all true, 2 = not completely true, 3 = somewhat true, 4 = absolutely true).
Self-rated health Self-rated health was evaluated by a single question measuring the individual's evaluation of their health and having four response options (1 = excellent, 2 = good, 3 = fair, 4 = poor) 62 .For the purposes of CFA the scale was reversed to have higher values indicating higher self-rated health (reversed values: 1 = poor, 2 = fair, 3 = good, 4 = excellent).

Demographic characteristics
Gender (1 = girl, 2 = boy) was self-reported.Age was computed based on the respondent's month and year of birth, and the date of the survey assessment.According to the HBSC protocol, respondents are assigned to three age categories, i.e. as being of ages 11 (≥ 10.5 and ≤ 12.5), 13 (> 12.5 and ≤ 14.5), and 15 (> 14.5 and ≤ 15.5).The age categories 13 and 15 years were used in this study.

Data analysis
First of all, regression analysis was used to determine the combination of five health literacy items that best predicted the original HLSAC instrument (consisting of 10 items).In the regression analyses, all the different sets of five items were tested for their predictive ability with regard to the original HLSAC.In line with the theoretical basis of the HLSAC 21 , all the 5-item sets were constructed in such a way as to contain one item from each theoretical core component included in the original HLSAC instrument.The rationale for choosing regression analysis as a tool to determine the items for the new HLSAC-5 measure was, in part, because we wanted to stay true to the original scale as far as reasonably possible, and hence achieve an explanatory power as close as possible to that of the original scale.We further theorized that closeness to the original scale would yield similar characteristics and associations to other variables (notably self-rated health) as the original HLSAC.Regression analysis was deemed a suitable tool, since it would allow us to use theoretical reasoning in choosing the restrictions that had to be in place for the items in the new HLSAC-5 measure.Only 32 models met the condition that there should be one item from each of the five core components.In total, 32 regression models were fitted for each country in the dataset, as were 32 models for all the countries together.In all the fitted models the original 10-item HLSAC instrument was considered to be the response variable, and the candidate short HLSAC instrument the predictor.The fitted models were ranked by their adjusted R 2 values, and the model with the highest coefficient of determination was regarded as comprising the best 5-item HLSAC instrument in terms of predicting the original HLSAC instrument.
To classify the responses into low, moderate, and high health literacy levels, sum scores were used.In the 10-item HLSAC measure, the classifications were built in the following manner: low (score 10-26), moderate (score 27-35), and high (score 36-40) 63 .For the 5-item HLSAC-5 measure, the thresholds for the levels were set by adopting a similar line of reasoning as for the longer measure: low (score 5-12), moderate (score 13-17), and high (score 18-20).
Structural validity was used to assess the extent to which the five selected items reflected the underlying dimension, i.e. health literacy.The structural validity was examined by confirmatory factor analysis (CFA).In the one-factor model, standardized factor loadings indicate direct structural relations between the latent factor and the item 64 .The CFA model fit was evaluated using the χ 2 -test and the following fit-indices: Tucker Lewis index (TLI), comparative fit index (CFI), standardized root mean square residual (SRMR), and RMSEA 34 .For TLI and CFI, values equal to or above 0.95 represent a good fit 48 .The RMSEA and SRMR values should be less than 0.06 and 0.08, respectively 48 .
The reliability of the five items was estimated by reliability coefficients (R 2 ), and the internal consistency of the scale by Cronbach's α.An alpha higher than 0.70 is regarded as indicating good internal consistency 65 .
Measurement invariance across gender, age, and country was determined through a multi-group analysis.We tested whether the underlying factor structure was consistent, regardless of whether the scale was used with girls or boys, with 13-or 15-year-olds, or with different countries.The models were first estimated for subgroups independently.The equivalence of the factor structure parameters was then tested in a hierarchical order (for configural, metric, and scalar invariance) as suggested by Byrne 66 .Configural invariance, with no restrictions, was first estimated.The metric invariance was tested by constraining the factor loadings to be equal across groups.Scalar invariance was examined by constraining the item intercepts to be equal across groups.Measurement invariance was established when the restrictions decreased CFI by not more than 0.010 and increased RMSEA by not more than 0.015, relative to the prior model 67,68 .
Criterion validity refers to the extent to which a construct relates to another construct that it should theoretically be related to Ref. 20 .As better self-rated health has been found to be moderately associated with higher health literacy in numerous studies 8,22,[25][26][27][28] , a positive association between these two factors was used as an indicator of appropriate criterion validity.In this study, structural equation modelling was used to analyse the association between health literacy and self-rated health; this was done by combining CFA and regression analysis within the same model.

find health-related information that is easy for me to understand
I can follow the instructions given to me by healthcare personnel (e.g.nurse, doctor) Critical thinking HL3 HL9 I can compare health-related information from different sources I can usually figure out if some health information is right or wrong

can give reasons for choices I make regarding my health
I can judge how my own behaviour affects my health Citizenship HL6 HL2 I

can judge how my own actions affect the surrounding natural environment When
necessary, I am able to give ideas on how to improve health in my immediate surroundings (e.g. a nearby place or area, family, friends)

Table 2 .
Adjusted amount of explained variance for the best 5-item HLSAC combinations in predicting the original 10-item HLSAC instrument.

Table 4 .
Model fit indices for the total sample and for each country.The p value is calculated from the Chisquare goodness-of-fit test.

Table 6 .
Measurement invariance by country, age, and gender.Results of configural, metric, and scalar invariance.