The Malay version of the attitudes and beliefs about cardiovascular disease (ABCD-M) risk questionnaire: a translation, reliability and validation study

A cross-sectional study was conducted in Kelantan, Malaysia, involving four districts: Bachok, Machang, Tanah Merah and Pasir Puteh. The respondents were adults (aged ≥18 years) who attended the government health clinics during the study periods. The health clinics that serve the most patients in every selected district were included in this study. The respondents who came to the chosen health clinics on recruitment days with no cognitive impairment (as judged by an attending researcher) and had good Malay language command were then invited to participate in the study. The chosen respondents were not restricted to those who are free of CVD risk or any stage of CVD risk. Data collection was performed between January 2020 to March 2020. The sample size was determined for the reliability testing and factor analysis, which required at least 164 respondents for a scale with ≤4 factors with the expected factor loading (FL) of ≥0.50 and item communality of < 0.45 [16, 17]. As regards the possible dropout, we oversampled the size by 10%; hence, the minimum required sample size was 173 respondents.

However, we recruited 179 randomly selected respondents based on the attendee list from the chosen health clinics. The study protocol and rights of the participants were explained. Those who consented to participate in the study signed informed consent forms and were given questionnaire forms to be completed.

The self-administered questionnaire consists of 26 items constituting four scales or domains: (a) knowledge of stroke risk and prevention, eight items; (b) perceived risk of heart attack/stroke, eight items; (c) perceived benefits and intentions to change, seven items; and (d) healthy eating intentions, three items. For the ‘knowledge of stroke risk and prevention’ domain, the type of answer is true/false/do not know, while the rest are according to the 4-point Likert scale: 1, strongly disagree; 2, disagree; 3, agree; 4, strongly agree; and 0, non-applicable. For the English version, the average time taken to complete the questionnaire (26 items) by each participant was between 10 and 15 min. The overall score was calculated from all item scores by the domains. The higher the score, the higher the awareness and the readiness to change the behaviour.

The ABCD risk questionnaire was translated into Malay using international guidelines for cross-cultural adaptation [18‐20] to ensure the quality of the translated version and its consistency of meaning to the original version [21]. There were four translators in this study (two performed the forward translation, and another two the backward translation). First, the forward translation process (from English to Malay) was conducted by two independent translators, and the translators produced a report of the translation. After a thorough discussion, the two translations were synthesised into one document by ZMS, who also addressed any gaps or differences between the two reports.

Then, the original and translated versions of the questionnaire were given to another two different translators. Both were native Malay speakers who spoke English as their second language. The translators who received either the original or translated questionnaire version then performed the forward or backward translation, respectively. Finally, the forward and backward translation discrepancies were reconciled, and cross-cultural adaptation was made to derive the final version. As the primary purpose of the ABCD risk questionnaire is to act as a tool to evaluate the awareness of stroke and CVD risk in the general population, the adapted questionnaire in the Malay version is called Soal Selidik Kesedaran mengenai Penyakit Kardiovaskular (ABCD-M).

The translated questionnaire subsequently underwent content validity, face validity, construct validity and reliability (internal consistency and composite) assessment. Content validation aims to assess the relevancy and representativeness of each item to a specific domain by a panel of experts. This context will evaluate the relevance of all 26 items in the questionnaire to represent each domain; ‘knowledge of stroke risk and prevention’, ‘perceived risk of heart attack/stroke’, ‘perceived benefits and intentions to change’ and ‘healthy eating intentions’.

The content validation of the questionnaire was conducted by five experts (a public health physician, neurologist, family health physician, statistician, and general practitioner). They were asked to give a score from 1, item not relevant, to 4, item very relevant, based on the relevancy of the translated items in the questionnaire. The scores of 3 and 4 were recategorised as 1 (relevant), and the scores of 1 and 2 were recategorised as 0 (not relevant).

Face validation testing, which aims to assess the clarity and comprehensibility of the translated items, was conducted by 10 potential target users in the adult population [22]. The users were asked to give a score from 1, item not clear and not understandable, to 4, item very clear and understandable, based on the clarity and comprehensibility of the translated items in the questionnaire. The scores of 3 and 4 were recategorised as 1 (clear and understandable), and the scores of 1 and 2 were recategorised as 0 (not clear and understandable).

Then, the construct validity of the translated questionnaire was evaluated by estimating its association with other variables (or measures of a construct) with which it should be correlated positively, negatively or not at all [23]. Theoretically, construct validity is the degree to which an instrument measures the trait or theoretical construct that it is intended to measure [23, 24]. It is the most valuable and most challenging measure of validity. The purpose of construct validity can be obtained using the factor analysis [24, 25]. It is usually employed when the construct of interest is in several dimensions, forming different domains of a general attribute [25]. In the analysis, several items put up to measure a particular extent within a construct of interest are supposed to be highly related to one another than those measuring other dimensions. An exploratory factor analysis (EFA) was conducted utilising principal component analysis with the varimax rotation method [26]. Items loaded above 0.30 and non-problematic cross-loading were considered for further analysis. The problem is indicated by having almost comparable FLs in two or more factors, indicating that the item is not specific for a construct and general [17, 26]. Therefore, the factor analysis results will satisfy the criteria of construct validity, including both the discriminant validity (loading of at least 0.40, no cross-loading of items above 0.40) and convergent validity (eigenvalues of 1, FL of at least 0.30, items that load on posited constructs) [25, 26].

The reliability was further tested using an internal consistency measure, the Cronbach’s alpha coefficient. The test is viewed as the most appropriate measure of reliability when using Likert scales [26]. Most of the literature agreed on a minimum internal consistency coefficient of 0.70 even though no absolute rules exist for internal consistencies [26‐28].

The suggested four cut-off points for reliability included excellent reliability (0.90 and above), high reliability (0.70–0.90), moderate reliability (0.50–0.70) and low reliability (0.50 and below) [26, 29].

Next, the composite reliability (Raykov’s rho) value was estimated using the confirmatory factor analysis (CFA). It is more general in that it does not assume that the items are unidimensional and considers the correlated errors [17, 30]. Raykov’s rho values of 0.7 and above were considered as good reliability [31, 32]. The flow of translation, validation process and reliability testing are depicted in Fig. 1.

For the validation test, the content validity index (CVI) and face validity index (FVI) were computed by calculating the scale average [22, 33]. The item-level CVI (I-CVI) was the proportion of content experts giving the item a relevance rating of 3 or 4, which is calculated by dividing the agreed item by the number of experts in the study. At the same time, the CVI average was estimated by dividing the sum of the I-CVI scores by the number of items. Next, the item-level face validity index (I-FVI) was defined as the proportion of respondents giving an item a clarity and comprehension rating of 3 or 4. By measuring the average of the I-FVI scores judged by all respondents, the FVI of the questionnaire was estimated. The acceptable CVI and FVI values were at least 0.83, corresponding to the number of experts and respondents involved in this study [22, 34].

The baseline characteristics of the respondents were presented according to the type of variables: frequency (n), percentage (%) and mean ± standard deviation (SD). The factor analysis was further conducted – a multivariate statistical analysis to determine the FLs, correlations among the items, internal consistency and composite reliability. The analyses were primarily performed with RStudio IDE [35] using the lavaan and semTools packages [36, 37]. All 18 items from three domains, ‘perceived risk of heart attack/stroke’, ‘perceived benefits and intentions to change’ and ‘healthy eating intentions’, were analysed using the factor analysis. In contrast, the eight remaining items under the ‘knowledge of stroke risk and prevention’ domain could not be entered into the factor analysis because of the different scales used [13]. In the EFA, the items with FLs above 0.30 were considered acceptable [17]. Multicollinearity between the factors was identified when the factor-to-factor correlation was above 0.85 [38]. Internal consistency reliability and composite reliability were estimated, and values ≥0.7 were considered to reflect good reliability [29, 32].

A robust maximum likelihood estimator was used for the CFA as we assumed that data were not normally distributed data [38]. The model fit assessment was based on the fit indices with the respective cut-off values: chi-square (χ²) (p-value > 0.05), comparative fit index (CFI) and Tucker–Lewis fit index (TLI) of ≥0.95 (good) or ≥ 0.90 (acceptable), root mean square error of approximation (RMSEA) of ≤0.08 and standardised root mean square residual (SRMR) of ≤0.08 [38]. A model-to-model comparison was based on the Akaike information criterion (AIC) and Bayesian information criterion (BIC). The model with the lowest values of the AIC and BIC was the best fitting model for the CFA [31, 39].

This study was approved by the Medical Review and Ethical Committee (MREC) of Ministry of Health (MOH) Malaysia (NMRR-19-3296-51,864-IIR), Human Research Ethics Committee Universiti Sains Malaysia (USM), Malaysia (USM/JEPeM/19110815); and Kelantan State Health Department. Data confidentiality was maintained, and only the researchers had access to the data. The permission to conduct translation and adaptation of the ABCD risk questionnaire was obtained from the original authors.