Development and validation of a structured survey questionnaire on knowledge, attitude, preventive practice, and treatment-seeking behaviour regarding dengue among the resident population of Sabah, Malaysia: an exploratory factor analysis

The survey study was approved by the Universiti Malaysia Sabah ethics committee [Ethics reference number: (53) dlm.JKN (SB)100-13] and complied with the Declaration of Helsinki. Permission to interview the respondents was obtained from the district police station [KPB Kota Kinabalu 10/8/11]. All respondents included in this study gave written informed consent before participation. Those individuals who participated remotely by completing an online survey form (Google) provided informed consent form immediately prior to clicking the “next page” button to start the survey.

The survey study was designed as a cross-sectional measurement to develop and validate a questionnaire on KAP and treatment-seeking behaviour regarding dengue among the Sabah resident population. It was divided into a pilot study and a validation study involving 30 and 468 respondents, respectively. The respondent inclusion criteria included being 18 years of age or older, the ability to read Sabah Malay creole and not being a health professional. The sample size required for verification of the survey was based on the ratio of the number of items in a questionnaire to the number of respondents, with 1:10 as a preferred rule-of-thumb [16‐18]. A sample size of 380 was required for exploratory factor analysis (EFA) whenever ten or more items were expected to have a factor loading of 0.4 [19]. The sample size was inflated to 494 in order to account for an anticipated 30% non-response rate. The required sample size for two-parameter logistic item response theory (2-PL IRT) followed that for EFA because there is no definitive size for IRT, although it may range from 100 to 500 samples [20].

A thorough literature review was conducted to collate and review existing questionnaires related to KAP and treatment-seeking behaviour on dengue. An assessment of the level of difficulty and adequacy of the questionnaire was performed by a panel of experts comprising an epidemiologist, infectious disease specialists, a statistician, and a biomedical scientist. From this process, an initial selection of 110 items was reduced to a total of 97. The items chosen for the questionnaire were adjusted according to the local setting and divided into four primary sections, as shown in Table 1. These were determined as follows: (1) Demographic factors (six items) examining demographic and socio-economic characteristics; (2) Knowledge (44 items), defined as respondents’ knowledge of dengue prevention, including possible causes of dengue recurrence, symptoms, transmission, treatment and prevention, requirement for dengue precautions and recommended/non-recommended practices. This section consisted of true/false response options with a “don’t know” category provided for each question. A correct answer (such as ‘mosquito’ as the transmission vector of dengue) received one point whereas an incorrect answer (such as ‘Aedes aegypti habitually bites at midday’) and a “don’t know” response each received zero points; (3) Attitude (15 items), defined as respondents’ opinions about dengue prevention, awareness, daily care and socio-cultural perspective. This section consisted of agree/disagree response options with a “not sure” category provided for each question. Responses deemed to be either good or acceptable were awarded one point each, while all other responses received zero points; (4) Practice (18 items), defined as respondents’ practice towards dengue prevention, such as action taken to avoid occurrence of infection. This section consisted of yes/ no response options with “not sure” and “not related” categories provided for each question. A point score for each respondent was determined based on their good practice, with none awarded for poor practice. A “do not have” answer did not count towards the cumulative score and the total score (denominator) was commensurately reduced. A secondary section of the questionnaire assessed treatment-seeking behaviour of respondents and consisted of a final five items. The questions used are included in an Additional file 1.

After assessment of the content validity by the expert panel, the pilot questionnaire was pre-tested on 30 respondents recruited randomly based on local acquaintances of the researchers (conducted in September 2019). The selection criteria included Sabahan (residents of Sabah) aged 18 years old and above, no known history of dengue infection, literate in Sabah Malay creole and working in non-medical-related professions. Respondents were recruited either physically or remotely. The former individuals were required to complete the questionnaire by hand (self-administered) on a hard copy, while the latter individuals did so online by using a Google survey form. For the self-administration method, a trained research assistant was assigned to check all items on-site before ensuring the return of each completed questionnaire to the researchers. The response rate was 96.67% (29/30 participants). Respondents took an average of 15 min to complete the questionnaire.

Based on the findings of the pilot study, a full validation study using IRT and EFA was conducted from October 2019 to February 2020. Respondents were recruited as described for the pilot study, either by convenience sampling, undertaken in Kota Kinabalu, the state capital of Sabah (self-administration), or through an online Google survey. Remote respondents came from all five divisions of Sabah state: Interior, Kudat, Sandakan, Tawau and West Coast. The timeline of questionnaire development and validation is shown in Fig. 1.

Data analysis was performed using SPSS Statistics for Windows (version 24.0) (IBM Corp., released 2011) and R software (version 3.3.2) in the R Studio environment. As the knowledge section of the questionnaire consisted of unidimensional items with dichotomous responses (correct/incorrect), it was analysed by 2-PL IRT using ltm package version 1.0 [21]. The metrics used to assess knowledge items were difficulty index (proportion of respondents who answered the item correctly) and discrimination index (how the ‘overall best’ respondents compared to the ‘overall worst’ respondents with regard to a particular item). A difficulty index in the range of − 3 to 3 and a discrimination index in the range of 0.35 to 2.5 were considered acceptable [22]. The Chi-square goodness-of-fit test determined if the observed frequencies fitted a specified distribution for each item [23]. IRT analysis makes a strong assumption of unidimensionality, a property that is used to determine whether items are suitable to be summed as a total score. Modified parallel analysis determined unidimensionality, with P-value ≥ 0.05 indicating that the assumption was met [24].

Attitude, practice and treatment-seeking behaviour were analysed by EFA using psy package version 1.8.4 to assess the construct validity and reliability of items in each section [25]. These were treated as continuous responses to allow evaluation of the dimensionality (number of factors) of items. The Kaiser-Meyer-Olkin (KMO) test measured the relative degree of correlation for a set of items in the questionnaire. A KMO value of > 0.7 indicated the adequacy and suitability of data for factor analysis. Communality (h2) is the proportion of variance that is shared among a set of items, for which an h2 value > 0.3 was considered acceptable. The principal axis factoring extraction method with oblimin rotation was applied in the EFA. Eigenvalues > 1.0 and a scree plot inspection determined the number of factors extracted. Factor loadings of > 0.4 were considered acceptable [26]. For indicators of internal consistency, a corrected item-total correlation of > 0.5 and a Cronbach’s alpha score > 0.7 were considered reliable [27].

The demographic characteristics of the 468 respondents who participated in this study are shown in Table 1. The most common age range of participants was 18–30 years old (41.2%) with a mean age of 28.43 (± 9.54) years. Interestingly, most were Kadazan-Dusun (62.3%), an ethnic group indigenous to Sabah, and predominantly females (60.5%). A high proportion of respondents received tertiary education (68.6%), while 26.5% completed secondary education.

Of a total of 44 items several were considered problematic after analysing by IRT (Table 2). Pertaining to the difficulty parameter only 31 items were within or close to the acceptable range of −3 to + 3. Two items (K1.9 and K6.8) each exceeded the cut-off value of + 3 by a large margin, indicating that the items were too difficult. Additionally, one item (K2) was, by 2.5, slightly above the cut-off margin. Eleven items (K1.1, K1.4, K1.12, K1.18, K3.3, K4.2, K6.1, K6.2, K6.4, K6.5 and K11) had a difficulty index below the -3 cut-off value, indicating that these items were relatively easy and should be considered for exclusion. In terms of discrimination, 29 items were within or close to the acceptable range of 0.35 to 2.5. In contrast, eight items (K1.2, K1.9, K1.10, K1.11, K1.15, K1.16, K1.17 and K6.3) and two items (K2 and K6.8) had a lower discrimination index by a small and large margin, respectively. Only three items (K4.1, K4.3 and K4.4) were higher than the cut-off value of 2.5 by a large margin.

The item goodness-of-fit statistics showed that 24 items did not fit well the 2-PL IRT model (P-value < 0.05). Furthermore, all items did not support the unidimensionality assumption of the model, as evidenced by modified parallel analysis (P-value = 0.01; Table 3). Based on these findings, in order to identify the best items representing knowledge that supports the unidimensionality assumption (P-value ≥ 0.05) the IRT analysis was rerun with several models using a different combination of items. Model 6 was selected in consideration of the importance of several poor fit items but with acceptable difficulty and discrimination values (Table 4). Hence, the final number of items in the knowledge section was 25, with an average Cronbach’s alpha score of 0.75 (95% CI 0.72–0.78) (Table 5).

The KMO value for the attitude section was > 0.7, indicating that data are suitable for factor analysis. Based on communalities, there were six items (A1, A3, A5, A6, A9 and A13) that had very low h2 values and thus were excluded from further analysis (as they would not correlate with other items that explain attitudes regarding dengue). Two items (A4 and A14) had a low factor loading (< 0.4) but they were retained in the analysis as a factor loading > 0.3 is considered as the minimal accepted value. Three items (A4, A8 and A14) had a low correlation (< 0.5) with the total, as shown in the corrected item-total correlations. Interestingly, by excluding item A4 the overall Cronbach’s alpha score increased while it remained the same after excluding items A8 and A14. Due to this, only item A4 was excluded from the analysis. Overall, only eight items (A2, A7, A8, A10, A11, A12, A14 and A15) retained in the attitude section were considered as reliable (Table 6).

For the practice section, the data were suitable for factor analysis as the KMO value was > 0.7. Seven items (P9, P10, P11, P12, P14, P16 and P18) had low h2 values (< 0.3) and thus were excluded from further analysis. All remaining items had acceptable factor loading values (> 0.4) and corrected item-total correlation values (> 0.5). In conclusion, 11 items (P1, P2, P3, P4, P5, P6, P7, P8, P13, P15 and P17) were retained in the practice section (Table 7).

In comparison to the attitude and practice sections, the KMO value of treatment-seeking behaviour section was < 0.7. Hence, this section was not suitable for factor analysis as the items did not measure an underlying construct that can explain behaviour regards dengue. However, factor analysis was still conducted to confirm the results. The factor loading and corrected item-total correlation values for four items (B3, B4, B5 and B6) were not acceptable as their factor loadings of > 0.4 and corrected item-total correlation of > 0.5, which was considered not reliable. Hence, this section was removed from further analysis.