Developing and validation of a questionnaire for assessment of individual’s perceived risk of four major non-communicable diseases in Myanmar

The adopting healthy life styles are greatly influenced by individual’s perceived risk of developing non-communicable diseases (NCDs). This study aimed to develop and validate a questionnaire that can assess the individual’s perceived risk of developing four major NCDs. Exploratory sequential mixed methods design was used. Qualitative part developed the question items pool by conducting two expert panels while quantitative part validated the questionnaire using both exploratory (EFA) and confirmatory factor analysis (CFA). Separate samples were used for EFA (n=150) and CFA (n=210). The participants were aged between 25-60 years of both sexes with no known history of NCDs. Face to face interview was conducted. Parallel analysis was done to decide the number of factors to be extracted. EFA was done using maximum likelihood method with Promax rotation to extract the underlying factors of perceived risk while CFA was done to assess the goodness of fit of proposed EFA Model using model fit indices. Based on literature search, 86-item questionnaire was firstly generated. During two expert panels, some overlapped items and items that did not represent the specific construct were removed. Experts made sure the content validity of developed 51-item questionnaire which was used to collect data from 360 participants. EFA revealed the five factors model with 22 high loading items which extracted 54% of total variance. CFA proved that hypothesized five factors model of 21-item questionnaire (one item was removed due to low loading) was satisfied with adequate psychometric properties and model fit indices (RMSEA=0.056, CFI=0.921, TLI=0.908, SRMR=0.063 & χ2/df=1.66). Developed 21-item questionnaire was shown to be valid and reliable to assess the perceived risk of developing NCDs among Myanmar population. Further research should be conducted to assess on the utility of the questionnaire in mismatch between risk perception and current risk and individualized counseling for behaviour change communication.


23
Abstract 24 The adopting healthy life styles are greatly influenced by individual's perceived risk of developing non-25 communicable diseases (NCDs). This study aimed to develop and validate a questionnaire that can assess 26 the individual's perceived risk of developing four major NCDs. Exploratory sequential mixed methods

86
The study used exploratory sequential mixed methods design and was conducted by 6 phases -3 phases 87 in qualitative approach and 3 phases in quantitative approach.

88
Qualitative approach (Questionnaire development) . These participants were randomly split into two groups -150 participants for exploratory factor 123 analysis (EFA) and 210 participants for confirmatory factor analysis (CFA).

124
The major objective of this phase was to reduce the number of items and evaluate the robustness 125 of the intended items. Univariate analysis was done among 150 participants to assess item facility and 126 item discrimination to ensure the selected items were appropriate for EFA. Items with reverse scoring 127 were recoded to get conceptual direction of the construct. Whether items were answered in the same 128 direction was examined using the facility index-approached extreme scores or had a low SD.

132
To determine the factorial structure of the questionnaire and which items together constituted a 133 particular construct, an EFA-a widely used technique in exploring theoretical construct was used [17].

134
EFA was done using maximum likelihood factor extraction method with Promax rotation. Parallel

142
Discrimination of items was measured by inter-factors correlation and if the study found some factors 143 were strongly correlated i.e. >0.7, EFA was rerun only with items of these factors to identify the item that validity. The items were extracted based on not only factor loading but also interpretability of the factors.

146
Items with low factor loading <0.40 and cross-loading with the difference below 0.2 were removed at 147 each step of iteration [19].

148
Phase 5. Reliability of the questionnaire 149 Internal consistency (reliability) of each latent factor was tested for developed questionnaires by 150 Cronbach's alpha (α) coefficient and α ≥0.70 indicates good reliability [20]. The items that affect the 151 reliability of latent factors were removed to get reliable factor. EFA was rerun excluding every item 152 deleted for reliability reasons to get final EFA model.

Development of question items pool (Qualitative Phase)
166 Figure 1 shows the sequential development of final 21-item questionnaire from 86-item questions pool 86-item questionnaire (S1 Table) Table) 184 During 2 nd Delphi round, the experts noticed that some items group had similar meaning;   Table). Further modifications of 208 wording of the items were done according to pilot testing results and suggestions not only from the 209 participants but also from the interviewers to assure face validity.

225
The study was based on health belief model to assess the perceived risk on developing non- Normalization. The first EFA output revealed that among 51 items, 10 items loaded in factor 1, nine items 239 loaded in factor 2, nine items in factor 3, eight items in factor 4 and seven items in factor 5. Among these 240 items, item sus_1 and sus_2 cross loaded in factor 3 and 5; and item intent_6 cross loaded in factor 1 and 241 4 (S4 Table -Pattern matrix 1). To get clean and theoretical meaningful results, the cross loading items 242 were removed from EFA one after another and repeated EFA again. First, sus_2 was removed since the and EFA was run again. During this analysis, previously unloaded item sus_4 was cross loaded between 245 factor 2 and 5 with the difference loading 0.075 so this item was removed and reanalyzed again. After 246 removal of intent_6, there was no more cross loading but two barrier items (bar_1, bar_2) were loaded 247 together with behavioural change items and one behavioural change item (intent_8) was also loaded with 248 barrier items. To get reasonable and theoretically interpretable constructs, these items were removed and 249 rerun EFA again. After these item purification steps, eight self-efficacy items were loaded in factor 1, six 250 benefit items and two severity items were loaded together in factor 2, eight barrier items loaded in factor 251 3, five behavioural change items loaded in factor 4 and; five susceptibility items loaded in factor 5.

256
To get reliable construct, reliability analysis was done for each factor using Cronbach's alpha and  PerBene factor and accounted 11% of total variance with alpha 0.831. Four behavioural change intention items were loaded to PerIntent factor with reliability alpha 0.854 and accounted for 9.9% of total variance. Four susceptibility items loaded strongly to PerSus factor which accounted 5.3% of variance of data with reliability alpha 0.792. Only 3 barrier items loaded to Factor PerBar with 0.683 and this factor 281 accounted 4% of total variance. More than 50% of total variance was extracted by 5 factor solution of 282 EFA. Average factor loading of each factor was greater than or equal to 0.65 and this finding point out 283 that convergent validity of each factor was satisfactory. Regarding to reliability, all the factors' reliability 284 exceed 0.7 except PerBar factor which has reliability nearly 0.7. These findings revealed that reliability of 285 all factors were satisfied to conduct CFA for validation process of 22 items questionnaire developed by 286 EFA.
287 Factor correlation matrix of final exploratory factor analysis was described in Table 2. It was found that between PerBene and PerBar (0.025). All these correlation coefficients were less than 0.7 which was the 293 upper limit that determine the discriminant validity issue; hence, the factors derived from EFA revealed 294 the adequate discriminant validity among the extracted factors.

295
Before CFA, Harman's single factor test was done to assess the common method bias which is a 296 systematic response bias and can occur when a single data collection method was used and that will either 297 inflate or deflate response. This test uses the maximum likelihood method and forced to extract only one 298 factor whether to assess a single factor contributes more than 50% of total variance. It was found that only 299 25.5% was accounted by one factor and this means that there was no problem with common method bias 304 developing non-communicable diseases among adults. These 5 factors have sufficient convergent validity 305 and discriminant validity to conduct the CFA. Moreover, all the factors satisfied with adequate reliability.

306
Whether EFA proposed 5 factors with 22 items questionnaire model can be used to assess the perceived 307 risk on developing non-communicable diseases at population level, CFA was done using different sample 308 of 210 participants.

309
CFA initial model was run with 22 items that hypothesized by EFA and it was found that effi_6  368 confirm EFA proposed model using CFA in which maximum likelihood method was used for parameter avoid these problems, we used the parallel analysis method instead of using Kaiser Criterion and scree 373 plot. It compares the eigenvalues produced be the real data with the eigenvalues estimated from Monte-

374
Carlo simulated matrix created from random data [22,25,26] and the number of factors above the 375 intersection point should be extracted. In our study, the simulated results cross the actual results between 376 factor 5 and 6, hence, we extracted 5 factors from the data. Moreover, to prevent underfactoring or 377 overfactoring, the study assessed not only 5 factors solution but also 4 and 6 factors solution. The findings 378 of 4 and 6 factors solutions were not good enough for theoretical interpretation and some items were cross 379 loaded with more than one factor (See detail in S7 Table). The study also used Promax rotation one of the 380 oblique rotation method which allowed to correlate the extracted factors each other instead of using 381 orthogonal rotation which does not allow to correlate among the factors since we belief the underlying 382 factors of perceived risk were somewhat correlated each other and EFA results also described the 383 significant correlation between some factors.

384
During EFA, the items that assessed the perceived severity of NCDs were dropped for many 385 reasons. First, the parallel analysis provided the evidence to extract 5 factors solution; hence, we forced to 386 5 factors model and some severity items were loaded together with the benefit items and some items were 387 low loading i.e. <0.4. Second, when we ran step by step EFA for item purification, these items were cross 388 loading with other items, that's why some of them were remove. Third, to get reliable construct, we 389 assessed the internal consistency of constructs and some of these items needed to remove to increase the 390 reliability of underlying constructs. These severity items were not able to strongly correlate with each 391 other to form a factor like other items (S4 Table & S7 Table). This might be due to the fact that the 392 participants were the one who had no known NCDs; hence they failed to perceive the severity of NCDs or 393 the developed items had lack of intrinsic ability to capture their perception regarding to severity of 394 diseases due to bias wording or ambiguous wording.