A confirmatory analysis of the knowledge, attitudes, and perceptions questionnaire for community-based surveillance of infectious diseases: a cross-sectional study in Malaysia

Abstract

Objectives

This study aimed to validate the psychometric properties of the recently developed knowledge, attitudes, and perceptions questionnaire for community-based surveillance of infectious diseases (KAP-CBS-ID questionnaire), using confirmatory factor analysis (CFA) and item response theory (IRT).

Methods

A cross-sectional study using multistage sampling recruited 470 schoolteachers from Kelantan, Malaysia. The self-administered KAP-CBS-ID questionnaire consists of 3 domains: knowledge (31 items), attitudes (23 items), and perceptions (21-items). Two-parameter logistic (2-PL) IRT analysis and CFA were performed to validate the knowledge section. For attitudes and perceptions sections, CFA proceeded using a 4-factor model to evaluate both model fit and construct validity.

Results

Two-PL IRT analysis of the knowledge section resulted in elimination of 14 items due to inadequate discrimination or difficulty parameters. The 3-factor CFA model demonstrated good fit indices for knowledge (root mean square error of approximation [RMSEA], 0.028; comparative fit index [CFI], 0.945; Tucker-Lewis index [TLI], 0.941) without any modifications. The attitudes section required re-specification, ultimately yielding 21 items across 4 factors with acceptable fit indices (standardized root mean square residual [SRMR], 0.067; RMSEA, 0.055; CFI, 0.937; TLI, 0.927). Similarly, the perceptions section was refined to 17 items across 4 factors, showing good model fit (SRMR, 0.055; RMSEA, 0.059; CFI, 0.962; TLI, 0.954). Factor loadings ranged from 0.33 to 0.98, while Raykov’s rho reliability estimates ranged from 0.71 to 0.93. Factor determinacy exceeded 80% for all factors.

Conclusion

The KAP-CBS-ID is a valid and reliable instrument for assessing community representatives’ knowledge, attitudes, and perceptions regarding community-based surveillance of infectious diseases.

Graphical abstract

Introduction

Infectious disease outbreaks and epidemics typically begin within the community. Community members are often the first to observe changes in health patterns and identify health threats, yet delays in transmitting this information to health authorities can hinder timely interventions [1,2]. Community-based surveillance (CBS) leverages the capabilities of community members to perform surveillance activities within their own communities, thereby enhancing the early detection and response to disease outbreaks [3]. The World Health Organization defines CBS as the systematic detection and reporting of public health events or cases of specific diseases by community members to health authorities [4]. This approach can substantially expedite the detection and response to outbreaks [5–11].

Multiple studies have shown that CBS can facilitate more timely responses and improved containment of disease outbreaks [1,9,10,12]. However, evidence supporting the overall effectiveness of CBS remains limited [13]. Documented cases have shown that community health workers can effectively implement CBS programs, even in challenging contexts such as conflict zones [10,13–15] or during natural disasters [16,17]. The International Federation of the Red Cross and Red Crescent (IFRC) maintains a network of more than 15 million volunteers worldwide who are deeply embedded in their local communities [18]. These volunteers, who possess linguistic and cultural familiarity and are physically present within their communities, represent a critical resource for implementing preparedness activities, including CBS, in remote regions with limited access to health services [9,13]. This highlights the vital role of community members in the early identification and reporting of health threats.

Malaysia has implemented event-based surveillance, defined as the organized and rapid collection of information regarding events that may pose risks or concerns to public health. These events, which arise within the community, may have actual, suspected, or potential impacts on human health, thereby generating concern, fear, or alarm [19]. Information about such events may be collected from rumors or other ad hoc reports via both formal channels (such as routine reporting systems) and informal sources (including non-governmental organizations, community members, and social media).

However, as noted in the 2018 Event-Based Surveillance Protocol [19], the flow of information from the community to health authorities remains slow. An administrative directive from the Deputy Director-General of Health (Public Health) issued in January 2014 highlighted that these events were often reported in the media well before they reached Ministry of Health officials. This delay can be attributed to a lack of community awareness regarding the CBS process. Although the community is recognized as the primary source of initial signals in the CBS system, the Event-Based Surveillance Protocol [19] also notes a lack of empirical studies evaluating community members’ knowledge, attitudes, and perceptions concerning this crucial surveillance mechanism.

Therefore, this study aims to confirm the structure of the knowledge, attitudes, and perceptions questionnaire for community-based surveillance of infectious diseases (KAP-CBS-ID questionnaire) [20] to assess these constructs among Malaysian community representatives. The selection of community representatives, rather than the general community, was intended to increase the signal-to-noise ratio‎ and provide adequate specificity, ensuring that reporting does not overwhelm the surveillance system [21].

Materials and Methods

Study Design and Setting

This study employed a cross-sectional design and was conducted among Malaysian community representatives residing in Kelantan from March 2024 to June 2024. It constitutes the third stage of a larger research project. The first stage involved the development and translation of the questionnaire [20], while the second stage focused on item response theory (IRT) analysis for the knowledge domain and exploratory factor analysis (EFA) for the attitudes and perception domains.

Setting

The study was conducted in primary and secondary schools across the state of Kelantan, Malaysia. Four districts—Kota Bharu, Bachok, Pasir Puteh, and Kuala Krai—were selected, with schools within each district chosen randomly. The KAP-CBS-ID questionnaire comprises 3 main sections: knowledge, developed based on a comprehensive literature review; attitudes, constructed using the theory of reasoned action (TRA); and perceptions, grounded in the health belief model (HBM). The questionnaire underwent thorough EFA in a previous phase, utilizing principal component analysis with varimax rotation to identify the underlying factor structure. This process revealed distinct factors within each domain, which were then confirmed in the current confirmatory factor analysis (CFA) study. Each scale underwent a rigorous validation process, including content validation by 11 experts, face validation, and a pilot study involving 30 participants, followed by EFA. The knowledge section used a dichotomous response format, whereas the attitudes and perceptions sections were rated on 5-point Likert scales. The questionnaire’s structure is as follows: the knowledge section contains 3 subdomains—knowledge of infectious diseases (18 items), knowledge of CBS of infectious diseases (3 items), and community-level case definition (10 items)—with responses of “yes,” “no,” and “I don’t know.” The attitudes section comprises 4 subdomains: negative attitudes (6 items), subjective norms (4 items), intention to participate in CBS (6 items), and behavioral likelihood to engage in CBS (7 items), each rated on a 5-point Likert scale ranging from “1-strongly disagree” to “5-strongly agree” or “1-never” to “5-always,” depending on the item. The perceptions section also includes 4 subdomains: perceived susceptibility (5 items), perceived benefit (5 items), perceived barriers (3 items), and self-efficacy (4 items), rated on 5-point Likert scales from “1-never true” to “5-always true” and from “1-extremely unlikely” to “5-extremely likely,” as appropriate [20]. Data collection commenced in May 2024 and concluded in June 2024, utilizing the KAP-CBS-ID questionnaire.

Participants

A total of 470 community representatives, including teachers from both public and Islamic schools, were recruited using a multistage sampling method. After obtaining informed consent, participants completed the self-administered questionnaires during school break times. To be eligible for inclusion, participants had to be at least 18 years old and proficient in both written and spoken Malay. Participants were encouraged to provide honest responses, and the estimated time to complete the questionnaire was approximately 20 to 25 minutes.

Bias

The researchers implemented several measures to minimize bias in the validation study of the KAP-CBS-ID questionnaire. The use of a multistage sampling technique ensured representative participation across varied geographic and demographic settings within Kelantan. Selection bias was reduced by applying clear inclusion criteria for schoolteachers. Comprehensive psychometric analyses, including 2-parameter logistic (2-PL) IRT and CFA, enabled the removal of items with poor performance. The self-administered questionnaire featured a balance of positively and negatively worded statements, and assurances of anonymity and confidentiality were provided to all participants. To further reduce cultural and linguistic biases, the questionnaire underwent a thorough translation and back-translation process with local experts, ensuring its suitability for the Malaysian context.

Study Size

Sample size determination for the CFA was performed using 2 approaches. First, the “rule of thumb” ratio of 5 participants per questionnaire item (N:P=5:1) was applied, indicating a need for 375 participants for the 75-item questionnaire. This calculation incorporated a 20% anticipated dropout rate. Second, Arifin’s web-based sample size calculator (2017) estimated a requirement of 463 participants [22]. Based on these calculations, a sample size of 470 participants was deemed adequate.

The sampling process comprised 3 stages. Initially, 4 districts in Kelantan were purposively selected: Kota Bharu and Bachok (urban), and Pasir Puteh and Kuala Krai (rural). While the rural-urban distinction was not directly analyzed in this validation study, it is considered relevant for future research, as some evidence suggests that CBS may be more effective in rural than in urban settings [3]. In the second stage, schools within each district were randomly selected. In the final stage, participants within each school were approached conveniently during their free periods. An overview of the study was provided, and verbal consent was obtained, with the process facilitated by a Malay-speaking research assistant.

Statistical Methods

To minimize data entry errors, all questionnaire items were compiled into a Google Form (Google LLC). Responses from the paper-based questionnaires were then manually entered into this Google Form. Subsequently, all responses were exported as an Excel spreadsheet for data cleaning and further analysis. After data cleaning and numerical coding, the dataset was transferred to RStudio ver. 4.4.1 (RStudio), for CFA and 2-PL IRT analysis.

Item response theory

The knowledge section of the questionnaire was analyzed using IRT due to its dichotomous response format. Specifically, a 2-PL IRT model was implemented using the ltm package, version 1.2-0. The 2-PL IRT model estimates 2 parameters for each item: difficulty (b) and discrimination (a). The difficulty parameter represents the ability level at which an individual has a 50% probability of answering the item correctly, with acceptable values typically ranging from –3 to +3. The discrimination parameter indicates how well an item distinguishes between individuals with varying ability levels; values above 0.25 are considered acceptable, and values above 1.0 are regarded as good. Unidimensionality was assessed using the modified parallel analysis method to ensure that items measured a single underlying construct. Cronbach’s alpha was calculated to evaluate the internal consistency reliability of the knowledge domain, with values above 0.7 considered acceptable. Items with low discrimination (a<0.25) or extreme difficulty (|b|>3) were considered for revision.

Confirmatory factor analysis

CFA was conducted following the exploratory study to verify the factor structure of the questionnaire. The CFA utilized the lavaan package ver. 0.6–18 [23]. Model fit was evaluated using a range of indices: absolute fit indices, including chi-square (χ²) and standardized root mean square residual (SRMR); parsimony-adjusted indices, such as the root mean square error of approximation (RMSEA); and comparative fit indices, including the comparative fit index (CFI) and the Tucker-Lewis index (TLI) [24].

Thresholds for model fit indices in this study were set as follows: a p-value for χ² greater than 0.05, RMSEA and SRMR less than or equal to 0.08, and CFI and TLI greater than or equal to 0.92, based on the model structure and sample size considerations [25]. Composite reliability (CR) was assessed using the semTools package ver. 0.5–6, within RStudio, with reliability indicated by Raykov’s rho [24,26]. A Raykov’s rho value exceeding 0.7 was considered acceptable for reliability [27]. Factor determinacy, representing the precision with which factor scores can be estimated from observed items, was evaluated, with values above 0.80 considered good. Average variance extracted (AVE), measuring the amount of variance captured by a construct relative to measurement error, was deemed acceptable at values above 0.50. CR, another measure of internal consistency, was also considered adequate if it was above 0.70.

For the knowledge section, CFA was conducted after IRT analysis. The combined use of IRT and CFA is generally recommended, as this approach enhances the quality of scale measurement and results in more robust assessments in applied research settings [28]. Due to the categorical nature of responses, the weighted root mean square residual (WRMR) was used in place of SRMR.

Ethics Statement

Ethical approval for this study was obtained from the Human Research Ethics Committee of Universiti Sains Malaysia (ref. no: USM/JEPeM/22050317). All participants were provided with information about the study, and informed consent was obtained from those who agreed to participate.

Results

Descriptive Data

As shown in Table 1, a total of 470 community representatives participated in this study. The demographic data show that the sample was predominantly female (75.3%) and Malay (99.4%), with a mean age of 43.3 years (standard deviation, 9.5). Most participants were married (79.4%) and employed as government servants (85.5%), while 87.2% held university degrees or higher qualifications. The majority worked as public schoolteachers (87.7%), resided in urban areas (64.0%), and reported having received information about infectious diseases (93.8%).

Table 1.

Demographic profile of the study participants (n=470)

Variable	Frequency
Age (y)	43.3±9.5
Sex
Female	354 (75.3)
Male	116 (24.7)
Ethnicity
Malay	467 (99.4)
Others	3 (0.6)
Marital status
Single	73 (15.5)
Married	373 (79.4)
Divorced	6 (1.3)
Widowed	18 (3.8)
Occupation
Government employee	402 (85.5)
Private sector	60 (12.8)
Pensioner	3 (0.6)
Others	5 (1.1)
Education level
Primary school	3 (0.6)
Secondary school (PMR/SRP)	2 (0.4)
Secondary school (SPM)	16 (3.4)
Religious school (Pondok)	2 (0.4)
Diploma	37 (7.9)
University and higher	410 (87.2)
Role
Public school teacher	412 (87.7)
Religious school teacher	58 (12.3)
Residency
Urban	301 (64.0)
Rural	169 (36.0)
Received information about infectious diseases
Yes	441 (93.8)
No	29 (6.2)

Data are presented as mean±standard deviation or n (%).

PMR, Penilaian Menengah Rendah (lower secondary assessment); SRP, Sijil Rendah Pelajaran (lower education certificate); SPM, Sijil Pelajaran Malaysia (Malaysian certificate of education).

Figure 1 illustrates the sources from which community representatives most frequently obtained information. The internet was the most widely used source, with 85% of respondents reporting that they obtain information about infectious diseases online. In contrast, the animal health sector was the least utilized, with only 5% receiving information from this source.

Figure 1.

Primary sources of information on infectious diseases.

Main Results

Knowledge section

The 2-PL IRT analysis indicated that the difficulty and discrimination parameters were generally acceptable for most items in the knowledge domain. This section includes 3 subdomains: information on infectious diseases (18 items), knowledge of CBS of infectious diseases (3 items), and community-level case definitions (10 items). Although a few items exhibited poor model fit, as indicated by a significant chi-square p-value, they were retained because their difficulty and discrimination parameters remained within acceptable ranges, as shown in Table 2. The proportion of information explained by the knowledge items was 93.57%, 89.35%, and 98.31% for the 3 subdomains, respectively. Both the knowledge of CBS and community-level case definitions subdomains met the unidimensionality assumption (p=0.53 and p=0.81, respectively). The internal consistency of the knowledge section was acceptable, with a Cronbach’s alpha of 0.71 in IRT analysis.

Table 2.

Results of the IRT analysis of the knowledge section (n=470)

Factor/items in original questionnaire	IRT results				CFA results
	b	a	χ2 (df=8)	p	Recoded items	Standardized item loading	AVE	CR
Information on infectious diseases							0.45	0.92
Q1	–7.81	0.43	9.20	0.33	Q1	0.33
Q3	–7.73	0.41	12.73	0.12	Q2	0.55
Q4	–2.58	1.09	11.62	0.17	Q3	0.67
Q7	–2.81	1.28	26.16	0.00	Q4	0.55
Q8	–2.24	2.54	10.83	0.21	Q5	0.65
Q9	–1.23	0.33	4.33	0.83	Q6	0.49
Q10	–0.76	11.43	3.05	0.93	Q7	0.91
Q11	–0.68	23.29	2.35	0.97	Q8	0.96
Q12	–1.06	6.90	0.96	0.99	Q9	0.98
Q14	–1.36	3.51	9.39	0.31	Q10	0.61
Q15	–1.37	29.73	2.08	0.98	Q11	0.92
Q16	–1.34	4.72	2.63	0.96	Q12	0.90
Q17	–4.79	0.78	44.87	<0.001	Q13	0.62
Q18	–3.52	0.78	65.61	<0.001	Q14	0.71
Q22	–1.55	1.70	21.63	<0.001	Q15	0.61
Q23	–1.54	1.47	23.42	<0.001	Q16	0.62
Q24	–2.63	1.17	26.59	<0.001	Q17	0.65
Q25	–2.94	0.72	6.48	0.59	Q18	0.60
Knowledge of CBS							0.58	0.78
Q33	–1.52	3.10	27.66	<0.001	Q19	0.68
Q34	–1.74	3.11	29.68	<0.001	Q20	0.85
Q35R	–1.19	2.07	51.82	<0.001	Q21R	0.74
Community-level case definition							0.50	0.93
Q36	–1.33	2.38	11.55	0.17	Q22	0.69
Q37	–2.44	1.66	9.48	0.30	Q23	0.74
Q38	–1.63	1.00	34.37	<0.001	Q24	0.81
Q39	–3.42	1.11	16.70	0.03	Q25	0.78
Q40	–1.09	3.31	4.60	0.80	Q26	0.79
Q41	–0.91	4.58	9.55	0.30	Q27	0.87
Q42	–0.91	5.92	6.60	0.58	Q28	0.91
Q43	–0.10	0.73	15.89	0.04	Q29	0.63
Q44	–3.52	0.78	11.34	0.18	Q30	0.70
Q45	–2.44	0.98	23.02	0.00	Q31	0.79

Items Q2, Q5, Q6, Q13, Q19, Q20, Q21, Q26, Q27, Q28, Q29, Q30, Q31, and Q32 were removed during IRT analysis.

IRT, item response theory; CFA, confirmatory factor analysis; χ², chi-square; b, difficulty; a, discrimination; df, degree of freedom; AVE, average variance extracted; CR, composite reliability; CBS, community-based surveillance.

Following the 2-PL IRT analysis, CFA was performed using the weighted least squares mean and variance adjusted (WLSMV) estimator to confirm the 3-factor structure of the knowledge section. The model demonstrated a good fit, with fit indices as follows: CFI=0.945, TLI=0.941, WRMR=1.301, RMSEA=0.028 (90% CI, 0.022–0.033), and an RMSEA p-value>0.950.

Attitudes section

CFA was conducted on the attitudes section. Since multivariate normality was not achieved, the robust maximum likelihood estimator was employed. The initial CFA model did not meet criteria for adequate fit, with a CFI of 0.874 and TLI of 0.858. However, SRMR and RMSEA values were within acceptable thresholds at 0.073 and 0.072 (0.066–0.079), respectively. To enhance model fit, item Q48 was removed due to a low factor loading (0.14), followed by the exclusion of item Q47, which had a high standardized residual (5.157). An error correlation was also added between items Q45 and Q46. These modifications were carried out iteratively in consultation with research team experts. As a result, model fit indices improved, as detailed in Table 3.

Table 3.

CFA model re-specification for the attitudes section

Model	Modification	CFI	TLI	SRMR	RMSEA (90% CI)	RMSEA p
Initial model (model 0)	-	0.874	0.858	0.073	0.072 (0.066–0.079)	<0.000
Model 1	Item Q48 removed (factor loading=0.140)	0.876	0.858	0.074	0.075 (0.069–0.082)	<0.000
Model 2	Item Q47 removed SR=5.157)	0.893	0.877	0.071	0.072 (0.064–0.079)	<0.000
Model 3	Correlated residual for item q45 with Q46 (MI=159.4)	0.937	0.927	0.067	0.055 (0.048–0.063)	0.126

CFA, confirmatory factor analysis; CFI, comparative fit index; TLI, Tucker-Lewis index; SRMR, standardized root mean square residual; RMSEA, root mean square error of approximation; CI, confidence interval; SR, standardized residual; MI, modification index; -, no modification was made.

The final model for attitudes comprised 21 items, organized into 4 distinct factors in alignment with the TRA: behavioral likelihood to be involved in CBS (7 items), subjective norms (4 items), intention to engage in CBS (6 items), and negative attitudes toward CBS (6 items). The CR coefficients (Raykov’s rho) for these factors were 0.860, 0.895, 0.714, and 0.546, respectively. Although the fourth factor’s reliability was relatively low, its factor determinacy value above 0.80 indicates good internal consistency reliability, as suggested by Wang and Wang [29].

Table 4 presents the initial and final CFA models for the attitude factors, including factor loadings, AVE, CR, and factor determinacy for all final model factors.

Table 4.

CFA of the attitudes section

Factor/item	Initial CFA model (model 0)			Final CFA model (model 3)			Factor determinacy for model 3
	λ	AVE	CR	λ	AVE	CR
Behavioral likelihood (BL)		0.54	0.859		0.58	0.860	0.85
Q45F1	0.484			0.441
Q46F1	0.565			0.529
Q47F1a)	0.530			-
Q49F1	0.872			0.879
Q50F1	0.830			0.835
Q51F1	0.902			0.906
Q52F1	0.865			0.870
Subjective norms (SN)		0.69	0.895		0.69	0.895	0.83
Q39F2	0.728			0.728
Q40F2	0.815			0.815
Q41F2	0.905			0.905
Q42F2	0.886			0.886
Intention to engage in CBS (INT)		0.31	0.714		0.31	0.714	0.94
Q32F3	0.652			0.651
Q33F3	0.627			0.628
Q34F3	0.642			0.642
Q36F3	0.334			0.332
Q38F3	0.491			0.490
Q43F3	0.388			0.387
Negative attitudes (ATT)		0.17	0.504		0.20	0.546	0.86
Q35RF4	0.365			0.366
Q37RF4	0.467			0.475
Q44RF4	0.356			0.364
Q48RF4b)	0.140			-
Q53RF4	0.424			0.415
Q54RF4	0.665			0.658

CFA, confirmatory factor analysis; λ, factor loadings, AVE, average variance extracted; CR, composite reliability.

^a)Item Q47 was excluded from the final model due to a high standardized residual;

^b)Item Q48 was excluded from the final model due to low factor loading; -, the item was removed.

Perceptions section

Initial CFA of the perceptions section, structured into 4 factors, did not produce satisfactory model fit. To improve the model, problematic items (Q55, Q66, Q69, and Q70) were removed based on both fit indices and theoretical rationale. Additional error correlations were introduced between items Q56 and Q57, as well as Q61 and Q62. Improved model fit indices are summarized in Table 5.

Table 5.

CFA model re-specification for the perceptions section

Model	Modification	CFI	TLI	SRMR	RMSEA (90% CI)	RMSEA p
Initial model (model 0)	None	0.850	0.827	0.091	0.094 (0.087–0.102)	<0.000
Model 1	Q70RF3 removed low factor loading	0.866	0.845	0.089	0.093 (0.085– 0.101)	<0.000
Model 2a)	Q66RF3 removed low factor loading	0.880	0.860	0.085	0.092 (0.084–0.101)	<0.000
Model 3	Q69RF3 removed low factor loading	0.884	0.862	0.084	0.096 (0.087–0.106)	<0.000
Model 4	Q55RF3 removed low factor loading	0.905	0.886	0.065	0.069 (0.063–0.075)	<0.000
Model 5	Correlate item 56 with 57 (MI=191.9)	0.952	0.941	0.068	0.066 (0.055–0.077)	0.011
Model 6	Correlate item 61with 62 (MI=63.4)	0.962	0.954	0.055	0.059 (0.047–0.070)	0.111

CFA, confirmatory factor analysis; CFI, comparative fit index; TLI, Tucker-Lewis index; SRMR, standardized root mean square residual; RMSEA, root mean square error of approximation; CI, confidence interval; MI, modification index.

^a)Model 2: items (Q55, Q66, Q69, Q70) were removed, and residuals of the following items were correlated (Q56–Q57) and (Q61–Q62).

The 4 factors in this section align with the HBM framework: perceived susceptibility (5 items), perceived benefit (5 items), perceived barriers (3 items), and self-efficacy (4 items). Each factor exhibited strong internal consistency reliability, with CR values of 0.816, 0.904, 0.808, and 0.820, respectively.

Discussion

Studies evaluating CBS programs have traditionally focused on metrics such as reporting completeness, report frequency, and related indicators, primarily emphasizing healthcare workers and health facilities [7,30–32]. Although comprehensive tools exist to assess the need for CBS, such as the Community-Based Surveillance Assessment Tool by IFRC [33], these instruments do not specifically target community awareness, attitudes, or perceptions toward CBS. To address this gap, we developed and validated a new questionnaire and used CFA to provide robust evidence for the validity of its factor structure [34]. This study presents the psychometric properties and model fit of the newly developed instrument using CFA.

The 3 domains of the KAP-CBS-ID questionnaire (knowledge, attitudes, and perceptions) were analyzed separately to evaluate their respective psychometric properties and factor structures. This approach ensures that each domain preserves its theoretical integrity while improving the overall construct validity of CBS assessment. Such a design is especially useful for targeted interventions, allowing for the independent assessment of specific components of knowledge, attitudes, and perceptions as needed.

The cut-off values for model fit indices used in this study were guided by the recommendations of Hair [35]. For questionnaires with more than 30 items and sample sizes exceeding 300, the suggested thresholds are: CFI and TLI above 0.92, SRMR no greater than 0.08 when CFI exceeds 0.92, and RMSEA below 0.07 if CFI is at least 0.92. These criteria ensure an adequate model fit for large-scale instruments. The fit indices observed in our models met these cut-off values. Several studies from Malaysia further support the use of these indices as standard benchmarks [36–38], with some reporting slightly lower values than those obtained in our study.

The IRT analysis of the knowledge section led to the removal of 14 items due to inadequate discrimination or difficulty, mirroring the process described by Hung et al. [39]. This refinement resulted in 31 items in the knowledge section, which were subsequently confirmed through CFA [40]. The 3-factor CFA model for the knowledge section demonstrated a good fit without requiring modification. The WLSMV estimator was employed due to the categorical nature of the data. As SRMR is not appropriate for categorical responses, the WRMR was used, where lower values indicate better fit, with a cut-off of 1.0 generally considered acceptable [41]. Although the WRMR value in our study did not reach the ideal threshold, prior research, such as Shuan [42], suggests that a less-than-ideal WRMR is not problematic when other fit indices confirm model adequacy.

The CFA for the attitudes section initially produced unsatisfactory fit indices. The re-specification process involved iterative adjustments to improve fit (Table 4). The initial model (model 0) exhibited suboptimal indices (CFI, 0.874; TLI, 0.858; RMSEA, 0.072; SRMR, 0.073). Modifications included sequential removal of items with low factor loadings (such as Q48 and Q47) and the addition of residual correlations between selected item pairs, such as Q45 and Q46 (modification index [MI], 159.4), based on factor loadings and modification indices. These changes resulted in a final re-specified model with a CFI of 0.962, TLI of 0.954, SRMR of 0.055, and RMSEA of 0.059 (90% confidence interval, 0.047–0.070; p=0.111), consistent with guidelines from Hair et al. [25], as summarized by Newsom [43], who recommend CFI and TLI>0.90, RMSEA<0.08, and SRMR<0.08 as thresholds for adequate fit.

The CFA for the perceptions domain also revealed initial model misfit, prompting a sequential re-specification process to improve fit (Table 6). The original model yielded fit indices of CFI=0.850, TLI=0.827, SRMR=0.091, and RMSEA=0.094. Although these indices were not ideal, similar studies have deemed such values acceptable [26]. Incremental improvements were achieved by removing items with low factor loadings (Q70RF3, Q66RF3, Q69RF3, and Q55RF3), with a significant improvement following the removal of Q55RF3 (CFI, 0.905; TLI, 0.886; SRMR, 0.065; RMSEA, 0.069). Further enhancement was accomplished by adding error correlations between Q56 and Q57, as well as Q61 and Q62. The final model (model 6) exhibited a CFI of 0.962, TLI of 0.954, SRMR of 0.055, and RMSEA of 0.059, indicating satisfactory fit. The justification for correlated errors was based on item similarity, with such method effects often arising from similarly or reversely worded items [34]. All factor loadings in the questionnaire exceeded the recommended threshold of 0.3 [25], with observed values ranging from 0.33 to 0.98. Internal consistency reliability across domains was satisfactory, with Raykov’s rho values ranging from 0.71 to 0.93. The exception was the negative attitudes domain, which fell below the recommended threshold. However, because of its conceptual importance, determinacy—a supplementary measure of internal consistency—was evaluated and surpassed the recommended value of 0.80, indicating adequate reliability [29].

Table 6.

Results of CFA of the perceptions domain

Factor/item	Initial CFA model (model 0)			Final CFA model (model 6)			Factor determinacy for model 6
	λ	AVE	CR	λ	AVE	CR
Perceived susceptibility (SUS)		0.58	0.862		0.545	0.816	0.83
Q56F1	0.629			0.621
Q57F1	0.609			0.594
Q61F1	0.841			0.787
Q62F1	0.854			0.799
Q63F1	0.842			0.873
Perceived benefit (BEN)		0.65	0.904		0.65	0.904	0.85
Q71F2	0.702			0.702
Q72F2	0.803			0.804
Q73F2	0.889			0.889
Q74F2	0.834			0.835
Q75F2	0.811			0.810
Perceived barriers (BARR)		0.23	0.504		0.58	0.808	0.82
Q58RF3	0.741			0.731
Q59RF3	0.755			0.765
Q60RF3	0.789			0.796
Q55RF3a)	0.274			-
Q66RF3a)	0.145			-
Q69RF3a)	0.160			-
Q70RF3a)	0.105			-
Self-efficacy (SE)		0.53	0.819		0.53	0.820	0.85
Q64F4	0.794			0.791
Q65F4	0.782			0.777
Q67F4	0.627			0.630
Q68F4	0.739			0.745

CFA, confirmatory factor analysis; λ, factor loadings, AVE, average variance extracted; CR, composite reliability; -, the item was removed.

^a)Item Q55, Q66, Q69, and Q70 were excluded from the final model due to a high standardized residual.

This study has a few limitations. First, the focus on schoolteachers as community representatives resulted in a relatively homogeneous and accessible sample; future validation should incorporate a broader range of community leaders, such as religious figures, local government officials, community health volunteers, and traditional healers, to enhance the tool’s applicability. Second, the sample was predominantly Malay, limiting generalizability across other ethnic groups. Third, the sample size calculations did not account for the cluster effect inherent in the multistage sampling design, which could affect the precision of the estimates. Future research should address these limitations by including more racially diverse and heterogeneous samples.

Conclusion

The KAP-CBS-ID questionnaire demonstrated strong validity, reliability, and psychometric properties for assessing the knowledge, attitudes, and perceptions of Malaysian community leaders regarding CBS of infectious diseases. The finalized model consists of 69 items, with 31 assessing knowledge, 21 evaluating attitudes, and 17 measuring perceptions. Given the relative homogeneity of the current sample, further validation is recommended to ensure generalizability to broader target groups and individuals with different racial backgrounds.

HIGHLIGHTS

• A psychometrically valid questionnaire on knowledge, attitudes, and perceptions regarding community-based surveillance of infectious diseases was previously developed.

• Item response theory and confirmatory factor analysis validated the 3-domain questionnaire among 470 schoolteachers in Malaysia.

• The final questionnaire retained 17 knowledge items, 21 attitude items, and 17 perception items that exhibited high reliability.

• This validated tool can effectively assess community leaders' preparedness for infectious disease surveillance.