ALL Metrics
-
Views
-
Downloads
Get PDF
Get XML
Cite
Export
Track
Research Article
Revised

A complementary approach to measure construct validity of WHO QOL BREF in Indian multidrug-resistant tuberculosis (MDR-TB) patients through psychometric property and factor analysis

[version 2; peer review: 1 approved with reservations]
PUBLISHED 03 Apr 2025
Author details Author details
OPEN PEER REVIEW
REVIEWER STATUS

This article is included in the Antimicrobial Resistance collection.

Abstract

Background

Multidrug-resistant tuberculosis (MDR TB) affects the physical, psychosocial and inter-relationship structure and thus quality of life (QoL) of an individual. WHO QOL BREF is presumed to capture the QoL construct. This study investigates the diagnostic accuracy and construct of WHO QOL BREF from a psychometric perspective and complements and converges findings through classical test theory.

Methods

The instrument validity study was conducted in a district of Central India amongst the microbiologically confirmed MDR TB cohort of year 2017 (n=98). We calculated global and domain-specific Cronbach alpha and Inter-domain Pearson correlations. The dichotomized items were fitted through Rasch model for item endorsement, response pattern and for variation inconsistencies. Item Characteristic Curves and person item maps were also plotted. We performed DIF (Differential Item Functioning) to check the effect of subgroups on underlying traits. This was complimented with an Exploratory Factor Analysis (EFA) using oblique ProMax rotation. The optimum number of factors were identified by Scree plot and parallel analysis approach and the emerging factor structure was compared with the result obtained through the Rasch model.

Results

The global Cronbach’s alpha was 0.94 (95% CI 0.92-0.96). Social relationship domain had poor correlations with all three domains (r=0.42, r=0.41 r=0.58), higher beta values and less discrimination. DIF showed a differential response by gender. There was a visual clustering and a non-uniform distribution of items across the perceived QoL. A 3-factor model emerged through EFA and was reframed on the notion of self-concept. Items related to pain, medical aid had significant misfit and weak factor loading while items of sexual activity and social support had relatively poor performance in Infit, Wald, DIF on factor loading parameters.

Conclusions

The study indicates the possible deviation of scale from theorized dimensional construct in Indian MDR TB context more with the items of the social relationship domain.

Keywords

Quality of Life, Tuberculosis- Multidrug-Resistant, Psychometrics, WHO QOL BREF, construct, Rasch, Factor Analysis

Revised Amendments from Version 1

We have made the changes that the esteemed reviewer suggested. We have written about incorporating the Hindi version of the tool, rearranged the method part, and elaborated on the ethical considerations. We have created a separate section on data collection. We have written more explicitly about how to handle missing data. In the Data management and analysis section, we have added the reasons in non-mathematical but logical ways for choosing the analysis methods and our approaches behind them. We have revised the discussion part with the change highlighted. We have explained the relatively low values of Cronbach’s alpha and added a section on the possible impact of cultural and linguistic factors.

See the authors' detailed response to the review by Kelechi Elizabeth Oladimeji

Introduction

Chronic diseases are characterised by their ability to influence the individual physically, psychologically, economically, and interpersonally.1,2 The long duration, complex treatment, muti-systems involvement, the costs involved in treatment and a negotiated social role may be ascribed to this.3,4 Tuberculosis (especially the Multi-Drug Resistant type, i.e. MDR TB) is one such disease where, in addition to these factors, social stigmatisations, the propensity of hitting vulnerable sections, and environmental support are also essential influencers.57 All these factors at an individual level may affect perceptual position in life (in the context of the culture and value systems), concerns, standards, and expectations.8 This emotive deviation may further affect the treatment outcome; thus, it seems vital to capture this phenomenon. This is precisely supposed to be measured by a Quality of Life (QoL) instrument.9 This notion of capturing QoL seems more relevant in the Indian context, where it is envisioned to address the social determinants of TB and to foster an enabling environment under a National Tuberculosis Elimination Program (NTEP).10,11

WHO QOL BREF is an instrument that measures QoL through 4 dimensions (physical, psychological, social relationship and environment domains) by 26 items.12,13

This scale has been used to capture QoL for wide-ranging diseases in a generic context like Diabetes Mellitus, Bronchial Asthma, substance abuse, and young onset dementia.1416 This fact may assign it a theoretical superiority over other scales. However, MDR TB has its specific QoL influencers like economic impact, polypharmacy, drug interactions, long duration regimen, compliance, taboos, discriminations, and above all, there is a well-conceived national program in India.6,7,1719 All these factors may interact and affect QoL in a complex manner in MDR-TB. Additionally, health-related quality of life may reflect one’s physical and psychological health in the backdrop of the disease. With this background, the WHO-QoL construct, to the best of our knowledge, is not validated for MDR-TB in the Indian socio-cultural context. Every disease condition may uniquely affect different spheres of QoL; a generic QoL tool like WHO QOL BREF must be validated for a specific disease context. Thus, it seems intuitive to inquire whether WHO QOL BREF can capture this convolution in the Indian MDR-TB context or if some customisation may be required. There appears to be a gap in this direction specially with the most vulnerable segment of TB, i.e., MDR TB.20,21

WHO QOL BREF is an instrument that measures QoL through 4 dimensions (physical, psychological, social relationship and environment domains) by 26 items.12,13 This scale till now has been used in several chronic conditions like Diabetes Mellitus, Bronchial Asthma, substance abuse, young onset dementia and others.1416 The claimed utility of the instrument to capture QoL for wide-ranging diseases in all cultural context and embracing different dimensions in general assigns it a theoretically superiority over other scales. However, MDR TB has its specific QoL influencers like economic impact, polypharmacy, drug interactions, long duration regimen, compliance, taboos, discriminations, and above all, a well-conceived national program in India which may affect QoL bidirectionally at any given point.6,7,1719 Thus it seems intuitive to inquire whether WHO QOL BREF can capture this convolution in the Indian MDR-TB context or it may require some customization. Health related quality of life is essentially a reflection of one’s physical and psychological health in the backdrop of the disease condition. Moreover, as every disease condition is unique in itself in terms of affecting different spheres of QoL, a generic QoL tool like WHO QOL BREF is needed to be validated for a specific disease context. There seems to be a gap in this direction specially with the most vulnerable segment of TB i.e., MDR TB.20,21

This study uses the Hindi-translated version of this tool, which was developed from its original English version in the Delhi field centre by the WHOQOL group and was checked for its cross-linguistic item equivalence against its English counterpart amongst the north Indian population.22,23

Thus, this study attempts to assess the construct of the WHO QOL BREF scale from psychometric Item-Response Theory (Rasch modelling) and further compliments the finding through the Factor analysis approach. Since Item Response Theory (IRT) captures the homogeneity, discrimination ability, and item linearity property of an instrument it offers additional information received from a classical text theory perspective. These two different approaches are converged further to draw an inference for sense-making in real-world context.

Methods

Study design and setting

A cross-sectional research design was used to recruit the participants for the study. All microbiologically confirmed Multi-Drug Resistant TB patients who were above 17 years of age and registered in the District Tuberculosis Centre (DTC) of Bhopal district (a district in Madhya Pradesh/Central Province) in any of the four quarters of the calendar year 2017 were included as participants. The data collection period was 2018-2019.

Ethical Consideration

This study was approved by the Institutional Human Ethics Committee, AIIMS, Bhopal (vide approval letter no. IHEC-LOP/2018/MD0009 Dated 30th June 2018). All the participants were told beforehand that they were not directly benefiting from the study. They were informed about the uses of data and potential indirect benefits to the community. The questions were asked with due consideration of local customs, beliefs, and norms and appropriate privacy and confidentiality were maintained during the interview. Confidentiality of the data sets were maintained and information was shared only with the study team. We ensured the anonymity of the data by replacing identifiers with alphanumeric numbers. The study was non-invasive and thus posed a less than minimal risk to the participants. All the investigators ensured that participation might not cause psychological or social harm

Instrument

The instrument is an adapted brief version of the original WHO QOL 100 survey, a multilinguistic instrument assessed in different countries. The scale consists of 26 items, with two enquiring about the overall quality of life and general satisfaction with health status. The remaining 24 items belong to 4 specific domains: physical health, environmental health, social relationships and psychological health. Each item in the instrument has the 5-point Likert-like response options from the lowest score of 1 to the highest score of 5. The 3 items on the scale have reverse scoring; the rest have positive scoring. The physical health domain incorporates 7 items covering daily living activities, dependence on medicinal substances/medical aid, energy /fatigue, mobility, pain/discomfort, sleep/rest, and work capacity. The psychological health domain assesses six facets of bodily image/appearance, negative feelings, positive feelings, self-esteem, religious/personal beliefs, thinking/learning, and memory/concentration. The social relationship domain assesses personal relationships, social support, and sexual activity through 3 items. The environmental health domain includes eight items related to financial resources, freedom/safety/security, accessibility/quality of health, home environment, opportunity for acquiring new information/skills, recreation opportunities, physical environment (pollution/noise/traffic/climate) and transport. The data was obtained using the Hindi version of the Health-Related Quality of Life WHO QOL BREF Scale with prior permission from the copyright holder.

Data collection

After giving written informed consent, the participants were interviewed during their scheduled visits to the centre for medication. The interviews were taken primarily by the leading author (SS) and supervised by the corresponding author (AJ) and co-author (AMK). The team had previous experience designing and conducting such studies and was affiliated with an apex teaching hospital as a public health faculty/resident. When participants could not visit the centre during the scheduled time, they were contacted at their place of convenience with the assistance of a Senior Treatment Supervisor (STS). There were 103 MDR TB patients registered in 2017, of which 98 participated in this study. The causes for not taking part in the survey were non-availability after attempted contact on two occasions (n=2), death during the study period (n=1), transfer out while on treatment (n=1) and refusal to participate (n=1).

Data analysis

Data management and analysis

The participants' responses on QoL items were transferred to an Excel sheet, and data was checked for missing values, duplicate entries and string inconsistencies. The missing values were duly checked for the response of each item and were associated while transferring to the Excel sheet. In the case of a blank cell, the response was back-traced from the participant’s form and connected to an Excel sheet with the unique alphanumeric identifiers. The further analysis of the data was done with R version 4.1.2 (2021-11-01) available at public domain.

In short, the study estimated the instrument's psychometric properties in terms of item recommending probability (in high or low perceived QoL), patterns of responses and response bias by Item Response Theory methods (Rasch model), the construct validity by the classical test theory method and further converged the two theories to draw translational inferences.

The reliability of the tool was estimated using Cronbach’s alpha for each domain and for the whole instrument. Point and 95% CI values of Cronbach’s alpha were calculated using bootstrapping with 1000 iterations. The inter-domain correlation structure of construct was evaluated by the Pearson correlation matrix.

The psychometric interpretation of the whole QoL construct was done by dichotomized item responses into low perceived QoL category (complete negative agreement/slightly negative agreement/neutral response) and high perceived QoL (moderately positive/complete favourable agreement) category. We used Rasch model to check the items endorsed spectrum about high to low perceived QoL, patterns of response biases (if any) or inconsistency and for fit and for checking the instrument fairness across the different groups. We calculated Beta estimates and corresponding 95% CI to understand the item specific QoL endorsement at 50th percentile level. The discriminatory capacity for each item to differentiate between low perceived QoL and high perceived QoL was represented by Item Characteristics Curve (ICC). ICC plotted the logit QoL values (X axis) against the endorsement probability (Y axis). A person item map was created to understand the distribution of item parameter’s location and person parameter along the linear logit QoL construct value. The Rasch model was applied to global diagnostic (Anderson likelihood ratio test) and local diagnostics (Wald test). We further performed a Differential Item Functioning (DIF) to check whether perceived QoL varies across gender, duration of treatment with category of patients (new/old) with same level of underlying trait. The operational definition for new/old category was decided per programmatic guidelines, where a new patient has either never taken anti -TB drug or has received it for less than a month. Previously treated patients for >1 months were classified as old patients.

The extent of factorability of the data and suitability for structure detection were determined respectively by the Kaiser-Meyer-Olkin (KMO) test and Bartlett’s test of Sphericity. The Scree plot and parallel analysis method decided the optimum number of factors. We chose the principal axis method as the factoring method due to the property of having robustness against sample size and multivariate normality violation. We used oblique ProMax rotation over orthogonal rotation as a factor axis rotation method due to its superior ability to conceptualise the representation of the construct. Items with factor loadings of 0.4 or higher were considered significant and were included in the interpretation. The items loading >0.4 on multiple emerging factors were assigned to higher loading factors. Both structure and pattern coefficients were calculated. The model fitting to data was determined by the chi-square test and RMSEA as the goodness of fit statistic. We calculated a Comparative Fit Index (CFI) and Tucker Lewis Index (TLI) to compare the hypothesised model with the null model. At last, we attempted to draw inferences by converging and comparing the responses received from these two approaches.

The reason for choosing the Rasch model as the preferred approach was that it allowed us to compare items’ ability to measure QoL on a transformed single logit interval scale while maintaining person and item independence. This common scale for item and person further allowed us to find items matching participants' abilities and the overall model structure. The other reasons for choosing this scale are its ability to measure a unidimensional construct and stability for cross-cultural comparisons. Another relevant property, helpful for MDR-TB QoL measurement (due to the inherent complexity), is its property of objective comparison independent of sample characteristics.

38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure1.gif

Figure 1. Flow diagram showing the process, sequence, and method in brief for instrument validation.

Results

The information on quality of life was collected from 98 participants in continuation phase (IQR-11 to 16 months) of treatment.35 The mean age of participants was 35.13(±14.11) years. Most of the participants were from urban area (n=79) and were unemployed or unskilled worker (n=65). Better quality of life was perceived amongst urban-living men, in higher education group and having higher socioeconomic status. The detailed descriptions of the participants' profile and their transformed score distributions are published elsewhere.22 This section focuses entirely on the validity and reliability perspective of the tool to measure QoL among MDR TB patients.

Instrument Cronbach’s alpha and Pearson corelations

The global Cronbach’s alpha of the instrument was 0.94 (95%CI 0.92-0.96). The same was calculated for physical domain (0.85, 95%CI 0.78-0.90), psychological domain (0.88, 95%CI 0.83-0.91), social relationship domain (0.59, 95%CI 0.33-0.74) and for environmental health domain (0.89, 95%CI 0.85-0.92). Thus, apart from the social relationship domain, the interrelatedness of the items to measure the respective domains can be interpreted as reliable. However, the relatively higher values of global alpha may indicate the redundancy of some items. The strength and patterns of Pearson corelations amongst different domains are shown as a correlation matrix plot in Figure 2. The environmental health domain had moderate to good correlation with other domains (r=0.65 for psychological health, r=0.64 for physical health) while the social relationship domain showed a relatively poor correlation (r=0.42 for psychological health, r=0.41for physical health) with other domains measuring the QoL construct.

38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure2.gif

Figure 2. Correlation matrix plot mapping the intercorrelations amongst different domains.

IRT (Rasch modelling)

The item difficulty parameters (β values) for each item showing the probability of endorsement for each QoL component along with 95% CI are shown in Table 1. The visual description of the same to discriminate between low and high perceived QoL component traits is shown in Figure 3 as Item Characteristic Curve (ICC). The items in social relationship domain have relatively lower beta values compared to other domains. The shape of the curves for the items in this domain is deviated from the classical S shaped curve thus may indicate poor discrimination ability in between high and low perceived QoL. Apart from this, the items related with dependence on medical aid (Q4), energy (Q10) and negative feeling (Q26) also had relatively poor discriminatory capacity. The most discriminating items were related to work capacity (Q18), activity of daily living (Q17) and pain (Q3) in the physical health domain. Items related with bodily image, spirituality, religion, and personal beliefs in psychological health domain also had fair discrimination capacity.

Table 1. Beta values (Item difficulty Parameters) with 95% Confidence Intervals of each item.

Item No.Item DescriptionItem Difficulty Parameters (beta) *Estimate Std. ErrorLower CIUpper CI
Global Item
Q1Global overall-1-2.050.45-2.93-1.16
Q2Global overall -2-1.500.40-2.27-0.72
Physical Health Domain
Q3Perceived physical pain-1.340.38-2.09-0.59
Q4Dependence-medical aid-3.170.62-4.39-1.94
Q10Energy-2.810.56-3.90-1.72
Q15Mobility-0.590.33-1.230.05
Q16Sleep/rest satisfaction0.630.270.101.16
Q17Activity of daily life-1.210.37-1.92-0.47
Q18Working capacity-0.940.35-1.62-0.26
Psychological Health Domain
Q5Positive feeling-0.690.33-1.34-0.05
Q6Personal belief0.010.29-0.580.57
Q7Ability to concentrate-0.700.33-1.35-0.05
Q11Bodily image0.160.29-0.400.72
Q19Self esteem0.480.28-0.061.02
Q26Negative feeling2.060.251.562.56
Social Relationship Domain
Q20Personal relationship3.160.272.633.69
Q21Sexual activity2.660.262.163.17
Q22Social support2.00.251.502.48
Environmental Domain
Q8Personal safety/security0.080.29-0.490.65
Q9Physical environment0.330.28-0.220.88
Q12Financial resource/support-1.500.40-2.27-0.72
Q13Access to information0.480.28-0.061.02
Q14Opportunity-recreation/leisure0.700.280.181.23
Q23Home environment0.560.270.021.09
Q24Access to Health care1.880.251.382.37
Q25Transport1.380.260.881.89
38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure3.gif

Figure 3. Item Characteristics Curves (ICC) showing the discrimination and dimensionality of each item in four different domains.

The person item map (Figure 4) plots the individual perceived QoL at the vertical axis and the component ability at the horizontal level to perceive a specific trait. Items were found to be non-uniformly distributed along the whole range of latent dimension construct and there was a visible clustering of (4 visual clusters) items along with same person ability measures. These clusters of items were heterogenous in nature as they had different domains membership. Notably, global items (Q1 and Q2) did not have a corresponding person distribution measure and thus might not capture the extra amount of information over and above the specified items. Similarly, there were zones of the item vacuum detected along -3 to -2 logit and again between 0 to 1 logit to which no items were mapped.

The fitting of the data to Rasch model was checked with Anderson -LR test (LR=49.60 LR=49.60, chi-square, df=16, p=0) globally and by Wald test chi-square statistic for each item. The goodness of fit, deterministic patterns, Wald chi-square statistic and Outfit/infit-t statistic is shown in Table 2.

38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure4.gif

Figure 4. Person Item Map showing the item and person distribution along the same latent quality of life (QoL) values.

Table 2. Rasch Diagnostic: Item wise goodness of fit, Wald chi-square statistic and Outfit/infit-t statistic.

Item No.Chi Sq (p -value)z statistic (p-value)Outfit MSQInfit MSQOutfit tInfit tDiscrimination
Q148.45 (1.00)0.500.86-0.17-0.370.52
Q265.36 (0.99)0.681.07-0.140.330.52
Q3479.78 (0.00)2.79(0.005)4.991.642.972.230.19
Q4179.84 (0.00)1.59(0.11)1.871.241.010.660.17
Q532.20 (1.00)0.340.61-1.44-2.130.78
Q650.62 (1.00)-1.52(0.13)0.530.84-1.32-0.970.67
Q743.07 (1.00)0.450.85-1.06-0.730.66
Q882.79 (0.81)-1.08(0.28)0.860.92-0.25-0.450.63
Q9122.51 (0.03)-0.31(0.75)1.271.140.860.920.51
Q1054.69 (1.00)0.570.870.240.220.37
Q1155.65 (1.00)-1.73(0.08)0.580.89-1.24-0.650.66
Q1239.68 (1.00)0.410.91-0.62-0.270.55
Q1347.13 (1.00)-1.24(0.22)0.490.72-1.89-2.070.73
Q14181.64 (0.00)2.05(0.04)1.891.412.542.660.37
Q1543.51 (1.00)0.450.76-1.13-1.210.70
Q1668.96 (0.98)-0.46(0.65)0.720.84-0.95-1.190.66
Q1724.00 (1.00)0.250.58-1.27-1.980.74
Q1844.73 (1.00)0.460.75-0.84-1.130.68
Q1976.62 (0.92)-0.66(0.51)0.790.97-0.58-0.170.62
Q2093.37 (0.53)1.33(0.18)0.971.040.120.370.34
Q21476.51 (0.00)3.87(0.00)4.961.595.394.220.05
Q22107.37 (0.18)1.95(0.05)1.121.220.481.750.41
Q2373.22 (0.95)-0.27(0.79)0.760.96-0.74-0.170.58
Q2469.01 (0.97)-0.41(0.68)0.720.87-0.99-1.050.57
Q2543.58 (1.00)-2.21(0.03)0.450.63-2.57-3.380.72
Q2671.92 (0.96)-1.37(0.17)0.750.89-0.79-0.82055

The infit statistics to measure the extent of unexpected response of the items was calculated. All the items except Q3 (χ2=479.78, p=0.000), Q4 (χ2=179.84, p=0.000), Q14 (χ2=181.64, p=0.000), Q20 (χ2=93.37, p=0.528), Q21 (χ2=476.51, p=0.000), Q22 (χ2=107.37, p=0.182) had acceptable infit values while these questions had high infit MSQ values. The responses to these items might influenced with unexpected inlying response patterns of participants. Similarly, item wise Wald test statistics to understand the fitting of Rasch model for that component were calculated. The p values of Q3 (z=2.79, 0.005), Q14 (z=2.05, 0.041), Q21 (z=3.87, 0.000), Q22 (z=1.95, 0.051), Q25 (z=-2.21, 0.027) were found to be significant. This may indicate the poor fit of Rasch model in reference to these items.

Differential Item Function (DIF) analysis

We performed DIF analysis for gender, duration of treatment (<14 months versus >14 months) and as per categorisation of patients (newly diagnosed versus old patients). Question related to personal safety and security (8), sexual activity (21), social support (22) and home environment (23) were perceived differently by male and female participants. This difference was found to be statistically significant. But we did not find any significant difference in the β values of the participants as per treatment category and observed treatment duration. The visual presentation of DIF analysis is given in Figure 5 and 6 as z-statistic coordinates with reference to subgroups and difference in item endorsements probability by subgroups.

38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure5.gif

Figure 5. DIF showing the Effect of Gender and Duration on Item responses- (A) item wise z statistics (B) item coordinates in reference to subgroup and (C) item wise endorsement logit differences between subgroups.

38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure6.gif

Figure 6. DIF showing the Effect of Treatment Category on Item response- (A) item wise z statistics (B) item coordinates in reference to subgroup and (C) item wise endorsement logit differences between subgroups.

Exploratory factor analysis

The optimum number of factors were determined through scree plot and Parallel Analysis method. Both the methods projected a three factor model instead of classical four factors model. Figure 7(A) depicts a scree plot that shows the eigenvalues on the y-axis and number of factors on the x-axis. The point on the elbow indicated the emergence of 3 factors might explain the maximum model variance. Similarly, parallel analysis plot showed that a point beyond 3 eigenvalues extracted from real data receded those eigenvalues extracted from random data.

38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure7.gif

Figure 7. Optimum factors number projections by (A) Scree plot and (B) Parallel analysis method.

Table 3 describes the loading of each item on the three derived corresponding factors with their communality and uniqueness score. The variance proportion explained by PA1, PA2 and PA3 were respectively 0.51, 0.33 and 0.16. The sum of square loading SS for all three factors was>1 (PA1-5.40, PA2-3.45, and PA3-1.71), thus all factors were found to be worth keeping. The intercorrelation between factors was calculated as 0.4 (PA1 to PA2), 0.5 (PA1 to PA3), and 0.4 (PA2 to PA3). The sufficiency of 3 extracted factors was detected by Tucker Lewis Index (0.84) and by the RMSEA index (0.07, 90% CI 0.06-0.09). The diagrammatic representation of the stronger loading items to the corresponding principal axis along with PA correlation values are shown in Figure 8.

Table 3. Loading of each item on the three corresponding factor with communality and uniqueness score.

Item No.Item DescriptionPA-1PA-2PA-3Communality score-H2Uniqueness Score-U2 com
3Perceived physical pain0.41-0.250.030.120.881.8
4Dependence on medical aid0.11-0.300.360.170.832.1
5Positive feeling0.740.18-0.090.630.371.1
6Personal belief0.400.030.380.480.522.0
7Ability of concentration0.500.170.080.410.591.3
8Personal safety and security0.680.010.020.480.521.0
9Physical environment-0.060.820.000.630.371.0
10Energy0.410.32-0.130.190.812.3
11Body Image0.62-0.040.170.500.501.2
12Financial support0.180.58-0.070.420.581.2
13Access to information0.290.610.050.660.341.4
14Opportunity -leisure activity0.080.080.430.180.821.3
15Mobility0.790.19-0.220.630.371.3
16Sleep/rest0.460.170.180.450.551.6
17Activity of daily life0.900.04-0.160.710.291.1
18Working capacity0.870.09-0.250.650.351.2
19Self-esteem 0.61-0.160.290.540.461.6
20Personal relationship0.060.130.400.170.841.4
21Sexual activity-0.100.010.420.040.971.4
22Social support-0.110.380.310.270.732.1
23Home environment-0.190.860.240.800.201.3
24Access to healthcare0.010.290.510.460.541.6
25Transport0.170.380.440.620.382.3
26Negative feeling0.410.240.200.340.662.7
38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure8.gif

Figure 8. Layout of items loading to corresponding component and emergence of the latent construct.

The factor loading shows the emergence of three latent constructs. The first construct (PA1) includes all the items of the psychological health domain and physical health domain except medical aid dependence (4). It also consists of one item from environmental domain i.e., personal safety and security (8). This dimension has been rephrased as “inner self”. The items loaded on the second construct are those items originally classified under Environmental domain of the WHO QOL BREF namely physical environment (9), financial support (12), home environment (23), accessibility to information (24). It also consists of one item from Social relationship domain i.e. social support (22). This dimension can be rephrased as “peripheral self”. The third construct (PA3) consists of 6 items in total that includes two items from Social relationship domain i.e. personal relationship (20) and sexual activity (21), four items from environmental domain i.e. access to health care (24), leisure activity (14), transport (25) and one item from the physical health domain i.e. medical aid dependence (4). This latent construct has been named as “personal or immediate self”.

Convergence and comparison

The findings of the Rasch modelling broadly corresponded with those of factor analysis, The items pertaining to perceived physical pain (Q3,0.41), medical aid dependence (Q4,0.36), opportunity and leisure activity (Q14,0.43) and all items of social relationship domain i.e. personal relationship(Q20,0.40), sexual activity (Q21,0.42) and social support(Q22,0.38) had weak loading. Four out of the six items loaded on PA3 showed overall suboptimal contribution of these items to reveal the QoL construct. Similarly the beta estimates for perceived physical pain (Q3, β=-1.34), medical aid dependence (Q4, β=-3.17) had a relatively extreme negative beta value while personal relationship (Q20, β=3.16), sexual activity (Q21, β=2.66) and social support (Q22, β=2.0) had somewhat extreme positive beta values. Both scenarios showed a relatively poor discrimination ability of items to differentiate in between high and low perceived QoL trait. The items in social relationship domain had a weaker inter item relationship and lesser inter -domain correspondence. It might had a propensity that these items are susceptible to subjective response patterns as shown by high infit MSQ values. Moreover, as of the items were culturally sensitive they might be responded differently by men and women as shown in DIF analysis. The deviation from the classical 4 factors model may also be ascribed to these facts where PA-1 clubs the classical physical and psychological domain items and PA-2 resemble to environmental health domain. The maximum item divergence can be seen in PA-3 which is an assortment of social relationship, environment and physical health domain and social relationship items are most unstable in this domain. The visual depiction of this methodological convergence and sense making at a glance is shown in the Figure 9.

38e08a1f-11c0-4cad-b8d3-9b7d01bb34bc_figure9.gif

Figure 9. Graphical representation of the convergence of methods.

Discussion

QoL is a profoundly subjective notion ingrained in one’s meta-cognition, where one attempts to understand both the self and the situation. Evolutionary biology terms it ‘reflection’, and this is linked with informed actions in similar future encounters. The reflections thus assign a primitive survival benefit.2527 Yet as civilization progressed, along with physiological and safety needs, psychological needs for belonging, intimacy, and self-esteem gradually became part of the self.28 The need for growth and self-fulfilment with social and cultural evolution further supplemented these deficiency needs. The mutual cohesive interactions of these “needs” in an individual human shaped the quality-of-life construct.29,30

Certain theoretical assumptions were made during the design phase. First, the participants in this study were chosen from the continuation phase of MDR TB treatment. We wanted the participants to have a long enough duration of treatment to absorb the experiences and maintain the stability of expression. Second, the quality of life in MDR TB is dynamic. Yet, we refrained from taking measurements longitudinally because of the inherent threat of answers by the participants from the memory of presuming an exaggerated sense of self/others. Third, the first two items of the scale are global items that deal with the overall quality of life and health. The response on these overall items may be understood with more lucidity through domain-specific items; thus, global items were not included in the Rasch model.

The items representing the social relationship domain in this study showed relatively weak internal consistency, weak correlations with other domains, less discriminatory power, and less loading strength with the underlying construct. The possible explanations behind this erratic performance are fewer items representing the domain and the relatively sensitive nature of the questions. The questions about sexual life and personal relationships are more likely to be misinterpreted and responded to in an unfitting manner.31 The beta values for this domain were relatively high, indicating that these items may not offer sufficient information to capture the differences in perceived QoL between respondents.32 The relatively inadequate performance of tool in the social relationship domain may be affected by potential cultural and linguistic issues like lacking the equivalent words, cultural beliefs and social desirability bias. The hierarchical societal structures with patriarchal dominance may also impact how openly participants express opinions. Variations like literacy levels may also affect comprehension patterns of personal/sensitive questions, and responses may be skewed as per prevalent social acceptance norms. The cultural adaptability of the tool may be enhanced by conducting cognitive interviews to check comprehension and testing the tool in diverse settings. The few items may be rearticulated using indirect cues and normalising language.

As observed in the person item map, some items were clustered at the same latent trait of perceived QoL. Although this clustering of items was heterogeneous, clustered items had different domain memberships. Hence, these clusters are more likely to be pseudo clusters or probably because of random variations in findings or due to item response collapsibility; however, if it is the true clustering that measures the same latent QoL trait, the interpretation of the question in the Indian context may be relooked from a linguistic and socio-anthropological angle.

We attempted to relook covariations of items and their loading on latent variables through Exploratory Factor Analysis (EFA). The structure model obtained was a three-factor model against the original four-factor model. Items from physical and psychological domains moved correlatedly and constructed the first factor (PA1), explaining the maximum variance in the model. Items from the original environmental domain had the most crucial source of variations from the original 4-factor model. Here, four items related to the external environment created a separate construct, and the rest of the items, along with all the items of the original social relationship domain, were covaried. This domain explained the least variance of the model, and the findings followed the Rasch model. These three constructs were renamed on self. At this moment, it is also vital to deliberate on ‘self’. Self is about the “who and what I am in my sight”. It consists of our rating of our behavioural abilities and assumed standing in the immediate environment.33,34 An event like MDR TB may generate or widen the agreement between the idealistic and accurate image and may negatively influence the component of self-concept (approval) and this is translated into overall low perceived QOL.35

This study's three latent domains constructed through factor loadings substantiate and align with the above argument. The inner/core self and bodily self are considered the appendages of the central self-scheme concept, where one acknowledges the constancy of his individualism and becomes aware of his own existential identity. The ‘peripheral self’ assigns validation to the categorical self-concept, where he realises he fits into societal/outer categories shaped by immediate environmental and prevalent socio-cultural and demographic milieu.1,34 This self-observed fit into the environment is not an absolute fit but comparative to the perceived fit by the others in the same peripheral environment. The whole construct is somewhat dynamic and moulded by significant events like Tuberculosis (for our study participants) and experiences thereafter in the context of disease. Thus, the corresponding QoL may also be perceived in negative or positive directions as per changes in the core, bodily, and peripheral self.

IRT measures have the property of item and ability invariance; hence, the evaluation of measure performance becomes immune to sample fluctuation, which may be considered as one of the study's strengths. Another strength is the methodology triangulation, which maps the construct validity of items and describes phenomenology in a rationalistic and explanatory mode. To the best of our knowledge, this study combines, converges, and complements two methods to measure construct validity in context with MDR TB QoL assessment. We also attempted to granularize analysis at the subgroup and item level to give it more robustness. This methodological strength may overcome the relatively small sample size, which should also be seen in the context of the rare incidences of MDR TB in the community, stringent inclusion criteria, and an attempt to restrict in a reasonably homogenous geo-environmental setting to avoid differential programmatic inputs. Another limitation is the inability to measure the test-retest validity due to the study's cross-sectional nature. Yet, we tried to overcome this by analysing the effect of duration on item responses through differential item functioning.

Conclusions

WHOQOL BREF is considered till now to capture the quality of life in a variety of diversified settings and in different population contexts. However, this study indicates the possible deviation from theorized dimensional constructs specifically when it is to be utilized in the MDR TB context. Moreover, the items in the social relationship domain may be reexamined again from the Indian subcultural context. With the caveat of single-center study and with a relatively lesser sample size these facts may be verified from similar studies on different disease contexts. However, as the findings were substantiated and converged using different methods this necessitates the need to explore it further. From the policy perspective, all behavioral change strategies in general should be tailored to the socio-cultural context, specific needs, and characteristics of the target audience. Items in QoL construct did identify and quantify the psychometric ‘self’ dimension. Thus, it would be meaningful to understand the interaction of ‘disease-self’ with different selves after necessary item customization. The possibility to incorporate a customized QoL tool may be explored further to address an individual need and to augment the probability of treatment success at a programmatic level.

Ethical considerations

This study was approved by Institutional Human Ethics Committee, AIIMS, Bhopal (vide approval letter no. IHEC-LOP/2018/MD0009).

Comments on this article Comments (0)

Version 2
VERSION 2 PUBLISHED 05 Sep 2023
Comment
Author details Author details
Competing interests
Grant information
Copyright
Download
 
Export To
metrics
Views Downloads
F1000Research - -
PubMed Central
Data from PMC are received and updated monthly.
- -
Citations
CITE
how to cite this article
Sharma S, Kokane AM, Krishna SN and Joshi A. A complementary approach to measure construct validity of WHO QOL BREF in Indian multidrug-resistant tuberculosis (MDR-TB) patients through psychometric property and factor analysis [version 2; peer review: 1 approved with reservations]. F1000Research 2025, 12:1108 (https://doi.org/10.12688/f1000research.138296.2)
NOTE: If applicable, it is important to ensure the information in square brackets after the title is included in all citations of this article.
track
receive updates on this article
Track an article to receive email alerts on any updates to this article.

Open Peer Review

Current Reviewer Status: ?
Key to Reviewer Statuses VIEW
ApprovedThe paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approvedFundamental flaws in the paper seriously undermine the findings and conclusions
Version 1
VERSION 1
PUBLISHED 05 Sep 2023
Views
12
Cite
Reviewer Report 25 Feb 2025
Kelechi Elizabeth Oladimeji, Ezintsha, University of the Witwatersrand Johannesburg, Johannesburg, Gauteng, South Africa 
Approved with Reservations
VIEWS 12
My summary of the paper
The study sought to examine the psychometric properties of the WHO QOL BREF in the context of multidrug-resistant tuberculosis (MDR TB). And I strongly believe this is worth the evidence and is not redundant. ... Continue reading
CITE
CITE
HOW TO CITE THIS REPORT
Oladimeji KE. Reviewer Report For: A complementary approach to measure construct validity of WHO QOL BREF in Indian multidrug-resistant tuberculosis (MDR-TB) patients through psychometric property and factor analysis [version 2; peer review: 1 approved with reservations]. F1000Research 2025, 12:1108 (https://doi.org/10.5256/f1000research.151486.r364613)
NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.
  • Author Response 03 Apr 2025
    Ankur Joshi, Community and Family Medicine, AIl India Institute of Medical Sciences, Bhopal, 462023, India
    03 Apr 2025
    Author Response
    1. Aim Statement and Use of the Hindi Translated Version:
      In the aim statement, it should be explicitly clarified that the Hindi version of the tool was used
    ... Continue reading
COMMENTS ON THIS REPORT
  • Author Response 03 Apr 2025
    Ankur Joshi, Community and Family Medicine, AIl India Institute of Medical Sciences, Bhopal, 462023, India
    03 Apr 2025
    Author Response
    1. Aim Statement and Use of the Hindi Translated Version:
      In the aim statement, it should be explicitly clarified that the Hindi version of the tool was used
    ... Continue reading

Comments on this article Comments (0)

Version 2
VERSION 2 PUBLISHED 05 Sep 2023
Comment
Alongside their report, reviewers assign a status to the article:
Approved - the paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations - A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approved - fundamental flaws in the paper seriously undermine the findings and conclusions
Sign In
If you've forgotten your password, please enter your email address below and we'll send you instructions on how to reset your password.

The email address should be the one you originally registered with F1000.

Email address not valid, please try again

You registered with F1000 via Google, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Google account password, please click here.

You registered with F1000 via Facebook, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Facebook account password, please click here.

Code not correct, please try again
Email us for further assistance.
Server error, please try again.