Possible relationship among socio-economic determinants, knowledge and practices on lymphatic filariasis and implication for disease elimination in India

Abstract

Objective

To assess the socio-economic determinants, knowledge and practices on lymphatic filariasis in India and discuss the implications for elimination.

Methods

A case–control study was undertaken to obtain knowledge and practice measures on various dimensions of the Wuchereria bancroftian filarial disease using a structured questionnaire. A structural equation model, a statistical technique for testing and estimating causal relationships using a combination of statistical data and qualitative causal assumptions, was developed.

Results

Among the affected individuals, the model was able to explain more than 90% of the variance in awareness about the lymphatic filariasis, 58% of the variance in prevention aspects of the disease and 24% of the variance in people’s knowledge about mosquitoes. The corresponding values in non-infected individuals were 49, 24 and 34%, respectively. A significant positive effect of education on awareness and prevention aspects of the disease was noted among the non-infected individuals. However, among the affected individuals, the awareness on various aspects of the disease was completely absent.

Conclusions

The present analysis highlights the crucial role played by formal education on creating awareness about lymphatic filariasis and how to prevent this vector-borne disease. The importance of education on intervention measures against mosquito breeding and biting is also dealt with in the analysis for planning an effective and sustainable control program.

Appendices

Appendix

The transformation of the likelihood function that is often used in maximum likelihood estimation (MLE) is given by

$$ {{H}} = { \log }\left[ {{ \det }\left( {\text{Z}} \right)} \right] + {\text{tr}}. \, \left[ {{\text{SZ}}^{ - 1} \left] { - { \log }} \right[{ \det }\left( {{S}} \right)} \right] - {{U}}, $$

where “log” means natural logarithm, “det” means determinant, “tr.” means trace (sum of main diagonal), U total number of measured exogenous (q) and endogenous (p) indicators, S unbiased sample estimate of covariance matrix of observed indicators, Z estimate of covariance matrix of measured indicators based on estimates of model parameters, and Z ⁻¹ inverse of Z matrix.

Minimization of this function is used to obtain estimates of model parameters from the sample data. As the sample size increases, the sample estimates become closer to the population values. The minimum value of H can be used to compute a likelihood ratio χ² statistic that allows us to test the degree of congruence of the covariance structure implies by the theory and that observed by empirical data. This χ² statistics is given by

$$ {{\chi^2}} = { \min }\left( {{H}} \right)\left( {{{N}}/ 2} \right), $$

with df = [(p + q)(p + q − 1)/2 + (p + q)] − f, where N sample size, and f number of free parameters estimated.

The emphasis was on understanding the causal sequence of relationships implies by the theory. In SEM for sound statistical reasons, it is usual to analyze covariance matrix and not correlation matrix. Therefore, observed variance covariance matrix between exogenous and endogenous variables was used as input data for fitting the full structural LISREL model. Initially, the measurement errors of exogenous and endogenous variables were assumed to be uncorrelated. Therefore, the corresponding covariance matrices of Θ_δ and Θ_ε (Table 1) were presumed to be diagonal. Accordingly, the model with the following matrix specifications was fitted.

Model (a)

Matrix	Form	Fixed elements	Value
Λ y	Rectangular	(1,1), (3,2), (5,3)	1.00
B	Rectangular	All	0.00
Γ	Rectangular	Nil	–
Φ	Symmetric	All	–
Ψ	Symmetric	Nil	–
Θε	Symmetric	Off-diagonal	–
Θδ	Diagonal	All	0.00
Θδε	Rectangular	All	0.00

The coefficients involved in this model were estimated and the model yielded a very high χ² value (310.7; P < 0.001 for 92 df) indicating that the fit was unacceptable. If the fit of a model is not adequate, it has been common practice to modify the model, by deleting parameters that are not significant and adding parameters that improves the fit. Therefore, statistical significance of individual parameter estimation was tested based on a null hypothesis that a parameter in a measurement model is zero. To reject this null hypothesis, a test criterion that the ratio of parameter estimation to its standard error approximately equals to 2 (fixing the confidence probability of 95%) was used. Accordingly, 14 parameter estimates of exogenous path coefficients in both affected and non-infected groups were not statistically significant. Therefore, a constraint was imposed that these 14 exogenous path coefficients were zero. While the model was fitted again the χ² value was 317.3 for 106 df The difference in χ² (317.3–310.7) between the two models was not statistically significant at 14 df confirming that these path coefficients are zero. Further, to improve the model fit, modification indices of all the fixed parameters were examined and the maximum modification index was 51.5 for Θ_ε (7, 1) in affected group and this parameter was set free. While the model was fitted again, as expected the χ² value was reduced to 248.9 for 105 df indicating that the parameter is significant for the refinement of the model.

Model (b)

At each step the parameter that showed maximum modification index was freed to obtain the largest improvement in the fit. This process is shown below until an adequate fit was reached.

Group	Fixed parameter	Maximum modification index	Action taken	Model χ2	Reduction in χ2
Non-infected	Θε (7, 1)	15.62	Freed	226.3	22.6
Affected	Θδε (3,6)	14.41	Freed	217.4	8.9
Affected	Λy (6,2)	14.02	Freed	197.4	20.0
Non-infected	Θε (6, 3)	11.25	Freed	187.3	10.1
Affected	Θδε (1,6)	11.07	Freed	174.2	13.1
Affected	Θδε (4,7)	9.53	Freed	162.7	11.5
Affected	Λy (2,2)	10.79	Freed	151.2	11.5
Affected	Λy (7,2)	9.37	Freed	141.9	9.3
Affected	Θδε (5,1)	8.76	Freed	133.2	8.7
Non-infected	Θδε (4,4)	8.71	Freed	123.9	9.4
Non-infected	Θε (7, 4)	6.61	Freed	115.3	8.6
Affected	Θδε (4,3)	5.54	Freed	109.7	5.6
Affected	Θε (7, 4)	6.83	Freed	103.7	6.0
Affected	Θε (4, 1)	7.82	Freed	96.8	6.9
Non-infected	Θδε (1,5)	5.22

Since the maximum modification index 5.22 for the element Θ_δε (1,5) was in the tolerable level of 5.0, further refinement of model was felt redundant. However, the parameter estimate for Ψ (1,1) was not statistically significant. Therefore, it was fixed to be zero and the resultant model χ² was 97.9 for 92 df. The good fit of the model and its assessment indices are presented in the result part.