Predictive Risk Area Modeling for Tuberculosis at the Provincial Level, Thailand
3
Issued Date
2022-05-01
Resource Type
ISSN
2697584X
eISSN
26975866
Scopus ID
2-s2.0-105016199267
Journal Title
Thai Journal of Public Health
Volume
52
Issue
2
Start Page
140
End Page
153
Rights Holder(s)
SCOPUS
Bibliographic Citation
Thai Journal of Public Health Vol.52 No.2 (2022) , 140-153
Suggested Citation
Winyouvongsiri P., Tipayamongkholgul M., Tornee S., Ekchariyawat P. Predictive Risk Area Modeling for Tuberculosis at the Provincial Level, Thailand. Thai Journal of Public Health Vol.52 No.2 (2022) , 140-153. 153. Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/112201
Title
Predictive Risk Area Modeling for Tuberculosis at the Provincial Level, Thailand
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Tuberculosis (TB) is a major public health problem globally and nationally. Although the Thai Ministry of Public Health has strengthened the National TB Control Program nationwide, the magnitude of the TB burden varies across the country. Identifying the high-risk areas of TB is crucial for public health prevention and control planning. A predictive model has been widely used in disease surveillance by identifying high-risk areas. Studies in Asia and western countries have found associations between the high prevalence TB and areas which are highly populated or have a high prevalence of diabetes, and people living with HIV (PLWH). However, these findings are inconclusive. This study aimed to determine factors associated with province-specific TB notification rate and develop scoring for assessing TB risk areas at the provincial level. An ecological study was conducted and used data at the provincial level of Thailand as a unit of analysis. This study was approved by the Ethics Committee of the Faculty of Public Health, Mahidol University (MUPH 2021/51). Data from 76 provinces of Thailand (except Bangkok) were retrieved from the Ministry of Interior and Ministry of Public Health including the number of people aged 40 years or more, those younger than 5 years, percentage of the population living in urban areas, percentage of the population under the poverty line, number of low-income communities, average number of household members, prevalence of PLWH, prevalence of diabetes mellitus, prevalence of chronic obstructive pulmonary disease (COPD), smoking prevalence, prevalence of alcohol consumption, and TB notification rates. Data were analyzed by percentage, median and interquartile, and negative binomial regression was used to develop risk models. A p value < 0.05 was considered statistically significant. Forward negative binomial regression was used to identify associated factors and to develop risk score. Data from 2017 to 2019 were used to develop the model, and it was verified by data from 2020. Factors with p < 0.10 were hierarchically selected into a forward model. The optimal model was considered to be a model with a lesser value of the Bayesian information criterion. Then relative risks of the optimal model were used to develop scores for TB risk area. Pearson’s correlation was used to examine the performance of the optimal model and risk score by checking the correlation between estimated province-specific TB notification rate and observed province-specific TB notification rate during 2017-2019. Finally, the risk scores for the year 2020 were calculated and used to predict TB notification rate in 2020. The predicted province-specific TB notification rate was validated with the observed province-specific TB notification rate in 2020 by Pearson’s correlation. Pearson’s correlation coefficient > 0.8 was accepted. The findings showed associations between the proportion of the population aged over 40 years, number of low-income communities, prevalence of PLWH, prevalence of diabetes mellitus, prevalence of COPD and province-specific notification rate of TB. The risk factors in the optimal model were used to develop score for TB risk area by using a weighted score method. The risk scores were used to estimate province-specific TB notification rate during 2017-2019 and the model’s performance was validated with observed data in the same year. The consistency yielded a correlation coefficient of 0.908. We used risk score of the year 2020 to predict expected province-specific TB notification rate in 2020 and validated it with observed data in the same year and found high consistency between expected and observed province-specific TB notification rate, Pearson’s correlation = 0.895. A high-risk area of TB comprises areas having low-income communities, prevalence of PLWH more than 650 per 100,000, and prevalence of diabetes mellitus more than 5,000 per 100,000. The present study provided useful information for a provincial health officer to perform a TB risk assessment in order to strengthen TB surveillance, prevention, and control and to allocate resources to operate an optimal TB plan.