Effect of modeling a multilevel structure on the Indian population to identify the factors influencing HIV infection

ABSTRACT

Many studies have addressed the factors associated with HIV in the Indian population. Some of these studies have used sampling weights for the risk estimation of factors associated with HIV, but few studies have adjusted for the multilevel structure of survey data. The National Family Health Survey 3 collected data across India between 2005 and 2006. 38,715 females and 66,212 males with complete information were analyzed. To account for the correlations within clusters, a three-level model was employed. Bivariate and multivariable mixed effect logistic regression analysis were performed to identify factors associated with HIV. Intracluster correlation coefficients were used to assess the relatedness of each pair of variables within clusters. Variables pertaining to no knowledge of contraceptive methods, age at first marriage, wealth index and noncoverage of PSUs by Anganwadis were significant risk factors for HIV when the multileveled model was used for analysis. This study has identified the risk profile for HIV infection using an appropriate modeling strategy and has highlighted the consequences of ignoring the structure of the data. It offers a methodological guide towards an applied approach to the identification of future risk and the need to customize intervention to address HIV infection in the Indian population.

KEYWORDS:

• HIV infection
• logistic regression
• multilevel modeling
• multistage sampling

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Nidhi Menon http://orcid.org/0000-0001-8400-6079

Binukumar Bhaskarapillai http://orcid.org/0000-0003-3056-941X

Alice Richardson http://orcid.org/0000-0001-7084-1524

Additional information

Notes on contributors

Nidhi Menon

Ms Nidhi Menon is a Ph.D. candidate at the Research School of Population Health at the Australian National University. Her research interests include multiple imputation, multilevel models and epidemiology.

Binukumar Bhaskarapillai

Dr. Binukumar Bhaskarapillai is an Associate Professor, Department of Biostatistics, National Institute of Mental Health & Neuro Sciences (NIMHANS), Bengaluru, India. He has been involved in teaching Biostatistics postgraduates and research of interest area includes Multivariate Methods, Categorical data analysis, Missing data analysis, Statistical epidemiology and Meta analysis.

Alice Richardson

Dr. Alice Richardson is biostatistician and leader of the Data Analysis in Population Health Hub in the Research School of Population Health at the Australian National University. Her research interests encompass many aspects of biostatistics including multilevel modeling, robust statistics and multiple imputation. She also investigates innovative classroom techniques to discover the most engaging ways to convey statistical concepts to non-statisticians.