Causal mediation analysis identifies the path of causality between exposure and outcome. Existing methods involve simple scenarios with one/limited number of univariate mediators given large sample. However often in biomedical studies this is not true. In the neuroimaging studies we often have high-dimensional data where number of variables are greater than number of samples and mediators are not measured directly but are latent. In this study we proposed a framework for causal mediation analysis in settings where we have latent mediator measured by multiple high-dimensional measured variables. We used data from BNS which includes the data for severe acute malnutrition (SAM) children and their matching controls. Exposure variable is SAM, outcome is school age cognitive score measured by WISC and the effect of SAM on cognitive scores is mediated by a latent variable Brain State measured by observable quantitative EEG, qualitative scales EEG and soft neurological signs. The proposed methodology consists of 4 steps: 1) Screening using high-dimensional multiple regression 2) PCA for the screened variables 3) Latent variable modeling using structural equation modeling approach to construct the latent mediator based on PCA score 4) Causal mediation analysis with univariate latent mediator. The results showed that the effects of malnutrition (in first year of life) on school-age cognition was fully mediated by Brain State with p-value <2e-16 and the direct effect was not significant