“…Three studies [ 9 , 23 , 28 ] used individual vulnerability indicator scores directly rather than constructing a composite vulnerability index. Of the 38 studies that constructed a composite vulnerability index, 16 used percentile-rank methods (eg, the CDC’s SVI) [ 4 - 7 , 12 , 22 , 25 , 26 , 29 , 30 , 33 , 34 , 36 , 41 , 44 , 51 ] which assumed equal contribution of the chosen indicators and components to the overall vulnerability, 11 used principal component analysis (PCA) or factor analysis [ 11 , 32 , 35 , 37 , 38 , 43 , 47 - 49 , 52 ] to explore main components of vulnerability from the chosen indicators and assign weights to each component regarding their contribution to the overall vulnerability, two [ 39 , 42 ] directly summed the indicator scores, and two [ 31 , 50 ] used the more sophisticated methods (ie, machine learning, generalized propensity modelling) to generate an overall vulnerability index. Thirteen studies provided insufficient justifications about the chosen statistical methods used to examine the associations between the vulnerability levels and health-related outcomes, and nine [ 6 , 11 , 14 , 24 , 31 , 37 , 39 , 42 , 48 ] only examined their univariate correlations.…”