Skip to main content
. 2021 Feb 25;13:43. doi: 10.1186/s13148-021-01030-8

Table 4.

Replication: predictive ability in the Framingham Heart Study (450 K) of the CpGs selected in the Strong Heart Study for lymphatic–hematopoietic, solid and overall cancers

N predictors Lymphatic–hematopoietic cancer Solid cancers Overall cancer
C index N coef > 0b C index N coef > 0b C index N coef > 0b
Risk factorsa 5 0.76 5 0.69 4 0.69 5
Risk factorsa + cell counts 11 0.79 10 0.69 11 0.69 10
Risk factorsa + cell counts + DMPs 132/379/399c 0.89 62 0.75 34 0.74 32

coef coefficient, PCs principal components, DMPs Differentially Methylated Positions

aAge (years), smoking status (current/former/never), sex (men/women) and BMI (kg/m2)

bVariables with coef 0 are considered not to play any role in prediction

cFor lymphatic–hematopoietic and overall cancers, among the 125 CpGs selected by elastic-net in the Strong Heart Study (restricting to CpGs included in 450 K), 123 were present in the Framingham Heart Study. In addition, the four risk factor variables and the five cell count variables were included in the elastic-net model: a total of 132 variables. For solid cancers, 373 CpGs were selected in the Strong Heart Study, 370 being present in the Framingham Heart Study. With the four risk factor variables and the five cell count variables, a total of 379 variables were included. For overall cancers, 395 CpGs were selected in the Strong Heart Study, 390 being present in the Framingham Heart Study. With the four risk factor variables and the five cell count variables, a total of 399 variables were included