Table 1.
Baseline characteristics of included patients.
Fig 1.
Predictive power of DFA alpha scaling exponent to forecast the development of T2DM on Cox survival analysis.
(A) Heatmap with Cox proportional hazard coefficient for different windowings. (B) Cox coefficient’s p-value for different windowings.
Fig 2.
Influence of integration on DFA alpha scaling exponent values.
(A) Rate of decline of alpha with the addition of an increasing component of randomness (white noise) in series pre–treated through integration (DFAint) or not (DFAraw). (B) Boxplot of the differences of the alpha exponent values between both methods.
Fig 3.
Error depending on the number and length of missing segments.
Each unit of missing segment represents 5 minutes. In 30 series originally with no missing values, an increasing number of randomly distributed segments of increasing length were deleted and interpolated. The process was repeated 30 times for each combination of length and number of deleted segments and for each patient, and DFA scaling exponent was calculated for each replica. The mean error (absolute difference with the real alpha value (complete series)) was recorded for each combination of length and number of missing segments.
Table 2.
Cox proportional hazard model for different metrics.
Table 3.
Principal Components Analysis of the variables selected in the Cox proportional hazard model.