Composite Measure of Physiological Dysregulation as a Predictor of Mortality: The Long Life Family Study

Abstract

Biological aging results in changes in an organism that accumulate over age in a complex fashion across different regulatory systems, and their cumulative effect manifests in increased physiological dysregulation (PD) and declining robustness and resilience that increase risks of health disorders and death. Several composite measures involving multiple biomarkers that capture complex effects of aging have been proposed. We applied one such approach, the Mahalanobis distance (D_M), to baseline measurements of various biomarkers (inflammation, hematological, diabetes-associated, lipids, endocrine, renal) in 3,279 participants from the Long Life Family Study (LLFS) with complete biomarker data. We used D_M to estimate the level of PD by summarizing information about multiple deviations of biomarkers from specified "norms" in the reference population (here, LLFS participants younger than 60 years at baseline). An increase in D_M was associated with significantly higher mortality risk (hazard ratio per standard deviation of D_M: 1.42; 95% confidence interval: [1.3, 1.54]), even after adjustment for a composite measure summarizing 85 health-related deficits (disabilities, diseases, less severe symptoms), age, and other covariates. Such composite measures significantly improved mortality predictions especially in the subsample of participants from families enriched for exceptional longevity (the areas under the receiver operating characteristic curves are 0.88 vs. 0.85, in models with and without the composite measures, p = 2.9 × 10^-5). Sensitivity analyses confirmed that our conclusions are not sensitive to different aspects of computational procedures. Our findings provide the first evidence of association of PD with mortality and its predictive performance in a unique sample selected for exceptional familial longevity.

Keywords: Long Life Family Study; aging; deficits index; mortality; physiological dysregulation; prediction; statistical distance.

Figure 1

Violin plots with box plots showing D_M for the total sample and by generation and spouse groups. The blue-colored shapes represent a kernel density plot of the distribution of D_M. Line, box, and points represent median, interquartile range (IQR), and outliers that are outside of 1.5 times the IQR.

Figure 2

Kaplan-Meier estimates of conditional survival function of females (A) and males (B) according to the quartiles of D_M. (A) Quartiles are calculated from females who survived until 80 years. (B) Quartiles are calculated from males who survived until 80 years. The numbers in the legend denote values of D_M in respective quartiles. The dark lines denote the point estimates of the survival functions and lighter colored areas denote their 95% confidence intervals.

Figure 3

Receiver operating characteristic (ROC) curves for different models applied to a sample of LLFS probands and their siblings. “D_M+DI” displays the ROC curve for the model with D_M also adjusted for DI and other relevant covariates (see Materials and Methods). “D_M” (“DI”) corresponds to the ROC for the model with D_M (DI) and the other covariates. “None” shows the ROC for the reference model including only the other covariates but not D_M and DI. Areas under the ROC curves (AUC) and p-values for the null hypotheses about a zero difference between AUCs in the respective model and the reference model (“None”) are presented in parentheses.