Predicting mortality with biomarkers: a population-based prospective cohort study for elderly Costa Ricans

Abstract

Background: Little is known about adult health and mortality relationships outside high-income nations, partly because few datasets have contained biomarker data in representative populations. Our objective is to determine the prognostic value of biomarkers with respect to total and cardiovascular mortality in an elderly population of a middle-income country, as well as the extent to which they mediate the effects of age and sex on mortality.

Methods: This is a prospective population-based study in a nationally representative sample of elderly Costa Ricans. Baseline interviews occurred mostly in 2005 and mortality follow-up went through December 2010. Sample size after excluding observations with missing values: 2,313 individuals and 564 deaths.

Main outcome: prospective death rate ratios for 22 baseline biomarkers, which were estimated with hazard regression models.

Results: Biomarkers significantly predict future death above and beyond demographic and self-reported health conditions. The studied biomarkers account for almost half of the effect of age on mortality. However, the sex gap in mortality became several times wider after controlling for biomarkers. The most powerful predictors were simple physical tests: handgrip strength, pulmonary peak flow, and walking speed. Three blood tests also predicted prospective mortality: C-reactive protein (CRP), glycated hemoglobin (HbA1c), and dehydroepiandrosterone sulfate (DHEAS). Strikingly, high blood pressure (BP) and high total cholesterol showed little or no predictive power. Anthropometric measures also failed to show significant mortality effects.

Conclusions: This study adds to the growing evidence that blood markers for CRP, HbA1c, and DHEAS, along with organ-specific functional reserve indicators (handgrip, walking speed, and pulmonary peak flow), are valuable tools for identifying vulnerable elderly. The results also highlight the need to better understand an anomaly noted previously in other settings: despite the continued medical focus on drugs for BP and cholesterol, high levels of BP and cholesterol have little predictive value of mortality in this elderly population.

Figure 1

Crude effects of blood pressure and lipid biomarkers on death rate ratios (Costa Rican males aged 80+). Notes: The numbers within parentheses in the legends are the mean values. The diamonds indicate the cutoff levels of metabolic risk used in clinical practice. Although these curves are for males aged 80+ years, the curves for females and ages 60 to 79 years are similar, given that the interactions with age and sex in Table 2 are small and nonsignificant.

Figure 2

Controlled effects of normalized biomarkers on death rate ratios (elderly Costa Ricans).

Figure 3

ROC curves for models predicting death with estimated death hazards. G = Gini coefficient or proportional discrimination area (between the curve and the no-discrimination line).

Figure 4

Controlled effects of increasing biomarker levels by 1 SD from the mean values on death rate ratios by all causes and CV diseases.