Early-adulthood Predictors of Mortality and Morbidity
Zachary Pincus, Ph.D.
Lay Summary: We propose to identify patterns in the way that human health changes from early adulthood to death, and to use these patterns to define ways of predicting an individual’s lifespan and future risk of disease. Such predictive tools may reflect factors that will eventually cause disease, or they may simply provide an early indication of the presence of disease. In the former case, the nature of the predictors we find will illuminate the causes of disease and may lead to new, lifespan-extending interventions. In the latter case, early predictors of known disease states are still extraordinarily useful. Clinically, early warning signs can be used to target individuals for additional care and intervention; actuarially, better predictors of lifespan and disease will enhance the ability of insurance pools to predict and manage risk.
Our proposed analysis will make use of data gathered by the long-running Framingham Heart Study, which was begun in 1948 and followed some 5000 individuals throughout their lives. Only recently has enough of the original Framingham cohort deceased to enable this proposed work, which requires both many decades of historical data and knowledge of the ultimate lifespans of those involved. Thus, we believe that this proposal is timely and may yield novel results.
In particular, we are interested in understanding how changes in simple physiological values over time relate to future lifespan and health. This topic has not been well studied: most “risk scores” combine multiple measurements (blood pressure, body mass index, etc.) made at one time to estimate the risk of a particular disease five or ten years hence. In contrast, we propose to investigate the utility of consulting several years’ or decades’ worth of history of any given clinical measurement in order to predict lifespan and health twenty or more years in the future.
The history of a given clinical measure will be useful in several different scenarios, which we will test. First, a measure’s rate of change over time may predict risk better than its current value. This is the case with a skin blemish: a small but rapidly growing lesion is much more dangerous than a large yet stable one. Alternately, perhaps total cumulative historical exposure, rather than current status, is the best predictor. This is certainly the case for cigarette smoking, for example.
Thus, our work will provide data on whether, for example, high blood pressure contributes to death and disease risk based only on its current value, cumulatively (like smoking), or only when changing (like skin blemishes). These results will enable better use of historical clinical data for longevity and disease risk estimation. Because such data are becoming increasingly available as part of routine practice due to the use of electronic medical records, our proposed analysis will have immediate practical implications.
We have found that approximately 10% of the variability in lifespan between individuals can be predicted while those individuals are still in their thirties, based on simple clinical parameters (body mass index, blood pressure, and blood glucose). Moreover, we find that later in life, incorporating the clinical history of these measures yields a better prediction of future lifespan or risk of death than does simply using their present value. In particular, summing the values over time to incorporate total past “exposure” to high blood glucose, high blood pressure, or obesity yields the strongest predictor. This implies that cumulative past damage from these conditions is not entirely erased by bringing blood glucose, blood pressure, or weight back under control. This work has been published as Zhang WB & Pincus Z, “Predicting all-cause mortality from basic physiology in the Framingham Heart Study”, Aging Cell (2015).
Subsequently, we attempted to use these measures to predict specific diseases rather than overall lifespan. Specifically, we examined cardiovascular mortality, time of onset of cardiovascular disease, and mental functional impairment. Unfortunately, we did not find that the clinical parameters we examined were of any more use in predicting these specific outcomes compared to overall lifespan. We also examined the possibility of creating a generic “frailty index” to measure overall participant well-being through time and compare that to future lifespan or disease risk. The fact that the Framingham Heart Study gathered dramatically different data over the decades of the study made this analysis challenging. We believe that the challenge is surmountable, but as a proof-of-concept, we first applied similar mathematical tools to analyzing health in a more homogenous longitudinal study: that of the roundworm C. elegans. We defined methods to aggregate multiple measures of physiological function into a single score of “health” in C. elegans, and then examined how health changes over time in long- vs. short-lived individuals. We found, surprisingly, that longer-lived individuals typically spend a larger fraction of their lives in “twilight” states of poor health / low physiological function. This work was published as Zhang WB, Sinha DB, Pittman WE, Hvatum E, Stroustrup N & Pincus Z, “Extended Twilight among Isogenic C. elegans Causes a Disproportionate Scaling between Lifespan and Health”, Cell Systems (2016).
We next plan to apply the statistical methods we perfected on our C. elegans data to the more challenging Framingham Heart Study data. We will use these methods to complete our efforts to create a “frailty index” from the Framingham data, allowing us to address two questions. First, we will determine the relationship between health/frailty and future lifespan / disease risk. Second, we will determine whether our findings in C. elegans regarding lifespan versus health hold in humans as well.
We intend to apply for a second year of LLF funding in 2017. We will also use our results in humans and C. elegans as the basis for an aim in an NIH R01 application.