Trajectories of changes over twelve years in the health status of Canadians from late middle age

Abstract

Aging in a given individual can be characterized by the number of deficits (symptoms, signs, laboratory abnormalities, disabilities) that they accumulate. The number of accumulated deficits, more than their nature, well characterizes health status in individuals — the proportion of deficits present in an individual to deficits considered is known as a frailty index. While on average deficits accumulate with age, individual trajectories in the number of deficits is highly dynamic. Transitions in the number of deficits over a fixed time interval can be represented by the Poisson law, with the Poisson mean dependent on the deficit numbers at baseline. Here we present an extension of the model to make possible predictions for any given time period. Using data from the Canadian National Population Health Survey of people aged 55 and over (n = 4330), followed during 7 cycles being the baseline and 6 cycles of follow-up every 2 years, we found that the transition in the number of deficits during any time period can be approximated using a time dependent Poisson distribution with the Poisson mean tending to decelerate over time, according to square-root-of-time kinetics characteristic for stochastic processes (e.g. diffusion, Brownian motion ) while the probability of death shows a pattern of time acceleration with a high degree of precision, “explaining” over 98% of variance. The model predicts a variety of changes in health status including the possibility of health improvement indicating the repair/remodeling abilities of the organism. The model is valuable for estimating how changes in health can influence mortality across the life course from late middle age.

Introduction

Aging is intrinsically associated with the accumulation of impairments, illnesses, disabilities, and with increasing risks of adverse outcomes such as death and loss of independence. It is well established that aging in a given individual can be characterized by the number of deficits (symptoms, signs, laboratory abnormalities, and disabilities) that they accumulate (Fulop et al., 2010, Goggins et al., 2005, Kulminski et al., 2007, Mitnitski et al., 2001, 2005; Rockwood and Mitnitski, 2007, Yang and Lee, 2010, Yashin et al., 2007a). Aging develops gradually and starts from small changes in health (Kulminski et al., 2007, Merrill et al., 2008, Mitnitski et al., 2001) which accumulate across the adult life course (Rockwood et al., 2011). While many of the variables, when considered in isolation from each other, have only small effects on health, their cumulative effect becomes significant (Kulminski et al., 2007). These cumulative effects can be quantified by combining health related variables in a so-called “frailty index” (FI is also sometimes referred to as a “deficit index”). Such measures have been investigated in both epidemiological surveys and clinical databases (Dupre et al., 2009, Kulminski et al., 2006, Mitnitski et al., 2005, Rockwood and Mitnitski, 2007, Rockwood et al., 2011, Woo et al., 2006, Yang and Lee, 2010). These indices are found to be useful indicators of aging and good predictors of adverse outcomes such as worsening health (Fallah et al., 2011, Mitnitski et al., 2006), institutionalization (Rockwood et al., 2007) and death (Kulminski et al., 2007; Mitnitski et al., 2005; Yashin et al., 2007a). Most importantly, the properties of the FIs depend more on the number of deficits from which the FIs are comprised rather than on their nature (Rockwood et al., 2006, Rockwood et al., 2007).

The concept of frailty as deficit accumulation has been further developed in a transition model which summarizes changes in the number of deficits. Specifically, we have developed a stochastic multistate model of transitions in health states during fixed periods of time (Fallah et al., 2011, Mitnitski et al., 2006, Mitnitski et al., 2007a, Mitnitski et al., 2007b, Mitnitski et al., 2011). Of note, in contrast to typical regression models which estimate only the average effects, this model allows the simultaneous estimation of the probabilities of changes in individual health states in all directions: improvement, deterioration and death. Here we report the extension of this model to any time interval, so that we can make predictions about changes in health status and mortality when the time period is not fixed. We apply the model to a representative Canadian cohort of people from late middle age on, who were followed for up to 12 years. In addition to this methodological development, our goal was to investigate the time kinetics of changes in the parameters of the model, so as to elucidate the laws governing dynamics of health deficits during aging.