Death and Cognition

A recurrent theme in aging research for the last 40 years is the extent to which years to live is a more prominent marker of functioning than years since birth (Kleemeier, 1962). Four decades of studies have indicated the central role of impending mortality for a variety of indicators. This research spans multiple laboratories across the world, research designs (e.g., cross-sectional vs. longitudinal; Bosworth & Siegler, 2002), statistical approaches (e.g., repeated measures ANOVA vs. Cox regression; Berg, 1996), and outcome measures including personality (Maier & Smith, 1999), affect (Eizenman, Nesselroade, Featherman, & Rowe, 1997), physiological indicators (e.g., event-related potentials, Van der Wal & Sandman, 1992; sleep parameters, Dew et al., 2003), and cognition (e.g., Bosworth & Schaie, 1999; Small & Bäckman, 1997; Wilson, Beckett, Bienias, Evans, & Bennett, 2003). Although the accumulated evidence suggests that the mortality-functioning link generalizes across multiple psychological and physiological domains, most extant research has focused on the association between cognitive performance and impending death. Two major concepts within this literature are terminal decline and terminal drop. Whereas the former denotes a gradual performance decrement as a function of proximity to death, the latter reflects a more precipitous death-related decline.

This phenomenon is important for theoretical and clinical reasons alike. Despite the demonstrated role of impending death as an indicator of cognitive performance and decline, this variable is rarely integrated into general theories of cognitive aging. This is an unfortunate omission because failure to account for time to death will inevitably result in an overestimation of age-related cognitive changes. Elderly adults are not a homogeneous group; time to death, preclinical dementia, and numerous other cognitively-relevant health-related indicators will shape performance trajectories (Bäckman, Small, Wahlin, & Larsson, 1999). Clinically, any abrupt change in cognitive performance, whether observed by a clinician or by a loved one, could merely reflect an acute illness (e.g., flu) or a protracted psychological condition (e.g., depression), but may also be indicative of accumulating underlying pathology (e.g., vascular disease) preceding death. In either case, such sudden changes should lead to careful scrutiny in a specialized setting.

Since the time of Kleemier (1962), considerable progress has been made toward understanding the link between cognition and mortality. Notwithstanding, many key issues remain unresolved. Several of these enduring questions were identified in a 1999 overview by Small and Bäckman. These include whether the effect of impending death on cognition: (1) decreases as a function of advancing age (Riegel & Riegel, 1972), (2) is more likely to be observed in cognitive tasks that are well preserved even up to age 70 (e.g., crystallized IQ, semantic memory; Rönnlund, Nyberg, Bäckman, & Nilsson, 2005) as opposed to tasks that exhibit an earlier and more marked age-related deterioration (e.g., fluid intelligence, episodic memory; see White & Cunningham, 1988), (3) are modulated by specific causes of death (e.g., vascular disease) or reflect a more global compromise of organ systems (i.e., biological vitality; Berg, 1996), and (4) are linked to causative factors (e.g., early-life cognitive ability, health-related conditions; Deary & Der, 2005). To be sure, the two latter points are intertwined. For example, some conditions (e.g., stroke) can be a source of both the terminal-decline deficit and an actual cause of death. In other instances (e.g., preclinical Alzheimer's disease), the relationship may be more indirect. Specifically, preclinical AD will affect the decline pattern, and although it does not represent a specific cause of death, it has immunological repercussions that increase the likelihood of dying (e.g., Jagger et al., 2000).

For several reasons, these pertinent issues in the literature on death and cognition can now be addressed in a comprehensive manner. First, the harvesting of mature longitudinal databases, spanning large age and time intervals, permit a more definitive test of Riegel and Riegel's (1972) hypothesis that terminal-decline effects are attenuated in very old age. Second, the richness of cognitive instrumentation in these large databases allows for determining whether mortality-related cognitive deficits are pervasive or specific to certain cognitive domains (White & Cunningham, 1988). Third, in these populations, there is rigorous assessment of various health conditions that can modify the mortality-cognition link, hence providing novel information pertaining to the issue of global breakdown vs. specific disease mediation. Finally, the widespread implementation of software for advanced statistical models of change (e.g., multilevel or hierarchical linear models; Bryk & Raudenbush, 1992; Singer & Willett, 2003) has obvious advantages for the analysis of terminal decline including: (1) a more accurate estimation of each individual's change trajectory in relation to time with disease (i.e., data from all individuals are examined independent of the number of assessment occasions, and individuals are not assumed to decline at the same rate), (2) differentiating terminal drop from decline (i.e., change can be specified to accelerate per year closer to death and is not assumed to be linear), and (3) examining the influence of global breakdown vs. disease states (within-person rates of decline can be examined in relation to systematic between-person differences in disease burden).

The five excellent and geographically diverse articles represented in this special section all provide unique and timely information pertaining to our understanding of terminal cognitive changes. The article by Anstey and colleagues (Anstey et al., 2006) reviews 47 longitudinal studies, examining mortality-cognition associations within three disease categories: stroke, cancer, and coronary heart disease. A chief objective is to determine whether a relationship between death and cognition could be documented within relatively homogeneous samples. This issue has an obvious bearing on the extent to which terminal cognitive deficits reflect specific disease states as opposed to a general breakdown of biological vitality. Using 20-year data from the Manchester Longitudinal Study, Rabbitt and colleagues (Rabbitt, Lunn, & Wong, 2006) address vitally important yet largely neglected issues in the terminal decline literature, namely the influence of dropout and practice on the strength of the mortality-cognition link. A critical feature of this research is the inclusion of multiple predictors (e.g., demographic, socioeconomic, intelligence) to examine their relative contribution to terminal cognitive decline, with the long time window enabling a glimpse of what cognitive performance would be like in the absence of pathology or death.

Sliwinski and colleagues (2006) make the important distinction between preterminal and terminal decline with a change-point model demarcating the border between two phases, using 25-year data from the Bronx Aging Study (BAS). This distinction resembles that between normative and nonnormative cognitive changes, with the former preterminal period largely reflecting age-normative effects and the latter terminal phase reflecting pathological influences. In a related article, Thorvaldsson and colleagues (Thorvaldsson et al., 2006) employed 29-year longitudinal data from the H70 study in Gothenburg, Sweden to investigate different ways of structuring time to account for terminal-decline effects. Specifically, the issue concerns whether the time metric of proximity to death provides a better account of variance in terminal decline than chronological age. Ghisletta and colleagues (Ghisletta et al., 2006) simultaneously combine longitudinal and survival models in analyzing 13-year follow-up data from the Berlin Aging Study (BASE). This novel, state-of-the-art statistical approach models individual change in cognitive performance as a predictor of mortality risk. The breadth of the cognitive battery and the extensive representation of variables from other domains of interest (e.g., sociodemographics, personality, sensory functioning) enables examining their relative importance in signaling impending death.