Capturing Processing Speed Changes

Selective cognitive decline is a robust phenomenon in aging (Craik & Salthouse, Reference Craik and Salthouse2008). Although well-practiced abilities and acquired knowledge show little decline until very late in life, mechanisms that underlie the efficiency of cognition, such as processing speed, executive function, and episodic memory, follow a linear decline that may start already in young adulthood (Hedden & Gabrieli, Reference Hedden and Gabrieli2004; Salthouse, Reference Salthouse2009, Reference Salthouse2010; although see Nilsson, Sternang, Ronnlund, & Nyberg, Reference Nilsson, Sternang, Ronnlund and Nyberg2009). Processing speed appears to be particularly sensitive to age and mediates at least some of the age-related declines in higher-level cognition (Salthouse, Reference Salthouse1996b; Verhaeghen & Salthouse, Reference Verhaeghen and Salthouse1997). Salthouse (Reference Salthouse1996b) proposed that reduced processing speed represents a core deficit in aging, which has stimulated a great interest in methods that can maximize the information gained from processing speed tasks (Balota & Yap, Reference Balota and Yap2011).

One important development in the characterization of processing speed is the increased appreciation of intra-individual variability of reaction times (RT) at a trial-to-trial level (RT inconsistency; Dykiert, Der, Starr, & Deary, Reference Dykiert, Der, Starr and Deary2012; Hultsch & Macdonald, Reference Hultsch and Macdonald2004). Above and beyond the average slowing of RTs (Verhaeghen & Salthouse, Reference Verhaeghen and Salthouse1997), RT inconsistency increases reliably in old age (Hultsch, MacDonald, & Dixon, 2002; MacDonald, Hultsch, & Dixon, Reference MacDonald, Hultsch and Dixon2003; Verhaeghen & Salthouse, Reference Verhaeghen and Salthouse1997). The importance of RT inconsistency as a cognitive construct has been demonstrated by evidence linking it to impairments in overall cognitive ability (Hultsch et al., Reference Hultsch, Macdonald and Dixon2002; MacDonald et al., Reference MacDonald, Hultsch and Dixon2003), physical function (Strauss, MacDonald, Hunter, Moll, & Hultsch, Reference Strauss, MacDonald, Hunter, Moll and Hultsch2002), and even to impending death (MacDonald, Hultsch, & Dixon, Reference MacDonald, Hultsch and Dixon2008). Consequently, RT inconsistency has been proposed to reflect a measure of general neurological integrity at a biological level (Hultsch & Macdonald, Reference Hultsch and Macdonald2004; Li, Lindenberger, & Sikstrom, Reference Li, Lindenberger and Sikstrom2001) and of general processing efficacy, including attentional and executive control, at a cognitive level (Duchek et al., Reference Duchek, Balota, Tse, Holtzman, Fagan and Goate2009; Schmiedek, Oberauer, Wilhelm, Suss, & Wittmann, Reference Schmiedek, Oberauer, Wilhelm, Suss and Wittmann2007).

The measurement of RT inconsistency warrants some methodological consideration. Since changes in RT inconsistency in aging are often accompanied by an average slowing of RTs, it is important to ensure that any RT inconsistency increase is not driven by changes in average RT (Hultsch et al., Reference Hultsch, Macdonald and Dixon2002). The coefficient of variation, which constitutes the intra-individual standard deviation divided by the intra-individual mean, has emerged as the standard measure of RT inconsistency (Jackson, Balota, Duchek, & Head, Reference Jackson, Balota, Duchek and Head2012). However, the suitability of this approach can be questioned based on the usual positive skew of RT distributions and the inherent assumption of distributional symmetry in the standard deviation (Balota & Yap, Reference Balota and Yap2011).

By fitting empirical RT data to an Ex-Gaussian function, it is possible to characterize dissociable components of the distribution and thereby capture RT inconsistency more fully (Hultsch & Macdonald, Reference Hultsch and Macdonald2004). Relative to other Ex-Gaussian parameters (mu, sigma), the parameter that describes the “slow tail” of the distribution (tau) appears to be a better predictor of general processing efficacy, as measured by tasks of working memory, reasoning and processing speed (Schmiedek et al., Reference Schmiedek, Oberauer, Wilhelm, Suss and Wittmann2007), and to better discriminate between age groups (Spieler, Balota, & Faust, Reference Spieler, Balota and Faust1996; West, Murphy, Armilio, Craik, & Stuss, Reference West, Murphy, Armilio, Craik and Stuss2002). Individual components of the distribution can also be studied directly by rank ordering and plotting RTs for individual participants as a function of condition, as in Quantile and Vincentile plots (Balota & Yap, Reference Balota and Yap2011). It is important to emphasize, however, that only computationally explicit models can be assumed to be commensurate with well-defined cognitive processes (Matzke & Wagenmakers, Reference Matzke and Wagenmakers2009; see diffusion model by Ratcliff, Van Zandt, & McKoon, Reference Ratcliff, Van Zandt and McKoon1999).

Capturing White Matter Changes

WM has been identified as an important contributor to processing speed slowing in aging (Madden et al., Reference Madden, Whiting, Huettel, White, MacFall and Provenzale2004). Whereas gray matter consists mainly of neuronal cell bodies, WM consists mainly of neuronal projections, which are coated with myelin to ensure efficient neural communication (Madden, Spaniol, et al., Reference Madden, Spaniol, Costello, Bucur, White, Cabeza and Huettel2009). Consequently, WM decline could contribute to a breakdown of the efficacy of neural communication and, ultimately, speed of processing.

Further to a reduction of WM volume (Jernigan et al., Reference Jernigan, Archibald, Fennema-Notestine, Gamst, Stout, Bonner and Hesselink2001; Raz et al., Reference Raz, Lindenberger, Rodrigue, Kennedy, Head, Williamson and Acker2005), old age is associated with an increased prevalence of white matter hyperintensities (WMH), which are best detected as areas of brightness on FLAIR magnetic resonance imaging (de Groot et al., Reference de Groot, de Leeuw, Oudkerk, van Gijn, Hofman, Jolles and Breteler2000). The increase in WM hyperintensity prevalence has primarily been observed in periventricular and frontal areas (Raz, Rodrigue, Kennedy, & Acker, Reference Raz, Rodrigue, Kennedy and Acker2007; Ylikoski et al., Reference Ylikoski, Erkinjuntti, Raininko, Sarna, Sulkava and Tilvis1995) and has shown robust associations with age-related cognitive decline (Gunning-Dixon & Raz, Reference Gunning-Dixon and Raz2000). In the context of processing speed specifically, negative associations have been demonstrated with speeded performance in older adults (Bunce et al., Reference Bunce, Anstey, Christensen, Dear, Wen and Sachdev2007; de Groot et al., Reference de Groot, de Leeuw, Oudkerk, van Gijn, Hofman, Jolles and Breteler2000).

An important limitation of volumetric and lesion techniques is that inferences are limited to regions affected by atrophy or WMH. Diffusion tensor imaging (DTI) provides a sensitive measure of microstructural properties and spatial organization of WM even when no significant disease is present (Madden, Bennett, & Song, Reference Madden, Bennett and Song2009). Fractional anisotropy (FA) represents the most common DTI index and approximates the ratio between diffusion of water molecules parallel to the tract (axial diffusivity) and perpendicular to the tract (radial diffusivity). In line with the lesion findings, DTI investigations have revealed an age-related reduction of FA, particularly in frontal and parietal tracts, with greater effects in anterior compared to posterior segments within the fiber tracts (Davis et al., Reference Davis, Dennis, Buchler, White, Madden and Cabeza2009; Gunning-Dixon et al., Reference Gunning-Dixon, Brickman, Cheng and Alexopoulos2009). A link between WM microstructure and cognitive decline has also been demonstrated, with particularly consistent effects in the domains of processing speed and executive function (Madden, Bennett, et al., Reference Madden, Bennett and Song2009).

Beyond average processing speed, RT inconsistency increases have been proposed to reflect the increased neural noise and less distinct cortical representations that could result from WM decline (MacDonald, Li, & Backman, Reference MacDonald, Li and Backman2009). Indicative of such a relationship, RT inconsistency appears to follow a U-shaped function across the lifespan (Williams, Strauss, Hultsch, & Hunter, Reference Williams, Strauss, Hultsch and Hunter2007), which approximates the inverted U-function of WM volume (Gogtay et al., Reference Gogtay, Giedd, Lusk, Hayashi, Greenstein, Vaituzis and Thompson2004). Furthermore, both reduced WM volume and an increased prevalence of WMH in older adults have been associated with measures of increased RT inconsistency, independent of mean RT, with promising results for the tau parameter of the Ex-Gaussian distribution (Bunce et al., Reference Bunce, Anstey, Christensen, Dear, Wen and Sachdev2007; Jackson et al., Reference Jackson, Balota, Duchek and Head2012; Walhovd & Fjell, Reference Walhovd and Fjell2007). Support for RT inconsistency as a reliable proxy of WM integrity across the lifespan has also started to emerge in DTI investigations (Fjell, Westlye, Amlien, & Walhovd, Reference Fjell, Westlye, Amlien and Walhovd2011; Tamnes, Fjell, Westlye, Ostby, & Walhovd, Reference Tamnes, Fjell, Westlye, Ostby and Walhovd2012).

Importantly, histopathological studies validate the neuroimaging findings by demonstrating age-related alterations in several aspects of WM, including axons, myelin and glia (Peters, Reference Peters2002). However, since such cellular components are not easily distinguished in vivo, the relative influence of the neurobiological mechanisms underlying DTI and WM hyperintensity findings remains largely unknown (Assaf & Pasternak, Reference Assaf and Pasternak2008; Maillard et al., Reference Maillard, Carmichael, Fletcher, Reed, Mungas and DeCarli2012). Some mechanistic information may be provided by the radial diffusivity measure in DTI, which has been linked to myelin-related effects in animal studies (Song et al., Reference Song, Sun, Ramsbottom, Chang, Russell and Cross2002). In aging, FA reductions are commonly accompanied by more prominent changes in radial diffusivity relative to axial diffusivity (Bennett, Madden, Vaidya, Howard, & Howard, Reference Bennett, Madden, Vaidya, Howard and Howard2010; Davis et al., Reference Davis, Dennis, Buchler, White, Madden and Cabeza2009) and indices of myelin breakdown, including radial diffusivity, have been found to be closely related to processing speed (Jacobs et al., Reference Jacobs, Leritz, Williams, Van Boxtel, van der Elst, Jolles and Salat2013; Lu et al., Reference Lu, Lee, Tishler, Meghpara, Thompson and Bartzokis2013; although see Burgmans et al., Reference Burgmans, Gronenschild, Fandakova, Shing, van Boxtel, Vuurman and Raz2011). Thus, although a mechanistic account for the WM change that underpins age-related slowing is largely incomplete, demyelination represents a likely contributor.

The Issue of Causality

As the evidence above suggests, correlational data often represent the primary means of investigating interrelations of age, brain structure and cognition. Although a correlation cannot prove causation, the implications of causal hypotheses can be examined with correlational data (Salthouse, Reference Salthouse2011). In the present context, the disconnection hypothesis represents the dominant causal hypothesis and postulates that WM changes mediate age-related cognitive decline (O'Sullivan et al., Reference O'Sullivan, Jones, Summers, Morris, Williams and Markus2001). According to the Age-Brain-Cognition triangle proposed by Salthouse (Reference Salthouse2011), the implication of such a hypothesis would be a reduction or elimination of the correlation between age and processing speed when variation in WM is statistically controlled. Consistent with this, a reduction in age-related variance of processing speed has been demonstrated when a WM index is added to hierarchical regression models (Rabbitt et al., Reference Rabbitt, Scott, Lunn, Thacker, Lowe, Pendleton and Jackson2007; Salami, Eriksson, Nilsson, & Nyberg, Reference Salami, Eriksson, Nilsson and Nyberg2012). Similarly, structural equation modeling has provided support for a mediating role for WM in age-related slowing (Burgmans et al., Reference Burgmans, Gronenschild, Fandakova, Shing, van Boxtel, Vuurman and Raz2011; although see Charlton et al., Reference Charlton, Landau, Schiavone, Barrick, Clark, Markus and Morris2008).

Although promising, an important caveat of mediation analyses is that because causality cannot be tested directly, examination of competing models is critical for enhancing confidence in the hypothesized model (Penke & Deary, Reference Penke and Deary2010; Salthouse, Reference Salthouse2011). For example, WM-cognition relationships are commonly reduced when age is controlled (Madden et al., Reference Madden, Bennett, Burzynska, Potter, Chen and Song2012), which would be suggested by a model in which WM and cognition are coincidentally related given independent relations to age. Examination of competing models is therefore critical for evaluating the plausibility of the causal connections postulated by the disconnection hypothesis.

The Issue of Specificity

The widespread WM change in aging necessarily has broad effects on cognition (Gunning-Dixon & Raz, Reference Gunning-Dixon and Raz2000). Nevertheless, the anterior-posterior gradient in WM decline and the particularly close relationship between WM and processing speed and executive function in old age suggest a degree of specificity (Davis et al., Reference Davis, Dennis, Buchler, White, Madden and Cabeza2009; Madden et al., Reference Madden, Bennett, Burzynska, Potter, Chen and Song2012). Comprehensive cognitive test batteries therefore remain essential for testing the domain specificity of any detected WM relationship. In the case of RT inconsistency, further mapping of the relationships with other domains has particular importance given its proposed role as a proxy of general processing efficacy (Schmiedek et al., Reference Schmiedek, Oberauer, Wilhelm, Suss and Wittmann2007).

Another important consideration is the degree to which a chosen processing speed task measures processing speed per se (Albinet, Boucard, Bouquet, & Audiffren, Reference Albinet, Boucard, Bouquet and Audiffren2012). The nature of the processing speed tasks in the context of age-related WM changes has ranged from visuospatial processing (Kerchner et al., Reference Kerchner, Racine, Hale, Wilheim, Laluz, Miller and Kramer2012) to psychomotor function (Charlton et al., Reference Charlton, Landau, Schiavone, Barrick, Clark, Markus and Morris2008) and various higher-order cognitive functions (Sasson, Doniger, Pasternak, Tarrasch, & Assaf, Reference Sasson, Doniger, Pasternak, Tarrasch and Assaf2013). Although this demonstrates the reliability of age-related slowing, it does not specify processing speed as the driving factor (Cepeda, Blackwell, & Munakata, Reference Cepeda, Blackwell and Munakata2013).

Fundamental processing speed is defined as the speed at which other cognitive operations can be executed, which indicates an independence from higher-level cognitive operations and motor operations (Salthouse, Reference Salthouse1996a). The age-related declines in executive function and motor speed (Joy, Fein, Kaplan, & Freedman, Reference Joy, Fein, Kaplan and Freedman2000; Salthouse, Reference Salthouse1993) and their associations with WM (Gunning-Dixon & Raz, Reference Gunning-Dixon and Raz2000; Sachdev, Wen, Christensen, & Jorm, Reference Sachdev, Wen, Christensen and Jorm2005), therefore, make a strong case for controlling for such confounding factors. When such separation is lacking, as in the case of complex processing speed tasks, WM findings may reflect associations with executive function rather than fundamental processing speed per se (Salami et al., Reference Salami, Eriksson, Nilsson and Nyberg2012). Similarly, since most tests of processing speed involve a motor response, motor slowing is likely to contribute to demonstrated relationships.

To capture processing speed independently of motor speed and executive processes, multiple speeded tasks can be combined in a composite score (Salthouse, Reference Salthouse1996a). By analyzing the commonalities of tasks with variable methodologies (e.g., motor requirements) and scope (e.g., cognitive load/domain), the contributions from individual task demands are minimized in favor of the processing speed component of interest (Cepeda et al., Reference Cepeda, Blackwell and Munakata2013). An alternative or complementary approach is to include additional task conditions to allow processing speed to be dissociated from confounding variables by a method of subtraction. As an example, the digit-symbol substitution test can be divided into influences of motor speed, perceptual speed, memory and higher-level coding processes (Joy et al., Reference Joy, Fein, Kaplan and Freedman2000; Joy, Kaplan, & Fein, Reference Joy, Kaplan and Fein2004). Demonstrating the relevance of subtraction in the present context, Madden, Spaniol, et al. (Reference Madden, Spaniol, Costello, Bucur, White, Cabeza and Huettel2009) demonstrated that the mediating role of WM was limited to a decisional component in a task-switching paradigm. Thus, to situate processing speed and RT inconsistency within the broader cognitive hierarchy and to evaluate their potential superiority over other cognitive variables in indicating age-related WM change tasks selection will be of utmost importance.