Behavioural Science Section / Subjective Memory Complaints: Their Nature, Correlates and Consequences Gerontology 2015;61:223–231 DOI: 10.1159/000369927

Received: August 5, 2014 Accepted: November 17, 2014 Published online: March 19, 2015

Subjective and Objective Memory Changes in Old Age across Five Years Daniel Zimprich Tanja Kurtz Institute of Psychology and Education, Ulm University, Ulm, Germany

Abstract Typically, subjective memory assessments (be it in form of single items or questionnaires) in old age only weakly correlate with the performance in objective memory tests at cross-section. It thus appears as if individual differences in subjective memory assessments hardly reflect individual differences in memory in old age. A shortcoming of cross-sectional studies, however, is that subjective assessments may rely on different individual standards, which are not taken into account. One solution to this problem has been to investigate subjective and objective memory longitudinally, thereby focusing on individual differences in intraindividual changes. Results from studies using this approach have been mixed, with some studies showing a significantly stronger relation between changes than between levels, and other studies showing no such significant difference. Using data from the Zurich Longitudinal Study on Cognitive Aging (n = 236), we find that 5-year changes in subjective assessments of memory capacity and memory changes correlate with objective memory changes of 0.54 and –0.44, respectively.

© 2015 S. Karger AG, Basel 0304–324X/15/0613–0223$39.50/0 E-Mail [email protected] www.karger.com/ger

These correlations are significantly stronger than at crosssection. After controlling for age, depressive affect, and subjective health at the first measurement occasion, correlations are slightly attenuated, but the basic findings remain the same. © 2015 S. Karger AG, Basel

Introduction

Is a person’s self-judgment of her memory associated with her objective memory performance? This straightforward question is motivated by the observation that objective memory performance declines in old age [1] whereas self-judgments of memory expressed as memory complaints increase in old age [2]. Given these findings, one might conjecture that, especially in old age, there is a relation between a declining objective memory performance and a person’s self-judgment of her memory. Cross-sectional studies have, indeed, provided support for such a relation, but the size of the relation is small at best. For example, Pearman and Storandt [3] found that objective measures of episodic memory were virtually unrelated to subjective memory. Similarly, Zelinski et al. [4] reported only a modest association between memory perDaniel Zimprich Department of Developmental Psychology Institute of Psychology and Education, Ulm University Albert-Einstein-Allee 47, DE–89081 Ulm (Germany) E-Mail daniel.zimprich @ uni-ulm.de

Downloaded by: Milton S.Hershey Medical Ctr. 198.143.32.33 - 1/27/2016 3:52:34 AM

Key Words Subjective memory · Memory complaints · Correlated change · Individual differences

224

Gerontology 2015;61:223–231 DOI: 10.1159/000369927

tween-person differences in within-person change can be examined. The critical outcome in longitudinal studies then consists of the between-person relation of objective memory performance changes and subjective memory changes across time.1 Several longitudinal studies have used this approach, although with different outcomes. Taylor et al. [14] found no statistically significant relation between subjective and objective memory changes. However, their study may have been underpowered due to a relatively small sample size. By contrast, using a latent change model, Zimprich et al. [13] reported a correlation of r = –0.50 between 4-year changes in subjective complaints and changes in memory performance, which was significantly larger than the correlation at cross-section (r = –0.25). Martin and Zimprich [12] showed that 4-year cognitive complaint changes and fluid intelligence changes were strongly correlated (r = –0.64), while at cross-section the correlation was small (r = –0.11). Similarly, Parisi et al. [15] found that, across 5 years, changes in subjective memory were related to changes in objective memory performance (r = 0.44). Across a longer time span, Mascherek and Zimprich [16] reported that 12-year changes in memory complaints were correlated to changes in memory performance (r = –0.39). To summarize, it appears as if the subjective-objective memory relation is stronger when longitudinal changes are examined rather than cross-sectional differences. More recently, however, Pearman et al. [17] have reported that no correlation could be estimated between subjective and objective memory changes. This was due to the unanticipated finding that there was no statistically significant variance in subjective memory changes across 6 years in a relatively large sample (n = 504) – which was subject to considerable attrition though. There are, however, some methodological limitations that may in part explain this result. The authors applied an age-convergence model, which represents a mixture of a cross-sectional and a longitudinal model. Although these models are useful in combining the longitudinal data of multiple age cohorts in describing the average trajectory of a variable (if conver-

1

Note that the longitudinal approach implicitly assumes that within a person the standards for evaluating one’s memory remain constant over time. Although this assumption may seem unrealistic, it is not indispensable for the longitudinal approach to be more ‘successful’ than the cross-sectional approach. It is sufficient to adopt a weaker assumption, namely, that withinperson changes of memory evaluation standards are smaller than betweenperson differences of memory evaluation standards at cross-section. We return to this issue in the Discussion.  

Zimprich/Kurtz

Downloaded by: Milton S.Hershey Medical Ctr. 198.143.32.33 - 1/27/2016 3:52:34 AM

formance and self-ratings of memory in a nationally representative sample of the oldest old. Recently, previous cross-sectional studies examining the subjective-objective memory relation have been combined in a meta-analysis [5]. Across studies, the average correlation between objective and subjective memory was 0.15. For other forms of memory, e.g., prospective memory, similar findings have been reported [6]. Although this correlation is not nil, it is too small to support the idea that subjective memory might serve as a (personal) diagnostic means for objective memory [7]. The fact that the relation between subjective and objective memory is small in the general population (of older adults) does not rule out the possibility that there are subpopulations in which this relation is stronger. One might, for instance, assume that older persons who show a more pronounced decline in memory are more accurate in their subjective memory judgments. However, Kliegel et al. [8], for example, have shown that objective memory performance was unrelated to subjective memory complaints in elderly diagnosed with aging-associated cognitive decline [9]. Reversing the coin, one may suspect that in those elderly who report pronounced memory complaints, accuracy is higher with regard to objective performance. However, Mascherek et al. [10], for example, have demonstrated that in a sample of memory clinic outpatients, cognitive functioning was unrelated to cognitive complaints. Instead of defining subpopulations a priori, there are statistical analysis approaches that aim at dividing a total sample into subsamples. Kliegel and Zimprich [11] used a mixture regression approach to examine cognitive complaints in older adults, which led to two distinct subgroups. In the smaller subgroup, cognitive performance and memory emerged as significant predictors of subjective cognitive complaints, also after controlling for depressive symptoms and neuroticism. Although such findings are promising, one limitation is that results based on data-driven analytic approaches are difficult to replicate. A different approach to examine the relation between subjective and objective memory relies on longitudinal data [12, 13]. In longitudinal data, individual differences between persons are controlled for. This seems even more important in terms of subjective assessments of one’s own memory functioning. One of the main intricacies in conjunction with these (and other) self-judgments is that different persons might rely on different standards to arrive at a judgment. Thus, the subjective-objective memory relation may be low because people use different standards to subjectively evaluate their memory. Longitudinally, instead of focusing on between-person differences, be-

gence assumptions are met, see [18, 19]), it is not clear how such converge assumptions could be tested for random effects. Moreover, the measure of subjective memory consisted of three 3-point Likert-type scale items and the authors did not report a reliability estimate of their composite score. Alternatively, the outcome variable could have been treated as being ordered-categorical [20]. Although in a second set of analyses, these 3 items (plus an additional item) were used as indicators of a latent variable, the according factor loadings were not reported, making it difficult to judge the quality of the so-defined latent variable. The goal of the present study was to add to the literature on the subjective-objective memory relation in old age. We used two measures of subjective memory, established strong longitudinal measurement invariance for the memory and subjective memory, and modeled the individual Likert-type scaled items tapping subjective memory as ordered-categorical. In these respects, our study is methodologically more precise than previous studies. The 5-year longitudinal data we use offer the possibility to include memory performance and two measures of subjective memory, namely, subjective memory capacity and subjective memory change. In accordance with previous studies from our lab [12, 13, 16], a latent variable approach will be used. Compared to mixed models, latent change models also allow for testing the properties of the measurement instruments. A key concern in modeling longitudinal changes in psychological variables is whether indicators of an underlying latent construct mean the same thing across time. In order to ensure that the same psychological construct operates in the same way at different measurement occasions, strong measurement invariance has to be established [21]. Meredith and Horn [22], in particular, have argued that the importance of measurement invariance as a tool in developmental psychology can hardly be overstated – even more so, one might add, if the ‘signal’ compared to ‘noise’ is small, such as in single-item indicators. In the analyses presented below, the measurement of memory complaints was based on individual Likert-type scaled items, which served as indicators for latent variables. If such Likert-type scaled items are factor-analyzed as if they were continuous or intervalscaled, there may be a critical mismatch between the information represented by the numbers assigned to the Likert-type scales and the nature of the factor model parameters on which statistical tests are based [23]. Besides the limitations arising from such a levels-of-measurement perspective, another problem associated with Li-

kert-type scaled items is that, frequently, they show departures from both univariate and multivariate normality. Previous studies have shown that this typically results in considerable negative bias of parameters and standard errors [24]. A methodologically more sound approach is to treat Likert-type scaled items as ordered-categorical [25], which we did in our investigation of the subjectiveobjective memory relation. Apart from objective memory performance, which cross-sectionally appears to be only weakly related to subjective memory measures, there are other variables that may influence subjective memory at cross-section. Typically, at cross-section, depressive affect and neuroticism are correlated with subjective memory [8, 11]. These relations may be explained by the fact that a person’s affective state colors the subjective evaluation of her memory performance [26]. A reversed causal mechanism has also been suggested, where depressive affect is considered a reaction to subjectively perceived memory performance deficits [27]. Longitudinally, Zimprich et al. [13] have shown that changes in depressive affect are related to subjective memory changes. We will not go into detail regarding the affect-subjective memory linkage here, but for the purposes of the present study, we included depressive affect as a control variable. In addition, we included subjective health as a general indicator of an individual’s concerns and attitudes about health and illness (of which memory problems might form one part) as a control variable.

Subjective and Objective Memory Changes in Old Age

Gerontology 2015;61:223–231 DOI: 10.1159/000369927

Methods

225

Downloaded by: Milton S.Hershey Medical Ctr. 198.143.32.33 - 1/27/2016 3:52:34 AM

Data were taken from the Zurich Longitudinal Study on Cognitive Aging (ZULU), a study conducted in the metropolitan region of Zurich, Switzerland. At the first measurement occasion (T1), the sample comprised 364 older adults born between 1925 and 1940. On average, participants were 73 years old at T1 (SD = 4.4 years), and 46% were female. The majority of the sample was married and resided with others. On average, participants reported about 13 years of formal education. There were no signs of cognitive impairment, as indicated by a mean value of 28.93 on the Mini-Mental State Examination (MMSE) [28], or pronounced depressive affect, as indicated by a mean value of 1.63 on a short form of the Geriatric Depression Scale (GDS) [29]. Subjective health was mostly judged as ‘good’. In addition, participants did not report any severe hearing or vision difficulties as judged on a 6-point Likert-type scale (mean = 4.6). For a more detailed description regarding the recruitment process of the baseline sample, see Zimprich et al. [30]. At the second measurement occasion (T2), about 18 months after T1, 336 participants returned for a second assessment. Finally, about 5 years (5.24 years, ranging from 5.07 to 5.49 years) after T1, 236 older adults participated in a third assessment (T3).

In the analyses reported below, we focus on 5-year longitudinal changes between T1 and T3.2 Hence, we only included those older adults who had complete data for T3 (n = 236). These individuals were, on average, 72.6 years old at T1 (SD = 4.4 years), and 45% were female. Although an extensive analysis on missing data is beyond the scope of the present paper, the 236 older adults who participated at T3 differed from those 127 who did not return for T3 in that they were slightly younger (Cohen’s d = –0.20), somewhat better educated (Cohen’s d = 0.25), showed a slightly better memory performance at T1 (average Cohen’s d = 0.25), and reported less symptoms of depressive affect (Cohen’s d = –0.32). With respect to subjective memory complaints, there were no statistically significant differences at T1 (Cohen’s d = 0.11 for capacity and Cohen’s d = 0.08 for change). Thus, as is typical in longitudinal studies on cognitive aging, the longitudinal ZULU sample represents a positive selection of the original sample, although effect sizes were, in general, small. Measures Memory Story Recall. This task consisted of story A of the Logical Memory subtest of the German version of the Wechsler Memory ScaleRevised (WMS-R) [31]. The 66-word story was read by an experimenter during 60 s. Participants were asked to listen closely and, when the story was finished, to immediately recall as many details as possible in any order. Scored was the number of correctly recalled propositions (possible range: 0–25). The test-retest reliability of the story recall test is 0.79 [31]. At T2 story B was used, and at T3 story A was administered again. Picture Recall. This task consisted of 12 pictures taken from the Nuremberg Age Inventory (Nürnberger-Alters-Inventar) [32]. For each item, a picture of a simple object was displayed for 2.75 s and participants were required to name the shown object aloud (e.g., ‘apple’). Followed by a pause of 1 s, the next picture was displayed. Immediately after presenting all 12 pictures, participants were asked to recall as many of the objects as possible. Scored was the number of correctly recalled objects (possible range: 0–12). The test-retest reliability of the picture recall task is 0.67 [32]. At T3, the same set of pictures was used as at T1. Verbal Memory. Verbal memory (VM) was assessed by five consecutive trials of a word list recall task. The task comprised 27 meaningful but unrelated 2- to 3-syllabic words that were taken from the Handbook of German Word Norms [33]. The 27 words appeared on a computer screen at a rate of 2 s each and participants were required to read them aloud. After the presentation of all 27 words, participants were asked to recall as many words as possible. This procedure was repeated five times, with a different order of word presentation for each trial. At each trial, the number of correctly recalled words was scored (possible range: 0–27). For the purposes of the present study, we selected the third VM score, that is, the number of words recalled at the third study recall cycle.3 The word lists were identical at T1 and T3. 2

The data from T2 were not used in the present study because this would have increased the number of analysis variables considerably, while at the same time adding only marginally to the change variance (e.g., in a linear change model, T2 would only contribute 9% to the change variance). 3 The reasons for selecting the third repetition of the VM score were that (1) it had a larger variance than the score of the first repetition, implying that it differentiated better between persons and (2) it had the highest average correlation with the four scores of the other repetitions, thus representing VM best.

Subjective Memory Subjective memory was measured using items of two subscales from the abridged Metamemory in Adulthood (MIA) questionnaire [34, 35]. In ZULU, the abridged MIA subscales tapping memory capacity and change were administered. Memory Capacity. The subscale Memory Capacity (MCap) of the abridged MIA is designated to assess a person’s knowledge of her/his memory capacity as evidenced by a predictive report of performance on a given task. A sample item is ‘I am good at remembering conversations that I have had’. In total, the subscale comprises 12 items, which participants are asked to answer on a 5-point Likert-type scale ranging from 1 = ‘agree strongly’ to 5 = ‘disagree strongly’. For the purposes of the present study, we selected those 6 items that showed the highest factor loadings. The 6 items selected correspond to items No. 49, 52, 88, 100, 104, and 106 in the original MIA [34]. The items were reversed such that for all of them higher scores indicated a larger subjective memory capacity. A sum score of the 6 items could range between 6 and 36. Memory Change. The subscale Memory Change (MCha) of the abridged MIA measures a person’s perception of memory abilities as generally stable or subject to long-term decline. A sample item is ‘The older I get the harder it is to remember clearly’. In total, the subscale comprises 10 items, which participants are asked to answer on a 5-point Likert-type scale ranging from 1 = ‘agree strongly’ to 5 = ‘disagree strongly’. For the purposes of the present study, we selected those 6 items that showed the highest factor loadings. The 6 items selected correspond to items No. 16, 28, 30, 56, 58, and 89 in the original MIA [34]. The items were reversed such that for all of them higher scores indicated a more pronounced subjective memory decline. A sum score of the 6 items could range between 6 and 36. Control Variables Depressive Affect. Depressive affect was measured using the short form of the Geriatric Depression Scale (GDS) [29]. The short form of the GDS contains 15 questions (e.g., ‘Do you feel that your life is empty?’), which participants are asked to answer using a yes/ no response format. Participants’ answers were combined to a sum score (possible range: 0–15), where higher scores indicate more pronounced depressive affect. In the present sample, Cronbach’s alpha was 0.79 at T1. Subjective Health. Subjective health was measured using a single item at T1. Participants were asked to rate their health on a 6-point Likert-type scale, ranging from 1 = ‘poor’ to 6 = ‘excellent’. Both depressive affect and subjective health were measured at T1 and T3. However, only T1 values were included in the analyses. In doing so, individual differences in the objective-subjective memory relations were controlled for due to initial individual differences in depressive affect and subjective health. Of course, there might also be a change in depressive affect and subjective health across 5 years, which, however, would be more difficult to model because there is only one manifest indicator of both variables. Although possible [20], this was beyond the scope of the present article.

 

Gerontology 2015;61:223–231 DOI: 10.1159/000369927

Zimprich/Kurtz

Downloaded by: Milton S.Hershey Medical Ctr. 198.143.32.33 - 1/27/2016 3:52:34 AM

226

Analysis Approach We used latent change models [36] to address the goals of the present study. In latent change models, the level of a latent construct (e.g., memory performance at T1) and the change of this latent construct over time (e.g., memory performance changes between T1 and T3) are estimated. More precisely, if (1) the indica-

Fig. 1. Latent change model of subjective

and objective memory. Mem = Objective memory performance; ΔMCap = change in subjective memory capacity; ΔMCha = change in subjective memory change; ΔMem = change in objective memory performance; DA = depressive affect; SH = subjective health; SR = story recall; PR = picture recall. Variables in grey are ordered-categorical. Numbers refer to item numbers of the MIA, and parameters are standardized.

Subjective and Objective Memory Changes in Old Age

Results

In reporting our results, we focus on the final models estimated (more detailed analyses are available from the authors upon request). Descriptive statistics of the analysis variables are shown in table 1. After having established strong measurement invariance for all indicators of memory, subjective memory capacity, and subjective memory change across time, a latent change score model was estimated. This model showed an acceptable fit [χ2 = 966, d.f. = 443, p < 0.01, CFI = 0.979, RMSEA = 0.071 (90% CI = 0.065–0.077)].4 On average, the subjectively assessed memory capacity of the 236 older participants decreased significantly across the 5 years of longitudinal follow-up (mean = –0.18). In terms of effect size, this factor mean change reflected a small effect (Cohen’s d = –0.21), indicating that participants perceived their memory capacity as having become somewhat smaller at T3 compared to T1. In contrast, the factor mean of the subjectively assessed memory change increased significantly over time (mean = 0.37), implying that participants perceived a more pronounced memory decline at T3 compared to T1. In terms of effect size, this 4 To compare, a model of configural invariance produced χ2 = 871, d.f. = 383, p < 0.01, CFI = 0.981, RMSEA = 0.074, whereas a model of weak measurement invariance obtained χ2 = 882, d.f. = 395, p < 0.01, CFI = 0.981, RMSEA = 0.072.  

Gerontology 2015;61:223–231 DOI: 10.1159/000369927

227

Downloaded by: Milton S.Hershey Medical Ctr. 198.143.32.33 - 1/27/2016 3:52:34 AM

tors at T1 and T3 load on one latent variable and the unstandardized factor loadings of the indicators and the intercepts are invariant over time (strong factorial invariance) and if (2) a second latent variable with equal factor loadings is introduced for the indicators at T3, the variance of this second latent variable captures interindividual differences in latent change over time. By modeling change on the latent level, change is modeled uncontaminated by measurement error (fig. 1). Since we used single items as indicator variables for subjective memory, these variables were analyzed employing factor analysis for ordered-categorical variables. Note that if ordered-categorical items are factor-analyzed as if they were interval-scaled, there may be a mismatch between the information represented by the numbers assigned to the Likert-type scales and the nature of the factor model on which statistical tests are based [23]. Moreover, orderedcategorical variables frequently show departures from normality. Previous studies have shown that this typically results in negative bias of parameters [37], even more so in multiple groups or longitudinal cases [24]. Factor analysis models for ordered-categorical variables date back to Bartholomew [38] and Muthén [39], among others. Millsap and Yun-Tein [25] have extended these models for the multiple-group case. Their approach will be used in the present study, with the necessary adaptations for longitudinal data. All analyses were conducted using Mplus, version 6 [40], employing a robust weighted least squares (WLSM) estimator adjusted for means. The goodness of fit of models was evaluated using the Satorra-Bentler rescaled χ2 test. In addition, we report the comparative fit index (CFI) and the root mean square error of approximation (RMSEA). CFI values >0.90 are considered to be adequate, whereas for the RMSEA values

Subjective and Objective Memory Changes in Old Age across Five Years.

Typically, subjective memory assessments (be it in form of single items or questionnaires) in old age only weakly correlate with the performance in ob...
395KB Sizes 1 Downloads 7 Views