DISCUSSION

The present study presents regression formulas that were designed to predict follow-up performances on the RBANS, which were derived from a large sample of community dwelling older adults. Consistent with previous research (McSweeny et al., 1993; Temkin et al., 1999), initial performance on a measure (i.e., Time 1 score) was the best predictor of follow-up performance (i.e., Time 2) on that same measure, although other factors, including demographic variables, also added to the predictive value of the formulas. Within the current study, predictor variables accounted for approximately 39–72% (M = 53%) of the variance of the criterion variables (i.e., R² of multiple regression equations). Although these findings do not capture the entirety of the Time 2 scores, they are similar to those reported by others (Hermann et al., 1996; Sawrie et al., 1996) using different patient samples. Temkin et al. (1999) reported slightly higher relationships between predictor and criterion variables, but they only presented on a limited set of neuropsychological measures. It is possible that the inclusion of other variables (e.g., education defined continuously rather than categorically, broader range of racial/ethnic status, medical conditions of the patient, psychometric stability of the measures, occupational status of the patient, or squared and cubed transformations of the Time 1 measures; see Temkin et al.) may further increase the robustness of these formulas.

An individual's Index score at Time 1 was consistently the best predictor of their Time 2 performance. For example, when looking at the standardized beta weights from the regression equations, nearly 80% of Total Score's Time 2 performance could be identified with its Time 1 score. Demographic variables, however, were less equitably distributed among the regression algorithms, in that age was utilized by three equations, education by three, while gender and race did not significantly contribute to any RBANS Index scores. The inclusion of age into equations was unexpected due to the relatively restricted age range of the sample (i.e., 64–90 years old) and given that Index scores are already age-corrected. Education's impact was less surprising, as its relationship with intelligence and other neuropsychological measures has been repeatedly demonstrated (e.g., Heaton et al., 1991). Test–retest interval was only used in one of the regression formulas, which may also be due to the restricted range in that the duration between testings was approximately 1 year for most participants. Past research (Temkin et al., 1999), however, suggests that some measures are more affected by the length of time between evaluations.

As noted above, it is not surprising that demographic factors contributed to the prediction of follow-up scores; however, the process by which they contribute is less clear. Unlike a single-session evaluation, where demographic factors affect performance at one point in time, repeated assessments are considerably more complex. In a re-evaluation, age and education could affect Time 1 scores, Time 2 scores, and the change in scores from Time 1 to Time 2. That is, there is likely some synergistic effect of age, education, and other factors on the change in performance across time (McCaffrey et al., 2000). For example, it is possible that change occurs differently in older individuals with lower education attainment. Indeed, intelligence has been shown to interact with repeated assessments, practice effects, and the determination of change across time (Rapport et al., 1997a, 1997b; Rentz et al., 2004).

In addition to developing the prediction formulas, the present study also empirically assessed the utility of these formulas by validating them in a separate sample of healthy older adults. The mean differences between observed and predicted Time 2 scores were minimal in the validation sample, and ranged from −0.80 (Delayed Memory) to 1.65 (Immediate Memory). Overall, the regression equations were accurate in predicting the observed score with the mean difference across Indexes being 0.48 Index points. Two of the predicted Indexes were significantly different from their observed counterparts, but the magnitude of these differences was small. Correlations between observed and predicted scores were also statistically significant for all measures, and indicated a strong relationship between these scores.

To further evaluate the clinical utility of the prediction equations, the percentage of cases whose predicted Time 2 score fell within ±5 or ±10 points of their observed Time 2 score was assessed in the validation sample. As expected, the Total Score had the greatest combined accuracy, with 86% of the cases falling within ±5 points, and 99% falling within ±10 points. Although the individual Indexes tended to have slightly lower hit rates, clinicians should feel confident in predicting Time 2 scores using the proposed algorithms.

Practitioners who have not used these techniques previously may also benefit from a brief case example illustrating their use. The patient is a 75-year-old retired White male with 10 years of education who was evaluated for complaints of memory problems approximately 1 year ago. His RBANS scores from that time are presented in Table 5. Due to his borderline memory performances, a diagnosis of mild cognitive impairment was made and a follow-up evaluation in 1 year was recommended. Observed Time 2 RBANS scores are also presented in Table 5, along with the difference between Time 1 and Time 2 observed scores. Predicted Time 2 scores (based on the equations in Table 2) and the difference between observed and predicted Time 2 scores are also presented. An initial review of the differences between observed Time 1 and 2 scores suggests that the patient has declined across all cognitive domains in the course of the year, from 1/3 to 4/5 of an Index standard deviation (SD = 15). Such a progressive decline, especially if accompanied by deterioration of activities of daily living, might lead a clinician towards the diagnosis of dementia.

Case example

If the clinician were to utilize the prediction equations from the current study, however, s/he might come to a different conclusion. Specifically, when Time 2 predicted differences (i.e., predicted Time 2 score − observed Time 2 score) are compared to observed differences (i.e., observed Time 1 score − observed Time 2 score), it is revealed that observed decline is greater than predicted decline for some of the Indexes (e.g., Immediate Memory: observed = −5, predicted = −.29). For other Indexes (e.g., Attention and Total Score), predicted and observed differences are similar (e.g., Attention: observed = −6, predicted = −6.24). Finally, the predicted difference (−20.27) is much greater than the observed difference (−12) for the Delayed Memory Index. When the predicted/observed differences are converted to standard metric by dividing them by their respective standard error of the estimates (Hermann et al., 1996; McSweeny et al., 1993; Sawrie et al., 1996), it is revealed that only the Delayed Memory Index demonstrates a statistically significant decline (i.e., z > |1.64|).

Hence, when the statistical data are considered, it appears that the patient in question only shows a significant decline in Delayed Memory, which might prompt the clinician to retain the diagnosis of mild cognitive impairment, rather than infer the expression of a full dementia. It should be noted, however, that we are not advocating for a strictly psychometric assessment of any patient (i.e., based solely on test data, without the complementing historical information or behavioral observations), rather we are attempting to provide clinicians with the necessary psychometric information to assist in the clinical decision-making process. A copy of the computer program used to calculate the predicted scores, difference scores, and test of the differences' significance can be obtained from the first author.

Several limitations of the present study should be noted. First, as with most regression-based prediction formulas (Tabachnick & Fidell, 1996), the current RBANS equations are likely to provide the best estimates of Time 2 performances for individuals who do not fall at the extremes of functioning (e.g., <2nd percentile or >98th percentile) at Time 1. In these cases, the prediction equations are more susceptible to regression to the mean effects, and will likely overestimate or underestimate predicted Time 2 scores. Similarly, the use of these formulas outside the demographic and situational parameters of the sample (e.g., <64 or >90 years old; relatively brief or extended retest intervals) may increase error. Readers are reminded that all Figure Copy and Figure Recall protocols were scored using slightly modified scoring criteria than those specified in the RBANS manual. Finally, participants in the current study do not represent “optimal” aging, as defined by the American Psychological Associate Working Group on the Older Adult (1998). Rather, they may be more representative of “normal” aging patterns, which include individuals with numerous medical conditions. In fact, the absence of a significant practice effect (see Table 1) in this sample raises the possibility that this group of older adults might be slightly declining in their cognitive abilities.

In conclusion, the present prediction algorithms have the potential to provide more accurate assessments of possible change in the individual patient by considering the influence of initial performance, practice effects, test–retest interval, and other demographic factors. These equations provided estimates that closely approximated actual Time 2 Index scores. In addition to providing prediction equations on a new measure, the RBANS, the current study also extends the existing literature in this area to include formulas relevant for practitioners who evaluate older adults. Future research should continue to expand this limited knowledge base, as this segment of the population continues to grow. Additionally, validation of the effectiveness of these formulas in a clinical sample would further extend their clinical utility.