Predicting normative data in healthy individuals on the computerized wisconsin card sorting test using regression models

1 Introduction

Neuropsychological tests have been used for a long time in the investigation of cognitive profiles occurring in healthy and neuropsychiatric disorders. Revealing these profiles gives information about the “normal-healthy” cognitive structure and helps us diagnose and follow the rehabilitation process for different mental and neurological disorders (Kessels & Hendriks, 2016; Nelson & Searl, 2002; Shan et al., 2008).

The Wisconsin Card Sorting Test (WCST) has been one of the most frequently used tests in evaluating frontal functions. To succeed in the WCST, an individual must employ many cognitive abilities concurrently, including attention, set shifting, perseveration, cognitive flexibility, inhibition of impulsive responses, problem-solving, and planning (Heaton et al., 1993). Hence, the WCST exhibits a considerably more intricate structure than other neuropsychological tests. When a deficit in cognitive abilities results from a mental or neurological disorder, there is a substantial decline in test performance compared to healthy controls (Gillan et al., 2020; Itzhaky et al., 2023; Petkus et al., 2020; Tercero et al., 2021; Venezia et al., 2018).

Psychometrically, different methods exist to compare individuals with their peers regarding specific cognitive abilities. Percentiles (Atalay & Cinan, 2007; Faustino et al., 2022) and Z and T transform scales created from Classical Test Theory are among the most common methods (Crawford, 2003; Guàrdia-Olmos et al., 2015). However, the most significant limitation of these methods is the need for a gold standard for classifying socio-demographic characteristics, especially education level and age, which are known to affect cognitive performance (Miranda et al., 2019; Rammal et al., 2019). WCST performance changes in different age groups due to age-related cognitive declines and individual differences such as education level and gender (Rhodes, 2004; Rammal et al., 2019). These variations highlight the need for age-appropriate norms and consider multiple factors when interpreting WCST results across different age groups. In this respect, the most common application is to categorize the sample arbitrarily according to age and education level and extract norm tables while creating norm values (Karakaş et al., 1999; Salthouse, 2001; Salthouse et al., 2003). However, this method can lead to wrong decisions about the individual’s cognitive performance; it does not allow us to know to what extent age and education level affect test performance. Another limitation is that age does not affect cognitive performance at the same level in every age range. In other words, while the decrease in performance in a test decreases with a particular slope, it can show a dramatic decrease after a specific age. Especially in neurodegenerative disorders, this situation is more evident as cognitive decline decreases after 65 years of age. To address these limitations, this study explores the application of regression models in cognitive assessment, drawing on recent advances in the field (Arango-Lasprilla et al., 2015; Guàrdia-Olmos et al., 2015). Using regression models also helps eliminate floor and ceiling effects often encountered in norm tables (Hartshorne & Germine, 2016).

2 Methods

Regression coefficients and R² values are given in Table 1. Age and education coefficients are significant at the 1% level for WCST 1, 2, 3, 4, 7, 10, and 11. There is a negative relationship between age and WCST 3, 4, 10, and 11, as well as between education level and WCST 1, 2, and 7. While establishing the regression models, gender was added to the model, but since its effect on WCST scores was found to be insignificant, it was removed from the model. This shows that gender has no effect on WCST scores, which means, it does not affect whether the scores increase or decrease. However, only WCST 2, 3, 4, 7, and 10, 11 models give more meaningful results when R² values are considered. Age and education variables were insufficient to explain the score of WCST 5, 6, 8, 9, 12 and 13. SDe represents the standard deviation of error terms.

Test statistics for OLS assumptions are given in Table 2. As is seen in Table 2, where VIF values are less than 10, there is no multicollinearity problem between the independent variables. However, there are non normality or heteroscedasticity problems for WCST 1, 5, 6, 7, 8, 9, 12, and 13 models. Although the OLS assumptions are not satisfied for WCST 3 and WCST 10, this issue has been overcome using WLS. In this study, Weights are determined as Weight = 1/Age for WCST 3 and Weight = 1/Fitted WCSTScore for WCST 10. Additionally, WCST 7 are not satisfied the OLS assumptions, and a suitable weight for WLS could not be identified. The OLS assumptions are satisfied by WCST 2, 4, and 11 among 13 different WCST models.

The primary objective of this investigation is to generate normative data for the WCST-CV for a sample of healthy Turkish adults using regression modeling. Our findings indicate that five distinct WCST-CV sub-scored conform to the prerequisites for regression analysis, with education level and age variables providing a satisfactory explanation for the variance observed. In contrast, when scrutinizing Tables 1 and 2 in tandem, it becomes evident that more than five WCST scores exhibit associations with education level and age. Nevertheless, norm scores were not derived for these particular score types due to their notably low explained variance rates. The existing literature reveals that in studies adopting a similar methodology, if the regression model attains statistical significance, estimated data are often presented without regard for the magnitude of explained variance (Arango-Lasprilla et al., 2015; Arango-Lasprilla et al., 2017). However, it is essential to emphasize that this practice represents an unsuitable approach within the broader context of regression analysis (Gujarati, 2022).

Norms for psychological tests should undergo periodic updates within the targeted sample. An illustrative example of this practice is found in the Wechsler Intelligence Tests for Children. The intelligence scores are periodically revised, considering biological factors (genetic factors) and development levels of countries change over time and, accordingly the growing accessibility to formal education commencing at increasingly early ages and technological advancements. This phenomenon is explained by the Flynn effect, which highlights a consistent upward trend in intelligence scores over time (Graves et al., 2021; Trahan et al., 2014; Bratsberg & Rogeberg 2018). The Turkish adaptation of the Wisconsin Card Sorting Test (WCST) was undertaken by Karakaş et al. in 1999. Nevertheless, this adaptation was limited to a relatively small sample of 69 healthy individuals, and it failed to address its applicability within clinical populations, particularly concerning ecological validity, which stands out as one of the pivotal methodological concerns. Furthermore, as alluded to in the introduction, the WCST ranks among the prominent neuropsychological tests internationally employed for the assessment of cognitive functioning in both healthy and clinical populations. However, its infrequent usage within our national context can be attributed to the scarcity of individuals possessing expertise in its administration and scoring, in contrast to more commonly employed tests. In light of this, our research team developed an extended version of the WCST in 2021 to facilitate its broader application in research and clinical contexts (Çelik et al., 2021). Nevertheless, psychometric investigations conducted among individuals afflicted by schizophrenia and those without such a diagnosis reveal disparities between the two versions. Consequently, this discrepancy necessitates formulating a distinct set of norms tailored explicitly to the WCST-CV, particularly when endeavors to assess an individual’s cognitive capacity within a clinical setting are at the forefront.

Using the computerized WCST version gives some advantages to clinicians, such as being more economical and minimizing errors in application and calculation. Due to these advantages, the production of the norms of the WCST-CV will enable the widespread use of this test in clinical and research. The increasing use of technology, especially in neuropsychological assessment tools, necessitates a review of norm studies. Having norm values of WCST-CV in a large sample will contribute to the widespread use of WCST-CV for both clinical and research purposes. Unlike other norm studies, we defined age as a continuous variable. Thus, we think that clinicians will make more precise decisions. Especially in the regression models, the gender variable was not significant for any score type. Therefore, gender was not included in the regression scores. Upon a comprehensive review of the literature, it becomes evident that a relatively limited number of studies demonstrate a substantial gender-related impact on WCST scores (Lineweaver et al., 1999). The consensus in the field is that gender is generally regarded as having no significant influence on WCST scores (Tercero et al., 2021; Norman et al., 2011; Caffara et al., 2004). Similarly, the previous norm study conducted in Türkiye found that gender did not affect WCST performance.

5 Conclusion

Compared to other studies, in this study, norm values for WCST-CV of healthy adults were calculated using a much larger sample population. Similarly, standardized tests such as the Beck Depression and Anxiety Scale and MOCA were used to determine whether the participants were mentally and neurologically healthy. Our aim is to establish norm values for the Turkish sample in order to popularize the use of the computer version of the WCST, which provides comprehensive information to clinicians and researchers about the frontal functions of the individual. Unlike other studies, instead of presenting norm lists/tables exceeding 200 pages for each score type, we developed a calculator. However, despite the unique aspects of our research, it also has some limitations. One of the most important limitations is that, as in other studies, education level was considered as a dummy variable. Dividing the education level into two groups: 12 years or less and above makes it difficult to investigate the real effect of education on WCST scores. For example, when Table 3 is examined, education level is 1–8; 9–12. Even when divided into more groups such as 13 or more years of education, a significant difference was found between education groups in terms of WCST scores. However, when the education level is wanted to be examined multivariately, the number of samples must be much higher than the current number, making the data collection process difficult and reducing the applicability of the research. In parallel with this, the number of healthy adults who have received 13 years or more of education in older ages is relatively lower than other groups. Therefore, as can be seen from Table 3, since the number of sample sizes is not sufficient for some specific education and age groups (at least 10 people for each age and education level were included in this study; if the number of people per group is less than 10 people, mean and standard deviation values are not given and are indicated as N.A.). A second limitation is that the norm values obtained are valid for the Turkish sample and cannot be generalized to other countries.