Cognitive reserve estimated with a life experience questionnaire outperforms education in predicting performance on MoCA: Italian normative data

Normative data of neuropsychological tests typically consider the effect of demographic variables like age and education on performance. However, a broad literature has shown that, after the school age, other cognitively stimulating experiences (e.g., occupational attainment and a variety of leisure-time activities) may increase and build up cognitive reserve (CR), which is positively associated with better performance in many neuropsychological tests. With these premises, we investigated the predictive capability of education and a life-experience proxy of CR on a widely used cognitive screening, i.e., the Montreal Cognitive Assessment (MoCA). Results show that models including the more comprehensive life-experience CR proxy performed better than models including Education. Based on the results of our analyses we provide normative data and cut-offs on 440 Italian individuals aged 50-90 years, by taking into account, for the first time for the Italian population, a CR index, together with demographic variables and Education, in the calculation of regression-based norms. Accounting for life-experience CR proxies can improve the accuracy of normative data and allow a finer estimation of cognitive performance, which lead to a more tailored approach to patient assessment.


Introduction
In clinical neuropsychology, observed scores at neuropsychological tests alone cannot be easily interpreted to draw meaningful conclusions. For example, a performance of 25 out of 30 at the MoCA test (Nasreddine et al., 2005) does not tell anything specific per se, unless there is a way to interpret what this observed score may imply. There are many ways to obtain some reference scores, often called clinical cut-offs, that may help the interpretation of an observed performance (e.g., Crawford et al., 1998Crawford et al., , 2010. If the aim of the test is a classification of some specific clinical conditions (e.g. having a Mild Cognitive Impairment vs having Alzheimer's Disease), particular methods, that are based on classification (as the Receiver Operating Characteristic, ROC), allow to identify the optimal scores that help in classifying the belonging to one condition or to another (see Nasreddine et al., 2005). On the other side, if a cognitive test aims to identify the potential presence of specific or general cognitive impairment, cut-off scores are typically obtained from a reference sample of healthy people often referred to as "normative data" (Borland et al., 2017;Cesar et al., 2019;Kopecek et al., 2017). Comparing a patient's performance with normative data makes possible to understand the potential meaning of an observed score: if that score is unlikely to be observed in healthy people, it could be a sign of a cognitive worsening, or an impairment.
To allow a meaningful comparison, normative data should be as much similar as possible to the patient, in order to accurately predict the expected performance and identify deviations that may be symptoms of cognitive impairment or decline. For this reason, normative data and cut-offs almost always take into account demographic variables, which are easy to collect in the context of a clinical setting, and are known to be associated with differences in cognitive performance (Strauss et al., 2006). In particular, the demographic variables that are typically considered in the development of normative data are age, sex, and education.
Regarding age, it is widely known that physiological agerelated brain changes can influence cognitive performance. Early studies about age-related effects on processing speed (Salthouse, 1993(Salthouse, , 1996 have set the ground for further research demonstrating that aging does indeed affect a wide range of behavioral, cognitive, and neurological functions (Hohman et al., 2017;Jockwitz et al., 2019;Madden et al., 2020). In general, as age increases performance worsens (Proust-Lima et al., 2007;Verhaeghen & Salthouse, 1997). However, some neuropsychological test may show a different effect of age, that is a better performance as age increases (like in the case of vocabulary, whose size may increase along with the words encountered during life; Verhaeghen, 2003). Importantly, the effect of age could be linear, that is with cognitive performance that increases at a fixed rate with age, but also non-linear, that is with a linear relationship reaching a plateau and stabilizing, or having different slopes over age (Arcara et al., 2019;Arcara & Bambini, 2016).
A further variable that, based on several studies, is included in normative data is sex (Halpern, 1996;Mann et al., 1990;Reilly et al., 2016). Although it is still under debate whether cognitive differences between sexes are related to brain differences or socio-cultural effects, as a matter of fact, males or females can perform differently on specific tests (Yeudall et al., 1986;Zhang et al., 2017). To improve normative data, sex is often taken into account. For example, it has been found that scores on the MoCA (but not on the Mini-Mental State Examination score) are significantly influenced by sex: with women showing a higher performance on delayed recall and men showing a higher score on visuo-construction and serial subtraction (Engedal et al., 2021).
Another variable almost always included in normative data is education, often expressed in terms of years spent at school (Kittner et al., 1986). Education is a variable that can be easily collected and many empirical studies showed that the higher the education is, the higher the expected performance in cognitive tests (e.g., Chan et al., 2018;Le Carret et al., 2003Stern et al., 1992). Also in the case of education, the effect on performance could be linear or nonlinear as has been shown in previous studies (e.g., Arcara & Bambini, 2016;Arcara et al., 2019).
It is important to underline that the positive relationship often expected between education and performance does not necessarily imply that education has a direct role on cognitive efficiency: other variables (e.g., socio-economic status) may be confounded with the possibility to attain higher education and thus to maintain higher cognitive performance along with the lifespan (e.g., socio-economic status; see Jefferson et al., 2011). The observed relationship between education and cognitive performance, however, has gained greater importance in recent times. Many findings indeed have shown that early-life cognitive stimulation/enrichment at school age (or deprivation, considering it on a continuum) is associated with an efficient cognitive functioning in later life (Stern, 2002) and with the degree to which brain resources are accumulated and deployed (Katzman et al., 1988(Katzman et al., , 1989Zhang et al., 1990), suggesting a potentially causal role in cognition. From this perspective, education is strictly related to the concept of the Cognitive Reserve (CR, Stern, 2002).
CR allows accounting for the discrepancy between agerelated brain changes and brain pathology and the relative manifestation in the cognitive domain. In other terms, CR "should help to account for individual differences in performance, given the same status of the brain" (Stern et al., 2018b). In fact, the same magnitude in brain disruption can result in different levels of cognitive performance, and such levels may increase with higher CR proxies (Barulli & Stern, 2013;Cabeza et al., 2018;Stern et al., 2018a). The most common proxy used to estimate CR is education (Borland et al., 2017;Busch & Chapin, 2008;Cesar et al., 2019;Malek-Ahmadi et al., 2015;Siciliano et al., 2019). Albeit the importance of education as a proxy of CR, its use has been considered conceptually problematic, as CR mechanism is unlikely to be represented only by the pathway between education and late-life cognition (Anatürk et al., 2021). Many studies in this field have underlined also the role of a 1 3 comprehensive range of further stimulating life experiences that occur along with adulthood (e.g., socio-behavioral factors; Fratiglioni et al., 2004;Livingston et al., 2017;Reed et al., 2011;Stern et al., 2018a;Ward et al., 2015). The main activities considered in this context are occupational attainment and leisure-time activities (Snowdon, 1997;Snowdon et al., 1996;Stern et al., 1995a, b). The potential contribution of occupational attainment and leisure-time activities to CR can be estimated by weighting the type of activity, the effort required to perform it, as also the regularity with which it is carried out (e.g., Scarmeas et al., 2001;White et al., 1994). For instance, this applies to the different degrees in responsibility and demand of a specific job, as also to physical activities (e.g., sports) requiring regularity, resistance, and concentration, or to hobbies entailing learning new skills, like for example music or art courses.
The importance of considering education together with other socio-demographic variables, proven to act as protective factors in longitudinal studies (Mondini et al., 2022;Ward et al., 2015), strongly suggests that normative data may improve when considering a composite proxy of CR in the calculation cut-off scores.
The Cognitive Reserve Index questionnaire (CRIq, Nucci et al., 2012) allows to estimate CR from education, occupational activity, and leisure-time activity as they concur in building up the reserve (Chan et al., 2018;Livingston et al., 2017;Scarmeas & Stern, 2003;Stern et al., 1995a, b;Ward et al., 2015). The CRIq is translated into many languages and it has been used both for clinical and research purposes (Artemiadis et al., 2020;Lavrencic et al., 2018;Lee et al., 2019;Maiovis et al., 2018;Montemurro et al., 2021a;Ozakbas et al., 2021). It has been also shown that, when using CRIq, the effect of CR significantly interacts with age in the expected cognitive performance (Montemurro et al., 2021b); in particular, as age increases performance is typically worse, but not when CRIq score is high. However, it is not clear how the interaction between age and CR may change between another relevant demographic variable, that is sex, which has been shown to have potentially complex interplay with CR (Levine et al., 2021;Myakotnykh et al., 2020;Subramaniapillai et al., 2021). There are different methods for estimating CR (e.g., Kartschmit et al., 2019). Education has the merit of being a practical and easy-to-record measure, but it only captures the formal educational training and at some point it is not effective (Berggren et al., 2018). IQ is also able to estimate CR but it should be used with caution, since not all the IQ measures have the same meaning as proxies of CR (see Nucci et al., 2012;Montemurro et al., 2021b for some considerations about this topic). The CRIq we chose has the advantage of being a validated and standardized measure that does not overlap with IQ (Nucci et al., 2012) and that it can adequately predict the MoCA score (Kang et al., 2018).
For the present study, the goal is two-fold. First, it aims to investigate whether a life-experience proxy of CR could be a better predictor of performance as compared to education in healthy controls for a popular screening test like the Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005). MoCA is, to date, used worldwide by psychologists, neurologists, geriatricians, occupational and speech therapists (e.g., O'Driscoll & Shaikh, 2017) and serves for a general screening, usually followed by a more detailed neuropsychological assessment of the cognitive profile. Initially created to distinguish among individuals with MCI, Alzheimer's disease, and healthy controls, now widely used for assessing the cognitive profile of healthy participants Montemurro et al., 2019a), populations with pathological aging (Roalf et al., 2013) and neurodegenerative disorders like Parkinson's Disease (Biundo et al., 2014;Montemurro et al., 2019b), Multiple Sclerosis (e.g., Al-Sharman et al., 2019), and Human immunodeficiency virus (HIV) (Rosca et al., 2019).
The second aim of the paper is to provide improved normative data and clinical cut-offs of MoCA for the italian population. Although other normative data are already available (Aiello et al., 2021;Conti et al., 2015;Pirrotta et al., 2015;Santangelo et al., 2015), this study aims to provide for the first time, normative data and cut-off scores accounting for a life-experience CR proxy.

Participants
The participant sample included a total of 440 Italian healthy individuals (292 females, 148 males, Age range = 50-90 years old), all Italian native speakers and autonomous in their daily living activities. The individuals with a medical history of stroke, traumatic brain injury, or any neurological or psychiatric disease requiring medical treatment were not included in the normative data sample. All participants took part in this study after signing the Informed Consent. The Research Ethics Committee of the University of Milano-Bicocca approved this study (permit number: RM-2017-84) as a minimal risk study that adheres to the Declaration of Helsinki. We report the descriptive results associated with the demographic variables (Age, Education, and CRI) and the MoCA raw scores in Table 1.
A series of t-tests showed that males had higher Education and higher CRI compared to females [Education: t(438) = 4.28, p < 0.001; CRI: t(438) = 2.68, p < 0.01), while MoCA did not differ between the two groups [t(438) = 0.31, p = 0.75)]. Figure 1 shows the distribution of these variables in the two Sex groups; Fig. S1 in Supplementary Materials shows more details about the distribution.

Materials
The MoCA test (Nasreddine et al., 2005) and the Cognitive Reserve Index questionnaire (CRIq, Nucci et al., 2012) were administered to all participants.
-The MoCA is a cognitive screening. It assesses visuospatial/executive abilities, naming, memory, attention,  language, abstract reasoning, and orientation (in time and space). Visuospatial and, to some extent, executive abilities are assessed using an adapted form of the clock-drawing task and a trail-making task (Reitan, 1958). Attention, concentration, and working memory are assessed through an auditory sustained attention task, a serial subtraction task, and forward and backward digit recall. MoCA has been proven to be more sensitive in detecting the level of general cognitive functioning, compared to the Mini-Mental State Examination (Ciesielska et al., 2016;Dong et al., 2010;Roalf et al., 2013). MoCA is available in over 30 languages and it can be administered in 10 min. CRIq is a semi-structured interview including 20 questions tapping into the formal educational experience (CRI-Education), working activity (CRI-WorkingActivity), and leisure-time activity (CRI-LeisureTime). CRI-Education refers to the years of formal education achieved plus any course carried out during adulthood (from 18 years old) for at least 6 months. CRI-WorkingActivity refers to years of working occupation carried out throughout the lifespan. CRI-LeisureTime refers to the amount of social, intellectual, and physical activities carried out in adulthood and its frequency in time (e.g., reading books, paintings, volunteering, doing sport, etc.). Regarding CRI-Education, one point is assigned for every attended year of schooling and 0.5 points are assigned to every 6 months of vocational courses attended in adulthood. Regarding CRI-WorkingActivity, one point is assigned for each year spent working. Jobs were weighted differently in the calculation of the total CRI score; they were divided into 5 groups depending on cognitive load, as also responsibility and mental resources required. CRI-LeisureTime accounts for activities with Table 2 Activities considered in the Cognitive Reserve Index questionnaire (CRIq). The first column reports the three components characterizing the CRIq: Education, Working Activities, and Leisure Activities; in the second column the activities considered for the computation of the score are listed; in the third column, the information regarding the regularity/frequency of occurrence of the activities accounted in the CRIq 1 3 usual weekly (e.g., reading newspapers, doing sport, using new technologies), monthly (e.g., voluntary work, playing music), and annual frequency (e.g., traveling, attending concerts, and/or conferences). We used the CRI total score for obtaining CR-based Italian normative data of MoCA. CRI is the average of CRI-Education, CRI-WorkingActivity, and CRI-LeisureTime (Nucci et al., 2012) and it provides a comprehensive/composite proxy of CR. Table 2 shows a description of activities considered in the calculation of CRI. Administration lasted on average 15 min. We administered MoCA and CRIq to each participant.

Statistical Analyses
All analyses were performed with the free statistical software R (R Core Team, 2021). The analysis strategy of this study was not pre-registered. The relationship between MoCA and Age, CRI, and Education (in terms of years) was first examined using a series of pairwise Pearson's correlations.
To investigate how CRI and Education predicted performance on MoCA, and to build the regression-based clinical cut-offs, we performed a series of regression analyses using demographic and socio-behavioural variables (i.e., Age, Sex, CRI, and Education) as predictors, and the score on MoCA as dependent variable.
Before running the regression analyses, we checked the potential occurrence of harmful collinearity between continuous predictors, which could harm model fit and interpretation. A preliminary check on potential collinearity was made by the inspection of the correlations, separately for each model. CRI showed a large correlation with Education  Table S1 in Supplementary Material shows VIFs. We then analysed a series of linear regression models including MoCA score as dependent variable: Model 1 included Age and Sex as predictors, Model 2 included Age, Sex, and CRI, Model 3 included Age, Sex, and Education. Model 1 was included to have a simpler reference model and to test if including either Education or CRI led to an actual improvement in prediction. Then, a fourth model (Model 4) included both Education and CRI as predictors of MoCA scores. We checked the model diagnostics to inspect whether the assumptions for these linear regression models were met and we looked for outliers that could have influenced the regression estimates (if this was the case, we would have re-fitted the model excluding the outliers): the diagnostics did not show harmful results and none of the data points were identified as outliers based on Cook's Distance. The best-fitting model was also evaluated through the inspection of the partial residuals to check whether any non-linearity (i.e., curvilinear relationship) emerged between one predictor and the MoCA. For variables that showed a non-linear trend, we would have tested whether adding quadratic terms yielded better models. Following the procedure used in Arcara and Bambini (2016) and Arcara et al. (2017), the quadratic term of CRI and Education were entered in Model 5 and Model 6, respectively. Finally, the quadratic and the cubic terms of both CRI and Education were added in Model 7. Quadratic values were mean-centered (Iacobucci et al., 2016) for a better interpretation of the results. Models with also cubic terms were built to check whether entering them or not would change the results (see Supplementary Material).
AICs were compared: the model with lower AICc was the one that best fit the data. The lower AICc value was shown for Model 7. Beyond the AICc values, probability information loss was further examined using AIC Weights and Delta AIC, (Burnham & Anderson, 2002;Symonds & Moussalli, 2011). AIC Weights are expressed in terms of the probability of a model (in this case, Model 7) to be the best among the set of given models; AIC Δ on the other hand shows how many times Model 7 is compared to all the Models. Model 7 showed the minimum loss of information (Burnham & Anderson, 2002) and it was used for fitting the data of our normative sample for generating cut-offs using the regression method of Crawford and Garthwaite (2006). Such method involves the calculation of the discrepancy between the observed and the predicted score from a regression model. The cut-offs are calculated as the scores delimiting a discrepancy that is expected to be obtained by less than 5% of the population, after predicting the score from the relevant variables. Shortly, a score falling below the cut-off indicates that the observed performance of an individual is unexpectedly low given the variables included in the model as predictors. The code used for cut-off calculation can be downloaded from https:// github. com/ giorg ioarc ara/R-code-Misc/ tree/ master/ Crawf ord_ and_ Garth waite_ 2006_R_ code.
Based on a previous investigation (Montemurro et al., 2021b), in which it has been shown a relevant and selective interaction between Age and a composite life experience CR proxy (but not for Education) in individuals without agerelated disease, an additional analysis was made to further explore such interaction. In line with the variables included in the study, we included in the additional analysis the variable Sex. More specifically, Generalized Additive Model analyses (GAMs; Hastie & Tibshirani, 1986) were used to investigate the potential interaction between Age and CRI. The primary advantage of GAMs over linear regressions is their ability to efficiently treat, through smooth functions, complex non-linear relations between predictors and the dependent variable. GAMs allowed investigating the interaction between continuous variables through tensor surfaces, (something that cannot be handled by GLM); for a recent application of GAM with neuropsychological tests see Yu et al. (2020); Montemurro et al. (2021b). The reported p values of the GAM analyses are corrected through the Bonferroni method according to the total number of tests performed. We used the package itsadug (van Rij et al., 2016) to inspect areas of significant interactions, and for data visualization, we overlaid a grey area when Confidence Interval (95%) included 0. The package mgcv version 1.8-24 (Wood, 2017) was used to implement GAM models. Diagnostics were performed for both GLM and GAM models; all statistical analyses were performed in R version 4.0.1 (2020-06-06).

Results
MoCA score showed a moderate correlation with Age [r(438) = −.41, p < .001], with CRI [r(438) = .42, p < .001], and with Education [r(438) = .42, p < .001]. The variable Age linearly and significantly predicted MoCA scores in all the models: the higher the Age, the lower the MoCA scores. Inspection of the partial residuals in Models 1-4 indicated that Education and CRI were non linearly related with MoCA scores. Model 5, 6, and 7 were built to check whether including non-linear terms would improve the model fit. According to the AICc comparison and to Adjusted R 2 , the best model fit was Model 7, which included the demographic variables (Age and Sex), the two CR proxies (Education and CRI), and their quadratic terms (Education 2 and CRI 2 ); see Table 3. Model 7 resulted as the model with the highest likelihood and lower probability information loss compared to all the other models (Burnham & Anderson, 2002). Figure 2 shows the effect of Age CRI and Education on the MoCA score, based on the best model fit (Model 7). Model 7 reached a satisfying power (0.80) considering the six predictors entered, the alpha-probability set to 0.05 and f-squared of 0.01.

Cut-off Scores
We calculated clinical cut-offs by applying the method of Crawford and Garthwaite (2006) to the results of Model 7. To calculate clinical cut-offs (i.e., reference scores that may help the interpretation of an observed performance in clinical contexts), values of MoCA associated with p = 0.05 (according to the Crawford method) were rounded to the nearest integer. Cut-offs of MoCA scores based on Age, Sex, CRI, and Education are reported in Table S2 of the Supplementary Material. This article is also accompanied by two online applications which allow to calculate (for a given observed score at MoCA) the exact p value associated with the performance. The observed p values and above/below cut-off performance can be automatically and easily calculated also using the Online Application available at: https:// osf. io/ y4jzs/

Exploratory Analyses: Interaction between Age and CRI by Sex, through Generalized Additive Models (GAMs)
In the first GAM model, Age and CRI were entered as interacting continuous variables (Mod_int_1), with MoCA as the dependent variable. The syntax of this model is reported below: In a second step, we entered Sex in the model as a factor with two levels, (i.e., males and females, Mod_int_2). The syntax of this model is reported below: Entering the factor Sex slightly improved the model fit (AICc of Mod_int_1 = 2007.57; AICc of Mod_ int_2 = 2006.55). Namely, the best fitting model assessing the interaction between Age and CRI was the one in which Sex was included. The results of this model showed no significant main effect of Sex (B = −0.39, t = −1.63, p = 0.31), and a significant interaction between Age and CRI in both sexes (females: edf = 5.32, F = 27.23, p < 0.001; males: edf = 3.01, F = 8.10, p < 0.001), see Fig. 3. GAM analyses are characterized by parametric (or linear) terms, and smooth terms. An additional simpler model assessing the interaction between CR and Sex without accounting the effect of Age is reported in Supplementary Materials. Details about the results of Mod_int_1 and Mod_int_2 are reported in Table 4. Figure 3 shows the effect of CRI on MoCA for three different values of Age (i.e., 50, 70, and 90 years): higher MoCA scores were observed at the lower values of Age and Mod int 1 <−gam MoCA ∼ te Age, CRI, k = c(5, 5) , data = dataset Mod int 2 <−gam MoCA ∼ Sex + te Age, CRI, k = c(5, 5), by = Sex , data = dataset the higher values of CRI. The interaction effect showed that the performance of participants with higher Age (that is when cognitive resources are decreasing with physiological decline) was more strongly and positively affected by CRI. Diagnostics of the GAM models showed non-harmful concurvity of the smooth terms, and the use of parameter k was adequate (based on gam.check R function).
For a better interpretation of the GAM interaction between Age and CRI in the two sexes (Mod_int_2) we provide in the Supplementary materials a contour plot (Fig. S2), which is used for interpreting GAM interactions among continuous variables.

Discussion
This study aimed at investigating whether CRI as a life-experience CR proxy predicted cognitive performance better than considering Education as CR proxy, for better normative data of the Montreal Cognitive Assessment (MoCA, Nasreddine et al., 2005). Both CRI and Education were positively associated with MoCA scores: the higher Education and the higher CRI, the better the performance, in line with previous findings in which a positive relationship between CR (estimated through Education and life-experience CR measures) and MoCA is highlighted (Kang et al., 2018;Kujawski et al., 2018;Peeters et al., 2020).
However, our data show that considering a composite score of CR, like CRI, results in a better prediction of the MoCA scores in healthy individuals, compared to considering only Education. This was attested by different metrics Table 4 Results of Generalized Additive Models with Age and Cognitive Reserve Index in interaction and MoCA as dependent variable. The table shows, for the two GAM performed (model name in the first column), the result associated with the parametric coefficients (leftmost part of the table) and the smooth terms (rightmost part of the table). Within the parametric coefficients, the table shows: names of parametric coefficients (terms, second column); coefficient estimate and standard error within round brackets (third column); t-value associated with the parametric coefficient (fourth column); p value associated with the parametric coefficient (fifth column), with an asterisk "*" denoting significant terms. Within the smooth terms, the table shows: names of smooth terms (sixth column); estimated degrees of freedom (seventh column); F-value associated with the smooth terms (eighth column); p values associated with the smooth terms, with an asterisk "*" denoting significant terms evaluating the quality of fit of our models (i.e., AICc and R 2 ). The predictive ability of CRI is in line with previous findings revealing that life-experiences proxies have a high influence on cognitive function (in line for example with Reed et al., 2011, in which some activities considered were reading or attending concerts). Moreover, also at a qualitative inspection, CRI showed a better distribution compared to the variable Education (Fig. S1 in Supplementary Material), suggesting CRI as a suitable variable for making predictions as it could potentially cover a more comprehensive range of influential life-experiences and differentiate the individuals of the normative sample.
The best-fitting model included Age and Sex together with CRI and Education. The variable Age exerted, as expected, a significant effect on MoCA scores, which were lower as Age increased, in line with previous literature (e.g., Malek-Ahmadi et al., 2018;Wöbbeking-Sánchez et al., 2020). The variable Sex showed a significant effect on cognitive performance, according to previous findings (e.g., Reilly et al., 2016) and it was an important variable we accounted for both cut-offs and exploratory GAM analyses.
The results of this study strengthen the importance of taking into account, in addition to the demographic characteristics (Age and Sex) and the educational level of the population, the inter-individual differences in carrying out a variety of potentially beneficial activities in adulthood: a point that has been considered crucial for prevention and healthcare (Livingston et al., 2017) and important for estimating more accurate normative data for clinical uses. The results presented here are focused on a specific test (i.e., the MoCA) and some specific proxies for CR (CRIq and Education), but they are already very encouraging in highlighting a potential importance of considering CR in building normative data. However, generalizability of the results is not warranted, and further research is needed to define whether this same result is shown with different cognitive tests, different life-experience CR proxies, as also different cultures.
Our best-fitting model with Age, Sex, CRI and Educations as predictors of MoCA was used for deriving cut-off scores based on the regression-based method from Crawford and Garthwaite (2006). All cut-off scores are reported in the Supplementary Material (Table S2).
We also explored the potential age-related effect on performance, together with the modulatory role of CR estimated by CRI, to better investigate the relationship between Age, CR, and performance, which has been recently shown to be potentially very complex (see Montemurro et al., 2021a, b). Such interaction was modelled through Generalized Additive Models (GAMs, Hastie & Tibshirani, 1986) which are suitable for evaluating complex relationships between continuous variables. This analysis showed that the patterns of MoCA smoothly changed along with the variations in Age and CRI. The higher the Age, the lower the performance at the MoCA test, but notably, this effect was modulated by CRI. On the other hand, the higher the CRI the higher the performance at the MoCA test, but this effect was significantly attenuated by Age.
Looking at demographic (like Age and Sex) and sociobehavioral variables (like CRI and Education), as potentially interplaying in determining the capacity of maintenance of the cognitive system (Anatürk et al., 2021;Cabeza et al., 2018;Stern et al., 2019), is relevant and has been considered vastly important in a prevention perspective of the health care (Livingston et al., 2017).
It is important to underline that different proxies of CR may be interconnected with each other (Steffener & Stern, 2012). We observed a correlation between CRIq total score with Education (not surprising as Education contributes to some extent to the score), and both might be thought as potentially related with variables like IQ or socio-economic status (e.g., Jefferson et al., 2011), which were not used in this study. Based on previous research, which did not show a relevant associations between CRIq scores and IQ measures, we might speculate that CRIq and IQ would not perform similarly if included in a model (Nucci et al., 2012). However, further investigation is needed to assess the different effect of proxies on cognition.
Interestingly, potential differences between sex could be expected in the compensatory effect of CR, in line with previous studies (e.g., Subramaniapillai et al., 2021). For this reason, in our additional analyses, we also explored the interaction between Age and CRI by entering Sex, in the GAM model of interest. Albeit no statistical differences involving Sex have been found, the group of females, especially at the oldest Age seemed to benefit more from higher CR, compared to the group of males. This result is compatible with previous studies showing how some relevant contributors to CR, like Education, occupation, as also physical activity) are highly gendered. Historically, as compared to males, females have had less access to CR determinants (see Cohen, 2007 for more details related to the Italian background) which might have contributed, for a certain period, to the higher rates of cognitive decline in women, compared to males (Farrer et al., 1997;Mielke et al., 2014). However, future studies are necessary to shed further light on this point.

Limitations and Further Implications
The calculation of normative data including life-experience proxies of CR has some limitations: while Age and Education can be easily collected in a few seconds, the administration of tools estimating life-experience CR proxies may require additional time, which should face possible constraints of clinical assessments.
It is also important to consider that the CRIq is not the only tool that can be used to obtain a life-experience proxy of CR (see for example Altieri et al., 2018;Leoń et al., 2014;Valenzuela & Sachdev, 2007); other questionnaires or scales may even outperform the predictive performance of CRIq (Kartschmit et al., 2019) and in turn, the accuracy of calculated cut-offs. We used the CRIq since it is a validated and versatile instrument that has been widely used for estimating CR in some previous researches (Artemiadis et al., 2020;Lavrencic et al., 2018;Lee et al., 2019;Maiovis et al., 2018;Ozakbas et al., 2021). Here, for the first time, we show how it can be used to build more accurate normative data for clinical uses.
In the analysis we included only main effects (that is without interactions) to calculate cut-offs. Future studies may consider the possibility of obtaining Italian normative data of MoCA based on the complex interaction between demographic and socio-behavioral variables, as those obtained with GAM, which may need the development of appropriate statistical models, for drawing useful clinical conclusions.
Anxiety, depression, or stressors have been previously shown to negatively affect MoCA performance (Blair et al., 2016). Although we did not assessed directly the presence of such conditions, none of participants reported psychological or psychiatric disorders, and the influence of such effects on MoCA performance is not always replicated in the literature (e.g., Anapa et al., 2021). Thus, psychopathological aspects might have played a negligible effect on performance.
potential contribution of occupational Overall, the present study suggests that preserving and promoting CR (i.e., with hobbies and intellectual activities) may potentially enhance cognitive efficiency, especially in the oldest years. Importantly, the study highlights that including more detailed demographic and socio-behavioral inter-individual information can foster the development of a more tailored approach in obtaining normative data, which is a crucial aspect for clinical assessments and health care.

Data Availability
The dataset analysed during the current study is available in the Open Science Framework (OSF) repository https:// osf. io/ y4jzs/

Declarations
Competing Interests On behalf of all authors, the corresponding author states that there is no conflict of interest that are directly or indirectly related to the work submitted for publication.