2.1 Study 1
From a cohort of 4,638 individuals, 3,081 (2,010 females) entered the study (mean age = 78.2 yrs, SD = 7.79, range = 45-99; mean education = 6.99 yrs, SD = 3.85, range = 0-25 yrs). Predictors of cognitive outcomes were age and sex as socio-demographic variables, comorbidity (number of concomitant relevant pathologies) as the clinical variable, and education and occupation as proxies of CR. Occupation was classified using the ISCO-08 code (International Standard Classification of Occupations, 200816).
At each assessment, participants underwent a complete neuropsychological evaluation, but only six tests routinely used in our clinical unit, and known to be sensitive in detecting cognitive decline in the elderly, were selected for the present analysis: 1) MMSE (Mini-Mental State Examination17, Italian version18) for a global measure of cognition; 2) Famous Face naming test to measure proper name retrieval19; 3) Memory test, to measure episodic memory; 4) Fluency test, to measure lexical access and executive function; 5) Trail Making Test-A (TMT), to assess selective and spatial attention and psychomotor speed; and 6) Clock test to measure visuo-spatial abilities, planning and praxis. The last three tests are all taken from the ENB-2 battery20. Three diagnostic groups were identified based on psychometric data and anamnestic and clinical information: 1) Subjective Cognitive Decline21 (SCD) with 507 participants (17%); 2) Mild Neurocognitive Disorder (MildND) with 584 participants (19%); and 3) Major Neurocognitive Disorder due to Alzheimer’s disease (MajorND) with 1980 participants (64%).
All patients gave their written informed consent for the use of their data for research purposes at the moment of their consultation. The study was conducted in accordance with the Declaration of Helsinki, and it was approved by the Ethical Committee of the School of Psychology of the University of Padua.
Generalized Linear Model (GLM) analyses were carried out to verify whether and how age, sex, education, occupation and comorbidities could predict participants’ cognitive performance in the six tests. Models with best fit (after entering one predictor at time, based on R-squared) were those with all five predictors and all these models were significant (see Model fit measure in Table 1). A consistent effect of age, education, and occupation was found on all tests except for Clock and TMT where occupation had no effect. Sex and comorbidity played a minor role and predicted only two tests each.
Some differences among the three diagnostic groups (SCD, MildND, MajorND) were found, but, importantly, age and education predicted performance in all tests for each of the three groups. Occupation did not play a role within each groups, but it differs significantly among them (F(2, 900)31.0, p<.001): SCD had the lowest score (i.e., more complex jobs), while MajorND had the highest score (i.e., less complex jobs) (see Figure 1).
2.2 Study 2
Five hundred and forty-three individuals from Study 1 underwent a second assessment (T2), and 125 underwent a third assessment (T3). In a series of GLM analyses, age, sex, education, occupation and comorbidity were introduced as predictors, and participants’ performance in the six cognitive tests as the dependent variable. At T2 all GLM were significant in predicting the outcomes in the cognitive tests. Education confirmed its effect on all six tests, and age had an effect only on two. No other effect was found. At T3, education was the only significant predictor in three tests.
Participants were then sorted into two groups based on clinical evidence and on their cognitive performance on the tests. At T2 and T3, individuals who performed worse than in the previous assessment, were put in a single group and named declining. This group also included MajorND participants who maintained the same clinical diagnosis but worsened their cognitive performance (no MajorND patient improved on cognitive tests). Participants who at T2 (or T3) did not worsen their profile were put in a single group and named resistant (see Supplementary Table 1 for more details about participants’ changing cognitive profiles). At each assessment, the declining group showed lower average scores at T2 than at T1 and at T3 than at T2, on 4 out of 6 tests. Instead, the average scores on all six tests of the resistant did not change at either assessment T2 or T3 (see Table 2).
The time-interval between assessments (from T1 to T2) was different among participants: Short (1 year; 242 participants), Medium (2-3 years; 207 participants) and Long (more than 3 years; 94 participants). Assuming that the longer the time, the higher the possible cognitive decline, the aim of this analysis was to verify if the effect of CR proxies changed depending on the time elapsed between assessments.
At T2, the resistant had higher education than the declining (t = 8.04, df = 1064, mean difference = 2.43, p < .001) and this difference was present at all time-intervals: Short (t = 5.67, df = 480, mean difference = 2.77, p < .001), Medium (t = 3.02, df = 398, mean difference = 1.34, p = .003), Long (t = 5.43, df = 182, mean difference = 3.77, p < .001).
Occupation showed the same trend as education: the resistant had higher occupations than the declining (t = -4.80, df = 924, mean difference = -979, p < .001) across the three time-intervals: Short (t = -2.69, df = 424, mean difference = -819, p =.007); Medium (t = -2.60, df= 342, mean difference = -858, p-value = .01); Long (t = -3.30, df = 154, mean difference = -1565, p = .001). Note that the resistant had the highest education and more complex occupations when the interval was the longest (see Figure 2).
Of the 125 who underwent a third assessment (T3) 104 participants were declining (83.9%) and only 21 (16.1%) were resistant. Once again, resistant were more educated (t = 3.86, df = 361, mean difference = 2.49, p <.001) and with better occupation than declining (t = 2.25, df = 319, mean difference = -941, p = 0.025), (see Figure 3). The time-intervals among assessments were not considered in this analysis.