We recruited healthy older adults in the German cities Cologne, Vechta, and Osnabrück. Individuals were recruited with flyers and posters distributed via health centers, advertisements in the local press, and senior representatives. Participants received no monetary compensation but did not have to pay for the interventions, which are usually fee-based. Individuals interested in the study were first screened for eligibility via phone and then invited for a neuropsychological assessment in which their eligibility was further evaluated. All participants gave written informed consent before the assessment following the Declaration of Helsinki. In case of cardiovascular disease, affirmation for the participation in a physical intervention was obtained from the participants’ general practitioner.

Inclusion criteria were age between 50 and 85 years, normal or corrected-to-normal vision and hearing, and German as the native language. Exclusion criteria were any past or present psychiatric or neurological disease, a condition that prohibited moderate physical activity, past or present intake of psychotropic drugs, and former participation in a cognitive group training. Further exclusion criteria were cognitive impairment as assessed with the cognitive screening DemTect (Kalbe et al., 2004; ≤12 points) and presence of clinically relevant depressive symptoms operationalized with the German version of the Beck Depression Inventory 2 (BDI2; Hautzinger et al., 2009; >19 points). Participants who attended <80% of the training sessions were subsequently excluded from the analyses.

N = 81 participants were recruited and allocated to three different training groups at baseline (CT: n = 23; CPT: n = 28; CPT+C: n = 30). The online Research Randomizer¹ was used to randomize the participants in blocks of three to the intervention types, separately for each study center. Participants were not blinded for group allocation and were told that the study aimed to compare different interventions. The interventions were carried out between October 2012 and June 2013. The single sessions of all nine training groups were led by a trainer (author JM), who was licensed both for the application of CPTs. Cognitive status was assessed at pre-test, post-test, and at FU 1 year after the intervention. For this study only data from pre-test and FU were analyzed.

The three interventions with a maximum of 10 participants per group had a frequency of two sessions per week and a duration of 7 weeks. Each of the 14 sessions lasted 90 min. Training amount was comparable between the three groups. All interventions were described in more detail in a previous report (Rahe et al., 2015a).

Primary cognitive outcomes were assessed with an extensive cognitive test battery in standardized test situations at baseline, post-test, and FU. Assessors had been trained in the test application and scoring and were blinded for training group allocation of the participants.

Primary outcomes of the study were performance changes in the domains of general cognitive status, memory, executive functions, attention, and visuo-construction. Only established test instruments with good test criteria were used. The general cognitive status was assessed with the already mentioned DemTect (Kalbe et al., 2004). The immediate and delayed recall of the wordlist subtests of the DemTect (Kalbe et al., 2004) and the delayed recall subtest of the Complex Figure Test (Strauss et al., 2006) were used to measure verbal memory and figural memory. Within the domain of executive functions, working memory was assessed with the subtest digit span backward of the German Wechsler Adult Intelligence Scale (Aster et al., 2009), and verbal fluency was assessed with the letter verbal fluency tasks S, P, M (total) and the alternating letter verbal fluency task G-R of the Regensburger Wortflüssigkeitstest (RWT, Aschenbrenner et al., 2000) and the semantic verbal fluency tasks supermarket or animal subtest of the DemTect (Kalbe et al., 2004). To assess planning, the Zoo Map subtest from the Behavioral Assessment of the Dysexecutive Syndrome (BADS; Wilson, 1996) was used. Inhibition was tested using the Stroop test (Bäumler, 1985). For the assessment of attention, we used the Brief Test of Attention (BTA, Strauss et al., 2006). The domain visuo-construction was assessed using the copy subtest of the Complex Figure Test (Strauss et al., 2006). To minimize retest effects parallel forms of the DemTect (Kalbe et al., 2004) and the RWT (Aschenbrenner et al., 2000) were used, and the use of versions A/B at all assessments (pre-test, post-test, FU) were randomized.

Secondary outcomes were only assessed in the pre- and post-assessment. Therefore, they could not be integrated into the pre-test to FU analyses, but secondary outcomes, assessed at pre-test, could be used as predictors in the regression analyses. These were physical fitness assessed with the Senior Fitness Test (Rikli and Jones, 2001), a reliable and valid measurement of all targeted domains of the conducted physical activity program (strength, endurance, flexibility, and coordination), peripheral blood levels of BDNF, IGF-1, and VEGF, as well as ApoE and BDNF polymorphism. For a detailed overview of the procedure see Rahe et al. (2015a).

Statistical analyses were performed using IBM SPSS Statistics 25 for Windows (2017). Data were tested for normal distribution with Kolmogorov–Smirnov tests and homogeneity of variances for between-group comparisons with Levene’s tests. For all statistical comparisons, the significance level was set at α = 0.05. The groups were analyzed for differences in the baseline demographic variables age, education, sex, depressive symptoms (BDI 2), apoE4, BDNF, and cognitive status (DemTect) using ANOVAs or Chi-square tests where appropriate. We applied the same procedure to compare participants who could be included in the current analysis and dropouts to FU. Participants who attended less than 80% of the sessions (11 sessions) were not included in the analysis and regarded as drop-outs, to guarantee comparable training duration. Post hoc power analyses were performed using G^∗Power (Faul et al., 2007) to estimate achieved power.

According to the hypothesis and referring to the pre-post comparison of Rahe et al. (2015a), we treated the study as two separate trials comparing CPT vs. CT (hypothesis 1) and CPT vs. CPT+C (hypothesis 2). Change scores for the different outcome measures (general cognitive status, verbal memory, figural memory, working memory, verbal fluency, inhibition, attention, and visuo-construction) from pre-test to FU were analyzed with ANOVAs for repeated measures (rANOVAs), if the assumptions of the ANOVA were fulfilled. The within-subject variable Time had two levels (baseline vs. FU). The between-subject variable Training also had two levels (trial 1: CPT vs. CT, trial 2: CPT vs. CPT+C). For all cognitive tests, positive change scores indicate better performance at FU, except for the Stroop test, where negative change scores indicate better performances. The effect size partial η² (ηp2) indicates a small (ηp2 > 0.01), moderate (ηp2 > 0.06), or strong effect (ηp2 > 0.14, Field, 2013). Within comparisons were conducted using a t-test for pre-test to FU comparisons for each group. If the assumptions of normal distribution and homogeneity of variances were violated or variables were non-parametric, Friedman’s ANOVA was used, and effect sizes ω were reported indicating a small (ω > 0.10), moderate (ω > 0.30), or strong effect (ω > 0.50; Field, 2013).

We calculated predictions of cognitive improvement only for CPT to compare the data to those presented for the pre- post-test results by Rahe et al. (2015a). Predictors of cognitive improvement were calculated using backward multiple regressions. The change scores (FU – pre-test) of cognitive variables were calculated. According to the current literature and following the analysis conducted by Rahe et al. (2015a), the predictors age, education, sex, baseline cognitive scores, ApoE, and BDNF polymorphisms, baseline levels of BDNF, IGF-1, and VEGF, and baseline overall fitness were integrated into the regression analyses. The assumptions of multiple regression models were checked according to the suggestions of Field (2013).

In sum, N = 81 participants fulfilled inclusion criteria at pre-test, were randomized to one of the three interventions, and attended at least one training session. As participants for only two training groups could be recruited in Osnabrück, an additional group was trained in Vechta. A total of n = 26 participants dropped out until FU because of health issues, personal reasons, time constraints or less than 80% attendance of sessions. Figure 1 shows the participants’ flow through the study. For the statistical analysis, the data of the remaining N = 55 participants were used. Participants of the three intervention groups did not differ significantly in the baseline demographic variables age, education, and sex, nor in the apoE and BDNF genotype, nor the handedness, nor in the FU variables age and cognitive status (all p > 0.05, see Table 1). Also, participants and drop-outs of the study did not differ significantly in the pre-test demographics (all p> 0.05, see Table 2), except in the CT group, where participants who dropped out of the study showed a significantly lower score in the DemTect total score at pre-test, t(21) = 3.47, p = 0.002.

Baseline to FU performances of the intervention groups are shown in Table 3. We achieved a 21% power to detect small interaction effects (ηp2 > 0.01), 82% power to detect moderate interaction effects (ηp2 > 0.06), and 99% power to detect strong interaction effects (ηp2 > 0.14) (Cohen, 1988), in the comparison of the groups for testing hypothesis 1 (N = 35, 2-tailed α = 0.05). For testing hypothesis 2 (CPT vs. CPT + C), a 22% power to detect small interaction effects (ηp2 > 0.01), 85% power to detect moderate interaction effects (ηp2 > 0.06), and 99% power to detect strong interaction effects (ηp2 > 0.14) was achieved (N = 38, 2-tailed α = 0.05).

rANOVAs were used to analyze the differences between groups in all primary outcomes. As indicated in the methods section, the within-subject variable Time had two levels, and the between-subject variable Training also had two levels (trial 1: CPT vs. CT, trial 2: CPT vs. CPT+C). A significance of the within-subject variable Time indicates cognitive changes for each group.

As indicated in the Methods section, we used backward multiple regressions to analyze predictors of cognitive improvement within the CPT group. Thereby, the achievement of the best model fit can be ensured while taking into account every relevant predictor. This study had a 35% power to detect a large effect (f² = 0.35, assuming a maximum of four predictors, based on n = 18 participants at the CPT group at FU, Field, 2009). Therefore, correlations between all ten predictor variables and (...) the different outcomes (difference scores of: general cognitive state (....), verbal short- and long-term memory, figural memory, working memory, verbal fluency, alternating letter verbal fluency, inhibition, attention, planning, and visuo-construction) were calculated. Variables that had a significant correlation with the outcome variable were integrated into the regression model assuming a maximum of four predictors. In cases, where more than four variables had a significant correlation with the outcome variable, the four variables with the highest significant correlation with the outcome variable were used.

The primary results of the predictor analyses within the CPT group are: (i) lower baseline performance was a predictor of gains in verbal short-term memory (ß = -0.44), letter verbal fluency (ß = -0.59), alternating verbal letter fluency (ß = -0.40), and attention (ß = -0.64), (ii) a lower educational level was a predictor of gains in working memory (ß = -0.59), (iii) low blood levels of BDNF were predictive of an improvement in alternating letter verbal fluency (ß = -0.35), and (iv) higher blood levels of IGF-1 were also predictive of an improvement in alternating letter verbal fluency (ß = 0.36). Statistical details are shown in Supplementary Table 1.

No significant interaction effects favoring CPT in comparison to pure CT could be observed. This fact contradicts our hypothesis 1 that CPT is superior to CT, but is in line with the findings of the pre-post comparison (Rahe et al., 2015a) and other RCTs failing to support the superiority of CPT at pre-post comparison (Legault et al., 2011; Barnes et al., 2013; Shatil, 2013) and pre-FU comparisons (Linde and Alfermann, 2014). In the meta-analysis of Zhu et al. (2016), a preliminary analysis of three studies (due to limited data) also showed no significant effects for the comparison between CPT and CT. Oswald et al. (2006), demonstrated that CT and CPT can improve cognitive performance in short-term as well as in the long-term in the outcome measure attention. The intervention groups in that study were contrasted with a passive control group, while the direct contrast (CPT vs. CT) was not analyzed. Our results are further supported by an exergaming study of Anderson-Hanley et al. (2018), who also found that physical activity with high and with low cognitive challenge yielded significant moderate effects on executive function, yet there was no significant interaction. In summary, the question of whether or not CPT is favorable concerning cognitive outcome cannot be answered conclusively, but our data support the notion that both trainings yield comparable results.

Besides the question which training is superior, it should be noted that both trainings, CPT and CT, can be regarded as efficient in stabilizing or even enhancing cognitive functions in healthy older adults at post-test (Kelly et al., 2014a; Law et al., 2014). Our data provide evidence that this is still the case 1 year after intervention, indicated by the found Time effects in both groups. Also, our conducted with-in comparison showed cognitive improvements for both types of training (figural memory and attention in both types of training, and additionally, verbal short-term memory after CPT) that go beyond expected retest effects. Also, the ACTIVE trial, the study with the largest sample of healthy older adults participating in CT, showed improvements in targeted cognitive abilities after reasoning and speed training (but not memory training) over a very long period of 10 years after the intervention (Rebok et al., 2014). However, as maintenance or even improvement of cognitive functions is a crucial indicator for intervention efficiency (Zhu et al., 2016), more studies with FU are needed to confirm the long-term effects of combined interventions and CT (e.g., Fabre et al., 2002; Legault et al., 2011; Theill et al., 2013).

A significant Time× Training interaction was found when comparing CPT vs. CPT+C in the domains of general cognitive status and verbal long-term memory in favor of the CPT group. This finding is not in line with the results of the pre-post comparison of the data, in which Rahe et al. (2015b) did not find significant interaction effects in favor of the CPT. However, that study showed a significant effect in alternating letter verbal fluency and planning, favoring the CPT+C group, and we also found a trend (p < 0.01) for the alternating verbal fluency, measured with the RWT G-R test, favoring the CPT+C training group in the FU analysis. We can only speculate about the reasons for this pattern. As discussed by Rahe et al. (2015b) it is possible that the CPT+C group was involved on a more “cognitive level” than the CPT group, because their task was to plan physical activity strategically, in contrast to the CPT group, which focused particularly on the behavioral level of being physically active. This hypothesis could explain the trend observed in the alternating letter verbal fluency task, favoring the CPT+C even 1 year after the intervention, as this task requires the use of a strategy (Hughes and Bryan, 2002). According to Rahe et al. (2015a) it is possible that participants’ complaints about higher strains caused by the additional individual counseling sessions in the CPT+C training might have resulted in less training motivation, therefore in less training and consequently fewer benefits compared to the CPT. Accordingly, it may be possible that a concrete training plan without additional individual effort in planning is more appropriate and more efficient for participants who are not highly motivated to individualize their training. In contrast, as “personal training” is an emergent training concept, it can be presumed that there are individuals for who CPT+C will be suitable and even more efficient than CPT. Future research is needed to elaborate on this topic. With regard to the interaction effect in favor of the CPT for the general cognitive status and verbal long-term memory, which is reflected in improved scores at FU measurements for the CPT group on a descriptive level, while the CPT+C group did not improve (and even slightly declined in verbal long-term memory), these data emphasize the potential of a CPT to enhance cognition in long-term. However, as for the comparison of CT versus CPT, it should be noted that within-comparisons indicated cognitive improvements for both types of training - figural memory and attention for both CT and CPT, short term memory for CPT, and verbal fluency for the CPT+C.

This is the first study that investigated predictors within a CPT 1 year after intervention. In line with the results of the pre-post prediction (Rahe et al., 2015a), a lower initial baseline performance predicted gains in the domains verbal short-term memory, letter verbal fluency, alternating letter verbal fluency, and attention. This finding is also consistent with several studies showing that lower initial baseline performance is predictive of CT improvement in healthy older adults (e.g., Whitlock et al., 2012; Rahe et al., 2015a). However, also contradictory results exist (e.g., Fairchild et al., 2013), showing that participants with an initially higher baseline benefit most from CT. It is important to consider the different statistical analyses used and to differentiate between the prediction of short and long-term effects when interpreting the different patterns of results (e.g., use of backward or forward multiple regression, latent growth models, etc.,) which has not been systematically conducted yet.

The fact that an initial low baseline performance was predictive in the present study might be explained by the compensation hypothesis (Lövdén et al., 2012). The compensation account implies that participants, who are already functioning at optimal levels, have less room for improvement in CT performance. Accordingly, participants who start with an initially lower performance will gain more from CT. It is not entirely clear under which conditions this account may or may not be applicable (Lövdén et al., 2012). Lövdén et al. (2012) showed that one factor that permits predictions about the empirical condition under which compensation is more likely to occur may be the theoretical distinction between flexibility [which “denotes the capacity to optimize performance within the limits of the brain’s currently imposed structural constraints”; Lövdén et al., 2010) and plasticity (the capacity for changes in the possible range of cognitive performance enabled by flexibility (e.g., Baltes, 1987; Baltes et al., 2006)]. Applying this theory to our results, one could argue that if the brain’s performance for a task was already optimized within current structural constraints, then fewer benefits could be expected from the CT and CPT. Therefore, participants with an initially high baseline performance did not profit directly right after the training, while participants with an initially low baseline performance did so, because the extensive practice pushed these participants beyond their initial range of performance, thereby inducing plastic changes (Lövdén et al., 2012). Another reason that may explain the results can be that the training was too easy for the participants with an initially high baseline performance, resulting in less training gain based on a lack of challenge.

In the present study, lower education was a significant predictor for a gain in the domain of working memory. There are only a few studies, which investigated the association between education and memory training gains, but yet, most of them did not find significant results (e.g., Dorfman and Ager, 1989; Wolters et al., 1996). However, Rebok et al. (2013) found a significant association, indicating that healthy older adults with more education show improvement in memory performance, a finding that is at odds with our data. Notably, one can conceive of explanations for both directions of the effect. On the one hand, it could be possible that individuals with more education benefit most from CT and CPT, as education is a driver of cognitive and neural plasticity (Mandolesi et al., 2017). On the other hand, it could also be possible, according to the compensation hypothesis (Lövdén et al., 2012), that healthy older adults with less education may benefit more, as there is more “room for improvement” (see above). Furthermore, in some tests ceiling effects are possible, so that for participants who show ceiling effects, there is no room for improvement (our data indicate that this could be the case, for example, for the visuo-constructive task, i.e., copying the Rey Complex Figure). Longitudinal studies are needed to determine whether education affects the maintenance of CPT gains and in which way CT gains are affected (Langbaum et al., 2009).

Interestingly, as already shown in the study of Rahe et al. (2015a), lower initial blood levels of BDNF were a predictor of an improvement in the domain alternating letter verbal fluency, showing that for this executive task BDNF level was essential for training-induced plasticity. However, contrary results exist (Voss et al., 2013), showing that higher blood levels of BDNF were associated with a greater change in functional connectivity as measured with structural and functional MRI in a group which was trained with non-aerobic exercises. In that study, however, the relationship between growth factors and cognitive outcomes was not assessed and, therefore, the association of changes of growth factors with cognitive outcomes still needs more investigation. BDNF is highly concentrated in the hippocampus, but is also distributed throughout the entire brain (Murer et al., 1999). BDNF is considered to mediate the effects of exercise on synaptogenesis, synaptic plasticity, and enhanced learning and memory (Cotman et al., 2007). Therefore, the expression of BDNF plays an essential role in regions that are vital to learning and memory (Bekinschtein et al., 2008). In particular, BDNF is critical for synaptic plasticity and memory-processing in the adult brain (Alonso et al., 2002; Tyler et al., 2002), because it induces long-term potentiation in the hippocampus, which is a form of synaptic plasticity thought to underlie long-term potentiation formation (Izquierdo and Medina, 1997). A recent systematic review on the effect of BDNF polymorphisms on cognition could show that several studies report an association between the Val66Met polymorphism (which is one of the most extensively studied single-nucleotide polymorphism in the BDNF gene, see Chen et al., 2006) and changes of various cognitive domains (Toh et al., 2018). More precisely, the highest percentage of positive associations were shown between Val66Met polymorphism and memory (41.3%), followed by executive functions (38.3%), and attention and concentration (17.4%). However, the neurobiological mechanisms for these benefits are not fully understood, and the reason why we found that BDNF was only a predictor of executive functions remains to be further investigated. Rahe et al. (2015a) could also show a moderation effect indicating that change in physical fitness × change in BDNF significantly predicted gains in this task suggestive of high changes in physical fitness being predictive of high cognitive gains and a strengthening of this association due to higher BDNF changes. However, we could not test this assumption for FU data, because data on BDNF and physical fitness at FU were not assessed.

Higher initial blood levels of IGF-1 predicted improvement in the alternating letter verbal fluency task in FU assessment. This result can be interpreted in line with previous studies reporting the association between IGF-1 and functional connectivity change (Voss et al., 2013). The IGF-1 plays an important role in the regulation of adult neurogenesis (e.g., Fernandez and Torres-Alemán, 2012; O’Kusky and Ye, 2012; Ziegler et al., 2015). A study of Liu et al. (2009) showed that the conditional deletion of the IGF-receptor gene in mice results in an almost complete loss of the dentate gyrus, which is part of the hippocampus and essential for learning and memory. Higher levels of IGF-1 seem to play an essential role in benefits from CT, although we found an effect on executive functions, rather than memory. Since the present and the original study of Rahe et al. (2015a) are the first that investigated the association between effects of a combined lifestyle interventions on growth factors, our results warrant further investigation.

In the present study, apoE4 was not predictive for any of the investigated outcomes, even though in the pre- post-test comparison of Rahe et al. (2015a), the apoE polymorphism was predictive for performance in an alternating letter verbal fluency task, indicating that non-carriers of the apoE 4 allele showed gains in executive functions compared to carriers of the ApoE 4 allele. Yet, there was an ever smaller sample in the FU sample for apoE4 carriers than in the pre-test – post-test comparison, which may explain the missing effect; further studies with larger sample sizes will have to shed light on the potential effect of apoE4 on training effects in healthy older adults.

Overall, the identification of predictors of cognitive interventions gains increasing interest to elucidate the question of who will benefit from CT or CPT interventions to optimize interventions for specific target groups in the future.

Some limitations have to be considered when interpreting our data. One limitation is the relatively small sample size in general and due to dropouts to FU: power analysis indicated a power of only 21% (hypothesis 1) and 22% (hypothesis 2) to detect small interaction effects, so that studies with larger sample sizes might be able to detect more differences between the effects of the different types of interventions.

Furthermore, our sample was highly educated, and our sample also represents highly motivated and active healthy elderly. As outlined by Rahe et al. (2015a), this constitutes a problem affecting most intervention studies of healthy older people, as participating in a study is always voluntary and it can be assumed that (...) volunteers differ from non-volunteers, not only in (...) motivation but also in outcome expectation, socio-demographic variables, and healthy lifestyles (Oswald et al., 2006; Unverzagt et al., 2009; Schubert et al., 2014). Also, our eligibility criteria might have limited the variability in our sample, e.g., no homozygote E4/E4-carrier was identified. This could be since homozygote E4/E4 carrier, who have the highest risk to develop cognitive dysfunctions or dementia, may have been excluded as a result of our screening with the DemTect (Kalbe et al., 2004). Notably, in the CT group, six individuals dropped out before FU, who had a significantly lower DemTect score at baseline (Table 2). Reasons for their refusal to participate in the FU can only be speculated, but it is possible that these individuals had developed cognitive decline, and might have been demotivated to participate in the FU. However, this aspect needs further research.

An overview about other activities in which the participants of all three groups might have engaged during the study period and the time period until the FU test was not obtained, so that we are unable to analyze which training (especially in the comparison of CPT and CPT+C) ultimately led to more physical activities in the long term. Detailed protocols of further off-study activities, as performed by, e.g., Graessel et al. (2011) in patients with dementia, which help to estimate their possible influence and protocols assessing leisure physical activity (e.g., self-report questionnaires or a physical activity diary) should be integrated into future studies. Besides, our study did not measure physical activities and peripheral growth factors at FU. Therefore, we could not conduct FU analyses for the secondary outcome parameters physical fitness and peripheral growth factors, as in the original study (Rahe et al., 2015a).

A further limitation of the study was the fact that we did not check how long and to what extend the participants conducted their homework. For future studies, a homework-diary should be implemented to control for intensity and total time spent on the cognitive homework as a further possible factor influencing CT performance.

A passive control group was not included in the original study so that cognitive improvements induced by any of our training types can only be estimated in a limited way. However, the focus of the study was on the hypothesis that the combination of CT with other lifestyle factors is superior to pure CT and to compare the results of the pre-post comparison to the pre-FU comparisons of the conducted training.

As a final limitation, there were some constraints in the blood sampling procedure (analyses were conducted in three different laboratories, no standardization of time of blood withdrawal) and the pre-analytics for blood analyses, which are described in more detail in Rahe et al. (2015a).