Lower cognitive baseline scores predict cognitive training success after 6 months in healthy older adults: Results of an online RCT

Identifying predictors for general cognitive training (GCT) success in healthy older adults has many potential uses, including aiding intervention and improving individual dementia risk prediction, which are of high importance in health care. However, the factors that predict training improvements and the temporal course of predictors (eg, do the same prognostic factors predict training success after a short training period, such as 6 weeks, as well as after a longer training period, such as 6 months?) are largely unknown.

training response, 4 few studies have addressed the question of who benefits most from cognitive training interventions. Possible prognostic factors for improvements after a cognitive training are sociodemographic factors, cognitive abilities at entry to the training, genetic parameters, blood factors, and personality traits. 5 Furthermore, results seem inconsistent: some studies state that higher age is a positive predictor for cognitive training success in healthy older adults, 5,6 whereas others indicate that younger individuals benefit more from training. 7,8 A recent systematic review on prognostic factors of changes after memory training in healthy older individuals showed that the tendency of the prognostic factor (the more of x/the more of y vs. the more of x/the less of y) is dependent on the used dependent outcome measure of the studies (eg, whether post-test scores or changes scores were used in calculations as the dependent variable). The use of these different dependent variables has led to seemingly contradictory results regarding prognostic factors for training success in the current literature. 9 After systemizing the included studies according to their dependent variables, the authors were able to draw the preliminary conclusion that older adults seem to benefit more from memory training than younger adults, when using the change scores (post minus preperformance) as the dependent variable, answering the specific question: "Who benefits from the training?". Yet, the review also emphasizes the need for elaborated prognostic factor studies with large sample sizes, clear descriptions of prognostic factor and confounder measurements, and clear reporting standards in the field of nonpharmacological interventions to shed further light on this important topic.
A further, under-investigated aspect of prognostic factor research on cognitive training success in healthy older adults is the temporal course of the prognostic factors: Do the same prognostic factors predict training success after a short training period (eg, 6 weeks), as well as after longer periods (eg, 6 months or even 1 year)? However, it is also important to consider the difference between predictors of training success of studies that provide an intervention for a specific time frame (eg, 6 weeks 10 ) and then investigate predictors for cognitive function at follow-up times in contrast to studies in which the participants have ongoing training, and predictors are investigated at different measurement periods throughout this training. To the knowledge of the authors, no study has focused on the latter aspect. Therefore, the present paper investigates who benefits from an online general cognitive training (GCT) intervention in healthy older adults by identifying predictors of the ongoing training intervention at 6 weeks, 3 months, and 6 months. For that purpose, data from an already published RCT were reanalyzed. 11 2 | METHODS

| Study design
Data were taken from a double-blind 6-month online randomized three-arm controlled trial with healthy older adults. In a previous paper, short-and long-term effects of this RCT were reported, 11 showing that GCT and reasoning cognitive training (ReaCT) conferred a benefit to self-reported instrumental activities of daily living scores as well as reasoning and verbal learning at 6months. In the present study, only data from the GCT and the active control group (CG) were used, but with four measurement times at baseline, 6weeks, 3months, and 6months. The present study only focused on the evaluation of predictors of changes after GCT, as the GCT targets multiple cognitive domains and, therefore, differs substantially in its concept from the ReaCT, which targets primary executive functions. The CGT resembles most cognitive trainings which are offered to older people in the context of prevention of cognitive decline. Thus, the identification of predictors of GCT is of high relevance. Even though, predictors of changes after ReaCT are also of interest, their analyses and discussions lay beyond the scope of the present paper.

| Participants
Eligible participants for the study were individuals older than 50 years of age with access to a computer and the internet. Through a partnership with the British Broadcasting Corporation (BBC), Alzheimer's Society (UK), and the Medical Research Council, all adults older than 50 years in the United Kingdom and internationally were invited to take part in this online RCT. Interested older individuals were invited to register and consent through a secure connection and an ethically approved online process to the study. Participants then received their own login details and were randomized to a study group (GCT, ReaCT, or CG). Throughout the intervention, participants received reminder emails to continue their training and complete their online cognitive assessments.

| General cognitive training
In the present study, only data from participants of the GCT compared to an active CG were investigated. Participants were recommended to

Key Points
• Prediction analysis of n = 4,185 healthy older adults revealed that sex and cognitive baseline performance were significant predictors of changes when performing a GCT.
• There is a time course underlying significant predictors for changes when performing a GCT: female sex was predictive for gains in grammatical reasoning after 6 weeks of training, and cognitive baseline level at study entry was predictive for GCT gains in tests for spatial working memory and verbal recall after 6 months, but not after 6 weeks or 3 months of training.
train for 10 minutes daily, even though flexibility in training duration was allowed. The GCT consisted of six cognitive tasks that trained attention, memory, mathematics, and visuospatial abilities. An overview of the tasks is provided in Table 1. The CG performed online tasks involving a game in which people were asked to put a series of statements in the correct numerical order. Outcome measures were changes in grammatical reasoning, spatial working memory, digit vigilance, verbal short-term memory, and verbal learning. Grammatical reasoning was measured using the total number of trials answered correctly in 90 seconds minus the number answered incorrectly in the Baddeley grammatical reasoning test. 12 Spatial working memory was measured with the widely used spatial working memory test 13 in which participants searched a series of on-screen boxes to find a hidden symbol. The main outcome was the change in the score of the average number of boxes in the successfully completed trials. Digit vigilance was measured through a version of the "digit span" task, in which each successful trial is followed by a digit span that is one digit longer than the last one, and each unsuccessful trial is followed by a digit span one digit shorter than the last. The main outcome measure was the average number of digits in all successfully completed trials. The paired associates test 14 was used to measure verbal short-term memory. In the test participants see a series of objects, one at a time, and select the correct location of each object in "windows" they had previously been shown.

| Outcome measures
The main outcome measure was the average number of completed correct object-place associations in the trials. Verbal learning was measured by changes in the recognition score on the revised Hopkin's Verbal Learning Test. 15 The test is comprised of six alternate forms, each containing 12 nouns and 4 words, which are taken each from one of three semantic categories to be learned over the course of three learning trials. This is followed by a recognition trial 20 to 25 minutes later composed of 24 words, including the 12 target words and 12 false positives.

| Statistical analyses
Statistical analyses were performed using R. 16 For all statistical comparisons, the significance level was set at α=.05. Descriptive statistics Training sessions included in the general cognitive training packages  20 We are particularly interested in the results of the interaction terms

| Predictors of cognitive training success at all three measurements
An overview of the results of the prediction analyses of all three time points (6 weeks, 3 months, and 6 months) is provided in Table 3. Furthermore, a simplified overview of the significant interaction terms (indicating significant predictors for the GCT compared to the CG) and their effect sizes are depicted in Table 4. Only significant interaction terms with an effect size that indicates at least a small effect (β ≥ .10) are reported in Table 4.

| DISCUSSION
The aim of the present paper was to identify predictors for GCT success in healthy older adults of an ongoing online GCT at 6weeks, 3months, and 6months. Our main results are that (a) sex and cognitive baseline performance were significant predictors of training success and (b) there is a time course underlying these predictors. More specifically, we found that female sex was predictive for gains in grammatical reasoning after 6weeks of training, but not after 3 or 6months T A B L E 4 A simplified overview of the significant interaction terms in the multiple regressions at 6 wk, 3 mo, and 6 mo of training. Furthermore, our results indicate that cognitive baseline level at study entry was predictive for GCT gains in tests for spatial working memory and verbal recall after 6months, but not after 6weeks or 3months of training.
Regarding predictors of training success, the fact that being female was a significant predictor for gains in a verbal task, measuring grammatical reasoning (though only after 6 weeks) of GCT, is remarkable. To date, sex differences in cognitive training interventions have rarely been studied in healthy older adults. However, one other study with older participants with mild cognitive impairment found that women showed stronger improvements in verbal tasks (immediate and delayed verbal episodic memory and verbal working memory) after a 6-week multidomain cognitive training program. 21 Notably, meta-analytic data demonstrate that healthy women perform better than men on tests of verbal learning and memory, 22 and women also outperform men in syntactic complexity and grammatical diversity. 23 Taken together, the data might point to "sex-specific plasticity," and more particular, stronger plasticity for verbal tasks in women. 24 Furthermore, a low cognitive baseline level at study entry was a significant predictor for gains in the GCT group (only at 6 months) in spatial working memory and verbal learning. This finding is in line with several other studies that found lower cognitive baseline level at study entry to be predictive of cognitive training improvement. 10,25 The compensation hypothesis 26 may account for this pattern; it implies that healthy older adults who are already functioning at optimal levels have less room for improvement in GCT performance, whereas those with low function may improve to a greater degree.
Regarding the time pattern of prediction of training success, our study showed that being female is only predictive for improvement in grammatical reasoning at the 6-week measurement, but not after 3 or 6 months of training. Yet, it may be possible that women might be more capable than men of activating their former resources in verbal domains immediately at the beginning of the training, 24 meaning that verbal resources are stronger in women and enable a faster activation of knowledge and strategies in this domain but that this sex-specific advantage diminishes over time. However, this aspect will have to be further investigated in future studies.
We also found that lower cognitive baseline performance at study entry is only a significant predictor after 6 months of training, but not earlier in the course of the training. This may be interpreted based on how participants profit in a comparable way during a longer period of time independently of their baseline level, but after 6 months, participating in a GCT is more successful for individuals starting with lower baseline performance. One explanation is that individuals with higher cognitive baseline levels reach their limit earlier, whereas those with a lower cognitive baseline level have a longer time period in which they may improve. It is important to note that in the literature, several studies have found that cognitive test performance at study entry is also predictive for gains after shorter periods of 6 to 10 weeks of cognitive training 10,27,28 -results that contradict our findings. Possible reasons for this inconsistency remain speculative but could lie in the use of different cognitive trainings or statistical methods (eg, the inclusion of the CG in the multiple regression in our study). However, a comparison of our study with other training studies is also difficult because our data refer to an ongoing training with an ongoing training also at follow-up measurements, whereas most other studies have a specific training duration (eg, two times a week 10 ) and predictors of training success then refer to postintervention (which would be comparable to our prediction analysis after a shorter period of time, for example, 6 weeks) or follow-up examination after a period of no training (which we do not have). In more detail, instead of a classical pre-intervention-post-FU design used in most studies in which no training is conducted between post-test and FU, the present study had several measurement points (at 6 weeks, 3 months, and 6 months) in which the intervention was still ongoing (pre-testintervention-6 week measurement-intervention-3 month measurement-intervention-6 months measurement-intervention).
Results showed that the CG trained significantly less than the intervention group, indicating a possible loss of motivation to participate in the study. As we did not collect data on training motivation, reasons for this remain speculative. Participants in the CG may not have enjoyed the offered games or may not have had the feeling of efficiency. Future studies need to ensure an active control group that has equally challenging and interesting tasks compared to the intervention group.
Particular strengths of the present paper are the fact that it reports on the first study to investigate predictors of cognitive training success over the time course in a large sample taken from an RCT.
Yet, as a possible limitation, it has to be kept in mind that the sample may be biased due to the fact that often highly educated and highly motivated participants conduct cognitive trainings, 29  to avoid such bias. As a further limitation, we did not correct for multiple testing (eg, by using the Bonferroni correction) due to the fact that it was an exploratory study investigating the time course of possible predictors for changes after memory training. Yet, we only discussed predictors with high effect sizes (>.1). However, future studies should imply corrections for multiple testing to confirm the found results.

| CONCLUSION
To conclude, our study showed that sex and cognitive variables may predict GCT success and there seems to be a differential time course for this prediction. As patterns of training success prediction might help to tailor cognitive trainings to individuals with different profiles, research should further unravel prediction patterns and their underlying mechanisms. Ultimately, this research might help to optimize the prevention of cognitive decline in a personalized medicine approach.