The systematic literature search was conducted using PubMed, Embase, Cochrane Library, Web of science, China National Knowledge Infrastructure, China Science and Technology Journal Database, Wanfang Database, and the Chinese Biomedical Database, covering the period from each database’s inception to September 30,2023. Additionally, a secondary search was conducted through the references of pertinent studies and reviews. The detailed search strategies are outlined in Supplementary Appendix 1.

The inclusion criteria were formulated based on the PICOS (population, interventions, controls, outcomes, study design) framework.

All retrieved literature was imported into EndNote X9. Duplicates were eliminated, and studies were preliminarily selected based on titles and abstracts. Subsequently, it is necessary to download and read the full texts to include primary studies that meet the research criteria. Before data extraction, an electronic information extraction form tailored to this project’s standards is used for evaluation. Two researchers (a and b) independently extracted data from the included studies, covering characteristics such as the first author’s name, year of publication, research location (country), citation, study type, participant characteristics, experimental and control group interventions, intervention duration, frequency, participant age, gender, and outcome measures. Discrepancies in literature screening and data extraction were resolved through discussion between the two researchers.

Two researchers independently assessed the risk of bias in all included studies using the Rob2.0 tool, as recommended in the Cochrane Handbook (Sterne et al., 2019). The assessment encompassed randomization methods, blinding, allocation concealment, intervention allocation, compliance, missing outcome data, outcome measurement, and selective outcome reporting. Each category was classified as low risk, unclear, or high risk. Any disagreements were discussed and resolved by the two researchers.

Initially, 4440 studies were screened. After removing duplicates, 3758 articles were reviewed, with 3484 excluded based on title and abstract. Consequently, 283 articles were assessed in full text, adhering strictly to the inclusion and exclusion criteria. Ultimately, 47 studies met the inclusion criteria. Additional searches in the references of published systematic reviews identified three more eligible studies, resulting in 50 randomized controlled trials included (Chen et al., 2006, 2019, 2020; Liu et al., 2006, 2019, 2022; Ou et al., 2007; Shin et al., 2008; Yang and Liu, 2008; Wang et al., 2009, 2013; Li and Gan, 2010; Kim et al., 2011; Zhang et al., 2012, 2018, 2020; Feng et al., 2014; Ye et al., 2014; Zucchella et al., 2014; Xu et al., 2015; Yoo et al., 2015; Li M. et al., 2016; Li X. et al., 2016; Cao et al., 2017; Gong, 2017; Jiang et al., 2017; Li et al., 2017; Lv et al., 2017; Maier et al., 2017; De Luca et al., 2018; Prokopenko et al., 2018, 2019; Xiao, 2018; Mao et al., 2019; Wang, 2019; Xiao and Liang, 2019; Yeh et al., 2019, 2022; Faria et al., 2020; Fu et al., 2020; Jung et al., 2020; Shao, 2021; Xuefang et al., 2021; Ho et al., 2022; Liao et al., 2022; Shang et al., 2022; Sun et al., 2022; Xia et al., 2022; Zhao et al., 2020, 2022). The specific screening process is depicted in Figure 1. These studies encompassed 3063 participants, with an average sample size of 61 subjects (range 11–149) and an age range of 42–74 years. Of these, 1274 (41.6%) were female and 1789 (58.4%) male. The geographic distribution of the 50 trials included 74% from China, 6% from Taiwan, 6% from South Korea, 4% each from Italy and Russia, and 2% each from the United States, Portugal, and Spain. Attempts to collect racial data were largely unsuccessful due to infrequent reporting. Eleven different cognitive interventions were administered, including: (1) VRCT (n = 2). (2) VRCT combined with traditional cognitive training (n = 6). (3) VRCT combined with CBCT (n = 1). (4) CBCT (n = 17). (5) CBCT combined with traditional cognitive training (n = 9). (6) Cognitive training (n = 42). (7) Exercise combined with CBCT (n = 2). (8) Exercise combined with traditional cognitive training (n = 1). The baseline characteristics of the subjects are shown in Table 1.

This study included 50 papers. While most studies detailed their randomization process, 13 did not specify the method of random sequence generation. Only one study (Sun et al., 2022), described allocation concealment, leading to the classification of the others as having an ‘unknown risk’ of bias in this regard. Due to the nature of the interventions, blinding of participants was not feasible. A total 13 studies described blinding methods, while the remaining were assessed as having an unknown risk. Concerning baseline data consistency, all studies reported comparable baseline data with no statistical differences, thus receiving a ‘low risk’ rating. In terms of outcome data completeness, most studies accounted for all participants at the time of outcome reporting, aligning with the number initially included and were thus rated as ‘low risk’. However, seven studies reported data loss or dropout rates exceeding 5% of the original sample size, which was assessed as ‘unknown risk’. Three studies were evaluated as ‘unknown risk’ due to the inaccessibility of their original statistical methods. The detailed methodological quality assessment is provided in a table, and the risk bias assessment is illustrated in Figure 2.

In total, 50 randomized controlled trials involving 3063 participants were included to assess the impact of various cognitive training methods on cognitive function in patients with PSCI. The network structure diagram, representing various interventions across different outcome indices, is presented in Figure 3. The line thickness in the diagram correlates with the number of pairwise intervention comparisons - the thicker the line, the greater the number of comparisons. Circle size corresponds to the sample size involved in each intervention. The study analyzed both the consistency and inconsistency of data across each outcome index. A difference of less than 5 in the DIC results between the consistency and inconsistency models suggested data consistency for each outcome. Closed loops were observed in the MoCA, MMSE, LOTCA, BI, and FIM scales network diagrams. The node-splitting method was employed for local inconsistency tests on these closed-loop outcomes, yielding P-values all above 0.05, indicating no significant local inconsistency among interventions forming these loops. Publication bias for the included studies was assessed using comparison-correction funnel plots for all outcomes, which demonstrated basic symmetry and a roughly symmetric distribution of studies on both sides of the central line, suggesting a low risk of publication bias (Figure 4).

To our knowledge, this study represents the first network meta-analysis comparing various cognitive training treatments for PSCI patients. It encompassed 50 randomized controlled trials involving 3017 subjects, analyzing nine cognitive interventions. Our study found that cognitive training can positively affect cognitive function, daily living function, functional independence and motor function in PSCI patients. The combination of CBCT and traditional cognitive training demonstrated the most significant impact on overall cognitive function improvement as assessed by the MMSE and LOTCA scales, ranking second only to the exercise combined CBCT group as assessed by the MoCA scale. The combined CBCT and traditional cognitive training group showed significant improvements in attention, abstract ability, memory, and orientation compared to other training groups. Furthermore, it ranked highest in terms of impact on daily living function as measured by the BI index, FMA-based daily living function, and FMI-based functional independence. This study primarily revolved around these simplified cognitive screening instruments and life function scales, which facilitate rapid and straightforward assessments of patients’ cognitive functions. Although the obtained results might be approximate, such screening tests are adequate when scores are low and the patient’s medical history strongly indicates dementia, along with staging and monitoring of cognitive impairment. Of course, within the literature we examined, several studies utilizing a variety of neurocognitive and neuropsychological measures have been conducted to uncover the potential benefits of cognitive training on various cognitive dimensions in patients suffering from PSCI. However, due to the heterogeneity of neuropsychological tests employed across studies, the limited recurrence of the same test, and the subtle variances in the focus of interventions and anticipated outcomes, we did not extract the results of specific tests individually and perform meta-analyses. Instead, we adopted a more macro and inclusive methodological strategy, aiming to capture the overall trend of cognitive training enhancing cognitive function in PSCI patients as a whole. By reviewing these articles, we discovered that the majority of studies employed the Digit Span (forward and backward) tests to assess patients’ short-term auditory memory and working memory capacity, as well as the Trail-Making Test (parts A and B) to evaluate the attention process of PSCI patients. Following the administration of these cognitive interventions, there was an improvement in clinical efficacy, but the statistically significant difference between groups was rare. Additionally, our study revealed that CBCT in conjunction with traditional cognitive training yielded a more significant impact on enhancing attention and memory compared to traditional cognitive training or rehabilitation alone. This finding contrasts slightly with the results of De Luca et al. (2018) and Xiao (2018) studies, which did not demonstrate a distinct advantage for the CBCT combined with traditional cognitive training group. This might be associated with the scale employed. The assessment of memory and attention using the MoCA scale is frequently simplistic and approximate, which can easily exaggerate or overlook the cognitive function scores of patients, leading to conspicuous data disparities. Conversely, neuropsychological tests are typically more meticulous and intricate, enabling the assignment of more precise scores and minimizing potential errors.

The inaugural ‘Adult Stroke Rehabilitation Guideline’ issued jointly by the American Heart Association (AHA) and the American Stroke Association (ASA) in 2016 recommended cognitive function rehabilitation training for post-stroke patients at a Level IA (Winstein et al., 2016). Jeffrey M. Rogers et al.’s systematic review highlighted that cognitive training effectively enhances cognitive function in PSCI patients. Our study’s findings align with these perspectives. In the 50 randomized controlled trials employing various cognitive training interventions for PSCI patients, we observed that both traditional cognitive training and CBCT combined with traditional cognitive training improved all study outcome indicators. These interventions notably enhanced overall cognitive function and daily living abilities. Furthermore, compared to traditional cognitive training alone, combined interventions often yielded greater cognitive benefits and effectiveness. Specifically, CBCT combined with traditional cognitive training emerged as the most effective in enhancing attention, abstract function, memory, orientation, daily living function, motor function, and functional independence. Our analysis revealed consistent results between the MMSE and MoCA scales. The top three interventions were identified as a combination of cognitive training and CBCT with traditional cognitive training, VRCT combined with traditional cognitive training, and exercise combined with CBCT. It suggests that these three combined interventions have advantages in improving the overall cognitive function of PSCI patients.

CBCT offers personalized and adaptive content tailored to the patient’s specific cognitive impairments, adjusting in real-time to the feedback received during training sessions. This adaptability ensures that tasks are neither too simple nor too complex, thereby enhancing patient engagement and concentration in the training process (Chuang et al., 2023; Maggio et al., 2023). The study observed that older adults demonstrate greater enthusiasm and compliance for CBCT compared to general cognitive training. Effective cognitive training requires participants to fully commit to the regimen to achieve an adequate ‘training dose’ and thereby reap the benefits (Turunen et al., 2019), which supports the finding that CBCT tends to result in better outcomes. The research indicates that computer games used in this training not only enhance visual-spatial abilities but also improve attention. This is achieved by training the spatial and temporal resolution of attention in these games, thus enhancing attention control and task focus (Bavelier and Green, 2019). Research by Richard E. Mayer suggests that games focusing on a single cognitive skill can enhance cognitive functions such as memory, attention, and spatial cognition. This enhancement requires participants to repeatedly practice the skill in varied environments and with progressively increasing challenges (Mayer, 2019). Cognitive training operates on repeated practice in specific cognitive domains, using tasks involving particular skills to achieve training objectives. Cognitive training is founded on the repetition of specific cognitive domain training, and the intended outcome is realized through the execution of tasks that necessitate specific skills. This renders CBCT, a training approach that incorporates computer games, more effective in enhancing the attention, memory, and visuospatial abilities of PSCI patients. This is in line with the fact that CBCT, alone or in conjunction with traditional cognitive training, can enhance visuospatial and executive function, attention, memory, and orientation. However, despite their adaptability, these technological tools cannot entirely replace traditional cognitive training, especially for more severe and subacute cases, where the role of neuropsychiatrists and speech therapists is crucial. In fact, combining computer-based with traditional cognitive training often results in optimal outcomes (Maggio et al., 2023). A meta-analysis of 32 studies involving 1837 subjects corroborates this study’s findings, revealing that combining computer-based and traditional cognitive training significantly enhances overall cognitive function and daily life function more than conventional cognitive training alone (Nie et al., 2022). In summary, CBCT synergistic intervention demonstrates effectiveness in enhancing cognitive function and mitigating neurological deficits in patients with PSCI. The underlying mechanism is potentially associated with enhanced cerebral blood flow, increased serum BDNF levels, suppressed VILIP-1 and NSE expression, and the repair of damaged neurons. These factors contribute to improving patient prognosis and accelerating the recovery process (Xuefang et al., 2021).

VRCT, akin to CBCT, offers high adaptability by providing personalized training tailored to the participant’s cognitive and physical conditions. This training method features enhanced ecological validity, creating a three-dimensional, digital environment where subjects engage in cognitive tasks within virtual recreations of familiar daily activities (Moreno et al., 2019). Ana Lúcia Faria et al.’s study highlighted that VRCT yields greater cognitive benefits than traditional methods (Faria et al., 2020), aligning with this study’s findings. The effectiveness of VRCT may stem from its immersive experience. By engaging participants in situational daily life scenarios, it offers multisensory, natural, and realistic stimulation. This immersion significantly heightens patients’ action awareness, self-identity, and self-cognition, fostering high interest and enjoyment in VR usage (Choi et al., 2019), thereby comprehensively exercising their abilities and enhancing task success through timely feedback (Faria et al., 2020). Additionally, the gaming aspect of VRCT can also improve cognitive functions. However, it’s noteworthy that participants tend to learn better on desktop computer screens compared to VR games (Mayer, 2019), potentially explaining why CBCT is more effective than its VR counterpart. To elucidate the advantages of VRCT, researchers have proposed various mechanisms. Carrieri et al. suggested that VR can stimulate the reactivation and enhancement of diverse cortical functions, particularly the dorsolateral/ventrolateral prefrontal cortex, optimize sensory cortex efficiency, and boost cognitive function (Carrieri et al., 2016). Flannery et al. noted that VRCT training activates brain metabolism, increases cerebral blood flow, and enhances neurotransmitter release (Flannery, 2002).

Sports combined with cognitive training encompasses two modalities: sequential exercise and cognitive training, and dual-task cognitive training, which concurrently emphasizes physical and cognitive training. This approach is more apt for enhancing the comprehensive skills required in daily life. Prior research indicates that such training is more advantageous in improving overall cognition, working memory, and executive function in elderly individuals with cognitive impairment than conventional treatments (Guo et al., 2020). I-Ching Chuang et al. discovered that exercise combined with cognitive training is more beneficial in enhancing overall cognitive function, particularly in visual-spatial working memory and language memory, compared to either physical or cognitive training alone. However, this approach does not demonstrate significant advantages in daily living abilities (Chuang et al., 2023), aligning with this study’s findings. It is posited that both simultaneous and sequential exercise and cognitive training may impact different cognitive aspects (McEwen et al., 2018). While there is ongoing debate regarding the superiority of each training mode, it is established that exercise combined with cognitive training constitutes an effective intervention. This efficacy could be attributed to exercise’s role in elevating brain neurotrophic factors in the peripheral blood (Marinus et al., 2019), crucial for the survival, maintenance, and regeneration of specific brain neuron populations (Allen et al., 2013). There is evidence indicating that physical exercise enhances the volume of the prefrontal cortex and the prehippocampus, fosters nerve and angiogenesis, and plays a crucial role in reducing cardiovascular risk factors. Furthermore, it has been discovered that the combination of exercise and cognitive training often yields more substantial benefits than either exercise or cognitive training alone. The neurological and cognitive benefits derived from exercise may be associated with increased environmental exposure during cognitive challenge (Karssemeijer et al., 2017). While environmental enrichment is considered to enhance language development in young children, improve cognitive abilities, and reduce the risk of dementia. Increased physical engagement in the form of exercise can enhance cognitive performance, mitigate memory impairment, and decrease the likelihood of Alzheimer’s disease in older adults (Figuracion and Lewis, 2021). In 2016, the inaugural “Adult Stroke Rehabilitation Guidelines” jointly published by the AHA and the ASA also issued Grade-A recommendations for cognitive activities within environmentally enriched settings (Winstein et al., 2016). However, only three related studies were included. Additionally, only the cognitive outcome of the MoCA scale demonstrates that exercise in conjunction with CBCT yields a more favorable effect than CBCT combined with traditional cognitive training. Further high-quality randomized controlled studies are required to ascertain the superiority and inferiority of exercise in combination with CBCT and CBCT in conjunction with traditional cognitive training in enhancing cognitive function and daily living abilities.

Regarding the improvement of motor function, interventions involving CBCT combined with traditional cognitive training are significant. The identical outcomes were obtained in the research, where the functional independence rating served as the outcome index. It should be highlighted that only CBCT in combination with traditional cognitive training, traditional cognitive training, and the conventional control group were incorporated in the study evaluating motor function. Both interventions demonstrated significant improvements compared to the conventional control group. The enhancement of motor function induced by CBCT and traditional cognitive training might be associated with the following reasons. One is the unique training mode of CBCT. CBCT engages participants in using specialized buttons and joysticks to exercise the movements of the wrist, thumb, and index finger of the impaired hand. This approach allows participants to enhance hand flexibility while engaging in cognitive tasks, thereby improving limb coordination in post-stroke patients (Otfinowski et al., 2006), which may account for the reason why the combination of CBCT with traditional cognitive training enhance motor function more effectively than traditional cognitive training alone. It was widely recognized that there is a connection between movement disorders and cognitive decline, although the underlying mechanism remains incompletely understood (Kobayashi-Cuya et al., 2018). In a prospective cohort study conducted by Zhangyu Wang et al., MRI was used to evaluate brain structural volume (including total brain volume, total white matter, total gray matter, hippocampus volume, and white matter high signal) and statistically analyzed in conjunction with motor function and cognitive function data. Ultimately, it was demonstrated that motor impairment is associated with declines in cognitive functions such as overall cognition, episodic memory, semantic memory, working memory, visuospatial ability and perceptual speed (Wang et al., 2023). Paul Rinne et al. identified a consistent and strong correlation between attention control and motor performance. They posit that poor attention control following stroke is one of the causes of limb paralysis after stroke, furthermore they suggest that attention control itself may be a therapeutic target for improving motor disorders post-stroke (Rinne et al., 2018). Additionally, previous research has shown that most patients with left hemisphere damage and a minority with right hemisphere damage exhibit impaired ipsilateral hand dexterity within a month post-infarction involving the parietal or posterior frontal lobes. This impairment may be linked to post-stroke cognitive impairment affecting action perception and control (Sunderland et al., 1999). Paul Rinne and colleagues interpreted this to mean that motor function impairment on the same side of the stroke focus was selectively associated with the lesion load on the attention network, but not with the anterior corticospinal tract (Rinne et al., 2018). In view of the relationship between motor function and cognitive function, it is possible that with the improvement of cognitive function, the motor function of PSCI patients will also be restored. Although the link/correlation is not clear and not necessarily causative.

In the assessment of cognitive domains using the MOCA scale, it is indicated that aside from language and naming functions, various cognitive training methods positively influence the improvement of visual-spatial and executive functions, attention, abstraction, delayed recall, and orientation. CBCT combined with traditional cognitive training is most effective in enhancing attention, abstract function, memory, and orientation. These results fit with the advantages brought by the unique cognitive training mode of CBCT. However, our findings suggest that various cognitive training programs did not show any advantage in language or naming. In contrast, computer-assisted cognitive training demonstrates superior benefits in improving visual-spatial and executive functions. This slightly diverges from Anastasia Nousia et al.’s study, which found that CBCT enhances language fluency, naming, and delayed memory (Nousia et al., 2021). Our study exclusively analyzed the various cognitive areas of the MoCA scale, suggesting the necessity for multiple outcome indicators within each cognitive domain, as improvements in a few indicators for the same function are less convincing than enhancements across multiple indicators within one function.

Our results align with most studies under similar outcome indicators, yet they diverge from some. A study affirmed the efficacy of CBCT and cognitive training in enhancing cognition in PSCI patients but did not demonstrate the superiority of CBCT in cognitive rehabilitation of post-stroke patients (Mingming et al., 2022). Compared to our study, their sample was smaller, encompassing only four studies from 2013-2021 with 96 participants, which may introduce bias. A network meta-analysis involving 1047 PSCI patients from 21 RCTs indicated that computer-based and VR-based cognitive training were superior to conventional training in MOCA, with the computer-based group showing the best therapeutic effect. However, this did not reach statistical significance under the MMSE index (Xiao et al., 2022). This study posits that, aside from combined interventions, CBCT surpasses conventional training in MoCA, possibly due to the limited number of studies directly comparing VRCT with traditional training included in this research.

This research entailed an extensive review of studies assessing the efficacy of various cognitive training methods in improving cognitive function in patients with post-stroke cognitive impairment (PSCI). The outcome measures included assessments of cognitive function, daily living, motor function, and functional independence. Given the large sample size and the narrow confidence intervals in this network meta-analysis, we consider the results to be reliable. A key strength of this study is its analysis of diverse cognitive training modalities and detailed examination of different cognitive domains, aspects not covered in previous meta-analyses of cognitive training for PSCI. However, there are several limitations. Firstly, there was no analysis of the heterogeneity of baseline characteristics. The included studies varied in intervention duration and frequency, total intervention time, measurement timings, and the computer platforms used in computer-based and VRCT, contributing to significant heterogeneity. Secondly, only three studies on sports cognitive training were included, with a relatively small sample size, raising concerns about the applicability and accuracy of the results. Thirdly, although all included randomized controlled trials provided conventional drug therapy and rehabilitation training in addition to cognitive training interventions, and studies significantly impacted by drug effects on cognitive function were excluded, the specific roles of these treatments remain ambiguous. This lack of clarity is due to insufficient details about the types and dosages of drugs used post-stroke. Fourthly, the influence of VRCT, which often involves whole-body engagement, is challenging to separate into effects attributable to cognitive training versus physical exercise. Fifthly, the included studies did not adequately describe allocation concealment, and since blinding participants to the interventions was unfeasible, only a few studies blinded the raters. Therefore, the study may contain some subjective biases. Finally, since 66% of the original studies included in this study were in Chinese, the extrapolation of the conclusions of this study has certain limitations.