The labels promise different designs
The paper starts from a practical problem: therapeutic exercises may help slow cognitive decline, but conventional paper-based or computer programs are often tedious and hard to sustain. VR is proposed as an alternative that can make training immersive and realistic, so users feel present and stay engaged.
Within that alternative, the literature splits systems into two types. Game-based systems use rewards, stories and play to keep users motivated. Training-based systems focus on structured, clinically grounded exercises aimed at functional recovery. The paper notes that prior reviews disagree about which works better, and that those reviews treat the two labels as different without checking whether the systems differ in any codeable feature. That gap is the question the study addresses directly.
How 34 studies were classified
The authors followed PRISMA 2020 guidelines and searched four databases for English-language studies published between 2015 and 2026. The initial search was deliberately broad, covering motor as well as cognitive VR rehabilitation; during title and abstract screening the scope was narrowed to cognitive rehabilitation for age-related conditions, mild cognitive impairment, dementia and Alzheimer's disease, and studies focused solely on motor or physical rehabilitation were excluded.
From 3,166 retrieved records, duplicate removal and screening left 34 studies that met all criteria. These were split into two tiers: 16 controlled efficacy trials (Tier 1) used to assess intervention effectiveness, and 18 design and feasibility studies (Tier 2) that describe system design without providing effectiveness evidence. Each study was described using five design features: intervention type, level of immersion, feedback modality, difficulty adaptation and hardware, plus population characteristics and targeted cognitive domains.
Where the two paradigms overlap
The central result is that the split did not hold. Immersion, feedback, difficulty adaptation and hardware were distributed similarly across game-based and training-based categories, and almost half of the studies used a hybrid approach. In the authors' reading, the paradigm label does not predict a system's technological configuration.
This matters for how earlier disagreements are interpreted. If the labels do not identify distinguishable systems, then inconsistencies across prior reviews cannot be explained by the paradigm label alone. The paper argues the differences more likely come from design features and study methodologies. That is the authors' interpretation of their taxonomy result, not a separate experimental finding.
What the controlled trials actually show
The 16 controlled studies indicate a consistent cognitive improvement, according to the paper's summary of their reported outcomes. The important qualification is that these benefits are not related to the design paradigm. Instead, the authors point to methodological choices — comparator selection, outcome measures and sample size — as playing a crucial role in the observed effects.
The paper also draws a specific link between effect size and comparison condition: the size of the effect is more connected with the comparator than with the intervention itself, and passive controls tend to provide exaggerated benefits. Its recommendation follows from that observation: use an active comparator equivalent in training dose and contact time. The two smallest studies, with 10 and 18 participants, failed to show clear results because they were too small.
A seven-dimension quality score
To assess study quality, the authors introduce VR-RES (VR Rehabilitation Evidence Score), a framework with seven dimensions, each scored from 0 to 2. The framework is meant to capture issues specific to software-based interventions that clinical trial tools such as RoB 2, the Downs and Black checklist and GRADE do not cover, including system reproducibility, long-term effectiveness and performance across different populations.
Applying VR-RES, the paper reports that methodological quality in earlier studies was moderate. Longitudinal follow-up was nearly absent: only one controlled trial reported follow-up measurements after the intervention. Blinding and reproducibility received limited attention, and none of the systems shared a complete protocol and runnable system. Demographic moderators such as age, sex and education were often reported but rarely analyzed; only two studies examined outcomes across demographic subgroups.
Why paradigm comparisons misfire
The paper's limitations section ties the taxonomy result back to the review literature. Some prior reviews reported a cognitive advantage for game-based VR, while others found no consistent paradigm effect. The authors identify a shared limitation in all of them: none evaluated whether the systems were genuinely different before comparing the two paradigms.
Because the paradigms are not distinguishable on the five coded features, the paper concludes that any paradigm-level advantage is associated with design features and trial conditions rather than with the label. It also notes that prior work tends to evaluate clinical effectiveness while rarely analyzing technological design systematically, which makes it hard to link outcomes to particular design characteristics.
What future trials should prioritize
The recommendations follow directly from the two findings. Rather than comparing whole paradigms, the paper suggests varying one design factor within a single system — for example adaptive versus fixed difficulty — so that a specific feature can be isolated. It also calls for active comparators matched in training dose and contact time, and for sample sizes planned around real effects rather than the very small groups that produced unclear results.
The broader priority is reporting and evaluation rigor: specific design features, reproducible systems, longer follow-up and analysis of demographic factors. The most decision-relevant point for readers comparing VR rehabilitation studies is that the game-based versus training-based label carries little information about what a system actually does; the coded design features and the trial's comparator and outcome choices carry more.
Sources used in this answer
Game-Based versus Training-Based Virtual Reality for Cognitive Rehabilitation in Mild Cognitive Impairment and Dementia: A Systematization of Design Paradigms, Outcomes, and Evaluation Rigor
Across 34 studies, game-based and training-based VR systems overlapped substantially on immersion, feedback, difficulty adaptation and hardware, so the labels do not clearly distinguish their technological characteristics; controlled trials showed consistent cognitive improvement not tied to paradigm.