PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
February 21, 2026BMC Medical Informatics and Decision Making0 citationsOpen Access

Prediction models for adherence to cardiac rehabilitation programs in patients with cardiovascular disease: a scoping review

View Full Paper
CXChengyu XiaGHGuo HtLJLiuxia Ji

Key Result

Prediction models for adherence to cardiac rehabilitation programs in patients with cardiovascular disease demonstrated AUROC values ranging from 0.62 to 0.893, with non-adherence rates varying from 41% to 61.4%, but existing models lack external validation and methodological rigor.

Key Points

  • The aim is to evaluate the quality of prediction models for adherence to cardiac rehabilitation in cardiovascular disease patients and suggest future research directions.
  • Conducted a scoping review based on the Arksey and O’Malley framework.
  • Systematically searched nine electronic databases for relevant studies published in English or Chinese.
  • Evaluated the methodological quality of prediction models using the PROBAST tool.
  • Included ten studies with varying adherence rates and methodological approaches.
  • Non-adherence rates to cardiac rehabilitation varied from 41% to 61.4%.
  • Significant methodological concerns were identified, including inadequate sample sizes and lack of external validation.
  • Logistic regression was the most common predictive method, with AUROC values ranging from 0.62 to 0.893.
  • Most models reviewed showed considerable heterogeneity and limited validation, hindering clinical applicability.

Study Design

Type

Scoping Review

Structured PICO

P
Population
10 studies including patients with cardiovascular disease (including AMI, post-PCI, post-cardiac events, stable angina, CABG, and heart failure) participating in or referred to cardiac rehabilitation programs. Study sample sizes ranged from 50 to 12,003 participants.
I
Intervention
Clinical prediction models (including logistic regression, decision tree, random forest, and artificial neural networks) for predicting adherence to cardiac rehabilitation programs
O
Outcome
Methodological quality (assessed via PROBAST framework) and predictive performance (AUROC, calibration, sensitivity/specificity) of prediction models for adherence to cardiac rehabilitation programs

Existing prediction models for cardiac rehabilitation adherence are at an early stage of development with high risk of bias and lack of external validation, meaning none can currently be recommended for clinical use.

Main Result

Effect estimate: AUROC 0.62 to 0.893

Absolute Event Rate: 41% vs 61.4%

Limitations

  • Included studies showed prevalent methodological concerns such as inadequate sample sizes and near-total lack of external validation.
  • Most studies relied on single-center, retrospective data with variable reporting quality.
  • Wide heterogeneity in sample sizes (50 to 12,003 participants) and predictor selection increases risk of overfitting in some models.
  • Lack of standardized adherence definitions and measurement methods across studies.
  • Limited use of machine learning models, with most studies using logistic regression and very few employing advanced algorithms.
  • inadequate sample sizes
  • near-total lack of external validation
  • reliance on single-center, retrospective data
  • variable reporting quality

Abstract

To critically evaluate the methodological quality and clinical readiness of prediction models for adherence to cardiac rehabilitation (CR) programs in patients with cardiovascular disease (CVD), and to propose a strategic roadmap for future research. This scoping review was conducted following the Arksey and O’Malley framework. Nine electronic databases were systematically searched from inception to June 2025 for studies published in English or Chinese. The methodological quality of included prediction models was critically appraised using the Prediction Model Risk of Bias Assessment Tool (PROBAST). Ten studies were included. CR non-adherence rates varied from 41% to 61.4%, measured via subjective scales, session completion rates, or wearable devices. Studies exhibited wide heterogeneity in sample sizes (50 to 12,003 participants) and predictor selection. Logistic regression was the most used predictive modeling method, followed by decision tree; random forest and artificial neural network were used in one study each. AUROC values ranged from 0.62 to 0.893. Critically, the PROBAST framework highlighted prevalent methodological concerns across all studies, including inadequate sample sizes, a near-total lack of external validation, and reliance on single-center, retrospective data. The application of prediction models for adherence to CR programs in patients with cardiovascular disease represents an emerging but methodologically heterogeneous research area. Mapping of the existing evidence indicates that most published models remain at an early stage of development, with limited validation and variable reporting quality. Consequently, no existing prediction model can be confidently recommended for clinical use. These findings highlight the need for future studies to prioritize external validation, model transparency, and adherence to established methodological guidelines to support potential translation into clinical contexts. Registered on the Open Science Framework (OSF) (https://doi.org/10.17605/OSF.IO/8JMDW).

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Xia et al. (2026) conducted a scoping review in Adult patients (≥18 years) with cardiovascular disease participating in or referred to cardiac rehabilitation programs. Prediction models for adherence to cardiac rehabilitation programs was evaluated on Adherence to cardiac rehabilitation programs, measured by session completion rates, validated scales, or wearable device tracking (AUROC 0.62 to 0.893). Prediction models for adherence to cardiac rehabilitation programs in patients with cardiovascular disease demonstrated AUROC values ranging from 0.62 to 0.893, with non-adherence rates varying from 41% to 61.4%, but existing models lack external validation and methodological rigor.

synapsesocial.com/papers/69994bef873532290d020131https://doi.org/10.1186/s12911-026-03391-7
Ask AI
Helpful
Bookmark
Share
View Full Paper