JMIR Publications

ABSTRACT

Background:

Biomedical engineering has historically operated under an assumption of gender neutrality, systematically overlooking fundamental biological and sociocultural differences between sexes. This oversight has profound implications for algorithmic fairness, model accuracy, medical device safety, and health equity across diverse populations.

Objective:

This paper comprehensively examines the integration of sex and gender variables into biomedical engineering research and practice. The objectives include: (i) identifying sources and consequences of sex-based bias in biomedical algorithms and machine learning models; (ii) evaluating the potential of digital twins and sex-specific computational frameworks for advancing precision medicine; (iii) assessing wearable sensor technologies with respect to sex-related variations in body morphology and physiology; and (iv) proposing inclusive research practices and reporting standards to promote equitable outcomes in biomedical engineering.

Methods:

This study synthesizes recent literature (2021–2026) on sex and gender dimensions in biomedical engineering, drawing upon peer-reviewed research spanning algorithm development, digital twin modeling, wearable device validation, and methodological tutorials. The analysis employs a critical evaluation framework adapted from the Sex Inclusive Research Framework (SIRF) to assess sex inclusion appropriateness across study designs.

Results:

The findings reveal that: (1) sex bias pervades machine learning models across multiple clinical domains, with 74.7% of evaluated algorithms demonstrating bias on sociodemographic factors; (2) shortcut learning mechanisms enable convolutional neural networks to infer sex from imaging data, leading to spurious correlations and unreliable diagnostic predictions; (3) sex-specific digital twins, particularly in cardiac modeling, achieve over 89% accuracy in drug response prediction, outperforming non-sex-specific models by approximately 7%; (4) wearable sensor validation studies exhibit significant gaps in reporting and representation, with only 14% reporting skin tone and median representation of older adults at 0%; and (5) the Sex Inclusive Research Framework provides a structured, traffic light–based decision tool for evaluating sex inclusion in experimental designs.

Conclusions:

The integration of sex as a biological variable and gender as a sociocultural construct into biomedical engineering is not merely an ethical imperative but a scientific necessity for achieving precise, safe, and equitable health technologies. This paper provides actionable recommendations for researchers, clinicians, and policymakers to systematically incorporate sex and gender variables across the biomedical engineering pipeline—from dataset design to clinical implementation.