Sex and gender classification in clinical decision-support tools: a cross-sectional review of tools on MDCalc

BACKGROUND: As algorithm-based decision support tools become increasingly integrated into clinical workflows, particularly with the popularization of artificial intelligence, the conceptualization and operationalization of demographic variables such as sex and gender have implications for how these variables are measured and interpreted. Misalignment between variable labels, response categories, and underlying data sources may introduce measurement error or ambiguity in algorithm inputs, potentially affecting clinical interpretation and downstream decision-making. This concern reflects issues of variable specification rather than terminology alone. The objective of this study is to evaluate how sex and gender are labeled, operationalized, and aligned with underlying source data in clinical decision-support tools. METHODS: In May 2025, we conducted a cross-sectional review of clinical tools available on MDCalc, a widely-used online repository of clinical decision-support tools derived from biomedical research and clinical guidelines. We assessed tools with a demographic input labeled as sex, gender, or both. Main measures were the presence and operationalization of sex or gender variables within tool interfaces and corresponding primary references. Variable labeling (sex or gender), response categories (e.g., male/female or man/woman), reported data collection methods in primary references, and concordance between tool interfaces and source literature. Tools were also qualitatively categorized by potential clinical impact (high, medium, low). RESULTS: Among 821 tools screened, 84 (10.2%) included a sex or gender variable. Of these, 69 (82.1%) labeled the variable as sex and 15 (17.9%) as gender. Binary response categories were used in 83 tools (98.8%). Sex-gender conflation occurred in 16 tools (19.0%), reflecting inconsistencies between variable labels and response categories used as inputs. Primary references were available for 76 tools (90.5%); among those that explicitly reported sex or gender collection, 18 (25.7%) demonstrated sex-gender conflation. Most tools were classified as having medium (51.2%) or high (33.3%) potential clinical impact. CONCLUSION: Sex and gender are inconsistently labeled and operationalized in clinical decision-support tools. One in five tools demonstrated misalignment between variable labels, response categories, and underlying source data. Greater conceptual clarity and transparency of sex and gender variables may reduce ambiguity and strengthen the validity, interpretability, and equity of algorithm-based clinical decision tools.