Healthcare AI systems can have large performance differences across demographic groups, despite the absence of comprehensive synthesis of fairness methodologies and evidence-based implementation recommendations. To review the research on fairness in healthcare AI while identifying key gaps, a PRISMA guidelines were followed, resulting in the identification of 129 records and analyzing 38 included papers using a comprehensive taxonomy, by searching IEEE Xplore, Scopus, and ACM Digital Library (2020–2025). This review indicates that publications increased sixteen-fold from 2020 reaching a high in 2023–2024. Medical imaging (47.4%) and EHR-based prediction (36.8%) predominated applications. While 57.9% assessed group fairness, only 15.8% included intersectional bias. Mitigation strategies and techniques included causal/counterfactual methods (21.1%), fairness aware training (18.4%), and data preprocessing (15.8%). Studies focused on race or ethnicity (76.3%) and gender (65.8%), with little emphasis on socioeconomic factors (21.1%). Calibration fairness (26.3%) stands out as important for clinical decisions. Research conducted on fairness in healthcare AI identifies gaps in intersectional analysis, real-world validation, and standardized frameworks. Lastly, we suggest fairness frameworks that are specific to each situation and that combine clinical usefulness with fairness standards, required intersectional evaluation, and global cooperation to make sure that AI reduces health disparities instead of making them worse.