Abstract:

Accurate reference intervals are crucial for disease diagnosis and health assessment in clinical laboratories. The traditional direct method is limited by its high cost and complex procedures，and indirect method leverages real-world data to establish reference intervals cost-effectively，offering greater potential for widespread adoption. Methodologically，the field has progressed from traditional graphical techniques reliant on manual interpretation （Hoffman and Bhattacharya） to robust statistical models （KOSMIC and refineR），and further to the integration of machine learning algorithms （mixture density networks and deep learning），enabling intelligent and automated estimation of the healthy population parameter distribution from mixed datasets. In application，the capacity of indirect methods to process vast datasets has successfully catalyzed a shift from generic reference interval towards precise stratification，effectively revealing and quantifying the specific influences of factors like age，sex and geography on physiological markers. Latest methods now transcend fixed partitions，utilizing techniques such as generalized additive models for location，scale and shape and quantile regression to construct dynamic reference curves that vary continuously with age. Furthermore，they facilitate the development of personalized reference interval by incorporating within-subject biological variation and longitudinal data. Despite persistent challenges related to data quality，model selection and clinical validation，the synergy between indirect methods and big data is unequivocally propelling laboratory medicine toward a more precise and personalized development.

Key words: Reference interval, Indirect method, Big datum