Risk prediction model for malnutrition in older adults: a systematic review

BACKGROUND: Global population aging is a major public health trend. The incidence of malnutrition among older adults is also continuously rising. The occurrence of malnutrition among older adults will affect their quality of life and disease prognosis. Although various risk prediction models have been developed to identify the risk of malnutrition in older adults, a comprehensive systematic review of these models is currently lacking. OBJECTIVE: The aim is to systematically evaluate the risk prediction models for malnutrition in older adults that have been published both domestically and internationally, assess their predictive performance, verification status, and methodological quality, thereby providing a basis for selecting appropriate models in clinical practice. METHODS: A systematic search was conducted in Chinese and English databases. These included PubMed, Web of Science, Cochrane Library, Embase, CINAHL, CNKI, Wanfang Database, VIP Database, and SinoMed. Relevant literature on predictive models for malnutrition risk in older adults was included. The search period covered the establishment of the databases to July 1, 2025. Only Chinese and English publications were considered. Two researchers independently screened the literature according to inclusion and exclusion criteria. They evaluated the quality of the included literature and extracted the data. The PROBAST risk of bias and applicability tool was used to assess the risk of bias and applicability of the included research models. Data extraction included the first author, publication year, country, study subjects, study type, predictive factors, model construction methods, and predictive performance. RESULT: A total of 27 articles were included, comprising 27 models. The sample size ranged from 115 to 3387 cases. Regarding model construction methods, Logistic regression models and Machine learning approaches were used. For model presentation, nomograms and regression equations were primarily used. The number of final included predictive factors ranged from 3 to 17, with common factors including age, BMI, hemoglobin level, serum albumin level, depression status, and daily activity ability (ADL). In terms of performance, the area under the receiver operating characteristic curve (AUC) ranged from 0.687 to 0.984. Validation was conducted internally in 16 studies, while 8 studies conducted external validation. Overall, the applicability of the included models was good; however, all studies demonstrated a high risk of bias. CONCLUSION: The included models have relatively good predictive performance. However, the overall study has a high risk of bias. In the future, it may be considered to adopt visual model presentation methods to build models with low bias risk, superior predictive performance, and strong clinical applicability.