Due to the higher birth rate of preterm infants and improvements in their management, metabolic bone disease of prematurity (MBDP) has a high incidence and is attracting attention. However, clear indicators for the early diagnosis of MBDP are lacking. We aimed to explore simple and feasible early warning indicators for diagnosing MBDP. Our study collected case data of premature infants from two medical centers in Chongqing from January 2020 to February 2022. According to the inclusion and exclusion criteria, data from 136 cases were collected. The correlation between 14 variables in each case and the occurrence of MBDP was analyzed. According to area under the receiver operating characteristic curve (AUROC) analysis, the best cutoff value for each variable was determined. Potential predictors were selected, and Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis was used to establish the association of two models with MBDP, whose results were used to develop a diagnostic nomogram. Furthermore, a model decision curve was analyzed. Four predictors were selected from 14 clinical variables by LASSO regression, and Model I was established, including the following characteristics:​ height (>36 cm), head circumference (≤29.49 cm), total serum calcium (Ca) (>2.13 mmol/L), and alkaline phosphatase (ALP) (>344 U/L) levels. A single predictor, the ALP level (>344 U/L), was used to establish Model II. The AUROC values of the two models were 0.959 for Model I and 0.929 for Model II. In conclusion, in this study, two diagnostic models of MBDP were developed using four combinations of predictors and ALP as a single predictor. Both models showed good sensitivity and specificity for the early diagnosis of metabolic bone disease (MBD), and an ALP level of 344 U/L was defined as a simple and effective diagnostic threshold. In future studies, using larger samples, diagnostic threshold values of ALP for premature infants of different ages should be established, and internal and external validations are needed to improve the adaptability of the current model.