Objective This study aimed to develop a risk stratification strategy for evaluating the relationship between complications of diabetes and the risk of diabetic-related hospitalization to accurately classify diabetes severity. Methods The study used administrative health records for 40,624 individuals with diabetes aged ≥ 18 years in Western Australian. The adapted Diabetes Complication Severity Index (DCSI), socio-demographic and clinical characteristics were used in random effects negative binomial and threshold effect models to determine the optimal stratification strategy for diabetes severity based on the homogeneity of the risk of hospitalization in response to variation of the DCSI. Results The optimal stratification of people with diabetes was specified by four sub-populations. The first sub-population was no complications with an inverse association with the risk of hospitalizations (coefficient − 0.247, SE 0.03). Further three sub-populations with DCSI at one (coefficient 0.289, SE 0.01), two (coefficient 0.339, SE 0.01) and three or more (coefficient 0.381, SE 0.01) were used to accurately describe the impact of DCSI on the risk of hospitalization. Conclusion A stratification into four subpopulations based on the homogeneous impact of diabetes DCSI on the risk of hospitalization may be more suitable for evaluating health care interventions and planning health care provision.