Endocrine Abstracts

Introduction: Lowering low-density lipoprotein cholesterol (LDL-C) via lifestyle change or pharmacologic therapy can reduce the cardiovascular risk (CV) in patients with type 2 diabetes (T2D). However, the response to lipid-lowering interventions is not uniform and a significant proportion of T2D patients do not reach the recommended LDL-C target. Elucidating the factors associated with lipid goal attainment represents an unmet clinical need.

Material and method: This is a retrospective, cross-sectional study based on data from the Annals of the Diabetes Medical Association (AMD) database, which includes the electronic medical records of 1.186.247 patients treated in various Italian diabetes clinics between 2005 and 2019. In this study, we used Rulex®, a type of Logic Learning Machine (LLM) approach to extract and classify the most relevant variables that could predict the achievement of an LDL-C value below 2.60 mmol/l within two years (defined as T2Y) from the time of lipid therapy prescription (T0).

Results: Overall, 11.252 patients (45.46% female) with T2D and dyslipidemia were evaluated. Stratification according to CV risk showed that, at T0, 97.7% of patients were at very high risk and 2.3% at high risk. At T0, 95.79% of patients were prescribed statins, and only 9.16% ezetimibe. The treatment goal at T2Y was achieved in 61.4% of cases. The LLM model was endowed with a precision of 69% and an accuracy of 67% (AUC-ROC: 0.775, P<0.001) to predict the attainment of LDL-C target. LDL-C values at T0 and six months later (T6M) exhibited the highest significance in this predictive model. Other predictors identified by LLM were: uninterrupted treatment between T0 and T2Y, higher HDL-C levels, younger age, anti-hypertensive drug administration, lower glycosylated hemoglobin levels, a higher Q-score, a lower BMI, lower micro-albuminuria levels and male gender. For each of the initial LDL-C ranges analyzed, the LLM also indicated the minimum reduction that should be achieved by T6M to increase the probability of reaching the therapeutic goal within T2Y, thus offering a new helpful tool to guide the physician’s therapeutic attitude.

Conclusion: This is the first study describing the application of a LLM approach on real-world data to identify variables involved in lipid goal attainment. Our results suggest that a follow-up visit should be recommended within six months after the start of lipid therapy in all T2D patients. The present study also underscores the importance of initiatives to establish a more aggressive lipid management strategy for specific patient subgroups.