OBJECTIVES: Heterogeneity of treatment effect (HTE) is the variation in how individuals respond to a treatment, which justifies the need for patient-centered care. We present a joint modeling method developed to help patients with type 2 diabetes mellitus (T2DM) initiating injectable therapies choose between basal insulin and premixed insulin.

METHODS: The outcome is HbA1c < 7% at 6 months after insulin initiation. Our method includes two base logistic regression models, one predicting the outcome if basal insulin were chosen and one predicting the outcome of premixed insulin, both using baseline characteristics as predictors. For joint modeling, relative risk is calculated as the quotient of the predicted probabilities of the two base models, which are further categorized into four recommendation groups: strongly favors premix, somewhat favors premix, somewhat favors basal, or strongly favors basal. The regimen can therefore be recommended based on the group. A decision tree model was also fit to classify the patients into recommendation groups using baseline characteristics, which provides a simpler and more interpretable way of model application.

RESULTS: A total of 1287 patients were included: 821 initiated with basal insulin and 466 with premixed insulin. Both base models had similar discrimination (AUC: 0.65) and good calibration (Hosmer-Lemeshow test P > 0.05). The decision tree model showed an accuracy of 0.69. The results showed that patients with baseline HbA1c < 9.7% can achieve a better outcome with basal insulin, especially those older than 57 years. Patients with baseline HbA1c ≥ 9.7% can achieve a better outcome with premixed insulin, especially those with HbA1c ≥ 12% or those with hypertension but without coronary artery disease.

CONCLUSIONS: Our method of building a joint model on top of base models can help with the choice of treatment to facilitate patient-centered care.