Assessing the effects of CFTR modulator therapies on weight change & T2D/CFRD development in a population of adult cystic fibrosis patients using real-world EHR data

INTRODUCTION

Since the introduction of cystic fibrosis transmembrane conductance regulator (CFTR) modulator therapies (MTs), the life expectancy of patients with cystic fibrosis (PwCF) has increased. Historically, PwCF have been prescribed high-calorie diets to help maintain healthy body weights. However, CFTR MTs necessitate refinement of traditional CF diets and treatments. Recently, an increase in the number of PwCF with type 2 diabetes (T2D) and cystic fibrosis-related diabetes (CFRD) has been noted. By leveraging large, multi-site EHR data, this study seeks to determine the relationship between CFTR MT prescription, patient weight change, and CFRD prevalence/development.

METHODS

The data for this study were obtained from the Greater Plains Collaborative’s (GPC’s) GROUSE environment. The data were retrieved using Structured Query Language (SQL) and analyzed using Python. Adult PwCF aged 18-65 were identified utilizing ICD codes, requiring ≥ 2 diagnoses. Weights were queried to build a record of patient weight trends. Visits with T2D/CFRD diagnoses were extracted utilizing an established computable phenotype method for PCORnet. Using the RxNorm API, ingredient and drug component Rx Concept Unique Identifiers (RxCUIs) were retrieved and used to identify patients prescribed a medication containing any of four CFTR MT ingredients and their prescription dates. Utilizing XGBoost with SHapely Addative exPLanations (SHAP), features associated with weight gain were extracted; features associated with T2D/CFRD prevalence were also identified. Weight gain thresholds were defined as indicator outcome variables; 15% weight gain attainment was used as the main model. Phecodes and electronic Clinical Quality Measure (eCQM) categories were used to translate diagnoses and prescriptions into covariates.

RESULTS

In total, 6,109 PwCF were identified with valid weight measurement data; after adding covariates and limiting the data to 2019-2023, 3,841 PwCF remained in the cohort. The mean weight difference of these patients was 3.9781 lbs. Outcome variables were based on percentage of weight gained from the initial weight measurement; mean weight difference was < 1%. 1,761 PwCF (45.8474%) were prescribed a CFTR MT. The prevalence of T2D/CFRD was 38.3234% (n = 1,472). 471 PwCF (12.2624%) attained at least 15% weight gain (CFTR MT n = 291/471). Trikafta and Kalydeco are both significant predictors for 15% weight gain; having a location of Missouri (SCHEMA_PCORNET_MU) was also a significant predictor of 15% weight gain attainment.

Symdeko provided the highest rate of 15% weight gain attainment with 20.6564% (MWD 8.5367 lbs) of recipients meeting the threshold. Recipience of a CFTR MT was utilized by the XGBoost models as a significant predictor variable both for 15% weight gain attainment and T2D/CFRD prevalence; Trikafta is positively correlated with 15% weight gain, while Kalydeco is largely negatively correlated. Model accuracy for 15% weight gain attainment was 86.2% (mean 5-fold cross validation 86.7%); accuracy for T2D/CFRD status was 74.9%.

CONCLUSION

While recipience of a Trikafta is still a strong positive predictor of 15% weight gain attainment, the current XGBoost models only reflect up to a 1.0 unit increase in SHAP values; Kalydeco is similar but correlates negatively. Symdeko was not significant, but had the highest MWD and 15% weight gain attainment n. Other Phecode and eCQM covariate categories are equally strong predictors of weight gain thresholds across all models tested. Geographic location, indicated by hospital visited, was also significant (specifically for The MU and KUMC).