(P 357) DEVELOPMENT OF AN ALGORITHM TO IDENTIFY PATIENTS (PT) WITH DESMOID TUMORS (DT) IN A UNITED STATES (US) CLAIMS DATASET

Objective: International Classification of Disease (ICD) codes 238.1 (ICD-9) and D48.1 (ICD-10), which represent neoplasms of uncertain behavior of connective and other soft tissue, were generally used for DT before Oct 2023 with limited precision. We aimed to construct an algorithm to more accurately identify pt with DT in a US claims dataset.

Methods: Adult pt in the University of Utah Health network who may have DT were identified via ICD codes 238.1/D48.1 in claims data from Jan 1, 2011 to Jul 31, 2023. Clinical notes or pathology reports were reviewed to confirm a past diagnosis (dx) of DT. Pt with confirmed DT dx and >2 DT-related encounters served as cases; pt without confirmed DT dx served as controls. Cases and controls were randomly assigned to the model construction or validation dataset on a 1:1 split. Stepwise logistic regression was performed using the construction dataset. Demographics, ICD codes and Current Procedural Terminology (CPT) codes were independent variables; case/control status was the dependent variable. The optimal threshold to identify pt with DT was determined using the receiver operating characteristic curve. Model performance was assessed in the validation dataset.

Results: A total of 1,593 pt were identified based on ICD codes 238.1/D48.1, of which 146 (9%) were eligible to serve as cases for algorithm development and 1,314 (91%) pt were eligible to serve as controls. The final algorithm consisted of 12 variables (age, sex, ICD codes for neoplasms and CPT codes for biopsy and imaging procedures [Table 1]). Among pt with a confirmed DT dx, the algorithm correctly identified 83.6% (sensitivity) [84.0% for specificity]. Individuals identified as a patient with DT by the algorithm have a 36.8% probability of a confirmed DT dx verified by chart review (positive predictive value, PPV) [97.9% for negative predictive value, NPV].

Conclusion: The predictive algorithm developed more accurately identified pt with DT compared to using the non-specific DT ICD-10 codes alone. As a result, this may help to further improve the precision of identifying pt with DT from future longitudinal analyses of claims data before Oct 2023.

Table 1. Model Variables of Algorithm to Identify Pt with DT