DEVELOPMENT AND VALIDATION OF A CLASSIFICATION ALGORITHM FOR PROPHYLACTIC VERSUS ON-DEMAND FACTOR VIII THERAPY IN PATIENTS WITH HEMOPHILIA A
Author(s)
Vekeman F1, Pocoski J2, Cheng W3, Trahey A3, Sarda SP3, Valluri S4, Preblick R2, Duh MS31Analysis Group, Inc., Washington, DC, USA, 2Bayer HealthCare Pharmaceuticals, Inc., Wayne, NJ, USA, 3Analysis Group, Inc., Boston, MA, USA, 4Bayer HealthCare Pharmaceuticals, Inc., Pine Brook, NJ, USA
OBJECTIVES: To develop and validate an algorithm enabling the classification of prophylactic (PPL) vs. on-demand (OD) factor VIII (FVIII) regimens for hemophilia A (HA) patients in health insurance claims databases, thereby enabling the assessment of these treatment approaches in real-world settings. METHODS: Prescription records from two 2010-2011 US specialty pharmacy dispensing databases were used. Males aged ≥2 years old with a HA diagnosis, ≥1 prescription for FVIII, no anti-inhibitor agents or mixed PPL/OD FVIII regimens were included. Four variables common to specialty pharmacy and standard claims databases were used to develop the algorithm: age at first observed dispensing (a weight proxy); total units of FVIII dispensed (TUFVIII); and if TUFVIII data were not available, vial potency (IU/vial) and number of vials dispensed were used to calculate TUFVIII dispensed (IU/vial*number of vials dispensed). Different TUFVIII thresholds per age groups 2-12, 13-16, 17-27, and 28+ formed the algorithm. Several TUFVIII thresholds for each age group were iteratively tested over various observation lengths for patients classified as on PPL (vs. OD) regimen. Each algorithm was assessed against actual regimens prescribed based on physician notes using sensitivity, specificity, positive and negative predictive values (PPV & NPV). RESULTS: 445 patients were identified with 65% on PPL regimens. Patients receiving OD regimens were significantly older than those receiving PPL regimens (Mean age [years], OD: 24.4±17.1; PPL: 19.6±13.0; p=0.02). The best-performing algorithm was based on TUFVIII thresholds of 43,920 (age group 2-12), 104,100 (13-16), 139,200 (17-27), and 176,970 (28+), over the twelve months of observation. Sensitivity, specificity, PPV, and NPV of this algorithm were 0.81, 0.82, 0.89, and 0.71, respectively. CONCLUSIONS: The best-performing algorithm showed promising performance validity with PPV>0.89 for ascertainment of PPL FVIII regimens in HA patients. Further refinement of the classification algorithm on larger populations is needed.
Conference/Value in Health Info
2012-06, ISPOR 2012, Washington, D.C., USA
Value in Health, Vol. 15, No. 4 (June 2012)
Code
PSY64
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
Systemic Disorders/Conditions