An Item Response Theory Model with Bounded Integer Subcomponents to Describe the Mayo Clinic Subscores in Patients with Ulcerative Colitis
Objectives: The Mayo Clinical Score (MCS) is widely used in clinical trials to describe the clinical status of patients with ulcerative colitis (UC). It comprises four subscores, each scored from 0 to 3: rectal bleeding (RB), stool frequency (SF), physician’s global assessment (PGA), and endoscopy (ENDO) subscores. Clinical “response” and “remission” rely on all scores being available. ENDO is typically measured at beginning and end of each study, while RB, SF and PGA are available more frequently. Item response theory (IRT) models [1] allow the shared information to be used for predictions of all subscores at each observation time; it could leverage information of RB, SF and PGA on clinical status of patients for prediction of endoscopy subscore. Therefore, an IRT model was developed to predict longitudinal MCS subscores and remission status at end of induction and maintenance in UC patients.
Methods: Four etrolizumab UC phase 3 studies were included [2]. Placebo data was used in the development of base model and active treatment data was added later for the purpose of increasing numbers of observed remissions to better evaluate the accuracy of the model. For each subscore, a bounded integer (BI) model [3] was used, consisting of a BASE and standard deviation (SD) parameter. The placebo effect was described by a monoexponential function acting on all subscores similarly. Interindividual variability (IIV) was estimated per parameter. Final model was externally evaluated by visual predictive check (VPC) using placebo arms from five UC studies from various pharmaceutical companies [4].
Results: BI models were developed with IIV on the BASE parameter per subscore and IIV for SD shared across scores. Subscore-specific parameters relied on baseline observations, while parameters describing post-baseline trends and variability were informed by all subscores equally. The model reliably predicted the analysis data. Remission at the end of induction was correctly predicted 183 out of 242 times (76%) and non-remission was predicted correctly 986 out of 1071 times (92%). End of maintenance remission was correctly predicted 65 out of 76 times (86%) and non-remission was predicted correctly 51 out of 75 times (68%). The external subscore data was predicted well for RB, PGA, and ENDO, but was underpredicted for SF due to lower mean SF baseline in external data vs analysis data. VPC for remission indicated adequate performance for the external data, and 95% confidence interval of simulated remission encompassed the observed remission for each study.
Conclusions: The IRT model adequately described the subscores of MCS and reliably predicted the remission status for both analysis and external datasets at the end of induction and maintenance. Other drug effects can be inserted as deemed appropriate, allowing for model informed drug development across a range of UC programs.
Citations:
[1] Ueckert, Modeling Composite Assessment Data Using Item Response Theory. CPT:PSP 2018; 7(4):205-218
[2] ClinicalTrials.gov (NCT02163759, NCT02171429, NCT02100696, NCT02165215)
[3] Wellhagen, A Bounded Integer Model for Rating and Composite Scale Data. AAPS 2019, 21(4):74
[4] ClinicalTrials.gov (NCT385740, NCT408630, NCT410410, NCT787200, NCT853100)
Author(s)
- Anita Moein, Genentech (Presenting Author)
- Jurgen Langenhorst , Pharmetheus (CoAuthor)
- Sami Ullah , Pharmetheus (CoAuthor)
- Mats Magnusson , Pharmetheus (CoAuthor)
- Jin Jin, Genentech (CoAuthor)
- Nastya Kassir , Genentech (CoAuthor)