Pre-diagnosis Predictors of High Health Care Cost Among Patients with Myasthenia Gravis Using a Combined Machine-learning and Regression Approach
Maryia Zhdanava1, Jacqueline Pesa2, Porpong Boonmak1, Samuel Schwartzbein1, Qian Cai3, Dominic Pilon1, Zia Choudhry2, Marie-Hélène Lafeuille1, Patrick Lefebvre1, Nizar Souayah4
1Analysis Group, Inc., 2Janssen Scientific Affairs, LLC, 3Janssen Global Services, 4Department of Neurology & Neurosciences, Rutgers-New Jersey Medical School
Objective:
To identify predictors of high costs among patients with generalized myasthenia gravis (gMG)
Background:
MG is an autoantibody disorder with high patient and health system burden, reflecting unmet needs. This study used random forest (RF) and Poisson regression methods to identify predictors of high costs in patients newly diagnosed with gMG.
Design/Methods:
Adults with gMG were identified from the IQVIA PharMetrics® Plus database (01/2017-12/2021) using MG diagnoses and physician specialty. The first MG diagnosis was the index date. Demographic and clinical characteristics were described during the 12-month baseline period, and monthly all-cause healthcare costs (2021USD) were estimated post-index. Those with costs in the top 15th percentile (≥$9,405) were defined as having high costs. An RF model using baseline characteristics was built to identify the top 50 predictors of high costs based on Shapley values; a modified Poisson regression was used to determine the magnitude and direction of each predictor.
Results:
2,739 patients with gMG were selected (mean age: 56.2 years; female: 50.6%). The RF model had good predictive power. The most important predictors were inpatient admissions, MG with exacerbation on the index date, number of outpatient visits, corticosteroid use, diagnosing physician specialty, and comorbidities (diabetes, fluid and electrolyte disorders, DSM-5). Adjusting for all top 50 predictors in the regression model, MG diagnosed by a cardiologist increased the risk of high costs by 177% compared to those diagnosed by a neurologist; index MG exacerbation increased the risk by 78%; metastatic cancer by 74%; weakness or fatigue by 54%; other autoimmune disorder by 42%; diabetes by 36%; every 10 claims of corticosteroids by 33%; inpatient admission by 27% (all p<0.05).
Conclusions:
This analysis shows the complexity of patients with gMG and the high degree of multimorbidity. By targeting amenable characteristics, healthcare providers can reduce the clinical and economic burden of gMG.