A Predictive Model of Initial Hospitalization Cost in Patients with Multiple Sclerosis

Background:

Recent seismic healthcare reforms are focused on curtailing rising healthcare expenditures. In patients with multiple sclerosis(MS), limited or no data exists identifying potential modifiable targets associated with high-hospitalization cost.

Objectives:

To create a baseline predictive model of initial cost for patients in patients admitted with MS.

Methods:

A cohort of patients hospitalized with a diagnosis of MS [ICD-9-CM 340] with complete data on hospital costs was extracted using the National Inpatient Sample [2001–2014]. A split-sample 1:1 randomization approach was utilized to create a derivation(model) and validation(training) cohort. Logarithmically transformed hospital cost data was modelled using ordinary least square to identity potential drivers impacting initial hospitalization cost. Subsequently, the model was applied to the validation cohort for internal validation.

Results:

Overall, 314,251 patients [mean age: 45.18 years; 58% female] with MS were registered with the NIS[2001-2014]. Median hospitalization cost was $7,726 (IQR: $3,179-$12,273). Pertinent drivers impacting cost include advancing age (+0.3%), female gender (-2.3%), Medicaid (-3.3%), African American (+6.6%), Hispanic (+9.8%), and Asian race (+10.0%), length of hospital stay (+4.8%/extra day hospital stay), patient comorbidities [paralysis (+3.0%), obesity (+3.1%), COPD(+4.1%), CHF(+4.5%),Seizure disorder (+2.2%), coagulopathy (+8.6%), previous ischemic stroke (+20.1%), and myelopathies (+21.4%), alcohol abuse(-14.2%)], complications[DVT(+4.9%), renal failure(+6.3%)], procedure related factors[lumbar puncture (+20.5%); plasmapheresis (+46.5%); CT scans (-18.0%); intravenous immunoglobulins(-24.0%); intravenous steroids(-26.8%)]. Our model could explain a considerable proportion of variance (R²⁼ 0.51). A variation of less than 2.0% was noted in the derived R² following model training (R² = 0.50) from that of model testing. Our model demonstrated a significant strength of association (p<0.001) to predict in an independent cohort as assessed by testing model fit by plotting predicted values against observed values using the validation cohort.

Conclusions:

The identified drivers of initial hospital costs in patients with MS could potentially be used for in-hospital auditing/budgeting, providing framework for creation of data driven policies, impact reimbursement criteria, and an adjunct in the cost containment debate.