An Algorithm to Detect Multiple Sclerosis Relapses

Multiple sclerosis (MS) is a multi-faceted autoimmune demyelinating disorder affecting >400,000 individuals in the United States. Prevention of relapses in patients with MS is a primary goal of MS therapy.

Confirmation of a relapse in multiple sclerosis is both time and labor intensive due to the need to review medical charts, according to researchers. An algorithm to identify relapses utilizing prescription and medical claims data has been validated in a large healthcare claims database.

Researchers recently conducted a retrospective database analysis to assess the cross-validity of an algorithm to detect relapses through claims data within an integrated healthcare system in central Texas. The research was supported by Novartis Pharmaceuticals Corporation.

They reported results of the analysis during a poster session at the Academy of Managed Care Pharmacy 2012 Educational Conference in Cincinnati, Ohio, in October. The poster was titled Cross-Validity of a Multiple Sclerosis Relapse Detecting Claims-Based Algorithm in an Integrated Healthcare System.

The database included information on inpatient, outpatient, and pharmacy claims accounting for >200,000 covered lives. The analysis included data from January 2005 through December 2010. Patients were grouped and compared within groups of relapsing and nonrelapsing MS patients.

Inclusion criteria included at least 2 International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes for MS at least 30 days apart during the study. The date of the first diagnosis code was the index date. Other inclusion criteria were continuous enrollment at least 3 months prior and 12 months postindex and ≥18 years of age.

The algorithm defined relapse as (1) an inpatient MS-related claim with MS as the primary diagnosis >30 days after the index date, and (2) outpatient MS-related diagnosis code and a prescription claim for a corticosteroid (dexamethasone, methylprednisone, prednisolone, prednisone, or ACTH) ≤7 days post outpatient visit. Patients with MS not meeting both criteria were classified as nonrelapsing.

Medical charts for 87 relapsing and 100 randomly selected nonrelapsing patients were reviewed. The algorithm identified a total of 166 relapse events. The positive predictive value (PPV) was low at 44.0%, whereas the negative predictive value was high at 91.0%.

To understand the low PPV, the researchers identified the reasons for misclassifications. Those reasons included steroid use for another disease, steroid use for routine MS treatment, relapse type symptoms that were not due to MS, insufficient documents within the established timeframe, regular nursing home check-ups, and hospitalizations or emergency department visits due to other health causes (miscodes).

The researchers completed a sensitivity analysis attempting to account for monthly steroid prescribing; an algorithm to “operationalize and capture routine monthly steroid use was applied, as well as manual removal of false positive relapse events due to routine monthly treatment and nursing home visit patients.” Application of the additional algorithms yielded a PPV of 61.9%.

The authors acknowledged that the algorithm may be limited to identify instances of moderate to severe relapse. Other limitations to the analysis are lack of comprehensiveness in the electronic medical record (EMR) for each patient, and use of the ICD-9-CM code 340.00, which does not distinguish between different types of MS. Patients with primary progressive/secondary progressive MS may be using steroids as routine monthly treatment, which has an impact on false positive relapse identification.

In summary, the researchers said, “Steroid use for routine MS treatment, insufficient documentation in the EMR, and identified events due to non-MS related issues were the main reasons for the low PPV. However, after adjustment for atypical events, the kappa value demonstrated fair agreement between algorithm-identified and actual events. With proper understanding of the layout of individual health systems claims data, this algorithm can be successfully employed for identifying MS relapses.”