Treatment Pathways Leading to Biologic Therapies for Ulcerative Colitis and Crohn's Disease in the United States.

OBJECTIVES: Biologic therapies have been available for inflammatory bowel disease for >20 years, but patient outcomes have not changed appreciably over this time period. To better understand medication utilization for this disease, we evaluated a novel technique for visualizing treatment pathways, including initial treatment, switching, and combination therapies. METHODS: This retrospective, observational study used administrative claims data from the Truven Health MarketScan Commercial and Medicare Database. Adult patients with ≥2 consecutive health claims and newly diagnosed with ulcerative colitis (UC) or Crohn's disease (CD) were evaluated. Treatment pathways were visualized using Sankey diagrams representing the number of patients receiving treatment and duration of each treatment. RESULTS: In all, 28,119 patients with UC and 16,260 patients with CD were identified. The most common initial treatment for UC was 5-aminosalicylic acid monotherapy (61% of the patients), followed by corticosteroid monotherapy (25%); <1% of patients were initially treated with biologics. The most common initial treatment for CD was corticosteroid monotherapy (42%), followed by 5-aminosalicylic acid monotherapy (35%); <5% of the patients were initially treated with biologics. Significantly fewer patients followed biologic vs nonbiologic treatment pathways (UC: 6% vs 94%, CD: 19% vs 81%, both P < 0.05). DISCUSSION: Significantly fewer patients with inflammatory bowel disease followed treatment pathways that included biologic therapies compared with nonbiologic therapies, and very few patients were ever initiated on biologic therapy. Although we have made significant progress in treatment, our most effective medications are only being used in a small proportion of patients, suggesting barriers prevent optimized patient management.

Megavariate data analysis of mass spectrometric proteomics data using latent variable projection method.

There are many data mining techniques for processing and general learning of multivariate data. However, we believe the wavelet transformation and latent variable projection method are particularly useful for spectroscopic and chromatographic data. Projection based methods are designed to handle hugely multivariate nature of such data effectively. For the actual analysis of the data we have used latent variable projection methods such as principal component analysis (PCA) and partial least squares projection to latent structures based discriminant analysis (PLS-DA) to analyze the raw data presented to the participants of the First Duke Proteomics Data Mining Conference. PCA was used to solve problem #1 (clustering problem) and the PLS-DA was used to solve problem #2 (classification problem). The idea of internal and external cross-validation was used to validate the model obtained from the classification analysis. The simple two-component PLS-DA model obtained from the analysis performed well. The model has completely separated the two groups from all the data. The same model applied on two-thirds of the data showed good performance by external validation with independent test set of remaining 13 specimens obtained by setting aside the spectra of every third specimen (accuracy of 85%).