Predicting Future High-Cost Schizophrenia Patients Using High-Dimensional Administrative Data.

BACKGROUND: The burden of serious and persistent mental illness such as schizophrenia is substantial and requires health-care organizations to have adequate risk adjustment models to effectively allocate their resources to managing patients who are at the greatest risk. Currently available models underestimate health-care costs for those with mental or behavioral health conditions. OBJECTIVES: The study aimed to develop and evaluate predictive models for identification of future high-cost schizophrenia patients using advanced supervised machine learning methods. METHODS: This was a retrospective study using a payer administrative database. The study cohort consisted of 97,862 patients diagnosed with schizophrenia (ICD9 code 295.*) from January 2009 to June 2014. Training (n = 34,510) and study evaluation (n = 30,077) cohorts were derived based on 12-month observation and prediction windows (PWs). The target was average total cost/patient/month in the PW. Three models (baseline, intermediate, final) were developed to assess the value of different variable categories for cost prediction (demographics, coverage, cost, health-care utilization, antipsychotic medication usage, and clinical conditions). Scalable orthogonal regression, significant attribute selection in high dimensions method, and random forests regression were used to develop the models. The trained models were assessed in the evaluation cohort using the regression R2, patient classification accuracy (PCA), and cost accuracy (CA). The model performance was compared to the Centers for Medicare & Medicaid Services Hierarchical Condition Categories (CMS-HCC) model. RESULTS: At top 10% cost cutoff, the final model achieved 0.23 R2, 43% PCA, and 63% CA; in contrast, the CMS-HCC model achieved 0.09 R2, 27% PCA with 45% CA. The final model and the CMS-HCC model identified 33 and 22%, respectively, of total cost at the top 10% cost cutoff. CONCLUSION: Using advanced feature selection leveraging detailed health care, medication utilization features, and supervised machine learning methods improved the ability to predict and identify future high-cost patients with schizophrenia when compared with the CMS-HCC model.

Computer-aided auditing of prescription drug claims.

We describe a methodology for identifying and ranking candidate audit targets from a database of prescription drug claims. The relevant audit targets may include various entities such as prescribers, patients and pharmacies, who exhibit certain statistical behavior indicative of potential fraud and abuse over the prescription claims during a specified period of interest. Our overall approach is consistent with related work in statistical methods for detection of fraud and abuse, but has a relative emphasis on three specific aspects: first, based on the assessment of domain experts, certain focus areas are selected and data elements pertinent to the audit analysis in each focus area are identified; second, specialized statistical models are developed to characterize the normalized baseline behavior in each focus area; and third, statistical hypothesis testing is used to identify entities that diverge significantly from their expected behavior according to the relevant baseline model. The application of this overall methodology to a prescription claims database from a large health plan is considered in detail.

Space-time clusters with flexible shapes.

INTRODUCTION: Detection of space-time clusters plays an important role in epidemiology and public health. Different approaches for detecting space-time clusters have been proposed and implemented. Many of these approaches are based on the spatial scan statistic formulation. One key aspect of these cluster detection methods is the choice of cluster shape. OBJECTIVES: In this report, the effect of using flexible shapes for clusters is explored by discussing the issues that need to be considered and evaluated. METHODS: The first issue is the flexibility of the shape and its ability to model the disease cluster being studied. Another subtle and related factor is that with a more flexible shape, clusters can appear more often by chance, which will be reflected in the p value obtained through Monte Carlo hypothesis testing. Choosing more complex cluster shapes can affect the computational requirements and also constrain the cluster detection approaches that could be applied. RESULTS: The New Mexico brain cancer data set is used to illustrate the tradeoffs. The analysis of these data should not be construed as a comprehensive investigation from the public health perspective. The data set is used to illustrate and compare clusters with two different shapes, cylinder and square pyramid. The results indicate the insights that can be gained from these shapes, individually and collectively. CONCLUSION: The domain expert should choose the cluster shape, being aware of the disease being modeled and the analysis goals. For example, a flexible shape like the square pyramid can model either growth or shrinkage and movement of the disease and might provide insights on its origin. In addition, performing the analyses with more than one shape can lead to increased insights regarding the disease cluster.

Testing the consistency of the Maxwell-Stefan formulation when predicting self-diffusion in zeolites with strong adsorption sites.

The Maxwell-Stefan (MS) formulation, as applied to zeolites that contain both weak and strong adsorption sites, such as ZSM-5, is compared to dynamic Monte Carlo simulations, for the limiting case of single-component self-diffusion. This study is intended as a consistency check, and as a step towards an analytical or semi-analytical theory for self-diffusion in zeolites with multiple types of sites. In its original form, when it is assumed that ζ, the ratio of the self-exchange coefficient to the corrected diffusivity, is equal to 1, the MS formulation performs well for silicalite, the all-Si version of ZSM-5. However, when there are lattice heterogeneities or the topology of the pore network differs from that of silicalite, it is necessary to assume [Formula: see text]. Because ζ is generally occupancy dependent, the theory is unsuited as a fully predictive theory for self-diffusion in heterogeneous microporous solids, unless a theory for ζ is derived. However, since several studies have demonstrated that the MS formulation is able to predict multi-component diffusivities from single-component diffusivities for zeolites with one type of site, an extension to zeolites with multiple types of sites would be very valuable.