Patients with “High Expense, Low Prevalence” (HELP) diseases such as rheumatoid arthritis, multiple sclerosis, and inflammatory bowel disease are at risk for exacerbations (flares), disease complications and treatment related side effects that can result in preventable mortality or major morbidity. Millions of dollars are spent yearly on medications for HELP diseases. Although some of these patients require lifelong, expensive, and potentially harmful medications to prevent serious complications, many others are at lower risk and are best treated with less expensive and less harmful medications, or by using “as-needed” therapy as flares occur.
Such complexities in patient circumstances and decision-making are very common in medical practice and therefore, risk-stratifying patients with HELP diseases for a disease exacerbation offers great promise to significantly improve both the quality and efficiency of patient care. Tools that more accurately predict the course of disease and offer advice on appropriate treatment could substantially improve the decision-making process. Developing tools and decision support systems to guide clinicians in personalizing medical decision-making for patients with HELP diseases is of particular importance in our modern, data-intensive and computing-intensive world, far more data is collected than can be fully evaluated by even expert healthcare providers. However, to implement this “targeted” or “tailored” prevention approach to risk stratifying individuals for disease exacerbation and treatment, a clinician must know both the individual’s baseline risk of disease complications and the probability that the individual would benefit (or suffer harm) from therapy. Having risk stratification tools developed and validated is an important first step towards realizing efficient patient-centered care for HELP diseases.
My research goals are to develop “targeted-prevention” prediction tools and decision support systems to facilitate the delivery of timely and cost-effective therapy for HELP with a focus on Inflammatory Bowel Disease (IBD) as a model condition for HELP diseases.
Predictive modeling includes both traditional “regression” models and novel “machine learning” approaches. Regression models are usually used to identify associations and causal pathways by testing specific hypotheses, an approach best suited to the examination of a limited number of variables with high data quality. In contrast, machine learning models can identify predictive patterns under the sole hypothesis that some predictive pattern exists, a technique intended to make sense of information on a large number of variables even when the source data are quite “dirty”. These models have not been widely applied in medicine, but are commonly used in other fields. For example, economic and marketing strategists use machine learning based approaches to analyzing large amounts of data, detecting patterns, and taking action on these patterns to understand consumer-spending behavior. The banking industry extends very specific credit card offers based on a person’s spending history, online advertisements after mining a person’s browsing history, and some companies recommend books based on previous purchases. It remains unclear whether traditional regression or machine learning models are more effective in evaluating patient risk and estimating response to therapy and my research focus is directed at providing a model for developing such tools for high-expenditure, low prevalence (HELP) conditions, using IBD as a model disease.