Statistical Areas of Interest:

• Statistical methods for precision treatment and precision prevention
• Design of clinical trials for both therapeutic and early detection purposes
• Methods for analyzing data from real-world studies, such as electronic medical records, administrative data
• Public health surveillance

Application Areas:

• Cancer (screening, surveillance, treatment, and survivorship)
• Diabetes
• Obesity

Projects

Improve the Design and Analysis of Randomized Screening Trial in a New Era of Cancer Early Detection

Novel blood tests offer a noninvasive way to detect cancer early. However, despite the simplicity of these tests, follow-up actions from the test results may be costly or harmful. The quick introduction of these tests requires reevaluating how novel early-detection biomarkers are assessed. 

Large screening trials are often needed when evaluating new cancer screening tests, where participants randomly receive new screening tests or usual care. These trials take a long time to assess if the new test lowers the chances of dying from cancer. Researchers have proposed to look at short-term results, but sometimes, big drops in these results may not mean significant reductions in cancer deaths. Better approaches to studying these tests, considering the benefits and potential harms they might cause, are urgently needed.  

This study aims to improve the design and analysis of randomized screening trials so that researchers can more quickly learn if the new tests can reduce the probability of cancer deaths and if they are harmful. Improving how the randomized screening trials are designed and analyzing the study results can speed up the evaluation process of new screening tests. This could lead to faster sharing of effective screening methods to the public and provide balanced information about their benefits and harms for shared and informed decision making. It will ultimately help reduce the disease burden and improve patients' outcomes.

Read more

Statistical Methods for Design and Analysis of Cancer Immunotherapy Trials 

In the US, about 1.7 million new cancer cases are expected to be diagnosed, and about 609,640 Americans are expected to die of cancer in 2018. This creates significant need for effective cancer medication. Cancer immunotherapy has led to a paradigm shift in oncology, where therapeutic agents are used to target immune cells rather than cancer cells. However, many challenges remain to optimize the use of cancer immunotherapies due to this ‘indirect’ effect. First, the efficacy-toxicity tradeoffs under the new immunotherapies are more complex, where the generated immune-related adverse events could be severe of even life-threatening. Given that every person’s immune system is slightly different, yielding a unique degree of sensitivity to cancer, determining who will benefit from immunotherapy is particularly challenging. Second, to date there is no criterion for providing guidelines on how to use surrogate endpoints in immunotherapy trials.

The proposed research will address some of these urgent needs, motivated by trial designs and data from SWOG, and in particular the Lung-MAP study, consisting of a biomarker platform for evaluating genomically-matched therapies and immunotherapies for the non-matched patients, to address the following aims: (1) to develop biomarker-assisted treatment rules by incorporating efficacy-toxicity profiles, and (2) to evaluate the clinical impact of different surrogate endpoints and design a clinical trial based on surrogate endpoints that optimizes the population impact. The proposed research is motivated from the practical perspective to help improve cancer care, and the developed and implemented strategies can benefit future cancer immunotherapy trials and treatment plans. 

Read more

Statistical Methods for Healthcare in Complex Patients with Diabetes

Diabetes affects 8.3% of the US population, and lead to costly adverse healthcare outcomes. Unfortunately, there may be a quality gap in the care of complex diabetes patients, that is, older patients (age>65 years) and those with comorbid conditions. Current practices, relying primarily on the presence of several factors, are not effective in capturing the risk of poor prognosis, i.e., multiple hospitalization and/or emergency department visits, and death. Hence, little evidence exists so far to help prioritize care for these patients. The diabetes guidelines recognize that tight control of glycosylated hemoglobin (A1c) may not be appropriate for complex patients, and recommend individualizations in tight A1c control. However, neither the outcomes of tight A1c control, nor the effects of the typical treatment regimens used to achieve tight A1c control can be evaluated in clinical trials, with minimal, if any, enrollment of complex diabetes patients due to either their restrictive inclusion criteria or lack of encouragement of the patient and/or clinical investigator to consider the RCT. In order to deliver more effective, efficient and accountable health cares, it is important to help clinicians to examine the relationship between patient complexity and patients' A1c control level, and to modify guideline appropriately with an evidence base. The proposed research will analyze a cohort of 8,304 Medicare beneficiaries with diabetes who were cared for by one of the country's 10 largest physician group practices, the University of Wisconsin Medical Foundation during 2003-2011 to address the following aims: (1) to conduct risk prediction incorporating longitudinal outcomes, (2) to inform guidelines for complex diabetes patients, and (3) to create a patient-centered surveillance tool for detecting short-term negative outcomes. Our analytic approach involves the use of state- of-the-art statistics and machine learning methods to take advantage of the large electronic health records data. The proposed methods and results will help clinicians to identify and quantify risks of tight A1c control in complex diabetes patients and potentially lead to improved patient experiences, and reduce medical expenditures from excess adverse events.

Read more

Childhood Obesity Surveillance Using Electronic Health Records Data

Childhood obesity is associated with high financial cost and increased morbidity and mortality in adulthood. Health care spending due to obesity is estimated to be as high as $210 billion annually, or 21% of total national spending. A childhood obesity surveillance system will discern disparities, detect aberrant signals, design targeted interventions, and track change over time. In addition to national public health surveillance systems, local data are increasingly necessary to reflect regional trends. Current systems in Wisconsin are subject to (1) bias estimation of obesity rates, (2) limited data in school-aged children ages 5-10 years, (3) insufficient power to estimate disparities in local subgroups, and (4) limited ability to track local changes over time. Therefore, the impact of investments and collective efforts in Wisconsin childhood obesity prevention has been difficult to measure. This, in turn, demands approaches that overcome and transcend the posited limitations. Since large electronic health records (EHRs) systems have been developed to routinely collect information over time, there has been a paradigm shift and accelerated improvements in the use of EHR data for the purpose of population health promotion and tracking of progress. The University of Wisconsin Population Health Information Exchange (PHINEX) database contains de- identified EHR data from a multicenter healthcare system located primarily in south central Wisconsin. In partnership with The Wisconsin Obesity Prevention Initiative, we aim to create an EHR model for a childhood obesity prevention surveillance system. We propose to develop innovative statistical machine learning methods using the platform of PHINEX for childhood obesity prevention purposes. This enhance and extend surveillance activities of childhood obesity conditions, and increase the effectiveness of evidence-based interventions aimed at changing policies, systems, and environments associated with childhood obesity. We plan to address the following aims: (1) robust estimation of the spatiotemporal prevalence at area level, (2) hot spot detection on areas with aberrant obesity incidence, and (3) prediction for childhood weight gain phenotypes. The developed and implemented surveillance system can benefit future planning, legislation and implementation in other states and/or surveillance of other acute and chronic health conditions.

Read more