Center for Predictive Medicine

Center for Predictive Medicine

How do we turn massive amounts of raw data into therapeutic treatment plans? This question guides the work of DCRI’s Center for Predictive Medicine.

Changes in federal regulations, financial incentives, patient engagement, and the availability of health information are driving revolutionary advances to the way that healthcare is delivered in the United States. The Center for Predictive Medicine can help healthcare systems to lead those advances through deeper understanding of their patient populations and optimal treatment regimens for the individuals within them. We do this by providing statistical, clinical, informatics, and project management expertise.

Using Predictive Analytics to Assess Risk of Future Disease

The availability of electronic health records (EHR), clinical trial data, claims data and other shared data has opened exciting new opportunities in predictive medicine. Applying predictive analytics to the data collected across our multinational clinical trials, national patient registries and outcomes research holds the key to developing novel approaches to quantify and communicate risk.

Lisa Wruck

The DCRI is a home to more than 100 statisticians and data scientists representing several functional and specialized groups. What we do:


Define the research question.

  • Collaborate with clinical partners to frame the research question and identify the population of interest.

Identify data sources.

  • Match appropriate data sources to the research question and population.
  • Create and validate analytic datasets.

Analyze the data.

  • Apply statistical methods and techniques appropriate to the research question.
  • Implement methodological research techniques related to meaningful use of EHR data, high-dimensional data, and risk prediction.

Disseminate results.

  • Collaborate with clinical partners to accurately interpret and present results.
  • Facilitate integration of risk-prediction models into EHR.
group photo

A Strong Tradition of Partnership

In addition to its experts’ own interests, the Center for Predictive Medicine partners with researchers at Duke and beyond the university. The center leads analytics for the Southeastern Diabetes Initiative.

Ultimately, we will apply predictive medicine across many aspects of clinical research and healthcare, streamlining the steps from discovery of novel risk algorithms to implementation of those algorithms in clinical practice. This transition from reactive to preventive medicine will lead to healthier patients and more efficient healthcare delivery. In an era of EHRs, for example, clinical decision aids can easily be built to calculate risk assessments that can be shared automatically with clinicians at the point of care.

Examples of our experience in applying novel data sources to answer complex questions include:


  • Link data sources within the Duke Health system, including the Duke Databank for Cardiovascular Disease, the Duke Echo Lab Database, MRI and Cardiac Computed Tomography databases, and healthcare encounters from Duke’s EHR to provide analyses for publications.


  • Support primary and secondary analysis of clinical trial data for more than 530 publications in peer-reviewed journals since 2000.
  • Pool clinical trial datasets around specific disease indications to leverage increased power to answer research questions not possible from a single data source.


  • Validate electronic phenotype algorithms for outcomes and procedures, including hospitalization for heart failure, stroke, acute MI, diabetes, CKD, HIV diagnosis, revascularization, and death.
  • Use Duke EHR data to provide clinicians with real-time risk assessment of sepsis and other adverse outcomes in the hospital setting.


  • Pool datasets from five NHLBI-funded cohort studies to develop and validate risk-prediction models for cardiovascular outcomes.
  • Use the CDC’s National Health and Nutrition Examination Survey data to describe the distribution of biomarkers in patients with a history of MI and correlate with standard risk factors.


  • Process MarketScan claims data using advanced programming techniques to handle large datasets.
  • Link Centers for Medicare and Medicaid Services claims data with clinical trial data.


  • Create and validate risk-prediction models across multiple data sources, including MarketScan, pooled NHLBI cohort data, and registry data.
  • Apply predictive value of discharge codes in Duke EHR data to calibrate estimated encounters rates based on MarketScan data.