Prof. Mark Weiner is currently the Assistant Dean for Informatics, and Professor of Clinical Sciences at the Temple University School of Medicine, charged with developing a curriculum to support the needs of the next generation of Clinical and Research Informatics leaders. He is also the Chief Medical Information Officer for the Temple University Health System, overseeing the clinical aspects of the inpatient Epic implementation, ensuring it achieves its goals of supporting the clinical and academic missions of the institution.

Throughout his career, he has conducted research and applied techniques that help bridge the gap between Health Services Research, clinical and research operations and Medical Informatics.

Here is my interview with him:

Anmol Rajpurohit: Q1. How has the recent increase in digitization and analysis of healthcare data impacted the drug development process? What are the top challenges and opportunities?

Mark Weiner: Acquisition and availability of increasing amounts of healthcare data is only beginning to be fully utilized by pharma in the drug development cycle. There is a great deal of focus on using data for study feasibility – ensuring that a reasonable number of potentially eligible patients exist for recruitment in clinical trials. While this is certainly valuable, a great deal more can be done with healthcare data for understanding where existing medications are working well, and where new medication development is truly needed.

Given the benefits of blockbuster drugs from years past, it is harder to develop a new drug that can replace these and capture a similar market share. The trick is finding the population that is not achieving the expected benefit, and develop new medicines that will work better for them.

AR: Q2. Data access has generally been one of the top challenges across healthcare analytics. Can you elaborate on the data desired by pharmaceutical industry and how readily that data is accessible?

MW: In the past, data available to the pharmaceutical industry has typically included administrative data sources that capture the billable markers of healthcare activity. Increasingly, clinical data captured at the time of the patient encounter are available, enabling analyses that include more direct clinical parameters like vitals signs, smoking status, and family history. To the extent the practices are collecting discrete data on patient status for registries or other purposes, it is sometimes possible to capture important, more direct patient parameters like functional status, and measures of disease severity.

Even when these data are available, the data collection is often captured when patients are not feeling well, and seeing the provider for a specific reason. With few exceptions like blood pressure and blood sugar monitoring, it is hard to capture data on regular intervals regardless of how the patient is feeling at that moment – and even there, the rigor and regularity of the data collection may differ widely for different patients. It is important to account for the significance of the variability in availability of data when conducting and interpreting an analysis.

For example, if a patient has a pattern of routine labs once every year, but then has a series of 4 sets of routine labs in a 1 month period, the fact that there was a change in pattern to the labs is clinically relevant, and may impact expected outcomes, even if the laboratory results themselves were normal.

I cannot discuss access to data without mentioning data privacy and HIPAA regulations. Generally, data must be de-identified to be used outside the provider or payer organization that is the source of the information. There can be no real medical record numbers, names, addresses, phone numbers, zip codes of more than 3 digits, or even exact dates. Data can be de-identified at the source so that it contains random pseudo-identifiers in place of real MRNs (Medical Record Numbers) and dates are shifted or rounded down to the first day of the month.

A good deal of research can be accomplished with de-identified data, but, by design, there is no way to follow up with patients. It is also difficult to merge de-identified data on the same patient across different institutions, though there has been some work on matching one-way hashed transformations of the original identifiable data to facilitate cross-institutional matching while still maintaining patient privacy.

AR: Q3. What measures would you recommend to address the data accessibility challenges?

MW: The same holes in data that exist for clinical research, and the needs of the pharmaceutical industry also impact the information needs of clinical care. While insurance data can capture the full spectrum of health utilization over time, the comprehensive, longitudinal availability of point-of-care clinical data is more elusive.

True data interoperability across different Electronic Health Records at different institutions is not as common nor easy as it needs to be. While standards for sharing of basic data exist, there is a difficult challenge with affirming the identity of the same patient across two institutions.

It is certainly feasible on a case by case basis to do this, but across the thousands of patients needed to do a good retrospective analysis, the problem becomes a lot more difficult. Again, patient privacy issues must be respected when conducting cross institutional analyses.

AR: Q4. In contrast to pharmaceutical industry, the providers and payers do have a good amount of healthcare data. What are your thoughts on the state of Analytics in those industries?

MW: Providers and payers have more direct access to patient data, but the quality of analytics is restricted by the data they do NOT have. Even the VA, which may consider to be a closed system, has difficulty accounting for the full spectrum of healthcare activity of their patients given the volume of care that takes place in non-VA facilities.

AR: Q5. During drug development and clinical trials what are the major ways in which Analytics can assist decision making?

MW: Analytics can assist decision making by providing objective, quantitative evidence to support expert opinion that provides too coarse an estimate of disease prevalence, activity and responsiveness to existing treatments. In design of clinical trials, analytics can do more than just tell you how many patients fit certain criteria, but help trial designers better predict outcomes within the heterogeneous sub-populations within the seemingly-homogeneous selection criteria.

With better information and analytics available today, the “discovery” at the time of an interim analysis of a low outcome rate in the control population should not occur. The midstream changes to study numbers or enrollment criteria required by these findings are expensive, and impact the eventual analyses, and are much more avoidable with better data and preliminary analytics.

Second part of the interview

Related:

• Xia Wang, AstraZeneca on Big Data and the Promise of Effective Healthcare
• Bill Moreau, USOC on Empowering World’s Best Athletes through Analytics
• Nicholas Marko, Geisinger on Building the Analytics Culture for Healthcare