Generating Real World Evidence form Real World Data

While completion of this assignment is required to pass the class, the goal of this final project is not primarily for evaluation.  Rather it is designed to get you to think about problems in your own area of interest, and how data can be brought to bear on the problem to help find a solution.  I am not asking you to solve the problem.  I am asking you to conduct the first steps toward a solution which I outline in the questions below.  Below each question, I have provided a representative (but incomplete) answer  to the question we discussed on day 1 – “Are men being overtreated or women being undertreated for heart disease.”  Your choice of question is totally up to you, though I expect that the fundamental question in your area of interest will be different from this particular cardiac issue.

Regarding the length, you can see that I have improvised a full page based on a very superficial overview.  For your assignment, I will say between 2 and 5 pages.

I want to see a submission from each of you, though I encourage you to work with your team on each submission.  Please indicate who the lead person is on each submission, and add the other names as collaborators.

My hope is that you will view this assignment as the start of a longer term project you will want to pursue on your own time and at your own pace (or I’d also accept leave for others to solve…).  I will provide feedback to help fine-tune your approach, and will be happy to support you on your journey to address the problem you specify if you wish.

What is one of the fundamental questions in your domain where the answer is truly unknown, or competing answers exist?

(eg – are men being over treated or are women being undertreated for coronary artery disease)

Frame this larger question in terms of smaller questions that will help you address the larger question.

(eg What are the raw numbers of men and women receiving the cardiac procedures?

What is the burden of disease for heart disease among men and women?

How effective are the therapies, and is there a gender difference in effectiveness?

What is the differential time course between initial presentation to intervention for men and women?

Are health seeking behaviors different for men and women in a way that may impact interventions?)

Based on what you learned in the course, what are the characteristics of the ideal data you would need to gather to help resolve the question?  Think about the breadth, depth, time interval, specific fields, granularity of records and responses  to individual fields.  How does this ideal data set compare with your sense of available data?

(eg I would love to see a comprehensive national registry of all people presenting with clinical complaints remotely attributable to hear disease.  Not everyone with chest pain has heart disease, and a comprehensive registry could show not only the total burden of disease, but the relative numbers of men and women who have symptoms that are possibly related to CAD, compared to the number of men and women who turn out to have true coronary artery disease.  The database should have information on comorbidities, medication ordering and use, laboratory parameters, and special study results.  Real data is far from this ideal – it is more localized and incomplete than the ideal data and it may reflect regional issues that do not generalize nationally (etc…)

What type of biases exist in the way the data are captured make creation of the ideal data set more challenging, and how remediable are those issues?

(eg It is possible that chest symptoms in men are taken more seriously by clinicians such that the documentation for a similar presentation in men and women appears different.  An analysis of the data would reflect that data bias in a way that may make it appear that differences in interventions  are appropriately attributed to a less severe presentation when in fact the disease may be equally sever, but the data do not reflect that.)

What are the subjective contextual issues that transcend the quantitative analytical results, even if you could conduct a perfect analysis of the perfect data set?

(eg – Different people have different belief systems and their tolerance of risk and benefit may impact their desire to undergo interventions that may cause harm, or not have the expected result. These beliefs may impact decisionmaking about the procedures in a way that affects results in ways that may be missed by the analyses.  It may be possible to develop a metric to assess this type or risk tolerance, and add this to the overall analysis)