The Downside To Clinical Trials

[gdpawel]

You reminded me of a posting I put on cancerfocus. I thought it is apropos.

Prominent Statistician Calls for Greater Collaboration With Clinicians

As a prominent statistician, Andrew J. Vickers, DPhil, at Memorial Sloan-Kettering Cancer Center in New York, sees a lot of data analysis and a lot of numbers. He was at the 2013 AUA (American Urological Association) Scientific Meeting in San Diego, California. Meetings like this are all about beta. People are presenting the results of their trials or some retrospective analysis of surgical databases. It is all about the numbers, the P values, the hazard ratios, and so on.

He had hoped to see a lot of statisticians but did not see many at a meeting like this. Everyone there was a urologist. Few others are specialists in analyzing data. In fact, if you go to the research meetings and see these presentations, and you look at the author list, you do not see statisticians on that list. Data analyses are being done by non-statisticians.

Vickers has a bit of a problem with that. He doesn't go into the operating room and perform urologic surgery, so why are the urologists going into the dry lab and doing the stats? A lot of those stats have really basic elementary flaws in them. Clearly, things have gone very badly wrong. We are at a major scientific meeting. People are presenting data that could affect the care of thousands or tens of thousands of men and there are elementary statistical errors that may seriously affect whether we believe their conclusions are right or not.

What Has Gone Wrong? First is what he calls the QuarkXPress problem. QuarkXPress is a computer program for assisting in the layout of articles for a magazine. If you are able to write, QuarkXPress enables you to publish a magazine.

Many people think that it is exactly the same with statistical software. You can load the statistics onto your laptop, press a few buttons, and get a statistical analysis. But statistics is a lot more than pressing the right buttons on statistical software. You really have to understand the underlying concepts, what the numbers actually mean, and whether you are selecting the right option, the right analysis, the right test.

The other big problem is the breakdown in the relationship between statisticians and clinical investigators. Vickers thinks that often the clinical investigators view statisticians as service providers: You order your sandwich and your coffee for lunch, and then you order up your T-test or your logistic regression when you return to the office. Statistical analysis is not seen as an equal collaboration between scientists interested in producing high-quality data together. Certainly some blame can lay at the feet of clinicians, but he does not think biostatisticians have been entirely free of blame.

Part of this problem is the inherent career structure of biostatistics. A statistician who wants to be promoted and be well known will not get there by being the second author on a paper about some clinical issue. Your career moves forward when you are the first author on a theory paper published in Biometrics or some other technical statistical journal. For many biostatisticians, having to work with clinical collaborators is actually somewhat of an irritation, essentially getting away from what they need to do to further their careers.

How can we actually get more statisticians at clinical meetings, working with clinicians, trying to solve clinical problems by doing high-quality research? If you want to address the problems, you have to go back to what they are. First, clinicians need to realize that analyzing statistics is not just a matter of operating statistical software. Clinicians need to treat statisticians as equal scientific collaborators.

Likewise, statisticians need to value working with clinicians and finding out information that actually helps patients. That must be seen as equally important as writing new theory papers about new statistical tests. Only by bringing together clinicians and statisticians to improve the statistical analyses that we are doing in clinical research, can we optimize the benefits of research for improving patient care.

Citation: Many Studies Have 'Elementary Statistical Errors'. Medscape. May 23, 2013.

[gdpawel]

Personalized Chemotherapy: Understanding Clinical Trials - the Kaplan Meier Graph

In this video of Personalized Cancer Chemotherapy, Dr. Larry M. Weisenthal explains the Kaplan-Meier graph. The graph is often used in clinical trials to compare survival times among patients with the same type of cancer who received different chemotherapy treatments. Understanding the graph is easy and also very useful as it will enable you to cut through the clutter in published clinical trial manuscripts and see at glance if any chemotherapy regimen provided a superior survival benefit.

http://www.youtube.com/watch?v=1Fz6kBXmAt0

Big Data Meets Cancer: Neil Hunt at TEDxBeaconStreet

https://www.youtube.com/watch?v=IWu6XWzbPBs

The consistent and specific cure or control of cancer will require multiple drugs administered in combination targeted to abnormal patterns of normal cellular machinery that effect or reflect malignant behavior, according to Dr. Arnold Glazier, former Oncology Fellow at Johns Hopkins. It is finding the patterns of malignant cells and developing a set of 5 to 10 drugs in order to cure or control cancer that classical clinical trials are not going to solve.

In clinical research, studies are deemed reportable when they achieve statistical significance. The so-called power analysis is the purview of the biostatistician who examines the desired outcome and explores the number of patients (subjects) required to achieve significance. The term “N” is this number. The most famous clinical trials are those large, cooperative group studies that, when successful, are considered practice-changing. That is, a new paradigm for a disease is described. To achieve this level of significance it is generally necessary to accrue hundreds, even thousands of patients. This is the “N” that satisfies the power analysis and fulfills the investigators expectations.

So what about Trials of N=1? This disrupts every tenet of cancer research, upends every power analysis and completely rewrites the book of developmental therapeutics, according to Laboratory Oncologist Dr. Robert A. Nagourney. Every patient is his or her own control. Their good outcome reflects the success or failure of "the trial." There is no power analysis. It is an "N" of 1.

This “breakthrough” concept however, has been the underpinning of the work of investigators like Drs. Larry Weisenthal, Andrew Bosanquet, Ian Cree, Robert Nagourney and all the other dedicated researchers who pioneered the concept of advancing cancer outcomes one patient at a time. These intrepid scientists described the use of each patient’s tissue to guide therapy selection.

They wrote papers, conducted trials and reported their successful results in the peer-reviewed literature. These results have provided statistically significant improvements in clinical responses, times to progression, even survival. By incorporating the contribution of the cellular milieu into clinical response prediction, these functional platforms have consistently outperformed their genomic counterparts in therapy selection.

With Cancer, Don’t Ask the Experts

http://robertanagourney.wordpress.co...k-the-experts/