Decision analysis and projection of prostate cancer risk and outcomes

In recent years, researchers have developed a variety of tools to help men and their doctors decide whether they need a prostate biopsy or what their outcome might be after a specific type of treatment for prostate cancer. However, the researchers who built these tools also have a sincere and serious appreciation of the limitations of the capabilities of these tools.

To be specific, and only as examples:

  • The PCPT Prostate Cancer Risk Calculator is only as good as the data from the Prostate Cancer Prevention Trial allow it to be.
  • The Partin tables are based on data from Johns Hopkins and certain other academic medical centers in the USA, so their accuracy when applied elsewhere are open to question.
  • The Kattan nomograms are also based on data from selected academic medical centers, so their accuracy may not be as high when applied to patients who get treated in community settings.

The bottom line is that our currently available predictive tools can only be as good as the data which have been used to construct them. As one researcher in this field has pointed out to me:

  • You can’t just look up a bunch of papers in the literature and copy over the estimates of (say) 5-year recurrence rates or long-term potency after surgery.
  • Different papers use different definitions for outcomes, and so literature estimates of potency after surgery vary from 20 to 90 percent, depending on how “potent” is defined.
  • The case mix in the published literature may differ (for example, two men with seminal vesicle invasion don’t necessarily have identical cancers; one man’s seminal vesicle invasion might be worse than another’s).
  • The actual radiation dose — and whether the patients received neoadjuvant or adjuvant hormone therapy — may dramatically affect outcomes to radiation therapy.
  • The probability of disease recurrence after surgical treatment for organ-confined prostate cancer varies by as much as 10-fold depending on surgeon experience.
  • There is still no consensus about whether PSA velocity is an important factor in determining the probability of a positive biopsy result as compared to the data required by the simple Prostate Cancer Risk Calculator.
  • For some forms of treatment, we don’t even have good quality, peer-reviewed, long-term follow-up data.
  • And on top of all of that, the long-term outcomes data that are available (for surgery, brachytherapy, and external beam radiation) are largely based on the treatment of patients diagnosed either before the PSA test was available or in the early years of the availability of that test; those patients therefore are not necessarily comparable to the patients faced with similar treatment options today.

Why is this important?

It is important because there is a web site now available that offers to assist patient decision making for prostate cancer and other prostate disorders using “decision analysis” systems.* We have inquired about access to detailed information regarding the basis of the guidance offered on this web site, but these inquiries have been ignored.

Now decision analysis is a very, very well developed science. It can be used for everything from programming sophisticated computer systems that make choices based on specific situations to helping patients make appropriate clinical choices about treatment when certain specific information is available. Indeed the science of decision analysis is critical to the ability of highly sophisticated computers to do things like play chess against (and beat) even chess grandmasters. But …

Decision analysis systems are only as good as the data that can be provided to build the decision processes. If you put garbage data into the system, then the guidance that the system provides you will also be garbage. GIGO:  garbage in, garbage out. This is not exactly a revelation. The reason that we can program a computer to play chess well is that, given enough time, one can accurately map every single possible chess move based on every single possible preceding chess move. As a consequence, in time, we will success in building a computer than can never be beaten at chess by a human. We aren’t there yet, but is certainly possible. Health care decision analysis is rather less precise!

People like Michael Kattan at The Cleveland Clinic, and many others, have spent years analyzing the quality of the data that is available and the ability to use those data to build decision analytic systems that would accurately predict probabilities 0f certain types of outcome associated with the diagnosis and treatment of prostate cancer. Their current conclusion is that we simply do not have good enough data to build accurate decision analysis systems yet. While these data aren’t complete garbage, they also aren’t exactly “clean.”

The “New” Prostate Cancer InfoLink has looked very carefully at the information that is available on the web site in question.* We have also looked carefully at the only published paper we are aware of that documents the decision analytic model on which (some) of that web site was constructed. The theoretical idea behind this web site is good. We would love to see a system like this that really helped patients to make the best possible decisions about the need for biopsy or outcomes after certain types of treatment. However, at this time, we do not believe it is possible to make accurate predictions for individual patients based on the available data, and we think that asking patients to pay an annual fee for access to such a system is highly inappropriate.

As we feel obliged to say fairly frequently in giving patients information about their options under certain specific circumstances: cavet emptor — let the buyer beware.

*We deliberately decided not to “make a big deal” out of the name of the web site or the company behind it. However, for those who are interested, the site in question can be accessed by simply clicking here.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: