Search Annals: Advanced search

Podcast Transcript - October 2, 2007

[ Return to Podcast Home Page ]

Hello, and welcome to this week’s Annals of Internal Medicine audio summary for our October 2, 2007 issue. I’m Michael Berkwits, Deputy Editor at Annals.

We have another fabulous issue for you this week, with articles on predicting breast cancer gene mutations in women; the diagnosis and management of low back pain; a new diagnostic for diagnosing some causes of biliary obstruction in patients referred for endoscopic retrograde cholangiopancreaticography, or ERCP; and the cost-effectiveness of different strategies for screening patients from Asia and the Pacific Islands for hepatitis B.

Our lead article this week is a study of the accuracy of methods that have been developed to help predict the presence of BRCA1 and BRCA2 mutations in asymptomatic women who come from families with high occurrence of breast and ovarian cancer.

BRCA1 and BRCA2 are only 2 of several susceptibility genes associated with breast cancer, but mutations in the two genes dramatically increase risk for breast and ovarian cancer, and they are responsible for most cases of inherited cancer in women. Since the genes were identified in the early ‘90s, tests for the most common mutations have been become available for clinical use. But because the decision to test is fraught with difficulties — what should a woman do if her test result is inconclusive, or if it’s positive? — researchers at US medical centers that do genetic testing and counseling have developed statistical models that use information from the patient and her family to predict the probability of a BRCA1 or 2 mutations in the populations they’re serving. However, there are at least a dozen models; they differ in the information they use and in what they predict; and in clinical practice, different models applied to the same person can give a wide range of probabilities that a BRCA1 or 2 gene mutation is present. So the question remains whether some models are more accurate than others, and which can or should be used in clinical practice.

In this week’s lead article, lead author Giovanni Parmigiani from Johns Hopkins University in Baltimore and his coauthors combined information from the approximately 3300 women for whom the different models were developed, and compared the models’ predictions and accuracy using the actual genetic test results for the women.

The findings of this study are pretty quantitative, and interested listeners should consult the many Tables for precise estimates of the models’ accuracy. But qualitatively, the study has 2 main findings.

First, the models differed in their predictions for individual women, and they differed in their ability to discriminate between women who did and did not have mutations. However, most did pretty good, and the better-performing models correctly distinguished women who did and did not have mutations about 80% of the time.

Second, the models tended to be more accurate when used in settings where high-risk women with extensive family histories of cancer seek genetic counseling, and less accurate when used in broader populations of women, such as those who already had breast cancer in a given geographic region. Measures of test performance, such as sensitivity, specificity, and positive and negative predictive value, changed by population, and the numbers of false positives and negatives was high even when using a high threshold probability of 20% to refer women in these general populations for genetic testing.

So the authors conclude that the statistical models are accurate enough to use in the counseling individual women, although some are more accurate than others in certain populations of women. There’s no best model, and their application should be individualized to women and settings based on the estimates provided by this study. They acknowledge that few of the women included in their dataset were from racial or ethnic minorities, so the findings might not apply to those groups, and that the decision to proceed with actual genetic testing should be based on other factors too, such as patient psychology, costs, and patient preferences in relation to other screening strategies, such as annual mammography. In an accompanying editorial, Susan Domchek of the University of Pennsylvania and Antonis Antoniou, of the University of Cambridge in the UK, note that the models studied in this article do not answer the more important question of how likely a women is to develop breast cancer; models to predict breast cancer itself have been developed – the Gail model is most famous -- but they’re probably not accurate enough yet for clinical use. And the question remains about what these findings mean to the average practitioner in practice. Parmigiani and his colleagues say that these models can be used in primary care to help gauge the appropriateness of referral for genetic counseling, while Domchek and Antoniou make an antecedent point, namely that these models all require information about family history, so that as a first step, providers might more conscientiously solicit and record a family history of cancer.

So I’ve just told you what this article’s about, but after reading it a few times I’m still wasn’t sure I got it. So I called two of the paper’s authors, and asked some pretty basic questions. Dr. Giovanni Parmigiani, the lead author, is a Professor of Oncology in the School of Medicine at Johns Hopkins University in Baltimore, MD, and Dr. David Euhus, the senior author of the article, is a Professor of Surgery and the Medical Director of Clinical Cancer Genetics at the University of Texas Southwestern Medical Center, in Dallas. They explained what the models are, when and how they should be used in practice, and where to find them.

Q: Dr. Parmigiani and Dr. Euhus, thank you for joining me.

GP: You’re welcome.

DE: Thank you.

Q: Your study looks at models that predict the probability of a BRCA1 or 2 mutation in women from families with extensive cancer histories. Can you give us a lay of the land? What are the models, and what do they do?

GP: Succinctly, what they do is they examine a family history of breast and ovarian cancer and they produce one of three things. The probability that an individual would carry a cancer-causing mutation of BRCA1 and BRCA2, or the probability that an individual would test positive for the test for BRCA1 or BRCA2 mutation, or a score, which we could think of perhaps as a grade, that describes the extent to which the family history suggests the presence of a deleterious mutation.

Q: When might you want to use these models to predict a woman’s change of having a mutation instead of just ordering genetic test itself?

DE: The models actually I think are most helpful for communicating to a woman why maybe it’s not a great idea to do the testing. Most women overestimate their risk of breast cancer and their probability of carrying a mutation, and you can sit down and show them the numbers and go, well, you know, there’s only a, less than a 1% chance that you would have a positive test, or that you would carry a mutation. It may not be reasonable to do the testing. So for educating the patient I think it’s helpful. The other place it really comes in handy is when we calculate a very high probability, and the test comes back negative. And that helps us recognize the families and individuals that really need to be continually scrutinized and continually followed as though they did have a mutation, because that, kind of alerts us to the fact that the gene test was probably a false negative.

GP: Familial risk prediction algorithms are among the most reliable computerized algorithms that are available these days in medicine, especially in the area of cancer. On the other hand, they cannot be used mechanically. They should be an aid in decision making much more than a mechanical rule that is applied to a population without any further discussion.

Q: So what information do you need to use these models, and are they the kinds of things that a primary care provider might use in his or her office or practice?

GP: Some models require some relatively simple input that could be ascertained maybe with 4 or 5 simple questions from a questionnaire. Some models require the ascertainment of an entire pedigree which requires knowing the exact relation of various relatives to the person that is being interviewed and as well as details of the individual family history. And so in a primary care setting it is far more practical to use the first type of model, while in a genetic counseling situation where a more careful examination of the pedigree is typically part of the routine, it is easier to use some of the more complex and at the same time perhaps slightly more accurate models. But in our study we identified the FHAT model as one that had a particularly good trade-off between accuracy and simplicity. And while every model have particular areas of strength, we thought that the FHAT might be a good one for a preliminary screen in a primary care context.

DE: I think in the primary care clinic most people don’t need to go as far as actually calculating probabilities. But they need to have some way of capturing family history information, and recognizing that there are several relatives with breast cancer in one generation, or there’s breast cancer in multiple generations, or the breast cancer seems to be occurring younger, like before the age of 50 in this family, or there’s ovarian cancer as well. Those are the types of things they need to recognize. I think that breast centers, where people are specialized in managing breast problems and breast histories, and also in cancer genetics clinics, is really where the models are best implemented.

Q: And in those clinics, can you make any recommendations for which models should be used in whom?

DE: We use all the models. I think that the study shows that for getting a ballpark figure, all of the models are reasonable. And so I wouldn’t come out and say, you know, if you’re not using BRCAPRO then you’re not doing an accurate enough calculation. So a package like Cancer Gene calculates multiple models. You can get a range of values and give you a sense of, you know, does this really like an inherited predisposition family, or not. The Myriad model is something we can show to a patient, and say, you know, of all the blood samples that were sent to Myriad that had a family history like yours, this is the percentage that had a positive test. That helps people decide, well, maybe I don’t want to go through the trouble of the testing, or the expense of the testing.

Q: Do the models relate in any way to other breast cancer prediction models, such as the Gail model that predicts a woman’s chance not of having a mutation, but of having breast cancer itself?

GP: Models in this study tend to be much more accurate than ones that predict risk, partly because predicting genetic status is a somewhat easier task overall. However having a mutation that puts someone at high risk is the first step toward that person developing cancer later on, and so some of the models that we examine also predict future risk of breast cancer. But they are particularly useful for families where this risk is driven by their inheritance of a high-risk variant of one of these genes, and so they are not as broadly applicable as something like the Gail model.

DE: The Gail model is a well-calibrated, well-validated model, but it doesn’t do well with women who have family history beyond a sister, or a mother, or a daughter, let’s say. The Gail model doesn’t see aunts or grandparents with breast cancer, wouldn’t see a father with a breast cancer, and those are important family history data points that could point to an inherited predisposition. And most of the models that were looked at in the current study weren’t capable of producing a risk of breast cancer like the Gail model does, the exception being the BRCAPRO model, which does generate a risk of breast and ovarian cancer, based on this probability of carrying a mutation. And what you would find if you were to compare like Gail and BRCAPRO is that for most women, BRCAPRO would be calculating a much lower risk, and for women with second-degree relatives, male breast cancer, ovarian cancer in the family, the Gail would be inappropriately calculating a lower risk than the BRCAPRO. So the models are doing different things, and need to be applied to the appropriate population.

Q: How would inconclusive results on the test affect the findings of your study?

GP: It is true that about 10% of tests come out as inconclusive, and we are progressively categorizing more and more of these inconclusive variants as deleterious or not. But currently the inconclusive are considered negative, and that’s what we did in the study. From a statistical standpoint, I think the results our study remain valid as long as all models are penalized more or less equally by the presence of inconclusive results. So I think that while it is true that we could perform more accurate validations if we had more accurate ways of testing for the presence of mutations and we could categorize more of the inconclusive as either deleterious or not, I think that the validity of the study remains untouched by this fact.

DE: When you mention inconclusive tests, I think of two different things. One is, there’s some change detected in the gene, BRCA1 or BRCA2, but it’s not the type of change that we’ve definitely associated with a high risk for breast or ovarian cancer. It’s like a single nucleotide substitution that may or may not change the amino acid sequence. And those are generally called variants of uncertain clinical significance, and to me they’re not that big of a deal. Gradually we’re getting them categorized as being disease-associated or not disease-associated and there’s very, very few of those single nucleotide substitutions that are associated with the kind of risk that you get with the large truncating mutations, which generally your genetic tests are going to tell you that yes, there is one of those large truncating mutations, or no there’s not. So for the most part we get an answer like that. The other thing I think though with an inconclusive test though is where, like I said, we’ve calculated a high probability of a mutation, maybe 90 or 95%, but the test is stone cold negative. And we go back and we order the extra testing, maybe the BART panel where they look for large exonic deletions and rearrangements, and that’s negative too. That’s inconclusive testing too in my mind because I know there’s something in this family, but the technology isn’t there to identify what it is. It’s probably BRCA3 or BRCA4 and we just don’t know which genes those are yet.

Q: And is it typical practice to run all the models on an individual woman?

DE: Well I think most of the cancer genetics clinics that I’m familiar with use this Cancer Gene suite, and so you don’t have a choice, They all come out at once. You put in your family history and hit the Done button and you get results for all of them at once.

Q: And finally, where can listeners who are interested in these models find them?

DE: One place you can find all of these models is with the software package called Cancer Gene, which is available at www4.utsouthwestern.edu/breasthealth/cagene. That’s a package that we put together here at UT-Southwestern that has essentially all the models that were used in this validation and they’re available free of charge.

Q: Dr Parmigiani and Dr. Euhus, thank you so much for talking to me.

GP: You’re welcome

DE: Thank you very much.

That was Giovanni Parmigiani of Johns Hopkins University and David Euhus of University of Texas-Southwestern discussing their lead article in this weeks issue, on the prediction of breast cancer gene mutations in women with extensive family histories of cancer.

We have 3 CME articles this week, 2 systematic reviews of therapy for acute and chronic low back pain; and recommendation for the Diagnosis and Treatment of Low Back Pain based on those reviews, put together in a Joint Clinical Practice Guideline issued by the American College of Physicians and the American Pain Society.

In the systematic reviews, authors Roger Chou and Laurie Hoyt Huffman of the Oregon Evidence-Based Practice Center in Portland, OR, searched Medline and the Cochrane database for high-quality evidence on the non-pharmacologic and pharmacologic treatments for low back pain.

For acute low back pain, they found good evidence that superficial heat, NSAIDS, acetaminophen, and skeletal muscle relaxants are moderately helpful, and fair evidence that spinal manipulation offers small to moderate benefits.

For chronic or subacute low back pain, they found good evidence that cognitive-behavioral therapy, exercise, spinal manipulation, and physical rehab were moderately helpful, and fair evidence that acupuncture, massage, therapeutic yoga, and physical conditioning were moderately helpful. They also found that tricyclic antidepressants were moderately helpful, as were opioids, tramadol, benzodiazepines, and gabapentin for radiculopathy.

These findings are summarized in a Table embedded in Figure 2 of this issue’s guidelines. There was poor reporting of harms, no testing of the sequencing of therapies, and no evidence supporting recommendations of one treatment over another.

The American Pain Society did their own review of the available evidence supporting the diagnosis of low back pain, and they and the ACP put it all together with the following recommendations:

Clinicians should perform a focused history and physical examination to categorize patients as having nonspecific low back pain, back pain associated with radiculopathy or spinal stenosis, or back pain associated with other specific causes, such as cancer or infection.

Routine diagnostic testing is not necessary for people with nonspecific LBP, but may be indicated for people with severe or progressive neurologic deficits, or when serious underlying conditions are suspected based on the history and physical exam. Figure 1 of the guideline provides a review of possible serious causes of low back pain, and what imaging or additional studies to obtain. Imaging in people with signs or symptoms of spinal stenosis or radiculopathy should only be performed if the patient is a reasonable candidate for surgery or epidural injections.

Clinicians should provide patients with information about the expected course of their symptoms and about self-care options, and advise them to remain active; they should give them drugs such as acetaminophen and NSAIDs as needed for acute pain relief, but not before assessing baseline pain severity and functional deficits; and they can guide patients who don’t improve to non-pharmacologic interventions that are supported by the evidence. For patients who do not seem to respond to this approach, other published guidelines suggest waiting at least 3 months and as long as 2 years before referring a patient with nonspecific pain for surgery.

Two final specific notes: a Cochrane review suggested that herbal therapies such as devil’s claw, willow bark, and capsicum are safe alternative treatments for acute exacerbation of chronic low back pain but they offer only small to moderate benefits; and oral corticosteroids are a for acute or chronic pain, as they have not been shown to be more effective than placebo.

All this is a very quick summary of a lot of work. Interested listeners should read the reviews themselves, and look especially at Figures 1 and 2 of the guidelines, that put all the information together in a diagnostic and management algorithm. And, Annals subscribers can receive CME credits by answering 2 quiz questions about this article at cme.annals.org. Just click the “My CME” link, to register or sign in.

Other articles in this week’s issue include:

A study of 73 patients with biliary obstruction identified by RUQ ultrasound who were referred for diagnostic ERCP, demonstrating that levels of insulin-like growth factor 1 obtained from bile samples during the procedure were very high in patients with extrahepatic cholangiocarcinoma, and perfectly distinguished patients with cholangiocarcinoma from those with pancreatic cancer and benign biliary pathologies.

A cost-effectiveness analysis of hepatitis B screening and vaccination strategies for people from Asia and the Pacific Islands, demonstrating that screening for HBsAg and treating people who are positive is cost-effective compared to the current strategy of voluntary screening or no screening at all, and that the same “screen and treat” strategy is also cost-effective even with a “ring vaccination” strategy in which close contacts of people who screen positive for sAg are in turn screened for HBsAb and vaccinated if they are antibody negative.

And we have an Update in Geriatrics, the 5th of 10 Updates appearing this year based on presentations given at the American College of Physicians annual Internal Medicine 2007 session held in April in San Diego.

And finally, it’s flu season, and here’s a pop quiz: when do you test for influenza, who do you treat with drugs, and what’s the difference between the two neuraminidase inhibitors zanamavir and oseltamavir? Run, don’t walk, to check out this month’s In The Clinic, on Influenza, to find the answers to those and other questions.

Well that’s it for this week. So far I’ve heard from a whopping one person about these summaries. If you’re not dozing off yet and you liked what you heard, of if you didn’t, write and tell me about it, at podcast{at}annals.org.

For full details of all of this week’s articles, please consult your print journal, or go to www.annals.org.

Technical support for this summary was provided by Andrew Langman, Neil Kohl, and Beth Jenkinson.

Special thanks to Kevin Stahl and all our friends at WHYY, public radio and television of Philadelphia, who helped produce this podcast.

Check back in 2 weeks for a complete summary of our regularly scheduled October 16, 2007 issue.

I’m Michael Berkwits, and thanks for listening.

[ Return to Podcast Home Page ]