Portal de Periódicos da CAPES

Cancer Prognostic Tools Free for Handheld & Desktop Computers

2002; Wolters Kluwer; Volume: 24; Issue: 4 Linguagem: Inglês

10.1097/01.cot.0000285558.08140.30

1548-4688

Marilynn Larkin,

Mobile Health and mHealth Applications

Nomograms to assist in treatment decision-making for patients with prostate cancer, renal cancer, and—most recently—sarcoma, are now available for downloading from a special area of the Memorial Sloan-Kettering Cancer Center Web site (www.nomograms.org or www.mskcc.org/mskcc/html/5794.cfm). The programs run on desktop computers (Windows 98 or higher, plus Microsoft Access needed), as well as handheld devices. Instructions for downloading, installing, and using the prognostic tools are posted on the site, along with PubMed-linked references to supporting literature. The nomogram software was developed by Michael Kattan, PhD, an Outcomes Research Scientist at Sloan-Kettering, who was diagnosed with Hodgkin's disease about 11 years ago. “I was studying artificial intelligence at the time, had electronic access to the literature, and was able to quickly review numerous statistical analyses of patients with my disease,” Dr. Kattan said in an interview. Nevertheless, prediction seemed kind of crude, and I thought the accuracy could be improved.” For example, the analyses provided stage-specific prognoses, but did not adjust for other prognostic factors. But Dr. Kattan had been told that although his disease was Stage IV, it was really “more of a ‘bad’ Stage II than a typical Stage IV.” How important is such information in predicting the probability of survival? “For purposes of patient counseling and treatment decision-making, ideally, you want the most accurate prediction you can get,” he explained. “If you can just put people in risk groups, it's easy. But if putting them into risk groups yields inferior prediction—because there's no adjustment for other factors—then I'd rather find a better method, as long as it's reasonable to program it, and just put in software to implement in more efficient manner.” “As every patient realizes sooner or later, and every oncologist knows too well, percentage predictions are valid for groups, but not for any one individual,” agreed Paul Meyers, MD, Vice Chairman of Academic Affairs in the Institute's Department of Pediatrics. “People are 80 percent well, not 100 percent or zero percent. What the nomogram does is give us a ball park estimate of their chances.” “In one sense, you lose the ability to just guess how a patient may do; you have to figure it out with the nomogram, and generally, we don't feel comfortable giving ourselves up to that,” observed Louis Potters, MD, Clinical Director of Radiation Oncology, who uses the Palm version of the prostate cancer nomogram. “But the more you use nomograms, the more you realize that you can't just rely on intuition or overly simplified categories, and you teach yourself that the intuitive approach [to prognosis] doesn't make sense.” Nomograms are also useful for defining groups of patients who are appropriate candidates for novel but potentially toxic therapies, Dr. Meyers added. For sarcoma, we'd like groups that are relatively homogeneous and at relatively high risk for failure with conventional therapy. Soft-tissue sarcomas are a heterogeneous group of diseases and cover a wide age span, but the nomogram takes all factors—age, site, grade, size—into consideration, and allows us to select patients for whom there is at least a 60% probability of death from sarcoma within 12 years. “This does several things,” Dr. Meyers continued. “It means we don't subject people with better prognoses to a higher risk of toxicity, and it also means that we can go to our statisticians and say, ‘here's a population for whom there's a high probability of adverse events,’ and this helps in determining the number of patients needed for a study to have sufficient power. We can do that with a smaller sample size, in a three-year time frame, before whole field changes and other potential therapies come to the fore.”Figure: Michael Kattan, PhDFigure: Paul Meyers, MDFigure: Louis Potters, MDTesting for Accuracy To develop the sarcoma nomogram, which was also published in the February 1st issue of the Journal of Clinical Oncology (2002;20:791–796) in an article that documents its utility in predicting 12-year sarcoma-specific death, Dr. Kattan and his colleagues D.H. Leung and Murray F. Brennan compared three different prediction models that could be used on future patients. “One is the old standby, the Cox Regression model, which is the standard for predicting outcome with multiple predictors,” Dr. Kattan explained. The second way was just going to the data set and pulling out patients who matched specific patients of interest, and plotting out the survival of those matches. That could be done because our data set was so large [2,136 adult patients]. “So if a new patient came in the door, you could see how they might do by going to the database; pulling out patients with the same histology, same site, same depth; plotting the Kaplan Meier curve for the matches; and using that as prediction for new patients. Intuitively, you would think that approach would be practically unbeatable, because you're pulling out exact matches—how could you do better than that?” But when the team compared the two approaches, the regression model was more accurate—probably because matching patients “is a very inefficient use of the data,” Dr. Kattan speculated. “You only pull out those that match exactly and leave all the rest behind. The regression approach takes all the patients and adjusts for any differences.” The third approach is a machine learning technique, recursive partitioning, which is a form of artificial intelligence. This method builds a tree out of the data set that splits with a yes or no at various points; it then creates risk groups of patients based on which “leaf” they end up on in the tree. Once again, the regression model worked better—most likely because it made use of all the data instead of creating subsets, said Dr. Kattan. With recursive partitioning, “patients in a particular leaf of the tree all get the same prediction. It doesn't adjust within that leaf, and there's probably heterogeneity left in the leaf.” Further details on the use of artificial intelligence versus standard statistical models for nomogram development are available in an editorial by Dr. Kattan in the February 15th issue of JCO (2002;20:885–887), which he titled “Statistical Prediction Models, Artificial Neural Networks, and the Sophism ‘I Am a Patient, Not a Statistic.’” Practical Considerations Dr. Meyers uses the paper version of the sarcoma nomogram, which is provided in the journal article, with new patients who ask about prognosis and to assist in decision making about therapies. “For a lot of the soft-tissue sarcomas, there are choices, and it really helps to know whether patients fall into a high-risk or lower-risk category,” he said. “Treatments can then be put into context, and patients can decide how much disability, down time, and toxicity they'd be willing to accept.” Recently, Dr. Meyers used the nomogram “and came up with a very high risk,” he explained. ”So the patient decided to accept all modalities available to him—chemotherapy, surgery, and radiation therapy. “Did we know about low- and high-risk patients before? Of course. But this tool allows us to give a numerical basis to the prediction, rather than simply saying, ‘in my experience, your type of tumor does well [or poorly].’” Dr. Potters uses the prostate cancer nomogram software routinely, as a sort of mental exercise, and to help me coach patients when it comes to discussing their options. Depending on the sophistication of the patient, I'll either use it directly [by showing them the results on the Palm] or in advance of my seeing them. “For the patient who can't understand it, or wants to be told what to do, or has a preconceived notion of what he wants, it's not going to matter what the nomogram says,” Dr. Potters continued. “But a lot of patients do understand and want to know what the risks are and are surprised to hear that there are more or fewer differences among treatment options than they had thought initially. For prostate cancer, short of having randomized trials with direct head-to-head comparisons, there's very little else out there that descriptively compares outcomes.” The bottom line, Dr. Potters stressed, is that “prognosis is not intuitive. It either requires the software program or the exercise of doing it on paper. But prostate cancer, you'd have to do three nomograms—brachytherapy, external beam, and surgery. Yes, you can do that on paper. But the software takes half the time, and with a sophisticated patient, you can do it right in front of them and show it to them.” Oncologists who don't feel comfortable at the computer should “have their son or daughter or relative download it,” Dr. Potters asserted. “It's really straightforward to use.” And, yes, he conceded, “it's counterintuitive to feel that we can't make the decision for the patient, especially if we consider ourselves experts. But the fact is, we can't. The reality is that by using the nomogram, you get a better result.” Oncology-Times.com Check www.oncology-times.com for basic information about OT. Although the articles are available as yet only in the print edition, the Web site does have a Table of Contents list of all articles starting in January 2001. Future Nomograms In the future, similar nomograms for other cancers—including breast and pancreatic—will also be available online, according to the Sloan-Kettering researchers.