Clinical Prediction Tools for Head and Neck Cancers

National Cancer Institute, Frederick MD

Investigators

Abstract

A growing number of clinical prediction models (CPMs) have been developed to provide individualized estimates of the risk of various clinical outcomes, including mortality (or survival), for cancer patients. By utilizing patient, disease, and treatment characteristics to estimate these outcomes at the individual—rather than aggregate—level, these CPMs produce the critical evidence that patients and clinicians need to make rational, personalized decisions about cancer treatment. The individualized risk information produced by CPMs thus holds great promise for improving the quality of cancer care. The clinical implementation and dissemination of CPMs in cancer care, however, faces a critical challenge. Existing CPMs are each limited in their focus. Some CPMs estimate mortality risk conditional on cancer patients’ comorbidities without adjusting for the treatments they undergo, while numerous other CPMs focus on mortality risk conditional on patients’ particular treatments, but irrespective of comorbidities. For example, separate CPMs have been developed to estimate mortality risk of clinically localized prostate cancer following radical prostatectomy and conservative treatment; however, these models do not account for comorbidities, and no CPMs estimate mortality risk following radiation therapy. Hence, there is no single CPM that can support most clinical decisions related to cancer management even for a single cancer site. The Surveillance, Epidemiology, and End Results (SEER) Program is one of the premier cancer surveillance programs in the world currently composed of population-based cancer registries covering 30% of the total US population. The information collected on each and every cancer patient in SEER coverage areas includes demographics, a description of their cancer, treatment, and patient follow-up including cause of death for deceased patients. Periodic reports on various aspects of the cancer burden on the population including incidence, survival, treatment, and mortality using the SEER database are made available to the general research community and the public and are routinely cited in the medical literature and the media. In most statistical reports on cancer provided by SEER in the past, information on survival has been net survival, i.e., survival which reflects the likelihood of dying of causes related to a patient's cancer in the absence of competing causes of death and is of limited value to clinicians and patients because it does not reflect a patient's actual survival experience. An initiative of the SEER Program has been to utilize the high-quality patient follow-up data (that SEER has collected since its inception in 1973) for the purpose of creating statistical tools that provide survival information that is more clinically relevant to physicians and individual patients. The approach taken was to develop the SEER*Cancer Survival Calculator (SEER CSC) which will include algorithms or nomograms for various cancers that will provide survival statistics that account for competing causes of death including the crude probabilities of dying from cancer and from other causes along with the probability of survival. These probabilities are provided in response to inputting a profile of prognostic variables for a patient. One unique aspect of the SEER CSC is the "life expectancy in the absence of cancer” which is derived from Medicare data linked to SEER and quantifies how much longer someone is expected to live if they did not have their cancer. The purpose of this agreement is to work with the Veterans Administration to further revise and refine the SEER*CSC module for oral cancer, the language and structure of which we now also plan to use as a model for the other cancers the CSC is addressing. SEER*CSC was originally developed with the idea that it would just be released for health care professionals, but after substantial discussion over the past several years, it was determined that it will be released on a public website for use by patients and their families. This has required new revisions to make the language and overall format more accessible to a general audience and making appropriate disclaimers so that people will not use the calculator for inappropriate purposes. Work involves discussions with head and neck patient advocacy groups which will lead to refinements to the user interface for the entire SEER*CSC program (to include visual aids, for example) to accurately place it in the landscape of other prognostic tools, ensuring the information links to correct deeper level information about cancer statistics and interpretation more broadly, and revising the academic paper summarizing the development of the oral cancer module, and its use in a general public setting. This work is of great import to the VA, because Veterans are disproportionately affected by head and neck cancer compared to other populations in the U.S. In addition, as a relatively rare cancer, the power of case capture by the NCI makes them an ideal partner for this kind of work, which could not be carried out by the Department of Veterans Affairs alone - there are not enough cases or a record keeping system in the VA to easily create the models that are needed to make these sorts of complex calculations.