Factors Associated With Oncologist Discussions of the Costs of Genomic Testing and Related Treatments

K. Robin Yabroff; Jingxuan Zhao; Janet S. de Moor; Helmneh M. Sineshaw; Andrew N. Freedman; Zhiyuan Zheng; Xuesong Han; Ashish Rai; Carrie N. Klabunde

Disclosures

J Natl Cancer Inst. 2020;112(5):498-506. 

In This Article

Sidebar

Supplementary Methods

National Survey of Precision Medicine in Cancer Treatment Survey Design and Analysis Weights. The National Survey of Precision Medicine in Cancer Treatment is a nationally representative survey of medical oncologists that was conducted between February and May 2017 and sponsored by the National Cancer Institute, National Human Genomic Research Institute, and the American Cancer Society. The survey included questions about oncologists' sociodemographic and practice characteristics and use of genomic tests. Information about multi-marker tumor panels used to inform treatment were collected, including BioSpeciFix, CancerSELECT or Cancer Complete, Caris Molecular Intelligence or Target Now, CGI Complete, DecisionDX, FoundationOne, FoundationOneHeme, FoundationACT, GPS Cancer, Guardant360, Mammaprint, Omniseq Comprehensive, Oncotype DX Breast, Oncotype DX Colon, OnkoSight Tumor Panels, Solid Tumor Mutation Panel (ARUP Laboratories) and non-commercial tumor panels performed at academic medical centers.

Prior to fielding the survey, three types of pretesting methodologies were conducted: expert review, cognitive testing, and usability testing. Survey experts and clinicians reviewed content and question and response wording. Cognitive interviewing in practicing oncologists was conducted prior to fielding the survey to ensure that questions were clear and responses were consistent with the intent of the questions. Usability testing was conducted to ensure survey navigation was simple and efficient.

Oncologists were selected from the American Medical Association Physician Masterfile, which covers all licensed physicians in the United States. Practicing physicians under the age of 75 were selected using probability sampling, stratified based on cross-classification of Census region (Midwest, Northeast, South, West), size of metropolitan statistical area (small/medium – fewer than 250,000 in population large -250,000 - 1,000,000 in population, and very large - 1,000,000 or more in population), specialty (Oncologists, Hematologists-oncologists, Hematologists), and sex by age category (female, male <55 years, male ≥55 years).

The sample of 4,904 oncologists was allocated proportionally to the 108 sampling strata with at least 2 oncologists per sampling stratum. In each sampling stratum, the probability of selection was the number of physicians allocated divided by the total number of physicians. Within a sampling stratum, the physicians were sorted in ascending order by a randomly generated uniform random number on the interval from zero to one. Physicians were sequentially selected to be in the sample until the number of physicians selected equaled the allocated number of physicians. For the selected physicians, the design weight was calculated as the inverse of the probability of selection. For the non-selected physicians, the design weight was zero.

Eligibility and contact information were verified by telephone for 3,465 oncologists (71%). The survey was fielded as a sequential mixed mode survey with mailed surveys to the confirmed eligible oncologists with a personalized invitation letter, and an endorsement letter from the NCI and ASCO, followed by email contact with a personalized link to the survey. Up to 2 email reminders and 2 follow-up mailed surveys were sent followed by telephone reminders. A total of 1,281 practicing oncologists completed the survey via mail or online with a cooperation rate of 38.0%. Participants received a $50 honorarium for completing the survey.

Sample Weights. Many of the sampling strata were collapsed to create variance estimation strata with sufficient respondents for stable estimates. A total of 51 variance estimation strata were created from the original 108 sampling strata. Noncontact and noncooperation adjustment factors were calculated within variance estimation strata. Within a variance estimation stratum, the noncontact adjustment factor shifted the weights from the unknown eligibility physicians to the known eligibility physicians, the ineligible physicians were removed from the sample, and noncooperation adjustment factor shifted the weights from the eligible non-respondents to the respondents. That is, for the respondents within a variance estimation stratum, the analysis weights were calculated as the product of the design weight, noncontact adjustment factor, and noncooperation adjustment factor.

Nonresponse Bias Analyses. Three techniques for evaluating nonresponse bias were conducted: comparison of response rates by subgroups, response propensity models, and a nonresponse follow-up study. Response rates were compared for Census region, MSA category, primary specialty, and gender/age category, the stratification variables for sample selection. There was little variation for Census region, MSA, or specialty. There was more variation for the gender/age strata, although differences were not extreme. Similarly, gender/age was the most important variable in the contact and cooperation propensity models. Thus, gender/age was considered the most important variable to retain in the collapsing of strata for noncontact and noncooperation adjustments.

A one-page follow-back survey was mailed to the nonresponding physicians. It was completed by 14.3% of nonresponding physicians to the main survey. Responses from respondents and non-respondents were compared on a question about the use of Oncotype DX that was asked of both groups. There were no statistically significant differences between the main study and the follow-back study for any of the subgroups used as stratification variables for sample selection.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....