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


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

In This Article


Data and Sample

The study sample was obtained from the 2017 National Survey of Precision Medicine in Cancer Treatment, a nationally representative survey of medical oncologists conducted between February and May 2017.[31,32] The survey was sponsored by the National Cancer Institute, National Human Genome Research Institute, and the American Cancer Society and collected information on oncologists' sociodemographic and practice characteristics and use of genomic tests.[32] Prior to fielding the survey, methodologists and clinical experts reviewed all content. Additionally, cognitive interviewing among practicing oncologists was conducted to ensure that questions were clearly worded and responses consistent with the intent of the questions. Oncologists were selected from the American Medical Association Physician Masterfile, which covers all licensed physicians in the United States. Practicing oncologists were selected using probability sampling, stratified by specialty, census region, size of metropolitan statistical area (MSA), and sex by age category. A total of 1281 practicing oncologists completed the survey via mail or online with a cooperation rate of 38.0%. We excluded oncologists who reported that they had not discussed genomic testing with patients or their families at all in the past 12 months (n = 61) and restricted our sample to the remaining 1220 oncologists who discussed genomic testing. More information about the survey design, sample weights, and analyses for nonresponse bias have been published elsewhere[31,32] and are summarized in the Supplementary Methods (available online). The survey protocol was reviewed by the Institutional Review Board (IRB) of RTI International, a nonprofit research organization. Survey data were deidentified and considered exempt by the National Institutes of Health IRB.


The measure of cost discussion frequency was based on the survey question, "In the past 12 months, when you or your staff discussed any form of genomic testing with your cancer patients or their families, how often did you discuss the likely costs of the testing and related treatment?" Response options among oncologists who discussed genomic testing within the past 12 months were never, rarely, sometimes, and often. Responses were categorized as "rarely or never," "sometimes," and "often."

We selected measures of physician-, practice-, and area-level characteristics previously shown to be associated with guideline-concordant practice[33–36] or hypothesized to be associated with cost discussions. Oncologist characteristics included age, years since medical school graduation, sex, and self-reported race and ethnicity, types of tumors treated (hematological cancers only, solid tumors only, or both hematologic cancers and solid tumors), percentage of time providing patient care, medical school affiliation, training in genomic testing, and use of next-generation sequencing (NGS) gene panel tests in the past 12 months. Practice-level characteristics were MSA, geographic region, and self-reported practice type, implementation of genomic testing services within the practice (internal policies or protocols for use of genomic and biomarker testing; electronical medical record [EMR] alerts for genomic test recommendations for particular patients or drugs; genomic/molecular tumor board), patient insurance status in practice (proportion of patients insured by Medicaid, self-pay, or uninsured), and patient volume (ie, 1–99 unique patients per month or ≥100 unique patients per month).

Area-level characteristics of the county of the physician's practice location were obtained from the 2016–2017 Area Health Resources Files;[37] these included county-level mean per capita personal income, percentage of individuals ages 25 years and older with at least 4 years of college, and median gross rent. Continuous measures of physician-, practice-, and area-level characteristics were categorized based on distributions within the sample. Exact wording of survey questions and response options are listed in Supplementary Table 1 (available online).

Statistical Analyses

We calculated descriptive statistics for physician, practice, and area-level characteristics. Associations between physician-, practice-, and area-level characteristics and frequency of cost discussion were assessed using polytomous logistic regression models.[38] A data-driven stagewise approach was used to identify physician-, practice-, and area-level covariates in developing parsimonious intermediate and final adjusted models. First, bivariable analyses were conducted to identify covariates statistically significantly associated with the frequency of cost discussions; those that were statistically significant at P less than .20 were included in one of three intermediate multivariable models of physician-, practice-, or area-level characteristics and cost discussions. The final multivariable model included covariates statistically significant at P less than .20 in any of the three intermediate models. Collinearity diagnostics were performed for the three intermediate and the final multivariable regression models. Statistical tests were two-sided, and statistical significance was defined as P less than .05. Analytic files were created with SAS 9.4 (SAS Institute, Cary, NC, USA) and analyses were conducted with STATA/IC 14.1 (StataCorp, College Station, TX, USA) Sample weights that accounted for the complex survey design and survey nonresponse were applied in all analyses.