Data Source, study sample and data collection procedure
This survey study is part (Project 4) of the larger multi-year Comparative Effectiveness Research on Cancer in Texas (CERCIT) program, funded by the Cancer Prevention and Research Institute of Texas (CPRIT) (https://www.utmb.edu/scoa/research/supported-research-programs/comparative-effectiveness-research-on-cancer-in-texas/current-projects).17 We recruited potential survey respondents from the state-wide population-based Texas Cancer Registry (TCR) ((https://www.dshs.state.tx.us/tcr/).18 A sample of 6222 TCR registrants 18 years of age and older with a diagnosis of a solid malignancy in the past 12 months were obtained. All stages of cancer were included. Of these, 5535 were determined to be alive and able to receive study questionnaires by mail delivery. In total, 1566 individuals responded between March 2018 and July 2020, yielding a response rate of 28.3%. Of these respondents, 1480 (93.2%) were included for this analysis as described below. We performed telephone follow-up early in the data collection time period, however yield was very low (<10%). For the first 2300 potential participants, we sent 3 follow-up reminders and two replacement study questionnaires at 2 weeks, 4-6 weeks, and 8-10 weeks to non-respondents. During the course of this study’s data collection efforts, the TCR/Texas Department of State Health Services changed their policy regarding the number of allowable attempted contacts to potential study participants and we limited contacts accordingly to a single follow-up reminder at 8-10 weeks after first mailing. The subsequent response rates were approximately 25% with either 1 versus 3 mail out attempts. Potential participants’ primary care physicians were notified of the study to ensure that there were no medical objections regarding participation; none objected. Due to resource constraints, questionnaires were available only in English. Individuals identified with Spanish surnames received an additional recruitment letter written in Spanish advising them to have an English-speaking family member to assist with translation of questionnaires.
EOL care preferences were assessed using a previously published and validated instrument that generates 7 dichotomous outcome variables related to EOL patient preferences. 19,20 For this is analysis, we focused on preferences regarding location of death and mechanical ventilation using a question stem that presented a hypothetical scenario of having an illness with certainty of < 1 year to live. Assessment of preferred location of death was one item that asked if the illness got worse “where would you like to spend your last days-in a hospital, a nursing home, or at home?” Responses could be: “Hospital”, “Nursing Home”, “Home”, “Don’t know”. “Home” and “Hospital” were used for this study. Preferences regarding mechanical ventilation were assessed by combining two items which asked respondents whether they would use a mechanical ventilator for extending their life by one week or one month (yes, no, unknown). A ‘Yes’ response to either item was categorized as a preference for mechanical ventilation. Health literacy was assessed using the Newest Vital Sign (NVS), a 6-item health literacy assessment structured around reading and understanding information on a nutrition label. The NVS has been validated with high reliability and validity against the Rapid Assessment of health Literacy in Medicine (REALM) and the Short Test of Functional Literacy in Adults (STOFLHA) and has a high sensitivity for ascertainment of health literacy.21,22 The scale classifies respondents’ high likelihood of limited health literacy (score: 0-1), possibility of limited health literacy (score: 2-3), and adequate health literacy (score: 4-6). For the purposes of this study we condensed this to two categories: adequate health literacy (score ³4) or limited health literacy (score < 4) and those that did not fill out the scale were classified as “unknown”. Decisional self-efficacy was assessed using the Decisional Self-Efficacy Scale (DSES) an 11-item, 5-point rating scale with statements assessing a patient’s feelings of adequacy and efficiency in dealing with life situations and perceived ability to engage in treatment decisions (score range: 0= “Not Confident”, 2= ”A Little Confident”, to 4=”A Lot Confident”). The total score is summed, then divided by 11 and then multiplied by 25, with scale score ranging from 0 (not confident) to 100 (extremely confident). The scale was developed by the Ottawa Hospital Research Institute Patient Decision Aids Research Group.23 It provides a measure of a patient’s ability to engage in procurement of information about treatment options, express concerns, and make treatment choices. The scale has a high reliability with a Cronbach’s alpha coefficient of 0.84.
Socio-demographic and cancer clinical characteristics were obtained from a combination of TCR registry data and self-reported information. Race/ethnicity was self-reported or, in cases of missing self-reported values was obtained from TCR registry. Only 64 of 1480 respondents (4.4%) did not self-report race/ethnicity and were thus assigned by the TCR data. Agreement between TCR race/ethnicity data for the remaining 1411 was high with 94% agreement on ethnicity and 98% agreement on race. Race and ethnicity groups are non-Hispanic white, non-Hispanic Black, and Hispanic. Additional items included: age, gender, primary language spoken in the home, income, marital status, whether the individual lived alone or with others, and highest-attained education level. Self-reported health was collected using the Medicare Health Outcomes Survey,24 which asks respondents to rate their health on a 5-point scale from poor to excellent. Rurality and Texas Health Service Region were derived by zip code at the time of diagnosis of cancer, with rurality being defined per the US Department of Agriculture 2013 Rural/Urban Continuum Codes with scores of 1-3 yielding an urban designation and 4-9 being considered as rural.25 Finally, Cancer type and stage at diagnosis were obtained from the TCR database.
Descriptive statistics were generated for demographic and socio-economic characteristics. In order to address potential selection bias from in our findings observed in this convenience sample, we performed an inverse probability weighting analysis.26 For each observed case (n=1460), we computed the probability of return of the survey form the pool of 5320 eligible non-missing cases. We determined the weights based upon five available factors: age, gender, race/ethnicity, cancer site, health services area in Texas. The inverse probability of response was derived to weight each observation attempting to balance the selection bias due to non-response. Finally, normalized inversed probability (inversed probability divided by the mean) was used in the final weighting analyses. Normalized inverse probability reduces the weight loading because weights can increase the standard errors of estimates and introduce instability in the data. Significance of differences between responses to items with regard to socio-demographic characteristics were assessed using a Rao-Scott Chi-square test for the weighted data.27 Logistic regression analysis was conducted to examine the significance of demographic, socio-economic factors, health literacy and DSES score on respondents’ preferences. All factors were included in the multivariable model, and odds ratios (OR), and 95% confidence intervals (CI) were reported for each covariate, a p-value of <0.05 was deemed to be statistically significant. Finally, responses to the items were assessed using a one-way analysis of variance (ANOVA) test to examine the association of EOL preferences for location of death and use of mechanical ventilation with decisional self-efficacy using the DSES aggregate scores of each respondent. DSES score was included in the multivariable models as a continuous variable. Data analyses were performed using SAS (version 9.4 SAS Institute Inc., Cary, NC, USA).