Comparing Decision Making Between Cancer Patients and the General Population: Thoughts, Emotions, or Social Influence?

Abstract

This study extends a risk information seeking and processing model to explore the relative effect of cognitive processing strategies, positive and negative emotions, and normative beliefs on individuals' decision making about potential health risks. Most previous research based on this theoretical framework has examined environmental risks. Applying this risk communication model to study health decision making presents an opportunity to explore theoretical boundaries of the model, while also bringing this research to bear on a pressing medical issue: low enrollment in clinical trials. Comparative analysis of data gathered from 2 telephone surveys of a representative national sample (n = 500) and a random sample of cancer patients (n = 411) indicated that emotions played a more substantive role in cancer patients' decisions to enroll in a potential trial, whereas cognitive processing strategies and normative beliefs had greater influences on the decisions of respondents from the national sample.

Acknowledgments

Data collection for this study was supported by a research grant from The Leukemia & Lymphoma Society. This article represents the views of the authors and not necessarily the views of the funding agency.

Notes

¹Respondents in this sample were not preselected by known presence or lack of medical conditions. On the basis of American Association of Public Opinion Research calculation standards, the response rate for the national sample was 24% and the cooperation rate was 54%. About one third of the national respondents reported having been diagnosed with a chronic or acute illness, and 11.4% were in treatment. However, mean comparison tests indicated that within-group variations were limited for key variables. Thus, we considered the national sample a healthy-adult sample in comparison with The Leukemia & Lymphoma Society sample of cancer patients.

²A total of 759 Leukemia & Lymphoma Society members expressed an interest in participating in our study when the society first contacted them in August 2007. For comparison purposes, respondents were randomly selected from this data pool to generate 500 completed telephone surveys. The response rate was 67%, and the cooperation rate was 99% for this sample.

³Respondents in the national sample appeared to be slightly younger, with less education and less income. Both samples were predominantly White, but The Leukemia & Lymphoma Society sample had more female respondents. Compared with the 2006 American Community Survey (U.S. Census Bureau, 2007), the national sample seemed to have slightly overrepresented individuals with higher income, but the other parameters were similar.

⁴Independent-sample t tests showed significant mean differences between the two samples on all six items that measured information processing strategies. The t values ranged from 3.24 to 11.19, with p < .001. Low reliability has been a concern for these information processing measures in studies that are based on the Risk Information Seeking and Processing model, but rather than using composite scales as done in previous studies, we specified a measurement model first to take into consideration the effect from measurement errors.

⁵Mean scores on all five attitude measures were significantly different for the two samples. The t values ranged from 6.06 to 9.96, with p < .001.

Note. Pearson correlation coefficients appear in the upper right triangular half of the matrix, variances appear on the diagonal (in bold), and covariances appear in the lower left triangular half.

*p < .05. **p < .01.

Note. Pearson correlation coefficients appear in the upper right triangular half of the matrix, variances appear on the diagonal (in bold), and covariances appear in the lower left triangular half.

*p < .05. **p < .01.

⁶LISREL provides tests of the adequacy of the entire model, simultaneous estimation of all structural coefficients, and tests of statistical significance for all coefficients. The goal was to find a parsimonious structural model that explained the data reasonably well and then compare the unstandardized solutions across the two samples.

⁷In general, Leukemia & Lymphoma Society (LLS) respondents were more likely to have heard about clinical trial opportunities: For the national sample, M = 1.54 (SD = 1.23); for the LLS sample, M = 2.47 (SD = 1.19), t(996) = 12.17, p < .001. In addition, LLS respondents had more experience with clinical trials: 42.4% for the LLS sample, 7.6% for the national sample; χ²(1) = 140.65, p < .001. LLS respondents also reported higher self-assessed knowledge about clinical trial enrollment: For the national sample, M = 26.27 (SD = 28.95); for the LLS sample, M = 50.68, (SD = 30.92); t(997) = 12.88, p < .001.

Note. CFA = confirmatory factor analysis; CFI = comparative fit index; GFI = goodness-of-fit index; RMSEA = root mean squared error of approximation. In multiple-sample structural equation modeling, the χ² goodness-of-fit statistic is a measure of model fit in all groups, where a nonsignificant value indicates good fit. Because χ² has been shown to be sensitive to sample size (Bollen, 1989), the χ²/df ratio was also reported, where a value less than five indicates a good fit (Kline, 2005). RMSEA values less than .08 suggest reasonable error of approximation (Browne & Cudeck, 1993). For CFI and nonnormed fit index (values ranging from .00 to 1.00), .90 and above is generally considered to represent good fit (Bentler, 1990).

^a Revision from baseline CFA model to final CFA model: allowed error variances and appropriate factor loadings to be different across groups.

^b Revision from baseline structural model to final model: deleted nonsignificant paths, allowed coefficients to be estimated differently across groups.

Note. Except where indicated otherwise, items were measured using 5-point Likert scales.