Loading [Contrib]/a11y/accessibility-menu.js
Bolcic-Jankovic, Dragana, Eric G. Campbell, Jessica L. LeBlanc, Manan M. Nayak, and Ilana M. Braun. 2021. “Using ‘Don’t Know’ Responses in a Survey of Oncologists Regarding Medicinal Cannabis.” Survey Practice 14 (1). https://doi.org/10.29115/SP-2020-0016.
Download all (8)
  • Figure 1
  • Figure 2
  • Figure 3
  • Figure 4
  • Table 1: Respondents’ characteristics.
  • Table 2: DK responses by respondents’ characteristics.
  • Table 3: Logistic regression results for control variables on DK responses.
  • Table 3 (continued): Logistic regression results for control variables on DK responses.

Abstract

Objective: To evaluate whether “Don’t Know” (DK) responses conveyed meaningful information when provided by oncologists in a national survey on medicinal cannabis (MC).

Study Setting: This study is a secondary analysis of national survey data (n=237) collected between November 2016 and January 2017.

Methods: The national survey asked oncologists about views regarding MC’s risks/benefits, and whether they had sufficient knowledge to make MC recommendations clinically. Cognitive testing of the survey instrument (n=5) suggested that physicians did not always feel that they possessed adequate MC knowledge in all domains, so DKs were added to six of 27 survey items. (Three items were batteries of questions while three were single questions.) We constructed bar graphs for the sum of DK responses in each battery and for the sum of all DK responses in the survey. Mann-Whitney tests compared medians for all DK responses within Yes/No responses to the sufficient knowledge question.

Principal Findings: Statistically significant associations between DK responses and the sufficient knowledge question indicate that DK answers improved data quality by providing all respondents with an answer category that they feel fits their “true” answer. Associations between DK answers and other background variables such as respondents’ age, sex, and race are also discussed. The logistic regressions found that possessing sufficient knowledge was the only variable significant across six of the nine regressions, and the direction was consistent with the bivariate findings.

Conclusion: In this study, DKs appear to be valid responses, improving data quality by providing some respondents with an answer category that best fits their “true” answer. Future surveys aiming to learn about physicians’ views regarding emerging treatments might consider including the DK response option in some items.

Accepted: December 14, 2020 EDT

Appendix

Table 1: Respondents’ characteristics.
N %†
Gender
Male 154 65.0
Female 80 33.8
Missing 3 1.3
Age
<40 51 21.5
40–49 80 33.8
50–59 55 23.2
60+ 42 17.7
Missing 9 3.8
Mean age (SD) 48.728 (10.728)
Race/Ethnicity
White non-Hispanic 135 57.0
Other 98 41.4
Missing 4 1.7
Type of practice organization
Non-Hospital 121 51.1
Hospital 108 45.6
Missing 8 3.4
Type of patients
Adults only 183 77.2
Children only 33 13.9
Both adults and children 14 5.9
Missing 7 3.0
Cancer patient volume
<40 70 29.5
40–59 65 27.4
60+ 93 39.2
Missing 9 3.8
Medicinal cannabis state
Yes 126 53.2
No 100 42.2
Missing 11 4.6

†Total % vary slightly due to rounding

Table 2: DK responses by respondents’ characteristics.
% Ever DK to effectiveness (B3) % Ever DK to risks (B6) % Ever DK to patient type (B8) % All DK to effectiveness (B3) % All DK to risks (B6) % All DK to patient type (B8) % DK to antineo plastic effects (B4) % DK to mode of MC use (B9) % DK to compound preferences (B10)
Gender *
Male 31.8
Female 45.0
Age
<40
40–49
50–59
60+
Race/Ethnicity #
White non-Hispanic 31.1
Other 42.9
Type of practice * ** #
Non-Hospital 43.8 11.8 43.8
Hospital 60.2 27.4 56.5
Type of patients # # #
Adults only 33.9 46.4 16.7
Children only 51.5 45.5 34.4
Both adults and children 21.4 78.6 14.3
Cancer patient volume **
<40 64.3
40–59 55.4
60+ 38.7
Medicinal cannabis state # * # ** *
Yes 47.6 7.1 6.3 12.1 44.4
No 59.0 16.0 14.0 28.3 58.0
Have sufficient knowledge *** ** ** * ** * *** ***
Yes 31.9 23.2 11.6 4.3 1.4 15.9 2.9 20.3
No 59.0 41.6 30.1 13.9 13.9 30.1 26.2 62.0

#p<0.10, *p<0.05, **p<0.01, *** p<0.001

Table 3: Logistic regression results for control variables on DK responses.
Ever DK to effectiveness (B3) Ever DK to risks (B6) Ever DK to patient type (B8) All DK to effectiveness (B3) All DK to risks (B6)
S.E. p O.R. S.E. p O.R. S.E. p O.R. S.E. p O.R. S.E. p O.R.
Female1 0.338 0.386 1.341 0.336 0.045 1.965 0.317 0.700 1.130 0.369 0.763 0.895 0.542 0.543 1.390
Age 40–492 0.412 0.290 0.647 0.422 0.368 1.463 0.388 0.211 0.615 0.45 0.583 1.28 0.726 0.185 2.618
Age 50–592 0.458 0.567 0.769 0.472 0.032 2.748 0.432 0.436 0.714 0.52 0.756 1.175 0.966 0.982 1.022
Age 60+2 0.502 0.972 0.982 0.516 0.127 2.197 0.473 0.352 0.644 0.553 0.768 1.177 0.821 0.182 2.990
Other Race3 0.312 0.864 1.055 0.320 0.122 1.640 0.298 0.922 0.971 0.351 0.923 0.967 0.536 0.757 0.847
Hospital4 0.329 0.084 1.767 0.337 0.595 0.836 0.315 0.658 1.149 0.360 0.181 1.618 0.54 0.434 0.655
Children only5 0.497 0.619 1.280 0.478 0.189 1.874 0.466 0.663 0.816 0.498 0.049 2.668 0.748 0.355 1.999
Both Adults and Children5 0.652 0.675 1.314 0.753 0.506 0.606 0.714 0.030 4.701 0.764 0.837 1.170 0.996 0.332 2.627
Patient volume 40–596 0.442 0.680 1.200 0.454 0.825 0.904 0.419 0.486 1.339 0.485 0.171 1.941 0.752 0.331 2.077
Patient volume 60+6 0.443 0.106 0.489 0.458 0.842 0.913 0.428 0.738 0.867 0.502 0.572 1.328 0.790 0.758 1.276
Non MC State7 0.315 0.194 1.506 0.322 0.332 1.367 0.299 0.595 1.172 0.345 0.572 1.215 0.521 0.237 1.852
No sufficient MC knowledge8 0.358 0.002 2.997 0.390 0.007 2.843 0.336 0.967 1.014 0.475 0.005 3.824 0.808 0.061 4.542
Cox & Snell R2=.148 Cox & Snell R2=.097 Cox & Snell R2=.047 Cox & Snell R2=.084 Cox & Snell R2=.064
Nagelkerke R2=.197 Nagelkerke R2=.134 Nagelkerke R2=.063 Nagelkerke R2=.125 Nagelkerke R2=.140

N=215
Missing=22
1Reference=Male; 2Reference age=<40; 3Reference=White non-Hispanic; 4Reference=Non-Hospital; 5Reference=Adults only; 6Reference=Patient volume <40; 7Reference=MC state; 8Reference=Yes, sufficient MC knowledge
Values in bold significant at #p<0.10, *p<0.05, **p<0.01, *** p<0.001

Table 3 (continued): Logistic regression results for control variables on DK responses.
All DK to patient type (B8) DK to antineo plastic effects (B4) DK to mode of MC use (B9) DK to compound preferences (B10)
S.E. p O.R. S.E. p O.R. S.E. p O.R. S.E. p O.R.
Female1 0.594 0.244 0.500 0.353 0.810 1.088 0.44 0.918 1.047 0.339 0.763 0.903
Age 40–492 0.686 0.670 0.747 0.431 0.388 0.689 0.573 0.497 1.476 0.415 0.221 0.602
Age 50–592 0.803 0.720 0.75 0.476 0.940 0.965 0.636 0.111 2.755 0.468 0.745 0.859
Age 60+2 0.746 0.706 1.326 0.541 0.716 0.821 0.647 0.072 3.200 0.517 0.658 1.257
Other Race3 0.56 0.514 0.694 0.337 0.394 1.333 0.432 0.902 0.948 0.319 0.481 0.799
Hospital4 0.537 0.787 0.865 0.357 0.881 0.948 0.438 0.010 3.068 0.337 0.080 1.803
Children only5 0.742 0.115 3.218 0.507 0.540 1.364 0.556 0.022 3.568 0.496 0.808 1.128
Both Adults and Children5 1.050 0.092 5.855 0.663 0.251 2.143 0.939 0.724 1.393 0.708 0.263 0.453
Patient volume 40–596 0.790 0.302 2.259 0.483 0.507 0.726 0.566 0.404 1.604 0.446 0.731 0.858
Patient volume 60+6 0.798 0.670 1.406 0.470 0.906 0.946 0.586 0.976 0.983 0.455 0.746 1.159
Non MC State7 0.533 0.393 1.577 0.339 0.649 0.857 0.421 0.044 2.329 0.318 0.821 1.074
No sufficient MC knowledge8 1.093 0.029 10.862 0.410 0.095 1.982 0.829 0.002 13.816 0.385 0.000 5.779
Cox & Snell R2=.076 Cox & Snell R2=.036 Cox & Snell R2=.172 Cox & Snell R2=.169
Nagelkerke R2=.166 Nagelkerke R2=.053 Nagelkerke R2=.282 Nagelkerke R2=.226

N=215
Missing=22
1Reference=Male; 2Reference age=<40; 3Reference=White non-Hispanic; 4Reference=Non-Hospital; 5Reference=Adults only; 6Reference=Patient volume <40; 7Reference=MC state; 8Reference=Yes, sufficient MC knowledge
Values in bold significant at #p<0.10, *p<0.05, **p<0.01, *** p<0.001


  1. The American Association for Public Opinion Research Response Rate 4