Gist Representations and Decision-Making Processes Affecting Antibiotic Prescribing for Children with Acute Otitis Media

Abstract

Objective. To test the predictions of fuzzy-trace theory regarding pediatric clinicians' decision-making processes and risk perceptions about antibiotics for children with acute otitis media (AOM). Methods. We conducted an online survey experiment administered to a sample of 260 pediatric clinicians. We measured their risk perceptions and prescribing decisions across 3 hypothetical AOM treatment scenarios. Participants were asked to choose among the following options: prescribe antibiotics immediately, watchful waiting ("hedging"), or not prescribing antibiotics. Results. We identified 4 gists based on prior literature: 1) "why not take a risk?" 2) "antibiotics might not help but can hurt," 3) "antibiotics do not have harmful side effects," and 4) "antibiotics might have harmful side effects." All 4 gists predicted risky choice (P < 0.001), and gist endorsements varied significantly between scenarios when antibiotics were indicated, F(2, 255) = 8.53, P < 0.001; F(2, 255) = 5.14, P < .01; and F(2, 255) = 3.56, P < 0.05 for the first 3 factors, respectively. In a logistic regression, more experienced clinicians were less likely to hedge (B = -0.05; P < 0.01). Conclusion. As predicted by fuzzy-trace theory, pediatric clinicians' prescription decisions are associated with gist representations, which are distinct from verbatim risk estimates. Implications. Antibiotic stewardship programs can benefit by communicating the appropriate gists to clinicians who prescribe antibiotics for pediatric patients.

Highlights: We found clinicians' antibiotic prescription decisions were driven by gist representations of antibiotic risks for a given hypothetical patient scenario, and clinicians' gist representations and verbatim risk estimates about antibiotic-related risks were distinct from each other.We showed that the effect of patient scenarios on clinicians' antibiotic prescription decisions was mediated by clinicians' gist representations.Less experienced clinicians tend to "hedge" in their antibiotic prescription decisions compared with more experienced clinicians.The broader impact of our study is that antibiotic stewardship programs can benefit by communicating the appropriate gists to clinicians who prescribe antibiotics for pediatric patients, rather than solely focusing on closing potential knowledge deficits of the clinicians.

Keywords: antibiotic stewardship; antibiotics; fuzzy-trace theory; medical decision making; medical expertise; risk perceptions.

Figure 1

Histogram of clinicians’ number of years of experience. Years of experience are defined as time from medical or professional school graduation to time of participation in the study. Ten participants did not report their graduation years; thus, the following analyses were done on a sample size of 250 (10 participants did not report). We did not detect a difference in the years of experience among the experimental groups, F(2, 247) = 0.6, P = 0.54.

Figure 2

Average gist endorsements for each patient scenario. ***P < 0.001; **P < 0.01. The values of all variables are centered at zero and scaled. The zero exploratory factor analysis score represents the following: a disagreement with the average Likert score of 2.41 (i.e., disagree) for the “Why not take a risk” gist, neutrality with the average Likert score of 4.09 (i.e., neither agree nor disagree) for “the antibiotics might not help but can hurt” gist, agreement disagreement with the average Likert score of 6.17 (i.e., agree) for the antibiotics might have harmful side effects gist, and disagreement with the average Likert score of 2.01 (i.e., disagree) for “the antibiotics do not have harmful side effects” gist.

Figure 3

The vertical line shows the realization of the estimated probability of patient recovery without antibiotic treatment. Specifically, the probability of patients’ recovery under the “indicated scenario” is 40%, the probability of patients’ recovery under the “maybe indicated” scenario is 59%, and finally the probability of patients’ recovery under the “not indicated” scenario is 73%. Kolmogrov-Smirnov test results showed that the distributions are statistically indistinguishable from each other (D = 0.12, P = 0.47; D = 0.14, P = 0.31; D = 0.09, P = 0.81). Based on data from placebo-controlled randomized clinical trials,^, patients would improve without antibiotics (i.e., the proportion from the placebo group who improved) with 40% probability under the “indicated” scenario, 59% probability under the “maybe indicated” scenario, and 73% probability under the “not indicated” scenario.

Figure 4

The mediation analysis for the “why not take a risk?” gist. * p < .05 **p < .01, ***p < .001. The independent variable IV, i.e., the “not indicated scenario” predicts the mediator gist in a linear regression (β_a = -0.38, t=-3.34). The mediator predicts the dependent variable DV, i.e., prescription decision (β_b =-1.71, t=-7.30) in an ordinal logistic regression. The “not indicated” scenario significantly predicts the decision (β_c = 1.93, t= 5.46) and this effect is smaller when controlling for the mediating variable (β_d =1.62, t=4.39).

Figure 5

The mediation analysis for the “Antibiotics might not help but can hurt” gist. * p < .05 **p < .01, ***p < .001. The independent variable IV, i.e., the “not indicated scenario” predicts the mediator gist in a linear regression (β_a =0.55, t=4.22). The mediator predicts the dependent variable DV, i.e., prescription decision (β_b =2.38, t=8.95). in an ordinal logistic regression. The “not indicated” scenario significantly predicts the decision (β_c = 1.93, t= 5.46) and this effect is smaller controlling for the mediating variable (β_d =1.54, t=3.88).

Figure 6

The mediation analysis for the “Antibiotics do not have harmful side effects” gist *p < .05 **p < .01, ***p < .001. The independent variable IV, i.e., the “not indicated scenario” predicts the mediator gist in a linear regression (β_a =-0.21, t=-2.17). The mediator predicts the DV dependent variable, i.e., prescription decision (β_b =-1.29, t=-5.23) in an ordinal logistic regression. The total effect of the “not indicated” scenario significantly predicts the decision (β_c = 1.93, t= 5.46) and this effect is smaller when the mediating variable is included in the ordinal logistic regression (β_d =1.78, t=4.97).

Figure 7

The mediation analysis for the “Antibiotics might have harmful side effects” gist. * p < .05 **p < .01, ***p < .001. The independent variable IV, i.e., the “not indicated scenario” is not a significant predictor of the mediator gist in the linear regression (β_a =0.06, t=0.59, p=0.55). On the other hand, the mediator predicts the DV dependent variable, i.e., prescription decision (β_b =0.76, t= 4.02) in an ordinal logistic regression. The total effect of the “not indicated” scenario significantly predicts the decision (β_c =1.93, t= 5.46) and this effect remained the same when the mediator is included in the ordinal logistic regression (β_d =1.93, t=5.43).