Procedure
We constructed 46 paper scenarios describing hypothetical children aged 15 months who might have ear infections. Each scenario displayed the values for 15 different cues, such as fever, redness of the tympanic membrane, or ear pain during examination (see the figure at www.jfponline.com for more information). These cues were selected after consultation with physicians and parents, who felt that they were important for decisions about the diagnosis and treatment of AOM. The diagnostic cues were based on our previous study of US pediatricians32 and on the medical literature in the United States,33-35 France,36,37 and elsewhere.14,38 We included the result of insufflation of the tympanic membrane, even though French primary care physicians are not taught to use pneumatic insufflation in their diagnosis of AOM.
The cue values for each case were generated randomly using a computer program. Very implausible combinations of cue values were excluded from the scenarios. We did not, however, attempt to create a set of cases that would reproduce the actual mix of cases seen by pediatricians in their offices. Indeed, the inclusion of unlikely combinations of cue values was useful in forcing the participants to choose which cues were most important. Intercorrelation between cues was small to moderate, ranging from -0.40 to 0.39. This resulted from the rules for excluding implausible cases (eg, 0.39 between bulging and mobility) or from chance (eg, -0.40 between a history of ear pain and ear pain during the examination). The participants were told to put themselves in the place of the examining physician. They were presented twice with the same 46 scenarios. For one set they were asked to judge the probability that each child had AOM. For the other, in which the cases were presented in a different order, they were to decide whether to treat with antibiotics or to observe the child. They were also asked in the second set to indicate on a 5-point scale their degree of certainty that this was the right choice. Half the participants completed the diagnosis set first; half did the treatment set first. They completed the study at home. They were instructed to refrain from looking back at the first set after finishing and to take a short break of up to a day between the 2 sets. At the end of the study session, the participants answered certain questions about attitudes toward health care and risk and about their background that might account for differences in their responses to the scenarios.*
Data analysis
Two multiple linear regression analyses were performed for each participant (physician or parent). In one regression, the predicted variable was the judgment of the probability of AOM; in the other, it was the choice of treatment. Before performing the second regression, the treatment choice was combined with the degree of certainty about it to create a treatment score on a 10-point scale ranging from -5 for “observe/completely sure” to 5 for “treat/completely sure” (with no 0). This score was used in the multiple linear regression analysis as the dependent variable. A participant was included in these analyses only if his or her R2 passed the F-test for fitness of the multiple regression model at P less than .05; failure of a model to pass the F-test meant that the individual’s judgments were not predictable by the cues. This could occur if he or she had answered randomly or had made mistakes.
Multivariate analysis of variance was used to test the differences in the mean responses to the questions on attitudes and opinions of each group of participants. The association of individual attitudes and opinions about AOM and health care with treatment choices was explored by correlating the percentage of cases treated with antibiotics with the responses to each of the questions about attitudes and opinions.
Results
The judgments of the parents were remarkably similar to those of the physicians, both in the United States and France. The means and the ranges of the mean probability judgments of the individual participants in all the groups were almost identical at 50% or just above (Table 1, row 1). The cue weights for diagnosis (Table 2) were also quite similar. The physical examination provided the key information for both the parents and physicians. The only differences were that French parents gave more weight than the other groups to a past history of ear infections (beta weight = 0.15) and focused more on fever (0.30) and ear pain during the examination (0.28) than on bulging (0.17).
In choosing treatment, the parents in each country had lower mean treatment scores (Table 1, row 2) than the physicians, but these differences were not statistically significant. Likewise, the overall percentages of cases judged as needing antibiotics, though lower for parents, were not significantly different between the physicians and parents of each country: 53.0% for US physicians, 44.6% for US parents, 53.4% for French physicians, and 51.1% for French parents. Among the 4 groups the important cues for treatment (Table 3) were similar, stressing the physical examination findings and de-emphasizing the history, including the parents’ report of ear pain. French parents gave atypical stress to the child’s temperature (beta weight = 0.52), so that they differed from other groups in their judgments of several individual scenarios. Both groups of parents assigned significantly less weight than the physicians of all groups to the cue of the parents’ position on antibiotics. Indeed, when asked to put themselves in the place of a physician, the parents did not give importance to any of the parent-sensitive cues, even when the question was treatment. These cues included ear pain noted by the parent, the parents’ personal position concerning antibiotics, whether caring for a sick child greatly upsets the parents’ ordinary schedule, and whether there are babies or other small children in the family.