Communicating diagnostic uncertainty reduces expectations of receiving antibiotics: Two online experiments with hypothetical patients
Funding information: This research was supported by grants from the German Research Foundation (DFG, grant numbers: BE 3970/8-2 and BO 4466/2-2) and the Leibniz Foundation (Leibniz Best Minds Program 106/2020). The funding source neither influenced the design of the study nor the analysis of the results. All authors approved the final version of the manuscript.
Abstract
The overprescription of antibiotics due to diagnostic uncertainty and inappropriate patient expectations influence antimicrobial resistance. This research assesses (i) whether communicating diagnostic uncertainty reduces expectations of receiving antibiotics and (ii) which communication strategies minimise unintended consequences of such communication. In two experimental online studies conducted in January and April 2023, participants read a vignette describing a doctor consultation for an ear infection and expressed their expectations of receiving antibiotics, trust in their doctor, rated the doctor's reputation and provided their intention to get a second doctor's opinion. Study 1 (N = 2213) investigated whether communicating diagnostic uncertainty and social externalities of antibiotic use (the negative social impacts of developing antibiotic resistance) decreases expectations for antibiotics and explores potential unintended consequences on the doctor–patient relationship. In Study 2 (N = 527), we aimed to replicate and extend the findings by adding specific treatment recommendations. Disclosing diagnostic uncertainty (vs. certainty) and communicating (vs. not communicating) the social externalities of antibiotic overuse reduced patients' expectations of receiving antibiotics. Yet, communicating uncertainty impaired trust in the doctor and the doctor's reputation. Combining the communication of uncertainty with specific treatment recommendations—particularly delayed antibiotic prescriptions—showed important to prevent these unintended consequences.
INTRODUCTION
Antimicrobial resistance (AMR) is among the top 10 threats to global public health (World Health Organization; WHO, 2020). In 2019, it was estimated that around 4.95 million deaths were associated with bacterial AMR (Murray et al., 2022). The overuse and misuse of antimicrobials are the main drivers in the emergence of drug-resistant pathogens (WHO, 2020), caused among other factors by unnecessary and inappropriate antibiotic prescribing in primary care (Smieszek et al., 2018). Unfortunately, the general population lacks knowledge about the negative social externalities of antibiotic overuse (Gualano et al., 2015), mainly that the inappropriate use of antibiotics contributes to the development of resistant bacteria that multiply and spread between people, animals and the environment. The term ‘social externalities’ refers to the unintended consequences imposed on others due to individual actions (Carande-Kulis et al., 2007). The development of drug resistance is often not directly felt by the patients, or the prescribing doctor, but AMR impacts the overall welfare of societies in the sense that it affects future patients who need antimicrobials and rely on healthcare systems for effective treatment (Coast et al., 1996, 1998). Specifically, AMR results in a large cost for society due to increased hospitalizations and mortality rates and decreased effectiveness of antimicrobials for current and future generations (Laxminarayan & Brown, 2001), as already reflected by the large burden of disease due to AMR (Murray et al., 2022).
Causes for the overprescribing of antibiotics in primary care are manifold. On the one hand, antibiotic prescriptions have been attributed to patients' expectations of receiving antibiotics (McNulty et al., 2013; Pan et al., 2016; Sirota et al., 2017; Wang et al., 2023). Clinically unjustified expectations to receive antibiotics can be driven by limited knowledge regarding the inappropriate use of antibiotics and the negative consequences of overuse (Thorpe et al., 2020b). A common misperception among the general population, for instance, is that antibiotics are effective against viral infections (Bakhit et al., 2019; Broniatowski et al., 2015; Gualano et al., 2015). On the other hand, doctors may encounter diagnostic uncertainty when making prescribing decisions: Infections often have nonspecific symptoms that resemble those of multiple diseases, and determining the source of infection is not easy, especially when doctors lack the resources to carry out testing (Crettez et al., 2020; Kotwani et al., 2010; Pulcini et al., 2017). However, to fulfil patient expectations (Ashworth et al., 2016; Crettez et al., 2020; Lucas et al., 2015; Md Rezal et al., 2015; Sirota et al., 2017) or prevent reputational harm (Broom et al., 2017), doctors tend to overprescribe antibiotics when experiencing diagnostic uncertainty (Kuehlein et al., 2010; Pandolfo et al., 2022).
This research aims to address these interrelated determinants of antibiotic overprescribing in primary care. Our objective is to examine whether patients' expectations of receiving a prescription for antibiotics can be reduced through communication about diagnostic uncertainty. The findings of this research have the potential to inform doctor–patient interactions and contribute to the reduction of antibiotic overuse. Additionally, we explore patients' perceptions of doctors who openly communicate diagnostic uncertainty, to identify communication strategies that effectively reduce patients' expectations of receiving antibiotics without negatively affecting the doctor–patient relationship.
Theoretical background
The doctor–patient relationship is characterised by an informational asymmetry, in which doctors have an informational advantage over patients regarding the identification and classification of health issues and the corresponding treatments. Due to this informational asymmetry, health care services have been described as credence goods: These are goods or services whose quality cannot be assessed by the customer (or patient, in this case) even after receiving it (Balafoutas & Kerschbamer, 2020; Dulleck & Kerschbamer, 2006). Credence good situations lead to inefficiencies such as undertreatment or overtreatment, with the latter being a prominent problem regarding antibiotics.
Patients may observe that their health problems resolve, but they may also experience difficulties in assessing whether the provided treatment was necessary. Imagine that a patient presents with acute otitis media (AOM). This is an ear infection that self-resolves in an estimated 85% of cases without the need for antibiotic intervention (Frost et al., 2019; Glasziou et al., 2011). Regardless, antibiotics are frequently prescribed for this condition (Hersh et al., 2011). A patient might take antibiotics for AOM and feel better within a few days—but will never know whether they feel better because the treatment was effective or the infection self-resolved.
Therefore, trust is crucial in credence goods settings (Dulleck & Kerschbamer, 2006). Specifically, a doctor's ability and certainty in diagnosing and treating illness is a factor that contributes to patients' trust in them (Anderson & Dedrick, 1990; Croker et al., 2013; Hsieh et al., 2010). Diagnostic uncertainty—that is, about the fact whether a disease is bacterial or viral—is common for many infectious diseases. Yet, communicating diagnostic uncertainty as to whether an infection is bacterial and whether antibiotics are needed may lower patients' expectations of receiving antibiotics, thus helping to decelerate AMR. Acknowledging diagnostic uncertainty can be an important aspect of shared decision-making (Politi, Clark, Ombao, Dizon, & Elwyn, 2011), as it promotes transparency, patient autonomy and the development of individualised treatment plans. On the negative side, communicating uncertainty may come with unintended effects, such as diminishing patients' trust in the doctor (Blanch-Hartigan et al., 2019; Politi, Clark, Ombao, & Légaré, 2015), harming the doctor's reputation (Om et al., 2016), for instance, when patients leave negative reviews online, or leading them to seek a second opinion elsewhere (Blanch-Hartigan et al., 2019). Such unintended consequences of disclosing diagnostic uncertainty could eventually undermine this practice.
STUDY 1
As the informational asymmetry between doctors and patients contributes to overtreatment with antibiotics (Balafoutas & Kerschbamer, 2020), the aim of Study 1 was to understand whether reducing this asymmetry by making the diagnostic uncertainty transparent can reduce patients' expectations of receiving antibiotics. We further reasoned that communicating both diagnostic uncertainty and the social externalities of antibiotic use would together have a stronger effect on patients' expectations, in line with research showing that knowledge about antibiotic resistance reduces the demand for antibiotics (Madle et al., 2004; Roope et al., 2020; Thorpe et al., 2020b, 2020a).
Specifically, we hypothesised that patients' expectations of receiving antibiotics are lower (i) in the presence of diagnostic uncertainty (vs. certainty) as to whether the infection is bacterial and (ii) when information about the negative externality resulting from antibiotic overuse is communicated (vs. not communicated). Additionally, we expected that communicating diagnostic uncertainty and social externalities together would attenuate antibiotic expectations more than what would be expected from the mere addition of both effects (attenuation interaction effect). Exploratively, we tested the potential unintended consequences of communicating about diagnostic uncertainty on perceptions of the doctor (in terms of trust and reputation) and the intention to get a second doctor's opinion as secondary outcomes.
Method
The study was pre-registered (https://aspredicted.org/yd7df.pdf). All materials, data and analysis code are publicly accessible (Open Science Framework: https://osf.io/kqvse/).
Participants
Based on an a priori power analysis for an ANOVA with main effects and interactions, the required sample size for a power of .80 was estimated to detect a medium effect size (f = .25), resulting in a minimum sample size of 158 participants. Considering recommendations for higher sample sizes to detect attenuation interaction effects (Giner-Sorolla, 2018), this number was multiplied by 14, leading to a large sample of N = 2212 at a power of .80. We recruited 2323 English-speaking adults residing in the United States via the recruitment panel Prolific (https://www.prolific.co). We recruited US participants because of the large amount of inappropriate antibiotic prescriptions (Young et al., 2020) and the market-based nature of the US healthcare system which makes it important to consider unintended consequences of communication strategies to reduce clinically inappropriate antibiotic use. Data collection took place in January 2023. Based on pre-registered exclusion criteria (upper and lower 2.5% based on completion time), 2213 participants who completed the full survey were included in the final analyses.
Participants were 18 to 93 years old (M = 40, SD = 14) and included 47.3% males, 50.4% females and 2.4% individuals who identified as non-binary or other. Of the participants, 55.2% had obtained at least a bachelor's degree, and 11.3% did not have health insurance. An overview of participants' characteristics is provided in Table 1.
| Study 1 (N = 2213) | Study 2 (N = 527) | |
|---|---|---|
| Age | ||
| Mean (SD) | 40 (14) | 38 (13) |
| Median [min, max] | 37 [18, 93] | 35 [18, 83] |
| Age distribution | ||
| 18–30 | 592 (26.8%) | 193 (36.6%) |
| 31–40 | 689 (31.1%) | 142 (26.9%) |
| 41–50 | 418 (18.9%) | 86 (16.3%) |
| 51–60 | 280 (12.7%) | 58 (11.0%) |
| 61–70 | 179 (8.1%) | 35 (6.6%) |
| 71–80 | 44 (2.0%) | 8 (1.5%) |
| 81–94 | 8 (1.4%) | 5 (1.0%) |
| Gender | ||
| Male | 1047 (47.3%) | 282 (53.5%) |
| Female | 1115 (50.4%) | 232 (44.0%) |
| Non-binary | 41 (1.9%) | 6 (1.1%) |
| Other | 10 (0.5%) | 7 (1.3%) |
| Education | ||
| Some high school | 13 (0.6%) | 2 (0.4%) |
| High school diploma or equivalent | 291 (13.1%) | 54 (10.2%) |
| Some college, no college | 462 (20.9%) | 115 (21.8%) |
| Associate's degree | 224 (10.1%) | 46 (8.7%) |
| Bachelor's degree | 837 (37.8%) | 212 (40.2%) |
| Master's degree | 310 (14.0%) | 78 (14.8%) |
| Doctorate | 65 (2.9%) | 17 (3.2%) |
| Other | 11 (0.5%) | 3 (0.6%) |
| Health insurance | ||
| Yes | 1963 (88.7%) | 466 (88.4%) |
| No | 250 (11.3%) | 61 (11.6%) |
Experimental manipulations
The study employed a 2 (diagnostic uncertainty: uncertainty vs. certainty) × 2 (social externalities: communicated vs. not communicated) between-participants design. Participants were randomly assigned to the conditions and had to imagine a visit at their GP where they present with an ear infection (adapted from Thorpe et al., 2020b). They received a vignette (Table S1) describing this visit with the manipulations being part of the doctor–patient conversation.
Diagnostic uncertainty
In the uncertainty condition, after the examination, the doctor stated that they could not determine whether the ear infection was bacterial or viral. The doctor then mentioned that antibiotics were only appropriate for severe bacterial infections. In the certainty condition, the doctor said that the infection was bacterial and that antibiotics were only appropriate for severe bacterial infections.
Social externalities
In the communication condition, the doctor explained that antibiotics can stop working due to antibiotic resistance and that this could mean that infections will be difficult to treat or even incurable in the future. The doctor pointed out the benefits of effective antibiotics for the well-being of society. In the no communication condition, no information about social externalities was provided.
Measures
The following constructs were assessed by using the mean score across several items.
Antibiotic expectations were assessed via six self-constructed items (each ranging from 1 = strongly disagree to 7 = strongly agree, adapted from Thorpe et al., 2021). Sample items are ‘I expect that my doctor will prescribe antibiotics’, ‘Antibiotics will be effective in treating my infection’ and ‘I actively request an antibiotics prescription from my doctor’. An exploratory factor analysis revealed a one-factor structure (Supporting Information, p. 4). Internal consistency was excellent (Cronbach's α = .92).
Trust in the doctor was measured via the 11-item Trust in Physician Scale (Anderson & Dedrick, 1990). Sample items are ‘My doctor is considerate of my needs and puts them first’ and ‘I trust my doctor's judgements about my medical care’. Participants were asked to rate their agreement with the statements from 1 = strongly disagree to 7 = strongly agree (Cronbach's α = .86).
Doctor's reputation was assessed by asking the participants to rate their experience with their doctor to help other people find a doctor. This was assessed like a Google star rating from one to five stars, where more stars indicate a higher reputation.
Intention to get a second doctor's opinion was assessed via two self-constructed items (‘I want to get a second doctor's opinion concerning my diagnosis’ and ‘I want to get a second doctor's opinion concerning my treatment’) on a 7-point scale ranging from 1 = strongly disagree to 7 = strongly agree (Spearman-Brown r = .96).
Procedure
After providing informed consent and answering demographic variables, participants read the hypothetical scenario. Depending on the experimental condition, the doctor then disclosed diagnostic uncertainty or not. Alongside (un)certainty, the social externalities of antibiotic overuse were communicated (vs. not communicated). Finally, participants indicated their expectations regarding receiving antibiotics, their trust in the doctor, the doctor's reputation and their intention to get a second doctor's opinion. The study was programmed using the non-commercial questionnaire web application Sosci Survey and conducted online. Participants were randomly assigned to the experimental conditions using Sosci Survey's built-in randomizer. The study had a median completion time of 4 min and 25 s. Each participant received a remuneration of £0.75 (equals US$0.92).
Statistical analyses
For the pre-registered analyses, a two-way ANOVA was employed using antibiotic expectations as the dependent variable. Independent variables were diagnostic uncertainty (uncertainty vs. certainty), social externalities (communicated vs. not communicated) and the interaction term. For exploratory purposes, separate two-way ANOVAs were employed using trust in the doctor, the doctor's reputation and the intention to get a second doctor's opinion as the outcome variables.
Results
As displayed in Figure 1a, participants reported lower antibiotic expectations when diagnostic uncertainty was disclosed, F(1, 2209) = 1080.38, p < .001, ηp2 = .328, in line with the first hypothesis. Participants also reported lower expectations of receiving antibiotics when the social externalities of antibiotic overuse were communicated, F(1, 2209) = 175.03, p < .001, ηp2 = .073, in line with the second hypothesis. The interaction effect of diagnostic uncertainty and social externalities was not significant F(1, 2209) = 0.27, p = .601, ηp2 < .001, contrary to our third hypothesis.
Exploratory analyses
Participants reported lower trust in the doctor when diagnostic uncertainty was disclosed, F(1, 2209) = 21.73, p < .001, ηp2 = .010. The communication of social externalities did not affect trust, F(1, 2209) = 1.49, p = .222, ηp2 < .001. However, there was a significant interaction effect of diagnostic uncertainty and social externalities: Information about the social externalities of antibiotic overuse buffered the negative effect of uncertainty on trust in the doctor, F(1, 2209) = 6.79, p = .009, ηp2 = .003 (Figure 1b). The same pattern was found for the doctor's reputation (Figure 1c): Participants rated the reputation of the doctor less favourably when diagnostic uncertainty was disclosed, F(1, 2209) = 222.16, p < .001, ηp2 = .091. The communication of social externalities did not affect reputation ratings, F(1, 2209) = 0.03, p = .872, ηp2 < .001. There was again a significant interaction effect of diagnostic uncertainty and social externalities; social externalities buffered the negative effect of uncertainty on reputation loss, F(1, 2209) = 15.52, p < .001, ηp2 = .007. Participants reported a higher intention to get a second doctor's opinion when diagnostic uncertainty was disclosed, F(1, 2209) = 282.13, p < .001, ηp2 = .113. Likewise, when social externalities were communicated, participants had a higher intention to get a second opinion, F(1, 2209) = 8.09, p = .005, ηp2 = .003. Here, the interaction of diagnostic uncertainty and social externalities was not significant, F(1, 2209) = 1.19, p = .275, ηp2 < .001 (Figure 1d).
Expectations to receive antibiotics were higher among certain age groups and lower when participants had a higher educational status. Furthermore, participants who reported that they had previously used leftover antibiotics as self-medication and had used antibiotics in the past 6 months also reported higher antibiotic expectations (Table S2).
Discussion
In summary, our findings demonstrate that disclosing diagnostic uncertainty and communicating the social externalities of antibiotic overuse lower patients' expectations of receiving antibiotics. However, contrary to our hypothesis, there was no interaction effect between communicating diagnostic uncertainty and using the explanation of the social externalities of antibiotic overuse on expectations to receive antibiotics. In contrast, explaining the social externalities helped to reduce antibiotic expectations irrespective of whether uncertainty was communicated or not. Additionally, we found that explaining the social externalities of antibiotic overuse yielded positive effects because revealing diagnostic uncertainty alone had unintended consequences: Disclosing diagnostic uncertainty was associated with lower trust in the doctor, a lower reputation rating and a higher intention to get a second doctor's opinion. Communicating social externalities alongside uncertainty buffered the unintended consequences on trust in the doctor and their reputation. In contrast, under diagnostic certainty, the additional communication of social externalities was associated with small effects of decreased trust in the doctor and a lower reputation rating, which could indicate that participants felt like the doctor weighed public health over individual outcomes. Therefore, the communication of social externalities seems particularly advisable in situations of diagnostic uncertainty.
It is typically recommended to pair the communication of diagnostic uncertainty with a recommendation on how to proceed (Bontempo, 2023). In Study 1, however, the doctor explained that antibiotics were only appropriate for bacterial infections but did not provide a treatment recommendation after the examination—neither for nor against antibiotics—which may partly explain the negative effects when communicating about diagnostic uncertainty. Therefore, Study 2 sought to address this limitation in the interpretation.
STUDY 2
We tested the effect of different treatment recommendations when communicating (vs. not communicating) about diagnostic uncertainty. In clinical practice, delayed antibiotic prescriptions have been introduced to deal with diagnostic uncertainty as to whether an infection is severe, bacterial and needs antibiotics to treat. This means that patients receive an antibiotic prescription but are instructed to wait and monitor their symptoms and only start the antibiotic treatment if their symptoms do not self-improve within a specified timeframe (Spurling et al., 2017). Delayed prescriptions lead to a significant reduction in antibiotic use when compared to immediate antibiotics prescriptions (Little et al., 2014; Spurling et al., 2017), making this practice an essential element of efforts to reduce antibiotic use in primary care.
Delayed prescribing is a strategy that corresponds to diagnostic uncertainty (Poss-Doering et al., 2020), whereas immediate antibiotic prescriptions are more common when diagnostic certainty is present (Rosenfeld, 2002). Theoretically building on cognitive dissonance theory (Festinger, 1957), we expected that matching the (un)certainty regarding diagnosis and treatment would be important to prevent unintended consequences on the patient–doctor relationship. Therefore, we conducted a pre-registered experiment to understand how the two treatment recommendations (immediate vs. delayed prescription) combined with diagnostic uncertainty (vs. certainty) affect the doctor–patient relationship, orthogonally crossing those two factors. Because the communication of social externalities alongside diagnostic uncertainty buffered the unintended negative consequences on trust in the doctor and their reputation, we always communicated diagnostic uncertainty alongside social externalities in Study 2.
Based on the results of Study 1, we hypothesised that participants would have lower expectations of receiving antibiotics in cases of diagnostic uncertainty (vs. certainty). Further, based on the explorative results from Study 1 and cognitive dissonance theory (Festinger, 1957), we expected that when a delayed (vs. immediate) prescription was recommended in the presence of diagnostic uncertainty, participants would (i) report greater trust in their doctor, (ii) rate the doctor's reputation more favourably and (iii) have a lower intention to seek a second doctor's opinion. Yet, when an immediate (vs. delayed prescription) was recommended in cases of diagnostic certainty, we expected that participants would have (i) higher trust in the doctor, (ii) rate the doctor's reputation more favourably and (iii) report a lower intention to seek a second doctor's opinion.
Method
The study was pre-registered (https://aspredicted.org/kh3mi.pdf). All materials, data and analysis code are publicly accessible (Open Science Framework: https://osf.io/kqvse/). Participants again received a vignette in which the manipulations were included.
Participants
Based on an a priori power analysis for an ANOVA with main effects and interactions, we estimated the required sample size for a high power of .95 (as recommended for replication studies, e.g. Anderson & Maxwell, 2017) to detect a medium effect size (f = .25) and an alpha of .0167 (using Bonferroni correction for multiple testing of three ANOVAs), resulting in a minimum sample size of 264 participants. Considering recommendations for higher sample sizes to detect reverse interaction effects (Giner-Sorolla, 2018), this number was multiplied by 2. Accordingly, we recruited N = 555 US participants via Prolific in April 2023. Individuals who had already participated in Study 1 could not participate in Study 2. Due to the pre-registered exclusion criterion of completion time, 527 participants who completed the full survey were included in the final analyses.
Participants were 18 to 83 years old (M = 38, SD = 13) and included 53.5% males, 44.0% females and 2.4% individuals who identified as non-binary or other. Of the participants, 58.8% had obtained at least a bachelor's degree; 11.6% did not have health insurance (see Table 1, for an overview of participants' characteristics).
Experimental manipulations
Participants were randomly assigned to one of the conditions in the 2 (diagnostic uncertainty: uncertainty vs. certainty) × 2 (treatment recommendation: delayed vs. immediate prescription) between-participants design. In all conditions, the social externality explanation was provided, and information on diagnostic (un)certainty and the treatment recommendations were varied as follows (overview see Table S1).
Diagnostic uncertainty
Diagnostic uncertainty was manipulated as in Study 1.
Treatment recommendation
In the delayed prescription condition, the doctor explained that ear infections often self-resolve but handed out a delayed prescription just to be sure. It was explained that the patient should wait a few days to see how symptoms developed. If the symptoms improved, they should not take antibiotics. If the symptoms did not improve or worsened, the patient was instructed to use the prescription to get antibiotics at a pharmacy. In the immediate prescription condition, the doctor also explained that most ear infections self-resolve but gave the patient a prescription for antibiotics just to be sure.
Measures
The same measures as in Study 1 were assessed. Internal consistencies of the scales were comparable to Study 1 (Cronbach's αexpectations = .92; Cronbach's αtrust = .89; Spearman-Brown rsecond opinion = .95). Additionally, we assessed perceived severity of symptoms with one item ranging from 1 = not at all severe to 7 = very severe.
Procedure
After providing informed consent and answering demographic variables, the same vignette for an ear infection as in Study 1 was used: The doctor first communicated diagnostic uncertainty or certainty, this time always accompanied by an explanation of the social externalities of antibiotic overuse. Then, the patients' antibiotic expectations were assessed. Next, the doctor recommended treatment by offering either a delayed or an immediate prescription. Finally, trust in the doctor, the doctor's reputation and the intention to get a second opinion were assessed. The study was conducted online, and participants were randomly assigned to the experimental conditions using Sosci Survey's built-in randomizer. The study had a median completion time of 4 min and 39 s, and each participant received a remuneration of £0.90 (equals US$1.12).
Statistical analyses
To investigate the effect of disclosing diagnostic uncertainty (vs. certainty) on antibiotic expectations, we conducted an independent samples t-test. Separate two-way ANOVAs were conducted for each dependent variable (trust in the doctor, doctor's reputation and intention to get a second opinion) using diagnostic uncertainty (uncertainty vs. certainty), treatment recommendation (delayed vs. immediate prescription) and the interaction term as independent variables.
Results
As displayed in Figure 2a, in line with the first hypothesis, participants reported lower expectations of receiving antibiotics when diagnostic uncertainty was disclosed (M = 3.74) in contrast to certainty (M = 5.45), t(525) = −16.19, p < .001, d = −1.410. Trust in the doctor did not differ depending on whether diagnostic uncertainty was disclosed, F(1, 523) = 3.13, p = .078, ηp2 < .001, and which treatment recommendation was given, F(1, 523) = 0.03, p = .868, ηp2 < .001. There was also no significant interaction of diagnostic uncertainty and treatment recommendations, F(1, 523) = 1.74, p = .187, ηp2 < .001, contradicting our hypotheses (Figure 2b). Due to the null findings, we carried out explorative equivalence tests to examine the equivalence to zero (Lakens et al., 2020): Previous studies did not yield information to define the smallest effect size of interest, which is why benchmarks of Cohen (1988) were used (Lakens, 2013). Therefore, the analysis was based on ηp2 = .010, indicative of a small effect. The equivalence test for diagnostic uncertainty was not significant, F(1, 523) = 3.13, p = .296, so we could not reject the null equivalence hypothesis. As the equivalence test for treatment recommendation was significant, F(1, 523) = 0.03, p = .010, we rejected the null equivalence hypothesis and considered the groups as equivalent. The equivalence test for the interaction of diagnostic uncertainty and treatment recommendation was not significant, F(1, 523) = 1.74, p = .163, so we could not reject the null equivalence hypothesis. Participants indeed rated the doctor's reputation less favourably when diagnostic uncertainty was disclosed, F(1, 523) = 19.49, p < .001, ηp2 = .040; there was no significant main effect of treatment recommendations, F(1, 523) = 1.33, p = .249, ηp2 < .001. However, as hypothesised, there was a significant interaction of diagnostic uncertainty and treatment recommendations, F(1, 523) = 11.00, p < .001, ηp2 = .020, in that participants rated the doctor's reputation more favourably when an immediate prescription was recommended in cases of diagnostic certainty and when a delayed prescription was recommended in cases of uncertainty (Figure 2c). Finally, participants had higher intentions to get a second doctor's opinion (Figure 2d) when diagnostic uncertainty was disclosed, F(1, 523) = 22.57, p < .001, ηp2 = .040, but which treatment was recommended did not influence the intention to get a second opinion, F(1, 523) = 2.20, p = .138, ηp2 < .001. We also found no significant interaction of diagnostic uncertainty and treatment recommendations, F(1, 523) = 3.56, p = .060, ηp2 < .001, contrary to what was expected.
Explorative analyses
When controlling for educational status, all previously reported results remained stable (see Tables S8 to S11). Furthermore, participants in the age group from 51 to 60 years had higher antibiotic expectations than those aged 18 to 30 (ß = 0.28, p = .021). Participants who had obtained a Master's degree (in contrast to those without a high school degree) reported lower antibiotic expectations (ß = −1.15, p = .047). Finally, perceiving the illness as more severe was associated with higher expectations to receive antibiotics (ß = 0.14, p < .001, Table S7).
Discussion
Replicating the results from Study 1, disclosing diagnostic uncertainty in the consultation significantly reduced patients' expectations of receiving antibiotics. When an explicit treatment recommendation was provided—by providing either an immediate or delayed prescription—trust in the doctor was not affected by diagnostic uncertainty communication. There are two possible explanations for this finding: First, we communicated the social externalities of antibiotic overuse alongside (un)certainty in this experiment because it had buffered the negative effect on trust in Study 1. Second, we combined uncertainty with a treatment recommendation in line with the recommendation to pair diagnostic uncertainty with further guidance (Bontempo, 2023), which could have prevented the previous unintended consequences on trust found in Study 1. Likewise, pairing diagnostic uncertainty with a delayed antibiotic prescription buffered the previous unintended consequences on the doctor's reputation. Yet, participants still wanted a second doctor's opinion considering diagnostic uncertainty. As the doctor's immediate or delayed prescription would have provided the patient with antibiotics, these results do not necessarily suggest that patients would seek antibiotics elsewhere; rather, they may point to dissatisfaction with the doctor's communication (Mellink et al., 2003; Shmueli et al., 2017).
GENERAL DISCUSSION
In two separate studies, we assessed whether disclosing diagnostic uncertainty can be used to reduce patients' potentially inappropriate expectations of receiving antibiotics and how this affects the patient–doctor relationship. In both studies, disclosing diagnostic uncertainty significantly decreased such expectations. As patient expectations have been characterized as a main driver of antibiotic prescriptions in primary care (McNulty et al., 2013; Pan et al., 2016; Sirota et al., 2017), our findings suggest that when diagnostic uncertainty is present, it can be used strategically to reduce patients' demand for antibiotics. Additionally, information on the negative social externalities of antibiotic overuse further reduced expectations to receive them. This finding is in line with previous research on the effect of educational interventions to reduce antibiotic expectations (Madle et al., 2004; Roope et al., 2020; Thorpe et al., 2020b, 2020a, 2021), and the theoretical assumptions of credence goods services that the information asymmetry causes overuse of antibiotics (Balafoutas & Kerschbamer, 2020). However, as shown in Study 1, disclosing diagnostic uncertainty can have unintended consequences on patients' trust in the doctor and the doctor's reputation, and it can increase patients' intention to get a second opinion. This corresponds to previous findings that being transparent about diagnostic uncertainty can harm the doctor–patient relationship (Blanch-Hartigan et al., 2019; Om et al., 2016; Politi, Clark, Ombao, & Légaré, 2011).
Study 2 offers insights into how to mitigate these effects: Diagnostic uncertainty should not be communicated alone but should be paired with specific recommendations on the next steps to prevent unintended consequences such as a loss of trust (Bontempo, 2023). The pairing of diagnostic uncertainty with a delayed antibiotic prescription was particularly promising: It did not lead to impaired trust and prevented reputational damage. Delayed prescriptions have been shown to reduce antibiotic use (Little et al., 2014; Spurling et al., 2017), but they do not reach their full potential as a considerable number of patients start using antibiotics on the same day they are prescribed them (Francis et al., 2012; Llor et al., 2022; Santana et al., 2023). Therefore, strategically disclosing diagnostic uncertainty as a reason why it is relevant to see how symptoms develop could prompt patients to refrain from antibiotics use longer. In practice, this could be accompanied by handing out a decision aid (Choosing Wisely Canada, 2022) or by advising patients to actively monitor their symptoms via a symptom diary before taking antibiotics (Santana et al., 2023).
However, whether communicating uncertainty helps to reduce antibiotic intake following a delayed prescription should be investigated further. In both studies, being transparent about diagnostic uncertainty increased participants' intention to get a second opinion. The desire for a second opinion could indicate patients' wishes to receive antibiotics elsewhere. We rule out this possibility in Study 2 because in both conditions participants received a prescription. Therefore, the results regarding a second opinion do not suggest that individuals want antibiotics, but rather that they may feel like the doctor did not examine them properly, or that they may experience difficulties when dealing with uncertainty (Mellink et al., 2003; Shmueli et al., 2017). A possible avenue to address difficulties in dealing with uncertainty related to antibiotic use is the uncertainty-normalizing strategy, which means to emphasise that existing uncertainty is an expected experience that is not reflective of a deficit in people's abilities (Han et al., 2018, 2021), and has been applied in health communication to buffer adverse cognitive and emotional responses to the communication of scientific uncertainty related to COVID-19 (Han et al., 2021).
It is important to differentiate between different types of uncertainty: The wording in the vignettes (‘either bacterial or viral’) could be seen as epistemic uncertainty and therefore perceived as a lack of competence (Bruine de Bruin et al., 2002; Fischhoff & Bruine De Bruin, 1999), which may explain the adverse results on the doctor–patient relationship in Study 1. In contrast, communicating uncertainty stemming from prognostic uncertainty where it is necessary to see how the symptoms progress (as communicated in Study 2), prevents reputational and trust damage. Likewise, the communication of external uncertainty, for instance, stemming from a borderline test value in a C-reactive point-of-care (CRP POC) test, is perceived as more competent than epistemic uncertainty (Løhre & Halvor Teigen, 2023).
Limitations
Our research comes with some limitations. First, we used two hypothetical vignette experiments to test our hypotheses, where a general concern is the potential lack of external validity due to the hypothetical scenarios. Yet, vignette-based studies are a commonly used tool in medical decision-making research, as they test patient perceptions of their doctor in controlled settings and allow the evaluation of doctor–patient interactions (Aguinis & Bradley, 2014; Hughes & Huby, 2002). The next step will be to test whether the findings hold as an intervention in a primary care context. Second, the data was collected online using Prolific and was not representative of the US population. Data collection via Prolific is not as controlled as in laboratory settings, which is why concerns about data quality have been raised (Chandler et al., 2014). Yet, in recent comparisons of different research platforms, Prolific provided high data quality in terms of comprehension, attention, and honesty (Douglas et al., 2023; Peer et al., 2022). Third, in our studies, we defined diagnostic certainty in relation to a control condition with a certain bacterial infection. Hence, we framed diagnostic uncertainty as positive because it was a deviation from a status quo where antibiotics are usually prescribed (despite potential diagnostic uncertainty) and could therefore be used to reduce antibiotic expectations. This differs from previous research, where the control condition was a viral infection, and diagnostic uncertainty was negative because it would lead to potentially inappropriate antibiotic use (Sirota et al., 2022; Thorpe et al., 2020a, 2020b). For a more comprehensive perspective, the study set-up would have benefitted from a second control group where the infection was certainly viral to measure how expectations change across the spectrum from certainly viral to certainly bacterial. Finally, some health systems such as in the United Kingdom and the Netherlands (Reibling, 2010; Reibling & Wendt, 2012) restrict patients' freedom to obtain second opinions. Therefore, the results regarding our measure of the intention to get a second doctor's opinion do not generalise to such contexts. Considering the specifics of the US health care system (Papanicolas et al., 2018), for instance, in terms of access to healthcare, racial discrimination (Nong et al., 2020) and regional differences in antibiotic prescribing (Hicks et al., 2015), it is crucial to replicate the findings with other populations because these potential confounders were not assessed in our study.
Conclusion
In two well-powered experimental online studies, we found that disclosing diagnostic uncertainty and communicating the social externalities of antibiotic overuse can be strategically used to reduce patients' expectations of receiving antibiotics—one of the main drivers of overprescribing in primary care. Further research should investigate whether the results replicate across other contexts and whether this also leads to reduced antibiotic use in the field. The results also suggest that it is important to provide treatment recommendations alongside uncertainty information to prevent unintended consequences on trust in the doctor and on the doctor's reputation. Delayed prescriptions seem particularly suitable, as their purpose (watching and monitoring symptoms) matches with the treatment requirements resulting from diagnostic uncertainty, where it is crucial to see whether the (non-severe) infection self-resolves.
ACKNOWLEDGEMENTS
Open Access funding enabled and organized by Projekt DEAL.
CONFLICT OF INTEREST STATEMENT
The authors have no conflict of interest to declare.
ETHICS STATEMENT
Ethical approval was granted by the University of Erfurt (2023-03 and 2023-12).
Open Research
DATA AVAILABILITY STATEMENT
The data, analysis code and materials are publicly accessible (Open Science Framework: https://osf.io/kqvse/). Both studies were preregistered (Study 1: https://aspredicted.org/yd7df.pdf, Study 2: https://aspredicted.org/kh3mi.pdf).
