The theory of planned behavior (TBP) framework was adopted in this study. Attitudes (favorability to perform a behavior), subjective norms (perceived social pressure to perform a behavior), and perceived behavioral control (perceived ease or difficulty to perform a behavior) in relation to SDM are linked with physicians’ SDM behavior.

The TPB model assumed that attitudes, subjective norms, and perceived behavioral control can influence behavioral intention, and the SDM behavior is influenced by behavior intention and perceived behavioral control (Armitage and Conner, 2001).

According to the TPB, the following hypotheses were examined in this study with patients presenting with ARIs.

H1: Physicians’ attitudes toward SDM positively impacted physicians’ intention to engage in SDM;

H2: Physicians’ subjective norms positively impacted physicians’ intention to engage in SDM;

H3: Physicians’ perceived behavioral control positively impacted physicians’ intention to engage in SDM;

H4: Physicians’ intention to engage in SDM positively impacted physicians’ SDM behavior;

H5: Physicians’ perceived behavior control positively impacted physicians’ SDM behavior.

A cross-sectional study was conducted in Hubei, a province with a middle level of social and economic development in central China (National data National Bureau of Statistics of China). The study was conducted in primary care sectors covering urban community health centers (CHCs) and rural township health centers (THCs). There were 1,161 CHCs and 1,139 THCs, serving 43.53 million outpatients in Hubei province in 2018 (National data National Bureau of Statistics of China).

A two-stage cluster sampling was applied in this study, which was described in detail and published in one recent research conducted by the research team (Wang et al., 2020). The first stage involved a random selection of five cities (the provincial capital Wuhan and four prefecture-level cities in Hubei province). The second stage involved one urban district and one rural county selection in the selected five cities. All the primary care facilities in each selected city/county were investigated from December 16, 2019 to January 17, 2020. The physician who were licensed to prescribe antibiotics and who had authorized ≥100 prescriptions over the past 3 months prior to the survey were eligible for this study. A total of 779 physicians were eligible for this study. Finally, 617 eligible questionnaires were collected, giving a response rate of 79.2%.

A self-administered questionnaire was applied in this study. The TPB model assessed attitudes, subjective norms, perceived behavior control, behavior intention of SDM, and physicians’ SDM behavior. The specific measurements were presented as follows. In addition, the barriers to the implementation of SDM were measured to supplement the results of TPB model (Section 2.5).

The questionnaire was piloted to evaluate the readability and reliability of the self-administered questionnaire. The questionnaire was tested using a sample of 30 primary care physicians in urban community health centers in Wuhan. Physicians were asked to complete the questionnaire and provide verbal feedback regarding the item’s readability. Some phrases were reworded according to the verbal feedback.

The trained research teams were paired and dispatched to the selected CHCs and THCs. The physicians who were licensed to prescribe antibiotics and the researchers administered the TPB survey to the physicians. Physicians’ demographic and work characteristics, such as professional titles and working years, were also collected.

The barriers to the implementation of SDM were designed to supplement for the results of the TPB with physicians’ perceived specific barriers to the implementation of SDM.

At the end of the TPB survey, physicians were asked to rate one or more barriers to the implementation of SDM in primary care. Several options were provided based on previous researches, such as lack of time or physicians’ perceived responsibility for making decisions for patients (Driever et al., 2020; Mathijssen et al., 2020). In addition, physicians are allowed to give answers outside the given options.

The TPB survey was administered in Chinese to the participants. As the original items from previous TPB studies and the Control Preference Scale were English written, a double translation was applied to ensure the consistency of translation, namely, the questionnaire was translated into Chinese and back-translated into English.

Some studies have reported associations between primary care physicians’ age or gender and their SDM behaviors (Brinkman et al., 2011; Sonntag et al., 2012). Also, clinical knowledge helps physicians weigh the treatment options and increase accuracy of risk perception, and physicians with different professional titles (such as resident physicians and senior physicians) and duration of working years may have different attitudes toward SDM based on previous studies (Alameddine et al., 2020; Oerlemans et al., 2021). These factors were considered to affect the physicians’ SDM behavior in this study.

Physicians’ age, gender, educational level, professional titles, working years, and receiving training in the past year were collected. The physicians were divided into subgroups based on the key characteristics in the TPB model. The professional titles were classified into three levels: resident physicians, attending physicians, and associate or chief physicians. Educational level was classified into senior high school and below, university degree, and graduate degree. The specific situation of medical education in China needs attention. Medical graduates with a non-bachelor degree can become assistant doctors. After accumulation of working experience, and if they pass an examination, they have the chance to get the full doctor license (Degner et al., 1997).

In addition, the key demographic and work characteristics were adjusted in the TPB model as covariates. The hypotheses of the effects of covariates on the TPB model are presented in Supplementary Material S2.

Goodness-of-fit indices were applied to evaluate the fit of the structural equation model: Tucker–Lewis index (TLI; >0.90 acceptable, >0.95 excellent), the comparative fit index (CFI; >0.90 acceptable, >0.95 excellent), root mean square error of approximation (RMSEA; <0.08 acceptable, <0.05 excellent), and weighted root mean residual (WRMR; <1.0 acceptable, <0.9 excellent) (Bentler, 1990; Hu and Bentler, 1999).

A total of 617 primary care physicians were investigated in the main study. The majority of physicians have a university or graduate degree (83.1%), while only a small proportion of physicians held senior job titles (13.0%). The median working years of the physicians was 16 years, and over 80% of physicians received training regarding antibiotics last year. The demographic and work characteristics of the physicians are presented in Table 2.

The physicians’ attitudes showed a strong inclination in favor of the SDM (mean = 3.92, SD = 0.70). They felt relatively high social pressure (subjective norms) of engaging in SDM (mean = 3.34, SD = 0.73) and their perception of behavioral control of engaging in SDM was positive (mean = 3.26, SD = 0.55). In addition, the physicians reported a relatively high intention to engaging in SDM (mean = 3.62, SD = 0.85) in primary care.

Most primary physicians (46.4%) reported SDM as their actual decision-making role with patients with ARIs, while 39.06% of physicians preferred the paternalistic approach when dealing with the decision-making with patients presenting with ARIs. As for the actual proportion of participation in SDM, only 19.8% of physicians reported they never or seldom engaged in SDM, while 11.7% of physicians reported that they engaged in SDM in more than 80% of medical encounters with patients presenting with ARIs (Table 2).

The final structural equation model had an excellent fit to these data (TLI = 0.98, CFI = 0.99, RMSEA = 0.07, WRMR = 0.89) and the detailed information is presented in Figure 1. Attitudes and subjective norms of SDM were identified as significant predictors of behavior intention, and behavior intention was demonstrated as one significant predictor of physicians’ SDM behavior (p < 0.001).

As initially hypothesized, attitudes toward SDM (β = 0.351, p < 0.001) and subjective norms (β = 0.755, p < 0.001) had direct effects on SDM behavior intention. As for the linkage between SDM behavior intention and SDM behavior, the higher intention of SDM will contribute to a more collaborative or active role with medical decisions (β = 0.265, p < 0.001) and a higher proportion of physicians engaging in SDM (β = 0.409, p < 0.001). Attitudes and subjective norms were shown to have indirect effects of 0.144 and 0.309 on the actual proportion of physicians engaging in SDM behavior, while they respectively had 0.090 and 0.200 indirect effects on the actual role of medical decisions (Figure 1).

To explore whether the relationship of SDM behavior model varied between subgroups of physicians, gender, age, educational level, professional titles, working years, and training were included as covariates based on previous researches (Kaplan et al., 1996; Menear et al., 2018). Multiple Indicators Multiple Causes (MIMIC) model assessed the effects of covariates on factor structure. This model had a good fit (TLI = 0.98, CFI = 0.99, RMSEA = 0.05, WRMR = 0.81) to the data. Educational level, age, and receiving training regarding antibiotics were found to be significantly associated with physicians’ perception of SDM. Specifically, not receiving training regarding antibiotic prescribing had significantly slightly lower scores on physicians’ attitudes (β = −0.025, p < 0.001) and subjective norms of SDM (β = −0.019, p = 0.021). Female physicians (β = 0.173, p = 0.009) with higher educational level (β = 0.231, p = 0.003) had a higher intention of engaging in shared decision-making. The details of the effects of covariates on the SDM behavior model are presented in Table 3.

The most frequently reported barriers were the physicians’ perceived patients’ expectations (n = 398) and patients’ incapability of making the decision (e.g., insufficient health literacy, the complexity of the medical problem) (n = 321). In addition, lack of time (21.0%) and lack of training of SDM (18.8%) were also reported. Also, 7.3% of physicians also proposed other barriers outside the given options, such as the discordance of clinical guidelines and patients’ expectations, the obstinacy of patients with medical decisions (e.g., patients’ strong demand of antibiotics), and patients’ uncertainty with medical decisions and distrust between physicians and patients (Table 4).

Our study served as one of the first series of studies to understand physicians’ SDM behavior based on TPB theory and identified barriers to the implementation of SDM with patients presenting with ARIs in primary care in China. Based on the well-fitting structural equation model, our study revealed that attitudes and subjective norms of SDM were identified as significant predictors of behavior intention, and behavior intention was one significant predictor of physicians’ SDM behavior. However, physicians’ perceived behavior control was not a significant predictor of SDM behavior intention and SDM behavior. Physicians’ perceived patients’ expectations, patients’ incapability of making the decisions, and lack of time and training were identified as main barriers.

Consistent with the result of Philippe et al.’s (2015) systematic review that subjective norms was most frequently and significantly associated with behavior intention, our study demonstrated the most significant effects of physicians’ subjective norms on SDM behavior intention (Thompson-Leduc et al., 2015). Subjective norm refers to the influence of the social environment (e.g., pressure from peers, patients, or administrators) on physicians. The influence of patients may explain, at least in part, the significant effects of subjective norms on physicians’ intention of SDM in primary care.

As the relationship between the physicians and the patients is the most fundamental unit of decision-making, this interpersonal and interdependent rapport between physicians and patients may help explain the significant positive effects of subjective norms on primary care physicians’ intention of engaging in SDM. The interpersonal influence of patients would contribute to increasing the physicians’ intention of engaging in SDM, especially under the context of the shift from the traditional biomedical model to patient-centered care in recent decades (Mead and Bower, 2000; Greene et al., 2012; Thompson-Leduc et al., 2015).

Congruent with previous studies, the increased intention of engaging in SDM was also determined by their attitudes toward SDM (Thompson-Leduc et al., 2015). One recent research showed physicians’ attitudes (being favorably or unfavorably disposed toward the behavior) significantly predicted their intention of participating in medical education programs (Rich et al., 2020). This seems to suggest exposing physicians to an SDM training program might improve their intention of engaging in SDM.

It is worthy to note that whether physicians received training regarding antibiotics was one significant predictor of physicians’ “attitudes” and “subjective norms” with patients presenting with ARIs. It seems plausible that increasing clinical knowledge would help increase physicians’ confidence and mitigate physician’s uncertainty of medical decisions, which was identified as one important contributor of the physician’s reluctance to participate in SDM, especially in primary care facilities with limited medical resources (Politi and Street, 2011; Bae, 2017).

As for the effects of educational level and gender of physicians on SDM perception of SDM behavior, previous researches have demonstrated that training and education played an important role in SDM (Chen et al., 2016), and female physicians were more likely to engage more in partnership with patients and more involved in medical encounters (Bertakis, 2009).

However, physicians’ perceived behavior control was not a significant predictor of SDM behavior intention and SDM behavior in our study. Existing researches applying TPB to understand physician’s SDM behavior have demonstrated the significant effects of perceived behavior control on behavior intention in oncology, mental health, or under another context (Sassen et al., 2011; Guerrier et al., 2013). Given the current lack of studies applying TPB in understanding physicians’ SDM behavior with patients presenting with ARIs in primary care, whether perceived behavior control is one significant predictor of behavior intention and SDM behavior remains unclear. One possible explanation is that physicians may perceive it is easy to engage in SDM; however, the perception of patients’ incapability of SDM in primary care may offset physicians’ willingness and intention to engage in SDM (Driever et al., 2020). Further researches are needed for exploring the potential factors affecting the association between perceived behavior control and SDM behavior.

In this study, we identified a range of barriers to the implementation of SDM corresponding with the literature (Légaré and Witteman, 2013; Driever et al., 2020). Over half of physicians rated the perception of patients’ expectations as delegates and patients’ incapable of making decisions as to the most frequent important barrier. As indicated in the NHS report, “patients don’t want shared decision making” was raised as one of the challenges of implementation of SDM. The preference of whether involving in medical decisions should be based on patients themselves, rather than presumption of physicians (Joseph-Williams et al., 2017).

On the other hand, patients’ inability to share in decision-making and insistent patient demand for antibiotics was also identified as main barriers to the implementation of SDM, which help provide implications for SDM training and educating the public (Mathijssen et al., 2020). For example, public campaigns that serve to educate the public about the use of antibiotics can be recommended. The reasons of “patients unable to participate in SDM” and “patients unwilling to participate in SDM” should be differentiated. Consistent with previous studies, lack of time and training was one of the main barriers in primary care and tertiary facilities (Huang et al., 2015).

There were also some limitations in this study. First, it relied on physician self-report outcome of the actual proportion of SDM behavior and thus may be at risk of social desirability bias (Boivin et al., 2008). However, relative to other source of information, such as audio and video recordings of physician–patient consultations, self-report outcomes are argued to overestimate the true proportion of SDM behavior (Gordon and Street., 2016). Second, as this study was conducted in primary care facilities in Hubei province in China, the results should be interpreted with caution with limited generalizability.