To identify which events are internationally influential, we visited well-known media sites, including the Wall Street Journal, New York Times, BBC, Economist, Washington Post, Telegraph, Times, New Yorker, South China Morning Post, International Business Times, National Public Radio, and The Atlantic and found a total of 17 medical violence incidents reported in China between October 19, 2011, and April 4, 2020. In total, there were 29 English reports. We then traced these incidents to Chinese website media and crawled (using a web robot to collect scripts at high speed) them for encoding and analysis to estimate online public attitudes toward doctors and patients (Multimedia Appendix 1).

We selected popular Chinese news media (N=6) from the New Media Influence Index Report [25], and considered their lasting influence from 2011, as shown in Table 1. We selected these 6 news media for 2 reasons. First, when we traced back the Chinese violence against doctors reports in international media sources, these media had a higher report rate than other media [26]; second, we referred to the media influence from 2011 to 2019 and, according to the Kantar China Media Impact Report 2016 and 2019, these 6 news media were more influential during these years [27].

We obtained 17 cases and their comments from the selected news media (see Multimedia Appendix 2 for details). As 2 of the 17 cases occurred in private hospitals, pseudonyms were applied to both private and public hospitals. There were 3 second-class A hospitals, 1 second-class B hospital, 2 top-class B hospitals, and 9 top-class A hospitals. Thus, the study covers different levels of hospitals, consistent with the distribution of hospital ownership status and hospital level in the total sample of doctor–patient disputes [28]. As the median number of comments was 400 on each incident, based on the total media coverage, if the count exceeded 400, we took a random sample of 400. The comments of each case are arranged in chronological order and divided into X/2n areas, and 2 comments were randomly extracted from each area. If the count was under 400, all coverage was included in the analyses. We then eliminated the repeated and unrelated comments. If the number of meaningless comments in the selected final comment sample exceeded 20%, these comments were removed. The number of comments required was then supplemented by the above sampling method. The sampling method is based on a real-world situation. We then collated and analyzed all the comments we collected. The final comment numbers for each case are listed in Table 2.

Ethical approval was obtained from the School of Public Health, Zhejiang University. As open data obtained through web crawlers do not include any personal data, informed consent was not required (waived by the ethics committee), and the study involved minimal risk.

We collected all the comments, marked each sentiment of each comment, and then segmented these comments into words, and removed some stop words to obtain the probability P_ij of each word appearing in each sentiment:

where P_ij represents the probability that w_i represents sentiment e_j, and C(w_i,e_j) represents the number of times that w_i appears in sentiment e_j.

Thereafter, when judging new comments, we first segmented each comment and removed some stop words. According to the naïve Bayes principle, a certain sentiment of each comment is the accumulation of the sentiments of all participles of the comment. To eliminate the polarization problem, the calculation is multiplied by the value related to the length of each comment:

Policy content quantitative analysis is a semantic analysis method that combines quantitative and qualitative analysis of the content of policy literature. A proper policy database could display the essential information of the content of violence against doctors–related policy literature objectively [37].

This study used Tsinghua University’s policy analysis system, “ipolicy,” to retrieve policies. “ipolicy” is a unique characteristic analysis system, which can perform the specified analysis function according to users’ needs at different levels. At present, there are approximately 1,200,000 policy texts from 1949 to 2019 (including all the organs of the State Council, other central bodies, the National People’s Congress, and more than 34 provinces), which are still being updated [38].

In the retrieval module, the system provides various retrieval modes, and the content of the retrieval can be selected as full text or title retrieval. Advanced retrieval and retrieval in the results are also available to facilitate the user to find the text more accurately.

We used interrupted time series analysis (ITSA), a kind of longitudinal quasi-experimental design [39,40]. ITSA provides a segmented linear regression model, which could assess the effect of each central policy on changes in the “support doctor” and “blame doctor” online public opinion before the policy promulgation date and at 30 time points after the promulgation date. We combined some of the rates of sentimental categories to understand netizens’ (Internet Citizens [41]) attitudes toward violence against doctors events. Because the meanings of “blame doctor” and “support patient” are similar, we combined the 2 categories of results into the new category of “blame doctor.” Likewise, we combined the results of “blame patient” and “support doctor” into the new category of “support doctor.” Originally, we crawled 19,220 comments. For the sake of ITSA, we averaged the sentiments of comments of the same date. For those with only 1 comment on each day, we framed adjected comments of 3-4 days and selected the second date to represent the time point. Finally, we obtained 151 time points representing the online public opinion change (Figure 1). As the policy promulgation date overlapped with the 151 points, we finally obtained 15 policy intervention dates for ITSA. The models quantify both a level and slope change following the intervention, while accounting for the autocorrelation of rates. Autoregressive integrated moving average models were used to adjust for residual autocorrelation. We prepared the data with MS Excel, version 2019, and completed all statistical analyses with R statistical software package, version 4.0.2 (R Foundation for Statistical Computing), using a Type 1 error rate of .05 as the threshold for statistical significance.

The regression model used to fit these data is straightforward:

In this specific example (using the variable names from Table 7), the model is:

Table 4 lists the preliminary results of coding, including the number of comments belonging to each category in the years 2011-2020, and the percentage of total comments in this period.

As shown in Table 4, the proportion of “blame big system” is more stable than other categories; “blame doctor” peaked at 39.2% (91/232) in 2013, but decreased to 12.97% (191/1473) in 2014 and 4.86% in 2020; “support doctor” decreased from 2011 to 2012, but rose significantly from 2015 to 2020; “blame patients” reached the minimum of 10.3% (24/232) in 2013 and the maximum of 24.58% (362/1473) in 2014; “support patient” showed a downward trend, from the overall in 2020, when it was only 2.52% (Figures 2 and 3).

The keywords we used to retrieve related policies were “doctor-patient,” “medical,” “violence,” “medical violence,” and “violence against doctors.” We then obtained related policies and regulations from 2011 to 2016 in China (Tables 5 and 6).

We browsed these 112 policies, and 19 policies were promulgated by the central government (Multimedia Appendix 3), of which the most relevant central policy documents are as follows:

As shown in Tables 7 and 8, for the first central policy (promulgated on October 26, 2012), at baseline (before the policy was promulgated), there was neither a significant level nor trend change in the rate of “support doctor.” However, there was a drop of 2% per time point in the “blame doctor” trend (Figures 2 and 3).

There was an increase in the “support doctor” trend after each policy promulgation (average trend changes of 0.011 of the policies 2-12, P<.05), but most of the slope change was not significant. At almost every time point in the policy promulgation, there was a decline in the proportion of “blame doctor” (average level drop of –0.784 of the policies 2-4, P<.05); however, there was also a trend change lag (average slope change of 0.018 of the policies 2-12, P<.05).

In 2017, discussion in The Lancet suggested that recommendations to rebuild patient–physician trust in China should not only focus on the physician’s side but also on the patient’s side [42,43]. Recently, particularly during COVID-19, voices to stop attacks against health care personnel have been raised worldwide [44-46]. All the health professional associations, societies, and organizations from all specialties and disciplines should unite to protect staff, while the government should play a major role by enacting strong and appropriate policies [47].

The current research has found that the 14 violence against doctors–related policies promulgated by the Chinese central government are associated with the sentiment expressed in online public opinions. The rate of online comments supporting doctors tends to increase significantly (see Table 7 for significant values) after almost each policy promulgation, while the rate of blaming doctors declined at each policy promulgation time point. However, the trend of blaming doctors then rises later after every policy is enacted. The overall declining trend of blaming doctors (39.22% in 2013 versus 4.86% in 2020) and the increased ITS trend for each policy time point imply that policy effects are only temporary. Particularly, in recent years (2017-2020), there have been few polices, and the ITS trends are not significant. It is time to reflect on the effects of policies.

The recent discussions regarding the relationship between violence against doctors events and health care insurance are very lively. In 2018, China’s Congress promulgated the Social Insurance Law of the People’s Republic of China (2018 Amendment) [48]. Although China’s medical security system is continually improving, there are still some problems. Furthermore, the severe current aging of our society and young people’s increasing work pressure have led to an increase in the incidence of some major diseases. For some significant diseases, many drugs are not included in the medical insurance coverage, and patients must bear most or all of the medical expenses alone. Patients are likely to experience significant imbalances, and it is easy to dissolve this inconsistency. The balance is blamed on the medical institution and the most directly contacted population—medical staff [49-51]. Our study has provided this new patient viewpoint from the health and science education for the public. In this study, we focused on netizens’ comments on major violence against doctors events and their attitudes toward doctors, patients, and the government as direct evidence to understand public opinions and the state of the physician–patient relationship. We found a significant relationship (see Tables 7 and 8 for significant P-values) between the time of implementation of government health reform policies and the shift of public opinion from “blame doctor” to “support doctor.” In 2011 and 2012, the proportion of “blame doctor” and “support doctor” opinions were relatively balanced. However, in 2013, the proportions began to shift: the proportion of “blame doctor” rose while “support doctor” dropped. Because of this differentiation, the government had to adopt some policies. By the end of 2013, the public mood had eased up after the implementation of related health policies. In 2014, the number of related health policies peaked but then decreased in 2015. Thus, the proportion of “support doctor” opinions fell in 2015. Again, the situation improved after the enactment of central regulations in 2016 (Figures 2 and 3).

From the “ipolicy” system, we retrieved related policies and regulations from 2011 to 2016 in China and found the most relevant central policies to be those decreeing stricter enforcement and severe punishment for those endangering the safety of doctors and other medical staff [27]. Furthermore, in 2014 and 2016, further policies showed more awareness of fighting illegal medical activities [23] and concerns for resolving mob-related criminal behavior at medical hospitals [24]. Before these enactments, the enforcement of punishment for violence against doctors and the level of security for hospital staff were weak, allowing doctors and other medical staff to become victims [52].

Nonetheless, it is the role of the government to enforce proper regulations to prevent danger to medical staff. Our results show that the only interaction between attitude and turning point occurred from 2013 to 2014, which is connected with relevant policies between 2014 and 2016 contributing to a significant drop in citizens’ “blame” of doctors (see Table 8 for significant P values). These efforts show that the Chinese government has heightened vigilance toward the problems displayed by increased violence against doctors incidents and more order in controlling the harms caused by such incidents. However, their efforts remain slightly below the US approach, suggesting that the Chinese government should continue to adopt more control and management roles in medical emergencies. According to the Chinese government, “Following the Occupational Safety and Health Act of 1970, 26 US states and two US territories require employers to develop detailed violence prevention programs. The Occupational Safety and Health Administration (OSHA) have also increased the number of visits to medical facilities, from 11 in 2010 to 86 in 2014. The American Government Accountability Office (GAO) gave a specific recommendation in its 2016 Work Report. Employers should be punished if they allow employees to be exposed to potential workplace violence, especially if they have a history of violence, assault or physical assault; provide additional information to inspectors in specific court summon cases to help develop a system of summonses; follow up confirmation of the need for further measures” [42,43,53].

In analyzing the comments rather than the news articles themselves, we provided a more direct measure of the public’s perspective and attitude toward the health care sphere and their stance on the patient–physician bond. The data we collected on these comments can further be used as a potential indicator of health care quality, as few reliable measures exist. Wang et al [54] utilized medical malpractice incidents as a probable marker of health care quality, but a more robust and direct measure is needed to probe the difficulties experienced by the Chinese medical field. We analyzed the comments based on the variable of time, so we can see changes and patterns through time, compared to previous studies, which examined more qualitative methods to seek reliable measures of the patient–doctor relationship [55,56]. Comments made by the general public can reveal their emotions and opinions, as well as their perceptions and viewpoints on the shortcomings of the health care field and expose needed reforms for the improvement of trust in medical professionals.

Our study also contributes to the qualitative study of feelings of mistrust in the physician–patient relationship in China. A survey of 107 health care physicians in Guangdong province, China, reported that physicians are caught in a “vicious cycle of mistrust” due to various contributing factors of work overload, conflicts of interest, unreasonable requests of patients, and more [57]. Restoring public trust in their physicians requires a strong foundation supported by government regulations for effective delivery and more transparent methods of health care quality assessments.

The media play an essential role in contributing to overall public attitudes toward the physician–patient relationship and overall health care system when it documents violence against doctors events for citizens. Both print and electronic media have a responsibility not to sensationalize the news. Health care workers and officials have cited negative media coverage and sensationalism of violence against doctors events as a prime reason contributing to the deteriorating doctor–patient relationship and mistrust in the medical field [58]. The practice of medicine is highly complex. Diagnosis of a patient is an essentially hypothetico-deductive process, and with new evidence presented through investigations and knowledge, this process continues to be refined. However, whatever the diagnosis may be, patient management generally considers such uncertainties and treatment continues. One of the purposes of this paper has been to analyze whether the government’s policies can affect public opinion on doctor–patient issues, mainly regarding the policies issued by the government on how to report violence against doctors events to influence public opinion. We expect that government policy could intervene in doctor–patient relationships through medical information and media, and our data demonstrate this. However, we do not expect that the decrease in netizens’ blame of doctors and the increase in support for doctors will immediately end the trend of violent acts against doctors. Another purpose of our study was to understand the current trends in the medical network for violence against doctors and hope to find a breakthrough to ease public opinion. Although the media is not the initiator of such medical violence incidents, they are a vital participant in the aftermath, playing the critical role of disseminator of information. Thus, information authenticity and other aspects need to be considered.

In China, the internet environment is managed and controlled by the government, which pays attention to the impact of the network on society. If news media views drive the mood of internet users, then internet users’ irrational speech will increase, also causing more extreme emotions and comments (broken window effect). Media reports tend to play an essential role in driving social norms, which influence people’s behavior and outlooks on topics [57]. To prevent the news of medical violence harming public opinion, network administrators may need to carry out monitoring measures. In our study, we observed numerous emotionally charged comments, which indicate that these news media network comments have not been filtered by management; otherwise, there would be fewer “blame-style” comments. However, measuring the extent of comment monitoring of online news platforms is beyond the scope of our study. When an outlet is provided for netizens to discuss current events, express their views, and vent their emotions, it should guarantee as much free speech as possible, but it is unclear whether the same should apply in terms of medical violence.

China’s intentional homicide has increased in our statistical data years—eg, from 10,285 in 2013 to 17,946 in 2015 [59]. The killing of a doctor is intentional homicide and with the increase in this crime, there has been an increase in the number of doctors killed. However, this does not mean that the relationship between doctors and patients has deteriorated further. The current research focuses on violence against doctors, not just homicides. According to statistics on other similar topics, the incidence of violent medical injuries in China has declined since 2015 [60].

Our study has several limitations. Our scope is limited to the violence against doctors events we selected for the study and to comments referring to those incidents, but we did not cover all types of medical incidents. Furthermore, the sample size, timeline, and number of comments are limited for objective reasons. Hence, we expect further research to be conducted to improve the study of people’s emotional valence in violence against doctors events. In future research, we hope to study the key factors influencing people’s emotions in violence against doctors events through randomized controlled trials. Future studies could also compare violence against doctors events across hospital types, for example, comparing violence against doctors events in private versus public hospitals. Examinations of violence against doctors events across departments within hospitals could also offer information on variations across specialties (eg, obstetrics, pediatrics, and other departments). Regional variations could be another area of inquiry. We seek to provide practical evidence and interventions to reduce the incidence of violence against doctors events and improve the doctor–patient relationship in China.

The causes of the rise and severity of violence against doctors in China are numerous and can be traced back to the systemic roots of inadequate investment in the health care system with subpar training and salaries for physicians, leading to higher chances of medical errors, administrative exploitation, and ineffective communication between patients and medical staff. Other factors may be cultural and societal, such as unfavorable and sensationalized news articles about doctors, the public’s limited knowledge of medicine, unrealistic expectations from doctors and treatments, and detrimental out-of-pocket expenses for patients. As such, the impact of violence against doctors is of great concern and should not be taken lightly for the future of Chinese medicine. Further studies highlighting the role of the government and media should be conducted to improve the physician–patient relationship.

By matching the policy documents for medical violence with the online commentary attitudes, we found that the state’s administrative intervention effectively guided public opinion. Notably, these findings provide scientific evidence and support for the government’s role in intervention and prevention of both cyber and physical violence.

During the COVID-19 pandemic, many countries have been advocating home isolation, leading to a major increase in people’s reliance on the internet for social support and health care information. During the pandemic, a harmonious patient–physician relationship has become an important guarantee for countries in responding to the epidemic. The results of this study show that the government’s reasonable and effective policy guidance for violence against doctors has a significant impact (see Tables 7 and 8 for significant values) on the public’s comments on online media reports, which in turn contribute to the normal medical order as well as epidemic prevention and control.