The research object is emergency patients in a large-scale tertiary hospital, which has 4,300 beds and receives approximately 240,000 emergency patients annually.

A total of 105 medical conflict cases were collected, of which 78 occurred in the daytime, and 27 occurred at night. Twenty-one conflict cases occurred on weekends, and 84 on weekdays. There were 56 conflict cases in the pre-examination and triage areas, 41 in the resuscitation rooms, and eight in the consulting rooms. Regarding the medical professionals, 80 conflict cases involved nurses and 16 involved doctors. The attackers were caregivers in 67 cases, patients in 31 cases, and both caregivers and patients in 7 cases. There were 49 cases with female attackers, 53 with male attackers, and 3 with both male and female attackers. See Table 1 for a summary of the conflict events.

This study included 105 patients with conflict events, including 58 males and 47 females. See Table 2 for a summary of the patients’ information.

A total of 105 conflict cases and 525 non-conflict cases were included in this study. Univariate analysis showed that there was no significant between-group difference in the gender of patients, way of payment, type of conditions, whether there were symptoms that might lead to harming or misunderstanding others, gender of caregivers, and whether there were precipitants (p > 0.05). However, statistically significant differences were found in waiting time, age, stage of the medical process, way of admission, triage level, history of psychosomatic diseases, caregivers (number of caregivers), and doctor-patient interaction (medical professionals’ failure to respond to patients’ needs timely due to heavy workload; patient skepticism toward treatment and care; patient non-compliance to diagnosis and treatment) (p < 0.05). See Table 3 for a summary of the results.

The final neural network model utilized the hyperbolic tangent as the activation function for the hidden output layers and the identity function for the output layer. Training was performed using two hidden layers, the first comprising five units and the second four units. The characteristics of the NN model are shown in ESF, including parameter estimates (Supplementary Table 1), neural network diagram (Supplementary Figure 1), model summary (Supplementary Table 2), and predicted versus observed chart (Supplementary Figure 2). During the training phase, the sum of squares error term was reduced from 14.425 to 4.626, indicating improved trend generalization by the model. Moreover, the model seems to have effectively minimized overfitting, as evidenced by the consistent percentage of incorrect predictions throughout training (5.2%), testing (3.2%), and holdout (2.7%) samples. These results indicate consistent, but not perfect, discrimination between the two groups. Figure 1 illustrates the importance chart, highlighting the significance of the key input variables. The model primarily recognized waiting time; patient non-compliance to the treatment process; patients’ skepticism, number of caregivers, medical professionals’ failure to respond to patients’ needs, and history of psychosomatic diseases as the most important predictors, followed at a distance by patient skepticism toward the cost; history of psychosomatic diseases; and number of caregivers. The cross-validated precision of the model (using a holdout sample) showed a predictive accuracy of 97.3%, with a sensitivity of 94.7% and a specificity of 97.7%. Figure 2 shows that the AUC of the ROC curve was 0.966.

The SVM model detected 76 support vectors, comprising 36 from the no-risk category and 40 from the risk category. The optimized model parameters were C = 10.0 and γ = 0.11111, employing an RBF kernel. The resulting decision function constant was 3.126763, signifying stable separation between classes. Model training attained an aggregate accuracy of 94.9%. In the independent test set, the model achieved an accuracy of 96.4%. The classification performance on the test set demonstrated a sensitivity of 91.2% and a specificity of 97.7%.

The results of the neural network analysis showed the importance of risk factors in predicting possible conflict: waiting time; patient non-compliance to the treatment process; patients’ skepticism, number of caregivers, medical professionals’ failure to respond to patients’ needs, and history of psychosomatic diseases.

Waiting time refers to the period a patient waits before their consultation. Waiting time was identified as the secondary risk factor. Prolonged waiting in emergency departments frequently triggers global dissatisfaction and conflicts among patients and their families. Our research results once again confirm this. Our results are similar to those reported by Vezyridis (16), who noted that patients in the emergency department considered their conditions urgent and believed they should receive comprehensive emergency treatment immediately. A longer waiting time, more severe conditions, and higher patient urgency for medical treatment increase the likelihood that the patient will be dissatisfied with the diagnosis and treatment process, leading to conflicts with medical professionals.

The second most important risk factor is “patient non-compliance to the diagnosis and treatment process”. It is related to two facts. First, in some hospitals, payment procedures, tests and examinations, result communication, and follow-ups are cumbersome, leading to patient drop-out and even conflict in emergent situations. Second, patients in the emergency department have severe conditions, and the pain makes them hope to get immediate treatment from medical professionals; when registration, triage, payment, or any other process does not go smoothly, they consider the medical process complicated and thus show poor compliance toward the diagnosis and treatment process, resulting in conflict, patients have high expectations of medical technology, while effective communication between doctors and patients is lacking (2). The verbal and nonverbal actions embodied in the doctor-patient interaction form the basis for establishing a harmonious relationship between the doctor and patient.

This study identified “patients’ noncompliance toward diagnosis, treatment, and care as the significant risk factor. This finding suggests that patients’ distrust and negative perceptions constitute the primary catalyst for conflicts. In the high-pressure setting of the emergency department, inadequate communication or a failure to meet patient expectations can readily trigger an eruption of such distrust. The reason may be that patients lack sufficient knowledge of emergency care, assuming they should receive immediate care and treatment in the emergency department. In fact, about 50% of the patients in the emergency department of a general hospital are non-emergency patients. Additionally, some patients or caregivers may not fully understand emergency treatment and hold overly high expectations for its effects. Consequently, patients may question the diagnosis, treatment, or care measures, which can potentially result in conflict.

Our study also indicated that a higher number of caregivers is related to a higher probability of conflict. It is due to caregivers’ concern about the patient’s condition. However, it could also result in an increased number of disputes and collective emotional outbursts, thereby elevating management challenges and the risk of conflict. A Chinese proverb states, “There is safety in numbers.” The caregivers will feel “impeccable” if there are many of them, and even slight dissatisfaction may make them prone to violent behaviors.

Our study also identified “medical professionals failure to respond to patients’ needs” as another significant contributor to conflicts. The emergency department is always busy with a heavy workload and high pressure. In a state of long-term mental tension, medical professionals may be irritable and impatient during work. They may sometimes fail to meet patients’ needs promptly, resulting in conflict.

In addition, patients with a history of psychosomatic diseases (of any level) have a potential risk of conflict with others and are prone to anxiety and irritability. Therefore, the history of psychosomatic diseases is one of the critical factors leading to medical conflict. Patients with this medical history may exhibit heightened anxiety and sensitivity to physical symptoms or pre-existing difficulties with emotional regulation, thereby increasing their susceptibility to conflicts.

Convergent performance of both neural network and SVM in independent holdout and testing samples (accuracy around 96–97%) suggests that our classification results may be reproducible. The prediction model can play a crucial role in the early detection of medical conflicts in emergency departments, enabling medical professionals to identify patients at risk of conflict and implement preventive measures and interventions in advance, thereby reducing the likelihood of medical conflicts. It has significant value for clinical practice.

In the emergency department, patients’ conditions are urgent, complex, and diverse, and patients and their families have an urgent need for medical treatment, making it prone to medical conflict. The study by Yang reported that the incidence of conflict in surgical emergencies was as high as 19%, and emergency services workers run the greatest risk of violence (17). The medical conflict significantly negatively impacts patients, family members, medical professionals, and hospitals. It disturbs hospitals’ regular medical services, damages medical professionals’ social reputation and image, causes significant psychological pressure to the professionals, and may even lead to violence. It harms the healthy development of the medical and healthcare industry and is not conducive to a harmonious society. Research has shown that acquiring relevant knowledge can prevent medical conflict (18). According to previous studies, the risk factors affecting the doctor-patient relationship primarily include a misunderstanding of medical behavior, overly high expectations for treatment, and inadequate or insufficient medical knowledge. For patients, prominent factors included the misunderstanding of medical behavior and high expectations for prognosis (10).

This study not only identified the key risk factors for doctor-patient conflicts but also ranked their relative importance. The findings clearly indicate that “soft” communication and psychological factors (e.g., distrust and perceived neglect) are more critical predictors of conflict than “hard” clinical factors (e.g., a history of physical or mental disorders). Consequently, for emergency care, targeted efforts should focus on improving communication, managing patient expectations, mitigating the perception of prolonged waiting, and training healthcare staff in effective emotional de-escalation techniques.

Establishing a prediction model for medical conflict in emergency departments is conducive to preventing medical conflict, improving the work environment of medical professionals, and safeguarding the legitimate rights and interests of both doctors and patients. This study included 630 patients in the emergency department. We constructed a prediction model by screening the risk factors of medical conflict through neural network analyses. The prediction model for medical conflict in emergency departments established in this study demonstrated good predictive performance in internal validation, and a good cross-validated predictive accuracy in an independent holdout sample. Based on this model, emergency medical professionals can identify the risk factors and prevent conflict by accurately judging the patient’s condition classification, minimizing their waiting time, and giving appropriate rescue measures. In addition, they should pay attention to the emotional status of patients with a history of psychosomatic diseases, strengthen communication with patients and caregivers, and make efforts to meet patients’ reasonable needs as much as possible and promptly. They can also provide a detailed explanation of the treatment methods, care measures, and medical process to patients and caregivers and appropriately limit the number of caregivers, thereby reducing the incidence of medical conflict.