We collected cases of violation in online drug sales from January 1, 2014, to June 30, 2024. Data sources included the Peking University Law Database, China Judgments Online, the National Medical Products Administration (NMPA) website, provincial/municipal medical product administration websites, and administrative punishment announcements from market supervision bureaus. The following search terms were used: “internet drug” (互联网药品) OR “online drug sales” (药品网络销售) OR “online pharmacy” (网上药店).

A total of 772 violation online drug sales cases were retrieved. After removing duplicates, 648 (83.94%) cases remained. Next, we conducted a detailed review of the cases and further excluded some based on the following exclusion criteria: (1) cases violation unrelated to online drug sales, such as online sales of medical devices, cosmetics, health supplements, or offline physical pharmacy operations; (2) cases where the information was too brief to identify a risk transmission chain; and/or (3) cases with extensive content but vague descriptions of illegal acts, using generalizations instead of process details, making it difficult to determine risks and their interrelationships. Finally, 458 (70.68%) cases of violation in online drug sales were obtained.

Two researchers jointly reviewed the facts, processes, and legal grounds for penalties in the 458 cases, identifying specific risk points for each case. Through discussion, similar items were consolidated and duplicates removed to obtain a unique set of risk factors. Subsequently, the two researchers independently grouped the risk factors based on semantic similarity. After resolving any disagreements through discussion, the factors were preliminarily categorized into several risk categories, and three experts were invited to review and confirm the classification results. Adjustments were made based on the experts’ feedback, resulting in a final risk list and risk categories.

Complex network theory commonly uses network diagrams to study system structures and the interrelationships between structure and system function. The key to complex network modeling lies in defining nodes and edges [21]. Based on the fundamental knowledge of complex network theory and the characteristics of online drug sales, we made the following assumptions:

Based on these two assumptions, risk transmission chain data are converted into an adjacency matrix. Using Gephi software, the number of times connections occur between risks is used as weighting to form a complex network composed of multiple risk transmission chains. Analyzing the network’s indicator parameters can reveal its structural characteristics and evolutionary patterns. Common indicators include average path length, diameter, degree, betweenness centrality, eigenvector centrality, and clustering coefficient [22].

Different metrics in complex networks analyze the effect of nodes from various perspectives. Relying solely on a single metric makes it difficult to fully measure the impact of nodes within a network [28]. To integrate multiple attribute metrics of a single node and thereby assess the effect of individual risks on the final outcome, we used the TOPSIS method to calculate the risk nearness degree C_i. The TOPSIS method is a classical multiattribute decision-making approach that comprehensively considers multiple attribute indicators, avoiding the one-sidedness of single-criterion decision-making. It is widely applied in multiobjective decision-making [29,30] and risk measurement studies [31,32]. The steps in calculation are detailed in Multimedia Appendix 1.

The fundamental approach of the TOPSIS method for risk measurement is as follows: Each risk node within the complex network model for safety risks in online drug sales is treated as a decision scheme in TOPSIS. Four metrics (degree, betweenness centrality, eigenvector centrality, and clustering coefficient) represent the four attributes of each decision scheme. A decision matrix based on the attribute values of each indicator was construct, the distance between each alternative and the ideal solution was measured, and their nearness degree was calculated. The nodes were ranked according to their nearness degree, where a higher degree indicated greater proximity to the positive ideal solution and a stronger influence.

This study did not involve any research on human or animal participants.

The case-screening process is shown in Figure 1. We collected a total of 458 cases of violation in online drug sales (Table 1). Over time, the overall number of cases showed an upward trend, accelerating significantly after 2018. This indicates that as internet penetration and e-commerce platforms develop, the scale of online drug sales continues to expand, and illegal incidents gradually increase. Cases of violation primarily occurred in economically developed or densely populated regions, such as Zhejiang Province, Guangdong Province, Fujian Province, and Beijing Municipality. These areas feature high internet penetration rates and substantial pharmaceutical market demand, making them potential hotspots for violations. The entities that commit violations include not only pharmaceutical trading enterprises (eg, retail pharmacies and wholesale drug distributors) and related individuals (eg, licensed pharmacists) but also enterprises that do not originally engage directly in drug trading, including technology companies, trading companies, and food stores. This underscores the need for enhanced oversight, not only of pharmaceutical businesses and individuals, but also of diverse entities operating in the field of online drug sales.

The importance of risk nodes in the complex network for safety risks in online drug sales is influenced by multiple factors. Different indicators in complex network theory analyze the effect of nodes from different perspectives. Relying solely on a single indicator makes it difficult to fully measure the importance of nodes in the network [28]. Therefore, we adopted the TOPSIS method, effectively integrating four indicators from the complex network model (degree, betweenness centrality, eigenvector centrality, and clustering coefficient) to scientifically and reasonably measure the importance of each risk node in the network. Based on the TOPSIS calculation results, the risk nodes were ranked in descending order of Cᵢ; the top 15 nodes are shown in Table 2, while the complete ranking of all nodes is provided in Multimedia Appendix 3.

In this paper, a higher relative nearness degree value for a node indicated that the node was more important in the complex network model for safety risks in online drug sales and represented a key node in the risk network. As shown in Table 2, the top-ranked risks in the complex network model included F15 (failure to strictly perform prescription review and dispensing duties), F01 (consumers failing to provide prescriptions), and F02 (failure to issue prescriptions through internet hospitals), with relative nearness degrees of 0.5408, 0.4638, and 0.4419, respectively. These risk nodes occupied critical positions within the network, and once manifested, they rapidly propagated to other risk nodes in the network. Among these, F15 (failure to strictly perform prescription review and dispensing duties) exhibited the highest relative nearness degree at 0.5408, making it the most critical risk factor in the complex network for safety risks in online drug sales that requires prioritized prevention and control.

With the rapid development of internet technology, significant changes have occurred in public drug-purchasing patterns. Online drug sales have gradually become a primary channel for the public to purchase drugs due to their convenience and efficiency [34]. However, this has also led to a series of issues and risks, such as the sale of prescription drugs without prescriptions and discrepancies between drug information displayed online and the actual products. Our study introduced complex network theory into the research on safety risks in online drug sales by constructing a network model to analyze risk propagation patterns and using the TOPSIS method to measure risks. The results showed that the key risk nodes in the model are F15 (failure to strictly perform prescription review and dispensing duties), F01 (consumers failing to provide prescriptions), F02 (failure to issue prescriptions through internet hospitals), B12 (failure to establish drug procurement and acceptance records), and H22 (failure to establish drug sales records).

We collected a total of 458 cases of violation. Based on the facts of violations and combined with legal and regulatory requirements, potential risk points in these cases were identified. Ultimately, 116 risks were extracted from the cases and categorized into 8 aspects: business qualifications, drug procurement, transportation and delivery, information display, advertising and promotion, prescription management, pharmaceutical services, and drug sales. Based on complex network theory, a complex network model for safety risks in online drug sales was established, comprising 116 nodes and 152 directed weighted edges.

Analysis of the topological properties of the complex network model revealed that A01 (failure to obtain a drug business license) has the highest total degree, with the maximum risk transmission or reception relationships. Although purchasing medicines from legal online pharmacies has been accepted in higher-income countries, the market has always been plagued by illegal suppliers, posing a significant risk to patient safety [35]. Burrell [36] analyzed the operational structure of illegal online pharmacies from the perspective of cross-border cybercrime, pointing out that it exhibits an adaptive strategy and high profit and low risk characteristics and is deeply intertwined with public health safety risks. Meanwhile, the network’s cumulative degree follows a power-law distribution, exhibiting scale-free characteristics. F15 (failure to strictly perform prescription review and dispensing duties) exhibits the highest betweenness centrality, serving as a crucial bridge in risk transmission. Consistent with our findings, Oriakhi et al [11] investigated the operation of online pharmacies and found that among the 116 online pharmacies evaluated, 47% were illegal, often selling prescription drugs without prescriptions or consultations. H22 (failure to establish drug sales records) has the highest eigenvector centrality, connecting to multiple important risk nodes in the network. Ten risk nodes, including B05 (using others’ information to buy drugs from various hospitals), B07 (procuring counterfeit drugs), and D04 (failure to display qualification information of pharmacists or other pharmaceutical technical personnel), have the highest clustering coefficients, indicating a clustering effect among adjacent risk nodes of these risks.

The TOPSIS method results indicate that in the complex network model, the top-ranked risks include F15 (failure to strictly perform prescription review and dispensing duties), F01 (consumers failing to provide prescriptions), F02 (failure to issue prescriptions through internet hospitals), B12 (failure to establish drug procurement and acceptance records), and H22 (failure to establish drug sales records). These risk nodes occupy crucial positions within the network and can rapidly propagate to other risk nodes. A study conducted in Saudi Arabia on the online purchase of antibiotics found that before selling antibiotics, up to 33.3% of online pharmacies do not require customers to provide a prescription [37]. The study further pointed out that 45% of these websites are fake or only sell local antibiotics [37]. Ahmed et al [38] conducted a systematic review on the issue of substandard and counterfeit drugs sold online, identifying challenges, such as the lack of knowledge and awareness among consumers and doctors, as well as the use of internet platforms and social media for direct-to-consumer advertising. Furthermore, Xiao [39] conducted a systematic analysis of the regulatory policies for online pharmacies in China and reported that the revision of the Drug Administration Law in 2019 was the first national-level legal framework for regulating online pharmacies in China. However, as artificial intelligence (AI)–driven online pharmacies are rapidly emerging, issues such as the responsibility boundaries of online platforms and the legal status of AI doctors and pharmacists still face legal and ethical controversies. These new business models will further exacerbate the complexity of safety risks in online drug sales. Digital health interventions, training programs for medical service providers, and public awareness campaigns tailored to specific situations are key approaches to bridging the gap between cognition and practice [40]. To mitigate the aforementioned risks, regulatory authorities should popularize the knowledge of prescription drugs to the public through various channels and media, including definitions, online purchasing procedures, and usage risks, thereby guiding the public to voluntarily purchase prescription drugs with prescriptions. Online pharmacies engaged in prescription drug sales should regularly organize licensed pharmacists to participate in relevant training or continuing education courses to enhance their sense of responsibility and professional competence. Furthermore, standardized and unified electronic prescription review processes should be established to ensure the professionalism and compliance of online prescription reviews. In addition, regulatory authorities should conduct regular inspections of online drug operating enterprises, focusing on whether they have established complete and accurate drug procurement, acceptance, and sales records, as required, and whether these records enable drug traceability.

This study quantitatively measured risks in online drug sales and identified key risks, providing scientific guidance for online drug sales regulation and helping enhance the risk management capabilities of regulatory authorities. Despite these contributions, this study has some limitations.

First, because the extraction of risk nodes was performed manually, coding bias may exist in the process of translating the contents of cases of violation into risk nodes due to subjective judgment. To mitigate this potential bias, the following measures were taken: two researchers independently coded the cases, followed by cross‑checking and discussion; three experts were invited to review and validate the risk classification results; and final adjustments were made based on expert feedback. Nevertheless, completely eliminating coding bias remains difficult. Future research could use automated text analysis methods for cross‑validation to further reduce coding bias.

In addition, the online drug sales process involves numerous uncertain risks, and risk evolution patterns are relatively complex. Future research should continuously supplement violation case data and use multiple methods to deeply explore the patterns of risk propagation and key risk nodes in online drug sales, such as infectious disease models and node simulation attacks.

As an emerging drug procurement method, online drug sales provide significant convenience to consumers, while also involving numerous risks. Based on the collected cases of violation, this study used complex network theory to construct a complex network model for safety risks in online drug sales. Using network indicators and the TOPSIS method, we measured the safety risks in online drug sales and identified key risks. The results showed that the key risk nodes in the complex network model include F15 (failure to strictly perform prescription review and dispensing duties), F01 (consumers failing to provide prescriptions), F02 (failure to issue prescriptions through internet hospitals), B12 (failure to establish drug procurement and acceptance records), and H22 (failure to establish drug sales records). Therefore, targeted risk prevention and control measures should be developed for these key risks to promote the healthy and orderly development of the online drug sales industry.