Development and pilot evaluation of an electronic trigger-based surveillance model for sedative-hypnotic drug misuse in a tertiary hospital Authors: Jianxi Liu, Ning Du, Xueyan Liu, Wenxiu Cao, Quan Zhao, Yinglin Wang Published: 2026 ABSTRACT Sedative-hypnotic drugs (SHDs) carry significant risks of misuse and dependence. However, current institutional pharmacovigilance relies heavily on passive reporting, leading to substantial underreporting and data gaps. This study aimed to develop and conduct a pilot evaluation of an electronic trigger-based active surveillance model integrated into a Hospital Information System (HIS) to proactively identify high-risk prescription patterns and suspected SHD dependence. This two-phase study was conducted at a tertiary hospital. First, a retrospective analysis of 16,609 outpatient prescriptions was performed to characterize anomalous utilization. Second, a “Trigger Dictionary” based on the Global Trigger Tool (GTT) was developed to flag suspicious cases for a 10-day prospective pilot evaluation. Patients with suspected dependence were screened through structured telephone interviews using ICD-10 diagnostic criteria among flagged patients with ≥ 1-year SHD use. Retrospective analysis revealed that 30.03% of patients (2,616/8,710) received at least one off-label prescription. It also identified 1,567 clusters in which multiple patient IDs shared identical residential addresses, serving as a potential, hypothesis-generating signal for drug diversion. During the prospective pilot, the system generated 200 automated alerts. Among the 63 high-risk patients interviewed, 17 were identified as meeting the screening criteria for suspected SHD dependence, yielding a Positive Predictive Value (PPV) of 26.98% within this selected high-risk interviewed cohort. In contrast, zero cases were reported via the passive system during the same period. Univariate analysis identified benzodiazepine use (OR 33.07, 95% CI: 3.95–276.81, P < 0.001), advanced age, and prolonged therapy duration as exploratory factors associated with suspected misuse. This 10-day prospective pilot evaluation suggests that the electronic trigger model can identify cases of suspected drug dependence that were not captured by traditional passive reporting during the same period. This automated framework offers a scalable approach for clinical pharmacists to triage high-risk patients and mitigate potential drug diversion risks, potentially enhancing medication safety in high-volume outpatient settings.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-63516-w. Introduction Sedative-hypnotic drugs (SHDs), including benzodiazepines (BZDs) and Z-drugs (non-benzodiazepines, nBZDs), are among the most frequently prescribed classes of medication globally1,2. They play a critical role in clinical practice for the management of insomnia, anxiety, and various neurological disorders3,4. However, their pharmacological profile—characterized by rapid onset of tolerance, high potential for psychological and physical dependence, and an increased risk of cognitive impairment—poses significant public health challenges5–8. The rising incidence of insomnia worldwide has led to an increase in SHD prescriptions, often resulting in prolonged treatment durations that deviate significantly from established clinical guidelines9–14. Suboptimal prescribing patterns, such as chronic long-term use and polypharmacy, have been increasingly linked to adverse clinical outcomes, including medication dependence, accidental falls, cognitive decline in the elderly, and, in severe cases, drug-related mortality and suicide15–18. Despite these risks, the current pharmacovigilance framework for SHDs remains insufficient. Drug misuse monitoring continues to rely heavily on passive, voluntary reporting systems. Such systems are inherently reactive, often resulting in substantial underreporting due to administrative burdens, clinical hesitation, and a lack of real-time clinical integration19–24. Furthermore, clinicians often fail to recognize sedative-hypnotic misuse as a reportable adverse event because its onset is insidious, and withdrawal or tolerance symptoms are frequently conflated with the patient’s underlying psychiatric conditions, such as insomnia or anxiety8. Consequently, a considerable time lag may occur between the emergence of drug misuse signals and regulatory or institutional intervention25,26. Indeed, a recent global bibliometric analysis mapping the evolution of sedative-hypnotic research over the past 15 years has confirmed a major paradigm shift, transitioning from retrospective molecular pharmacology to proactive, database-driven digital and computational surveillance27. Medical institutions (specifically tertiary hospitals and affiliated clinical pharmacies), which serve as the primary nodes for SHD distribution, currently lack proactive, automated tools capable of identifying patients at risk of dependence within the daily clinical workflow. While the Global Trigger Tool (GTT) methodology has been widely pilot-tested for detecting hospital-wide adverse events and medical errors, its application in the dynamic, high-volume environment of outpatient SHD monitoring remains largely unexplored, leaving a critical gap in institutional drug safety programs26,28–30. To address this significant unmet need in pharmaceutical care, we developed and conducted a pilot evaluation of an electronic trigger-based active surveillance system integrated within our hospital’s Health Information System (HIS). The HIS is a comprehensive centralized database integrating electronic medical records (EMRs), outpatient prescriptions, and patient demographic profiles, which provides a rich repository for real-time clinical data mining. Our study aimed to: (1) analyze the prevalence of potentially inappropriate SHD prescription patterns, such as off-label use and indicators of potential doctor-shopping (defined as patients seeking redundant prescriptions from multiple physicians or clinical departments concurrently); (2) construct a predictive electronic trigger model based on clinical keywords and high-risk prescription behaviors; and (3) conduct a pilot evaluation of the clinical utility of this model through structured follow-up interviews to identify screen-positive cases for suspected SHD dependence. By demonstrating that an automated, data-driven approach can complement existing surveillance methods, this study provides a scalable framework for clinical pharmacists to triage high-risk patients efficiently, thereby optimizing SHD use and institutional drug safety management. Methods Study design and data source This study employed a two-phase design: a retrospective cross-sectional analysis of prescription patterns, followed by the prospective development and pilot-testing of an active surveillance system for drug misuse. The study was conducted at Yantai Yuhuangding Hospital, which is a large tertiary Grade-A general hospital and Shandong Regional Medical Center with approximately 4,000 authorized beds and over 50 clinical departments. This study evaluated outpatient prescriptions exclusively. Data were extracted from the hospital’s Hospital Information System (HIS). To protect patient privacy, all data were anonymized and assigned unique patient identifiers (IDs) prior to analysis. Study population and data processing For the retrospective prescription analysis, we extracted all outpatient prescriptions containing sedative-hypnotic drugs (SHDs) issued between January 1, 2023, and June 30, 2023. We included four types of benzodiazepines (BZDs: alprazolam [ATC: N05BA12], clonazepam [ATC: N03AE01], diazepam [ATC: N05BA01], estazolam [ATC: N05CD04]) and two types of non-benzodiazepines (nBZDs: zopiclone [ATC: N05CF01], zolpidem [ATC: N05CF02]). To ensure analytical precision, the extracted data were evaluated at both the prescription level and the patient level. Prescription-level analysis was used to quantify total prescription counts, drug-class distribution, and off-label rates. Patient-level analysis was performed by consolidating these prescription records using unique anonymized patient IDs to evaluate patient demographics, polypharmacy, and address-sharing behaviors. Definition of variables and clinical indicators To evaluate SHD utilization and identify potential misuse, several key indicators were defined: Clinical Diagnoses and Off-label Use: Patient diagnoses were standardized using the International Classification of Diseases, 10th Revision (ICD-10). We evaluated seven clinical indications: anxiety, sedation, hypnosis, epilepsy, anticonvulsion, panic disorder, and alcohol withdrawal. A prescription was defined as “off-label” if the patient’s documented ICD-10 diagnoses did not match any of the approved indications specified by the National Medical Products Administration (NMPA) for that specific drug (detailed in Table S1). Polypharmacy: Defined as a single patient being prescribed two or more different types of SHDs (concurrently or sequentially) during the study period. High-Frequency Visits: Defined as a patient receiving SHD prescriptions during six or more separate outpatient visits within the 6-month study period, indicating sustained or chronic use. Multiple IDs per Address: Defined as different patient IDs registering the exact same residential address to obtain SHDs, a potential signal for drug diversion or “doctor shopping” using family members’ medical cards. Development of the active surveillance system Traditional drug misuse monitoring relies heavily on passive, voluntary reporting, which is subject to underreporting. To address this, we developed an active surveillance model based on the Global Trigger Tool (GTT) methodology. The GTT retrospectively or prospectively identifies adverse drug events using clinical markers (‘triggers’) to scan medical records. We adapted this methodology because of its proven efficiency in converting massive electronic health data into manageable, high-yield clinical alerts, which we structured into five quantitative risk identification rules (see Supplementary Table S2). By analyzing historical electronic medical records (EMRs) and previous drug misuse reports, we established a “Trigger Dictionary.” The triggers included specific drug names, clinical symptoms (e.g., insomnia, anxiety, headache, nausea), and high-risk prescription behaviors (e.g., remaining prescription days > 7, ≥4 prescriptions per month, annual cumulative use > 180 days) (Table S2). The system was designed to automatically scan the HIS and flag suspicious cases for further pharmacist review. Within the system’s surveillance logic, all trigger keywords and quantitative rules listed in Supplementary Table S2 carried equal weight; meeting any single trigger condition (e.g., a cumulative annual use > 180 days, or a patient ID accessing different departments for SHDs) was sufficient to generate an automated alert (OR-gate logic). No cumulative scoring system was applied. To minimize alert fatigue, duplicate alerts generated for the same patient within the pilot period were automatically deduplicated and consolidated by their unique anonymized patient ID. Pharmacists prioritized these consolidated alerts for clinical follow-up based on two criteria: (1) a continuous SHD prescription history exceeding one year (as detailed in Sect. 2.5), and (2) the presence of multiple concurrent triggers, such as an anomalous prescription trigger combined with a patient registration trigger (same address with multiple IDs). Pilot evaluation via structured telephone interviews To evaluate the positive predictive value (PPV) of the electronic triggers within the selected high-risk cohort during this pilot phase, prospective pilot surveillance was conducted in the psychological counseling clinic from October 24 to November 2, 2023. Patients flagged by the system who had a history of SHD use for ≥ 1 year were selected for follow-up. We developed a 10-item structured interview outline based on the ICD-10 diagnostic criteria for “Drug Dependence Syndrome” (Table S3). After obtaining informed consent, telephone interviews were conducted by licensed clinical pharmacists under the supervision of senior clinical pharmacists. To ensure clinical and terminological consistency, we strictly defined and separated our study terms as follows: (1) ‘high-risk prescription patterns’ refer to outpatient records that triggered any of our quantitative HIS rules (Table S2); (2) ‘SHD misuse’ is defined as the use of SHDs outside approved indications, in excessive doses, or without proper medical supervision; and (3) ‘suspected SHD dependence’ refers to patients meeting the screening proxy criteria of our structured interview. Because face-to-face psychiatric consultation was not feasible in this pilot study, these telephone interviews served as a screening proxy for suspected SHD dependence rather than a formal, clinical psychiatric diagnosis. A patient was classified as having suspected SHD dependence or misuse if they answered “Yes” to at least three of four core questions evaluating tolerance, withdrawal symptoms, loss of control, and psychological craving (Questions Q5, Q6, Q7, and Q8 in Supplementary Table S3, mapped to the ICD-10 diagnostic criteria for drug dependence syndrome). Statistical analysis Continuous variables were presented as medians with interquartile ranges (IQR), and categorical variables were presented as counts and percentages. Wilcoxon rank-sum tests were used to compare the distribution of continuous variables between groups. Fisher’s exact test was employed for categorical variables due to the limited sample size in the pilot evaluation cohort. A two-tailed P-value < 0.05 was considered statistically significant. Results Study population and baseline characteristics The patient selection process and the progression from retrospective analysis to prospective pilot evaluation are detailed in Fig. 1. Between January 1, 2023, and June 30, 2023, a total of 775,700 outpatient prescriptions were issued across all clinical departments in Yantai Yuhuangding Hospital. We identified 16,609 outpatient prescriptions for sedative-hypnotic drugs (SHDs) (representing 2.14% of the total hospital outpatient prescriptions during this period), involving 8,710 unique patients. Prescribing activities were geographically dispersed across more than 50 clinical departments (reflecting the multidisciplinary use of SHDs to manage sleep disturbances or secondary anxiety in patients admitted to non-psychiatric departments such as oncology, cardiology, and gastroenterology); however, the volume was disproportionately concentrated in the Psychological Counseling Clinic (18.26%, n = 3,032), Neurology (17.99%, n = 2,988), Emergency Department (6.76%, n = 1,122), Cardiology (4.63%, n = 769), and the Traditional Chinese Medicine Outpatient Clinic (4.60%, n = 764) (Figure S1). Fig. 1Flowchart of study population selection and active surveillance system pilot evaluation. Flowchart of study population selection and active surveillance system pilot evaluation. Table 1Demographic and Clinical Characteristics of the Overall Study Cohort (N = 8,710).CharacteristicsTotal Patients, n (%)Demographic Characteristics (Patient-Level)GenderMale3,471 (39.85%)Female5,239 (60.15%)Age groups (years)≤ 18257 (2.95%)19–441,984 (22.78%)45–643,209 (36.84%)65–741,939 (22.26%)≥ 751,321 (15.17%)Anomalous Prescription Patterns (Aggregated at Patient-Level)Polypharmacy (≥ 2 SHDs)821 (9.43%)High frequency visits (≥ 6 times)346 (3.97%)Off-label use (at least one off-label prescription)2,616 (30.03%) Table 1Demographic and Clinical Characteristics of the Overall Study Cohort (N = 8,710).CharacteristicsTotal Patients, n (%)Demographic Characteristics (Patient-Level)GenderMale3,471 (39.85%)Female5,239 (60.15%)Age groups (years)≤ 18257 (2.95%)19–441,984 (22.78%)45–643,209 (36.84%)65–741,939 (22.26%)≥ 751,321 (15.17%)Anomalous Prescription Patterns (Aggregated at Patient-Level)Polypharmacy (≥ 2 SHDs)821 (9.43%)High frequency visits (≥ 6 times)346 (3.97%)Off-label use (at least one off-label prescription)2,616 (30.03%) Demographic and Clinical Characteristics of the Overall Study Cohort (N = 8,710). Demographic and clinical data (summarized in Table 1) revealed that the cohort had a median age of 43.5 years, with the largest proportion (36.84%) aged between 45 and 64 years. Women constituted 60.15% of the study population. Benzodiazepines (BZDs) were the most commonly prescribed class (53.12%, n = 8,822). Notably, 30.03% of patients (2,616/8,710) had at least one SHD prescription classified as off-label, with clonazepam demonstrating an exceptionally high off-label prescribing rate of 93.18%. Specifically, our clinical database revealed that clonazepam was most frequently prescribed off-label for anxiety-depressive states (n = 966) and sleep disorders (n = 646), which are off-label in China because its NMPA-approved indication is strictly limited to epilepsy (Table S1). Similarly, zopiclone and zolpidem (which are NMPA-approved strictly for the treatment of insomnia/hypnosis) were frequently flagged as off-label when prescribed under the clinical diagnoses of anxiety-depressive states (n = 1,615 and n = 142, respectively) or neurasthenia (n = 124 and n = 70, respectively) without a co-documented diagnostic code for insomnia. Identification of anomalous prescription patterns To develop the electronic trigger model for active surveillance, we analyzed prescription behaviors to identify potential misuse signals. First, continuous heavy utilization was observed, with 3.97% of patients (n = 346) receiving SHD prescriptions for 6 or more visits during the 6-month study period. Second, polypharmacy of SHDs was identified in 9.43% of patients (n = 821) who were concurrently or sequentially prescribed two or more different SHDs. Fig. 2Bar chart of address distribution / Multiple patient identifiers (IDs) registered under identical residential addresses. Bar chart of address distribution / Multiple patient identifiers (IDs) registered under identical residential addresses. In addition to clinical risk indicators, we observed a frequent occurrence of multiple patient IDs registered under identical residential addresses (1,567 groups). As illustrated in Fig. 2, the distribution of these IDs reveals that a substantial number of residential addresses are associated with multiple patient identifiers. While this pattern is suggestive of potential drug diversion or doctor-shopping, it should be interpreted as a hypothesis-generating signal derived from administrative data, warranting further investigation with privacy-compliant methods in future prospective studies. Performance and pilot evaluation of the active surveillance system Based on clinical keywords and the anomalous patterns identified above, an electronic trigger-based active surveillance system was deployed. During a 10-day prospective pilot phase (October 24 to November 2, 2023) in the top prescribing clinics, the system automatically screened all incoming records and flagged 200 suspicious cases of drug misuse/dependence (averaging 20 alerts per day). This automated filtering significantly reduced the manual review burden for clinical pharmacists, as reviewing these 20 flagged alerts required less than 1 h per day, whereas manually screening all daily outpatient records in high-volume clinics would typically require several hours of continuous chart review. For model pilot evaluation, we excluded patients with an SHD prescription history of less than one year, yielding a pilot evaluation sample of 63 highly suspicious cases for structured telephone interviews. Of the 63 patients successfully interviewed, 17 cases met the structured interview screening criteria for suspected drug dependence/misuse. The overall Positive Predictive Value (PPV) of the surveillance triggers within this selected high-risk interviewed cohort was 26.98% (17/63). Among these screen-positive cases, BZDs were the primary misused drugs, specifically clonazepam (n = 8), alprazolam (n = 6), and diazepam (n = 2), while only 1 case involved an nBZD (zopiclone). Univariate analysis of factors associated with suspected SHD misuse To further understand the factors associated with suspected SHD misuse and to refine future surveillance weights, a univariate analysis was performed comparing the suspected misuse group (n = 17) with the non-misuse group (n = 46) within our pilot evaluation cohort (Table 2). Due to the limited pilot evaluation sample size and the quasi-complete separation of data, exact and non-parametric tests were employed. Table 2Characteristics and Univariate Analysis of the Pilot Evaluation Cohort (N = 63).VariablesTotal(N = 63)Misuse Group(n = 17)Non-Misuse Group(n = 46)Odds Ratio(95% CI)P-valueAge (years), Median [IQR]46.0 [27.0–60.0]69.0 [62.0–72.0]43.0 [27.0–56.0]N/A*< 0.001Gender, n (%)0.63 (0.18–2.15)0.548Male21 (33.3%)7 (41.2%)14 (30.4%)Female42 (66.7%)10 (58.8%)32 (69.6%)Drug Class, n (%)33.07 (3.95–276.81)< 0.001BZDs31 (49.2%)16 (94.1%)15 (32.6%)nBZDs32 (50.8%)1 (5.9%)31 (67.4%)Duration of Use (years), Median [IQR]0.5 [0.5-3.0]6.0 [4.0–10.0]0.5 [0.5–0.5]N/A*< 0.001*Note: Odds Ratios (OR) were calculated using Fisher’s exact test for categorical variables. The Wilcoxon rank-sum test was used for continuous variables (Age and Duration). Table 2Characteristics and Univariate Analysis of the Pilot Evaluation Cohort (N = 63).VariablesTotal(N = 63)Misuse Group(n = 17)Non-Misuse Group(n = 46)Odds Ratio(95% CI)P-valueAge (years), Median [IQR]46.0 [27.0–60.0]69.0 [62.0–72.0]43.0 [27.0–56.0]N/A*< 0.001Gender, n (%)0.63 (0.18–2.15)0.548Male21 (33.3%)7 (41.2%)14 (30.4%)Female42 (66.7%)10 (58.8%)32 (69.6%)Drug Class, n (%)33.07 (3.95–276.81)< 0.001BZDs31 (49.2%)16 (94.1%)15 (32.6%)nBZDs32 (50.8%)1 (5.9%)31 (67.4%)Duration of Use (years), Median [IQR]0.5 [0.5-3.0]6.0 [4.0–10.0]0.5 [0.5–0.5]N/A*< 0.001*Note: Odds Ratios (OR) were calculated using Fisher’s exact test for categorical variables. The Wilcoxon rank-sum test was used for continuous variables (Age and Duration). Characteristics and Univariate Analysis of the Pilot Evaluation Cohort (N = 63). *Note: Odds Ratios (OR) were calculated using Fisher’s exact test for categorical variables. The Wilcoxon rank-sum test was used for continuous variables (Age and Duration). Statistical analysis revealed strong associations for SHD misuse. The duration of continuous use was identified as a critical associated factor; patients in the misuse group had a significantly longer duration of SHD prescriptions (median: 6.0 years) compared to the non-misuse group (median: 0.5 years) (Wilcoxon rank-sum test, p < 0.001). Similarly, patient age was significantly higher in the misuse cohort (p < 0.001). Regarding the pharmacological class, the misuse of BZDs was strikingly more prevalent than nBZDs. Among the 17 screen-positive cases, 94.1% (16/17) involved BZDs. Fisher’s exact test revealed a highly significant association between BZD prescription and suspected misuse (Odds Ratio [OR] = 33.07, 95% CI: 3.95–276.81, p < 0.001). In contrast, patient gender did not show a statistically significant association with SHD misuse (p = 0.548). These findings support the integration of specific clinical parameters—particularly prolonged duration and the BZD drug class—into our automated surveillance algorithm to efficiently triage high-risk patients. Discussion This study demonstrates the feasibility and clinical value of transitioning from a passive to an active, trigger-based pharmacovigilance model for SHDs in outpatient settings. By mining real-world EHR data, our surveillance algorithm effectively identified anomalous prescribing patterns and captured hidden cases of suspected drug dependence. The value of active surveillance Currently, pharmacovigilance for SHDs heavily relies on spontaneous, passive reporting systems, which notoriously suffer from underreporting and latency. Remarkably, during our 10-day prospective pilot phase, the spontaneous reporting rate for SHD dependence from the involved departments was 0%, whereas our active trigger-based system achieved a detection rate of 8.5% (17 screen-positive cases out of 200 alerts) among high-risk flagged records. While this stark contrast demonstrates the practical utility of our active surveillance model, it should be interpreted as supportive evidence of its screening efficiency in clinical workflows rather than absolute proof of superior diagnostic performance. Spontaneous passive reporting is highly dependent on clinician awareness, busy outpatient clinical routines, and institutional reporting culture, whereas the trigger model utilizes automated electronic screening to actively triage high-risk records. This observation highlights the potential of proactive, EHR-integrated surveillance to complement existing monitoring approaches and help address gaps in current drug misuse surveillance networks. While the system’s Positive Predictive Value (PPV) within the selected high-risk interviewed cohort was 26.98% (17/63), approximately three out of four flagged high-risk patients did not meet the screening criteria for suspected dependence during follow-up. Nonetheless, this level of precision is highly acceptable for preliminary clinical pharmacovigilance screening. In a clinical landscape where spontaneous passive reporting yielded zero cases during the same period, a screening tool with a 26.98% PPV represents a substantial diagnostic advancement. In terms of clinical workload, the system generated 200 alerts over 10 days, averaging 20 alerts per day. Reviewing these 20 automated alerts is highly manageable for clinical pharmacists, typically requiring less than 1 h of rapid electronic medical record (EMR) screening per day. Pharmacists can easily filter out clinical false alerts through quick electronic reviews before deciding which patients require targeted telephone follow-ups, demonstrating that the model’s workload is sustainable for routine hospital medication safety programs. Furthermore, these validated alerts directly trigger targeted patient-care interventions: clinical pharmacists conduct standardized screening interviews (Supplementary Table S3), deliver patient education on SHD risks, collaborate with prescribing physicians to design gradual drug tapering protocols, and refer high-risk patients to psychological counseling clinics. This closed-loop clinical workflow successfully translates automated HIS alerts into actionable, patient-level care. Pharmacological disparities: BZDs vs. nBZDs Our univariate analysis revealed an association between BZD use and a higher rate of meeting dependence criteria compared to nBZDs. This aligns with the pharmacological consensus that while both classes act on GABA-A receptors, BZDs typically exhibit broader receptor affinity and longer half-lives. Prolonged use of BZDs significantly increases the risk of tolerance and withdrawal symptoms upon cessation. In contrast, Z-drugs (nBZDs) generally demonstrate a lower propensity for physical addiction, although they still require monitoring. Consequently, clinical guidelines advocate for strictly limiting the duration of BZD prescriptions—a principle that our system aims to enforce algorithmically. Detecting drug diversion via socio-demographic triggers An observation from our retrospective analysis was the identification of anomalous patient registration patterns, specifically 1,567 groups of multiple IDs sharing identical residential addresses. In the context of stringent outpatient prescription limits for psychotropic substances in China, this pattern may indicate possible doctor-shopping or drug diversion—where patients use family members’ medical cards to accumulate quantities of SHDs. Incorporating such socio-demographic indicators into clinical surveillance algorithms offers a promising direction for providing hospital administrators with additional tools to help mitigate potential drug diversion. Furthermore, under China’s current regulatory framework, benzodiazepines and Z-drugs are strictly classified as Class II psychotropic substances, which cannot be sold online and face extremely stringent registry and prescription audits in community pharmacies. Consequently, patients almost exclusively obtain these controlled medications through tertiary hospitals like ours, making hospital-centric outpatient databases the primary and highly representative data source for capturing sedative-hypnotic misuse. Limitations This study has several limitations. First, the ‘multiple IDs per address’ metric is a surrogate signal and may capture legitimate cohabitation cases where multiple family members require SHD therapy. However, in China’s clinical context, patients also use family members’ registration cards to obtain additional medications to bypass strict single-prescription dosage limits (household-level card sharing). Therefore, address sharing can indicate either potential indirect diversion or legitimate household co-prescribing, which requires future clinical verification. Also, this was a single-center study conducted at one tertiary hospital in Shandong Province, China, which may limit the generalizability of the findings. Second, the pilot evaluation cohort sample size was relatively small (n = 63) and highly selected (restricted to flagged patients with ≥ 1 year of SHD use). This focus introduces selection bias, limits generalizability, and prevents the calculation of overall sensitivity, specificity, or negative predictive value (NPV) since non-flagged patients were not interviewed. Consequently, our performance metrics are limited to estimating the Positive Predictive Value (PPV) in this high-risk subset, and the model should not be assumed to capture all or most cases of clinical dependence across the broader outpatient population. Furthermore, this small sample size led to quasi-complete separation during multivariable logistic regression attempts, necessitating the use of exact univariate tests. As such, the present study should be regarded as a proof-of-concept investigation. Furthermore, our method of classifying dependence relied on telephone screening interviews rather than formal, in-person psychiatric diagnostic evaluations, which may have led to potential misclassification. However, as an active pharmacovigilance screening tool, its primary clinical utility is to flag suspected high-risk cases for further clinical triage rather than definitive diagnostic confirmation. Additionally, the “multiple IDs per address” trigger, while identified retrospectively, was not directly pilot-tested in the prospective telephone interviews due to data privacy constraints. Finally, the recorded off-label use rate of 30.03% may be partially influenced by incomplete or nonspecific diagnostic documentation in outpatient electronic medical records, where busy clinicians might fail to record secondary diagnoses of insomnia or anxiety, leading to the potential misclassification of clinically appropriate prescriptions as off-label. Future multi-center studies with larger sample sizes and the integration of machine learning techniques are warranted to refine trigger specificity and generalizability. Conclusion In this study, we developed and piloted an electronic trigger-based active surveillance system to monitor SHD misuse in a tertiary hospital setting. Our retrospective analysis of 16,609 prescriptions revealed prescription anomalies, including off-label use and the phenomenon of multiple patient IDs registered under the same residential address, which may serve as a hypothesis-generating marker for potential drug diversion. The prospective pilot demonstrated that the active surveillance model was able to identify screen-positive cases of suspected SHD dependence that were not captured by concurrent passive reporting, detecting 17 such cases within a 10-day period. This approach provides a scalable framework for clinical pharmacists to triage high-risk patients efficiently, thereby reducing the burden of manual chart reviews. Our findings suggest that active, trigger-based surveillance may serve as a useful complement to, rather than a replacement for, traditional spontaneous passive reporting systems in monitoring SHD misuse. Future research should focus on integrating machine learning algorithms to refine trigger specificity and extend this surveillance framework to other high-risk pharmaceutical classes. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1