Hospital EHR alert system cuts dangerous drug–herb interactions in Thai study

Herbal medicines are no longer fringe therapy. Across hospitals in Southeast Asia, patients routinely combine conventional prescription drugs with herbal preparations, often without telling their physicians, and the pharmacological consequences can be serious. Now researchers in Thailand have shown that a carefully curated, severity-stratified drug–herb interaction alert system, built directly into a hospital electronic health record, can measurably reduce co-prescribing of interacting pairs. The study, conducted at Warinchamrab Hospital, a 260-bed general district hospital in Ubon Ratchathani Province, tracked more than half a million prescription encounters over 24 months and found a statistically significant decline in drug–herb interaction events after the alert system went live.

The scale of the problem the team set out to address is considerable. Concurrent use of herbal and conventional medicines is common worldwide, with documented prevalence of 33.6% among older adults using prescription medications in high-income community settings, and potential herb–drug interaction exposure reaching 79.8% among cancer patients using herbal medicine in lower-income clinical settings. A systematic review and meta-analysis of clinical case reports identified 176 potentially life-threatening drug–herb interactions across 23 countries, with anticoagulant, antidiabetic, immunosuppressant, and cardiovascular drug classes disproportionately represented. In Thailand specifically, surveys have found that 35.9% of outpatients with chronic diseases used herbal medicines in the preceding 12 months, and among Thai outpatients with chronic kidney disease, 45% reported herbal and dietary supplement use, with 72% not informing their physicians.

The mechanisms underlying drug–herb interactions parallel those of conventional drug–drug interactions but remain substantially understudied in controlled clinical settings. Herbal phytochemicals can induce or inhibit cytochrome P450 isoenzymes such as CYP3A4, CYP2C9, and CYP2D6, modulate P-glycoprotein efflux transport, produce direct pharmacodynamic synergism or antagonism, and alter drug absorption through phytochemical chelation or changes in gastric pH. Despite these well-characterized pathways, the evidence base relies disproportionately on isolated case reports and preclinical animal models rather than prospective pharmacokinetic trials. Compounding the problem, a systematic review of 44 knowledge, attitude, and practice studies encompassing 16,929 participants found that healthcare professionals, including pharmacists, consistently underestimate their knowledge deficits regarding herb and supplement interactions and rarely proactively inquire about patients’ herbal use during consultations.

Existing screening resources have proven inadequate for the task. An evaluation of eight interaction tools found sensitivity below 0.20, with Lexicomp showing the highest overall performance score of only 0.54. Computerized provider order entry and clinical decision support system interventions can reduce medication errors, but drug–herb interaction-specific EHR alert systems remain underdeveloped, mostly limited to prototype or non-prospective evaluations. Against this backdrop, the Thai research team constructed an evidence-based interaction database through a standardized seven-step development and validation process, drawing on internationally recognized tertiary references including Stockley’s Herbal Medicines Interactions, Lexicomp, Micromedex Drug-REAX, and the Natural Medicines Comprehensive Database, supplemented by primary literature searches in PubMed and MEDLINE to capture emerging phytopharmacological evidence on regionally specific herbs.

Crucially, the database was restricted to herbal medicinal products listed in the Thai National List of Essential Medicines or those routinely stocked within the hospital formulary, including Andrographis paniculata, Cassia angustifolia, Phyllanthus emblica, Curcuma longa, and standardized Cannabis preparations. Interacting pairs were classified into three operational clinical tiers. Major-severity interactions, coded red, involved narrow therapeutic index medications such as warfarin, chemotherapeutic agents, or combinations with potential for severe toxicity, and triggered an interruptive hard stop requiring mandatory clinical justification and prescriber confirmation before the order could proceed. Moderate-severity interactions, coded orange, could cause altered therapeutic efficacy or symptom exacerbation and triggered an interruptive soft stop requiring acknowledgment and clinical review. Minor-severity interactions, coded yellow, were supported by preclinical data or mild pharmacological shifts and were routed as non-interruptive passive notifications to dispensing pharmacists to support patient counseling without creating prescriber workflow friction.

A multidisciplinary clinical committee comprising two clinical pharmacists, one general physician, one licensed Thai traditional medicine practitioner, and one hospital informatics specialist independently reviewed each interaction rule, with discrepancies reconciled through formal consensus. The finalized database, clinical alert rules, and interface warning messages were formally ratified by the hospital Pharmacy and Therapeutics Committee before production deployment. The alert system operated in two complementary detection modalities: intra-prescription detection, triggering when an interacting pair was entered within a single prescription encounter, and cross-prescription detection, triggering when a newly prescribed herbal medicine interacted with an active conventional medicine from a prior prescription still within the patient’s active medication record. Before full implementation, the system underwent pilot testing with 30 healthcare professionals, and all five satisfaction domains, including alert accessibility, ease of order confirmation, clinical usefulness, ease of use, and response speed and accuracy, surpassed the predetermined acceptability threshold of 3.50 out of 5.00.

The results over 24 months were telling. Of 510,910 prescription encounters recorded, 7,278 contained both conventional and herbal medicines, and 544 drug–herb interaction events were identified overall. The drug–herb interaction event rate decreased significantly from 312 out of 239,684 encounters, or 0.13%, before implementation to 232 out of 271,226 encounters, or 0.09%, after implementation, a difference significant at p < 0.001. Major-severity events decreased from 28 to 13, representing a 53.6% relative reduction, and moderate-severity events decreased from 284 to 122. Minor-severity events increased from zero to 97, likely reflecting expanded database coverage rather than clinical safety deterioration. The most prevalent pair across both periods was Milk of Magnesia with Senna, a moderate-severity dual laxative combination risking excessive fluid and electrolyte depletion, which accounted for 72.6% of all drug–herb interaction events and decreased from 278 to 117 events, a 57.9% reduction.

Among major-severity pairs, Naproxen with Andrographis paniculata and Etoricoxib with Andrographis paniculata were most frequent, with naproxen events decreasing from 12 to seven and etoricoxib events decreasing from 12 to two. These pairs are biologically plausible because preclinical rat studies have shown that Andrographis paniculata extract and andrographolide can alter the pharmacokinetics of naproxen and etoricoxib. Warfarin with Andrographis paniculata emerged as a new major-severity pair identified exclusively in the post-implementation period, warranting close monitoring because warfarin has a narrow therapeutic index, although preclinical evidence remains inconsistent. Post-implementation detection of minor-severity events predominantly involved Phyllanthus emblica with metformin, aspirin, or clopidogrel, signals that are pharmacologically plausible given demonstrated antiplatelet activity of the herb but should be interpreted as monitoring signals rather than confirmed clinical harm.

The three-tier severity architecture was designed to concentrate prescriber interruption at the highest-risk encounters while reducing low-value alert exposure, an approach consistent with recommendations for severity-based alert tiering and workflow-sensitive design. Alert fatigue remains a central operational challenge in medication-related clinical decision support implementation, with a recent systematic review and meta-analysis of drug–drug interaction alerts reporting a pooled physician override rate of 90%. Because the EHR system did not systematically capture alert override rates or mandatory override rationales, the exact extent to which prescribers intentionally bypassed versus adhered to the alerts cannot be definitively quantified. The persistent co-prescribing of the dominant laxative pair highlights an operational challenge in distinguishing clinically justifiable combination laxative therapy from unintentional co-prescribing, underscoring the necessity of embedding structured, forced-choice override documentation into clinical workflows.

The authors are careful to distinguish prescribing-process improvements from actual clinical patient safety outcomes. The primary outcomes reflect electronic order-modification behaviors at the point of prescribing, and while 10 clinical adverse event reports were identified in the hospital EHR during the 24-month period, these data originated from routine spontaneous pharmacovigilance documentation rather than a prospectively powered, adjudicated clinical trial. The findings demonstrate that the alert system successfully reduced patient exposure to potentially hazardous drug–herb combinations at the point of prescribing but should not be construed as direct evidence of reduced clinical adverse drug reactions, hospitalizations, or patient morbidity. Limitations include the observational before–after design lacking a contemporaneous parallel control group, the single-center setting operating an institution-specific EHR platform, and the database curation restricted to Thai essential medicines, which may limit generalizability to jurisdictions where different phytotherapeutic agents such as St. John’s Wort, Ginkgo biloba, or Echinacea predominate. To maintain and extend the prescribing safety benefits, the team recommends annual database maintenance, monitoring of alert overrides, prospective evaluation of direct clinical patient outcomes, and integration with the national pharmacovigilance infrastructure.

Subject of Research: Development and evaluation of a severity-stratified drug–herb interaction alert system integrated into a hospital electronic health record in Thailand

Article Title: Development and evaluation of a drug–herb interaction alert system integrated into a hospital electronic health record

Article References: Songmuang, T., Songmuang, A., Pitchayajittipong, C., & Supapaan, T. S. (2026). Development and evaluation of a drug–herb interaction alert system integrated into a hospital electronic health record. Pharmacia, 73, Article e203882. https://doi.org/10.3897/pharmacia.73.e203882

Image Credits: AI Generated

DOI: 10.3897/pharmacia.73.e203882

Keywords: drug–herb interactions, electronic health records, clinical decision support, medication safety, pharmacovigilance, Thailand, herbal medicine, prescription safety, Development, evaluation, drug, herb