Prevalence, predictors, and clinical relevance of drug-drug interactions in outpatient prescribing: A national cross-sectional study

Drug–drug interactions (DDIs) pose a major challenge in healthcare, resulting from one medication altering the pharmacological effect of another drug that is given at the same time [ 1 ]. A High prevalence of the DDI has been reported across different populations, with recent analyses showing rates of 12.12% and 10.06% in Indianapolis (USA) and Catalonia (Spain), respectively [ 2 ]. In low and middle-income countries (LMICs), the situation is even more concerning. This disparity is often attributed to limited resources for medication monitoring and insufficient clinical decision support systems in resource-constrained settings [ 3 , 4 ]. While several studies have explored DDIs within the Iranian healthcare context, their findings show significant variability. For instance, a systematic review of Iranian studies reported a median DDI rate of 8.5% in the outpatient setting. However, individual reports vary widely. This wide disparity is largely related to the small or single-center sample size and the lack of a unified approach [ 5 ]. There remains a critical need for large-scale, methodologically standardized pharmacoepidemiologic analyses to bridge these gaps and provide a baseline for national health policy reforms. The clinical consequences of DDIs vary from minimal effects to significant adverse events that may include life-threatening ones [ 6 ]. Exposure to DDI is associated with an increased number of adverse drug events and decreased health‐related quality of life [ 7 ]. Pharmacokinetic interactions can result in altered drug absorption, distribution, metabolism, or excretion, while pharmacodynamic interactions may produce additive, synergistic, or antagonistic effects [ 8 ]. Severe DDIs may present as cardiac arrhythmias, seizures, respiratory depression, hypotension, kidney injury, liver toxicity, or bleeding [ 9 ]. DDIs could be responsible for a large number of hospital admissions, with anticoagulants, cardiovascular drugs, and antimicrobials commonly involved [ 7 , 10 ]. In addition to physiological damage, DDIs could have a significant impact on treatment adherence since patients suffering from adverse effects could stop taking life-saving medications without consulting their medical doctor [ 11 ]. DDIs also represent a critical threat to patient safety within polypharmacy, defined as concurrent use of five or more medications [ 12 ]. Likewise, such a threat exists regarding multimorbidity, where complex regimens amplify DDI risk, adverse drug reactions, and hospitalizations, particularly in older adults. Patient safety frameworks, such as WHO’s Medication Without Harm initiative, emphasize clinical decision support systems and deprescribing protocols to mitigate these risks, emphasizing the need for tailored pharmacoepidemiologic strategies in at-risk populations [ 13 ]. Discovery and management of DDIs can substantially decrease healthcare utilization and costs [ 7 ]. This economic impact is particularly pronounced in healthcare systems with limited resources, where preventable DDIs divert scarce funding from other essential services [ 14 ]. Several tools are available to healthcare providers for the timely identification of potential DDIs, as early interventions prevent malefic outcomes [ 15 ]. The Micromedex Drug Interactions database is one of the most commonly used and extensive resources available, providing evidence-based interventions to support clinical decisions across the patient care continuum [ 16 ]. Screening of DDIs confirmed that although newer solutions are promising, established databases like Micromedex remained unrivaled in accuracy, breadth of coverage, and showed the best performance [ 15 , 17 ]. Therefore, this study aimed to evaluate outpatient medication safety in Iran by achieving three primary objectives: (1) determining the prevalence and patterns of potential DDIs at patient and prescription levels; (2) classifying these interactions by severity patterns. These include contraindicated, major, and moderate interactions based on Micromedex® criteria (a top clinical decision support package); and (3) identifying demographic and clinical predictors of DDI occurrence and severity. Characterizing DDIs at a national level provides the essential evidence required to refine clinical prescribing guidelines, optimize decision support within e-prescribing systems, and implement health policy reforms to improve medication safety.

From an initial 3,093,481 prescriptions (16,421,497 drugs), 727,670 were excluded. Specifically, 35,913 prescriptions were removed due to incomplete demographic data (missing age or sex), 128,638 were excluded because they contained only a single drug, and 563,119 were removed as they did not include any of the 100 most frequently prescribed medications. This resulted in 2,365,811 prescriptions comprising 11,499,301 drugs from 982,102 patients. The cohort had a median age of 51 years (IQR: [31–64]). On average, each patient received 2.41 ± 2.35 prescriptions, containing 4.86 ± 2.43 medications in each prescription. Nearly half of all prescriptions and more than 60% of patients involved polypharmacy (≥ 5 drugs), underscoring substantial exposure to potential drug interactions. Sex-specific patterns showed females had more prescriptions per patient, while males had more drugs per prescription and higher polypharmacy prevalence (p < 0.001). Furthermore, both prescription frequency and polypharmacy rates rose significantly with age. Patient characteristics are detailed in Table 1 . N: number, SD: standard deviation, IQR: Interquartile Range. *Polypharmacy: prescriptions with ≥ 5 medications. Bold values indicate statistically significant differences (p < 0.001) for all comparisons between sexes and across age groups (t-test, ANOVA, or Chi-squared). The five most frequently prescribed medications were acetaminophen (n = 422,819; 17.87% of all prescriptions), atorvastatin (303,720; 12.84%), azithromycin (269,454; 11.39%), acetylsalicylic acid (ASA, aspirin) (260,600; 11.02%), and the adult cold (acetaminophen + chlorpheniramine maleate + phenylephrine hydrochloride) (250,285; 10.58%). Outpatient prescriptions yielded 2,606,201 DDIs from 4,725 unique drug pairs, representing 28.59% of all possible pairs (16,529). Nearly half of prescriptions (46.06%, n = 1,089,718) contained at least one interaction, with an average of 2.39 ± 2.29 DDIs and 5.97 ± 2.58 drugs per prescription. At the patient level, 57.77% (n = 567,407) had at least one interacting prescription, averaging 1.92 ± 1.66 such prescriptions per patient. By severity, major interactions predominated (62.58%; n = 1,630,882), followed by moderate (32.82%; n = 855,443) and contraindicated (4.60%; n = 119,876). At the prescription level, 36.12% included at least one major DDI, 19.34% at least one moderate, and 4.73% at least one contraindicated combination. Detailed classifications are shown in Fig 1 , with drug-specific analyses in Supplementary File 1 ( S1 Table ). Fig 1 illustrates the proportion of all prescriptions assigned to eight mutually exclusive DDI categories. Over half of the prescriptions contained no interaction (dark green, 53.94%). Among those with DDIs, the largest segment consisted of prescriptions with only major interactions (orange, 22.48%), followed by those with both moderate and major interactions (dark blue, 11.32%) and only moderate interactions (yellow, 7.52%). Prescriptions containing only contraindicated interactions accounted for 2.25% (red). Less common patterns included major + contraindicated interactions (brown, 1.99%), moderate + contraindicated interactions (purple, 0.34%), and prescriptions featuring all three severities, moderate, major, and contraindicated (black, 0.16%). A limited number of drug pairs accounted for most DDIs. Among the 20 most prevalent interactions, 10 were moderate, eight major, and two contraindicated. The leading contraindicated pair was naproxen–ketorolac (1.77% of all interactions), with additional risks from other Non-steroidal anti-inflammatory drugs (NSAID)–ketorolac combinations (diclofenac, ibuprofen, celecoxib), highlighting duplicative NSAIDs use. Major interactions were dominated by acetylsalicylic acid, especially with metformin (2.36%), clopidogrel (1.46%), and hydrochlorothiazide (1.36%), reflecting frequent co-prescription of cardiovascular and antidiabetic agents. Moderate DDIs were likewise centered on acetylsalicylic acid, most often with metoprolol (2.25%), nitroglycerin (2.08%), and bisoprolol (1.17%). Table 2 summarizes the 10 most common DDIs, ranked by severity and frequency. The most frequent interactions involve agents with significant clinical safety profiles. For instance, the high prevalence of ASA + clopidogrel interactions carries a well-documented risk of major hemorrhage. Similarly, the ASA + metformin pair necessitates close monitoring for hypoglycemia, particularly in elderly populations. The top contraindicated pairs, predominantly NSAID duplications involving ketorolac, represent a high-risk prescribing pattern that significantly elevates the danger of acute kidney injury and gastrointestinal perforation. For a comprehensive drug-specific analysis Supplementary file 2 ( S2 Table ) provides detailed information on the top three interacting medications for each of the 100 most frequently prescribed drugs, stratified by interaction severity level. Network analysis ( Fig 2 ) depicts a densely connected network dominated by two interaction hubs, acetylsalicylic acid and metformin, whose node diameters denote the highest interaction burdens. The acetylsalicylic-acid–metformin pair shows the most significant edge weight, while similarly high-volume major interactions linked acetylsalicylic acid to clopidogrel and hydrochlorothiazide. Contraindicated interactions are confined to a compact NSAID cluster, with the thickest red edge between ketorolac and naproxen and additional red links involving diclofenac, ibuprofen, and celecoxib. Beyond these high-risk pairs, orange edges (major severity) connect cardiovascular, metabolic, and gastrointestinal agents, while the surrounding lattice of green edges (moderate severity) spans multiple therapeutic areas. Fig 2 visualizes the top high-frequency drug–drug interactions detected in outpatient prescriptions. Node diameter is proportional to the total number of interactions attributed to each drug (larger nodes indicate higher interaction counts). Node color reflects the primary therapeutic class. Edge width is scaled to interaction frequency. Edge color denotes interaction severity—red = contraindicated, orange = major, green = moderate. DDI prevalence was comparable between males (46.05%, n = 443,434) and females (46.07%, n = 646,284; p = 0.87). However, females had a higher rate of contraindicated prescriptions (4.94% vs. 4.43%), while males showed a greater mean number of interactions per prescription (1.15 vs. 1.07), driven by a higher prevalence of major (36.56% vs. 35.83%) and moderate (20.27% vs. 18.71%) interactions (all p < 0.001). Age analysis revealed a progressive rise in DDI prevalence from children (0–18 years) (24.92%) to older adults (65–100 years) (52.48%), paralleled by an increase in mean interactions (0.38 ± 0.88 to 1.45 ± 2.31). Contraindicated interactions were highest in the 19–39-year group (7.27%), while major (39.62%) and moderate (30.27%) interactions peaked in the elderly (65–100 years) (all p < 0.001) ( Table 3 ). Additionally, a comprehensive stratified analysis of the most frequent DDIs by sex and age groups is provided in Supplementary File 3 ( S3 Table ). N: number, SD: standard deviation. Bold values indicate statistically significant differences (p < 0.001) for all comparisons between sexes and across age groups (t-test, ANOVA, or Chi-squared). Among the ten highest-prescribing specialties, general practitioners issued the largest share of prescriptions, with more than half involving polypharmacy second only to cardiology. Despite this, DDI prevalence in general practice (43.09%) was lower than in several specialties. Cardiology showed the most intense DDI exposure: over two-thirds of prescriptions contained interactions (70.29%), with the highest mean drug count (5.60 ± 2.67) and average number of interactions per interacting prescription (3.34 ± 2.72). Psychiatry had a similarly high prevalence (81.99%) and the second-highest DDI average (3.24 ± 3.16), though with lower polypharmacy rate (37.63%). Pediatrics had the lowest DDI burden (20.14%) and fewer interactions per prescription (1.89 ± 1.76). Urology also recorded low polypharmacy (14.2%) and interaction prevalence (28.31%). In terms of severity, orthopedics (30.24%) and emergency medicine (21.74%) had the highest rates of contraindicated interactions, while neurology and psychiatry were dominated by major DDIs (≥ 90%). In contrast, moderate interactions were a particularly prominent feature in cardiology, where their prevalence surpassed that of major interactions ( Table 4 ). A comprehensive list of the five most frequent interactions for each specialty and the three leading contraindicated, major, and moderate pairs can be found in supplementary file 4 ( S4 Table ). DDI: drug-drug interaction, N: number, SD: standard deviation. After excluding specialties with fewer than 5,000 prescriptions (constituting roughly 0.5% of all prescriptions) to improve representativeness and stability of estimates, multivariable linear regression identified several independent predictors of prescription-level DDI counts. The number of drugs per prescription remained the dominant predictor. Each additional medication increased the expected interaction count by β = 0.460 (95% CI 0.459–0.461; p < 0.001). Physician specialty showed the widest effect range. Compared with general practitioners, prescriptions written by psychiatrists had the largest incremental DDI load (β = 2.191, 95% CI 2.181 to 2.200; p < 0.001), followed by cardiology (β = 1.143, 95% CI 1.136 to 1.151, p < 0.001) and neurology (β = 1.042, 95% CI 1.031 to 1.052, p < 0.001). Conversely, radiation oncology was associated with a significant decrease in DDI load (β = −0.899, 95% CI −0.926 to −0.871; p < 0.001). Age demonstrated a graded positive association. Relative to children (0–18 years), prescriptions for adults aged 19–39, 40–64, and 65–100 years increased DDI numbers by β = 0.105 (95% CI 0.097–0.113), β = 0.373 (95% CI 0.366–0.380), and β = 0.520 (95% CI 0.513–0.528), respectively (all p < 0.001). Sex had a modest inverse effect, with female prescriptions showing β = –0.042 (95% CI –0.046 to –0.038; p < 0.001) compared with males. The model accounted for 40.2% of the variance in interaction counts (adjusted R² = 0.402). Fig 3 illustrates the relative magnitude and direction of each predictor’s effect on the number of DDIs. Plot of standardized β coefficients (dots) with 95% confidence intervals (horizontal bars) for predictors of prescription-level drug–drug interaction counts in the multivariable linear regression model. Estimates are shown relative to their reference categories: male sex, age 0–18 years, and general practice. The vertical dashed line at β = 0 denotes no association; coefficients to the right indicate an increase and those to the left a decrease in the expected number of interactions. In the interaction-adjusted linear model, the incremental effect of drug count on DDI burden differed markedly across age groups and physician specialties. The baseline slope for drug count was β = 0.152 (95% CI 0.149–0.154; p < 0.001); however, this slope increased progressively with patient age, rising by an additional 0.057, 0.180, and 0.279 units in the 19–39, 40–64, and ≥ 65-year groups, respectively (all p < 0.001). A comparable pattern emerged across specialties: relative to general practice, the drug-count slope was steepest in psychiatry (β = 0.812, 95% CI 0.808–0.816) and remained elevated in cardiology (β = 0.442, 95% CI 0.440–0.445), general surgery (β = 0.385, 95% CI 0.378–0.391), and neurology (β = 0.328, 95% CI 0.323–0.332) (all p < 0.001). Sex did not significantly modify the drug-count effect (β = –0.004, p = 0.063). Incorporation of these interaction terms improved model fit, yielding an adjusted R² of 0.477. Interaction effects between drug count and key predictors are illustrated in Fig 4 . This figure presents the estimated coefficients and 95% confidence intervals from a multivariable linear regression model including interaction terms between drug count and key predictors. The vertical dashed line at zero indicates no effect. Positive coefficients reflect a greater increase in DDIs per additional medication, conditional on age group or physician specialty. Notably, interaction slopes were steepest for psychiatry, cardiology, and older age groups, indicating a stronger marginal effect of drug count in these subgroups. In the multivariable logistic regression model predicting the presence of contraindicated DDIs, physician specialty emerged as the strongest predictor. Compared with general practitioners, prescriptions issued by orthopedic specialists had the highest odds of containing contraindications (OR = 2.305, 95% CI 2.251–2.360; p < 0.001), followed by emergency medicine (OR = 1.417, 95% CI 1.390–1.446; p < 0.001) and physical therapy and rehabilitation (OR = 1.260, 95% CI 1.148–1.381; p < 0.001). In contrast, specialties such as dermatology (OR = 0.036, 95% CI 0.021–0.062; p < 0.001), ophthalmology (OR = 0.038, 95% CI 0.024–0.060; p < 0.001), and obstetrics and gynecology (OR = 0.050, 95% CI 0.042–0.058; p < 0.001) were associated with significantly reduced odds of contraindicated prescriptions. Drug count was another strong and independent predictor. Each additional medication increased the odds of a contraindicated interaction by 27.5% (OR = 1.275, 95% CI 1.272–1.278; p < 0.001). Age also showed a graded association with contraindications. Compared to individuals aged 0–18 years, those aged 19–39 years had the highest odds (OR = 2.796, 95% CI 2.720–2.874; p < 0.001), followed by 40–64 years (OR = 2.245, 95% CI 2.188–2.304; p < 0.001) and 65–100 years (OR = 1.345, 95% CI 1.307–1.384; p < 0.001). Female sex was associated with a modest but statistically significant increase in the likelihood of contraindicated interactions compared to males (OR = 1.032, 95% CI 1.019–1.046; p < 0.001). The model demonstrated good overall fit (AIC = 766625), with specialty and drug count identified as the most influential predictors of contraindication presence. Fig 5 shows the direction and magnitude of each predictor’s influence on contraindicated DDIs. This figure presents the odds ratio and 95% confidence intervals from the multivariable logistic regression model assessing the association between patient and physician characteristics and the presence of contraindicated DDIs. Estimates are shown relative to the reference categories: male sex, age group 0–18 years, and general practice. Predictors with positive coefficients increased the likelihood of contraindications, while those with negative coefficients were associated with reduced odds.

This large-scale analysis highlights a high burden of potential DDIs in Iranian outpatient prescriptions, particularly among elderly patients, men, and those receiving multiple medications. The dominance of major interactions and the presence of over 100,000 contraindicated prescriptions underscore the urgent need for comprehensive prescribing reform. Polypharmacy and age are the principal drivers, and a few common drug classes (NSAIDs, antiplatelets, antidiabetics) dominate the risk. Health systems should prioritize e-prescribing, clinician education, pharmacist integration, and patient engagement to mitigate DDI risks and improve medication safety.

This was a retrospective, cross-sectional study that analyzed data from the Iran Health Insurance Organization (IHIO) database for the year 2022–2023. The primary objective was to investigate the prevalence and patterns of potential DDIs, categorize them by severity, and identify associated demographic and clinical predictors. The dataset was accessed for research purposes on February 19, 2025. All data were fully anonymized prior to analysis, and the authors had no access to any information that could identify individual participants during or after data collection. Therefore, in accordance with institutional and national ethical guidelines, formal ethics approval and informed consent were not required. This study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for cross-sectional studies. The study protocol was reviewed and conducted in accordance with the national ethical standards of the National Center for Health Insurance Research and the Iran Health Insurance Organization (IHIO) allowing the use of administrative health data. As this was a retrospective, cross-sectional analysis utilizing a fully de-identified national dataset, the research was classified as exempt from individual informed consent. Data processing and cohort cleaning were conducted using Python software (Version 3.12). We included all IHIO-registered outpatient prescriptions in Tehran between March 26, 2022, and February 25, 2023. Prescriptions were excluded if they had incomplete information on patient sex or age, contained only a single drug, or did not include at least one drug from a list of the 100 most frequently prescribed medications in the database. A list of the 100 most frequently prescribed drugs was used as a reference. A comprehensive analysis was performed to identify all possible DDIs among these drugs using the Micromedex Drug Interactions database. Micromedex categorizes drug interaction severity into four levels: contraindicated, major, moderate, and minor. Contraindicated means the two drugs should not be taken together because the risk of serious harm is very high. Major interactions may be life-threatening and/or demand urgent medical management to prevent severe adverse outcomes. Moderate interactions may worsen the patient’s condition and/or necessitate adjustments in treatment [ 18 ]. The DDI information containing contraindicated, major, and moderate was first exported from Micromedex online database as PDF files for each of these 100 drugs, which were manually reviewed, cleaned, and semi-automatically converted to CSV files for use in the analysis. Minor interactions were excluded from the analysis because they are generally considered clinically negligible and typically do not necessitate treatment modifications. To ensure the consistency and accuracy of this process, two authors (S.S.A. and F.S.) independently performed the data extraction and cleaning. A cross-check of the converted CSV files against the original PDF records was conducted, and any inconsistencies in drug mapping or severity grading were resolved through consensus. To accurately identify interactions, a multi-step process was implemented. Regular expressions were used to match generic drug names from the Micromedex data to the various brand names and formulations found in the IHIO prescription records [ 19 ]. For generic drugs like “Ferrous Sulfate,” a list of alternative names such as “IRONFORTE,” “FERFOLIC,” and “FOLIRON” was manually curated and used in a regex pattern to ensure all relevant prescriptions were captured. The analysis focused on systemic drugs, and non-systemic drugs were excluded by identifying and omitting prescriptions containing keywords like cream, lotion, and drops in the drug name. Additionally, to avoid double-counting, drug pairs were standardized by sorting the names alphabetically (e.g., Drug A-Drug B was treated the same as Drug B-Drug A). For prescriptions with multiple interactions of varying severity for the same drug pair, the highest severity level was retained for analysis. All Python scripts used for data processing and analysis are openly available in our GitHub repository ( https://github.com/SahabAarabi/Drug-Drug-Interaction-Project ). The prevalence of DDIs was calculated at both the patient and prescription levels. The frequency and proportion of DDIs by severity level were also analyzed. Based on potential violations of the normality assumption assessed using Shapiro-Wilk tests, data are reported as either mean ±SD or median and Inter-quartile range (IQR). Differences between groups were assessed using student's t-tests, analysis of variance (ANOVA), and chi-squared (χ²) tests. A p-value threshold of 0.05 was utilized for statistical significance. To identify factors independently associated with the number and severity of DDIs, multivariable regression models were employed using R software (R Core Team (2024). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/ , version 4.5.0). R scripts used for analysis are also openly available in the mentioned GitHub repository. A linear regression model was developed to predict the total count of DDIs per prescription. The model included independent variables such as age group, sex, drug count, and physician specialty. An additional model was constructed to explore the interaction effects between drug count and age group, as well as drug count and specialty, to determine how the effect of polypharmacy varies across different patient and physician groups. Moreover, a logistic regression model was utilized to predict the presence of a contraindicated DDI (a binary outcome of yes or no). The same set of independent variables from the linear model was used to identify key predictors of these high-risk interactions. Results are presented as odds ratios (OR) with 95% confidence intervals (CI) to quantify the magnitude of each predictor’s effect. Model robustness was evaluated using the adjusted R 2 for linear regression models and the Akaike Information Criterion (AIC) for the logistic regression model. Multicollinearity was assessed for all multivariable models using the Variance Inflation Factor (VIF). All VIF values were found to be below 3, confirming that collinearity did not significantly bias the model estimates. Wherever model assumptions (e.g., normality, homoscedasticity, or linearity of predictors) were not fully met, we applied a nonparametric bootstrap approach with 1,000 resamples to obtain robust estimates and validate model stability. A drug interaction network was created to visually represent the most common DDIs. This was accomplished using Cytoscape software [ 20 ]. In this network, drugs were represented as nodes, with the size of each node scaled to reflect its total interaction count. The connections between drugs (edges) were weighted by the frequency of co-prescription and colored according to the severity of the interaction (red for contraindicated, orange for major, and green for moderate).

(DOCX) This table presents the three most frequent drug interaction partners for each of the 100 most commonly prescribed medications in the IHIO database, categorized by severity level (contraindicated, major, and moderate). Numbers represent the absolute frequency of co-prescription, while percentages (in parentheses) indicate the proportion relative to all prescriptions containing the target drug listed in the first column. Empty cells indicate no interactions were identified in that severity category. (DOCX) (DOCX) (DOCX)