Predictors of drug–drug interactions in patients with heart failure

ABSTRACT

Introduction:

Heart failure (HF) patients are frequently exposed to polypharmacy, placing them at increased risk for drug–drug interactions (DDIs), especially in an outpatient setting where monitoring may be limited. The aim was to assess the prevalence and predictors of clinically relevant DDIs in HF patients and identify patterns associated with higher interaction risk.

Methods:

This cross-sectional observational study included 670 HF patients, treated at a single outpatient cardiology practice. Clinical and pharmacologic data were extracted from medical records. DDIs were identified using the Epocrates interaction checker and classified by severity. Logistic regression and generalized linear models were used to evaluate predictors of avoidable and monitor-required interactions.

Results:

No contraindicated DDIs were found, but 72.7% of patients had at least one avoidable interaction, and 100% had at least one requiring monitoring. Anticoagulants, mineralocorticoid receptor antagonists, proton pump inhibitor (PPI) use, older age, and HFmrEF and HFpEF phenotypes were significantly associated with increased risk of clinically relevant DDIs. Polypharmacy strongly influenced interaction burden, particularly when ramipril, metformin, atorvastatin, SGLT2 inhibitors, or PPIs were prescribed. Beta-blockers were mildly protective.

Conclusion:

Clinically significant DDIs are common in HF patients in outpatient care. Risk is highest in older patients, patients with comorbidities requiring multiple pharmacological agents, regimens, and patients with HFmrEF and HFpEF. Structured medication review and attention to high-risk drug combinations are essential to improve safety in HF pharmacotherapy.

Introduction

Polypharmacy is frequently observed in patients with heart failure (HF), particularly in those with heart failure with reduced ejection fraction (HFrEF), where guideline-directed medical therapy (GDMT) commonly necessitates the concurrent use of multiple pharmacologic agents. These typically include an angiotensin-converting enzyme inhibitor (ACEi) or angiotensin receptor–neprilysin inhibitor (ARNI), a beta-adrenergic receptor blocker (beta-blocker), a mineralocorticoid receptor antagonist (MRA), and a sodium-glucose co-transporter 2 (SGLT2) inhibitor.[1] This treatment complexity is compounded by frequent coexisting conditions such as atrial fibrillation, diabetes, chronic kidney disease, and thyroid disorders, all of which necessitate additional medications.[1,2] As a result, patients are exposed to a heightened risk of potentially inappropriate prescribing (PIP) and drug–drug interactions (DDIs), both of which can negatively impact treatment outcomes.[2,3]

PIP refers to the use of medications where the potential risks outweigh clinical benefits, through overprescribing, underprescribing, or misprescribing in the context of an individual patient’s clinical profile.[2] This is especially concerning in patients with HF, where the therapeutic window is narrow for many commonly used drugs, such as digoxin or MRAs. Certain medications, including non-steroidal anti-inflammatory drugs (NSAIDs), corticosteroids, non-dihydropyridine calcium channel blockers, and some antiarrhythmics, are known to exacerbate HF symptoms or interfere with GDMT and are therefore concerned inappropriate in this context.[4] DDIs, on the other hand, occur when two or more coadministered medications affect each other’s pharmacokinetics or pharmacodynamics in a way that reduces efficacy or increases toxicity. In patients with HF, especially those on complex protocols, the clinical significance of these interactions can range from mild to life-threatening. Monitoring for DDIs is therefore a clinical component of safe prescribing yet often underemphasized in routine outpatient care.[3] This study was conducted to better understand the frequency and predictors of potentially inappropriate prescribing in heart failure patients, with a specific focus on clinically relevant DDIs. By analyzing data from a single, outpatient cardiology center, we aimed to explore whether patient characteristics, such as age, sex, HF phenotype, and common comorbidities, as well as cerain high-risk drug classes [e.g., MRAs, anticoagulants, proton pump inhibitors (PPIs), antihypertensives], were associated with an increased likelihood of encountering such prescribing risks. Our aim was to identify patterns that could inform safer, more individualized treatment strategies for this high-risk patient group.

Methods

Patients and study design

This was a cross-sectional observational study conducted on a cohort of patients diagnosed with heart failure (HF) who were treated at the private medical practice “Professor Dr. Edin Begić”, patient records spanning a 2-year period, from December 2022 to December 2024.

The inclusion criteria were age ≥18 years, confirmed diagnosis of HF according to European Society of Cardiology guidelines, and availability of complete medication records. Patients with terminal illness or incomplete data were excluded. Out of over 3000 medical records screened, a total of 670 patients met the inclusion criteria and were included in the final analysis.

Methods

Demographic and clinical data, including age, sex, comorbidities (e.g., diabetes mellitus, atrial fibrillation, thyroid disease), HF phenotype, and pharmacological therapy, were extracted from electronic medical records. Medications were categorized by pharmacologic class (e.g., ACE inhibitors, beta-blockers, MRAs, anticoagulants). The number of concomitantly used drugs was recorded for each patient. Potential DDIs were identified using the Epocrates interaction checker and were classified by clinical relevance. DDIs requiring avoidance, monitoring, or modification of therapy were defined as “clinically relevant” and were used as the primary outcome.

Statistical analysis

Dependent variables: for logistic regression – presence or absence of at least one clinically relevant DDI (binary outcome); for generalized linear modeling – number of clinically relevant avoid/monitor/modify DDIs per patient (continuous outcome). Independent variables: demographics (age, sex), clinical factors (HF phenotype, comorbidities (e.g., diabetes, AF, thyroid disease, etc.)), pharmacotherapy (use of specific drug classes, total number of medications).

Baseline characteristics were summarized using means and standard deviations for normally distributed continuous variables, medians and interquartile ranges for skewed data, and frequencies with percentages for categorical variables. Group comparisons were performed using an independent samples t-test or Mann–Whitney U test for continuous variables, and Chi-square test or Fisher’s exact test for categorical variables.

To identify independent predictors of the presence of at least one clinically relevant DDI, a multivariate logistic regression model was constructed. Variables with P < 0.10 in univariate analysis were entered into the model using backward stepwise deletion. Adjusted odds ratios (ORs) with 95% confidence intervals (CIs) were reported. Model fit was assessed using the Hosmer–Lemeshow test and Nagelkerke’s R².

To analyze predictors of the number of clinically relevant DDIs per patient, a generalized linear model (GLM) was fitted using a gamma distribution and log link function, appropriate for modeling positively skewed count data. Exponentiated coefficients were interpreted as multiplicative changes in the expected number of DDIs. Model adequacy was assessed using deviance, Akaike Information Criterion (AIC), and Omnibus likelihood ratio tests. All statistical analyses were conducted using IBM SPSS Statistics version 18.0, with a significance level set at P < 0.05.

Results

In total, 670 patients were enrolled in the study, with an average age over 70 years and almost equal distribution of males and females (334:336). The detailed characteristics of the study cohort are shown in Tables 1 and 2. A multivariate logistic regression analysis was conducted to examine the factors associated with the emergence of „Avoid/use alternative“ DDIs. The logistic regression model built by the backward deletion method was statistically significant, χ²(2) = 219.282, P = 0.000, indicating that the model reliably distinguishes between patients who experienced „Avoid/use alternative“ DDIs and those who did not. Hosmer–Lemeshow test was 2,687, P = 0.953, Cox and Snell R square was 0,284, and Nagelkerke R square was 0,411. The model correctly classified 72,6% of cases. Effects of significant predictors are presented in Table 3. Patients with HFmrEF have 2,22 times higher odds of experiencing "Avoid/use alternative" DDI than patients with HFrEF, and the same is true for patients with HfpEF when compared to patients with HfmrEF. The patients on anticoagulant, MRA, or PPI therapy have 16,1, 8,1, and 1,2 higher odds of „Avoid/use alternative“ DDIs, respectively. Each year of age increases the odds of DDIs for 3%. The only protective factor was therapy with beta blockers, decreasing the odds of „Avoid/use alternative“ DDIs for 13.3%. The Generalized Linear Model built to predict a number of “Avoid/use alternative” DDIs using a Gamma distribution and log link function was a significant improvement over the intercept-only model: Omnibus Test: χ²(43) = 389.08, P < 0.001. The model also fitted well the observed data: Deviance/df = 0.098, Pearson Chi-Square/df = 0.100. Significant predictors of the number of „Avoid/use alternative“ DDIs per patient are shown in Table 4. The patients who were treated by PPIs or anticoagulants, or had thyroid disease, had a higher number of „Avoid/use alternative“ DDIs (15.8%, 23.6%, and 34.4% more DDIs, respectively). The Generalized Linear Model built to predict the number of “Monitor/modify” DDIs using a Gamma distribution and log link function was a significant improvement over the intercept-only model: Omnibus Test: χ²(45) = 643.39, P < 0.001. The model also fitted well the observed data: Deviance/df = 0.128, Pearson Chi-Square/df = 0.112. Significant predictors of the number of „Monitor/modify“ DDIs per patient are shown in Table 5. A higher number of „Monitor/modify“ DDIs could be expected in patients who were prescribed more drugs (10.4% DDIs more for each additional drug), especially by PPIs (17.2% more DDIs), atorvastatin (20.1% more DDIs), metformin (12.7% more DDIs), ramipril (37.3% more DDIs), SGLT2 inhibitor (20.1% more DDIs), or anticoagulants (39.8% more DDIs), or had either thyroid disease (19.5% more DDIs) or HFrEF (9.9% more DDIs). The ten most frequent clinically significant DDIs observed in the study cohort are presented in Table 6.

Table 1.

Descriptive statistics of continuous variables in the study cohort (n=670)

Variable	Valid N	Mean	Std. Deviation	Median	Interquartile range	Skewness	Kurtosis
Age (years)	668	70.76	10.28	72.00	12.75	-0.600	0.844
Contraindicated interactions	670	0.00	0.00	0.00	0.00	0.000	0.000
Avoid/use alternative interactions	670	1.13	0.95	1.00	2.00	0.783	0.749
Monitor/modify interactions	670	13.49	7.04	13.00	9.00	0.972	1.911
Caution (advised) interactions	670	0.00	0.00	0.00	0.00	0.000	0.000

Table 2.

Descriptive statistics of categorical variables in the study cohort (n=670)

Variable	Category	Frequency (n)	Percentage (%)
Sex	Male	334	49,9
	Female	336	50,1
Contraindicated drug–drug interactions	Not present	670	100,0
	Present	0	0
Avoid/use alternative drug–drug interactions	Not present	183	27,3
	Present	487	72,7
Monitor/modify drug–drug interactions	Not present	0	0
	Present	670	100,0
Caution advised drug–drug interactions	Not present	668	99,7
	Present	2	,3
Angiotensin-Converting Enzyme Inhibitor (ACE-I)	Not present	172	25,7
	Enalapril	1	,1
	Ramipril	290	43,3
	Lisinopril	1	,1
	Perindopril	196	29,3
	Captopril	1	,1
	Trandolapril	9	1,3
Angiotensin receptor blocker (ARB)	Not present	618	92,2
	Valsartan	33	4,9
	Losartan	4	,6
	Olmesartan	12	1,8
	Telmisartan	3	,4
Angiotensin receptor-neprilysin inhibitor (ARNI)	Not present	595	88.8
	Sacubitril/Valsartan	75	11.2
Beta blockers (BBs)	Not present	55	8,2
	Bisoprolol	321	48,0
	Metoprolol	29	4,3
	Carvedilol	143	21,4
	Nebivolol	111	16,6
	Atenolol	1	,1
	Propanolol	9	1,3
Mineralocorticoid receptor antagonist (MRA)	Not present	165	24,6
	Spirinolactone	497	74,2
	Eplerenone	6	,9
	Finerenone	2	,3
Calcium channel blocker (CCB)	Not present	412	61,6
	Amlodipin	124	18,5
	Lercandipine	114	17,0
	Diltiazem	18	2,7
	Verapamil	1	,1
Digoxin	Not present	639	95,4
	Present	31	4,6
Ivabradine	Not present	669	99,9
	Present	1	,1
Nitrates	Not present	555	82,8
	Nitroglycerine	43	6,4
	Isosorbide mononitrate	72	10,7
Other anti-anginal	Not present	448	66,9
	Trimetazidine	218	32,5
	Ranolazine	4	,6
Statins	Not present	15	2,2
	Atorvastatin	206	30,8
	Rosuvastatin	448	67,0
Sodium-glucose cotransporter 2 inhibitor (SGLT2i)	Not present	522	77,9
	Dapagliflozin	13	1,9
	Empagliflozin	135	20,1
Biguanides	Not present	479	71,5
	Metformin	191	28,5
Glucagon-like-peptide-1 (GLP-1)	Not present	638	95,2
	Liraglutide	5	,7
	Semaglutide	5	,7
	Dulaglutide	22	3,3
Insulin	Not present	625	94,3
	Lispro	8	1,2
	Aspart	6	,9
	Glulisine	1	,2
	Glargine	9	1,4
	Detemir	2	,3
	Degludec	5	,8
	Gensulin	3	,5
	Human analog	4	,6
Other antidiabetic drugs	Not present	644	96,1
	Sulfonylureas	6	,9
	DPP-4 inhibitors	19	2,8
Proton Pump Inhibitors (PPI)	Not present	320	47,8
	Pantoprazole	225	33,6
	Rabeprazole	125	18,7
Diuretic	Not present	0	0
	Present	670	100
Heart failure phenotype	HFrEF	187	27,9
	HFmrEF	191	28,5
	HFpEF	292	43,6
Atrial fibrillation (AF)	Not present	488	73.0
	Present	181	27,0
Myocardial ischemia (MI)	Not present	391	58,4
	Present	279	41,6
Thyroid disease	Not present	582	86,9
	Present	88	13,1
Benign prostatic hypertrophy	Not present	599	89,4
	Present	71	10,6
Diabetes mellitus	Not present	437	65,2
	Present	232	34,6
Antithrombotic⁄anticoagulant therapy	Not present	21	3,1
	Present	649	96,9
Antiarrhytmic therapy	Not present	634	94,7
	Present	35	5,3

Table 3.

Predictors of „Avoid/use alternative“ DDIs (n=670)

Predictor	Adjusted Odds Ratio	95% Confidence interval	P
Heart Failure type	2,220	1,193 - 4,128	,012
Antithrombotic⁄anticoagulant therapy	16,090	4,148 - 62,412	,000
Age	1,031	1,009 - 1,054	,006
Beta-blocker (BB) therapy	,867	,752 - 1,000	,049
Mineralocorticoid Receptor Antagonist (MRA) therapy	8,130	5,142 - 12,854	,000
Proton Pump Inhibitor (PPI) therapy	1,225	1,100 - 1,365	,000

Table 4.

Predictors of the number of „Avoid/use alternative“ DDIs per patient (n=483)

Predictor	B Coefficient	Wald Chi-Square	P
Not having PPI therapy	- 0.171	16.812	0.000
Not having Thyroid Disease	-0.268	37.216	0.000
Not having antithrombotic⁄anticoagulant therapy	-0.421	7.588	0.006

Table 5.

Predictors of the number of „Monitor/modify“ DDIs per patient (n=664)

Predictor	B Coefficient	Wald Chi-Square	P
Ramipril	0.317	6.823	0.009
Atorvastatin	0.183	29.532	0.000
Not having SGLT2 inhibitor	-0.225	21.645	0.000
Not having metformin	-0.136	5.826	0.016
Not having PPI inhibitor	-0.188	21.074	0.000
HFrEF	0.095	4.009	0.045
Not having thyroid disease	-0.204	22.676	0.000
Not having antithrombotic⁄anticoagulant therapy	-0.508	31.159	0.000
Number of drugs per patient	0.099	36.284	0.000

Table 6.

The ten most frequent „Avoid/use alternative“ or „Monitor/modify“ DDIs observed in the study cohort

Interacting drugs	Consequence of the interaction
ACE inhibitor + MRA	Hyperkalemia
Angiotensin receptor blocker + MRA	Hyperkalemia
Sacubitril/valsartan + MRA	Hyperkalemia
Bisoprolol + apixaban	Increase in apixaban plasma concentration
Bisoprolol + rivaroxaban	Increase in rivaroxaban plasma concentration
Carvedilol + apixaban	Increase in apixaban plasma concentration
Acetylsalicylic acid + furosemide	Impairment of kidney function
Amiodarone + rivaroxaban	Increase in rivaroxaban plasma concentration
Clopidogrel + rabeprazole	Rabeprazole decreases effects of clopidogrel by affecting hepatic enzyme CYP2C19 metabolism
Furosemide + levothyroxine	Furosemide may inhibit binding of thyroid hormones to carrier proteins and result in increase in free thyroid hormones

Discussion

This study confirmed that possible DDIs are highly prevalent in HF patients, especially in those treated with complex medication protocols. Every patient in this study had at least one DDI that required close monitoring, which highlights just how common these issues are in everyday practice. Age, heart failure type, and certain medications, particularly anticoagulants, MRAs, and PPIs, were strong predictors of clinically relevant interactions. While combinations such as ACE inhibitors or ARBs with MRAs were among the most frequently observed and clinically significant DDIs, primarily due to the risk of hyperkalemia, anticoagulant use was the strongest independent predictor in regression analysis, likely reflecting the variety of potential interactions these agents pose with other cardiovascular drugs. These findings align with broader literature showing that anticoagulants often interact with agents that alter absorption, hepatic metabolism, or bleeding risk thresholds.[5] For instance, rivaroxaban or apixaban coadministered with beta-blockers, amiodarone, or antiplatelets has been flagged for increased exposure or bleeding risk, reinforcing the critical need for individualized therapeutic assessment.[9,10] The predictive impact of HF phenotype, specifically HFpEF and HFmrEF, suggests that these populations may be more exposed to avoidable DDIs than those with HFrEF.[10] This trend could possibly be due to often, less structured pharmacologic management in these subtypes and the higher burden of noncardiovascular comorbidities, particularly thyroid disorders, which were associated with higher DDI risk.[11,12,13,14,15] These findings are consistent with prior studies from Ethiopia and Spain, which identified comorbid conditions as key predictors of drug-related problems and highlighted thyroid disease as a frequently overlooked risk factor.[5,16,17] Polypharmacy remains a central issue; each additional prescribed drug was associated with an approximate 10% increase in the number of clinically relevant DDI requiring monitoring or modification, with combinations like ramipril and PPIs, metformin, atorvastatin, or SGLT2 inhibitors frequently contributing to the overall interaction burden. This underscores the message from Papotti et al.,[5] who noted that most clinically significant DDIs in patients with polypharmacy fall within moderate severity and are often manageable with appropriate surveillance. Another critical layer in understanding inappropriate prescribing in heart failure is medication adherence. Even when pharmacotherapy is following the guidelines, patients who fail to take their medications as prescribed face a significantly higher risk of poor outcomes and recurrent hospitalizations.[6] Meta-analyses of adherence interventions in heart failure confirm that such programs significantly reduce mortality and readmissions in this patient group.[7] From the patient’s perspective, the complexity of regimens, cognitive impairment, and adverse effects represent major barriers to adherence. These challenges often result in skipped doses or unintentional overdosing, situations that directly amplify the likelihood of harmful DDIs.[8,9,10] Equally important is the issue of dose optimization. Both underdosing and overdosing are well-documented reasons for inappropriate prescribing. In older adults, failure to tailor doses to renal function or age-related pharmacokinetics can transform otherwise safe prescriptions into clinically significant risks.[17] In this regard, explicit criteria such as Beers and STOPP/START highlight frequent errors in dosing and intervals, reinforcing the importance of systematic medication review and individualized prescribing.[11,12]

The presence of polypharmacy itself, combined with irregular dosing intervals, reflects the similar patterns described in prior literature as precursors of clinically significant DDIs.[13,14,15] Another interesting observation was the slightly protective effect of beta-blocker therapy, which may reflect the lower DDI potential of this class and its well-integrated role in HF treatment protocols. This echoes findings by Tsige et al., who reported fewer drug-related problems (DRPs) in patients on beta-blockers compared to those on diuretics, ACEIs, and MRAs.[16,17]

Sacubitril/valsartan deserves special attention due to its growing role in guideline-directed therapy for HFrEF and its unique dual mechanism. Its superiority over enalapril in reducing mortality and hospitalizations, as shown in the PARADIGM-HF trial, has driven its widespread adoption.[18] In our study, this ARNI appeared in several flagged interactions, especially with MRAs, raising concern for hyperkalemia. Additionally, it is known to reduce furosemide absorption and natriuresis, though this does not typically require dose adjustments in practice, likely due to its compensatory natriuretic effects. Even though dose adjustments are generally not necessary, these observations reinforce the importance of careful clinical monitoring when introducing sacubitril/valsartan, particularly in patients receiving MRAs or loop diuretics, due to the potential for pharmacodynamic interactions such as hyperkalemia or altered diuretic response.[18] This study has several important limitations. As a single-center, cross-sectional analysis, the findings reflect prescribing patterns from a specific outpatient setting and may not be broadly generalizable to other healthcare environments. Since this was a cross-sectional observational study, we could identify associations between predictors and DDIs, but we cannot prove direct causality. Although the use of structured interaction databases ensured consistency, it may have overestimated the clinical significance of certain flagged interactions or missed patient-specific variables that influence actual clinical risk. Additionally, we lacked comprehensive data on medication adherence, over-the-counter drug use, and renal function beyond heart failure type, all of which are known to modify drug interaction profiles, especially in patients with chronic conditions such as HF or CKD (5, 16, 17). In summary, this study reinforces that DDIs are a common and clinically relevant aspect of HF management, rather than isolated events. By pinpointing at-risk patient groups and frequently problematic drug combinations, it provides a practical basis for improving prescribing practices. Approaches like regular medication assessments, closer collaboration with clinical pharmacologists, and the use of reliable drug interaction checkers can meaningfully improve prescribing safety and help prevent avoidable harm in patients with complex heart failure therapy regimens. This study showed that DDIs are quite common in patients with heart failure, especially those on complex treatment protocols. Even though no contraindicated combinations were found, almost three-quarters of patients had at least one interaction requiring an alternative therapy, and all patients had interactions that needed monitoring. Older age, HF type, and the use of anticoagulants, MRAs, and PPIs were linked to a higher chance of clinically relevant interactions, while the beta-blockers use appeared mildly protective. Patients with HFmrEF and HFpEF were more likely to experience avoidable DDIs, possibly due to treatment complexity and noncardiac comorbidities such as thyroid disease. The most common clinically significant combinations involved MRAs with ACE inhibitors, ARBs, or sacubitril/valsartan. Additionally, polypharmacy played a major role, with the number of prescribed medications correlating with an increased interaction burden, especially when ramipril, proton pump inhibitors (PPIs), metformin, atorvastatin, or sodium-glucose co-transporter (SGLT) 2 inhibitors were used. These findings highlight the importance of structured medication reviews and greater awareness of high-risk combinations, especially in elderly patients and patients with multiple comorbidities, to improve safety in heart failure pharmacotherapy. The research can serve as a foundation that could be expanded to the national level, and in general, the results could be used to guide practice in a unified direction and optimize state funds.

Ethical approval

The study was approved by the Ethics Committee of University Sarajevo School of Science and Technology, decision No 3/25.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

Nil.