Abstract
Background:
Polypharmacy increases the risks of drug–drug interactions (DDIs), which increase the risk of adverse drug events (ADEs). Data mining techniques have described high-order (three or four) drug combinations as high risk for ADE and suggested alternative low-risk combinations, but were not able to establish clinical feasibility.
Objective:
The objective of this study is to evaluate the clinical feasibility of potentially low-risk drug combinations identified in previous work through a data mining technique as substitutes for high-risk high-order drug combinations.
Methods:
This expert-review study utilizes potentially high-risk high-order drug combinations and complementary low-risk combinations identified in previously published work from Medicare fee-for-service (2018) and MarketScan Medicare Supplemental claims emergency department (ED) data (2012–2020). We convened a panel of clinical experts to adjudicate the list for clinical feasibility. A standard rubric was developed, and the reviewers indicated whether the data-driven substitutions were always, sometimes, or never clinically acceptable and provided comments. Two experts, not involved in the initial panel review, reviewed the results produced by the panel to determine agreement or disagreement among reviewers. The results of this clinical review are presented.
Results:
Of the 1904 high-/low-risk combinations reviewed, 606 were deemed always acceptable; 588 were anticoagulant ADEs, 18 were opioid, and none were antidiabetic. The 20 most frequently observed high-risk combinations were all anticoagulant ADEs; 11 of the top 20 involved proton pump inhibitor substitution and 9 involved statin substitutions.
Conclusion:
High-risk high-order DDIs with low-risk alternatives as identified in Medicare fee-for-service and MarketScan databases are identifiable, are clinically acceptable, and could decrease ADE-related ED visits.
Plain language summary
Large data sets (tens of thousands of observations) can help identify trends that smaller data sets cannot. However, sometimes these trends are artifacts of the large data set and not clinically applicable. We previously used 2 large data sets to identify groups of 3 or 4 drugs that increase patient risk for certain drug side effects in geriatric patients and drugs that could be substituted to decrease these risks. In this manuscript, physicians reviewed the suggested drug substitutions to make sure they were clinically possible. We found many substitutions that would be clinically possible and could reduce patient risk for side effects.
Introduction
Adverse drug events (ADEs) lead to more than 1.3 million older adult emergency department (ED) visits and 350,000 hospitalizations per year in the United States. 1 High-risk individual medications have been described by the American Geriatrics Society Beers Criteria 2 and the recent Geriatric Emergency Medicine Safety Recommendations (GEMS-Rx). 3 In addition to these individual drug considerations, drug–drug interactions (DDIs) of otherwise appropriate medications are also critical to consider. DDIs occur when the combination of two or more drugs increases the risk for an ADE.
The risk of ADE due to DDIs increases with age 4 and the number of medications a patient is prescribed.5 –8 Although most research has focused on two medication combinations, the potential for unrecognized ADEs from the use of ⩾3 medications, “higher-order drug combinations,” is an area of critical importance. 9 Identifying appropriate and safer higher-order drug combinations to reduce the use of medication combinations that increase the risk of ADEs would have a direct clinical impact.
In previous work, 10 we applied the mixture drug-count response model to Medicare fee-for-service and MarketScan Medicare supplement data on ED visits to identify higher-order drug combinations with high risk of ADEs as defined by Digmann et al. 11 associated with anticoagulants (e.g., gastrointestinal bleeding), antidiabetic drugs (e.g., hypoglycemia), and opioids (e.g., sedation) in older adults. These medication classes were chosen due to established literature regarding two-drug combinations interacting and increasing the risk of ADEs related to these drug classes.12 –17 This analysis was powerful because it harnessed therapeutic class-based mining to identify potential ADE-causing high-order DDIs that would go undetected with other methods, and for its ability to suggest alternative lower risk drug combinations. However, this data analytic technique suggests substitutions by drug classes. This results in proposed low-risk drug substitutions that may not be clinically acceptable. For example, it may propose drugs with different modes of delivery within the same therapeutic class that are not interchangeable, such as oral and ophthalmic drops. Therefore, we undertook an expert review of each potential drug substitution to determine the clinical acceptability of the substitutions suggested by the data analytics.
The objective of this study was to determine the clinical feasibility of the high-to-low-risk drug substitutions in three- and four-drug combinations identified through the data mining technique in our previous work. 10
Methods
As described briefly above, our group’s previous work examined 5.1 million ED visits in Medicare fee-for-service and 3.3 million in MarketScan Medicare supplement data to identify high-risk drug combinations associated with anticoagulants, antidiabetic drugs, and opioids in older adults. Briefly, this was a case–control design that identified the high-risk drug combinations based on drug exposure in the 30 days prior to ED visit using the mixture drug-count response model. This prior work identified 1904 high-risk three- or four-drug combinations with a potential low-risk alternative drug combination. 10
The analysis presented here builds upon this large database work. For each three- and four-drug high-risk drug combination with an identified lower-risk combination in previous work described above, we identified the individual drug substitution(s) to create the lower-risk combination. This reduced the number of substitutions to be reviewed by experts to 286. 10 This study was certified as non-human subjects research by our institutional IRB.
The methodology employed in our previous work 10 relied on drug classes that could result in proposed substitutions that are not clinically feasible, for example, treating antiplatelet and anticoagulant medications as interchangeable. Two investigators (J.C. and K.H.) reviewed all individual drug substitution dyads and removed any that were deemed clearly not acceptable. Reasons a dyad was defined as clearly not acceptable included: medications were different clinical classes (e.g., antiplatelet and anticoagulant) or different modes of delivery that were not clinically interchangeable (e.g., ophthalmologic drops and intravenous).
The two investigators then reviewed the remaining dyads and classified the dyads by the specialty most appropriate for review. They determined that specialists required for review included emergency medicine, cardiology, pulmonology, gastroenterology, endocrinology, and psychiatry. Of note, the medicine subspecialists were also board-certified in internal medicine, which provided an important perspective on medications not specific to a medical subspecialty. Each substitution was reviewed by appropriate specialist(s) in addition to an emergency medicine physician to determine whether the drug substitution suggested by the database analysis was clinically acceptable. Therefore, each substitution was reviewed by at least two reviewers, but no limit was placed on the number of reviewers. Each reviewer indicated if the proposed substitution was always, sometimes, or never clinically acceptable and had the opportunity to provide comments. Each reviewer made this determination based on their clinical expertise. Given knowledge and medication use differences between specialties, disagreements were expected. In cases of disagreement among the initial reviewers, two investigators (J.C. and B.B.) reviewed the selections and comments to make a final determination. In this determination, justified specialty-specific knowledge was favored. For example, if the emergency medicine physician reviewer said something was “always” acceptable, but the cardiologist picked “sometimes” and provided a reason specific to cardiology knowledge, sometimes was the final determination. If during investigator review, there was no consensus reached, a group meeting was held to reach consensus.
We present the reviewed individual medication substitutions in three categories: always acceptable, sometimes acceptable, and never acceptable with the following assumptions: (1) appropriate dosing adjustments can be made between medications for potency, half-life, and patient factors such as renal function; (2) patient can take the medication in the required form, for example if a PO medication replaces an IV medication; (3) medications are available on the US market; and (4) clinicians would review full medication lists for other DDIs prior to the substitution.
Finally, we present the 20 most frequent higher-order drug substitutions identified in previous work and deemed “always” accepted by this review of individual drug substitutions and the associated relative risks as calculated in previous work. 10 Most frequent were defined as the substitutions in three- or four-drug combinations with significant p values that occurred with the highest frequency when Medicare fee-for-service and MarketScan Medicare supplement data were summed.
Results
Of the 286 dyads from previous work, 58 were removed on initial review as clearly not clinically acceptable by the initial two investigator review. Of the 228 remaining, after expert review, 84 dyads were categorized as always, 123 as sometimes, and 21 as never appropriate for substitution (Tables 1–3). There were no discrepancies remaining after final review by the two investigators considering specialty-specific knowledge, and, therefore, no consensus meeting was needed.
Individual drug substitutions are determined to be always acceptable and therefore clinically relevant among identified in three- and four-drug combinations with lower-risk substitutions.
Individual drug substitutions determined to be sometimes acceptable and therefore potentially clinically relevant among identified in three- and four-drug combinations with lower-risk substitutions.
BPH, benign prostatic hyperplasia; CAD, coronary artery disease; CHF, congestive heart failure; DKA, diabetic ketoacidosis; GERD, gastroesophgeal reflux disease; LV, left ventricular; LVAD, left ventricular assist devices; PRN, pro re nata; VT, ventricular tachycardia.
Individual drug substitutions are determined to be never acceptable and therefore clinically irrelevant among identified in three- and four-drug combinations with lower-risk substitutions.
Of the 1904 high-risk combinations with lower-risk alternative combinations identified in our prior work, 10 606 had suggested drug substitutions that were deemed always acceptable; 588 were anticoagulant ADEs, 18 were opioid, and none were antidiabetic. The 20 most frequently observed high-risk combinations and the proposed substitutions are listed in Table 4; the top 20 all involved anticoagulant ADEs. Eleven of the top 20 involved suggested proton pump inhibitor substitutions, and 9 suggested statin substitutions. The complete list of high- and low-risk combinations with always acceptable substitutions is provided in Supplemental Table 1. For example, for a patient taking furosemide, metoprolol, and pantoprazole, if pantoprazole is replaced with dexlansoprazole, the associated relative risk of anticoagulant side effect is 0.689 in Medicare fee-for-service and 0.488 in MarketScan Medicare supplement data.
The 20 most frequent high-risk three- and four-drug combinations and low-risk alternative drug combinations with high-risk drug combination and substitution that makes it a low-risk combination, as identified in previous work. 10
Relative risks are reported for Medicare fee-for-service and MarketScan Medicare supplement data separately. All top 20 are anticoagulant-induced adverse drug events.
In the Medicare fee-for-service data, there were 148,098 ED visits with a potential anticoagulant ADE. The 588 always acceptable high-/low-risk substitutions correspond to 46,665 ED visits. Therefore, 31.5% of the anticoagulant ADEs with a high-risk combination present had an acceptable low-risk alternative. Similarly, in MarketScan, this corresponded to 113,531 with ED visits with potential anticoagulant ADE, 50,087 with always acceptable substitution, and, therefore, 44.1% with a low-risk alternative.
In the Medicare fee-for-service data, there were 146,245 ED visits with a potential opioid ADE. The 18 always acceptable high-/low-risk substitutions correspond to 1048 ED visits. Therefore, 0.7% of the opioid ADEs with a high-risk combination present had an acceptable low-risk alternative. Similarly, in MarketScan, this corresponded to 116,781 ED visits with potential opioid ADE, 1441 with always acceptable substitution, and, therefore, 1.2% with a low-risk alternative.
There were no clinical substitutions for diabetic ADEs that were deemed always acceptable.
Discussion
We demonstrate that the findings of the mixture drug-count response model yield clinically relevant results. Almost one-third of the identified high-/low-risk substitutions were deemed always acceptable with appropriate assumptions. If applied to reduce the use of high-risk three- and four-drug combinations, these results could have immediate positive clinical impact on patients. Though not explicitly presented here, many more substitutions may be acceptable in certain scenarios.
We estimated that 31.5% and 44.1% of anticoagulant ADEs observed in the Medicare (2018) and MarketScan (2012–2020) datasets, respectively, had an acceptable lower-risk alternative regimen. This corresponds to 46,665 and 50,087 visits in the databases, respectively. Annually, this is 46,665 and 5565. While this would be the upper bound of effect, substitution of a lower-risk alternative has the potential for large clinical significance of preventing tens of thousands of ADEs per year. Importantly, these lower-risk alternatives have equal therapeutic efficacy, are widely available, are highly familiar to prescribers, are generally inexpensive, and should be acceptable substitutions to the vast majority of prescribing physicians.
Similarly, we estimated that 0.7% and 1.2% of opioid ADEs observed in the Medicare and MarketScan datasets, respectively, had an acceptable lower-risk alternative regimen. This corresponds to approximately 1048 and 1441 visits in the databases, respectively. Annually, this is 1048 and 160. While this is a lower potential clinical impact than anticoagulation ADEs, this still has clear clinical applicability.
The primary limitation of this work is related to the use of large databases to identify the high- and low-risk drug combinations. There is no way to know if the potential ADEs were related to the medications or not. For example, a gastrointestinal bleed could occur as an ADE or due to the development of a bleeding ulcer that would have occurred with or without the high-risk medications. However, the relative risks are significant and imply that there is a real relationship. Furthermore, we do not know if the patients were taking the medications as prescribed.
Further work should seek to determine whether the high-risk combinations were felt to contribute to the side effect in real patient encounters, examine the identified high- and low-risk combinations to determine if there is an underlying pharmacologic explanation for the differences, and seek to change prescribing patterns to lower-risk combinations.
Conclusion
High-risk high-order DDIs with low-risk alternatives as identified in Medicare fee-for-service and MarketScan databases are clinically acceptable as determined by expert physician review of proposed substitutions. Substituting high-risk combinations for low-risk combinations could decrease ED visits resulting from anticoagulant and opioid ADEs; in particular, substitutions of certain statins and PPIs could have the greatest effect.
Supplemental Material
sj-docx-1-taw-10.1177_20420986251374931 – Supplemental material for Data mining results for reduction of adverse drug–drug interaction events in older adults are clinically applicable
Supplemental material, sj-docx-1-taw-10.1177_20420986251374931 for Data mining results for reduction of adverse drug–drug interaction events in older adults are clinically applicable by Katherine M. Hunold, Yi Shi, Pengyue Zhang, Macarius M. Donneyong, Alexander Ulintz, Benjamin H. Buck, Joshua I. Gordon, Antoinette Pusateri, Melanie Rayan, Katherine Brownlowe and Jeffrey M. Caterino in Therapeutic Advances in Drug Safety
Footnotes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
