Biomedical Informatics Approaches to Identifying Drug-Drug Interactions: Application to Insulin Secretagogues

Thinking Three-Dimensionally: A Self- and Externally-Controlled Approach to Screening for Drug-Drug-Drug Interactions Among High-Risk Populations

The global rise in polypharmacy has increased both the necessity and complexity of drug-drug interaction (DDI) assessments, given the growing potential for interactions involving more than two drugs. Leveraging large-scale healthcare claims data, we piloted a semi-automated, high-throughput case-crossover-based approach for drug-drug-drug interaction (3DI) screening. Cases were direct-acting oral anticoagulant (DOAC) users with either a major bleeding event during ongoing dispensings for potentially interacting, enzyme-inhibiting antihypertensive drugs (AHDs) (Study 1), or a thromboembolic event during ongoing dispensings for potentially interacting, enzyme-inducing antiseizure medications (ASMs) (Study 2). 3DI detection was based on screening for additional drug exposures that served as acute outcome triggers. To mitigate direct effects and confounding by concomitant drugs, self-controlled estimates were adjusted using negative cases (external "control" DOAC users with the same outcomes but co-dispensings for non-interacting AHDs or ASMs). Signal thresholds were set based on P-values and false discovery rate q-values to address multiple comparisons. Study 1: 285 drugs were examined among 3,306 episodes. Self-controlled assessments with q-value thresholds yielded 9 3DI signals (cases) and 40 DDI signals (negative cases). External adjustment generated 10 3DI signals from the P-value threshold and no signals from the q-value threshold. Study 2: 126 drugs were examined among 604 episodes. Assessments with P-value thresholds yielded 3 3DI and 26 DDI signals following self-control, as well as 4 3DI signals following adjustment. No 3DI signals met the q-value threshold. The presented self- and externally-controlled approach aimed to advance paradigms for real-world higher order drug interaction screening among high-susceptibility populations with pre-existent DDI risk.

The State of Use and Utility of Negative Controls in Pharmacoepidemiologic Studies

Uses of real-world data in drug safety and effectiveness studies are often challenged by various sources of bias. We undertook a systematic search of the published literature through September 2020 to evaluate the state of use and utility of negative controls to address bias in pharmacoepidemiologic studies. Two reviewers independently evaluated study eligibility and abstracted data. Our search identified 184 eligible studies for inclusion. Cohort studies (115, 63%) and administrative data (114, 62%) were, respectively, the most common study design and data type used. Most studies used negative control outcomes (91, 50%), and for most studies the target source of bias was unmeasured confounding (93, 51%). We identified 4 utility domains of negative controls: 1) bias detection (149, 81%), 2) bias correction (16, 9%), 3) P-value calibration (8, 4%), and 4) performance assessment of different methods used in drug safety studies (31, 17%). The most popular methodologies used were the 95% confidence interval and P-value calibration. In addition, we identified 2 reference sets with structured steps to check the causality assumption of the negative control. While negative controls are powerful tools in bias detection, we found many studies lacked checking the underlying assumptions. This article is part of a Special Collection on Pharmacoepidemiology.

Application of an Innovative Data Mining Approach Towards Safe Polypharmacy Practice in Older Adults

INTRODUCTION

Polypharmacy is common and is associated with higher risk of adverse drug event (ADE) among older adults. Knowledge on the ADE risk level of exposure to different drug combinations is critical for safe polypharmacy practice, while approaches for this type of knowledge discovery are limited. The objective of this study was to apply an innovative data mining approach to discover high-risk and alternative low-risk high-order drug combinations (e.g., three- and four-drug combinations).

METHODS

A cohort of older adults (≥ 65 years) who visited an emergency department (ED) were identified from Medicare fee-for-service and MarketScan Medicare supplemental data. We used International Classification of Diseases (ICD) codes to identify ADE cases potentially induced by anticoagulants, antidiabetic drugs, and opioids from ED visit records. We assessed drug exposure data during a 30-day window prior to the ED visit dates. We investigated relationships between exposure of drug combinations and ADEs under the case-control setting. We applied the mixture drug-count response model to identify high-order drug combinations associated with an increased risk of ADE. We conducted therapeutic class-based mining to reveal low-risk alternative drug combinations for high-order drug combinations associated with an increased risk of ADE.

RESULTS

We investigated frequent high-order drug combinations from 8.4 million ED visit records (5.1 million from Medicare data and 3.3 million from MarketScan data). We identified 5213 high-order drug combinations associated with an increased risk of ADE by controlling the false discovery rate at 0.01. We identified 1904 high-order, high-risk drug combinations had potential low-risk alternative drug combinations, where each high-order, high-risk drug combination and its corresponding low-risk alternative drug combination(s) have similar therapeutic classes.

CONCLUSIONS

We demonstrated the application of a data mining technique to discover high-order drug combinations associated with an increased risk of ADE. We identified high-risk, high-order drug combinations often have low-risk alternative drug combinations in similar therapeutic classes.

Real-World Trends in the Evaluation of Medical Products

There is a compelling need to evaluate the real-world health effects of medical products outside of tightly controlled preapproval clinical trials. This is done through pharmacoepidemiology, which is the study of the health effects of medical products (including drugs, biologicals, and medical devices and diagnostics) in populations, often using nonrandomized designs. Recent developments in pharmacoepidemiology span changes in the focus of research questions, research designs, data used, and statistical analysis methods. Developments in these areas are thought to improve the value of the evidence produced by such studies, and are prompting greater use of real-world evidence to inform clinical, regulatory, and reimbursement decisions.

Skeletal muscle relaxant drug-drug-drug interactions and unintentional traumatic injury: Screening to detect three-way drug interaction signals

AIM

The aim of this study was to identify skeletal muscle relaxant (SMR) drug-drug-drug interaction (3DI) signals associated with increased rates of unintentional traumatic injury.

METHODS

We conducted automated high-throughput pharmacoepidemiologic screening of 2000-2019 healthcare data for members of United States commercial and Medicare Advantage health plans. We performed a self-controlled case series study for each drug triad consisting of an SMR base-pair (i.e., concomitant use of an SMR with another medication), and a co-dispensed medication (i.e., candidate interacting precipitant) taken during ongoing use of the base-pair. We included patients aged ≥16 years with an injury occurring during base-pair-exposed observation time. We used conditional Poisson regression to calculate adjusted rate ratios (RRs) with 95% confidence intervals (CIs) for injury with each SMR base-pair + candidate interacting precipitant (i.e., triad) versus the SMR-containing base-pair alone.

RESULTS

Among 58 478 triads, 29 were significantly positively associated with injury; confounder-adjusted RRs ranged from 1.39 (95% CI = 1.01-1.91) for tizanidine + omeprazole with gabapentin to 2.23 (95% CI = 1.02-4.87) for tizanidine + diclofenac with alprazolam. Most identified 3DI signals are new and have not been formally investigated.

CONCLUSION

We identified 29 SMR 3DI signals associated with increased rates of injury. Future aetiologic studies should confirm or refute these SMR 3DI signals.

Population-based screening to detect benzodiazepine drug-drug-drug interaction signals associated with unintentional traumatic injury

Drug interactions involving benzodiazepines and related drugs (BZDs) are increasingly recognized as a contributor to increased risk of unintentional traumatic injury. Yet, it remains unknown to what extent drug interaction triads (3DIs) may amplify BZDs' inherent injury risk. We identified BZD 3DI signals associated with increased injury rates by conducting high-throughput pharmacoepidemiologic screening of 2000-2019 Optum's health insurance data. Using self-controlled case series design, we included patients aged ≥ 16 years with an injury while using a BZD + co-dispensed medication (i.e., base pair). During base pair-exposed observation time, we identified other co-dispensed medications as candidate interacting precipitants. Within each patient, we compared injury rates during time exposed to the drug triad versus to the base pair only using conditional Poisson regression, adjusting for time-varying covariates. We calculated rate ratios (RRs) with 95% confidence intervals (CIs) and accounted for multiple estimation via semi-Bayes shrinkage. Among the 65,123 BZD triads examined, 79 (0.1%) were associated with increased injury rates and considered 3DI signals. Adjusted RRs for signals ranged from 3.01 (95% CI = 1.53-5.94) for clonazepam + atorvastatin with cefuroxime to 1.42 (95% CI = 1.00-2.02, p = 0.049) for alprazolam + hydrocodone with tizanidine. These signals may help researchers prioritize future etiologic studies to investigate higher-order BZD interactions.

Opioid Drug-Drug-Drug Interactions and Unintentional Traumatic Injury: Screening to Detect Three-Way Drug Interaction Signals

Growing evidence suggests that drug interactions may be responsible for much of the known association between opioid use and unintentional traumatic injury. While prior research has focused on pairwise drug interactions, the role of higher-order (i.e., drug-drug-drug) interactions (3DIs) has not been examined. We aimed to identify signals of opioid 3DIs with commonly co-dispensed medications leading to unintentional traumatic injury, using semi-automated high-throughput screening of US commercial health insurance data. We conducted bi-directional, self-controlled case series studies using 2000-2015 Optum Data Mart database. Rates of unintentional traumatic injury were examined in individuals dispensed opioid-precipitant base pairs during time exposed vs unexposed to a candidate interacting precipitant. Underlying cohorts consisted of 16-90-year-olds with new use of opioid-precipitant base pairs and ≥1 injury during observation periods. We used conditional Poisson regression to estimate rate ratios adjusted for time-varying confounders, and semi-Bayes shrinkage to address multiple estimation. For hydrocodone, tramadol, and oxycodone (the most commonly used opioids), we examined 16,024, 8185, and 9330 drug triplets, respectively. Among these, 75 (0.5%; hydrocodone), 57 (0.7%; tramadol), and 42 (0.5%; oxycodone) were significantly positively associated with unintentional traumatic injury (50 unique base precipitants, 34 unique candidate precipitants) and therefore deemed potential 3DI signals. The signals found in this study provide valuable foundations for future research into opioid 3DIs, generating hypotheses to motivate crucially needed etiologic investigations. Further, this study applies a novel approach for 3DI signal detection using pharmacoepidemiologic screening of health insurance data, which could have broad applicability across drug classes and databases.

Population-Based Signals of Antidepressant Drug Interactions Associated With Unintentional Traumatic Injury

Antidepressants are very widely used and associated with traumatic injury, yet little is known about their potential for harmful drug interactions. We aimed to identify potential drug interaction signals by assessing concomitant medications (precipitant drugs) taken with individual antidepressants (object drugs) that were associated with unintentional traumatic injury. We conducted pharmacoepidemiologic screening of 2000-2015 Optum Clinformatics data, identifying drug interaction signals by performing self-controlled case series studies for antidepressant + precipitant pairs and injury. We included persons aged 16-90 years codispensed an antidepressant and ≥ 1 precipitant drug(s), with an injury during antidepressant therapy. We classified antidepressant person-days as either precipitant-exposed or precipitant-unexposed. The outcome was an emergency department or inpatient discharge diagnosis for unintentional traumatic injury. We used conditional Poisson regression to calculate confounder adjusted rate ratios (RRs) and accounted for multiple estimation via semi-Bayes shrinkage. We identified 330,884 new users of antidepressants who experienced an injury. Among such persons, we studied concomitant use of 7,953 antidepressant + precipitant pairs. Two hundred fifty-six (3.2%) pairs were positively associated with injury and deemed potential drug interaction signals; 22 of these signals had adjusted RRs > 2.00. Adjusted RRs ranged from 1.06 (95% confidence interval: 1.00-1.12, P = 0.04) for citalopram + gabapentin to 3.06 (1.42-6.60) for nefazodone + levonorgestrel. Sixty-five (25.4%) signals are currently reported in a seminal drug interaction knowledgebase. We identified numerous new population-based signals of antidepressant drug interactions associated with unintentional traumatic injury. Future studies, intended to test hypotheses, should confirm or refute these potential interactions.

Drug-Drug Interaction Surveillance Study: Comparing Self-Controlled Designs in Five Empirical Examples in Real-World Data

Self-controlled designs, specifically the case-crossover (CCO) and the self-controlled case series (SCCS), are increasingly utilized to generate real-world evidence (RWE) on drug-drug interactions (DDIs). Although these designs share the advantages and limitations of within-individual comparison, they also have design-specific assumptions. It is not known to what extent the differences in assumptions lead to different results in RWE DDI analyses. Using a nationwide US commercial healthcare insurance database (2006-2016), we compared the CCO and SCCS designs, as they are implemented in DDI studies, within five DDI-outcome examples: (1) simvastatin + clarithromycin and muscle-related toxicity; (2) atorvastatin + valsartan, and muscle-related toxicity; and (3-5) dabigatran + P-glycoprotein inhibitor (clarithromycin, amiodarone, and verapamil) and bleeding. Analyses were conducted within person-time exposed to the object drug (statins and dabigatran) and adjusted for bias associated with the inhibiting drugs via control groups of individuals unexposed to the object drug. The designs yielded similar estimates in most examples, with SCCS displaying better statistical efficiency. With both designs, results varied across sensitivity analyses, particularly in CCO analyses with small number of exposed individuals. Analyses in controls revealed substantial bias that may be differential across DDI-exposed and control individuals. Thus, both designs showed no association between amiodarone or verapamil and bleeding in dabigatran-exposed but revealed strong positive associations in controls. Overall, bias adjustment via a control group had a larger impact on results than the choice of a design, highlighting the importance and challenges of appropriate control group selection for adequate bias control in self-controlled analyses of DDIs.

Comparison of Self-controlled Designs for Evaluating Outcomes of Drug-Drug Interactions: Simulation Study

BACKGROUND

Self-controlled designs, both case-crossover and self-controlled case series, are well suited for evaluating outcomes of drug-drug interactions in electronic healthcare data. Their comparative performance in this context, however, is unknown.

METHODS

We simulated cohorts of patients exposed to two drugs: a chronic drug (object) and a short-term drug (precipitant) with an associated interaction of 2.0 on the odds ratio scale. We analyzed cohorts using case-crossover and self-controlled case series designs evaluating exposure to the precipitant drug within person-time exposed to the object drug. Scenarios evaluated violations of key design assumptions: (1) time-varying, within-person confounding; (2) time trend in precipitant drug exposure prevalence; (3) nontransient precipitant exposure; and (4) event-dependent object drug discontinuation.

RESULTS

Case-crossover analysis produced biased estimates when 30% of patients persisted on the precipitant drug (estimated OR 2.85) and when the use of the precipitant drug was increasing in simulated cohorts (estimated OR 2.56). Self-controlled case series produced biased estimates when patients discontinued the object drug following the occurrence of an outcome (estimated incidence ratio [IR] of 2.09 [50% of patients stopping therapy] and 2.22 [90%]). Both designs yielded similarly biased estimates in the presence of time-varying, within-person confounding.

CONCLUSION

In settings with independent or rare outcomes and no substantial event-dependent censoring (<50%), self-controlled case series may be preferable to case-crossover design for evaluating outcomes of drug-drug interactions. With frequent event-dependent drug discontinuation, a case-crossover design may be preferable provided there are no time-related trends in drug exposure.

Screening to identify signals of opioid drug interactions leading to unintentional traumatic injury

BACKGROUND

Efforts to minimize harms from opioid drug interactions may be hampered by limited evidence on which drugs, when taken concomitantly with opioids, result in adverse clinical outcomes.

OBJECTIVE

To identify signals of opioid drug interactions by identifying concomitant medications (precipitant drugs) taken with individual opioids (object drugs) that are associated with unintentional traumatic injury DESIGN: We conducted pharmacoepidemiologic screening of Optum Clinformatics Data Mart, identifying drug interaction signals by performing confounder-adjusted self-controlled case series studies for opioid + precipitant pairs and injury.

SETTING

Beneficiaries of a major United States-based commercial health insurer during 2000-2015 PATIENTS: Persons aged 16-90 years co-dispensed an opioid and ≥1 precipitant drug(s), with an unintentional traumatic injury event during opioid therapy, as dictated by the case-only design EXPOSURE: Precipitant-exposed (vs. precipitant-unexposed) person-days during opioid therapy.

OUTCOME

Emergency department or inpatient International Classification of Diseases discharge diagnosis for unintentional traumatic injury. We used conditional Poisson regression to generate confounder adjusted rate ratios. We accounted for multiple estimation via semi-Bayes shrinkage.

RESULTS

We identified 25,019, 12,650, and 10,826 new users of hydrocodone, tramadol, and oxycodone who experienced an unintentional traumatic injury. Among 464, 376, and 389 hydrocodone-, tramadol-, and oxycodone-precipitant pairs examined, 20, 17, and 16 (i.e., 53 pairs, 34 unique precipitants) were positively associated with unintentional traumatic injury and deemed potential drug interaction signals. Adjusted rate ratios ranged from 1.23 (95 % confidence interval: 1.05-1.44) for hydrocodone + amoxicillin-clavulanate to 4.21 (1.88-9.42) for oxycodone + telmisartan. Twenty (37.7 %) of 53 signals are currently reported in a major drug interaction knowledgebase.

LIMITATIONS

Potential for reverse causation, confounding by indication, and chance CONCLUSIONS: We identified previously undescribed and/or unappreciated signals of opioid drug interactions associated with unintentional traumatic injury. Subsequent etiologic studies should confirm (or refute) and elucidate these potential drug interactions.

Medication class enrichment analysis: a novel algorithm to analyze multiple pharmacologic exposures simultaneously using electronic health record data

Objective

Observational studies analyzing multiple exposures simultaneously have been limited by difficulty distinguishing relevant results from chance associations due to poor specificity. Set-based methods have been successfully used in genomics to improve signal-to-noise ratio. We present and demonstrate medication class enrichment analysis (MCEA), a signal-to-noise enhancement algorithm for observational data inspired by set-based methods.

Materials and Methods

We used The Health Improvement Network database to study medications associated with Clostridium difficile infection (CDI). We performed case-control studies for each medication in The Health Improvement Network to obtain odds ratios (ORs) for association with CDI. We then calculated the association of each pharmacologic class with CDI using logistic regression and MCEA. We also performed simulation studies in which we assessed the sensitivity and specificity of logistic regression compared to MCEA for ORs 0.1-2.0.

Results

When analyzing pharmacologic classes using logistic regression, 47 of 110 pharmacologic classes were identified as associated with CDI. When analyzing pharmacologic classes using MCEA, only fluoroquinolones, a class of antibiotics with biologically confirmed causation, and heparin products were associated with CDI. In simulation, MCEA had superior specificity compared to logistic regression across all tested effect sizes and equal or better sensitivity for all effect sizes besides those close to null.

Discussion

Although these results demonstrate the promise of MCEA, additional studies that include inpatient administered medications are necessary for validation of the algorithm.

Conclusions

In clinical and simulation studies, MCEA demonstrated superior sensitivity and specificity for identifying pharmacologic classes associated with CDI compared to logistic regression.

Clopidogrel Drug Interactions and Serious Bleeding: Generating Real-World Evidence via Automated High-Throughput Pharmacoepidemiologic Screening

Few population-based studies have examined bleeding associated with clopidogrel drug-drug interactions (DDIs). We sought to identify precipitant drugs taken concomitantly with clopidogrel (an object drug) that increased serious bleeding rates. We screened 2000-2015 Optum commercial health insurance claims to identify DDI signals. We performed self-controlled case series studies for clopidogrel plus precipitant pairs, examining associations with gastrointestinal bleeding or intracranial hemorrhage. To distinguish native bleeding effects of a precipitant, we reexamined associations using pravastatin as a negative control object drug. Among 431 analyses, 28 clopidogrel plus precipitant pairs were statistically significantly positively associated with serious bleeding. Ratios of rate ratios ranged from 1.13-3.94. Among these pairs, 13 were expected given precipitant drugs alone increased and/or were harbingers of serious bleeding. The remaining 15 pairs constituted new DDI signals, none of which are currently listed in two major DDI knowledge bases.

A Case-Crossover-Based Screening Approach to Identifying Clinically Relevant Drug-Drug Interactions in Electronic Healthcare Data

We sought to develop a semiautomated screening approach using electronic healthcare data to identify drug-drug interactions (DDIs) that result in clinical outcomes. Using a case-crossover design with 30-day hazard and referent windows, we evaluated codispensed drugs (potential precipitants) in 7,801 patients who experienced rhabdomyolysis while on cytochrome P450 (CYP)3A4-metabolized statins and in 15,147 who experienced bleeding while on dabigatran. Estimates of direct associations between precipitant drugs and outcomes were used to adjust for bias and precipitants' direct effects. The P values were adjusted for multiple testing using the false discovery rate (FDR). From among 460 drugs codispensed with statins, 1 drug (clarithromycin) generated an alert (adjusted odds ratio (OR) 5.83, FDR < 0.05). From among 485 drugs codispensed with dabigatran, 2 drugs (naproxen and enoxaparin, ORs 2.50 and 2.75; FDR < 0.05) generated an alert. All three signals reflected known pharmacologic interactions, confirming the potential of case-crossover-based approaches for DDI screening in electronic healthcare data.

Serious Hypoglycemia and Use of Warfarin in Combination With Sulfonylureas or Metformin

Prior research suggests that warfarin, when given concomitantly with some sulfonylureas, may increase the risk of serious hypoglycemia. However, the clinical significance remains unclear. We examined rate ratios (RRs) for the association between serious hypoglycemia and concomitant use of warfarin with either sulfonylureas or metformin using a self-controlled case series design and US Medicaid claims (supplemented with Medicare claims) from 1999 to 2011. Across all risk windows combined, warfarin was associated with an elevated rate of serious hypoglycemia when given concomitantly with glimepiride (RR, 1.47; 95% confidence interval (CI), 1.07-2.02) and metformin (RR, 1.73; 95% CI, 1.38-2.16). Particularly in the late risk window (>120 days since beginning concomitancy), most of the RRs for warfarin were elevated: glipizide (RR, 1.72; 95% CI, 1.29-2.29), glyburide (RR, 1.57; 95% CI, 1.15-2.15), metformin (RR, 2.26; 95% CI, 1.67-3.05), and glimepiride (RR, 1.56; 95% CI, 0.97-2.50). These results are consistent with a previously hypothesized hypoglycemic effect of warfarin in patients with type 2 diabetes through inhibition of the carboxylation of osteocalcin.

Comparative risk of serious hypoglycemia with oral antidiabetic monotherapy: A retrospective cohort study

PURPOSE

To examine and compare risks of serious hypoglycemia among antidiabetic monotherapy-treated adults receiving metformin, a sulfonylurea, a meglitinide, or a thiazolidinedione.

METHODS

We performed a retrospective cohort study of apparently new users of monotherapy with metformin, glimepiride, glipizide, glyburide, pioglitazone, rosiglitazone, nateglinide, or repaglinide within a dataset of Medicaid beneficiaries from California, Florida, New York, Ohio, and Pennsylvania. We did not include users of dipeptidyl peptidase-4 inhibitors, glucagon-like peptide-1 agonists, or sodium-glucose co-transporter 2 inhibitors. We identified serious hypoglycemia outcomes within 180 days following new use using a validated, diagnosis-based algorithm. We calculated age- and sex-standardized outcome occurrence rates for each drug and generated propensity score-adjusted hazard ratios vs metformin using Cox proportional hazards regression.

RESULTS

The ranking of standardized occurrence rates of serious hypoglycemia was glyburide > glimepiride > glipizide > repaglinide > nateglinide > rosiglitazone > pioglitazone > metformin. Rates were increased for all study drugs at higher average daily doses. Adjusted hazard ratios (95% confidence intervals) vs metformin were 3.95 (3.66-4.26) for glyburide, 3.28 (2.98-3.62) for glimepiride, 2.57 (2.38-2.78) for glipizide, 2.03 (1.64-2.52) for repaglinide, 1.21 (0.89-1.66) for nateglinide, 0.90 (0.75-1.07) for rosiglitazone, and 0.80 (0.68-0.93) for pioglitazone.

CONCLUSIONS

Sulfonylureas were associated with the highest rates of serious hypoglycemia. Among all study drugs, the highest rate was seen with glyburide. Pioglitazone was associated with a lower adjusted hazard for serious hypoglycemia vs metformin, while rosiglitazone and nateglinide had hazards similar to that of metformin.

Pharmacoepidemiologic Methods for Studying the Health Effects of Drug-Drug Interactions

A drug-drug interaction (DDI) occurs when one or more drugs affect the pharmacokinetics (the body's effect on the drug) and/or pharmacodynamics (the drug's effect on the body) of one or more other drugs. Pharmacoepidemiologic studies are the principal way of studying the health effects of potential DDIs. This article discusses aspects of pharmacoepidemiologic research designs that are particularly salient to the design and interpretation of pharmacoepidemiologic studies of DDIs.