eTable. Association of Patient Characteristics and Antimicrobial Exposure With Hypoglycemic Events Among Glipizide or Glyburide Users, Stratified by Sex
Parekh TM, Raji M, Lin Y, Tan A, Kuo Y, Goodwin JS. Hypoglycemia After Antimicrobial Drug Prescription for Older Patients Using Sulfonylureas. JAMA Intern Med. 2014;174(10):1605-1612. doi:10.1001/jamainternmed.2014.3293
Certain antimicrobial drugs interact with sulfonylureas to increase the risk of hypoglycemia.
To determine the risk of hypoglycemia and associated costs in older patients prescribed glipizide or glyburide who fill a prescription for an antimicrobial drug.
Design, Setting, and Participants
This was a retrospective cohort study of Texas Medicare claims from 2006 to 2009 for patients 66 years or older who were prescribed glipizide or glyburide and who also filled a prescription for 1 of the 16 antimicrobials most commonly prescribed for this population.
We assessed hypoglycemia events and associated Medicare costs in patients prescribed 1 of 7 antimicrobial agents thought to interact with sulfonylureas, using noninteracting antimicrobials as a comparison. We used a repeated measure logistic regression, controlling for age, sex, ethnicity, Medicaid eligibility, comorbidity, prior emergency department visits for hypoglycemia, prior hospitalizations for any cause, nursing home residence, and indication for the antimicrobial. We estimated odds of hypoglycemia, number needed to harm, deaths during hospitalization for hypoglycemia, and Medicare costs for hypoglycemia treatment.
Main Outcomes and Measures
Any hospitalization or emergency department visit owing to hypoglycemia within 14 days of antimicrobial exposure.
In multivariable analyses controlling for patient characteristics and indication for antimicrobial drug use, clarithromycin (odds ratio [OR], 3.96 [95% CI, 2.42-6.49]), levofloxacin (OR, 2.60 [95% CI, 2.18-3.10]), sulfamethoxazole-trimethoprim (OR, 2.56 [95% CI, 2.12-3.10]), metronidazole (OR, 2.11 [95% CI, 1.28-3.47]), and ciprofloxacin (OR, 1.62 [95% CI, 1.33-1.97]) were associated with higher rates of hypoglycemia compared with a panel of noninteracting antimicrobials. The number needed to harm ranged from 71 for clarithromycin to 334 for ciprofloxacin. Patient factors associated with hypoglycemia included older age, female sex, black or Hispanic race/ethnicity, higher comorbidity, and prior hypoglycemic episode. In 2009, 28.3% of patients prescribed a sulfonylurea filled a prescription for 1 of these 5 antimicrobials, which were associated with 13.2% of all hypoglycemia events in patients taking sulfonylureas. The treatment of subsequent hypoglycemia adds $30.54 in additional Medicare costs to each prescription of 1 of those 5 antimicrobials given to patients taking sulfonylureas.
Conclusions and Relevance
Prescription of interacting antimicrobial drugs to patients on sulfonylureas is very common, and is associated with substantial morbidity and increased costs.
Sulfonylureas are used in the management of type 2 diabetes mellitus. Hypoglycemia is a known adverse effect of sulfonylureas,1 with a reported rate of 1.23 hospitalizations per 100 patients per year (95% CI, 1.08-1.38 hospitalizations).2 Glyburide use is associated with a greater hypoglycemic risk than glipizide (odds ratio [OR], 1.9 [95% CI, 1.2-2.9]).3 Hypoglycemia can result in significant morbidity, including deterioration in cognitive function, higher risk of dementia, and stroke, as well as death.4- 7
Investigators have documented increased hospitalizations for hypoglycemia with glipizide or glyburide following coadministration of several antimicrobial agents with sulfonylureas, using analyses of Medicaid data8 and population-based data from Ontario, Canada.9 The existence of Medicare Part D drug data allows us to evaluate the frequency of coprescription of antimicrobials and sulfonylureas and their association with hypoglycemia in the older Medicare population. In addition, we can assess how specific patient characteristics affect the risk of hypoglycemia and estimate downstream Medicare costs owing to hypoglycemia. Our aim was to estimate the proportion of all hypoglycemic events in patients prescribed glipizide or glyburide that were attributed to an interacting antibiotic and also to estimate the downstream Medicare costs associated with a prescription for an interacting antibiotic in a patient taking sulfonylureas.
We followed the general approach of prior studies8,9 to make these estimates, assessing the risk of emergency department visits or hospitalizations for hypoglycemia after use of antimicrobials previously linked to potential interactions with sulfonylureas and comparing them with the rates with antimicrobials with no known interactions with sulfonylureas. The antimicrobials are shown in Table 1, along with a description of the evidence of their interaction (or lack of same) with sulfonylureas.
The research protocol was considered exempt by the University of Texas Medical Branch institutional review board. Claims from the years 2006 to 2009 for Texas Medicare beneficiaries were used, including Medicare beneficiary summary files, Prescription Drug Event files, Medicare Provider Analysis and Review (MedPAR) files, Outpatient Standard Analytical Files (OutSAF), and Medicare Carrier files. For 2007 through 2009, we identified all prescriptions for glipizide or glyburide. Then we identified those with concurrent use of any oral antimicrobial agent, defined as at least 1 day of overlapping supply. We then selected the 16 most frequently prescribed (Table 1). We next selected those from beneficiaries 66 years or older who had Medicare Parts A and B enrollment but not enrollment in any health maintenance organization (HMO) in the 12 months prior to and 14 days after the antimicrobial filling date, and had no additional antimicrobial prescribed in the 14 days after the initial script for an antimicrobial.
We also performed patient level analyses using 2007 and 2009 data to estimate the prevalence of individuals taking glipizide or glyburide who were also prescribed an interacting antibiotic. The inclusion criteria were beneficiaries who had completed 12 months of Part D enrollment and who used glipizide or glyburide.
Demographic information, including age, sex, race/ethnicity, and Medicaid eligibility, was collected from Part D beneficiary summary files. We used the Medicaid indicator as a proxy of low socioeconomic status.29 The Charlson Comorbidity Index, excluding diabetes mellitus, was calculated using claims from MedPAR, Carrier, and OutSAF files in the year prior to the antimicrobial drug use.30 Previous emergency department visits for hypoglycemia in the prior year were identified by using the algorithm of Ginde et al.31 The number of acute hospitalizations for any cause in the prior year was determined from MedPAR files. We determined residence in skilled nursing facilities and long-term care nursing homes by identifying skilled nursing facility billings from MedPAR files in the 14 days before antimicrobial use or Evaluation and Management charges for any nursing home care with Current Procedural Terminology codes 99304 to 99318 from Carrier and OutSAF files in the previous 3 months.32 We examined diagnoses from all claims in MedPAR, Carrier, and OutSAF files on the antimicrobial prescription date and in the prior 7 days to determine the possible indication for antimicrobial use..
The RED BOOK Select Extracts database was used to identify the drug class.33 Based on a literature review, the 16 antimicrobial drugs were divided into 2 groups: (1) those with prior evidence that or a plausible mechanism by which they might potentiate sulfonylureas to cause hypoglycemia and (2) those with no prior evidence and no plausible mechanism linking them to hypoglycemia (Table 1).
We identified hospitalization or emergency department visits due to hypoglycemia within 14 days of antimicrobial exposure, using validated algorithms.8,31 We also measured all Medicare payments for emergency department services, hospitalizations, and professional services associated with the hypoglycemia.
We first calculated the rates of hypoglycemia in the 14 days after filling the prescriptions for each of the 16 antimicrobial drugs in patients using either glyburide or glipizide. We chose 14 days because more than 90% of the antimicrobial prescriptions were for 14 days or less. We next examined the association of each of the drugs with an emergency department visit or hospitalization for hypoglycemia. We used a repeated measure logistic regression to account for multiple episodes of antibiotic use within 1 patient. The analyses controlled for age, sex, race/ethnicity, Medicaid eligibility at the year of antimicrobial use, comorbidity, any prior emergency department visits due to hypoglycemia, any acute hospitalization for any cause in the prior year, nursing home residence, and the indication of antimicrobial use. We used azithromycin, amoxicillin, and cephalexin as comparison drugs because they were the 3 most commonly used antibiotics with no link to hypoglycemia. In additional analyses, we estimated the odds of hypoglycemia for each of the drugs in group 1 using all of the drugs in group 2 combined as the control. We also estimated the number needed to harm for each antimicrobial for which there were significantly increased odds of hypoglycemia.34 95% Confidence intervals of the numbers needed to harm were estimated using the bootstrap method with 1000 bootstrap samples.34
To estimate the excess Medicare costs from each prescription of an interacting antimicrobial agent given to a patient taking sulfonylureas, we summed all Medicare costs for hypoglycemia treatment after prescription of ciprofloxacin, clarithromycin, levofloxacin, metronidazole, or sulfametoxazole-trimethoprim and expressed this as average cost per prescription. We then calculated the average cost after prescription for any of the noninteracting antibiotics. The difference between the 2 average costs was the excess cost attributed to the interacting antibiotics. All analyses were performed by one of us (Y.L.L.) using SAS statistical software (version 9.2; SAS Inc).
There were 68 186 episodes of an overlapping prescription for glipizide with 1 of the 16 antimicrobial agents in 31 184 patients, and 65 349 episodes involving glyburide and an antimicrobial in 30 411 patients. In more than 90% of the instances, the number of days of overlap was at least 3. Table 2 lists the number of exposures to each of the 16 antimicrobials and the rate of emergency department visits or hospitalization for hypoglycemia in the 14 days following the antibiotic prescription filling. The rate of hypoglycemic episodes varied from 0.17% to 1.44% in glipizide users and 0.32% to 1.87% in glyburide users after treatment with 1 of the 16 antimicrobials.
Table 3 presents the results of multivariable analyses estimating the odds of a hypoglycemic event within 14 days after filling a prescription for 1 of the 16 antimicrobial agents in patients taking either glipizide or glyburide, adjusted for patient characteristics. We present 3 models for each sulfonylurea, using either azithromycin, amoxicillin, or cephalexin as the comparison antimicrobial. Of the antimicrobials in group 1, clarithromycin, levofloxacin, and sulfamethoxazole-trimethoprim were associated with significantly higher odds of hypoglycemia in all 6 models with the 3 different control antimicrobials. Fluconazole and moxifloxacin were not significantly associated with hypoglycemia in any of the 6 models. Ciprofloxacin and metronidazole were associated with higher odds of hypoglycemia in all 6 models, but not all differences were statistically significant. None of the 9 drugs in group 2 were associated with significantly higher odds of hypoglycemia in any of the 6 models.
We combined glipizide and glyburide users for further analyses because no consistent differences were found between the 2 sulfonylureas in Table 3. Table 4 presents a multivariable model in which the comparison group was all the antimicrobials in group 2 of Table 3. In this analysis, ciprofloxacin, clarithromycin, levofloxacin, metronidazole, and sulfamethoxazole-trimethoprim were significantly associated with hypoglycemia, whereas moxifloxacin and fluconazole were not. The number needed to harm ranged from 71 prescriptions (95% CI, 43-157 prescriptions) for clarithromycin to 334 prescriptions (95% CI, 223-595 prescriptions) for ciprofloxacin. Other factors associated with hypoglycemia after use of an antimicrobial include increasing age, female sex, black or Hispanic race/ethnicity, higher comorbidity, prior episodes of hypoglycemia, and prior hospitalizations. There were no significant interactions between patient age, sex, or race/ethnicity, and use of an interacting vs noninteracting antimicrobials, on odds of subsequent hypoglycemia. We have included an eTable in the Supplement that presents the analyses in Table 4 stratified by sex.
We repeated the analyses in Table 4, restricting the outcome to hospitalization for hypoglycemia. We also conducted analyses stratifying by indication for the antimicrobial agent, or after excluding patients with renal disease or patients prescribed insulin during the year. In all cases, the results were similar to those shown in Table 4.
Of the episodes associated with any hypoglycemic event after antibiotic administration, 39.8% were associated with hypoglycemic hospitalizations and 60.2% with only emergency department visits. There were 9 deaths during hospitalization for hypoglycemia after an overlapping prescription for 1 of the 5 antimicrobial agents significantly associated with hypoglycemia (out of 54 028 overlapping episodes) and 3 deaths after an overlapping prescription for the noninteracting antimicrobials (74 481 episodes) (P = .02 by χ2 test).
Table 5 presents patient-level analyses. There were 140 174 patients who were prescribed glipizide or glyburide in 2009, and 28.3% received at least 1 of the 5 antibiotics associated with increased risk of hypoglycemia. Those 140 174 patients had 5541 episodes of hypoglycemia from any cause requiring an emergency department visit and/or hospitalization, or 3.9 episodes per 100 patients per year. Of those hypoglycemic episodes 13.2% were preceded by a prescription for 1 of the 5 interacting antimicrobials.
Finally, we calculated the costs to Medicare associated with hypoglycemia after prescription of 1 of the 5 interacting antimicrobial drugs. Compared with the noninteracting antimicrobials, the excess Medicare payments for the emergency department and hospital treatment of hypoglycemia after prescription of 1 of the 5 antimicrobials was $30.54 per prescription. The 140 174 patients prescribed sulfonylureas in 2009 filled 69 537 prescriptions for 1 of the 5 interacting antimicrobials that overlapped with a sulfonylurea. This totaled approximately $2 124 000 in additional Medicare costs from treatment of subsequent hypoglycemia, compared with prescriptions of a noninteracting antibiotic.
Our report adds to a growing literature documenting risk of hypoglycemia after certain antibiotics are given to patients prescribed sulfonylureas. Schelleman et al8 examined Medicaid data and found that in glipizide users, sulfamethoxazole-trimethoprim, clarithromycin, fluconazole, and levofloxacin were associated with 2- to 3-fold higher odds of an episode of severe hypoglycemia compared with patients using cephalexin. In glyburide users, clarithromycin, levofloxacin, sulfamethoxazole-trimethoprim, fluconazole, and ciprofloxacin were associated with 2- to 5-fold higher odds of an episode of severe hypoglycemia.8 Juurlink et al9 found that patients who were prescribed glyburide and were admitted to the hospital for hypoglycemia were 6 times more likely to be treated with sulfamethoxazole-trimethoprim in the previous week.
Our analyses generally confirm those of earlier reports, assessing the 16 most commonly prescribed antimicrobials. We estimate that 13.2% of all hypoglycemic events in patients prescribed sulfonylureas were associated with 1 of 5 interacting antimicrobials. On average, each prescription of an interacting antimicrobial was associated with $30 in additional in Medicare costs for subsequent hypoglycemia, which in some cases is more than the cost of the drug.
Recently, a large population-based cohort study19 of patients in Taiwan found that diabetic patients prescribed moxifloxacin had higher rates of hypoglycemia than patients given macrolides. Moxifloxacin did not have a significant association with increased hypoglycemic events in the current study or in a prior study.21 The genetics of cytochrome P450 may differ between Asian and European populations, which may explain the discrepant results.35,36 We also did not replicate the earlier finding of fluconazole’s association of hypoglycemia.8 Fluconazole is an antifungal agent, as are none of the control antimicrobial agents, which calls into question whether the fluconazole and control groups were comparable in underlying risk of hypoglycemia.
Our study had limitations. It is possible that the acute infection was responsible both for the antibiotic prescription and hypoglycemia, with no causal connection between the latter 2 events. We feel that using noninteracting antibiotic drugs as controls and also controlling for the indication for the antibiotic reduced this possibility. In addition, we controlled for comorbidity, prior episodes of hypoglycemia, prior hospitalizations, and other factors that might have confounded any association between an antimicrobial drug and subsequent hypoglycemia. We also performed sensitivity analyses deleting patients with renal disease or those who were using insulin. None of these factors changed the association. It is difficult to obtain data on serious drug interactions from randomized clinical trials, which tend to exclude patients with multiple comorbidities and polypharmacy.
Another limitation is that hospitalization and emergency department visits represent a minority of total hypoglycemic events.1 The current study did not assess hypoglycemic events treated without formal medical interventions; those treated in outpatient settings such as urgent care clinics; and those treated in emergency departments and hospitals but coded with diagnoses such as syncope or fall. Thus, the estimated numbers needed to harm for the 5 antibiotics are conservative.
The study also lacks information on the acute or chronic health effects of hypoglycemia. Hypoglycemia acutely can cause myocardial infarction, stroke, and death.1,4 Repeated hypoglycemia events contribute to cognitive decline, depression, and lower quality of life.4- 6 We also lacked information on the level of control of diabetes mellitus. Our study was limited to Texas and excludes persons younger than 66 years, those in Medicare HMOs, and those without Part D coverage. Use of antibiotics is somewhat higher in the South than in other regions.36,37 Approximately 50% of Texas Medicare enrollees had Part D coverage in 2009. Also, we cannot determine adherence to either the antimicrobial or sulfonylureas, and there may be confounders not available in claims data. It is reassuring; however, that none of the known confounders substantially altered the associations when included in the analyses.
While there is general recognition that adverse drug reactions are common and serious in elderly patients, estimates of the incidence of such events vary substantially, depending on the methodology used.9,38- 42 Methods using administrative data, such as the one used herein, have the advantage of large populations available for study but would miss milder but still clinically significant manifestations of toxicity.
The initial report on sulfamethoxazole-trimethoprim interacting with glyburide was published in 2003 in JAMA9 and is also noted in commonly used drug references such as Micromedex24 and ePocrates.43 This might have been expected to lead to greater use of noninteracting antimicrobial medications. However, more than 10% of sulfonylurea users were prescribed this drug in 2009 (Table 4). One reason may be that it is inexpensive. However, after factoring in the excess Medicare expenses from hypoglycemia, the cost is considerably greater.
Interactions with certain antibiotics are a major cause of hypoglycemia in older patients with diabetes mellitus treated with sulfonylureas. Hospitalizations for hypoglycemia are now more common than for hyperglycemia and are associated with higher resulting morbidity rates.44,45 Greater efforts are required to limit the use of these antibiotics in this population.
Accepted for Publication: May 31, 2014.
Corresponding Author: James S. Goodwin, MD, Sealy Center on Aging, Departments of Internal Medicine, Preventive Medicine, and Community Health, University of Texas Medical Branch, 301 University Blvd, Galveston, TX 77555-0177 (firstname.lastname@example.org).
Published Online: September 1, 2014. doi:10.1001/jamainternmed.2014.3293.
Author Contributions: Dr Parekh had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Parekh, Raji, Lin, Tan, Goodwin.
Acquisition, analysis, or interpretation of data: Raji, Lin, Tan, Kuo, Goodwin.
Drafting of the manuscript: Parekh, Raji, Lin, Goodwin.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Tan, Kuo.
Obtained funding: Goodwin
Administrative, technical, or material support: Parekh.
Study supervision: Raji, Kuo, Goodwin.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was supported by grants from the Agency for Health Care Research and Quality (AHRQ) (1R24HS022134-0) and by the National Institutes of Health (NIH) (K05 CA134923, R01 AG033134, P30 AG024832, and UL1TR000071).
Role of the Sponsors: The AHRQ and NIH had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.