Association Between Bisphosphonates and Hospitalized Clostridioides difficile Infection Among Frail Older Adults

Abstract

Objectives:

Clostridioides difficile infection is a major source of morbidity and mortality among frail older adults, especially those in nursing homes (NHs). Safety reports have signaled that bisphosphonate use may be a contributing cause. We therefore evaluated the risk of C difficile hospitalization associated with oral bisphosphonate use in the NH.

Design:

Observational, retrospective new-user cohort study.

Setting:

The cohort included US NH residents aged ≥65 years who became a long-stay resident (>100 days in the NH) between January 1, 2008 and December 31, 2009.

Methods:

We conducted a study of NH residents using linked Medicare claims and Minimum Data Set records. Residents were new users of an oral bisphosphonate 1:1 matched to new calcitonin users (“active” comparator) on propensity scores controlling for more than 100 covariates. The outcome was risk of hospitalization for C difficile infection in a Cox proportional hazards model adjusted for previous antibiotic and proton pump inhibitor use.

Results:

Our final analytical cohort included 17,753 bisphosphonate and 5348 calcitonin users. In the matched cohort, 84/5209 (1.6%) vs 71/5209 (1.4%) C difficile–related hospitalizations occurred in bisphosphonate and calcitonin users, respectively. We observed no significant difference in the risk of hospitalization among bisphosphonate users (hazard ratio: 1.11, 95% confidence interval: 0.80–1.51). Antibiotic and proton pump inhibitor exposure before and after osteoporosis treatment was also similar between bisphosphonate and calcitonin users.

Conclusions and Implications:

C difficile infection should not be a consideration when prescribing bisphosphonates to frail older adults given the lack of a significant association.

Frail older adults bear significant morbidity and mortality from Clostridioides difficile infection (CDI). As the oldest and frailest US subpopulation, nursing home (NH) residents are often at greatest risk because of their age-related physiology and frailty. CDI rates range from 1 to 2 cases per 10,000 patient-days, with most cases happening during the first 30 days following NH admission.^1–4 NH residents with CDI are at increased risk of poor health outcomes compared with community dwellers, including 3 times higher odds of severe disease CDI, hospitalization, longer hospital stays, and disease recurrence.⁵

The primary risk factor for a CDI case is previous antibiotic exposure, particularly recent exposure (ie, in the 3 months prior to infection) to broad-spectrum agents. Older age, white race, presence of a feeding tube, pressure ulcers, and multimorbidity are also associated with increased risk of CDI in frail older adults, especially among NH residents.⁴ Many factors, like antibiotics and other medications (eg, gastric acid suppressants), increase CDI risk by modifying intestinal microflora, which can allow C difficile spores to germinate and progress to an infection with severe, watery diarrhea.

Bisphosphonates (BPs) were previously associated with an unusually high number of “C difficile infection” or “pseudomembranous colitis” adverse drug events in the FDA spontaneous adverse event reporting system.⁶ It was estimated that alendronate users had 2-fold greater odds of reporting a CDI event vs other drug users. BPs possess weak antimicrobial activity but have poor oral bioavailability, and so it is plausible that they could alter the intestinal microflora, predisposing an individual to CDI.^7,8

FDA adverse event reports are spontaneous and voluntarily submitted, so they can be substantially biased because of underreporting of adverse drug reactions and lack of a denominator. Therefore, any safety signal should be confirmed with more rigorous pharmacoepidemiologic studies. Particularly in the NH, limiting risk of nosocomial infections like CDI is an important public health goal. Therefore, it is important to understand the true adverse effects of preventive therapies in order to adequately consider risks and benefits in a frail population.

We evaluated the risk of C difficile hospitalization associated with oral BP use among frail, older adults residing in a NH. To explore the potential mechanism between BPs and CDI, we also compared the incidence of antibiotic use and proton pump inhibitor(PPI) use following osteoporosis medication prescription. We hypothesized that BPs would be associated with an increased risk of use of PPIs and CDI events.

Methods

Data Source and Study Design

We conducted a retrospective new-user cohort study design. The data source included national Medicare parts A and D data linked to the Minimum Data Set (MDS) version 2.0 and Online Survey Certification and Reporting System (OSCAR) data on facility level information. The MDS is a quarterly clinical assessment tool that is required for all NHs certified to receive Medicare or Medicaid funding.⁹ This study was approved by the Brown University institutional review board.

Study Population

The cohort comprised new users of a BP or calcitonin (CA; active comparator, see Exposures section) aged ≥65 years, residing >100 days in a US NH between January 1, 2008 and December 31, 2009, followed through December 31, 2013. The index date was defined as the first eligible dispensing of a BP or CA. Individuals with a dispensing for either in the 365 days before this date were excluded. Study subjects were excluded if they had less than 365 days of continuous enrollment in fee-for-service Medicare or were enrolled in Medicare Advantage at any time. Prior literature contains additional details on the cohort design and exposure definitions.^10,11 Follow-up started on the day of BP or CA dispensing and continued until the first occurrence of Medicare disenrollment, death, occurrence of C difficile hospitalization, or study end (December 31, 2013).

Exposures

CA is approved for the treatment of osteoporosis, and has a different pharmacology profile with no theoretical or known relationship with C difficile, making it a suitable “active” comparator. The effect of initiating BPs vs CA was measured in an “intention to treat” fashion regardless of subsequent treatment.

Outcomes

The primary outcome was hospitalization for C difficile infection, measured using ICD-9 008.45 in any coding position. This definition was 78% sensitive and 99.7% specific compared with toxin assays results.¹² Additionally, as a secondary outcome, we also evaluated prescriptions for antibiotics and PPIs up to 60 days after the incident BP or CA exposure to identify if the incident prescription was associated with traditional risk factors for CDI.

Covariates

In addition to 111 covariates¹¹ including information on demographics, comorbidities, cognitive and functional status, medication use, and facility quality, we measured oral antibiotic and PPI use up to 1 year before the index date. Covariates were obtained from the NH Minimum Data Set assessments prior to the index date of new use, with medication use from Part D administrative claims.

Statistical Analyses

We used propensity scores to match 1 new user of BPs to 1 new user of CA with a 5:1 matching algorithm, where treated cases were iteratively matched to the control person with the nearest propensity score without replacement (ie, “Greedy” algorithm). Our 111 propensity score estimation covariates included demographic characteristics, medical history, medication use, health services utilization, and NH facility characteristics (see previous reports).^10,11 The propensity score model selection was based on overall model classification error (fit), variable importance, and consensus among the coauthors. The final model was selected after examining distributions between treatment groups using histograms and reducing covariate balance by standardized mean differences to a maximum of 0.2. We calculated incidence rates (IRs) of CDI in the matched and unmatched cohorts. We estimated hazard ratios with 95% Huber-White confidence intervals (CIs) using Cox proportional hazards regression models to compare BP vs CA users. Antibiotic use and PPI use in the prior 180 days were included as covariates in the final models, and the relative risk for a new prescription of either was estimated at 60 days after the main exposure.

Results

Our final analytical cohort included 23,101 users, 17,753 BP and 5348 CA.¹¹ Before matching, BP users were younger (83.4 vs 85.5 years), less female (84.6% vs 87.4%), less white (80.2% vs 88.2%), and had less cognitive and functional impairment (Table 1).¹¹ After matching, all but 4 of 111 variables had standardized mean differences of 0.04 or less. The final matched cohort included 5209 NH residents.

Table 1.

Characteristics of Bisphosphonate Users and Calcitonin Users Before and After Propensity Score Matching

Characteristics	Before Matching, %		After Matching, %
	BP (n = 17,753)	CA (n = 5348)	BP (n = 5209)	CA (n = 5209)
Age, y, mean (SD)	83.4 (7.8)	85.5 (7.8)	85.2 (7.7)	85.3 (7.8)
Female sex	84.6	87.4	86.8	87.4
White race	80.2	88.2	88.7	88.0
Time between becoming long-stay resident and index date, d, mean (SD)	303.9 (265.0)	292.8 (262.6)	290.0 (252.0)	293.2 (264.3)
Cognitive function
Intact to mild impairment (CPS 0–2)	50.2	46.7	47.9	47.6
Moderate impairment (CPS 3–4)	44.3	46.4	45.7	45.9
Severe impairment (CPS 5)	5.7	6.7	6.4	6.5
Number of activities of daily living with functional independence (out of 6)
4–5	10.2	8.6	8.3	8.8
2–3	14.3	12.8	12.8	13.0
0–1	75.4	78.6	78.9	78.3
Moderate to severe instability	8.9	11.6	11.0	11.4
Do not resuscitate order	49.1	60.0	59.7	59.6
Do not hospitalize order	1.8	2.4	2.1	2.3
History of falls	47.6	50.6	50.7	51.0
Comorbidities
Number of comorbidities, mean (SD)	10.1 (3.5)	10.7 (3.6)	10.8 (3.6)	10.8 (3.5)
Osteoporosis	14.2	19.0	18.0	18.5
Arthritis	37.5	41.5	42.5	41.6
Diabetes	31.4	30.0	29.9	30.4
Bone disorder	1.3	1.6	1.4	1.6
Renal disease	6.1	8.9	8.8	8.9
Dual-energy X-ray absorptiometry	19.4	6.5	6.7	6.6
Number of medications, mean (SD)	12.4 (4.9)	13.0 (5.2)	12.9 (5.0)	13.0 (5.2)
Oral glucocorticoids	17.9	19.9	20.1	20.1
Loop diuretics	41.1	46.4	47.1	46.3
Opioids	25.1	33.4	32.8	33.3
Non-benzodiazepine hypnotics	13.3	12.4	12.5	12.8
Oral antibiotics	74.4	78.9	77.3	78.7
Proton pump inhibitors	38.6	43.7	43.8	43.3
Number of physician visits, mean (SD)	3.1 (4.0)	2.9 (3.9)	2.9 (3.9)	3.0 (3.9)
Number of hospitalizations, mean (SD)	1.1 (1.4)	1.2 (1.5)	1.2 (1.5)	1.2 (1.5)

CPS, Cognitive Performance Score; SD, standard deviation.

Overall, there were 369 CDI events (1.6%): 296/17,753 in BP users (1.7%) vs 73/5348 in CA users (1.4%) before matching. In the matched cohort, 155 events occurred (1.5%): 84/5209 in BP (1.6%) vs 71/5209 in CA users (1.4%). BP users had a mean exposure time in years of 2.71 (IR, 0.61 per 100 person-years) and 2.51 (IR, 0.64 per 100 person-years) in the unmatched and matched cohort, respectively. CA users had a mean time of 2.31 (IR, 0.59 per 100 person-years) and 2.33 (IR, 0.58 per 100 person-years) in the unmatched and matched cohort, respectively. Table 2 summarizes our findings: we observed a numerically higher hazard of CDI with BP use in each cohort and model specification, but the findings were highly compatible with chance. The Cox proportional hazards ratio was 1.07 (95% CI: 0.83–1.4) for unmatched and 1.11 (95% CI: 0.81, 1.52) for matched cohorts, respectively.

Table 2.

Unadjusted and Adjusted Risk for Clostridioides difficile Hospitalization Among Bisphosphonate (vs Calcitonin) New Users

	Bisphosphonate, Events/n (IR)*	Calcitonin, Events/n (IR)*	Hazard Ratio (95% CI)
Unmatched	296/17,753 (0.62)	73/5348 (0.59)	1.07 (0.84, 1.39)
Matched	84/5209 (0.64)	71/5209 (0.58)	1.11 (0.81, 1.52)

^*Values indicate number of C difficile hospitalization events per group, with the IR per 100 person-years within parentheses.

In the matched cohort, antibiotic exposure up to 1 year prior was 4024/5209 (77.3%) among BP users vs 4101/5209 (78.7%) in CA users (RR 0.98, 95% CI: 0.96–1.00). Antibiotics were prescribed within 60 days of BP or CA prescriptions in 2130/5209 users (40.9%) and 2224/5209 users (42.7%), respectively (RR 0.96, 95% CI: 0.92, 1.00). PPI use up to 1 year prior was 2280/5209 among BP users (43.8%) vs 2257/5209 among CA users (43.3%) (RR 1.01, 95% CI: 0.96–1.05). PPIs were prescribed within 60 days of BP or CA prescriptions in 3494/5209 (67.1%) and 3478/5209 users (66.8%), respectively (RR 0.99, 95% CI: 0.94, 1.04).

Discussion

This national study of frail older adults in NHs did not observe a significant association between BP use and hospitalized CDI in a matched cohort. The prior evaluation of this association used adverse safety reports, which are subject to significant biases because of the nature of voluntary reporting; for example, lack of information on exposures, recall bias, and no denominator for patients exposed. Potential mechanisms that explain a link between BPs and CDI include both noncausal and causal pathways: BPs are commonly used for postmenopausal women who may be at increased risk of CDI based on population-based surveillance studies; however, we adjusted for age, gender, and multiple comorbidities in our analysis.¹³ Additionally, gastric acid suppressants may be used concomitantly with oral BPs to manage reflux symptoms. We found no difference in the receipt of PPIs following BP prescription. BPs also have antimicrobial activity and are poorly bioavailable.^7,8,14 Low bioavailability of alendronate may lead to high gut concentrations in the intestine, predisposing frail older adults to CDI. It should also be noted BPs have complex effects on monocyte cell lines, for example, induction of apoptosis in macrophages, but the relevance of this is unclear.¹⁵

Despite these plausible mechanisms, we did not observe a statistically significant increase in hospitalized CDI among BP users. Our findings include several important caveats: our ascertainment of CDI cases is based on administrative billing codes for hospitalizations, representing the more severe cases of CDI, and may not reflect the entire burden of disease in NHs. Also, although the use of an active comparator may be preferred to avoid treatment selection bias vs an untreated cohort, new CA users may be systematically different from new BP users even after our propensity score matching. The propensity score model also may not be able to provide an unbiased estimate if significant confounding variables are omitted, or a complex and poorly understood causal relationship exists, such as adjusting for common effects of BPs and CDI (ie, collider bias). Although our analysis includes a national cohort of NH residents, our final sample size of 5209 residents per arm may not be large enough to identify small treatment effects of BP medications with CDI, though it is the largest study to date. Alternatively, this population of frail older residents is among the highest at risk of adverse drug events and hospitalization due to infection. The competing risk of death is also important to consider in evaluating causal relationships in frail older adults. This would be an important concern if BP treatment was associated with differential risk of mortality, but in our prior work, adjustment for the competing risk of death only resulted in marginally different treatment effects among BP users and hip fracture.¹⁰ Absence of a strong association in this cohort may suggest low likelihood of finding an association in healthier, community-dwelling older adults.

Conclusions and Implications

In summary, we do not observe a significantly increased risk of C difficile infection in frail older adults using BPs, nor did we observe an increase in antibiotic use or acid-suppressing drug use following BP initiation. C difficile infection should not be a consideration when prescribing BPs to frail older adults given the lack of a significant association. We continue to advocate for evaluating nontraditional and modifiable risk factors to reduce burden of C difficile infection in frail older adults.