Antibiotic-Specific Risk for Community-Acquired Clostridioides difficile Infection in the United States from 2008 to 2020

ABSTRACT

Antibiotic exposure is a crucial risk factor for community-acquired Clostridioides difficile infection (CA-CDI). However, the relative risks associated with specific antibiotics may vary over time, and the absolute risks have not been clearly established. This is a retrospective cohort study. Adults were included if they received an outpatient antibiotic prescription within the IBM MarketScan databases between 2008 and 2020. The primary exposure was an outpatient antibiotic prescription, and the receipt of doxycycline was used as the reference comparison. The primary outcome was CA-CDI, defined as the presence of an International Classification of Diseases (ICD) diagnosis code for CDI within 90 days of receiving an outpatient antibiotic prescription, and subsequent treatment for CDI. There were 36,626,794 unique patients who received outpatient antibiotics, including 11,607 (0.03%) who developed CA-CDI. Relative to doxycycline, the antibiotics conferring the highest risks for CA-CDI were clindamycin (adjusted odds ratio [aOR], 8.81; 95% confidence interval [CI], 7.76 to 10.00), cefdinir (aOR, 5.86; 95% CI, 5.03 to 6.83), cefuroxime (aOR, 4.57; 95% CI, 3.87 to 5.39), and fluoroquinolones (aOR, 4.05; 95% CI, 3.58 to 4.59). Among older patients with CA-CDI risk factors, nitrofurantoin was also associated with CA-CDI (aOR, 3.05; 95% CI, 1.92 to 4.84), with a smaller number needed to harm, compared to the fluoroquinolones. While clindamycin, cefuroxime, and fluoroquinolone use declined from 2008 to 2020, nitrofurantoin use increased by 40%. Clindamycin was associated with the greatest CA-CDI risk, overall. Among older patients with an elevated baseline risk for CA-CDI, multiple antibiotics, including nitrofurantoin, had strong associations with CA-CDI. These results may guide antibiotic selection and future stewardship efforts.

INTRODUCTION

Clostridioides difficile infection (CDI) is a continuing public health concern, with a national burden of nearly 500,000 cases and 15,000 to 30,000 deaths per year (1–4). Whereas CDI was once predominantly health care-associated, there is a rising incidence in community-acquired CDI (CA-CDI) (5–7), which comprises 20 to 30% of all of the CDI cases in North America and Europe (8) and approximately half of the CDI burden in the United States (4). Despite the increasing incidence of CA-CDI, its epidemiology and risk factors remain less well-understood than those of health care-associated CDI (9, 10).

Antibiotic exposure has a uniquely important role among the CA-CDI risk factors. Antibiotics are the risk factor most highly associated with CA-CDI and, unlike other crucial CA-CDI risk factors, such as age, are modifiable (11, 12). Even in situations when antibiotics cannot be withheld, low-risk antibiotics could often be reasonably substituted for high-risk antibiotics, especially among the patients who are at the greatest risk for CA-CDI. However, these clinical decisions can only be made effectively if the relationships between specific antibiotics and CA-CDI are known.

Prior studies have found that clindamycin, fluoroquinolones, and later-generation cephalosporins confer the greatest risk for CDI, whereas nitrofurantoin, macrolides, penicillins, and sulfonamides/trimethoprim confer lower risk (5, 13–15). Tetracycline antibiotics (specifically doxycycline) have typically been associated with little to no risk for CDI (16). However, the relative risks associated with these antibiotic classes have varied across prior studies, their absolute risks have not been clearly established, and whether they have changed over time remains largely unknown. As outpatient antibiotic use and microbial ecology inevitably change within the population at large, the fraction of CA-CDI that is associated with specific antibiotics also has the potential to change. Moreover, many population-based CDI studies have focused exclusively on patients who were ≥65 years old and/or hospitalized patients.

This study sought to define the relationship between specific antibiotics and the risk for CA-CDI, utilizing a large commercial insurance database with a national ascertainment of outpatient prescription records for adults ≥18 years old through 2020 (17). We aimed to evaluate the risk associated with CA-CDI for specific antibiotics, compared to a doxycycline reference group, to inform outpatient antibiotic prescription choices in light of CA-CDI.

RESULTS

Population.

The final study population consisted of 36,624,794 continuously enrolled, unique patients who received an outpatient antibiotic prescription, 11,607 (0.03%) of whom developed CDI within 90 days after receiving the antibiotic. The flow of patients into the study is shown in Fig. 1, and the patient characteristics are shown in Table 1. Compared to patients without CDI, patients with CDI were more likely to be older (25.5% versus 7.7% over the age of 65), be female (62.1% versus 56.3%), have comorbidities (40.7% versus 19.1% with a Charlson comorbidity index of >0), have had a recent hospitalization (17.1% versus 2.6%), and have received additional antibiotics (28.3% versus 16.9% receiving at least one additional antibiotic).

FIG 1.

Flow of patients into the study.

TABLE 1.

Characteristics of patients within the MarketScan database who received outpatient antibiotics from 2008 to 2020, stratified based on the subsequent development of C. difficile infection (CDI)

Variable	No. of infections (N, %)	C. difficile infections (N, %)	90-day CDI rate per 10,000 antibiotic prescriptions
Total	36,615,187	11,607	3.17
Age (yrs)
18–34	12,710,653 (34.7%)	2,199 (18.9%)	1.73
35–44	7,062,692 (19.3%)	1,467 (12.6%)	2.08
45–54	7,591,155 (20.7%)	2,203 (19.0%)	2.90
55–64	6,418,140 (17.5%)	2,781 (24.0%)	4.33
≥65	2,832,547 (7.7%)	2,957 (25.5%)	10.43
Sex
Male	15,999,198 (43.7%)	4,404 (37.9%)	2.75
Female	20,615,989 (56.3%)	7,203 (62.1%)	3.49
Charlson comorbidity index
0	29,617,711 (80.9%)	6,881 (59.3%)	2.32
1	4,417,844 (12.1%)	1,869 (16.1%)	4.23
2	1,516,818 (4.1%)	1,152 (9.9%)	7.59
≥3	1,062,814 (2.9%)	1,705 (14.7%)	16.02
Region
Northeast	6,485,903 (17.7%)	2,543 (21.9%)	3.92
North Central	8,469,842 (23.1%)	3,257 (28.1%)	3.84
South	15,356,370 (41.9%)	3,755 (32.4%)	2.44
West	6,303,072 (17.2%)	2,052 (17.7%)	3.25
Hospitalization within the prior 90 days
No	35,660,721 (97.4%)	9,618 (82.9%)	2.70
Yes	954,466 (2.6%)	1,989 (17.1%)	20.80
Additional antibioticsa
0	30,394,993 (83.0%)	8,323 (71.7%)	2.74
1	5,472,349 (14.9%)	2,767 (23.8%)	5.05
≥2	747,845 (2.0%)	517 (4.5%)	6.91

^aAll patients in this study received at least 1 antibiotic. Patients with 0 additional antibiotics received 1 antibiotic in total. Patients with 1 additional antibiotic received 2 distinct antibiotics in total. Patients with ≥2 additional antibiotics received ≥3 distinct antibiotics total.

Use of outpatient antibiotics over time.

Prescription rates for the top antibiotics within MarketScan fluctuated throughout the 13-year timespan of this study, even after adjusting for the total number of participants within the data set (Fig. 2). Amoxicillin was the most prescribed antibiotic at over 200 prescriptions per 1,000 person-years for every year of the study, followed by azithromycin and fluoroquinolones. Doxycycline and clindamycin ranged between 50 and 100 prescriptions per 1,000 person-years, with nitrofurantoin at just under 50. From 2008 to 2020, the greatest increases in antibiotic prescription rates were seen in nitrofurantoin (+40.2%), doxycycline (+35.3%), and cefdinir (+15.7%). The greatest decreases in antibiotic prescription rates were seen in clarithromycin (−82.7%), fluoroquinolones (−63.0%), and penicillin VK (−61.2%). Prescriptions for clindamycin decreased by 7.7% from 2008 to 2020.

FIG 2.

Use of outpatient antibiotics within the MarketScan database from 2008 to 2020. These figures describe the use of outpatient antibiotics within the MarketScan database from 2008 to 2020, adjusted for the total number of MarketScan participants. The top panel presents these data in absolute terms (prescriptions per 1,000 patient-years over time). The bottom panel facilitates the visualization of trends by presenting the same data in relative terms, referenced to 2008 (prescription rate of each antibiotic, relative to its own 2008 rate).

Risk factors for C. difficile infection.

A total of 11,607 patients developed CDI within 90 days of receiving an antibiotic prescription. There was an increased risk for CA-CDI associated with older age groups (Table 1; Table S1), with the highest risk being found among patients over the age of 65 (adjusted odds ratio [aOR], 3.23; 95% confidence interval [CI], 3.04 to 3.44). There was also increased risk for CA-CDI associated with the presence of comorbidities, recent hospitalization, and the receipt of additional antibiotics. Compared to the northeast region, the north central, south, and west regions were associated with lower CA-CDI risk.

Antibiotics and risk for C. difficile infection.

Fluoroquinolones, amoxicillin, and clindamycin were the antibiotics associated with the greatest number of CDI cases (Table 2). After adjusting for other factors, the antibiotics most strongly associated with an increased risk for CA-CDI were clindamycin (aOR, 8.81; 95% CI, 7.76 to 10.00, relative to doxycycline), cefdinir (aOR, 5.86; 95% CI, 5.03 to 6.83, relative to doxycycline), cefuroxime (aOR, 4.57; 95% CI, 3.87 to 5.39 relative to doxycycline), and fluoroquinolones (aOR, 4.05; 95% CI, 3.58 to 4.59 relative to doxycycline) (Fig. 3; Table 3). Medium-risk antibiotics included amoxicillin, cephalexin, and nitrofurantoin. Low-risk antibiotics included clarithromycin, penicillin VK, and azithromycin. Among patients who did not receive any additional antibiotics following their first prescription (i.e., those who received one and only one antibiotic), the hierarchy of relative risk was unchanged, with clindamycin (aOR, 11.72; 95% CI, 9.87 to 13.92 relative to doxycycline) remaining the antibiotic most strongly associated with CA-CDI risk (Table S2). In terms of the absolute risk associated with CA-CDI, the antibiotics with the highest 90-day incidence were clindamycin (incidence of 9.74/10,000), fluoroquinolones (incidence of 6.62/10,000), cefuroxime (incidence of 6.48/10,000), and cefdinir (incidence of 6.22/10,000) (Table 2). Relative to doxycycline, these four high-risk antibiotics all had high attributable risks and low numbers needed to harm [NNH]: clindamycin (NNH of 1,164, relative to doxycycline), fluoroquinolones (NNH of 1,828, relative to doxycycline), cefuroxime (NNH of 1,876, relative to doxycycline), and cefdinir (NNH of 1,972, relative to doxycycline).

TABLE 2.

Crude relationships between the most prescribed outpatient antibiotics and C. difficile infection^a

Antibiotic	Total prescriptions (N)	C. difficile infections (N, %)	90-day CDI rate per 10,000 prescriptions	Attributable risk per 10,000 prescriptions relative to doxycycline	No. needed to harm relative to doxycycline
Any antibiotic	36,626,794	11,607 (100%)	3.17	—	—
Clindamycin	2,205,494	2,149 (18.5%)	9.74	8.59	1,164
Fluoroquinolones	5,430,988	3,596 (31.0%)	6.62	5.47	1,828
Cefuroxime	459,679	298 (2.6%)	6.48	5.33	1,876
Cefdinir	683,161	425 (3.7%)	6.22	5.07	1,972
Cephalexin	3,261,406	908 (7.8%)	2.78	1.63	6,135
Nitrofurantoin	1,176,060	300 (2.6%)	2.55	1.40	7,143
Amoxicillin	10,575,062	2,669 (23.0%)	2.52	1.37	7,299
Clarithromycin	559,260	92 (0.8%)	1.65	0.50	20,000
Penicillin VK	1,365,023	171 (1.5%)	1.25	0.10	100,000
Doxycycline	2,333,933	268 (2.3%)	1.15	Reference	Reference
Azithromycin	8,576,728	731 (6.3%)	0.85	−0.30	—

^aDashes indicate that there is no calculable value for the category of antibiotic.

FIG 3.

Relationship between the exposure to outpatient antibiotics and the risk for C. difficile infection within 90 days. Adjusted odds ratios for CA-CDI for each of the 10 antibiotics studied. Results are adjusted for age, sex, Charlson comorbidity index, region, hospitalization within the prior 90 days, and additional antibiotics.

TABLE 3.

Relationship between exposure to outpatient antibiotics and odds of C. difficile infection within 90 days

Antibiotic	Unadjusted odds ratio (95% CI)	Adjusted odds ratioa (95% CI)
Doxycycline	Reference	Reference
Clindamycin	8.49 (7.48, 9.64)	8.81 (7.76, 10.00)
Cefdinir	5.42 (4.65, 6.32)	5.86 (5.03, 6.83)
Cefuroxime	5.65 (4.79, 6.66)	4.57 (3.87, 5.39)
Fluoroquinolones	5.77 (5.10, 6.53)	4.05 (3.58, 4.59)
Amoxicillin	2.20 (1.94, 2.49)	2.29 (2.02, 2.60)
Cephalexin	2.43 (2.12, 2.78)	1.97 (1.71, 2.25)
Nitrofurantoin	2.22 (1.88, 2.62)	1.92 (1.63, 2.26)
Clarithromycin	1.43 (1.13, 1.82)	1.31 (1.03, 1.66)
Penicillin VK	1.09 (0.90, 1.32)	1.21 (1.00, 1.47)
Azithromycin	0.74 (0.65, 0.85)	0.77 (0.67, 0.88)

^aAdjusted for age, sex, Charlson comorbidity index, region, hospitalization within the prior 90 days, and additional antibiotics.

High-risk stratum.

To facilitate clinical decision-making regarding antibiotic selection and the risk for CA-CDI, a restriction analysis was performed that included only the highest-risk stratum of patients who were ≥65 years old and were either hospitalized within 90 days before their antibiotic prescription or had a Charlson comorbidity index of ≥3. After adjusting for other factors, clindamycin (aOR, 4.40; 95% CI, 2.85 to 6.81, relative to doxycycline) remained the highest-risk antibiotic within this cohort, followed closely by cefdinir and cefuroxime (Table S3). However, the risk associated with nitrofurantoin (aOR, 3.05; 95% CI, 1.92 to 4.84 relative to doxycycline) was now similar to that of the fluoroquinolones. The antibiotic with the highest absolute risk for CA-CDI was clindamycin, followed by cefuroxime, cefdinir, and nitrofurantoin (NNH of 403, relative to doxycycline) (Table 4).

TABLE 4.

Crude relationships between the most prescribed outpatient antibiotics and C. difficile infection among patients with multiple risk factors for C. difficile^a

Antibiotic	Total prescriptions (N)	C. difficile infections (N, %)	90-day CDI rate per 10,000 prescriptions	Attributable risk per 10,000 prescriptions relative to doxycycline	No. needed to harm relative to doxycycline
Any antibiotic	550,169	1,391 (100%)	25.28	—	—
Clindamycin	20,967	92 (6.6%)	43.88	34.18	293
Cefuroxime	10,551	45 (3.2%)	42.65	32.95	303
Cefdinir	5,915	25 (1.8%)	42.27	32.57	307
Nitrofurantoin	17,393	60 (4.3%)	34.50	24.80	403
Fluoroquinolones	185,369	618 (44.4%)	33.34	23.64	423
Amoxicillin	105,349	238 (17.1%)	22.59	12.89	776
Cephalexin	73,875	166 (11.9%)	22.47	12.77	783
Clarithromycin	5,883	9 (0.6%)	15.30	5.60	1,786
Penicillin VK	7,969	10 (0.7%)	12.55	2.85	3,509
Azithromycin	90,100	102 (7.3%)	11.32	1.62	6,173
Doxycycline	26,798	26 (1.9%)	9.70	Reference	Reference

^aThe risk factors used to define this high-risk stratum were age of ≥65 and either a Charlson comorbidity index of ≥3 or a hospitalization within the prior 90 days. Dashes indicate that there is no calculable value for the category of antibiotic.

Additional analyses.

To account for possible CA-CDI case misclassification due to hospitalization, we conducted two additional analyses with revised case definitions. In our first analysis, we reclassified patients who were prescribed an outpatient antibiotic and were subsequently hospitalized with a primary diagnosis of CDI as CA-CDI patients (Table S4). In our second analysis, we reclassified patients who received an interim hospitalization between their antibiotic prescription date and their outpatient CDI diagnosis date as non-CA-CDI patients (Table S5). In both analyses, our main findings were substantively unchanged, both for the full cohort and for the high-risk cohort.

DISCUSSION

This large, retrospective cohort study estimated the risk of specific antibiotics for CA-CDI among adults who received outpatient antibiotics from 2008 to 2020. Some of these estimates were surprising and differed from conventional thinking regarding antibiotics and CA-CDI. Overall, clindamycin remained as the antibiotic with the highest overall risk for CA-CDI, as in the past, followed by cefdinir, cefuroxime, and fluoroquinolones. Among those who were ≥65 years old with CDI risk factors (medical comorbidities or a recent hospitalization), clindamycin, cefuroxime, and cefdinir remained the top three antibiotics that were most associated with CA-CDI. Surprisingly, nitrofurantoin was also associated with substantial excess risk within this subpopulation. The number needed to harm was approximately 300 for clindamycin, cefuroxime, and cefdinir and approximately 400 for nitrofurantoin and fluoroquinolones, sufficiently low to affect antibiotic choices in situations where there is clinical latitude. Importantly, while clindamycin, cefuroxime, and fluoroquinolone use declined during the years of the study, nitrofurantoin use increased by 40%. These results revise older estimates for the antibiotic-specific risks for CA-CDI and may guide future decision-making in terms of antibiotic selection, especially among older patients who have a high baseline risk for CA-CDI.

With respect to clindamycin, our main results are consistent with those of prior meta-analyses (5, 11, 14, 18) and cohort studies (13, 15, 19), which have ranked clindamycin as the highest-risk antibiotic for CA-CDI for decades (5, 11) while placing macrolides, penicillins, and tetracyclines among the lower-risk classes (16). Despite the high relative risks of clindamycin, it was associated with far fewer cases of CA-CDI, compared to more commonly used antibiotic classes, such as fluoroquinolones. Therefore, our results for clindamycin are consistent with those of previous studies, emphasizing its risk for CA-CDI. Additionally, these results put this risk in an appropriate clinical perspective. If one of the goals of antibiotic stewardship is the prevention of CA-CDI, effective stewardship cannot focus solely on the reduction of clindamycin use.

Nitrofurantoin was associated with a 90% increase in the risk for CA-CDI, relative to doxycycline, within the complete cohort and an over 200% increase in risk, relative to doxycycline, among patient subpopulations with high baseline rates for CA-CDI. Among these high-risk patients, nitrofurantoin conferred the fourth-highest risk for CA-CDI, surpassing the fluoroquinolones. Importantly, the association between nitrofurantoin and CA-CDI remained after adjusting for additional antibiotics and when the analysis was restricted to those who received antibiotic monotherapy. This finding differs from those of most of the prior population-based studies, which have identified nitrofurantoin as one of the antibiotics associated with the least risk for CDI. Prior studies focused on nitrofurantoin have largely been conducted within specialized cohorts, such as ambulatory urology clinics (13, 20, 21), and differences in our findings compared to these studies could be related to differences in the patient populations or in center-specific practice patterns.

Nitrofurantoin has preferential urinary excretion and achieves high urinary concentrations in humans. Few prior studies have examined the effect of nitrofurantoin on the human fecal microbiota. Vervoort et al. found that modest microbiota changes were detectable in human stool up to 43 days after nitrofurantoin exposure (22). In poultry, low dose nitrofurans are given widely as medicated feeds to prevent avian Salmonella. In medicated chickens, nitrofuran derivates are found at up to 330 μg/kg within meat and gizzards, suggesting that there is at least some penetration into the gut (23). (The minimum inhibitory concentration [MIC] of nitrofurantoin for E. coli is 4 to 16 μg/mL.). Our study may spark a greater interest related to the impact of nitrofurantoin on the gut microbiota. In these data, the use of nitrofurantoin rose by 40% during the 13 years of the study, so the awareness of the potential risk is important for this increasingly popular antibiotic. There was also some evidence of effect modification (i.e., more harm among high-risk patients), which may reflect differing biological effects of the drug across patient types. The gut microbiome and C. difficile colonization resistance differ in older, recently hospitalized patients, compared to young, healthy patients (24, 25). Nitrofurantoin, which has activity against both Gram-positive and Gram-negative bacteria, may have a larger impact on the gut microbiome in the former, but not the latter, patients, with a downstream effect on the risk for CA-CDI. In studies that have directly tested the effect of nitrofurantoin on the gut microbiome, nitrofurantoin had a modest impact (22, 26). However, these studies do not account for differences between young versus older populations. Future studies could investigate this hypothesis by directly testing the effects of nitrofurantoin on the gut microbiome in differing patient populations.

This study, which was nested within the large MarketScan population, also produced precise estimates for the absolute excess risk for CA-CDI associated with specific antibiotics. Prior studies have often used a case-control design to examine antibiotics and CA-CDI, which precludes the estimation of the absolute risk and the number needed to harm (27). Our study design allowed us to calculate the excess risk for antibiotics, relative to doxycycline. In the entire patient cohort, the numbers needed to harm for most of the antibiotics were relatively large; for example, 1,164 patients treated with clindamycin instead of doxycycline to cause one excess case of CA-CDI. Within an important high-risk patient subset, the numbers needed to harm dropped considerably and were in a range that may impact clinical decision-making: 403 for nitrofurantoin, 307 for cefdinir, 303 for cefuroxime, and 293 for clindamycin. These data may encourage providers to consider the use of lower-risk antibiotics when patients have multiple CDI risk factors. Low-risk antibiotic substitution would also be reasonable among patients who have a history of recurrent C. difficile infection, although such patients were not included in this study.

Our study had several strengths. The MarketScan database covers a large patient population, spanning the entire United States, and was current through 2020 so that we could update the results of prior studies, many of which were completed over a decade ago. The use of doxycycline as the comparator ensured that all patients in our cohort were being treated for some type of infection, rather than comparing antibiotic recipients to nonantibiotic recipients. However, the study also had limitations. We were unable to differentiate between durations of antibiotics, nor did we attempt to adjust for antibiotic indication (which is likely to be highly correlated with antibiotic type). Another limitation is that we did not entirely account for frequently used combinations of antibiotics, a common limitation of CDI studies and one that is interwoven with the thorny issue of antibiotic indications (13, 28). An analysis of those who received antibiotic monotherapy was in line with the main results and provides some reassurance in this regard. Last, there is the possibility of case misclassification related to hospitalization, although additional analyses suggested that our results were robust when CA-CDI was operationalized in an alternative manner.

In summary, this population-based, retrospective study found that clindamycin remained, overall, as the antibiotic most strongly associated with CA-CDI in the United States from 2008 to 2020, followed by cefdinir, cefuroxime, and the fluoroquinolones. Among the older patients at the highest risk for CDI, nitrofurantoin was associated with an over 3-fold relative risk. These results may inform clinicians in their outpatient antibiotic prescribing choices and may guide future antibiotic stewardship that is seeking to mitigate CA-CDI.

MATERIALS AND METHODS

Data source.

This study used the IBM Watson Health MarketScan Research Databases (29, 30) (formerly Truven), which provide deidentified claims-based health care data. For this study, we utilized two of the six MarketScan databases: (i) the Commercial Claims and Encounters (CCAE) database and (ii) the Medicare Supplemental and Coordination of Benefits (MDCR) database. The CCAE database contains over 40 million unique patients per year, consisting of active employees, early retirees, Consolidated Omnibus Budget Reconciliation Act (COBRA) continuees, and dependents who are ineligible for Medicare. The MDCR database contains over 4 million unique patients per year, consisting of Medicare-eligible retirees with employer-sponsored Medicare Supplemental plans.

Study design and population.

This was a retrospective cohort study. Adult patients (age of ≥18 years) were initially included if they filled a prescription for any one of the most commonly used outpatient oral antibiotics within MarketScan from 2008 to 2020. They were then excluded if they were missing an enrollee ID (a unique identifier for each patient), a service date, a relevant drug code, or a region. Additionally, patients were excluded if they were not continuously enrolled in either the CCAE database or the MDCR database from 90 days before until 90 days after the prescription index date. For each patient, the first antibiotic prescribed within the study time frame was selected as the primary exposure so that each prescription record represented a unique patient. The primary analysis then examined the relationship between these specific antibiotics and CA-CDI, after adjusting for additional antibiotics. The study was approved by the institutional review board of the Columbia University Irving Medical Center.

Community-acquired C. difficile infection.

The primary outcome was community-acquired C. difficile infection (CA-CDI). CA-CDI was classified as present when any of the ICD diagnosis codes for CDI were recorded (ICD-9 008.45, ICD-10 A04.7, A04.71, or A04.72) within 90 days following the receipt of the first outpatient antibiotic prescription as well as when an outpatient prescription for an anti-CDI antibiotic appeared within a 14-day window that was centered on the CDI diagnosis. Anti-CDI antibiotics included oral vancomycin, metronidazole, or fidaxomicin.

Antibiotics.

The primary exposure was the antibiotic type, compared against doxycycline as a reference. Doxycycline was selected as the reference because it does not alter susceptibility to C. difficile in animals (31) and has not been associated with risk for CDI in prior retrospective studies (5, 13, 16, 32). To determine which antibiotics to study, we generated a list of the 250 most prescribed medications within MarketScan and found that there were 16 distinct antibiotics on this list. From this list of 16 antibiotics, we excluded metronidazole because it was used as a treatment for CDI, trimethoprim-sulfamethoxazole (TMP-SMZ) because it is used for prophylaxis among those who are immunosuppressed (33), and minocycline because it is used long-term for acne and for other skin conditions (34). Additionally, since fluoroquinolones share similar spectra of activity (35), we elected to group ciprofloxacin, levofloxacin, and moxifloxacin into a single category. Our final list constituted 83.1% of all outpatient antibiotic prescriptions within the MarketScan data, consisting of the following 10 antibiotics, plus doxycycline: amoxicillin, azithromycin, fluoroquinolones (ciprofloxacin, levofloxacin, and moxifloxacin), cephalexin, clindamycin, penicillin VK, nitrofurantoin, cefdinir, clarithromycin, and cefuroxime.

Covariates.

6 patient-level covariates that could potentially influence the association between antibiotics and CA-CDI were captured: age (classified as 18 to 34, 35 to 44, 45 to 54, 55 to 64, or ≥65 years), sex (classified as male or female), Charlson comorbidity index (classified as 0, 1, 2, or ≥3 points), region (classified as northeast, north central, south, or west), hospitalization within the prior 90 days (classified as present or absent), and additional antibiotics received during the follow-up period (classified as 0, 1, or ≥2 distinct additional antibiotics). The Charlson comorbidity index (CCI) was determined by aggregating comorbidities, based on ICD codes (36, 37), for the year prior to the antibiotic prescription date. Prior hospitalization was determined by verifying the presence of an inpatient service record within 90 days prior to the index antibiotic prescription date. Additional antibiotics were determined by accumulating any new prescriptions for antibiotics after the index prescription and before 90 days of a follow-up or a CDI diagnosis, whichever occurred first.

Stratified analysis.

In addition to the full-cohort analysis, we performed a stratified analysis to delineate the absolute and relative risks associated with specific antibiotics or antibiotic classes within a preselected high-risk patient stratum. The a priori high-risk stratum was comprised of patients with age ≥65 years and either a prior hospitalization within 90 days or a Charlson comorbidity index of ≥3.

Statistical approach.

Multivariate logistic regression modeling was used to determine the independent risk associated with each antibiotic for CDI, relative to doxycycline. The 90-day incidence rates of CDI were calculated for each antibiotic and were used to assess the absolute risk, attributable risk [AR], and number needed to harm [NNH], relative to doxycycline. The statistical analyses were performed using the SAS software package, version 9.4 (SAS Institute Inc., Cary, North Carolina), using an α value of 0.05 as the threshold for statistical significance.

Data availability.

For more information about the IBM MarketScan Research Databases, please reference https://www.ibm.com/products/marketscan-research-databases. The code implemented in this study is available at https://github.com/jimmyzhang2003/antibiotics-cdi-risk.