The major finding from this FAERS database study is that there are differential reporting rates of hypersensitivity and anaphylactic reactions for the four studied IV iron products; reporting rates were highest for ferumoxytol and iron dextran, intermediate for iron sucrose, and lowest for ferric carboxymaltose. This finding from the national US spontaneous reporting system, FAERS, is consistent with prior real-world studies, both using spontaneous reporting databases and claims databases [4, 8, 15, 16]. In addition, our analysis used the ROR, which is less likely to result in biased estimates than other disproportionality measures , as well as providing new cost estimates associated with these AEs. This type of assessment differs from clinical trials, which are designed to increase the likelihood that drug efficacy signals can be detected. Specifically, most inclusion and exclusion criteria of clinical trials are designed to include a relatively homogenous study population, that may comprise individuals at lower risk of AEs. For example, a head-to-head clinical trial whose primary endpoint was evaluation of safety, and specifically hypersensitivity and anaphylactic reactions, among patients treated with either ferumoxytol or ferric carboxymaltose, excluded patients with a prior history of allergies to drugs that did not include those under investigation . This, together with the small sample sizes in most clinical trials, means that uncommon yet clinically important AEs may not be detected before drug approval. Therefore, careful post-approval monitoring is vital to the ongoing safety evaluation of approved therapies. The FDA professionals and pharmacovigilance experts routinely look to the FAERS data for post-marketing drug safety surveillance and as a guide to and signal generator of drug safety issues .
Approved IV iron preparations are an important component of current therapeutic armamentariums , but are known to be associated with rare and/or serious hypersensitivity reactions and anaphylaxis [4, 5]. Using a disproportionality analysis of FAERS primary suspect cases, we found that all IV iron preparations were associated with signals for overall hypersensitivity and anaphylactic reactions, but ROR05 values were highest for ferumoxytol and iron dextran followed by iron sucrose, with ferric carboxymaltose having the lowest ROR05 values. Signals for death associated with hypersensitivity and anaphylactic reactions were present for ferumoxytol, whereas only one death associated with a hypersensitivity reaction occurred with ferric carboxymaltose. Signals for hospitalization associated with hypersensitivity and anaphylactic reactions were present for ferumoxytol, iron dextran, and iron sucrose, with ferric carboxymaltose having the lowest ROR05 value. Reporting rates were also lowest with ferric carboxymaltose in various supplementary analyses, including an analysis of total suspect case counts and analyses of primary case counts stratified by age group and sex. In addition, these lower reporting rates with ferric carboxymaltose results could not be explained by masking owing to the presence of other IV iron preparations reporting similar AEs in the same database; MRCI values were all very close to unity.
An alternative assessment of a drug’s potential safety risk can be achieved by estimating the magnitude of downstream direct medical costs based on AE and outcome costing data using the methods of Hoffman et al. . These methods provide an accessible reference point regarding real-world differences in post-marketing drug safety . The results obtained can be used to improve patient safety by identifying therapies that cause an undue burden on patients and healthcare providers. The current analyses showed that over the time period examined, total downstream medical costs associated with anaphylaxis and manifestations of hypersensitivity reactions were highest with ferumoxytol (US$1,240,963). Total costs were about 42–66% lower with the three other IV iron preparations. When average costs per AE were considered, which accounts for differences in AE reporting rates, iron dextran and ferumoxytol were associated with the highest costs per AE (US$8615 and US$8164, respectively). When average costs per treated patient associated with these AEs were considered, ferumoxytol again had the highest cost, followed by iron dextran, ferric carboxymaltose, and iron sucrose. Thus, using the system of Hoffman et al. , results suggest that ferumoxytol is associated with the greatest burden of anaphylaxis and manifestations of hypersensitivity reactions, and that ferric carboxymaltose and iron sucrose are associated with the lowest burden of these events. The low burden of events with ferric carboxymaltose is not unexpected because this preparation is characterized by a tight binding of elemental iron to the carbohydrate polymer shell  and has the lowest labile-free iron release of the IV iron preparations considered .
While real-world studies, both using spontaneous reporting databases and claims databases have shown results consistent with our analysis [4, 8, 15, 16], other analyses have not . In a meta-analysis of randomized controlled trials examining AEs across trials comparing ferumoxytol with other pooled IV iron preparations (iron sucrose and ferric carboxymaltose), oral iron, and placebo, an elevated rate of hypersensitivity reactions and hypotension was noted with ferumoxytol vs oral iron, but not vs the pooled IV iron group. However, only three studies contributed to this latter comparison and all three studies specifically excluded patients with a prior history of allergies to drugs that did not include those under investigation . In an earlier, but much larger, meta-analysis of 103 clinical trials, infusion reactions were increased with all IV iron preparations examined (iron sucrose, ferric gluconate, and ferric carboxymaltose) compared with placebo, oral iron, or no iron, and ferric carboxymaltose was associated with a reduced rate of cardiovascular AEs and AEs leading to discontinuation . However, this analysis did not include ferumoxytol and made no comparisons among the IV iron preparations examined.
A number of limitations of our analyses must be acknowledged. These analyses used FAERS data that have well-recognized limitations: the FDA does not require a causal relationship for an event to be reported; the “primary suspect” designation in FAERS is subjective; many AEs are not reported to FAERS; and reporting rates to FAERS are likely to be low in general, may not be similar across the included drugs, and may vary as a function of drug approval dates. Estimates of the impact of FAERS underreporting vary in the literature, but one study found that roughly 20–33% of the expected serious events were reported to FAERS for selected biologics and narrow therapeutics index drugs examined . In addition, although reports lodged in FAERS are evaluated by clinical reviewers to monitor the safety of products after they are approved by the FDA, there is no certainty that the reported event was caused by a named product or that all AEs relating to a product are reported . However, we conducted a sensitivity analysis of total suspect case reports, regardless of attributed causality as in the primary suspect case analysis, and found similar results. Moreover, spontaneous reporting databases suffer from the general limitation that only a numerator is available, which may itself be subject to reporting bias. Therefore, in the absence of denominator information on the overall population exposed to the drug, the specific risk associated with the drug is difficult to estimate. While disproportionality measures aim to overcome this limitation by quantifying to what extent reported AEs are occurring more frequently than expected with a specific drug vs all drugs, they can only indicate differential reporting rates and not actual differences in the occurrence of the event [24, 25, 49]. Thus, spontaneous reporting databases are useful for signal detection , whereas other data sources should be used for signal validation or comparative safety—ideally large, adequately powered, randomized controlled trials, or in the case of rare events, large prospectively designed observational studies conducted in representative populations.
Another general limitation of a disproportionality analysis is potential masking, where a disproportionate signal for a given drug–event pair can be masked because of the presence of a disproportionate signal for another drug with the same event of interest in the same database . However, our analysis of MRCI did not show any evidence of masking or dilution of the signal for any of the IV iron preparations examined. An additional general limitation is that information on demographic and pre-existing medical history is limited in spontaneous reporting databases, making adjustment for confounding factors difficult. Age and sex information is available in FAERS, and while this demographic information is limited, our stratified analyses by age group showed that reporting rates of anaphylaxis tended to be higher among patients aged ≥ 65 vs < 65 years for all IV iron preparations and reporting rates for hospitalization due to hypersensitivity or anaphylactic reactions with iron sucrose and ferumoxytol were higher among patients aged ≥ 65 vs < 65 years and among male vs female patients. Nevertheless, it is important to note that signals indicating a likely drug—AE association also occurred among patients aged < 65 years and among female patients, meaning that this analysis is relevant to patients of all ages and sex. However, these stratified analyses were limited by missing information on age and sex in 47.2% and 11.1% of cases, respectively. In addition, although outcome data are collected in FAERS, this information was missing in a substantial number of reports (e.g., for hospitalization and death in our analysis, coverage was only 42.8%); thus, this limitation should be acknowledged when assessing reporting rates for hospitalization and death in this analysis. Other relevant limitations specific to this analysis include the low number of overall reports with iron dextran during the timeframe of the study. It is well known that reporting frequency varies as a function of time since approval, which should be considered when assessing iron dextran results (see Fig. S1 of the ESM). It is also possible that FDA boxed warnings concerning some IV iron preparations may have led to increased reporting.
Findings for the cost estimates may not be comprehensive because they were obtained by mapping AHRQ HCUP cost survey data to MedDRA® terms found in FAERS; not all FAERS hypersensitivity MedDRA® terms had an associated cost mapping available. Thus, potential variations between the FAERS patient population and those used for HCUP surveys could influence estimates. In addition, the most recent available data from the AHRQ HCUP cost survey data were from the year 2016 [32, 33], and for hospital stays resulting in death, the year 2007 . Because of the limited available costing data, we operated on the assumption that the estimated hospitalization costs for the given AE remained the same year to year; however, we adjusted these calculations for inflation on a yearly basis using the medical care services component of the consumer price index available from the US Bureau of Labor Statistics. We specifically analyzed downstream medical costs only as we wished to focus on costs attributable to AEs not confounded by the acquisition costs of the various IV iron preparations. Finally, patients reporting AEs for each IV iron preparation were not matched on baseline characteristics, thus differences in prescribing patterns could account for some differences in AEs (e.g., one preparation might be used more frequently for patients with comorbidities or severe iron deficiency anemia).