Drug-induced interstitial lung disease: a real-world pharmacovigilance study of the FDA Adverse Event Reporting System from 2004 to 2021

Drug-induced lung injury is a specific adverse event in the pulmonary system caused by drug use, which may involve the airway, parenchyma, pleura, pulmonary vascular system, and neuromuscular system.^1,2 It has different clinical phenotypes and histopathologic patterns, including pneumonia (autoimmune, eosinophilic, and hypersensitive), pulmonary fibrosis, sarcoidosis, and pleural effusion.³ Among these, drug-induced inflammation and fibrosis, also known as drug-induced interstitial lung disease (DILD), are the most common and severe forms of drug-induced lung injury. It can damage gas exchange, resulting in dyspnea, in severe cases, respiratory failure, and death.⁴ Compared with other involved pulmonary clinical phenotypes, it has the strongest drug-induced association and is an increasingly common cause of morbidity and mortality. The incidence of DILD varies between 4.1 and 12.4 cases per million per year, accounting for 3–5% of prevalent interstitial lung disease (ILD).^5,6 Moreover, the clinical, pathological, and radiological features of DILD are less specific, making the diagnosis of DILD an enormous challenge.⁷ Consequently, our knowledge of the demographics and characteristics of DILD is limited, which hinders our ability to develop precise policies to reduce the risk of this severe drug-induced disease.

Big data analysis has promising applications in drug-induced disease prevention and treatment. The Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) is the largest public pharmacovigilance database for spontaneous reporting of adverse events. And it is also the primary source for post-market safety monitoring and evaluation of drugs and therapeutic biologics.^8,9 Recently, some studies have evaluated drug-specific lung injury through the FAERS database.^10,11 However, as far as we know, no one has used this database to identify the most frequently reported drugs related to DILD. Therefore, we sought to analyze reporting trends of DILD and determine the most frequently reported drugs, drug classes, and drugs with the highest strength of association with DILD by sex and age, respectively. Additionally, we aimed to assess the differences in DILD reports across countries.

DILD has a significant impact on the health and quality of life of patients. More than 350 drugs may cause DILD, but most drugs cannot be discovered until later in development or after post-marketing.⁵ Due to the unpredictability of DILD, the lack of specific biomarkers, histological manifestations, and diagnostic clinical features, the unclear pathogenesis, and the possibility of life-threatening or disabling conditions, the prediction and characterization of DILD in preclinical drug development and post-marketing have brought extreme challenges to clinicians, the pharmaceutical industry, and health regulators.²²

From 2004 to 2021, reports of DILD increased year by year. The reasons for this may be likely multifactorial. First, the drug safety awareness of the public and the pharmaceutical industry has gradually enhanced, resulting in greater attention to adverse events.²³ And second, several contributing facilitators might promote the reporting of adverse events, including institutions engaged in risk evaluation and mitigation strategies, more stringent regulatory authorities, and the convenience of the Internet.²⁴

During the study period, we observed significant differences related to age and sex in all DILD cases submitted to the FDA. In total, 57.17% of all reports of DILD were over 60 years, and males were more frequently reported than females. We have also observed the age- and sex-related risks in clinical outcomes, with higher death rates in older/male patients. It has been reported that smoking is an important risk factor for patients with DILD, and male smoking preference is more than female, which may be the reason for the higher frequency of male reporting.⁵ However, in some drug classes, female patients reported more than male patients, such as antirheumatic agents. The possible reason is that the drugs themself for diseases with female susceptibility or unique (such as rheumatoid arthritis and menstrual disorders).^25,26 The mechanism of DILD remains unclear, which may be mainly related to the allergic reaction and direct cytotoxicity caused by drugs and their metabolites.²² Of those, oxidative stress is considered a risk factor for the occurrence and progression of lung tissue damage.²⁷ With age, the antioxidant capacity of the human body, the activity of drug-metabolizing enzymes, the anti-inflammatory reaction, and the repair ability are all reduced, so more DILD may occur. In addition, the underlying diseases of the elderly are one of the significant reasons for DILD. For example, patients with underlying respiratory disease may die from mild DILD^28–30; for patients with underlying renal disease, prolonged excretion of certain drugs may increase the risk of DILD.^31,32 It is worth paying attention that the elderly, as a high-risk group, will increase the number of cases and social burden of DILD as the global population ages, and global public health management will face more challenges.

In our study, antineoplastic agents were the most frequently reported drug class that induced DILD. A study found that the incidence of respiratory failure caused by DILD was 6.6 per 100,000 patients per year, with more than half of the cases related to chemotherapy.³³ It also had been reported that antineoplastic agents were the primary cause of DILD, accounting for 23–51% of all reported cases.³⁴ As a major complication in cancer treatment, DILD increases the risk of drug withdrawal and death.^35,36 Moreover, many new antineoplastic agents have been widely used in medicine over the past few years. In our study, we found that monoclonal antibodies, as an emerging therapy against cancer, had a high volume in the frequency of DILD reported in multiple countries. However, this scenario may correspond to the increased use of monoclonal antibodies in treatment. Some monoclonal antibodies have been reported in association with true DILD, such as fam-trastuzumab deruxtecan-nxki, nivolumab, pembrolizumab, and others.^37–43 Of these, fam-trastuzumab deruxtecan-nxki was the top drug related to DILD reported to the FDA, with an incidence of 15.8% and a mortality of 2.4% in clinical trials.^44,45 While some reports may be confused by its off-label therapy.^39,46 On the other hand, we also noted that half of the top 20 drugs with high strength of association did not mention DILD in the labels. It draws attention that healthcare professionals should enhance monitor related to drugs of potential importance. In short, DILD brings a higher risk of drug withdrawal and death, as well as the emergence of many new antineoplastic agents with the lack of real-world safety data makes patients confused about drug selection. The results of our study can help physicians choose drugs with low DILD risk from similar effects to improve patient compliance and survival rate.

At the same time, we observed significant country-related differences in DILD reports in the FAERS. In Japan, the United States, France, and Canada, 60–70% of the top 10 drugs most related to DILD were antineoplastic agents, consistent with their high tumor incidence rate.⁴⁷ The proportion of cardiovascular agents in the United States was significantly higher compared to other countries, which was related to its higher cardiovascular burden.⁴⁸ Four and five antirheumatic agents were found in the frequently reported top 10 drugs in the United States and Canada, respectively. It may be related to the higher prevalence in North America compared to other regions.⁴⁹ These differences across countries can be attributed to differences in disease maps, demographics, genetic effects, drug use patterns, and reporting rates. Furthermore, our research showed that more than one-third of the reports came from Japan, seeming to indicate a higher prevalence of DILD in Japan. However, due to coding and spontaneous reporting, most of them are considered to be man-made rather than biological.⁵⁰ In contrast to Japan, many regions of the world have reported little data on DILD to FAERS, including Africa and much of Asia.

We respectively calculated the ROR, PRR, and BCPNN to detect the correlation strength between an individual drug and DILD. van Puijenbroek et al.²¹ found that these methods have different advantages and limitations. Different health regulators use various statistical indicators to measure the correlation between drugs and adverse events, such as ROR currently used by Lareb in the Netherlands, PRR by the UK Medicines and Healthcare Products Regulatory Agency, and BCPNN by the World Health Organization Drug Monitoring Centre in Uppsala. Each of these methods may lead to different sensitivities and specificities in signal risk detection, with no single measure of effect superior to others.⁵¹ Therefore, we utilized ROR, PRR, and BCPNN to detect the risk drug in the DILD database to reduce the generation of false-positive signals. Moreover, many previous studies and papers have used these methods to detect adverse event signals in spontaneous reporting systems.^52–54

There were limitations in this study. FAERS is an adverse drug reaction spontaneous reporting system with a lack of denominator data and a high false report rate, limited by uncertain causality, repeated reports, missing information, and spelling errors. Therefore, the data are biased against the true incidence of DILD.^55,56 We obtained the ROR, PRR, and IC values according to the reported frequency of the combination of drugs and adverse events. This value represents the adjustment between the DILD ratio reported by a drug and the ratio of all other adverse events reported for the target drug. They calculate the reporting risk rather than the risk of DILD occurrence.⁵² There are differences in drug use and race in various countries and regions. Nevertheless, the FAERS database lacks ethnicity data, an indispensable part of considering the environmental and genetic factors that lead to adverse events. Moreover, there was a ‘Weber effect’ in the FAERS database, where adverse event reports peaked in the initial post-marketing phase and following declined year after year.⁵⁷ Nonetheless, it still provides references to formulate precise and targeted policies to prevent DILD in the real world.

In a review of the FAERS database, we uncovered that the top drug classes associated with DILD in FAERS were antineoplastic, cardiovascular, and antirheumatic agents, in varying order in different sexes/age. Fam-trastuzumab deruxtecan-nxki, ramucirumab, and eribulin had the highest strength of association with DILD. We also found some unexpected drugs without DILD in the labels, such as amiodarone, doxorubicin, and ursodiol. Our findings provided real-world data for the development of novel management strategies of drug-related adverse events.

None.