Big Data Research on Severe Asthma

1. Electronic health record data

Recent advances in artificial intelligence (AI) technology and cloud computing have led to the emergence of commercial big data companies, and several software programs have been developed for extraction of various types of information from EHRs [12]. These technological breakthroughs have enabled the conduct of research on severe asthma utilizing large-scale EHR data. One weakness is that the quality of data can vary across the healthcare providers. In addition, the cost of extracting EHRs data can be substantial. The specific clinical setting from which EHR data originate may affect the representativeness of the data, which could lead to selection bias. However, due to the abundance of clinical information, including pulmonary function tests, large number of patients, and long-term duration of follow-up, this dataset can be considered useful for providing real-world evidence.

Optum, a healthcare services and technology company based in the United States, which offers ‘Optum EHR data [13],’ is one of the largest healthcare organizations worldwide, with a database of nearly 4 million EHR data, of which 3.2 million are linked to claims data. The dataset includes a wide range of covariates, including pulmonary function tests, laboratory findings, and healthcare costs, thus, it is regarded as a valuable source of medical information on severe asthma. In addition, the Optimum Patient Care (OPC) dataset, a not-for-profit social enterprise based in the United Kingdom (UK) that manages the OPC research database, is also available [14]. This dataset contains de-identified EHR data from general practices across the UK, comprising 22 million patients from over 1,000 general practice sites. The dataset has a well-established collaborative network, which offers a unique advantage. Collection of data by the OPC has expanded beyond the boundaries of UK, even including Australia. The International Severe Asthma Registry (ISAR) is also based on the OPC system for collection of data [15].

2. Claims data

Claims data is automatically generated in the healthcare process, and well-organized nationwide claims datasets have already established in some countries [16]. Despite limitations to accessibility of data, its strengths include size, speed, cost-efficiency, and expandability to other datasets. However, the lack of clinical details may reduce its usefulness, necessitating the use of operational definitions for identifying specific patient groups or events for study. For example, differentiating between difficult-to-treat asthma and severe asthma can be challenging based solely on International Classification of Disease 10th revision codes. Integrating prescription data and healthcare utilization patterns may be helpful in the effort to address this issue, making the data more suitable for use in conducting of robust research.

The USA Centers for Medicare and Medicaid Services (CMS) [17], a federal agency that offers a government-funded healthcare program covering almost 95% of United States citizens aged 65 or older, is a well-known example of claims data. Data collected from this program include healthcare utilization, costs, disease outcomes, and beneficiary demographics. Due to the long history of this system, longitudinal studies have been conducted over extended periods. However, a significant limitation of CMS data is its primary focus on the elderly population, which may not fully represent the broad spectrum of the disease. IBM (International Business Machines Corporation) MarketScan Research Databases, formerly known as Truven Health MarketScan Research Databases, which also provide claims data in the United States [18], includes reimbursed healthcare claims data from more than 250 private health insurance plans for employees, retirees, and their dependents. They are also linked to other useful data, including EHR and mortality. This dataset covers more than 30 million people, excluding those enrolled in medicaid health insurance plans. However, accessing this dataset can be costly, and it raises issues regarding generalizability due to limited coverage from private insurance plans, which may result in selection bias.

The Korean National Health Insurance Service (NHIS) and Health Insurance Review and Assessment (HIRA) database, in South Korea, is a comprehensive claims dataset [19]. This database, which is administered by the Korean government, covers almost 97% of the Korean population through a compulsory universal health insurance program. The greatest strength of these datasets is the universal coverage of Korean citizens, ensuring a high level of representativeness. It also includes regular health check-up data, providing valuable information, including anthropometric measures, changes in personal habits (smoking, alcohol drinking, and physical activity), and laboratory data [20,21]. In addition, it can also be linked to other datasets, including the Korea National Health and Nutrition Examination Survey [22]. However, its historical span is not as long as that of CMS data. Like other claims datasets, the lack of clinical details, including pulmonary function tests, is also a weakness of the NHIS and HIRA databases.

3. Cohort data

Although cohort data are not typically considered big data, due to the relatively lower threshold for what is considered ‘big’ in such diseases, they may be considered in the case of certain rare diseases. Collaboration among multinational cohorts can generate a large dataset comparable to that of conventional big data. Establishment and maintenance of a nationwide severe asthma cohort has been achieved in recent years. Efforts to combine severe asthma cohorts from multiple countries for creation of an international registry, the ISAR, are underway. The goal of the ISAR, which plays an important role in conduct of research on severe asthma [15]. One significant advantage of this international cohort is accuracy in diagnosing severe asthma and the abundance of severe asthma-related data. However, the number of patients may be smaller when compared with EHR data, and primary care patients are generally not included. Nevertheless, during the first three years, more than 5,000 patients with severe asthma were recruited by the ISAR. Because this cohort is ongoing, it is expected that its volume will eventually be comparable to that of other big data. Of particular importance, the ISAR also provides detailed demographics and medication data, including the use of biologics, an important medication in management of severe asthma. Therefore, the results of novel treatment, such as newer biologics, can be rapidly updated in dataset. The remaining challenges is recruitment of patients from diverse ethnicities and regions to ensure acquisition of unbiased data for the specific race or country.

1. Example 1

EHR data can be used for acquisition of real-world data on treatments for severe asthma. Using the Optum dataset, Jeffery et al. [23] attempted to determine the risk of exacerbation in patients with severe asthma following discontinuation of biologics. In the results of the study, which included approximately 2,500 patients with severe asthma who had used biological agents for more than 6 months, a similar risk of asthma exacerbation was observed between stoppers and continuers.

By contrast, the results of a randomized controlled trial (RCT) examining the use of mepolizumab conflicted with findings reported by Jeffery et al. [23], suggesting a potential association between discontinuing mepolizumab and an increased risk of exacerbation [24]. This discrepancy could be due to differences in the study populations, methodologies, or definitions. Despite the possibility of biased outcomes, the use of well-organized study methods using EHR data could potentially lead to replacement of several RCTs [25].

EHR dataset can also be helpful in the conduct of a novel type of research on severe asthma. For example, Ryan et al. [26], examined the treatment patterns and outcomes for patients with the potential for having severe asthma based on their referral status. The study, which combined data from Optimum Patient Care Research Database (OPCRD) and ISAR, enrolled more than 200,000 patients who underwent treatment for asthma in primary care and referral hospitals. Based on the findings of the study, many patients with the potential for having severe asthma in primary care who were not acknowledged could benefit from referral to an asthma specialist. Conduct of similar study, including a large number of patients representing different levels of care, using traditional datasets is not possible.

2. Example 2

Comprehensive information on drug use nationwide from claims data may be useful in the conduct of pharmacoepidemiologic research on severe asthma. Hong et al. [27] used claims data in evaluating the cost-effectiveness of adding tiotropium to the inhaled corticosteroid/long-acting beta-2-agonist in elderly patients with severe asthma. According to their findings, the incremental cost-effectiveness ratio was 60,074 US dollars/quality-adjusted life years, suggesting the cost-effectiveness of add-on tiotropium for treatment of severe asthma. Such findings can provide knowledge that can be used in decision making in treatment of severe asthma, including approval for new drugs and national insurance coverage for costly treatments.

Due to large size and extended follow-up period, claims data can be used for evaluating long-term outcomes, and it can be regarded as an effective and affordable approach to conduct of mortality studies. Lee et al. [28] conducted a study using Korea NHIS data for tracking the long-term outcomes for patients with oral corticosteroid (OCS)-dependent asthma. The result of the study, which included almost 10,000 patients, indicated that death was more common among these patients compared to those with OCS-independent asthma during follow-up periods of more than 10 years. These types of studies are considered more reliable due to their larger sample size and cost-effectiveness compared to the cohort study or RCTs.

3. Example 3

Data from a global cohort can be utilized to study the characteristics of severe asthma worldwide. Wang et al. [29] revealed on the diverse nature of severe asthma across different countries using the ISAR data. The strengths of this study include the diverse registry data, and representative of real-world patients compared to data collected in RCTs. The global cohort also can be used in examination of racial and ethnic differences in severe asthma. Chen et al. [30] reported significant disparities in the use of biologics among different racial groups. The findings of these studies demonstrated that global data on severe asthma can be helpful in the effort to understand how regional and cultural characteristics are compared internationally, which can be regarded as valuable information in the practice of precision medicine [31,32].

1. Current engagement in big data research

The usefulness of data based on the purpose should be determined prior to initiation of research on severe asthma using big data [33], which can be critical in development of a meaningful and effective research plan. Use of prospective data can provide a valuable opportunity for evaluating treatment outcomes, enabling robust examination of causal relationships [34]. However, when the objective is to evaluate epidemiological outcomes, use of large-scale cross-sectional data would be the preferred choice [35]. In addition to data compatibility, close collaboration between clinicians and data scientists is important for optimizing the outcomes. The success of the collaboration will depend on effective communication and mutual understanding between experts. Clinicians provide in-depth knowledge of the domain, insights regarding patients, and medical expertise, while data scientists contribute their analytical knowledge and data manipulation skills. Collaborative synergies of these two different domains could enable a comprehensive approach to treatment of severe asthma, which will lead to determination of the phenotype and endotype of severe asthma, identification of novel biomarkers, and prediction of the medication effects (Figure 1).

Figure 1.

Precision medicine for management of severe asthma using big data.

2. Artificial intelligence

With the continuing development of AI technology, both the generation and utilization of big data will include appropriate use of AI (Figure 2). Optical character recognition (OCR) can be utilized effectively when using non-digital data. OCR is a technology used to convert different types of documents, including hand-written documents, facilitating the transition of traditional medical records into EHRs [36]. Use of natural language processing, a branch of AI, can enables comprehension of human language based on algorithms by the computer for identification of natural language rules. Use of these new technologies can enable interpretation and organization of unstructured data, including free-text narratives in medical charts, thereby optimizing the use of EHRs in conduct of medical research. This effort may lead to attainment of critical clinical insights from large and non-standardized EHR datasets for improvement of research and clinical practice in treatment of severe asthma [37].

Figure 2.

Generation and utilization of big data using artificial intelligence (AI). OCR: optical character recognition; NLP: natural language processing; EHR: electronic health record.

Phenotyping, diagnosis, and management of severe asthma pose numerous challenges, which might be supported with use of AI technologies. Clustering of diseases, extraction of essential features, and prediction of outcomes can be performed without the direction of the researcher using unsupervised AI algorithms [38]. Combining new AI techniques and big data could lead to development of an innovative approach to treatment of severe asthma research [39]. However, because a significant portion of the available data, which contains an abundance of clinical information, is cross-sectional or retrospective, prospective longitudinal data essential for developing robust prediction models of treatment responses is limited. In addition to the longitudinal dataset, concerted efforts toward integration and consolidation of sparse data into a unified framework will also be required. Utilization of big data, along with advancements in AI techniques, will be helpful to clinicians in making prudent decisions in management of severe asthma.