Skip to main content
NCBI home page
Journal of Managed Care & Specialty Pharmacy logoLink to Journal of Managed Care & Specialty Pharmacy
. 2026 Aug;32(8):968–978. doi: 10.18553/jmcp.2026.32.8.968

Evaluating the feasibility of health care databases for assessing the safety of switching between originator biologics and biosimilars

Aaron B Mendelsohn 1,✉, Terese A DeFor 2, Djeneba Audrey Djibo 3, Scott M Myers 4, Pamala A Pawloski 1, Catherine M Lockhart 1
PMCID: PMC13403258  PMID: 42504812

Abstract

BACKGROUND:

Real-world data (RWD) have received considerable attention recently, namely, with regard to their value in providing further evidence about the benefit-risk profile for medicinal products beyond clinical trial data. RWD may be particularly useful for establishing interchangeability and supporting switching decisions between originator biologics and their biosimilars.

OBJECTIVE:

To evaluate the fitness of 3 US health care databases—a national, commercial health plan, a regional integrated delivery network (IDN), and a multipayer, national claims-based database (henceforth, multipayer database)—for capturing data from clinical trials assessing interchangeability between originator biologics and biosimilars across multiple therapeutic areas.

METHODS:

We identified 8 clinical trials examining switching between originator biologics and biosimilars from the published literature and ClinicalTrials.gov. All variables representing inclusion/exclusion criteria, interventions, and outcomes from these trials were recorded and grouped into 8 categories: assessment, behavior, demographic, diagnostic, laboratory, procedure, treatment, and vital signs. The 3 databases were then individually evaluated based on the availability of variables identified from the clinical trials and by calculating the percentage of observed data in each database for all relevant variables across the clinical trials, overall and within the above categories.

RESULTS:

The population size varied across the databases (4 million for the regional IDN through 170 million patient-lives in the multipayer database). All databases had complete (100%) capture for procedure, treatment, and vital signs data and performed well for capturing diagnostic information (78%-100%). Most demographic information (eg, age, sex) was captured; however, race and ethnicity was not available for all databases. Seventy-one percent of the behavior data (eg, whether the patient was sexually active) was captured by the commercial health plan and the regional IDN, but only 29% by the multipayer database. Assessment data (eg, survival, functional status) varied across the databases, with the regional IDN having 93% data capture vs 37% and 16% in the commercial health plan and multipayer database, respectively. Notable differences among the databases were also observed for laboratory data; the regional IDN had complete capture vs only 6% in the multipayer database.

CONCLUSIONS:

Health care databases provide information such as diagnoses, treatments, and some outcomes that may be useful for generating real-world evidence relevant to biosimilar regulatory assessment. However, details on treatment effectiveness may be limited. Specific databases should be evaluated according to their unique attributes to select the most appropriate source(s) of information for a given research need. Further studies are warranted to evaluate data accuracy and timeliness in health care databases.

Plain language summary

Real-world medical care data may be useful for evaluating the safety and benefit of medicines, including biosimilars or medicines that are highly similar to existing biologics. This study describes how health care databases capture information such as diseases and treatments that could support regulatory decisions about biosimilars. As different health care databases have unique limitations with respect to data availability, the study emphasizes the importance of evaluating individual databases to select the most appropriate data source for a given purpose.

Implications for managed care pharmacy

Evaluating the fitness for use of real-world data (RWD) sources is critical for using RWD to inform about the benefit and risk of medicinal products. This study demonstrates that health care databases provide information that may support biosimilar regulatory assessment and potentially facilitate broader adoption of biosimilars. However, as some data elements are limited in RWD sources, investigators should evaluate the unique attributes of specific databases to select the optimal data source(s) for given research priorities.

Real-world data (RWD), namely, data derived from routine clinical care, and real-world evidence (RWE), that is, clinical evidence on the benefits and risks associated with medicinal products based on RWD,1 have increasingly gained attention in the life cycle of medicinal products with applications in early development, regulatory approval, and market access.2–5 This shift is fundamental for biosimilars where demonstrating interchangeability between originator biologics and biosimilars to enable switching decisions remains a regulatory challenge in the United States.6–10

Biosimilars must be highly similar to licensed biologics with no clinically meaningful differences in safety, purity, and potency.6,8,9,11,12 Interchangeability requires evidence that switching between the biosimilar and the originator biologic must not reduce efficacy or magnify safety risks.9-11,13 According to US Food and Drug Administration (FDA) guidance, interchangeability studies should explore whether switching leads to differences in immunogenicity, pharmacokinetics, and pharmacodynamics. Although these outcomes would typically come from randomized clinical trials, the FDA revised their guidance on whether switching studies may be needed and recognizes that RWD, and the resulting RWE, may inform about the benefit-risk for biosimilars and provide safety data for certain switching scenarios.14 RWD may therefore provide a promising pathway for helping to establish interchangeability and ultimately facilitate broader adoption of biosimilars. However, as a prerequisite, RWD must satisfy the conditions of high data reliability (ie, data accuracy, completeness, and traceability) and data relevance (ie, data are fit for use and capture the critical elements, namely, exposure(s), covariates, and outcomes of interest).15 Furthermore, RWD should come from data sources that approximate the population of interest and can fortify the data from the clinical trial setting. Data accuracy and traceability is beyond the current project’s scope; thus, we will focus on RWD completeness and relevance.

Different RWD sources offer unique attributes, including data types, population size, patient characteristics, and geographic distribution, that can influence selection of the data source most suitable for specific research questions. The objective of this study was to evaluate the feasibility of using different health care databases for assessing the safety between originator biologics and their biosimilars. To do this, we examined the availability of data from these RWD sources relative to the inclusion/exclusion criteria, treatment information, and key outcomes included in clinical trials specifically designed to assess interchangeability and switching between originator biologics and their biosimilars across multiple therapeutic areas.

Methods

We included 3 distinct health care databases—a large national, commercial health plan with administrative claims and some clinical data, a regional integrated delivery network (IDN) with administrative claims and clinical data with a moderate population size, and a very large multipayer, national claims-only database (multipayer database) (Table 1). These databases were selected as they all have administrative claims but vary in size and access to other data (eg, electronic health records [EHR]). Our goal was to describe the nuances of available data from different claims-based data sources that may inform optimal database selection for observational research.

TABLE 1.

Characteristics of Health Care Databases Evaluated for Assessing the Safety of Switching Between Originator Biologics and Biosimilars

Data source
National, commercial health plan Regional integrated delivery network Multipayer, national claims-only database
Description Administrative claims (medical, pharmacy) and enrollment data from a large, national, commercial health plan. Possible to obtain medical record data for ∼70% of patients Administrative claims (medical, pharmacy) linked to EHR from a regional health system and insurance plan Administrative claims data from multiple payer types, including commercial, Medicaid, Medicare, employer, self-pay
Date availability From Jan 2008 From Jan 2000 From Jan 2010
Data lag Up to ∼11 weeks Claims: Up to ∼3 months EHR: 1 day Average: ∼20 months
Population size >44 million patient-lives >4 million patient-lives >170 million patient-lives
Average follow-up time 2 years 5 years 9 years
Geographic region All 50 states and territories Midwestern United States All 50 states
Open in a new tab

EHR = electronic health records.

We searched PubMed (MEDLINE) and ClinicalTrials.gov to identify switching trials that have been conducted, or are planned, and may be used to evaluate switching between originator biologics and biosimilars. For this targeted literature search, we used the following keywords: biosimilars, switching, interchangeability, clinical trials. Identified studies were not required to have been used as part of regulatory applications for interchangeability but instead were included to reflect the information likely to be relevant for FDA assessments, to facilitate our evaluation of the fitness of RWD for biosimilar regulatory decisions. We identified a total of 8 clinical trials focusing on switching between originator and biosimilar products across multiple therapeutic areas, including type 1 diabetes mellitus (insulin glargine),16 psoriasis (2 adalimumab studies for chronic plaque psoriasis,17,18 1 ustekinumab study for moderate to severe plaque psoriasis19), ophthalmology (ranibizumab for neovascular age-related macular degeneration20), rheumatoid arthritis (infliximab,21 rituximab22), and oncology (bevacizumab for metastatic colorectal cancer23). We extracted a comprehensive set of variables detailing the inclusion and exclusion criteria, treatment information, and measured outcomes from each trial. Experts from the 3 data sources provided information on the availability of each variable within their respective databases.

The variables identified were grouped into 8 distinct conceptual categories:

  • (1)

    Demographic: For example, age, sex, race and ethnicity, geographic region.

  • (2)

    Diagnostic: Any disease diagnosis/condition (eg, herpes zoster infection, immunodeficiency), identified through clinical evaluation, diagnostic tests, or imaging results, including therapeutic areas of focus and comorbid conditions.

  • (3)

    Treatment: Drug therapies (current or historical).

  • (4)

    Procedure: Medical or surgical interventions, including vaccinations.

  • (5)

    Laboratory: Clinical laboratory tests (eg, hemoglobin A1c, absolute neutrophil count [ANC], white blood cell count [WBC]).

  • 6)

    Vital Sign: Clinical measurements (eg, weight, body mass index [BMI]).

  • (7)

    Behavior: Participants’ actions, habits, lifestyle factors (eg, birth control, pregnancy/lactation).

  • (8)

    Assessment: Clinical health scales or functional status measures (eg, Eastern Cooperative Oncology Group [ECOG] performance status, best-corrected visual acuity [BCVA], quality of life [QoL] data).

Supplementary Exhibit 1 (229.4KB, pdf) (available in online article) contains further details on the specific variables included within the above categories.

The 3 databases were evaluated based on the availability of variables identified from the clinical trials by calculating the percentage of observed data (ie, data element present) in each database relative to the total number of variables from the clinical trials. These percentages were calculated for all variables combined, and within each of the 8 core categories described above. We explored results aggregated across trials and within each individual trial to check for consistency. It is worth noting that we only examined whether a given data element was present in the databases. We did not examine the extent to which the field was populated or the accuracy of the data—these are important considerations but are beyond the scope of the present study.

Results

Characteristics of the 3 databases included in this study are presented in Table 1. Intentionally, all databases had administrative medical and pharmacy claims data. The regional IDN had linkage to EHR, and the commercial health plan had access to enrollment data and some EHR data (∼70% of patients). The population size varied from more than 4 million (IDN), to more than 44 million (commercial health plan), and more than 170 million patient-lives (multipayer database) with the larger databases having geographic representation throughout the United States (although not technically nationally representative). The average follow-up time was shortest in the commercial health plan (∼2 years) and longest in the multipayer database (∼9 years).

The IDN had the highest overall data capture (98%) across all 8 variable categories and all included clinical trials compared with 75% for the commercial health plan and 55% for the multipayer database. The capture rates varied based on the variable category (Figure 1). All databases had complete capture (100%) for procedure and treatment data. Vital signs were also fully captured, noting, however, that the only 3 variables in this category across the clinical trials were weight/BMI and QT measurement. We considered weight/BMI to be captured by a given database if the relative BMI or weight could be determined, for example, obese or not, and did not require the precise weight or BMI to be recorded. Similarly, for QT measurement, we considered this variable to be captured if it could be determined that the patient had a long QT interval.

FIGURE 1.

Percentage of Variables Captured Across Categories and Overall (Total) by Health Care Database

FIGURE 1

Variables within each category are listed in Supplementary Exhibit 1 (229.4KB, pdf) . Percentages were calculated by dividing the number of captured variables within each category (or across categories/total) by the total number of variables within each category (or across categories/total).

IDN = integrated delivery network.

Open in a new tab

The databases also performed well with capturing diagnostic information, with the commercial health plan and the IDN capturing more than 90% of diagnostic data. The multipayer database captured 78% of diagnostic variables, missing data elements such as drug hypersensitivity and cobalamin deficiency that lack specific diagnosis codes or may not consistently be coded.

Most demographic information (eg, age, sex, geographic region) was captured for all databases; however, race and ethnicity was not available for the multipayer database. Although the commercial health plan had race and ethnicity data, these fields were missing for approximately 10% of patients. The commercial health plan and IDN each had 71% capture of behavior data compared with 29% in the multipayer database. None of the databases included information as to whether the patient was sexually active, and the multipayer database was also missing other behavior-related data such as whether the patient was pregnant or breastfeeding and whether the patient was enrolled in another investigational drug/device trial.

Laboratory data capture varied by data source. The IDN had 100% capture of laboratory variables vs 53% in the commercial health plan. Both of these databases included results for common laboratory measures such as WBC count, ANC, platelet count, albumin, serum creatinine, and total bilirubin, as well as measures relevant to specific diseases (eg, A1c) that were not available in the multipayer database. One caveat related to the commercial health plan is that laboratory measures were not available for all patients; their inclusion in the database depends on the individual health plan under this insurer that the patient is enrolled in. The multipayer database did not capture actual values for the laboratory measures. However, it did include either ranges or normal/abnormal determination (through diagnostic and/or procedure codes) for 6% of the laboratory measures including erythrocyte sedimentation rate (ESR), quantiferon test, and folic/folinic acid. To clarify, elevated ESR was captured by the multipayer database, but not the precise ESR level. Additionally, for the quantiferon test, a claim for test administration, coupled with a tuberculosis diagnosis (within relative chronological proximity to the test), could be used to indicate a positive test: hence the rationale for considering this variable to be captured by the multipayer database.

Assessment data also varied across the databases; the IDN captured 93% of assessment data vs 37% (commercial health plan) and 16% (multipayer database). All databases captured assessment variables related to the necessity of various treatments (eg, need for prednisone, nonsteroidal anti-inflammatory drugs, methotrexate), defined as a filled script for a given treatment or a claim for a recommended treatment. The commercial health plan and IDN also captured data related to inadequate response or intolerance to medication (eg, disease-modifying antirheumatic disease agents), defined as a switch or addition of a new medication, or other surrogates (eg, diagnostic codes, length of treatment), that would be suggestive of an initial therapy not being fully effective. The IDN had complete capture of clinical and functional status measures, including ECOG, BCVA, Disease Activity Score-28,24 and Health Assessment Questionnaire,25 as well as QoL data, namely, the Short-Form–3626 for selected patients (ie, where the assessment was collected as part of routine care at selected facilities fully owned by the IDN) that were not available in the other databases.

We found consistency in the percentage of variables captured (overall and within categories) across trials for the commercial health plan and IDN, with overall data capture ranging from 62% to 88% and 88% to 100%, respectively (Table 2). The regional IDN had more than 90% data capture for 7 (87.5%) trials. For the multipayer database, overall data capture ranged from 22% to 76%, with lower capture rates for trials involving greater numbers of variables in the assessment, behavior, and laboratory categories.

TABLE 2.

Percentage of Clinical Trial Variables Captured by Variable Category for the Selected Health Care Databases

Trial (reference number) Database Overall capture, % Variable category capture, %
Assessment Behavior Demographic Diagnostic Laboratory Procedure Treatment Vital signs
Type 1 diabetes trial
 1 (16) Commercial health plan 88 0 — 100 100 100 — 100 100
Regional IDN 88 0 — 100 100 100 — 100 100
Multipayer, database 65 0 — 75 100 0 — 100 100
Psoriasis trials
 2 (17) Commercial health plan 62 33 100 100 100 0 100 100 100
Regional IDN 100 100 100 100 100 100 100 100 100
Multipayer, database 47 33 50 60 71 0 100 100 100
 3 (18) Commercial health plan 86 100 — — 83 75 — 100 —
Regional IDN 100 100 — — 100 100 — 100 —
Multipayer, database 29 0 — — 50 0 — 100 —
 4 (19) Commercial health plan 81 60 75 — 100 75 100 100 100
Regional IDN 96 100 75 — 100 100 100 100 100
Multipayer, database 67 40 25 — 100 25 100 100 100
Ophthalmology trial
 5 (20) Commercial health plan 70 14 – 100 71 0 100 100 —
Regional IDN 100 100 — 100 100 100 100 100 —
Multipayer, database 76 0 — 100 90 0 100 100 —
Rheumatoid arthritis trials
 6 (21) Commercial health plan 77 0 — — 100 67 100 100 —
Regional IDN 100 100 — — 100 100 100 100 —
Multipayer, database 69 0 — — 0 67 100 100 —
 7 (22) Commercial health plan 72 44 — — 100 64 — 100 —
Regional IDN 100 100 — — 100 100 — 100 —
Multipayer, database 45 25 — — 82 0 — 100 —
Oncology trial
 8 (23) Commercial health plan 83 20 0 — 100 78 100 100 —
Regional IDN 95 80 0 — 100 100 100 100 —
Multipayer, database 48 0 0 — 68 0 80 100 —
Open in a new tab

Variables within each category are listed in Supplementary Exhibit 1 (229.4KB, pdf) . Percentages were calculated by dividing the number of captured variables within each category (or across categories for the overall capture) by the total number of variables within each category/overall.

IDN = integrated delivery network.

Discussion

We evaluated the fitness for use of 3 distinct RWD sources to inform biosimilar regulatory assessment by examining the availability of variables aligned with those collected in clinical trials assessing switching between biosimilars and their reference biologics. The primary objective was to assess whether RWE may contribute to evaluating the safety of biosimilars relative to their reference products and thereby inform regulatory decision-making, including considerations of interchangeability. RWD is not intended to replace clinical investigations—particularly studies of pharmacokinetics, pharmacodynamics, and immunogenicity that require controlled experimental settings—but rather to complement clinical trial evidence.

We found that all 3 databases captured relevant data for identifying patient cohorts and exposures through medical and pharmacy claims using common coding systems such as the International Classification of Diseases, Ninth and Tenth Revisions (ICD-9, ICD-10), Healthcare Common Procedure Coding System (HCPCS), National Drug Codes, and Uniform System of Classification codes. Although cohort identification is nontrivial in many disease categories, all sites demonstrated the capacity and capability to apply algorithms based on these coding systems to consistently identify and describe patients with breast cancer treated with prophylactic pegfilgrastim according to our use case.27 Although all sites had a foundation of administrative claims, they differed on some key characteristics that may inspire database selection for research purposes. First, available population size ranged from approximately 4 million patient-lives (IDN) to 170+ million (multipayer database). This has important implications when selecting a database depending on the expected number of treated patients and the anticipated effect size of interest. For biosimilar research, a large database may be desirable to detect rare events related to safety and effectiveness, but a smaller database with rich clinical data may be preferred if the population of interest is expected to be proportionally large, or to answer more detailed questions regarding nuances of the patient journey. Second, the IDN only included patients treated in the Midwest United States, whereas the other databases had national coverage. This has implications in available sample size, as well as broad generalizability to a population that may have wider variability in demographics and clinical characteristics. Third, duration of follow-up ranged from 2 years to 9 years, the latter based on a proprietary tokenization scheme that allows for tracking patients across health plans. This feature is particularly important in the setting of delayed outcomes or if the diagnostic journey for patients is long.

Most demographic information was readily available in all databases; however, variables such as race and ethnicity and other determinants of health were not consistently collected. The multipayer database did not include race and ethnicity data, although it was possible to apply some algorithms based on census data for the geographic area. The commercial health plan included race and ethnicity data for approximately 90% of their patient population owing to access to enrollment files and deliberate efforts to encourage recording these important demographic data (unpublished data, personal communication, DA Djibo). The IDN had the most robust data on patient characteristics as patient claims data were linked to EHR that are more likely to contain race and ethnicity information. For biosimilar research, selection of a data source should take into consideration the availability of data such as race, ethnicity, or other social factors that may impact health outcomes, for example, in diabetes and cancer.28,29

Diagnoses, treatments, and records of procedures were widely available across databases, as these variables are routinely available in medical claims. Some diagnosis-related variables are not consistently or accurately reflected in medical claims, however, namely, severity or disease stage (eg, metastatic disease) and disease-related symptoms. For certain conditions, reliance on diagnosis codes alone without a validated algorithm to identify specific disease features may limit patient characterization. Although there may be information in administrative claims that are suggestive of disease stage (eg, duration of disease, use of certain therapeutic interventions), there is no guarantee that these surrogate measures will yield the same patient population as that from a clinical trial setting. For biosimilars, this is particularly relevant in oncology and inflammatory conditions where measuring treatment effect is largely dependent on disease progression (or remission).30,31 If these measures are essential to the research question, it is worth considering a database where a more complete characterization of outcomes is likely to be available; however, this may come at the cost of sample size, for example, the use of a smaller IDN with integrated EHR and claims data. If the goal is also detecting a rare event, then it may be difficult to use RWD without access to a detailed data source, for example, a registry.32

Not surprisingly, the availability and completeness of laboratory results varied widely and was low except for the IDN, which consistently had recorded test results through linkage with EHR. The national commercial health plan had laboratory results for a portion of patients (an approximate percentage of patients for whom laboratory data would be available is uncertain as this depends on the specific health plan for which the patient is enrolled and the current, contractual obligations for the health plans for sharing supplemental clinical information). In the multipayer database, we could identify when a test was ordered and, for limited laboratory measures (eg, ESR), whether an abnormal level was found (provided there was a specific ICD or HCPCS code available), but precise results were not available. Generally, describing the patient journey relies on subsequent treatment and other measurable outcomes to generate assumptions about what the test results may have shown. This is imperfect and relies on assumptions that cannot be validated without further information such as the correlation or causality with treatment choices and related outcomes. Some laboratory information might not be reported in a structured data format (eg, included in pathology reports) and would therefore require some effort to extract and use the data. Also, although the commercial health plan had lower capture for laboratory data than the IDN, it did include results for common laboratory measures such as WBC count, ANC, and platelet count, that are relevant across diseases and patient populations. Thus, the commercial health plan and the IDN could be viable data sources when laboratory results are required. The availability of laboratory data certainly could impact the robustness of research findings, although this depends on the evidentiary needs to describe the clinical effect. For example, when evaluating biosimilars such as insulin glargine for treatment of insulin-dependent diabetes, having A1c may be an important component, but only if rates of hypo- or hyperglycemia are inadequate for measuring clinical effectiveness, as those outcomes are readily available in claims data.

The category we defined as vital signs included measures of weight or BMI, which were generally available in all databases, but with some major caveats. In the IDN data, measured values of weight and calculated BMI were available; however, in the other databases, the information came from administrative claims, which only indicate if there is an abnormality, but not the precise value (eg, ICD-10 code of E66 indicates overweight or obese but does not yield the exact BMI). When assessing biosimilars using RWD and RWE, granular measures such as specific BMI may not be essential when assessing overall treatment effect. For example, in the insulin glargine switching trial,16 BMI was only used to describe the treatment cohorts and specifically to demonstrate that randomization had successfully balanced the cohorts according to demographic and clinical variables. There was no attempt to evaluate how BMI may have influenced differences in treatment effect across groups. Thus, and arguably, simply having percentages of obese patients would have been sufficient for ensuring that comparator groups are balanced. This vital signs category needs more investigation as only a limited number of variables were called out in the studies included in this analysis; examining other vital signs that may influence health status, for example, blood pressure, and the impact on analyses that do not properly account for these measures would be useful.

Similarly, behavior data were limited as only a few behavior-related variables were included in these studies (eg, birth control methods, pregnancy/breastfeeding status, whether the patient was sexually active, whether the patient was enrolled in another investigational study). Other behavior data that might be relevant for evaluating treatment effect and controlling for potential confounding, such as exercise, smoking status, alcohol use, and health-seeking behaviors (eg, receiving vaccines, annual physicals) were not evaluated and would warrant further exploration. Whether the participant was sexually active was missing from all 3 databases. The multipayer database was also missing other behavior-related variables that were captured by the IDN and the commercial health plan, although only for some patients depending on their specific health plan, including whether the patient was pregnant or breastfeeding and whether the patient was enrolled in another investigational trial. In general, behavior data are especially challenging as these data are typically collected through direct observation or patient/proxy report so these variables may not be available in claims data. Although they may be found in EHR, the information may not be recorded in a standardized or consistent manner.33

A related measure we described as assessment data includes measures of disease severity or progression. Only the IDN database captured clinical and functional measures such as the ECOG and BCVA. Even though the measures may be captured in the EHR, there are still limitations in terms of the completeness of records available. Only approximately 25% of oncology patients had ECOG scores, depending on clinician discretion and whether the patient received care in a facility owned by the IDN (unpublished data, personal communication, T. DeFor). The percentages of ophthalmology patients with BCVA scores and general patients with QoL is uncertain (although presumed to be no more than 25%) and is also dependent on whether the treating health care provider felt that the information would be useful to obtain and record and where the patient received care.

One major challenge facing broader functional use of RWD for biosimilar regulatory evaluation is variability in data availability. This is, in part, an artifact of using data that are collected for health care billing or recordkeeping and not specifically intended for research. These real-world databases demonstrate that data provenance varies, inviting opportunities to improve data collection at the source (eg, during real-world clinical encounters). Encouraging clinicians to recognize the importance of collecting certain data elements that may not be routinely considered, (eg, assessments, behavior data) could also help enrich available data in both claims and EHR. Fortunately, such information would be relevant for ongoing clinical care as well as for research purposes that rely on these data. Additionally, further standardizing how clinical care is recorded by improving documentation practices in medical record completion (trigger questions, education of clinicians) should also serve to enhance the value of RWD.

Researchers should consider not just overall rates of data capture but the availability of data elements in specific categories that may vary across RWD sources. Although the IDN had detailed clinical data, it is limited by its relatively small sample size and regional geographic representation. The commercial health plan and the multipayer database provide large populations that are geographically diverse but may not readily offer data that are not captured in claims.

LIMITATIONS

This research had limitations. We examined the presence of data elements but did not evaluate other data quality dimensions including data accuracy and timeliness. Additionally, not all data elements are collected for all patients (eg, clinical measures). Researchers should therefore consider not just whether a given data element is present but how well populated the variable is and whether there exist systematic differences between patients who have the field populated and those who do not. Even in cases where a data element might be available and well populated, there may be challenges with applying the data for trial emulation or regulatory assessment, for example, inconsistency in data collection including different time points or frequency of sample collection for laboratory or clinical assessments. There also may be a lack of standardization in how measures might be obtained or, in the case of laboratory tests, inconsistency in the units of measure.34 Additionally, we used only 1 expert for each database for assessing the presence or absence of the data elements. Using multiple evaluators could have added to the rigor of this study but would certainly add to the complexity of the investigation. Although we did not conduct adjudication of the information reported from the experts, we evaluated differences across the databases as a means to confirm findings. Another consideration for these findings is the varying amount of time and effort required to obtain certain information, for example, obtaining medical record data through the commercial health plan can be cumbersome and expensive, take several months, and not yield records for all patients.35 There were some data elements from the clinical trials that are challenging to clearly define in a real-world setting, for example, “inadequate response to therapy” and “necessity of treatment.” Investigators should aim to align the information from real-world sources as closely to the underlying clinical trials as possible. Data lag is another consideration as delays in the availability of the most recent patient information are common in observational data. This was evident in the multipayer database, with an average data lag of 20 months. Finally, this analysis only included 8 trials and was limited with regard to the diversity of disease areas covered (eg, 3 of the trials were for psoriasis). This may impact the generalizability to other therapeutic areas and medicinal products. There also were limited variables for certain categories; for example, vital signs in the trials did not include common assessments such as blood pressure, and behavior data measured did not include exercise or health-seeking behaviors. Future research should expand the scope to evaluate additional trials and therapeutic areas (and consequently, a larger, more diverse sample of variables), providing a broader understanding of health care databases’ capabilities and fitness evidence generation.

We focused on 3 specific health care databases to assess how their similarities and differences based on their purpose and source may impact RWD biosimilar research. Our findings may not be entirely applicable to the myriad health care databases available to researchers. However, we would expect consistent findings between similar databases. For instance, data sources with integrated claims and EHR data would likely yield similar results to the IDN by capturing laboratory and disease-related measures (eg, ECOG) that would be available in medical records. Additionally, other national, commercial insurers would likely provide large sample sizes and inclusion of patients from broad geographic regions similar to this study’s commercial health plan. The multipayer database was uniquely able to track patients transitioning between health plans or insurance types offering an exceptionally long follow-up period; other databases comprising data from multiple payers and employing similar tokenization approaches may be able to do the same.

Conclusions

We found that currently available US health care databases provide information that may be useful for generating RWE relevant for biosimilar regulatory assessment; however, additional analyses should be conducted regarding data accuracy, completeness, and timeliness for additional therapeutic areas. We also recommend further studies to improve the consistency and transparency of data provenance, improve data collection in clinical settings, and encourage pursuing broad linkage among disparate data sources to enrich relevant RWE generation.

Disclosures

The authors report no disclosures. This project is fully supported by a cooperative agreement (Award #1U01FD007757) with the Food and Drug Administration (FDA) of the US Department of Health and Human Services (HHS). The contents are those of the authors and do not necessarily represent the official views of, nor an endorsement by, FDA/HHS or the US Government.

References

  • 1.Food and Drug Administration . Real-World Evidence. Accessed April 22, 2026. https://www.fda.gov/science-research/science-and-research-special-topics/real-world-evidence
  • 2.Timbie JW, Kim AY, Concannon TW. Use of real-world evidence for regulatory approval and coverage of medical devices: a landscape assessment. Value Health. 2021;24(12):1792-8. doi: 10.1016/j.jval.2021.07.003 [DOI] [PubMed] [Google Scholar]
  • 3.Arondekar B, Duh MS, Bhak RH, et al. Real-world evidence in support of oncology product registration: a systematic review of new drug application and biologics license application approvals from 2015-2020. Clin Cancer Res. 2022;28(1):27-35. doi: 10.1158/1078-0432.CCR-21-2639 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Klonoff DC. The new FDA real-world evidence program to support development of drugs and biologics. J Diabetes Sci Technol. 2020;14(2):345-9. doi: 10.1177/1932296819832661 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Duffield S, Jónsson P. The real-world impact of National Institute for Health and Care Excellence’s real-world evidence framework. J Comp Eff Res. 2023;12(11):e230135. doi: 10.57264/cer-2023-0135 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Biologics Price Competition and Innovation Act of 2009 (BPCIA) . Pub. L. No. 111-148, § 7001, 124 Stat. 119, 804 (2010).
  • 7.O’Callaghan J, Barry SP, Bermingham M, Morris JM, Griffin BT. Regulation of biosimilar medicines and current perspectives on interchangeability and policy. Eur J Clin Pharmacol. 2019;75(1):1-11. doi: 10.1007/s00228-018-2542-1 [DOI] [PubMed] [Google Scholar]
  • 8.Ebbers HC, Schellekens H. Are we ready to close the discussion on the interchangeability of biosimilars? Drug Discov Today. 2019;24(10):1963-7. doi: 10.1016/j.drudis.2019.06.016 [DOI] [PubMed] [Google Scholar]
  • 9.Alvarez DF, Wolbink G, Cronenberger C, Orazem J, Kay J. Interchangeability of biosimilars: what level of clinical evidence is needed to support the interchangeability designation in the United States? BioDrugs. 2020;34(6):723-32. doi: 10.1007/s40259-020-00446-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Rathore AS, Stevenson JG, Chhabra H, Maharana C. The global landscape on interchangeability of biosimilars. Expert Opin Biol Ther. 2022;22(2):133-48. doi: 10.1080/14712598.2021.1889511 [DOI] [PubMed] [Google Scholar]
  • 11.Kurki P, Barry S, Bourges I, Tsantili P, Wolff-Holz E. Safety, Immunogenicity and interchangeability of biosimilar monoclonal antibodies and fusion proteins: a regulatory perspective. Drugs. 2021;81(16):1881-96. doi: 10.1007/s40265-021-01601-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Hung A, Vu Q, Mostovoy L. A systematic review of U.S. biosimilar approvals: what evidence does the FDA require and how are manufacturers responding? J Manag Care Spec Pharm. 2017;23(12):1234-44. doi: 10.18553/jmcp.2017.23.12.1234 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Dowlat HA, Kuhlmann MK, Khatami H, Ampudia-Blasco FJ. Interchangeability among reference insulin analogues and their biosimilars: regulatory framework, study design and clinical implications. Diabetes Obes Metab. 2016;18(8):737-46. doi: 10.1111/dom.12676 [DOI] [PubMed] [Google Scholar]
  • 14.U.S. Food and Drug Administration . Guidance for Industry: Considerations in Demonstrating Interchangeability With a Reference Product: Update. Accessed May 18, 2025. https://www.fda.gov/media/179456/download
  • 15.U.S. Food and Drug Administration . Framework for FDA’s Real-World Evidence Program. 2018. Accessed July 13, 2025. https://www.fda.gov/media/120060/download
  • 16.Blevins TC, Barve A, Raiter Y, et al. Efficacy and safety of MYL-1501D versus insulin glargine in people with type 1 diabetes mellitus: results of the INSTRIDE 3 phase 3 switch study. Diabetes Obes Metab. 2020;22(3):365-72. doi: 10.1111/dom.13904 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Menter A, Cohen S, Kay J, et al. Switching between adalimumab reference product and BI 695501 in patients with chronic plaque psoriasis (VOLTAIRE-X): a randomized controlled trial. Am J Clin Dermatol. 2022;23(5):719-28. doi: 10.1007/s40257-022-00708-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Mylan Pharmaceuticals IA. Multicenter, Randomized, Blinded, Parallel Group, Interchangeability Study in Moderate to Severe Chronic Plaque Psoriasis Evaluating Pharmacokinetics, Efficacy, Safety, and Immunogenicity Between Subjects Receiving Humira® Pre Filled Syringe (40 mg) Continuously and Subjects Undergoing Repeated Switches Between Humira® Pre Filled Syringe (40 mg) and Hulio Pre-filled Syringe (40 mg). Accessed April 17, 2025. https://clinicaltrials.gov/study/NCT05637515
  • 19.Amgen . A Study to Investigate Interchangeability of ABP 654 for the Treatment of Participants With Moderate to Severe Plaque Psoriasis. Accessed April 17, 2025. https://clinicaltrials.gov/study/NCT04761627
  • 20.Holz FG, Oleksy P, Ricci F, Kaiser PK, Kiefer J, Schmitz-Valckenberg S; COLUMBUS-AMD Study Group . Efficacy and safety of biosimilar FYB201 compared with ranibizumab in neovascular age-related macular degeneration. Ophthalmology. 2022;129(1):54-63. doi: 10.1016/j.ophtha.2021.04.031 [DOI] [PubMed] [Google Scholar]
  • 21.Nichi-Iko Pharmaceutical Co., Ltd . A Phase 3 Study of NI-071 in Participants With Rheumatoid Arthritis (RADIANCE). Accessed April 17, 2025. https://clinicaltrials.gov/study/NCT02990806
  • 22.Biocad . Study of Safety and Efficacy of BCD-020 Comparing to MabThera in Patients With Rheumatoid Arthritis (BIORA). ClinicalTrials.gov. Accessed April 17, 2025. https://clinicaltrials.gov/study/NCT01759030
  • 23.Zhejiang Teruisi Pharmaceutical Inc . A Study Assessing the Interchangeability Between TRS003 and Bevacizumab® For Colorectal Cancer (CRC). ClinicalTrials.gov. Accessed April 17, 2025. https://clinicaltrials.gov/study/NCT05378867
  • 24.van der Heijde DM, van ’t Hof MA, van Riel PL, et al. Judging disease activity in clinical practice in rheumatoid arthritis: first step in the development of a disease activity score. Ann Rheum Dis. 1990;49(11):916-20. doi: 10.1136/ard.49.11.916 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Bruce B, Fries JF. The Stanford Health Assessment Questionnaire: dimensions and practical applications. Health Qual Life Outcomes. 2003;1(1):20. doi: 10.1186/1477-7525-1-20 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Rand . 36-Item Short Form Survey (SF-36). Accessed April 17, 2025. https://www.rand.org/health-care/surveys_tools/mos/36-item-short-form.html
  • 27.Lockhart CM, McDermott CL, Mendelsohn AB, et al. Identification of cancer chemotherapy regimens and patient cohorts in administrative claims: challenges, opportunities, and a proposed algorithm. J Med Econ. 2023;26(1):403-10. doi: 10.1080/13696998.2023.2187196 [DOI] [PubMed] [Google Scholar]
  • 28.Macias-Konstantopoulos WL, Collins KA, Diaz R, et al. Race, healthcare, and health disparities: a critical review and recommendations for advancing health equity. West J Emerg Med. 2023;24(5):906-18. doi: 10.5811/WESTJEM.58408 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Hirko KA, Rocque G, Reasor E, et al. The impact of race and ethnicity in breast cancer-disparities and implications for precision oncology. BMC Med. 2022;20(1):72. doi: 10.1186/s12916-022-02260-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Huang B, Tian L, McCaw ZR, et al. Analysis of response data for assessing treatment effects in compartive clinical studies. Ann Intern Med. 2020;173(5):368-74. doi: 10.7326/M20-0104 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Blumenthal GM, Pazdur R. Response rate as an approval end point in oncology: back to the future. JAMA Oncol. 2016;2(6):780-1. doi: 10.1001/jamaoncol.2015.6352 [DOI] [PubMed] [Google Scholar]
  • 32.Berlin JA, Glasser SC, Ellenberg SS. Adverse event detection in drug development: recommendations and obligations beyond phase 3. Am J Public Health. 2008;98(8):1366-71. doi: 10.2105/AJPH.2007.124537 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Lasser EC, Kim JM, Hatef E, Kharrazi H, Marsteller JA, DeCamp LR. Social and behavioral variables in the electronic health record: a path forward to increase data quality and utility. Acad Med. 2021;96(7):1050-6. doi: 10.1097/ACM.0000000000004071 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Holmes JH, Beinlich J, Boland MR, et al. Why is the electronic health record so challenging for research and clinical care? Methods Inf Med. 2021;19;60(1-02):32-48. doi: 10.1055/s-0041-1731784 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Djibo DA, Margulis AV, McMahill-Walraven CN, et al. Validation of an ICD-10 case-finding algorithm for endometrial cancer in US insurance claims. Pharmacoepidemiol Drug Saf. 2024;33(1):e5690. doi: 10.1002/pds.5690 [DOI] [PubMed] [Google Scholar]

Articles from Journal of Managed Care & Specialty Pharmacy are provided here courtesy of Academy of Managed Care Pharmacy

RESOURCES