What real-world data and evidence are, and what they add
Real-world data are collected outside clinical trials, in the routine course of care, and real-world evidence is what analysing them produces. Big data are one form of these data. Each source of real-world data records some things well and others poorly, and the sources are increasingly joined up. A real-world study differs from a randomised trial in how it is run and in what it can show. It answers its own kinds of question, and the people who commission and read it expect the writer to convey what the data mean and what they do not.
Real-world data and real-world evidence
Real-world data are the raw material and real-world evidence is what analysing them produces. Real-world data are data collected in healthcare settings outside the scope of clinical trials, from routine clinical care, patient registries, electronic health records, insurance claims, mobile health apps, wearable devices, patient networks, social media and other sources. They are valuable because they reflect how treatments, interventions and health systems work in real-world settings rather than in a trial's idealised environment, and they offer a richer, more diverse data pool than controlled clinical environments. Real-world evidence is the actionable insight derived from analysing those data systematically.
The two terms are generally understood to mean data and evidence from sources other than traditional clinical trials, but it has not always been clear exactly what the categories include, and regulators were pressed over a number of years to delineate ways to incorporate such data in their evaluations. In the framework for its Real-World Evidence Program, released in 2018, the US Food and Drug Administration defines real-world data as data relating to patient health status and/or the delivery of health care routinely collected from a variety of sources. The Food, Drug and Cosmetic Act gives the same definition and lists among those sources electronic health records, medical claims and billing data, product and disease registries, patient-generated data, data from in-home use settings and data gathered from mobile devices. The framework defines real-world evidence as the clinical evidence about the usage and potential benefits or risks of a medical product derived from analysing real-world data. Such evidence can be generated by different study designs or analyses, including but not limited to randomised trials such as large simple and pragmatic trials, and prospective or retrospective observational studies. It serves, for example, to characterise a population and a disease and to evaluate the utilisation, benefits and risks of medicinal products.
Real-world data and real-world evidence differ in four respects.
| Real-world data | Real-world evidence | |
|---|---|---|
| Stage of processing | Raw, unprocessed data | Analysed, interpreted evidence |
| Data against insight | Diverse, unstructured, often complex; meaning is hard to extract without analysis | Distils the complexity into insight professionals, policymakers and researchers can use |
| Purpose | Collecting comprehensive data on practice, patient behaviour and outcomes | Applying the data to improve outcomes, guide policy and inform pharmaceutical development |
| Complexity | Heterogeneous, from sources with varying formats and standards | Directly applicable to real healthcare situations |
Big data
Big data is data whose size or type is beyond the ability of traditional relational databases to capture, manage and process with low latency, with one or more of the "three Vs": high volume, high velocity and wide variety. Artificial intelligence, mobile, social media and the Internet of Things drive its complexity, with data arriving from sensors, devices, video and audio, networks, log files, transactional applications, the web and social media, much of it in real time and at very large scale. Big health data exist in different structures and are stored in different repositories, and despite the common use of computers and artificial intelligence in healthcare, the three Vs still present a major challenge. Big data analytics is the use of advanced analytic techniques on very large, diverse datasets of structured, semi-structured and unstructured data from different sources, sized from terabytes to zettabytes. The European Medicines Agency defines big data as "extremely large datasets which may be complex, multi-dimensional, unstructured and heterogeneous, which are accumulating rapidly and which may be analysed computationally to reveal patterns, trends, and associations".
Big data has also been defined as datasets whose size is beyond the ability of typical database software to capture, store, manage and analyse. Some accounts give four Vs rather than three: volume, the scale of the data; variety, the types available beyond the structured data of a typical database; velocity, the speed at which data are acquired and used, often in real time; and veracity, their uncertainty. Data analytics is what is done with the data, and in population health there are four kinds, each with a low-blood-sugar example: descriptive (how many patients had low blood sugar?), diagnostic (what correlated with it?), predictive (how many will have it next year?) and prescriptive (what change will reduce it?). SAS, R, Python and MATLAB are among the common programs used in data analysis.
The Internet of Things is the growing number of network-connected objects that send and receive data, including the devices patients, physicians and researchers use to diagnose illness, monitor health parameters and store information. A report for the European Commission estimated that medical images alone would soon make up 30% of the world's stored data.
Sources of real-world data
Real-world data come from several diverse and interconnected sources, which together offer a comprehensive view of patient care that can be used to evaluate healthcare trends and outcomes. Electronic health records are providers' digital records of demographics, diagnoses, treatment histories, laboratory results and prescriptions. They give a continuous, longitudinal view of care, which makes them among the most valuable sources, and they are used to track disease progression, treatment efficacy and outcomes across settings. Health insurance claims record service utilisation, such as hospitalisations, visits, prescriptions and diagnostic procedures, with reimbursement details, and serve studies of care patterns, cost and the impact of treatment strategies. Statutory health insurance claims in the German Pharmacoepidemiological Research Database (GePaRD), which covers about 20% of Germany's general population, have been used to estimate the trend of sickle cell disorders in Germany from 2006 to 2016. Patient registries collect detailed information about people diagnosed with a specific disease or condition and are particularly valuable for understanding rare diseases or conditions that may not be adequately represented in trials. Wearables and mobile health apps give real-time measurements of activity, heart rate, sleep and blood pressure, and pharmacy records show what is prescribed, dispensed and used, informing adherence and treatment patterns.
The sources also differ in structure. Medical and pharmacy claims offer structured data attached to reimbursement requests for a diagnosis; electronic health records are unstructured but give far more detail about each patient's health and care; patient-reported data from surveys or interviews are highly variable yet show a disease's or treatment's true impact from the patient's side. To include comprehensive information in a study, data on individual patients are often combined from different sources, which may be formatted differently and need harmonisation before use, and the techniques for converting data into evidence increasingly use big data, artificial intelligence and machine learning.
Strengths and limitations of each source
Myriad big data sources are now available, including some considered fit for regulatory decision-making, and each has its main strengths and limitations.
- Clinical trial data, interventional and non-interventional: well structured, of high internal quality, shared non-selectively, with public trial documents such as protocols and analysis plans; but data formats and variable definitions are not standardised across trials.
- Spontaneous adverse drug reports: a legal and regulatory framework, consistency at global level, good at detecting new risks, and many dimensions of safety concern, from medication errors and quality defects to abuse, misuse and occupational exposure; but under- and over-reporting, and bias when a concern reflects media attention.
- Drug consumption data: large populations, but no individual patient data.
- Administrative claims data: consistent coding, longitudinal records with longer follow-up in the EU and in countries with public healthcare, national representativeness, linkage to other sources, complete drug exposure and location data for geocoding; but heterogeneity, misclassification of diagnosis, exposure or outcome, possibly out-of-date data, possible gaps in secondary care, little clinical detail, data protection law that may block linkage between providers, no lifestyle or socio-economic factors to control confounding, and no over-the-counter drugs.
- Electronic medical records: diverse clinical data that complement claims, longitudinal, with more valid diagnoses than claims and linkage to other sources; but heterogeneity, privacy concerns, often only one care setting, and the same gaps in lifestyle factors and over-the-counter drugs.
- Patient registries: consistent data, established large programmes, and the ability to observe disease course and the effects of new treatments; but limited to particular procedures, diseases or settings, possibly out of date, with discrepancies between the data collected and the data a regulator requests, inconsistent in quality across registries, and possibly needing source data verification.
- Biomarkers, including "omics" data: precision medicine and markers of disease or response; but high genetic variation, false positives and negatives, further validation needed to connect markers to outcomes, little publicly accessible clinically meaningful genomic information, no standardisation, privacy concerns, especially in rare diseases, and heterogeneity.
- Medical imaging: generally consistent and widely used in trials, with unexplored potential in many therapeutic areas; but hard to access, ethically complicated to share, and hard to integrate with other data.
- Social media: the reach of the internet, many kinds of data, and data patients chose to share; but heterogeneous, unspecific and prone to bias, with limited follow-up, hard to verify, silent on patient characteristics, outside good clinical practice, of uncertain validity and reliability, and a privacy risk.
- Mobile health and wearable devices: biometric data that may allow confounding to be controlled, patient-centric and continuous real-life data rather than episodic clinic data, data readily available for research on platforms that often link directly to an electronic case report form, and monitoring that can guide dosing; but further validation needed to tell clinically important signals from noise, unknown sensitivity, precision that is not the same as accuracy, output that varies across devices and with how much users interact, unfamiliarity with interpreting the data, awkward timing of surveys against other healthcare data, and privacy and security risks such as hacking.
About 90% of medical data are images of ever higher quality and resolution, and much of it is stored unanalysed. To use these data, the UK Biobank Imaging Study aims to build longitudinal records combining volunteers' brain, heart and body imaging with biomarker and genetic results, physical measurements and self-reported health and lifestyle data, linkable to their National Health Service records.
The World Health Organization defines mobile health as medicine and public health supported by mobile devices collecting data through symptom-monitoring applications, implantable diagnostics and wearable motion detectors. In the 12-week Lilly Exploratory Digital Assessment Trial sponsored by Eli Lilly and Apple, a model applied to data from phones and wearables distinguished people with mild cognitive impairment or early Alzheimer's disease from those without, suggesting impairment might be detected early. Such tools should be further assessed to ensure they are "reliable, validated, reproducible, and predictable" before they are used for regulatory decision-making. The US Food and Drug Administration's MyStudies app, released in 2018, is a platform for multi-site or multi-database studies that collects primary data directly from patients' mobile devices, and it was to be linked to the individual's electronic medical record and enhanced with electronic consent, eligibility testing, survey delivery, notifications and data validation. Data collection through mobile devices was expected to become more common in clinical studies following the app's release, and the app was seen as holding great potential for pragmatic trials, which give evidence on an intervention used in clinical practice and so may guide policy, and for observational trials, which have no intervention. During COVID-19 the Scripps Research Translational Institute launched the DETECT study in March 2020 to collect data from smartwatches and activity trackers for early detection of viral illness, Stanford Medicine started a COVID-19 wearables study for the same purpose, and in April 2020 the two formed a consortium with Fitbit to pool data for wearables research.
Linked data sources
Big data sources are multiplying as existing sources are linked across networks into larger platforms that serve wider areas of expertise, and clinical study designs incorporating big data have started to appear, a phenomenon expected to grow. Much remains to be done in assuring the accessibility, validity, quality and privacy protection of big data. The US Food and Drug Administration's Sentinel system monitors its regulated products by extracting electronic health records from many networks, mostly health insurers, and collaborates with over 40 other networks through three centres for operations, innovation, and community building and outreach. One Sentinel collaborator, the Patient-Centered Outcomes Research Network (PCORnet), joins over ten networks holding richer medical-record data, such as laboratory results, vital signs, biospecimen, genomic and patient satisfaction data. Its pragmatic ADAPTABLE study (NCT02697916) compared 81 mg and 325 mg of aspirin in about 20,000 patients, using existing health records and a web-based patient portal to identify eligible patients, obtain consent, randomise and follow them. In the future it may be possible to create a complete, longitudinal record of an individual starting from the omics level. Collaborations between academia, companies and regulatory authorities culminated in the Electronic Medical Records and Genomics (eMERGE) Network, which since 2007 has brought together researchers in genomics, statistics, ethics, informatics and clinical medicine to combine a biological materials repository with medical-record systems for research at the genomic level. In April 2020 the UK Biobank announced that it would grant researchers access, for health-related research, to the health data of its 500,000 participants, including COVID-19 test results, primary care, hospital and intensive care data. Personal omics profiling, high-frequency wearable data, imaging and ever larger platforms with patient-centred interfaces are expected to change greatly how clinical research is done.
Big data raises questions of regulation. Privacy is one: in 2014 the editors of Scientific American judged current law not even remotely prepared for biometric data such as DNA and iris scans, whose legal status was unclear while the potential for misuse was obvious. Two other important considerations are interoperability, the ability to exchange and use information safely, securely and effectively between devices, products, technologies or systems, and usability, the extent to which designated users can use a product with effectiveness, efficiency and satisfaction in a specified context. Several government agencies and nonprofit organisations set rules and standards for health data exchange, among them the US Office of the National Coordinator for Health Information Technology, whose 2015 roadmap aimed at a learning health system continuously improving data exchange, information governance, interoperability, health record usability, patient safety, provider satisfaction and population health; Health Level Seven International, a not-for-profit providing standards for sharing health information; Integrating the Health Enterprise, a nonprofit initiative of health care professionals and industry to improve how computer systems share health care information; and the family of ISO and IEEE standards for the interoperability of personal health devices.
From data to evidence
The conversion of real-world data into evidence runs in stages, each needed for the result to be reliable and actionable. Data are collected from many sources, including behaviour, socio-economic status and treatment adherence as well as clinical records; integrated and harmonised, since sources use different formats and standards; analysed with machine learning, regression and other methods to find patterns and relationships; and applied to individual clinical decisions, policy design and the safety and efficacy of treatments. Real-world evidence is not static: data keep arriving and being analysed, so the evidence base evolves as a continuous feedback loop.
Real-world studies and randomised trials
For more than five decades randomised controlled trials have been the gold standard for demonstrating the therapeutic benefit of medical interventions before marketing authorisation, using standardised methods such as randomisation and blinding to minimise bias. They are typically resource-intensive and time-consuming and often run in relatively homogeneous populations defined by restrictive eligibility criteria, so their results may not always generalise to broader, more diverse populations, which can leave questions about effectiveness and long-term safety unanswered. The overall low generalisability of trial results to routine clinical practice requires new approaches in clinical research.
A trial is planned in meticulous detail, from selecting similar patients against strict criteria to prespecified assessments at regular timepoints, and treatment can be randomised; a real-world study can also be planned in detail, but it is observational, and neither its environment nor the patients who seek care can be strictly controlled. Treatment and assessment are at the discretion of the professionals caring for each patient, shaped by clinical experience, local considerations such as availability and reimbursement, and the patient's goals, lifestyle and preferences, which also influence how often patients see their doctor, are tested, fill prescriptions and take their medicines. So where trials typically offer gold-standard treatment and regular, detailed assessment for every participant, comparable care in the real world is unrealistic to expect, and documentation of care varies between settings, with many real-world sources unstructured and more variable than the standardised outputs of trials.
| Randomised trial | Real-world study | |
|---|---|---|
| Classification | Interventional | Non-interventional (observational) |
| Patients | Homogeneous; often healthier than the average patient | Heterogeneous; may have multiple comorbidities and variable presentations |
| Population size | Small relative to the affected population; set by power for the primary outcome | Can be much larger; often unspecified and set by data availability |
| Care team | Experienced in the condition; supported by detailed protocols and a study team | No guaranteed disease-related experience |
| Treatments | Prespecified; often an investigational product; restricted concomitant therapy | Local standard of care, depending on availability and accessibility; wide range of concomitant therapies |
| Comparisons | Placebo or another medication; designed to detect differences between groups | Often designed to find associations rather than conclude causality |
| Treatment assignment | Randomised | At the treating physician's discretion |
| Blinding | Double-blind, single-blind or open-label | None |
| Data collection | Prospective; structured, at consistent intervals | Retrospective or prospective; often unstructured, at various intervals |
| Assessments | Prespecified to collect the data that answer the study question | Part of routine care at variable times; often not made with research in mind |
| Key outcomes | Efficacy and safety | Effectiveness and safety; natural history; disease burden and unmet needs; treatment patterns; costs |
| Duration | Short; up to several years depending on the research question; limited by feasibility and cost | Can be much longer; may cover decades in a retrospective analysis or prospective registry |
| Documentation | Protocol, statistical analysis plan, final study report | Varies in level and detail |
Strict eligibility can make a trial population more homogeneous, which can reduce variability, increase precision and power and dampen known and unknown confounding, but can make the results less generalisable to routine practice. Rare-disease studies often struggle to enrol enough participants, rigorous trials are often so expensive that in effect only large pharmaceutical sponsors or consortia can run them, and trials may be too short to show late adverse events or effects in incurable chronic conditions needing lifelong treatment. Trial findings often fail to apply to populations varying in genetic background, comorbidity and treatment regimen, an efficacy–effectiveness gap that limits the generalisability of trial outcomes and challenges regulatory decision-making. To bridge it, pharmaceutical companies, regulatory agencies and health technology assessment organisations increasingly integrate real-world data with randomised trials. Trials remain indispensable for understanding how a new treatment behaves, with a notable level of internal validity, but in many cases they cannot provide enough external validity to extrapolate their conclusions point by point to the whole population.
Cooperative research among Spanish haematologists began in the early 1970s with the care protocol, a consensus document setting out the key guidelines for diagnosing and treating a disease. Participating clinicians collected an essential data set on their patients with that disease and sent it, by fax, to a national coordinator, whose aggregate analysis yielded conclusions about response to the treatments used, while clinicians monitored their results in real time and agreed improvements to the procedures. By modern standards it was rudimentary, with no prior authorisation, no informed consent and no external monitoring, but it covered every patient with a given clinical profile without special selection and studied daily practice directly, so that it could be called something like a prototype of modern real-world research.
As research activity developed, clinical research came under legal and ethical regulation and new drugs advanced, trials became particularly important, planning optimal experimental conditions and selecting patients by rigid homogeneity criteria to test the efficacy and safety of new therapies. Because trials in many cases cannot be extrapolated to the whole population, it has become necessary to return to these origins and recover real-world research, which by the agreed definition includes observational and registry studies, patient self-reported data such as surveys, quality-of-life questionnaires and social media testimonials, administrative databases, and electronic registry data including laboratory data, digital records and patient complaints.
What real-world data can and cannot show
Post-marketing phase 4 studies usually focus on safety, yet patients, payers, government agencies and companies each want to know how well a drug works in the real world. Registries, electronic health records, claims databases and mobile devices promise abundant, cheap and fast data from thousands or millions of patients, apparently more representative than trial populations, which appeals to companies keen to cut the cost and time of bringing treatments to market. Such easy access may sound too good to be true, and several major obstacles stand in the way of reliable conclusions about effectiveness.
The first is data quality. A trial protocol specifies which data are collected, when and how, on a case report form designed for the trial under Good Clinical Practice, with monitors verifying completeness and accuracy, and real-world data are unlikely to share any of these qualities. They are collected for purposes other than research and will not be optimal for the question of interest, and they risk letting the available data set the research question; they may be inaccurate through human error or faulty devices, cannot be verified against source data, and carry random errors and systematic bias that are difficult to identify and impossible to correct; they have many values missing, with no information as to why; and the variables recorded, measurement methods, interoperability, formats and standards usually vary greatly across patients, providers, devices and companies. The second is the lack of randomisation: whenever doctor or patient chooses the treatment, groups taking different treatments are likely to differ systematically, often in ways difficult to identify, and the reasons for the choice are almost never recorded. The comparison is then likely to be biased, and there have been well-publicised cases in which observational studies and trials reached opposite conclusions. Although regulators have long used such data to evaluate safety, using them for unbiased evaluations of effectiveness is challenging and needs sophisticated methodology. The third is methodology: no amount of data can in itself evaluate a treatment, and data become evidence only inside a research study answering a specific question, with defined dataset formats, a comprehensive protocol and detailed analysis plans for reproducibility. Data from novel sources such as wearables must accurately reflect the clinical outcome studied.
Real-world data are informative as a supplement to randomised trials in certain contexts but are unlikely to replace them. Until a treatment is widely used there are no real-world data on it, so first approvals still rest on trials, whose close monitoring also protects patients taking an unapproved drug; the picture differs when a new drug is compared with a widely used one or established drugs are tried in a new disease. In an idealised theoretical framework, evidence can be placed on a continuum running from the classic randomised trial at the efficacy end, through hybrid, pragmatic and large simple trials, to observational studies on electronic records and claims at the effectiveness end. Integrating real-world data alongside randomised trials can enhance evidence generation with insight into broader populations, long-term effects and real-world treatment effectiveness, so appropriately designed and executed analyses of real-world data can complement trials and fill their gaps.
Questions real-world studies answer
The questions real-world studies answer fall under three headings. The first is understanding a disease, population and standard of care: how many people are affected and how that will change, the stages of a disease and the patient journey, and unmet needs under current treatment. Such studies are especially valuable for rare diseases, and their data can be reused elsewhere, as when natural-history data form an external control within a single-arm trial. The second is treatment use, safety and effectiveness in larger, more diverse populations over longer periods: who is on long-term treatment, whether they take it as regularly as expected, what else is prescribed, rare adverse events, risk-management measures and safety signals after authorisation, and effectiveness that confirms or extends trial efficacy. The third is burden and healthcare impact: direct costs of medicines, procedures and visits, indirect costs such as lost work and caregiver support, cost savings, and quality of life, used to show a medicine's clinical and cost-effectiveness or affordability. Real-world data may, for example, quantify the burden of disease in years of life lost or lived with disability, supporting value comparisons between an existing and a future treatment.
Real-world evidence gives insight into real-world treatment patterns and unmet medical needs, helping to prioritise research and allocate resources. Real-world research is particularly valuable for evaluating safety and effectiveness in rare or severe diseases, paediatric populations and other groups often under-represented in trials, because it shows treatment effects in diverse populations receiving care in conventional settings. Evidence from real-world studies has informed disease histories, post-marketing safety surveillance, quality-of-life outcomes and treatment effectiveness in clinical practice.
Key areas of application include the following. In medication safety, claims and health records can be used to identify potential adverse drug reactions and detect safety signals that may not be evident in controlled trials; in chronic disease management, records and registries track long-term outcomes of diabetes, heart disease and asthma and help refine guidelines; in pharmaceutical development, prevalence, demographics and treatment patterns inform trial design and post-marketing surveillance monitors long-term safety and efficacy; and in healthcare policy, analysing data from different populations can give policymakers insight into the social, economic and geographic factors that contribute to inequities.
Real-world studies examine practice without leaving out any type of patient. They cannot replace the safety and efficacy data of trials, but they can help to consolidate them by comparison across many settings, so they tend to give fuller information on safety and effectiveness. They are therefore key to subtle trends that need samples of thousands or hundreds of thousands, or decades of follow-up. This is especially necessary in haematological malignancies, where the constellation of biological variants makes it very hard to assemble enough cases of one entity, such as a mutation or set of mutations, in a trial, and where the genetic and epigenetic phenomena require far more information and cross-referencing of data. Real-world studies are also needed where only a single arm is possible because the disease is serious and rare and randomisation is often infeasible or unethical, so that external controls from trial databases or earlier real-world studies are the only comparison. They also capture patient-perceived outcomes, of particular interest in haematological malignancies, which often involve considerable morbidity.
In haematology the priority needs differ by disease. In monoclonal gammopathies, especially multiple myeloma, real-world research completes the evaluation of new therapies in patients with comorbidities who are often under-represented in trials and helps build predictive models of medium- and long-term response, notably in smouldering myeloma and specific genotypes. In acute myeloblastic leukaemia it clarifies the prognostic value of genetic mutations, the right indication for haematopoietic stem cell transplantation and comparisons across diverse populations; in acute lymphoblastic leukaemia, long follow-up of new targeted therapies and comparisons of polychemotherapy regimens across age strata; and in bone marrow failure syndromes, chiefly aplastic anaemia, a database large enough to study immunosuppression and transplantation in a disease where sample size is always the greatest handicap. In chronic myeloproliferative and lymphoproliferative disease it shows the real-life impact of inhibitors and advances more personalised treatment, and in lymphomas it allows long-term follow-up after immunotherapy and emerging therapies in refractory disease. Across all of them, extensive biological research is crucial for discovering biomarkers that predict treatment response more precisely and for collecting fuller information on idiosyncratic toxicities and delayed adverse reactions, and here real-world data prove particularly useful for tracking second malignancies and for studying peculiar and unexpected toxicities, such as those of CAR-T therapy.
Clinical research is trending towards maximum sample universalisation, aiming not to lose a single case worth studying, and this will exponentially increase the complexity of research systems and make them ever more dependent on technological advances. Nano-robotics, already explored for some therapeutic purposes, may have a role in collecting data from inside cells; clinical and laboratory diagnostics need to become more reliable, faster, more efficient and less invasive; and wearables need to become more discreet and less disruptive to daily life so that patients comply with them.
Purpose and ethics of real-world evidence
One editorial holds that real-world data and evidence are no longer auxiliary concepts but are becoming central to how health, disease and care are understood. One author describes integrating real-world data and evidence into healthcare practice as a monumental shift with the potential to transform how health systems operate, enabling a more personalised, data-driven approach, and sees the future of healthcare as integrating them into every aspect of decision-making. Randomised trials answer whether a treatment works; real-world data and evidence say how it works in real life, raising the questions of how patients experience treatment in the everyday world and what happens between visits, outside hospitals and beyond clinical endpoints, and they shift the focus from the controlled to the lived, from the prescriptive to the observed and from the theoretical to the ethical. Patient-generated data, whether from electronic health records, mobile health devices or social determinants of health, are people's lives in data form, so privacy, autonomy and agency must be respected. Real-world evidence gives rise to a new ethics of knowing, in which evidence is not neutral but shaped by who collects it, how it is interpreted and for what purpose. Developing guidance on real-world data and evidence is about safeguarding trust as well as technical rigour, and transparency, inclusivity and participatory design must become the ethical cornerstones of integrating such evidence into health systems.
Diverse sources such as wearables, patient-reported outcomes and behavioural data show how treatments work for real people in real contexts, and capture lifestyle, social environment and personal goals usually left out of trial data. They let in populations often excluded from trials: older adults, people with comorbidities, rural communities and ethnically diverse groups. In health economics real-world evidence helps allocate resources by actual utilisation, in rare diseases real-world data are the only practical route to understanding natural history and treatment response, and in pharmacovigilance real-world evidence captures safety signals trials miss. In each case the opportunity is not only to observe but to act: to refine policy, improve practice and empower patients. The tools exist to detect patterns across vast datasets, tailor interventions and predict outcomes, and the question is whether this opportunity will be used to reinforce equity or to widen gaps. In the same editorial view, real-world data and evidence are less a technical evolution than a cultural shift in how evidence is thought about, emphasising practicality, responsiveness and accountability. The same editorial urges that real-world evidence be about what matters as well as what works: it must address real needs, reflect lived realities and be applied with practical wisdom, and evidence should be reimagined as a commitment to accountability and care rather than only a pathway to approval.
The medical writer's role
Medical writers must understand big data and the strengths and potential limitations of the data used, and should consider the impact of big data on study design and on authoring protocols and clinical study reports. Leveraging big real-world data is a challenge for writers more accustomed to smaller clinical trial databases, who need to learn about the reliability of such data and how they can be used. Everyday practice often reveals different patterns from controlled trials, and writers communicating real-world insights need a nuanced understanding of how to present them in ways that are both scientifically rigorous and accessible to diverse audiences.
In one practitioner's view, big data means more job opportunities for medical writers, but also changing clients and a changing regulatory environment. The same practitioner expects the exponential growth of healthcare data to present medical writers with two main challenges: learning to communicate with data professionals, and understanding the regulatory rules for sharing and documenting medical data. Facing the growing volume of medical data, medical writers should learn to speak the language of data and catch up with the regulatory rules, so that they can convey the information clearly and accurately to their readers.
One editorial holds that writers are not merely translators of data but stewards of meaning: it is up to them to help clinicians, patients, payers and regulators understand what the data mean and what they do not, and they must advocate for ethical reporting, resist oversimplification, and ensure that patients are co-authors of understanding whose realities shape the decisions made and the words chosen as well as the data collected. With data faked, manipulated, misused, stolen and deleted, especially in the age of AI, writers are also urged to act as data stewards, helping to protect the quality and integrity of the data they handle.