Healthcare has no shortage of data. The harder question is whether that data is reliable enough to influence decisions about the safety and performance of a medical device. Kiran Chitambare, Senior Manager at Intuitive Surgical, brings more than 15 years of clinical data management experience spanning data quality, regulatory compliance, and clinical research. Her experience with standards and requirements including ICH-GCP, 21 CFR Part 11, CDISC and CDASH has given her a close view of what it takes to turn healthcare information into evidence that can withstand regulatory scrutiny.
That distinction is becoming increasingly important as healthcare organizations generate information from electronic health records, device registries, insurance claims, medical records, digital health technologies, remote monitoring systems, and even patients themselves.
Collectively, these sources can provide real-world data, or RWD, about what happens when medical technologies move beyond controlled clinical studies and into routine healthcare.
For medical device companies, the opportunity is considerable. Real-world data can potentially help researchers understand how a device performs across broader patient populations, different clinical settings, and longer periods of time.
But having more data does not necessarily mean having better evidence.
For regulators such as the U.S. Food and Drug Administration, the key question is whether the available data are reliable, relevant, and scientifically sufficient to answer a particular regulatory question.
In other words, “good enough” for regulatory use is not simply about the size of a database. It is about whether the information can credibly contribute to evidence about safety, performance or effectiveness while preserving data integrity and traceability.
The Regulatory Question Has to Come First
A dataset containing millions of patient records may appear impressive, but its value depends on what researchers are trying to establish.
If the regulatory question concerns the performance of a device in a particular patient population, the first consideration is whether that population is actually represented in the data.
Are the patients similar to those for whom the device is intended? Are the outcomes captured by the dataset clinically meaningful? Does the information clearly establish exposure to the device? Is the follow-up period long enough to observe the outcomes researchers need to study?
A dataset can be large and technically sophisticated yet still be poorly suited to a particular question.
Consider follow-up time. A relatively short period may provide useful information about immediate procedural outcomes. The same dataset may be inadequate if the question concerns complications that emerge six months or a year later.
This is why relevance is fundamental. Instead of starting with the data that happen to be available and asking what can be done with them, regulatory-quality research should begin with the question and determine whether the data are fit to answer it.
Accuracy Means Connecting the Right Patient to the Right Device
Accuracy in real-world data extends well beyond correcting data-entry errors.
Medical device evaluation may require researchers to establish a dependable connection between the patient, procedure, device, and clinical outcome.
Was the correct patient identified? Was the procedure recorded accurately? Can researchers determine exactly which device was used? Are important outcomes documented correctly? Have critical variables been verified when necessary?
These questions become more complex when information is assembled from multiple sources.
A patient record might be combined with information from a device registry, claims system or remote monitoring platform. If those records cannot be linked accurately, the analytical dataset may present a distorted picture even when each individual system contains relatively good data.
The quality of the final evidence therefore depends partly on maintaining that chain of accuracy from the original clinical event through the eventual analysis.
Device Identification Deserves Particular Attention
Medical devices present another challenge: the technology itself can change.
A single device family may include several models. Software-enabled devices may receive updates. Components can change between generations. Different manufacturing lots may sometimes become relevant to safety investigations.
That means researchers often need more than a general statement that a patient received a particular type of device.
Depending on the regulatory question, relevant information may include a Unique Device Identifier, device model, software version, lot number or serial number.
Without sufficient device identification, researchers might identify a safety signal but struggle to determine whether it applies broadly or only to a particular model, version or production lot.
As devices become increasingly software-driven and connected, this issue may become even more important. A patient's clinical outcome may need to be interpreted not simply in relation to a device name, but to the exact configuration or software version in use at the time.
Missing Data Can Change the Story
No real-world dataset is likely to be perfectly complete.
The regulatory concern is therefore not simply whether information is missing. It is what is missing and whether those gaps could materially change the conclusion.
Regulators may consider whether key endpoints are adequately captured, whether sufficient follow-up exists, whether adverse events are documented, and whether important variables are available consistently across the patient population.
Suppose a study examines outcomes after a device procedure, but a significant proportion of patients disappear from the dataset before the relevant follow-up period.
The patients who remain may have excellent outcomes. But without understanding what happened to those who were lost to follow-up, confidence in the overall conclusion may be reduced.
Adverse-event capture creates a similar concern. If complications are systematically missing or documented inconsistently, the dataset could create an incomplete picture of device safety.
Not every missing field carries the same weight. An absent administrative variable may have little influence on the analysis. Missing a primary safety or effectiveness outcome can be far more consequential.
Clinical data management therefore requires understanding the significance of missing information rather than looking only at a percentage of completed fields.
Chitambare's experience has included activities such as query management, serious adverse event reconciliation, laboratory reconciliation and preparation of clinical data documentation and submission packages, processes designed to identify and resolve exactly these kinds of data-quality concerns.
Consistency Makes Information Comparable
Real-world data often come from multiple hospitals, physicians, information systems and geographic regions.
That diversity is one of its strengths because it can reflect actual clinical practice. It is also a source of potential inconsistency.
One hospital may define a complication differently from another. A clinical variable captured in a structured field at one institution may appear only in physician notes at another. Coding practices can vary between systems or change over time.
Even the same organization may modify its data collection methods during the period being analyzed.
This makes standardization important.
Consistent definitions, uniform collection procedures, standardized coding practices, and version-controlled forms can help researchers determine whether information collected in different places or at different times is genuinely comparable.
Version control is particularly important.
If the definition of an endpoint changes midway through a study period, investigators should know when the change occurred, why it happened and whether it affected the resulting analysis.
Otherwise, an apparent shift in patient outcomes could actually be a shift in documentation.
Traceability Builds Confidence in the Final Evidence
One of the simplest questions about regulatory-quality data may also be one of the most important:
Where did this information come from?
By the time real-world data reach an analytical dataset, they may have gone through numerous transformations.
Information might be extracted from a clinical system, mapped into another format, cleaned, linked with another database, recoded, and then prepared for statistical analysis.
Every transformation introduces another point where errors or misunderstandings can occur.
Traceability allows reviewers to work backward.
Can the original source be identified? Can important values be verified? Is there an audit trail? Were transformations and corrections documented?
If a variable was recoded, researchers should be able to explain the methodology. If two datasets were linked, the linkage process should be clear. If information changed during cleaning, there should be documentation of what was changed and why.
The principle is straightforward: a final analytical value becomes more credible when the path that produced it can be reconstructed.
That discipline is already familiar in clinical data management. Chitambare's professional background includes clinical research form design, data dictionaries, validation checks, user acceptance testing, and oversight of documented database changes.
Representativeness Defines How Widely Evidence Can Be Applied
One of the strongest potential advantages of real-world data is its ability to capture patients who may not always be represented fully in tightly controlled clinical studies.
Once a medical device enters routine clinical practice, it may be used across different age groups, demographics, geographic regions, and levels of clinical complexity.
Patients may have multiple comorbidities. Physicians may use technologies somewhat differently. Healthcare infrastructure may vary substantially between locations.
Those differences can make real-world evidence valuable, but researchers need to know who the dataset actually represents.
Demographics, comorbidities, geographic distribution, and variation in clinical practice can all influence the interpretation of results.
A database containing hundreds of thousands of patients may still have limited generalizability if most patients come from a narrow geographic or clinical environment.
Representativeness does not mean every dataset must perfectly mirror every potential device user. It means researchers should understand its boundaries well enough to know where the findings can reasonably be applied.
Different Data Sources Answer Different Questions
Real-world data are not one uniform category.
Electronic health records may provide rich clinical detail about diagnoses, procedures and outcomes.
Device registries can offer structured longitudinal information about particular technologies.
Claims databases can provide large-scale information about healthcare encounters and utilization.
Traditional medical records may contain clinical context that structured databases miss.
Disease registries can provide longitudinal information about defined patient populations, while digital health technologies, patient-generated data, and remote monitoring systems increasingly capture what happens between conventional healthcare visits.
Each source has different strengths and limitations.
The objective is therefore not to find the largest available source. It is to identify the source, or combination of sources, that contains the information necessary to answer the regulatory question with sufficient reliability.
From Healthcare Data to Regulatory Evidence
The growth of real-world data is expanding what medical device researchers can study, but it also creates a risk of confusing information abundance with evidence quality.
A credible real-world dataset should allow researchers to explain who the patients are, which devices they received, what happened to them, how important outcomes were measured, what information is missing, and how the data moved from their original source into the final analysis.
Producing that evidence is rarely the work of one discipline. It requires collaboration among clinical professionals, data managers, statisticians, programmers, medical experts and regulatory teams. Chitambare's own work has involved coordination across clinical teams, biostatisticians, programmers, medical monitors and external vendors, illustrating the cross-functional structure behind reliable clinical evidence.
As real-world evidence assumes a larger role in medical device development and evaluation, organizations with the most data will not automatically have the strongest evidence.
The advantage will belong to those that can demonstrate why their data can be trusted.
Ultimately, the regulatory question is not simply:
“How much data do we have?”
It is:
“Do we have the right data, with enough quality, relevance, and traceability to support the decision we are asking it to support?”
