Four-Stage Data Quality Framework for Multi-Source Oncology Real-World Data Integration: Validation Using Health Datasets Mapped to the OMOP Common Data Model

Imagine you are a detective trying to solve a mystery, but instead of a crime scene, your crime scene is a mountain of medical records. In the world of modern medicine, especially when studying cancer, researchers don't just rely on controlled lab experiments anymore; they look at "Real-World Data." Think of this as the messy, everyday paperwork from hospitals, clinics, and pharmacies—the actual notes doctors write, the lab results they print, and the prescriptions they hand out. This data is gold because it shows how treatments work in the real world, not just in a perfect test tube.

However, this gold is often buried in dirt. One hospital might write "heart attack," while another writes "myocardial infarction," and a third just uses a code like "410.0." If you try to mix these records together to find patterns, it's like trying to build a house with bricks, LEGOs, and smooth river stones all mixed up. The structure collapses. To fix this, scientists use a "Common Data Model," which is like a universal translator or a standard blueprint that forces every piece of information into the same shape and label. But here's the catch: just because the bricks look the same shape doesn't mean they are the right bricks. You could have a brick that says "Patient died in 2020" but also has a note saying "Patient visited the doctor in 2021." That's impossible, but a computer might not notice unless someone checks very carefully. This is where data quality comes in: making sure the story the data tells is actually true.

---

The Four-Stage Detective Squad

In this study, a researcher named Samuel Maniraj Selvaraj built a super-organized, four-step checklist to make sure two different piles of cancer patient data were ready to be mixed together. He didn't just glance at the data; he ran it through a rigorous "quality control" tunnel, treating the data like a high-stakes product that needs to be perfect before it can be used to make life-or-death decisions.

Think of the process like preparing two different batches of ingredients to bake a giant, complex cake for a pharmaceutical company. You have Batch A (from one set of hospitals) and Batch B (from another). Before you can mix them, you have to make sure they aren't spoiled, that they are actually the ingredients you think they are, and that they haven't been swapped out in the middle of the process.

Stage 1: The Recipe Check (Mapping File Validation)
First, the team checked the "recipe cards." These cards tell the computer how to translate the messy hospital codes into the clean, standard language (called the OMOP Common Data Model). The rule here is simple: Does every single code have a valid translation? The team ran exhaustive checks on every record. While the study confirmed that the mapping files met the structural rules required to proceed, the specific percentage of records that passed these checks was not disclosed in the final report. The key takeaway is that the foundation was deemed valid enough to move forward, but the exact "score" of the recipe cards remains private.

Stage 2: The Assembly Line Check (Structural Mapping Validation)
Next, they watched the assembly line. This is where the actual data gets moved from the messy hospital files into the clean, standard database. They checked to make sure no records fell off the conveyor belt and that no records got duplicated (like a patient appearing twice by accident). They also checked that the data landed in the right boxes. For example, a "drug" record shouldn't end up in the "surgery" box. The study reported that the data successfully moved from source to target for all validated domains, achieving a "Pass" status. However, while the structural transformation was confirmed as successful, the precise rate of data loss or the total volume of records processed was not explicitly calculated as a percentage in the public findings.

Stage 3: The Meaning Check (Conceptual Mapping Validation)
This is where things get tricky. Just because a record moved to the right box doesn't mean it makes sense. Imagine a label that says "Apple" but is actually a picture of a banana. The computer might think it's an apple because the box is labeled "Fruit," but a human needs to check if the label is actually correct. The team had experts review the data to ensure the medical terms actually matched what the doctors meant. They confirmed that the data was semantically correct enough to pass, but the study explicitly noted that the underlying "Conceptual Concordance Rate"—the exact percentage of records that were perfectly mapped—was not disclosed. The experts gave the green light, but the granular statistics on how many "apples" were actually "bananas" remain hidden.

Stage 4: The Delivery Check (Data Synchronization Validation)
Finally, they checked the delivery truck. The data had been cleaned and organized in a secure vault (the main database), but researchers access it through a public portal (a website). The team made sure that what was in the vault was exactly the same as what was on the website. They wanted to ensure that no one was looking at old, stale data while the vault had fresh data. This "synchronization" was also a perfect pass, confirming that the data available to users matched the certified database.

The Hidden Glitches: When "Perfect" Isn't Perfect

Even though the data passed all four stages and got a "Green Light" to be used, the team didn't stop there. They ran a special "plausibility" test, which is like asking, "Does this story make sense?" This is where they found some very strange, funny, and concerning glitches that a simple computer check might have missed.

They found three main types of "time-travel" errors:

1. The Time-Traveling Patient: Some records showed patients having medical events before they were even born. For example, a patient had a "Condition Occurrence" (like a diagnosis) dated before their birth date. In Data Source B, there were 9,213 patients with this error, and in Data Source A, there were 63.
2. The Ghost Visits: They found patients visiting the doctor after they had already died. In Data Source B, 354 patients had a visit record dated after their recorded death date. This is a huge problem for studies trying to figure out how long people live after a diagnosis.
3. The Magic Pills: They found drug records with impossible amounts. Some patients were recorded as taking 50,000 units of a drug (which is physically impossible), while others were recorded as taking 0 or even negative amounts.
   • In Data Source A, a massive 129,923 patients had drug records with zero or negative amounts.
   • In Data Source B, 6,033 patients had this same issue, and another 1,772 had the "too many pills" error.

Why This Matters

The most important finding of this paper isn't that the data was "good" (it was), but that even "good" data can hide massive, silly mistakes that would ruin a study if left unchecked. If a researcher tried to study how well a drug works without catching these errors, they might conclude that a drug is amazing because the "ghost patients" who visited after death seemed to survive longer, or that a drug is useless because the "negative pill" counts messed up the math.

The paper proves that you need a strict, four-step rule-based system to catch these errors. It's not enough to just say "the data is clean." You have to check the recipe, the assembly, the meaning, and the delivery. And even then, you have to look for the "time-travel" glitches.

The author concludes that this four-stage framework is a reliable, repeatable way to certify that cancer data is ready for use. It's a safety net that catches the "impossible" stories before they can trick scientists. While the study didn't calculate the exact percentage of errors (because the total number of patients and specific coverage rates were not shared), the sheer number of "ghost visits" and "negative pills" found shows that this kind of deep checking is absolutely necessary. Without it, the real-world evidence we rely on to fight cancer could be built on a foundation of time-traveling patients and magic medicine.