Two histories with unequal observation
Imagine two people with similar symptoms. One has several visits and testing; the other has no contact with the system. A database may contain a diagnosis for the first and nothing for the second.
The difference in documentation does not prove that only the first person had the condition.
The observation process can be part of the result
Access, insurance, illness severity, service availability, and care preferences may influence visit patterns. Those factors can also relate to the outcome being studied. Simply adjusting for the number of visits is not guaranteed to solve the problem and can introduce other complications depending on the causal question.
Define what the dataset sees
- Which organizations contribute records?
- What counts as an observation period?
- Can care outside the network be detected?
- Does absence of a code mean absence of disease, or only absence of documentation?
These questions belong in the study design, not just a limitations paragraph after analysis.
A large dataset can still have uneven visibility. More rows improve neither completeness nor representativeness automatically. Interpreting healthcare data requires understanding the process that generated the records as well as the values stored inside them.
