- Published
- Updated
A diagnosis or procedure code used as an endpoint does not automatically establish accurate capture of a clinical event. Recording purpose, coding practice and database coverage can affect sensitivity and the risk of misclassification in the resulting outcome algorithm.
Define the question and assumptions
Write the clinical event definition before the code list. Identify source records, expert review and a sampling approach for validation. Reviewing only algorithm-positive records cannot answer every performance question. Assess whether the reference review itself could be incomplete or biased and report performance measures alongside the actual scope of validation performed.
Hypothetical study example
In a hypothetical study, one code represents suspicion at some centres and confirmed diagnosis at others. The team reviews sampled records by centre and evaluates the algorithm. It does not transfer a high-accuracy claim from one source directly to every setting with a different recording process.
Design review checks
- Record the clinical definition and version of the algorithm code list.
- Explain validation sampling and the source used as the reference assessment.
- Assess possible misclassification by centre, period and treatment comparison group.
Record the decision and its limits
Accepted evaluation presents performance estimates with uncertainty and retains the population and period in which validation occurred. Do not modify an algorithm to increase a treatment difference. Record the rationale and impact of justified changes. Use a suitably limited outcome definition where the database cannot support a definitive clinical-event claim.
Frequently asked questions
- Does a code establish a confirmed clinical event?
- Not always. Examine its recording purpose and validation evidence.
- Is validation from another source sufficient?
- Assess transferability when local recording processes or populations differ.