A zero, an unreported value and a value you have not checked are different entries. When extracting numbers from clinical papers, preserve that distinction before exporting a spreadsheet or making a chart.
This small Python example uses synthetic teaching records. It does not describe findings from real trials.
rows = [ {"study": "Teaching A", "events": 0, "status": "reported", "locator": "Table 2"}, {"study": "Teaching B", "events": None, "status": "not reported", "locator": ""}, {"study": "Teaching C", "events": None, "status": "not verified", "locator": ""}, ] for row in rows: if row["status"] == "reported": if row["events"] is None or not row["locator"].strip(): raise ValueError("Reported values need a value and a source locator") display = str(row["events"]) else: if row["events"] is not None: raise ValueError("Unverified or unreported rows must not carry a value") display = row["status"] print(f"{row['study']}: {display}") Output:
Teaching A: 0 Teaching B: not reported Teaching C: not verified The validation deliberately does not replace None with 0. A reported zero retains its source locator. A row marked “not reported” means your extraction found no reported value in the material you checked; it does not prove that no event occurred. “Not verified” records an unfinished check.
If you inspected only an abstract, record that access level alongside the status. Before analysis, decide how missing entries will be handled in the analysis plan. This display check does not choose a statistical missing-data method or assess study quality.
For a handoff, keep the paper identifier, source locator, access level and review status with each extracted value. Someone else should be able to reopen the same passage without guessing which denominator or outcome you meant.
We are the RigorMed team. Our public clinical evidence example provides a two-paper starting point. It prepares an editable task draft; it does not run a task automatically. Research execution uses credits.
The code and records above are a teaching example prepared with AI assistance.