The Minimum Data Dictionary Every Medical Study Needs

A variable name is not a definition. In one medical dataset, `visit_date` may mean the scheduled appointment date; in another, the date a measurement was recorded; in a third, the date later chosen as analytically representative. Those columns can all be valid, yet combining or analyzing them as if they mean the same thing creates an error that statistical software cannot detect.

The practical defense is a study data dictionary: a controlled specification of what each variable means, how it is represented, where it came from, and how it may change. The NIH Common Data Elements Repository defines a common data element as a precisely defined question paired with allowable responses. That principle scales down to an individual study. Even when no external standard fits perfectly, each locally defined variable should be precise enough that a second analyst can reproduce its meaning without interviewing the person who created it.

01Start with the scientific meaning

Every variable should have a human-readable label and a complete operational definition. “Blood pressure” is insufficient. Specify systolic or diastolic pressure, the measurement device or source, participant position if relevant, number of readings, selection or averaging rule, and time relative to an intervention or visit.

The definition should also distinguish data collected directly from data derived later. Age at enrollment, for example, may be calculated from date of birth and enrollment date. The dictionary should identify the source fields, formula, rounding rule, and handling of incomplete dates. If a derived variable changes when code changes, the team needs to know which rule produced each analysis dataset.

02Record representation, not just description

A useful dictionary answers how the value appears in the file. At minimum, record the field name, data type, permitted values or range, units, date or time format, and missing-value codes. Controlled terminology matters because labels that look similar may not be equivalent. CDISC terminology provides standardized codelists and valid values for clinical-research data, while the NIH repository allows teams to find and compare established elements. Local studies need not copy a standard blindly, but they should reuse a relevant standard when it matches the intended concept and document any adaptation.

Units deserve explicit attention. A laboratory measure stored as `5.6` is uninterpretable without knowing whether the unit is mmol/L or mg/dL. Do not rely on a column heading, spreadsheet color, or team memory. Store the unit in the dictionary and, when records may contain different units, store the unit with each observation or convert values through a documented rule.

03Make timing and provenance visible

Clinical meaning often depends on time. Define the target time point, allowable window, and rule for multiple measurements. If baseline is “the last non-missing value before first dose,” say so. If the closest measurement within seven days is selected, state how ties are resolved. This turns a silent analyst choice into a reviewable study rule.

For provenance, identify the original source, collection form or system, source field, transformation, and responsible owner. A field extracted from an electronic health record should name the table or interface and extraction version. A manually adjudicated outcome should identify the adjudication procedure and status codes. Provenance is what lets a reviewer trace a reported result backward through the analysis dataset to the source record.

04Treat missingness as designed data

Blank, unknown, not applicable, not collected, and not yet available are different states. A single empty cell collapses those distinctions and can make quality problems look like participant characteristics. The dictionary should list allowed missingness codes, their meanings, and which are valid for each variable. The analysis plan can later decide how to handle them, but it cannot recover distinctions that were never recorded.

05Control definitions through versions

A data dictionary should have a version, effective date, change history, and named approver. Changes after data collection begins are sometimes necessary, but overwriting a definition erases the meaning under which earlier records were created. Keep prior versions, document the reason for change, identify affected variables and records, and decide whether data must be recoded or analyses repeated.

The minimum dictionary is therefore not a decorative spreadsheet. It is a compact scientific contract: concept, operational definition, representation, timing, provenance, missingness, ownership, and version. Before analysis, select ten critical variables and ask whether an independent analyst could implement each one exactly. Any answer that depends on oral explanation identifies a data-quality task that should be completed before modeling begins.

This article is educational and intended for research purposes. It does not provide individual medical advice, diagnosis, or treatment. No patient data were used.