Data Quality and Missingness
Data quality depends on fitness for purpose, while missing information can reflect both recording problems and meaningful patterns in care.
#Quality means fitness for a purpose
Weng describes completeness, correctness, agreement, plausibility and whether information represents the relevant time. A dataset can be suitable for one research question but inadequate for another. A file that passes a format check is not automatically reliable evidence.
WHO's data-quality toolkit addresses routine health-facility reporting through regular checks and reviews. Its scope is not a universal certificate for every research dataset. Read which dimensions were assessed and whether those checks fit the question being asked.
Evidence: Weng: clinical data quality across its life cycle / WHO: routine health-service data quality assurance
#A missing measurement can reflect the care process
Groenwold explains that the presence, timing or absence of a measurement in electronic health records can carry information about how care was delivered. A missing value is not always an accidental blank.
The commentary uses synthetic data to show how a model relying on missingness patterns can lose performance if measurement practices change. Read why information was absent and whether that mechanism is likely to stay the same. This does not make missingness a universally useful predictor or justify filling a blank with a normal result.
Evidence: Groenwold: informative missingness in health records
#Handling gaps requires assumptions
The BMJ missing-data tutorial distinguishes missingness mechanisms and explains that excluding incomplete records can reduce precision and sometimes bias results. It also warns that simple substitutions and single imputation are not generally valid solutions.
Multiple imputation creates several plausible completed datasets to reflect uncertainty, but its validity depends on suitable models and assumptions. It does not recover the true missing values. Read how gaps were handled and why that approach was considered appropriate rather than treating an imputed field as an observed measurement.
Evidence: BMJ: missing data and multiple-imputation assumptions
#Read the missing-data report, not only the result
The BMJ tutorial recommends reporting the amount of missing information, reasons when known, exclusions, methods and assumptions. It also discusses checking how conclusions depend on different assumptions. Those details matter when deciding how much confidence to place in an analysis.
Groenwold adds that a predictive pattern may depend on the original measurement process. A result can change when the workflow changes even if the model code stays the same. These are general research-reading principles, not an imputation recipe, diagnosis of a missing result or Mynd data-cleaning service.
Evidence: BMJ: missing data and multiple-imputation assumptions / Groenwold: informative missingness in health records
Source note
The sections above were checked against the linked sources. No clinical review has been performed. This is general research education, not a clinical guideline.