Missing data and detection limits in a table

Zero is a numerical value; an empty cell lacks an interpretable meaning without a legend. Treating them as equivalent can change averages, dispersion and relationships between variables.

A useful table distinguishes not measured, not detected, outside the quantifiable range and excluded for quality reasons. There is no need to invent an amount for every case. Preserve the reason so subsequent analysis can identify which observations actually support each comparison.

ICH Q2(R2) defines the detection limit in terms of a detectable amount, which is not necessarily quantifiable with suitable accuracy. A not-detected label therefore needs the method and applicable limit alongside it.

If two tables use different analytical capabilities, their labels do not automatically describe the same situation. Replacing them with a common number also requires explaining the assumption introduced. Official guideline.

A data-dependent-acquisition proteomics study used public datasets and simulations to examine the composition of missing values and the effect of different imputations on differential analysis.

The question was not simply which algorithm filled the most cells. The assumptions about missingness mattered because they changed later conclusions. The evaluation applies to those datasets and scenarios; it does not establish that every blank in a table results from low abundance. Primary study.

In targeted proteomics using proximity extension assays, another study compared imputed values with new measurements. Algorithms differed in how they preserved variability and biased subsequent analyses, with performance also varying between proteins.

This explains why a favorable overall correlation is insufficient to treat measured and estimated data as equivalent. New measurements provide a particularly useful check, but even that comparison does not guarantee precision for every individual protein. Primary study.

Readers need to distinguish observed values from estimates and understand the criteria for exclusions. Report how many observations support each summary: two means displayed with the same decimal precision may rest on very different amounts of data.

As a hypothetical example, if one group contains many nondetects, filling them with a constant can create apparent homogeneity. That regularity comes from an analytical decision. Presenting results under alternative assumptions helps assess how much the conclusion depends on that choice without turning imputations into new observations.