Missing and censored values

A blank, <0.5 or hemolyzed is kept as written and its row is flagged. Correct each flagged row, or exclude it with a reason, before you calculate. See Corrections and exclusions.

When a file holds such values, Import a file asks what to do with them: exclude them all with one reason, or keep them flagged to correct in Data. See importing a file.

Blank results#

A row with an empty result, on either side in a paired study, shows Fix errors. A notice above the grid counts the blank rows and offers one button for all of them.

Leave out blank rows in one click#

1. Choose the button in the notice, such as Exclude 3 rows. Blank rows are excluded with the reason Not measured. No value entered., or with the reason you already typed; the report lists them as not measured. Completely empty rows are removed instead. If all of them are empty, the button reads, for example, Remove 2 empty rows.
2. Choose Save changes or Calculate results.

Blanks are flagged in these fields:

• method comparison and qualitative agreement: both results
• linearity and near-cutoff: expected concentration and result
• stability: elapsed time and result
• interference: result, and interferent concentration on test aliquots
• other studies: the result

In a 21-position carryover study, a missing or excluded result leaves that run incomplete.

Results reported as less than or greater than a limit#

<0.5, ≤0.5, >1000 and ≥1000 are censored values. The true value lies somewhere beyond the limit. The row stays flagged until you give an exact value or exclude it. Detection limits can rank < results instead, as described below. In qualitative studies, <0.5 is flagged as an undeclared label.

• Exclude the row with a reason such as "Below the measuring range (<0.5 mmol/L)".
• Correct the value, if an exact result exists.
• If many results are censored, consider a detection limits study for the low end.

Limit of blank: the rank-below option#

In a detection-limits study, set Set up › Censored results › Censoring policy to one of:

• Refuse “<” results (row errors), the default. Each < result is flagged until you exclude or correct it.
• Rank “<” results below exact results. For a nonparametric LoB or LoB verification, each <x blank ranks below every exact result. The blanks' mean, SD and minimum are then not calculated.

The calculation still stops if:

• x is above an exact blank result
• the LoB percentile falls among the censored results. Add more blanks.
• a censored result would enter a parametric LoB or an LoD

An LoQ material with a censored result is indeterminate. In detection limits, > and ≥ results are always flagged. See Detection limits.

Text, flags and formulas in a result column#

Qualitative results that are not reportable#

Set up lists the result categories, by default negative and positive, and the nonreportable labels, by default equivocal, invalid and indeterminate. Case and surrounding spaces don't matter.

A nonreportable label is valid and counted separately:

• In qualitative agreement, a specimen whose first result is nonreportable goes in the Nonreportable row or column, outside the main table. It counts as a disagreement in the all-pairs agreement.
• In near-cutoff precision, it is counted in the Nonreportable column and left out of that concentration's hit rate.

Any other label, or a blank category, is flagged. Correct it, or add the label on Set up.

How rows are counted#

Several rows can belong to one specimen or subject. See Uncertainty and denominators.

One side missing in a paired study#

In a method comparison, a row with one result blank is flagged. Exclude it, or add the missing result to that row when it arrives. Importing the missing result as a new row doesn't complete the pair. See Reviewing an append or a replacement.

In an instrument comparison, a specimen missing on one instrument is still used wherever both results exist. For that instrument's specimen limit it counts as not evaluated, so the limit can't be met. In a lot comparison, a patient specimen measured on one lot only is listed as unmatched and counts as not evaluated, so the lot can't pass.