Reference interval verification

Run this study#

What you are checking: whether an existing reference interval fits healthy people from the population your laboratory serves.

Suggested starting plan: a common starting cohort is 20 eligible subjects per partition, with one result per person. The CLSI implementation-guide sample describes small-cohort verification. Two partitions start with 40 people and 40 results. Follow your laboratory's written rule for the final count and any further cohorts.

Calculation minimum: one subject gives counts. A limit on the share inside the interval needs at least 2 subjects in the partition, and each stage of a staged rule needs at least 2.

1. Decide first. Define who is eligible, any partitions, the required count and the accept/reject rule. Create the study with the claimed limits and unit. Enter your staged rule on Set up, or set the smallest acceptable share under Acceptance limits.
2. Recruit and measure. Select healthy subjects by those rules, following the stated preparation, collection and handling conditions. Give each person a coded ID and one partition, and measure under routine conditions.
3. Import one result per subject. On Data, choose Import a file and choose your file (start from Blank template (CSV) or reference_verify.csv). Answer any question the dialog asks, then choose Import into this study. Check repeat IDs.
4. Calculate and apply your rule. Choose Calculate results. Review the below, within and above counts for each partition. Staged rules count subjects in entry order, so use one only if your written rule works the same way. A rule that needs a separate new cohort needs a separate study. On Report, choose Download PDF.

What you get: the below, within and above counts for each partition and the outcome of your limit and any staged rule. If the interval doesn't fit, investigate and follow your planned next step before adopting it.

Purpose#

You have a reference interval from a package insert, the literature or an earlier study. Before adopting it, test a small group of healthy subjects from your own population and ask: how many fall below, within and above the claimed limits?

The study checks a two-sided interval, for example 4.0 to 11.0, or a one-sided lower or upper limit.

When to use it#

Use it when the limits already exist. To calculate new limits from your own subjects, use Reference interval establishment.

Study setup#

1. In your project choose New study. Enter Analyte and Unit, and select Verify a reference interval under Reference intervals.
2. Choose Reference limits: two-sided, one-sided lower or one-sided upper. Enter the Claimed lower reference limit and Claimed upper reference limit: both for a two-sided check, with the lower below the upper, and only its own for a one-sided check. Both may stay blank while you plan. Choose Create study, or Create and import data if you already have results. If you later switch to a one-sided check, clear the field labelled Unused lower limit — leave blank (or Unused upper limit — leave blank).
3. On Set up, enter the staged rule if you use one (see Staged verification rules).
4. If you use a percentage rule instead, under Acceptance limits choose Enter your own limit and Share of results inside the interval, then enter the smallest acceptable percentage. See Acceptance limits.

The design needs:

• Healthy reference subjects who meet your eligibility rules, each with a coded ID.
• One result per subject. Repeat draws are averaged, and the subject counts once.
• Optionally a partition for each subject, for example adult and senior. Partitions are counted separately. Without partitions, all subjects are in partition 1.

Entering data#

Choose Import a file; see Importing and mapping.

The analysis needs exact values. Correct a censored value such as <0.01, or exclude it with a reason; see Corrections and exclusions.

Fix these before you can calculate:

• missing claimed limits;
• a missing analyte or unit;
• a subject entered in two partitions;
• a cohort with no eligible subjects;
• marked rows, for example a missing subject ID or an exclusion without a reason.

Statistics#

• Subject value = mean of that subject's included draws. Limits are inclusive: below means value < lower, above means value > upper.
• Within-interval share = 100 × within / eligible subjects.
• Review fences: type-7 quartiles, Q1 − 1.5·IQR and Q3 + 1.5·IQR, flagged only when strictly beyond (NIST/SEMATECH box plot; R boxplot.stats).
• Q–Q positions: Blom (i − 3/8)/(n + 1/4). Histogram: ceil(√n) equal-width bins, at most 40.

See Methods and sources.

Staged verification rules#

Enter your written staged rule on Set up:

• For each stage: Subjects tested, Accept when ≤ this many outside and Reject when ≥ this many outside. Choose Add stage for each further check (up to 10) and Remove stage to drop one. Use whole numbers, at least 2 subjects per stage. Subjects tested rise from stage to stage, the accept number is below the reject number, and the reject number can't exceed the subjects tested.

Within each partition, subjects are taken in entry order. At each stage, the number outside the claimed limits among the first n subjects is compared with the stage's numbers. At or below the accept number accepts. At or above the reject number rejects. Anything between moves to the next stage. With too few subjects for the next stage, the rule is inconclusive and says how many more are needed.

Example (synthetic rule, same file as below): stage 1 = 10 subjects, accept ≤ 0, reject ≥ 3; stage 2 = 20 subjects, accept ≤ 2, reject ≥ 3.

The rule's own check on the Why panel is Met for senior and Undecided for adult.

Acceptance limits#

The dialog and its choices are in Acceptance limits. For this study:

• Share of results inside the interval is the percentage of eligible subjects within the claimed limits, for example "at least 90".
• Each partition is compared separately, and each must meet the limit.
• A staged rule is added to the limits as Recorded verification rule. A rejection fails the study. An unresolved rule leaves it Undecided, even when overall coverage passes.

See Criterion outcomes.

Worked example#

reference_verify.csv holds 25 synthetic white blood cell counts (10^9/L) from 24 subjects: 14 adult (subject S003 has two draws, 5.4 and 5.6) and 10 senior. Several values sit exactly on 4.0 or 11.0.

Setup: two-sided, claimed 4 to 11 10^9/L; no staged rule; one limit, Share of results inside the interval at least 90 %.

What the saved result shows#

• Adult: 15 rows became 14 subjects because S003's two draws averaged to 5.5.
• Values exactly on 4 or 11 count as within.
• The review list flagged no values.

The study status is Criteria not met.

Reading the results#

Results show one partition at a time. With several, choose the Partition at the top.

Verification counts gives the claimed limits, Eligible independent subjects, the below, within and above counts, and Results exactly on a claimed limit. A one-sided check shows only its side.

Why shows the share of your subjects inside the claimed limits, which is what your limit is checked against. It has no confidence interval and describes your cohort, not the population coverage. With 14 subjects, one subject moves it by about 7 percentage points.

Verification conclusion appears when you entered a staged rule: Within the accept bound, At or beyond the reject bound or Inconclusive under the rule, with the reason.

Subject results and claimed limits plots Raw points or a Histogram, with dotted lines at the claimed limits.

Outside 1.5·IQR whiskers — review lists values beyond the fences with View row links. Flagged values stay in the result. Check the record behind each flag. To remove a subject, exclude the row with a reason.

Including excluded rows (for information only) appears when a partition has excluded rows. It shows the counts with them added back.

Distribution diagnostics describe shape only; see Reference interval establishment.

Watch for many subjects outside on one side, which suggests a shifted population or a calibration difference, several flagged values, or a partition with very few subjects.

What to do when the study fails#

• Criteria not met: check the claimed limits and unit against the source, the partition column, and each subject's eligibility. If all are right, the interval doesn't fit your population. Test more subjects, or establish your own with Reference interval establishment.
• Inconclusive under the rule: if it asks for more subjects, test them and recalculate. If the rule is incomplete, fix the staged-rule fields.