Background: A complete blood count (CBC) is the most commonly ordered laboratory test for patient diagnosis and treatment.1 Clinical decisions are often made on the basis of a single result which is rarely repeated. The availability of more rapid tests for CBCs that are accessible and approved for Clinical Laboratory Improvement Amendments (CLIA)-waived use at the point of need increases the likelihood that serial testing may occur over short periods of time.2 Understanding natural CBC variation in healthy individuals may help clinicians interpret serial test results which may have a range of values.

Objective: We sought to evaluate CBCs from blood drawn over a short time period and assessed in separate, accredited labs to understand biological and laboratory-based variability.

Methods and findings: Ten healthy adult volunteers with a self-report of being healthy were recruited by a study nurse and signed informed consent between April and May 2017. At baseline, 1 hour and 2 hours, six 3 mL tubes were drawn each time and sent as routine tests in duplicate to three College of American Pathologist certified clinical core laboratories for CBC: Lab 1 used a Beckman Coulter LH750 (Beckman Coulter, Atlanta, GA), Lab 2 used a Sysmex XE2100 (Sysmex America, Lincolnshire, IL) and Lab 3 used a Sysmex XN9000 (Sysmex America). The total time range covered by this study was 120 min, consistent with waiting and treatment times in US Emergency Department visits.3 Participants were not permitted to eat or drink for the entire 2-hour study period. All tubes were drawn in the morning to avoid the previously noted diurnal variation.4 5 The Latin square design was used within each subject to counterbalance order effects between blood draws and between tubes within blood draws (three pairs of tubes within three draws). Using the Latin square design, participants were randomised to blocks of six to achieve balance across orderings. While order effects are assumed to be small, this design was used to mitigate any potential bias in the case that they were non-trivial. Linear mixed effects models were used to estimate the variability between draws within subjects, and variability between replicates within blood draws. Nested random effects were used to account for correlation within subjects across repeated measurements, and a random intercept was added to account for correlation due to between-lab variability. Intraclass correlation coefficients (ICC) were estimated from the random effects, giving the correlation within each level of the data: correlation within subjects, correlation within a particular draw of a subject and the correlation between a pair of tubes from a participant within a single blood draw. These represent the average correlation between all pairs of observations in each hierarchy of the data.