An account of data entry inconsistencies and their impact on positron emission tomography quantification

Abstract

Reproducibility is essential to clinical application of positron emission tomography (PET) quantification. Human lapses in data registration and protocol compliance are pervasive sources of intrasite quantification variability. Although rarely assessed or reported, these lapses are ultimately a limitation to harmonization in multicenter clinical trials. A comprehensive account of their possible extent is relayed here.This is a retrospective audit of errors in manual registration of study parameters and in protocol adherence across a sample of in-center research projects over one year (201 patients, 222 PET/CT scans). Discrepancies in patient height and weight; tracer type; dose; injection; and scan times were listed. Correspondent variances in standardized uptake values (SUVs) normalized by body weight, SUV (BW), and body surface area, SUV (BSA), were assessed.Manual misregistrations totalled 41.8%. These were mainly small, but with a few large deviations, and most significant in weight (range: -1-100 kg) and dose (-19 to 12 MBq). Errors were more frequent and generally larger in non-routine studies. This also applied to protocol compliance. A 50.7% noncompliance was found with significant deviations in dose (-106 to 208 MBq) and especially in early scan uptake times (-37 to 54 min). Although misregistrations did not overall translate into significant SUV variability, noncompliance did. These errors contributed a factor 0.02 to 1.45 and 0.71 to 3.09 SUV (BW) change, respectively. SUV (BSA) saw a significant 21% to 22% decrease with mistyped height and weight.Inconsistency was frequent but less prominent in data entry than in protocol compliance. As both caused some substantial SUV variances, intra-site assessments and data checking are required for clinical trials.

Figure 1

Bar diagrams of percentage inconsistent data registrations (excl. missing entries) between patient records and DICOM metadata for all studies (A) and individual study types (B) in entries of patient height and weight, tracer type and dose, tracer injection time (T_inj) as well as early (T₁) and late (T₂) scan times.

Figure 2

Bland-Altman plots of differences in all registered patient height (A), weight (B), tracer dose (C), injection time (D), scan times (E), and uptake times (F) between data entries in the patient records, image DICOM metadata, and study protocol prescriptions, respectively. Solid and dashed lines indicate mean ± 2SD.

Figure 3

Number of entries noncompliant with protocol prescriptions shown as bar charts for patient record (A) and DICOM metadata (B) registrations, with and without deviations within allowed limits. Corresponding boxplots (C, D) of discrepancies, including (gray) and excluding (white) allowed deviations, display the range of variation in tracer dose [MBq] and uptake times [minutes] at early (ΔT₁) and late (ΔT₂) scans.

Figure 4

Variation in normalization factors, Ω, for SUV (BW) calculations with deviating parameters in FDG (A) and NaF (B) studies as well as (C) between entries most representative of true levels in acquired scans (Ω_acq) vs. protocol targets (Ω_proc), indicating parameter compliance variance. Similar plots for SUV (BSA) between discrepant patient record (Ω_rec) and DICOM (Ω_dic) entries of height and/or weight (D). BSA = body surface area, FDG = 2-deoxy-2-fluoro-D-glucose, NaF = sodium fluoride, SUV = standardized uptake value.