Indirect CT venography procedure
Contrast medium injection
The intravenous CM, iopamidol 350 mgI/mL (Iomeron 350H, Bracco) or iohexol 350 mgI/mL (Omnipaque 350H, GE Healthcare), was administered using a power injector (CT Exprès TM 4D, Bracco Injeneering S.A.) through the antecubital vein followed by a saline flush.
CM dose was estimated based on the perceived body composition of the patients, rated as obese, average, or muscular. Obese patients received an approximation of 1.3 mL/kg, average 1.8 mL/kg and muscular 2.3 mL/kg. According to the standard protocol, the total amount of administered CM ranged from 90 to 180 mL.
CT acquisition parameters
All CT scans were performed on a 256-slice multi-detector CT scanner (Brilliance iCT; Philips Healthcare).
The main scan covered the region from the diaphragm to the feet and was started 120 s after the start of CM injection with a fixed CM injection time of 40s and a fixed post-CM injection delay of 80s. In addition, a single acquisition was taken at the level of the proximal popliteal veins at a fixed scan delay of 30 s, 60 s, 90 s, 150 s, 180 s and 210 s after the start of CM injection.
The tube voltage was chosen based on the patient’s body mass index (BMI) to achieve highest possible vessel enhancement without compromising image quality. At the level of popliteal veins, 80 kilovolt (kV) was used for patients with BMI ≤ 25 and 100 kV for patients with BMI > 25.
Patient monitoring
During contrast administration and scanning, hemodynamic data (heart rate, blood pressure and CO) were measured continuously and non-invasively by Nexfin HD monitor (BMEYE), a photo-plethysmography device using an inflatable finger cuff as the only interface with the patient [
29,
30]. Means of recorded samples (4 min) covering from before the start of CM injection to after completed scanning were used in the analyses.
Patient variables
Data on the patients’ age, sex, body weight and self-reported height were collected.
BMI was calculated as (body weight in kg) / (height in m)2.
Image reconstructions and evaluations
Image reconstructions
Conventional images were reformatted to a slice thickness of 3 mm. All images were sent to a commercial workstation (IntelliSpace Portal 10, Philips Healthcare) and a Carestream Vue picture archiving and communication system version 12 (Carestream Health) for evaluation.
Objective measurements of examination quality
CE was measured in the popliteal veins bilaterally by placing a circular region of interest cursor on the single acquisitions taken at 30 s, 60 s, 90 s, 150 s, 180 s, 210 s and from the full acquisition scan taken at 120 s, and registering the attenuation values (in Hounsfield units). The circular regions of interest were drawn as large as possible within the anatomic limits of the vessel lumen, avoiding artefacts. The CE measurements were performed by a radiographer with more than 10 years of experience in CT cardiovascular imaging and served as the objective measurement of examination quality.
Subjective rating of examination quality
The subjective rating was performed independently by two cardiovascular interventional radiologists with more than 10 years’ experience. Examinations were rated at each scan delay from 30 to 210 s, and the raters were not blinded to the scan delay.
Examinations were rated adequate if opacification of the popliteal veins and image quality were sufficient for detecting or ruling out DVT; otherwise, they were rated inadequate. Disagreement between the two radiologists was independently rated by a third radiologist and resolved by the median score for all three radiologists.
Statistical analysis
Descriptive statistics are reported using mean (SD) or number (%), as appropriate. The time-density curves for the popliteal veins were illustrated using a boxplot (median, 25% and 75% percentiles).
Objective evaluation of popliteal CE was performed by estimating the proportion of examinations exceeding predefined cut-offs of 70, 80, 90 and 100 HU at scan delays between 90 and 210 s. For standardisation, only patients with a complete set of measurements at scan delays between 90 and 210 s were included.
Differences between scan delays in proportions exceeding the cut-offs were tested with a logistic mixed model trained on binary endpoints with patient identity as a random effect and time as a fixed effect. The binary endpoint was whether the popliteal CE exceeded the given threshold.
We also estimated the peak CE defined as the highest attenuation value for each patient, and the TPCE defined as the scan delay at which the highest attenuation was achieved.
The subjective rating of examination quality according to scan delay was reported as the percentage of adequate examinations between 90 and 210 s. The inter-rater agreement (%) and Cohen’s kappa score were estimated [
31]. The strength of agreement was rated as poor, slight, fair, moderate, substantial or almost perfect according to the Kappa statistic scores < 0.00, 0.00–0.20, 0.21–0.40, 0.41–0.60, 0.61–0.80 and 0.81–1.00 respectively [
32]. Differences between scan delays in the proportions of adequate examinations were tested with a logistic mixed model trained on the subjective ratings by the raters, using patient and rater identity as random effects and time as a fixed effect.
A Bayesian mixed-effects non-linear spline regression model was used to model the variation of CE and subjective rating of examination quality with time [
33]. We used fixed effects in the form of an I-spline basis of degree three and knots at 90 s and 150 s, and random effects of a constant and linear term for each individual and a constant term for rater identity for subjective ratings [
34]. The I-spline fixed effects provide a basis for six monotonic functions that can capture the global non-linearity in the relationship between the outcome variable and time, while the random effects allow for variation in scale between individuals and raters. The I-spline basis is shown in Fig.
3a. A Gaussian likelihood was used for the CE measurements and a logistic likelihood for the binary subjective ratings.
The trained mixed-effects spline model allows for estimation of the time at which the latent variables underlying the models reach a plateau: for the Gaussian likelihood, this can be interpreted as a smoothed value of the contrast enhancement, while for the logistic likelihood the latent variable can be interpreted as the log odds ratio of the rating being judged acceptable.
For both CE and subjective rating, we defined the plateau threshold as the maximum of the model latent variable mean across scan delays minus half the predicted standard deviation, with the prediction including all fixed and random effects. The time until the plateau was reached was defined as the minimum time point at which the predicted mean value is greater than the plateau threshold.
A linear mixed-effects model was used to evaluate associations between patient factors and the TPCE. A significance level of 0.05 was chosen for the analysis.
Patients with missing data on heart rate (6%), systolic blood pressure (8%) or CO (6%) were excluded from the analyses of baseline characteristics and predictors of TPCE.
Patients with missing attenuation measurements (19%) (due to thrombus at one or both sides (n = 3) or missing image acquisitions at any scan delay (n = 7)) were excluded from the cut-off analyses to archive a complete set of data for comparison.
Statistical analyses were performed with R version 4.1.0 (R Foundation for Statistical Computing) using the libraries brms and lme4, or Stata version 15.1 (StataCorp) [
35].