The research project was approved by the Institutional Review Board of our institution (Ethics Committee on research involving humans of the University Hospital of Geneva, CER 13_027_R), in compliance with the Helsinki Declaration. Since a better characterization of the packets by a targeted DECT series was considered useful to improve the clinical management of the conveyors, and therefore their safety, the Ethics committee waived the need for consent.
All consecutive adult persons suspected of having ingested drug packets within the Geneva State territory (n = 720) during a 30 months period of time (November 2010 to May 2013) were brought to our emergency department for a LDCT examination. LDCT images were immediately interpreted for the presence of drug containers by the radiology resident and the attending physician on call, while the suspect was still on the CT table.
Whenever LDCT was reported positive for body-packets (n = 120, 16.5 %), a 3-cm thick CT series was performed, targeted on a drug container, using a dual-energy protocol (Gemstone imaging, GSI).
When the presence of body-packets was reported at LDCT, the conveyors were hospitalized until all packets were collected. For each patient, four of these packets were analyzed by a dedicated laboratory, which evaluated their size and weight, the percentage of cocaine and also performed a qualitative analysis of the cutting agents. These data were retained by the scientific police up to the end of the study.
Exclusion criteria
Cases were excluded from the study when laboratory data could not be obtained or when dual energy series could not be performed. When the cocaine concentrations measured in various packets from the same conveyor were dissimilar (variation of more than 20 % between the highest and the lowest drug concentration), the case was also excluded.
Reference standard
The mean percentage of cocaine obtained from the 4 measures from samples of homogenous drug packets was considered reference standard.
Technical imaging parameters
LDCT were performed with a 64-rows GE 750 HD CT (Discovery 750 HD CT, GE Healthcare, Milwaukee, USA), from lung bases to pelvis, without intra-venous, oral or rectal contrast material, using the following parameters: 64 ×1.25 mm collimation, pitch 1.375, gantry rotation period 0.7 s, tube potential 120 kV, tube charge per gantry rotation 25.2 mAs, reconstruction slice thickness 2.5 mm, using 40 % ASIR (adaptive statistical iterative reconstruction).
GSI series were performed on the same CT scanner as LDCT, on a 3 cm thick area, using the following parameters: 64 ×1.25 mm, pitch 1.375, 140 kV, gantry rotation period 0.6 s, variable potential output of 80 kV and 140 kV, tube charge per gantry rotation 180 mAs.
Effective dose of LDCT and of GSI series
The dose of radiation delivered by low-dose CT and the dual energy series were estimated from the mean normalized values of effective dose per dose-length product (DLP) for the abdomen, as described by Shrimpton et al. [10]. The effective dose for LDCT was 1.4 mSv (DLP = 94.15 [mGy.cm]) for men, and 1.2 mSv for women (DLP 83.54 [mGy.cm]). For the dual energy series, the effective doses were 1.27 mSv and 0.9 mSv for men and women respectively.
CT images analysis
LDCT images and dual CT data were transferred and analyzed on a dedicated GE Advantage workstation (ADW, GE Healthcare, version 4.6), using a GSI viewer 2.0-2L.
LDCT and dual-energy CT images were prospectively analyzed by one of the two board certified attending radiologist of the emergency radiology unit, with 15 and 17 years of experience respectively in reading abdominal CTs. Both radiologists were blinded to the results of the chemical analysis of the packets. The radiologist measured the mean density (HU) in a packet on the LDCT series, using a relatively small (60 mm2) circular region of interest (ROI), to reduce the risk of partial volume averaging (Fig. 1). The same packet was selected within the 3 cm thick GSI series; the mean effective atomic number Z(eff) (Fig. 2), as well as the mean density (HU) at 40 and 140 keV, were measured in a 60 mm2 circular ROI. The difference between the attenuations at 40 and 140 keV was calculated and referred to as ∆HU.
Statistical analysis
We obtained mean values and standard deviations of all continuous variables. Then we examined associations between cocaine content (expressed in percent) and various measures of radiologic density using scatterplots and summarized them through Pearson correlation coefficients.
Since the median packet weight in our study population is 10 g, a cocaine content of 35 % or higher is close to the reported lethal threshold (about 3 to 4 g) in case of packet rupture. Therefore, we dichotomized cocaine content as below or above 35 %. We also dichotomized cocaine content as below or above 50 %, to focus on packets with major lethal risk in case of rupture. Then we examined associations between the cocaine concentration at these thresholds with measures of radiologic density by means of receiver operating characteristic curves (ROC) and corresponding areas under the curve (AUC). An AUC of 0.5 means no association while an AUC of 1 reflects perfect discrimination between low-content and high-content containers.
For the most promising predictor, we looked at cutoff values that achieved a good compromise of sensitivity and specificity, and reported statistics of test performance.
Analyses were performed using SPSS version 18 (PASW Statistics 18, SPSS Inc, Chicago, USA).