Five hundred thirty-six patients, referred by their general practitioner (GP) to the occult cancer pathway, were prospectively included in the study between May 2017 and November 2018. The ethical committee waived ethical issues and the study was approved by the Danish data authorities. Five hundred three patients gave written consent to participate in the project, and allowed access to image and clinical data.
Patients referred from the GP to the fast track package for suspected serious illness that could be cancer were eligible for inclusion. Exclusion criteria were missing written consent, scan protocol differing from the national guidelines, and allergies to iodine contrast media. Figure 1 provides an overview of the study design.
Patient in the study population presented themselves to the GP with numerous symptoms and risk factors (see Table 1). Unfortunately, the referral information is often incomplete.
An overview of the basic demographics is included in Table 2.
Spectral contrast-enhanced CT and contrast-enhanced CT
SCE-CT of the chest, abdomen, and pelvis was acquired on a 64-row dual-layer detector CT scanner (Philips IQon; Philips Healthcare). CT acquisition parameters were 64 × 0.625-mm collimation, kilovolt peak 120–140, milliamperes per second/slice 150–250, rotation time 0.75 s, reconstruction thickness 2 mm, increment 1 mm, pitch 1.078, FOV 35 cm, and matrix 512 × 512. Iodixanol 270 mg/ml (Visipaque® 270; GE Healthcare), or iohexol 300 mg/ml (Omnipaque® 300; GE Healthcare), was injected intravenously in weight-adjusted doses of 2 ml/kg body weight to compensate for differences in distribution volume, with an injection rate of 4 ml/s. A bolus tracking technique was used with a ROI in the descending aorta on the level of carina to compensate for differences in cardiac output. A threshold of 150 HU was used and CT was performed after a delay of 15 s for the chest and upper abdomen (late arterial phase), and 65 s for the abdomen (portal venous phase). The mean dose length product (DLP) of CT scans performed on the population was 2104 mGy·cm (CI95% 2064 to 2144). By spectral separation of the CT signal in the two detector layers, a spectral CT dataset was reconstructed. By weighted addition of the signal of the two layers together before reconstruction, a conventional CT dataset was reconstructed that possesses all features of a normal single energy CT in terms of dose  and image quality .
Reading of examinations
Scan data were transferred to a dedicated spectral workstation (Intellispace 9; Philips Healthcare) and divided into primary readings and secondary reading folders. The reading folders contained either the CE-CT or a superset of the conventional results and the CE-SCT. The provided spectral data include the following: virtual monoenergetic images (ranging from 40 to 200 keV), effective atomic number (Zeff, reporting the atomic number of the tissue), iodine-no-water (pure spectral decomposition of iodine and water. Calcium remains visible and is mimicked by iodine), iodine density (similar to iodine no water, but calcium is masked out), contrast enhancing structures (masking of iodinated tissues), uric acid (masking of uric acid containing tissue), and virtual non-contrast (VNC, a 70 keV map without the signal of iodine contrast).
Overlay images on the conventional series or the VNC was available for all datasets.
The scans were hereafter read in consensus by two experienced radiologists, with respectively 9 and 33 years of experience. In case of disagreement, a third radiologist would determine the outcome of the proposed findings.
All organ systems covered in the scans were reviewed. Initial reading was performed with virtual monoenergetic 40 keV to identify lesions that required further attention. In suspicion of disease, other spectral results were investigated. For instance, VNC was used in the case of a suspicion of calcification to discriminate calcium from iodine. In contrary, for low- or hyperattenuated lesions, iodine density, iodine no water, contrast enhancing structures, and Zeff were used to prove the presence of iodine.
To prevent recall bias, the primary and secondary reading were performed with an interval of at least 3 months and in random order. The radiologists were blinded from patient identifiers and earlier imaging, but had access to symptoms and a concise medical background of the patient mentioned in the referral.
All findings were entered into a RedCAP database . Up to 7 findings per patients were classified according to their severity and were scored for malignancy of the finding (1 = “certainly malignant,” 2 = “probably malignant,” 3 = “probably benign,” and 4 = “certainly benign”) and certainty of the finding (1 = “certain,” 2 = “almost certain,” 3 = “somewhat uncertain,” and 4 = “very uncertain”).
During the course of the study, we decided to record additional information. For a subset of the readings (308 spectral and 304 conventional), we recorded the reading time. For another subset (418 spectral and 414 conventional), we recorded the need for supplementary examinations. For the last 221 patients (221 spectral and 221 conventional readings), we also recorded whether the indicated supplementary examinations were part of routine or needed to improve the certainty of the finding.
Routine follow-up procedures included a wide range of examinations, including CT thorax in case of pulmonary nodules according to Fleischner Society criteria, multiphase CT or ultrasound in case of hyperdense lesions in the kidney, PET/CT in the case of suspected lung cancer, transvaginal ultrasound for ovarian lesions, dedicated adrenal CT for adrenal lesions, ultrasound and possible biopsy from thyroid lesions, etc. In the case of uncertainty, the follow-up procedures include contrast-enhanced ultrasound in liver or kidney lesions, biopsy or follow-up on enlarged/suspicious lymph nodes, endoscopies when intraluminal lesions, or mural thickening was suspected etc.
Review of findings
Individual findings were matched between the two readings, according to disease type and associated organ.
After a median follow-up of 21.3 months, the clinical status was recorded for every patient. When the clinical status could not validate a recorded serious first finding (cancer or another disease that would need immediate medical treatment), a more thorough search in the medical records was conducted for validation. When a serious finding was not considered in the medical history of the patient, the case was reported to the Ethical Committee for additional follow-up.
Every patient would now have an outcome value (1 = “cancer, with recorded proof,” 2 = “likely cancer, without recorded proof,” 3 = “unlikely cancer,” 4 = “proven benign,” − 1 = “finding not investigated,” O = “other proven serious disease”).
Most of the benign findings were not reported in the patient dossier and would get a score of “− 1.” There were also potential malignant findings without follow-up that received a score of “2” or “3.” Very unlikely cancers were often not investigated and patients often did not want follow-up diagnostics or were not able to cooperate/receive the suggested procedure. Other serious diseases (“O”) were conditions that required treatment within a short period of time.
Cancer, with recorded proof, was obtained by the following methods: biopsy in 48 cases, surgery in 9 cases, multidisciplinary team decision in 11 cases, follow-up in 2 cases, and in 2 cases the patient had a diagnosed cancer not reported in the initial referral.
We hypothesized that CE-SCT would detect more findings than CE-CT. We used McNemar’s test to identify types of findings that differed in frequency between the two modalities.
To test the hypothesis that spectral findings provide a higher confidence to the radiologist in the diagnoses compared to conventional findings, we used a two-sided two proportion z-test to find the types of findings for which the proportion of certain and uncertain findings differed between CE-SCT and CE-CT.
We compared sensitivity and specificity of malign findings between CE-SCT and CE-CT and tested for significance by means of McNemar’s test.
We performed descriptive statistics and box-and-whisker plots for the reading time per patient of both CE-SCT and CE-CT. Differences in reading time between CE-SCT and CE-CT based on the certainty of the first finding were also compared using a box-and-whisker plot.
Furthermore, we compared the frequency of supplemental procedures between CE-SCT and CE-CT by means of a two-sided t test.