The Additive Value of Semi-Quantitative Analysis of 18F-FDG PET/CT For The Diagnosis of Device Related Infections In Patients With A Left Ventricular Assist Device.


 Background: Left ventricular assist devices (LVADs) improve quality of life and long-term survival in advanced heart failure, but device related infections (DRI) remain cumbersome. We evaluated the diagnostic capability of FDG-PET/CT and the additive value of semi-quantitative analysis for the diagnosis of DRI. Methods: LVAD recipients undergoing FDG-PET/CT between December 2012 and August 2020 for suspected DRI were retrospectively included. FDG-PET/CT was performed and evaluated according to EANM guidelines and assessors were blinded to the clinical context of included patients. Final clinical diagnosis of driveline infection and/or central device infection, based on multidisciplinary consensus and findings during surgery whenever performed, was used as the reference for the diagnosis. Results: 44 patients were evaluated for 62 episodes of suspected DRI. Clinical evaluation established driveline infection in 33 (54%) episodes, central device infection in 8 (14%) and combined infection in 2 (4%). Visual analysis of FDG-PET/CT achieved a sensitivity and specificity of 0.83 and 0.73, respectively, for driveline infections, while semi-quantitative analysis found comparable results (p=0.77). For central device component infections, visual analysis of FDG-PET/CT achieved a sensitivity and specificity of 1.0 and 0.26, respectively, while semi-quantitative analysis using SUVratio (background liver) achieved a sensitivity and specificity of 1.0 and 0.8, respectively. Both SUVmax and SUVratio outperformed visual analysis (p<0.001). Conclusions: FDG-PET/CT is a valuable tool for the assessment of DRI in LVAD recipients. Semiquantitative analysis significantly increases specificity of FDG-PET/CT for the analysis of central device components and should be considered in equivocal cases after visual analysis.


Full-text Introduction
Left ventricular assist device (LVAD) therapy increases survival and quality of life in patients with advanced heart failure, with median 5-year survival now approaching 50% [1]. However, device related infections (DRI) occur frequently, with an incidence of 18.1% during the first year after implantation and 11.9% per year in the following years [1]. Establishing the presence, extent and the severity of DRI, especially in deep and central infections, can be difficult as conventional imaging modalities such as echocardiography and CT are hampered by device-related artefacts [2,3]. Most infections occur at the site of the driveline exit from the abdominal wall. These driveline infections have the potential to become life-threatening in case the infection reaches the central device components (pump housing, inflow tract and outflow tract), especially when this is complicated by device-related bloodstream infection, which is accompanied by an in-hospital mortality rate of up to 50% [4][5][6]. Simultaneously, complicated device related infections can form a contraindication for heart transplantation, further increasing the risk of an unfavourable patient outcome.
Several studies assessed the diagnostic performance of [ 18 F]-fluorodeoxyglucose positron emission tomography combined with low dose computed tomography (FDG-PET/CT) for diagnosing driveline and/or central LVAD component infections [7][8][9][10][11][12][13][14][15] and two meta-analyses combined their results [15,16]. Visual analysis of FDG-PET/CT has a high sensitivity for establishing LVAD infections and a high but variable specificity, with a pooled sensitivity of 0.95 (95% confidence interval (CI) 0.89-0.97) and a pooled specificity of 0.91 (95% CI 0.54-0.99). The variable specificity may be caused by differences in patient selection, scan interpretation or specific technical issues. Furthermore, semi-quantitative analysis might increase the specificity of FDG-PET/CT. So far, only one study investigated the role of semiquantitative analysis for diagnosing infections of both driveline and central LVAD components [8], while two studies focused on driveline infections alone [7,10]. The results of these studies were mixed. Therefore, we evaluated whether (1) factors that may affect FDG-PET/CT quality, such as patient preparation using a high-fat, low-carbohydrate (HFLC) diet and antibiotic use prior to PET/CT, affected visual analysis, and (2) whether quantification of FDG uptake around the LVAD and driveline improves the specificity and overall diagnostic accuracy of FDG-PET/CT.

Patients
In this retrospective, dual-centre study, all consecutive patients that underwent FDG-PET/CT for assessment of suspected LVAD and/or driveline infections in the University Medical Center Groningen (UMCG) and Erasmus MC University Medical Center (EMC) were included from the start of using electronic patient files (first recorded case December 2012) until 31 August 2020.
Initial suspicion of infection was based on clinical history and presentation. Presenting symptoms of patients were recorded alongside general demographic data and information about the LVAD; e.g. indication for implantation, type and brand of LVAD and the time interval between LVAD implantation and FDG-PET/CT. Findings from clinical investigations such as inflammatory markers, blood cultures, cultures from exit-site swabs and findings during surgery (if performed) were recorded, together with information on the use of intravenous antibiotics at the time of the FDG-PET/CT, including the duration in days. If patients underwent FDG-PET/CT scans during multiple episodes of suspected infection, the later scans were only included if there were recurring symptoms after a symptom-free period of at least 1 month and after treatment for the first episode was fully completed (minimum interval between consecutive FDG-PET/CTs was 2 months) and special care was taken to avoid FDG-PET/CT scans that were performed to establish treatment effect. The study was approved and the need for informed consent waived by the Local Medical Ethics Review Committees of both centres due to the non-WMO (Dutch law on studies involving human subjects) nature of this study using retrospective data: protocol nr M19.223017 (UMCG) / MEC-2019-0613 (EMC).

FDG-PET/CT protocol
The protocol for FDG-PET/CT preparation that was followed for each scan was documented.
This included the duration of the fasting period before FDG-PET/CT, whether or not an HFLC diet had been used prior to the scan, patient blood glucose levels at the time of FDG injection, the injected FDG activity, and the used vendor type of PET/CT camera system.

Semi-quantitative analysis
After delineation of volumes of interest (VOI) the SUVmax and SUVmean were measured at 6 predefined areas of interest in each patient. These regions comprised three areas alongside the peripheral driveline tract (driveline exit site, suprafascial driveline tract, and subfascial driveline) and 3 areas around the central device components (intrathoracic driveline, inflow canula / pump housing, and the outflow tract) (Figure 1 and 2). Although the pump-pocket of the Heartmate II is technically located abdominally, we defined it as a central device component because of its direct connection to inflow and outflow tracts, and the clinical severity when the pocket is infected. The highest SUVmax value alongside the driveline and the highest value around the central device components were used for establishing the diagnostic accuracy of semi-quantitative analysis.
After delineation of reference regions in the thoracic aorta and liver, the SUVratio was calculated by dividing the highest SUVmax value alongside the driveline and the highest SUVmax value around the central device components by the mean activity in both reference regions separately. The aorta was carefully delineated to exclude the aortic vascular wall, and it was verified that the liver function tests were normal at the time of FDG-PET/CT in the included patients. All calculations were performed on EANM research ltd (EARL) reconstructed images [18].

Potential confounders
Factors that could potentially affect the assessment of either driveline infection or infections of central device components were identified for further analysis. These included patient age, gender, BMI, diabetes mellitus, the indication for LVAD implantation (ischemic or dilated cardiomyopathy), type of LVAD (either Heartmate II or Heartmate 3), type of PET/CT system, patient preparation using a high-fat low carbohydrate (HFLC) diet and the duration of antibiotic use prior to FDG-PET/CT.

Statistics
For demographic data, continuous variables are presented as mean +/-SD, while categorical variables are reported as frequencies. We used logistic regression to evaluate the accuracy of visual analysis and semi-quantitative analysis diagnosis driveline and/or central device infections. Potential confounders were tested using univariate logistic regression for effect on visual analysis and final diagnosis. Multivariate logistic regression was not performed on account of insufficient case numbers. For semi-quantitative analyses receiver operating characteristics (ROC) analysis was performed to determine the optimum threshold values to maximize sensitivity and specificity. For all statistical analyses a two-tailed P-value <0.05 was considered statistically significant. Analyses were performed using IBM® SPSS 26 [IBM Corp].

Patient characteristics
In total, 44 patients underwent a total of 70 FDG-PET/CT scans because of a clinical suspicion of LVAD and/or driveline infection. Eight scans were excluded because they were follow-up scans performed to establish treatment effect, leaving 62 scans for analysis. Of these, 20 patients (25 scans) were evaluated in the EMC, while the remainder of 24 patients (37 scans) were evaluated in UMCG. The indication for LVAD implantation was non-ischemic cardiomyopathy in 25 of the 44 patients (57%) and ischemic cardiomyopathy in the remaining 19 (43%) patients. The implanted LVAD system was either the Heartmate 3 (32/44 or 73% of patients) or the Heartmate II (12/44 or 27% of patients). For an overview of demographic data see Table 1.

FDG-PET/CT protocol
Preparation for the FDG-PET/CT was different for the two participating centres: While in both centres a fasting period of at least 6 hours was standard, in the UMCG the 24-hour high-fat low-carbohydrate diet was only formalized at the beginning of 2020, while in the EMC this had been the standard since October 2016 [14]. The HFLC diet was used to prepare patients for Optimal sensitivity and specificity was achieved using the SUVratio with the liver as a reference region. At a SUVratio cut-off value of 2.45 the corresponding sensitivity and specificity were 1.00 and 0.80, respectively. Using the thoracic blood pool also achieved good sensitivity and specificity. At a cut-off value of 3.04 these were 0.89 and 0.83, respectively. SUVmax alone reached its optimum at a cut-off value of 5.14, achieving a sensitivity and specificity of 0.67 and 0.80, respectively. The improvements of sensitivity and specificity compared to visual analysis were statistically significant: p<0.001, regardless of the semi-quantitative method used. The corresponding ROC curves for both driveline and central device components are shown in Figure 3.

Discussion
In this multicentre retrospective study the diagnostic accuracy of FDG-PET/CT for LVAD device related infections (DRI) was evaluated in 44 patients and 62 episodes of suspected infection, making it one of the largest cohorts for this indication to date [7][8][9][10][11][12][13][14][15]. Additionally, we specifically stratified for driveline infections and infections involving central device components, the assessors were fully blinded to the clinical context of patients and used both AC and NAC images. Furthermore, we included semi-quantitative analyses at predefines areas and performed in accordance with EARL recommendations.
Visual analysis also showed good sensitivity for diagnosing a central device infection.
However, specificity was extremely poor. Semi-quantitative analysis clearly outperformed visual analysis for this specific indication, with best results achieved using the liver as the reference region: specificity of FDG-PET/CT improved from 0.26 to 0.80 in central device compartments. By contrast, its high sensitivity remained unchanged at 1.0, compared with visual analysis, and this improvement was statistically significant. These results indicate that in all cases in which infection of central device components cannot be ruled out based on visual analysis, semi-quantitative analysis may significantly increase the diagnostic accuracy.
There are multiple explanations for the low specificity of visual analysis of FDG-PET/CT for central device components [21]. Firstly, in our study the assessors were blinded to the patients' clinical context, while in about half of earlier studies, assessors were not (fully) blinded, which may have led to observer bias. Secondly, many patients in our study had a heartmate 3 and distinguishing between sterile inflammation at the insertion of the pump in the left ventricle proved difficult by visual analysis, even while using NAC images. Another explanation for the low specificity and the heterogenous uptake in many patients could be that the outflow tract for both Heartmate II and 3 consists of polyethylenterephthalat (PET; Dacron®), a synthetic material known to cause reactive FDG uptake on FDG-PET/CT in up to 97% of Dacron® vascular graft recipients [21]. This increased uptake can be heterogeneous in up to 30% of patients and remain unchanged for at least 16 years. Therefore, a follow-up study to assess the value of baseline FDG-PET/CT after LVAD implantation would be of importance, since baseline scans could potentially help distinguish sterile inflammation from infectious processes, increasing the discriminative power of FDG-PET/CT.

Limitations
This study had a retrospective design which makes it prone to inclusion bias, though care was taken to include every consecutive patient with a suspicion of infection. Another limitation concerns the gold standard for the diagnosis: direct culture and molecular diagnostics using 16S PCR of the suspected device parts. This gold standard for the diagnosis is frequently unattainable because surgery is not always performed and other methods of retrieving the target tissues (e.g. CT-guided biopsy) are only very rarely used. While exit site swabs and blood cultures can give an indication for device infection, they are often insufficient to fully ascertain the diagnosis, especially considering infections of the central device components. To minimize the risk of incorporation bias, findings during surgery were considered as the true gold standard for the diagnosis whenever available, and when these were unavailable, all clinical findings, such as blood cultures, driveline exit site swabs, results from other available imaging modalities and outcomes during follow-up were considered together as a composite gold standard to avoid reliance on FDG-PET/CT alone.
The effect of (non-)adherence to the HFLC diet on FDG-PET/CT accuracy could not be statistically verified in our study, likely due to insufficient case numbers in this group as only 4