Accepted Manuscript Usefulness of F-18-fluorodeoxyglucose Positron Emission Tomography to confirm suspected pancreatic cancer: A Meta-Analysis Anton P. Rijkers, MSc Roelf Valkema, MD, PhD Hugo J. Duivenvoorden, PhD Casper H.J. van Eijck, MD, PhD PII:
S0748-7983(14)00378-3
DOI:
10.1016/j.ejso.2014.03.016
Reference:
YEJSO 3797
To appear in:
European Journal of Surgical Oncology
Received Date: 19 December 2013 Revised Date:
18 March 2014
Accepted Date: 19 March 2014
Please cite this article as: Rijkers AP, Valkema R, Duivenvoorden HJ, van Eijck CHJ, Usefulness of F-18-fluorodeoxyglucose Positron Emission Tomography to confirm suspected pancreatic cancer: A Meta-Analysis, European Journal of Surgical Oncology (2014), doi: 10.1016/j.ejso.2014.03.016. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Title Usefulness of F-18-fluorodeoxyglucose Positron Emission Tomography to confirm suspected
RI PT
pancreatic cancer: A Meta-Analysis
Authors
Anton P. Rijkersa, MSc; Roelf Valkemab, MD, PhD; Hugo J. Duivenvoorden, PhDa; Casper
a
SC
H.J. van Eijcka, MD, PhD
Department of Surgery, Erasmus MC, University Medical Center, Rotterdam, The
b
M AN U
Netherlands
Department of Nuclear Medicine, Erasmus MC, University Medical Center, Rotterdam, The
Netherlands
TE D
Correspondence Prof. Casper H.J. van Eijck, PhD Department of Surgery
EP
Erasmus MC Rotterdam ‘s Gravendijkwal 230
AC C
3015 CE Rotterdam The Netherlands
Email: [email protected]
Type of publication Meta-Analysis, Systematic review
1
ACCEPTED MANUSCRIPT
Abstract
Introduction
RI PT
Pancreatic cancer is among the five most lethal malignancies in the world. Unfortunately,
many malignant tumors go undetected by the current primary diagnostic tools. 18FDG-PET
SC
and 18FDG-PET/CT might be useful to confirm suspected pancreatic cancer.
Methods
M AN U
A meta-analysis was performed using all major search engines. Methodological quality of included studies was assessed as well as quality of the PET-protocol. The following pooled estimates served as primary outcome measures: sensitivity, specificity, positive predictive
Results
TE D
value (PPV), negative predictive value (NPV) and accuracy.
Thirty-five studies were included. Pooled estimates for 18FDG-PET were: sensitivity 90%, specificity 76%, PPV 90%, NPV 76% and accuracy 86%. Pooled estimates for 18FDG-
EP
PET/CT were: sensitivity 90%, specificity 76%, PPV 89%, NPV 78% and accuracy 86%. The pooled sensitivity and specificity for 18FDG-PET to differentiate between pancreatic cancer
AC C
and chronic pancreatitis were 90% and 84%, respectively.
Conclusion
Both 18FDG-PET and 18FDG-PET/CT offer no benefit over the current primary diagnostic tools in diagnosing pancreatic cancer. However, the 18FDG-PET/CT systems are still improving. We should investigate the sensitivity and specificity of these new systems while reevaluating the tradeoff between false positive and false negative results. Yet, 18FDG-
2
ACCEPTED MANUSCRIPT
PET/CT may have a role in the staging of pancreatic cancer, in survival prediction, and may add to other diagnostic information, like histology.
RI PT
Keywords
Diagnostic Imaging; Meta-analysis; Pancreatic Neoplasms; Positron-Emission Tomography;
SC
Review.
M AN U
Introduction
Pancreatic cancer is among the five most lethal malignancies in the world1 and survival has not increased substantially in the past 30 years2. Its overall 5-year survival rate is 6%2, increasing to 10% - 25% after intentionally curative resection 3-8. Unfortunately, at the time of
TE D
diagnosis over 80% of all tumors are unresectable9. Therefore, an early diagnosis of pancreatic cancer is crucial.
EP
Overlooking a malignant tumor is a great concern to both physician and patient, as it may withhold early and live-extending treatment. The ideal diagnostic tool should be able to
AC C
identify every malignant tumor, but keep the number of false positive results to a minimum. Surgeons are dealing with this dilemma of false negative and false positive results on a daily base. However, it is often beneficial to perform surgery in patients with suspicious pancreatic head lesions, despite the risk of performing an unnecessary pancreaticoduodenectomy10. Besides, the impairment of quality of life after pancreaticoduodenectomy is not per definition irreversible11. From a clinical and social point of view, lowering the risk of missing a malignant pancreatic tumor is more valuable than the risk of overtreatment. Against this
3
ACCEPTED MANUSCRIPT
background, clinicians would be willing to accept 95% sensitivity and 80% specificity. thus allowing for four times more false positive results than false negative results.
RI PT
Helical computed tomography (CT) and magnetic resonance imaging (MRI) are still the
primary diagnostic tools, with sensitivity 91% and 84%, respectively, and specificity 85% and 82%, respectively12. This means that with the use of these imaging techniques 10% to 20% of
SC
all malignant tumors may be missed (false negatives). Therefore, it would be desirable to have a diagnostic tool with higher sensitivity and specificity, thereby, most important that would
M AN U
enable to identify a higher percentage of malignant pancreatic tumors.
One candidate is positron emission tomography (PET) with F-18-fluorodeoxyglucose (18FDG), which has been proven to be of clinical value in several gastrointestinal
TE D
malignancies13, 14.It is based on the principle that photons resulting from the annihilation of the positrons emitted by specific radioisotopes (e.g. 18Fluorine) are detected by the PETscanner and subsequently computed to a tomogram.. A more recent tool, the 18FDG-PET/CT, combines in a single gantry both 18FDG-PET and CT and produces one combined image15,
EP
see Supplement 1. Currently, 18FDG-PET/CT is quickly replacing stand-alone PET in clinical
AC C
practice. Various studies have examined the ability of 18FDG-PET to diagnose pancreatic cancer. The reported sensitivity and specificity of
18
FDG-PET is up to 95% and 100%,
respectively16. Furthermore, multiple studies showed that the sensitivity and specificity of 18
FDG-PET is superior to CT alone17-20. The 18FDG-PET/CT is stated to be more sensitive
than conventional imaging techniques regarding the detection of pancreatic cancer21. Nonetheless, the clinical usefulness of both 18FDG-PET and 18FDG-PET /CT in patients with suspected pancreatic cancer is yet to be determined.
4
ACCEPTED MANUSCRIPT
Main objective of this study is to review the relevant literature and to determine from these sources the usefulness of 18FDG-PET and 18FDG-PET/CT to confirm suspected pancreatic cancer. The focus is on identifying malignant tumors and limiting the number of false
RI PT
negative results to a minimum. Secondary objective is to evaluate the extent to which 18FDG-
AC C
EP
TE D
M AN U
SC
PET(/CT) can differentiate between pancreatic cancer and chronic pancreatitis.
5
ACCEPTED MANUSCRIPT
Methods
Relevant literature was searched in Medline via OvidSP 350, PubMed not MEDLINE,
RI PT
Embase and the Cochrane Central Register of Controlled Trials. The search period was from
January first 1990 to April 10th 2013. Main keywords for the search-queries were: pancreatic neoplasm or cancer, pancreas, tumor and fluorodeoxyglucose. Studies, written in English,
SC
which fulfilled our search query and investigated the diagnostic accuracy of 18FDG-PET(/CT) in patients with primary pancreatic cancer were eligible for inclusion. Besides we had to be
M AN U
able to calculate the sensitivity and specificity from these studies. We excluded primarily results and pilot studies. The full search strategy is given in Figure 1. All included studies were screened for cross references.
Data extraction and management
TE D
Two authors (APR, CHJE) independently analyzed results and characteristics of each included study. They were not blinded for information about the journal, author, institution and date of publication, as it has been reported that blinding does not affect the result of meta-
EP
analysis22. Discrepancies were resolved by consensus. Data from each included study were
AC C
extracted using a standardized form.
Quality assessment
Methodological quality of the included studies was assessed with the QUADAS tool23, 24, which was designed to be used in systematic reviews of diagnostic accuracy studies23. In addition, quality of the PET-protocol was assessed with the EANM procedure guidelines for tumour PET imaging15 .
6
ACCEPTED MANUSCRIPT
Statistical analysis The following pooled estimates served as primary outcome measures: sensitivity, specificity,
calculated using a bivariate mixed effect analysis.
RI PT
positive predictive value (PPV), negative predictive value (NPV) and accuracy. These were
Heterogeneity among studies was assessed by means of a meta-regression analysis based on
SC
the methodological quality assessment. In addition, the influence of the covariates year of publication and impact factor on the heterogeneity among studies was analyzed, also by
M AN U
means of a meta-regression analysis. A sensitivity analysis was performed to gain insight into the effect of removing possibly influential studies on the results. First, the pooled outcome variables were calculated using all included studies. Secondly, the pooled outcome variables were calculated after excluding the studies indicated as highly influential by the Cook’s distance. Highly influential was defined as a Cook’s distance of >1/n. The presence of
TE D
publication bias was visually assessed in a funnel plot. A symmetric funnel shape is characteristic for the absence of publication bias. A test for inconsistency was performed to check whether discrepancies between study results were due to fundamental differences
EP
(heterogeneity) or to chance alone (basically homogeneity). Inconsistency was assessed statistically by calculating the I2-statistic An I2-statistic of 25%, 50% and 75% was considered
AC C
as low, moderate and high inconsistency, respectively25.
Statistical significance was set at P0.05). In only one study pancreatic cancer was suspected based on solely clinical findings47 and not
RI PT
on other imaging modalities. Hence, analogue to 18FDG-PET, in the remaining studies the prior chance of pancreatic cancer was possibly increased by other imaging techniques.
However, the pooled accuracy did not differ significantly between the two studies in which
SC
suspicion was based on solely clinical findings and the other studies (P=0.60).
M AN U
None of the covariates was able to explain the heterogeneity among the studies, see Table 2. The funnel plot was slightly asymmetrical, suggesting publication bias, see Supplement 6. The I2-statistic was 32%. Therefore, small to moderate inconsistency was considered present.
18
FDG-PET
TE D
Differentiation between pancreatic cancer and chronic pancreatitis
A total of eight studies evaluated the ability of 18FDG-PET to differentiate between pancreatic
EP
cancer and chronic pancreatitis19, 20, 51-56. The total study population was 454 and the prevalence of pancreatic cancer was 55%. The pooled estimates were: Sensitivity 90% (95%
AC C
CI: 83% to 93%). Specificity 84% (95% CI: 75% to 91%). PPV 89% (95% CI: 82% to 93%). NPV 85% (95% CI: 67% to 93%). Accuracy 87% (84% to 90%). See Figure 3a and Table 1. Accuracy of 18FDG-PET to differentiate between pancreatic cancer and chronic pancreatitis did not differ from the ability to diagnose pancreatic cancer in patients with suspected pancreatic cancer (P=0.60), also after controlling for year of publication, methodological quality, quality of the PET-protocol or impact factor. However, in most other studies on the ability of 18FDG-PET to diagnose pancreatic cancer in patients with suspected pancreatic
11
ACCEPTED MANUSCRIPT
cancer, the most prevalent disease among controls was chronic pancreatitis. This is a possible explanation for the absence of a difference. None of the covariates was able to explain the heterogeneity in the ability of 18FDG-PET to
RI PT
differentiate between pancreatic cancer and chronic pancreatitis, see Table 2. The funnel plot clearly showed asymmetry. See Supplement 5. The I2-statistic was 64.0%. Hence, moderate to
18
SC
high inconsistency was present.
FDG-PET/CT
Only one study evaluated the usefulness of 18FDG-PET/CT to differentiate between
M AN U
pancreatic cancer and chronic pancreatitis52. For the sake of completeness we present the pooled estimates: Sensitivity 96% (95% CI: 80% to 100%). Specificity 17% (95% CI: 0% to 64%). PPV 83% (95% CI: 69% to 97%). NPV 50% (0% to 100%). Accuracy 81% (95% CI: 67% to 95%). The authors themselves of this study did not report the specificity as the
EP
Discussion
TE D
number of histopathological benign findings was too small52.
This meta-analysis revealed that both 18FDG-PET and 18FDG-PET/CT have a pooled
AC C
sensitivity of around 90% and a pooled specificity of around 80% to diagnose pancreatic cancer in patients with suspected pancreatic cancer. As mentioned in the introduction, missing a malignant tumor is a great concern for both physician and patient. Unfortunately, 10% of all patients with a malignant tumor are not classified as such by 18FDG-PET. We also suggested that we are willing to allow four times more false positive results than false negative results. This is not in agreement with the results of our meta-analysis. We think there is still a margin to lower the false negative results by allowing more false positive results.
12
ACCEPTED MANUSCRIPT
The performance of 18FDG-PET to differentiate between pancreatic cancer and chronic pancreatitis did not exceed its diagnostic performance in patients with suspected pancreatic
RI PT
cancer.
Regarding the question how 18FDG-PET and 18FDG-PET/CT compare to CT and MRI we
refer to a meta-analysis published in 2005, which reported a pooled sensitivity and specificity
SC
of 91% and of 85%, respectively, for helical CT and 84% and 82%, respectively for MRI12.
18
M AN U
These values differ little from those we found for 18FDG-PET and 18FDG-PET/CT. However, FDG-PET does not provide anatomical information, which information is crucial for staging
and of utmost importance for determining resectability.
No improvement in diagnostic performance over time
TE D
The diagnostic performance of both 18FDG-PET and 18FDG-PET/CT has not improved over the years, despite ongoing technical innovations. Regarding 18FDG-PET the diagnostic performance did even decrease significantly over the years. A funnel plot suggested moderate
EP
publication bias. Still, neither the methodological quality nor the quality of the PET-protocol could explain this lack of improvement. Regarding
18
FDG-PET/CT none of the covariates
AC C
could explain the absence of improvement in diagnostic performance over the years. Yet, the presence of modest publication bias was reflected in the funnel plot.
Considering all this and also taking into view that the overall diagnostic accuracy between 18
FDG-PET and 18FDG-PET/CT did not differ significantly, it would seem that not the
technique itself, but rather the characteristics of pancreatic cancer and local inflammation are
13
ACCEPTED MANUSCRIPT
the limiting factors for diagnosing pancreatic cancer in patients with suspected pancreatic cancer.
RI PT
Physiology 18FDG uptake and 18FDG-PET image analysis The most frequently used radiotracer is 18FDG. This glucose analogue is imported into the cell via glucose transporters (GLUT) and is than phosphorylated by hexokinase to FDG-6-
SC
phosphate. Unlike glucose, 18FDG is not further metabolized and accumulates inside the cell. The increased accumulation in cancer cells is related to the increased expression of GLUT-1 molecules, the increased activity of hexokinase and reduced levels of gluco-6-phosphatase14,
M AN U
57
. However, infections, inflammatory processes and granulation tissue can result in an
increased 18FDG uptake15. The analysis of the images produced by 18FDG-PET can be performed visual and semiquantitative. For the semiquantitative analysis the standardized uptake value/ratio (SUV or SUR) is commonly used, which is a measure for glucose
TE D
metabolism. The equation of SUV is the activity concentration in the tissue divided by the ratio of injected dose and body size. The SUV is often presented as the mean SUV (SUVmean) or maximal SUV (SUV-max) of the region of interest. The latter is most often used.
EP
The measurement of SUV can be affected by the patients’ blood glucose level15. This should
AC C
be taken into account when performing a 18FDG-PET study15
Methods and Limitations The broad search strategy in all major search engines was aimed not only at minimizing selection bias, but also on analyzing the ability of 18FDG-PET to: (i) Detect metastasis. (ii) Determine resectability. (iii) Determine survival. (iv) Evaluate treatment response. Unfortunately, the studies were too heterogeneous in study design and outcome measures to properly perform meta-analyses on these outcomes. A broad search strategy is
14
ACCEPTED MANUSCRIPT
disadvantageous in that many studies have to be evaluated for eligibility. Cross references were screened to minimize the risk of wrongly excluding studies. Language bias may have been present as only studies in the English language were included. By not excluding earlier
RI PT
published studies we were able to evaluate the influence of year of publication on the pooled diagnostic performance. Data-extraction and management was performed using a pre-set
standardized form designed on the basis of expertise and comparable meta-analyses. Quality
SC
assessment was performed with the QUADAS tool23, 24. The requirements per item were
determined by consensus between the authors after screening the studies. Quality of the PET-
M AN U
protocol was assessed with the EANM procedure guidelines for tumour PET imaging15, thus not with a validated scoring tool. This approach still provided an impression of the quality of the PET-protocol. We chose not to exclude studies based on the quality assessment, but to evaluate the influence of study quality on the pooled diagnostic performance. In this manner we prevented selection bias. Unfortunately, the majority of all studies reported a moderate
TE D
description of their method section and information on variables like tumor stage and size were not reported unambiguously. For this reason we did not perform a meta-regression analysis based on individual parameters, but only on the total scores on both the QUADAS
EP
tool and the quality of the PET-protocol. Furthermore, the available method sections did not enable us to differentiate between the value of 18FDG-PET as a primary diagnostic tool and
AC C
its additional value after a negative CT or MRI scan. The diagnosis pancreatic cancer can for example be hard to make when a double duct sign on CT is absent. The incremental value of 18
FDG-PET is still under debate. Standardization in conducting diagnostic performance
research is highly recommended. Furthermore, multicenter projects could provide better generalization. Important problems and proposed solutions regarding the use of standardized uptake values (SUV) in multicenter trials are addressed by Westerterp et al58.
15
ACCEPTED MANUSCRIPT
The pooled outcome measures were determined with a bivariate mixed effect analysis, which is one of the recommended methods for a diagnostic performance study. It has the advantage of incorporating a possible correlation between sensitivity and specificity, and also accounts
RI PT
for randomly distributed unspecified differences between studies59-63.
In this meta-analysis we did not review other diagnostic modalities like endoscopic
SC
ultrasound (EUS), fine needle aspiration (FNA) and endoscopic retrograde
cholangiopancreatography (ERCP) with biopsy or brush cytology. These modalities are
M AN U
beneficial in case of an uncertain diagnosis after CT64. A comparison with these modalities was not within the scope of our meta-analysis. Still, they must be taken into account when determining the overall value of 18FDG-PET(/CT) in the diagnostic management of pancreatic cancer.
18
TE D
Additional utilities
FDG-PET and 18FDG-PET/CT are also used to detect metastatic disease in patients with
pancreatic cancer but studies and subsequent data are limited. Accuracy to diagnose
EP
metastasis varied per metastatic site. For bone metastasis it ranged from 24%65 to 100%66. Additionally, all studies which compared the detection rate of bone metastasis between FDG-PET (/CT) and CT reported that 18FDG-PET (/CT) was superior over CT65-70. The
AC C
18
detection rate for liver metastasis widely ranged from 0%67 to 90%71. It is harder to detect smaller than larger liver metastases. For example, one study reported that the detection rate of 18
FDG-PET dropped from 88% to 50% when comparing tumors of >1 cm with tumors of
S0748-7983(14)00378-3
DOI:
10.1016/j.ejso.2014.03.016
Reference:
YEJSO 3797
To appear in:
European Journal of Surgical Oncology
Received Date: 19 December 2013 Revised Date:
18 March 2014
Accepted Date: 19 March 2014
Please cite this article as: Rijkers AP, Valkema R, Duivenvoorden HJ, van Eijck CHJ, Usefulness of F-18-fluorodeoxyglucose Positron Emission Tomography to confirm suspected pancreatic cancer: A Meta-Analysis, European Journal of Surgical Oncology (2014), doi: 10.1016/j.ejso.2014.03.016. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Title Usefulness of F-18-fluorodeoxyglucose Positron Emission Tomography to confirm suspected
RI PT
pancreatic cancer: A Meta-Analysis
Authors
Anton P. Rijkersa, MSc; Roelf Valkemab, MD, PhD; Hugo J. Duivenvoorden, PhDa; Casper
a
SC
H.J. van Eijcka, MD, PhD
Department of Surgery, Erasmus MC, University Medical Center, Rotterdam, The
b
M AN U
Netherlands
Department of Nuclear Medicine, Erasmus MC, University Medical Center, Rotterdam, The
Netherlands
TE D
Correspondence Prof. Casper H.J. van Eijck, PhD Department of Surgery
EP
Erasmus MC Rotterdam ‘s Gravendijkwal 230
AC C
3015 CE Rotterdam The Netherlands
Email: [email protected]
Type of publication Meta-Analysis, Systematic review
1
ACCEPTED MANUSCRIPT
Abstract
Introduction
RI PT
Pancreatic cancer is among the five most lethal malignancies in the world. Unfortunately,
many malignant tumors go undetected by the current primary diagnostic tools. 18FDG-PET
SC
and 18FDG-PET/CT might be useful to confirm suspected pancreatic cancer.
Methods
M AN U
A meta-analysis was performed using all major search engines. Methodological quality of included studies was assessed as well as quality of the PET-protocol. The following pooled estimates served as primary outcome measures: sensitivity, specificity, positive predictive
Results
TE D
value (PPV), negative predictive value (NPV) and accuracy.
Thirty-five studies were included. Pooled estimates for 18FDG-PET were: sensitivity 90%, specificity 76%, PPV 90%, NPV 76% and accuracy 86%. Pooled estimates for 18FDG-
EP
PET/CT were: sensitivity 90%, specificity 76%, PPV 89%, NPV 78% and accuracy 86%. The pooled sensitivity and specificity for 18FDG-PET to differentiate between pancreatic cancer
AC C
and chronic pancreatitis were 90% and 84%, respectively.
Conclusion
Both 18FDG-PET and 18FDG-PET/CT offer no benefit over the current primary diagnostic tools in diagnosing pancreatic cancer. However, the 18FDG-PET/CT systems are still improving. We should investigate the sensitivity and specificity of these new systems while reevaluating the tradeoff between false positive and false negative results. Yet, 18FDG-
2
ACCEPTED MANUSCRIPT
PET/CT may have a role in the staging of pancreatic cancer, in survival prediction, and may add to other diagnostic information, like histology.
RI PT
Keywords
Diagnostic Imaging; Meta-analysis; Pancreatic Neoplasms; Positron-Emission Tomography;
SC
Review.
M AN U
Introduction
Pancreatic cancer is among the five most lethal malignancies in the world1 and survival has not increased substantially in the past 30 years2. Its overall 5-year survival rate is 6%2, increasing to 10% - 25% after intentionally curative resection 3-8. Unfortunately, at the time of
TE D
diagnosis over 80% of all tumors are unresectable9. Therefore, an early diagnosis of pancreatic cancer is crucial.
EP
Overlooking a malignant tumor is a great concern to both physician and patient, as it may withhold early and live-extending treatment. The ideal diagnostic tool should be able to
AC C
identify every malignant tumor, but keep the number of false positive results to a minimum. Surgeons are dealing with this dilemma of false negative and false positive results on a daily base. However, it is often beneficial to perform surgery in patients with suspicious pancreatic head lesions, despite the risk of performing an unnecessary pancreaticoduodenectomy10. Besides, the impairment of quality of life after pancreaticoduodenectomy is not per definition irreversible11. From a clinical and social point of view, lowering the risk of missing a malignant pancreatic tumor is more valuable than the risk of overtreatment. Against this
3
ACCEPTED MANUSCRIPT
background, clinicians would be willing to accept 95% sensitivity and 80% specificity. thus allowing for four times more false positive results than false negative results.
RI PT
Helical computed tomography (CT) and magnetic resonance imaging (MRI) are still the
primary diagnostic tools, with sensitivity 91% and 84%, respectively, and specificity 85% and 82%, respectively12. This means that with the use of these imaging techniques 10% to 20% of
SC
all malignant tumors may be missed (false negatives). Therefore, it would be desirable to have a diagnostic tool with higher sensitivity and specificity, thereby, most important that would
M AN U
enable to identify a higher percentage of malignant pancreatic tumors.
One candidate is positron emission tomography (PET) with F-18-fluorodeoxyglucose (18FDG), which has been proven to be of clinical value in several gastrointestinal
TE D
malignancies13, 14.It is based on the principle that photons resulting from the annihilation of the positrons emitted by specific radioisotopes (e.g. 18Fluorine) are detected by the PETscanner and subsequently computed to a tomogram.. A more recent tool, the 18FDG-PET/CT, combines in a single gantry both 18FDG-PET and CT and produces one combined image15,
EP
see Supplement 1. Currently, 18FDG-PET/CT is quickly replacing stand-alone PET in clinical
AC C
practice. Various studies have examined the ability of 18FDG-PET to diagnose pancreatic cancer. The reported sensitivity and specificity of
18
FDG-PET is up to 95% and 100%,
respectively16. Furthermore, multiple studies showed that the sensitivity and specificity of 18
FDG-PET is superior to CT alone17-20. The 18FDG-PET/CT is stated to be more sensitive
than conventional imaging techniques regarding the detection of pancreatic cancer21. Nonetheless, the clinical usefulness of both 18FDG-PET and 18FDG-PET /CT in patients with suspected pancreatic cancer is yet to be determined.
4
ACCEPTED MANUSCRIPT
Main objective of this study is to review the relevant literature and to determine from these sources the usefulness of 18FDG-PET and 18FDG-PET/CT to confirm suspected pancreatic cancer. The focus is on identifying malignant tumors and limiting the number of false
RI PT
negative results to a minimum. Secondary objective is to evaluate the extent to which 18FDG-
AC C
EP
TE D
M AN U
SC
PET(/CT) can differentiate between pancreatic cancer and chronic pancreatitis.
5
ACCEPTED MANUSCRIPT
Methods
Relevant literature was searched in Medline via OvidSP 350, PubMed not MEDLINE,
RI PT
Embase and the Cochrane Central Register of Controlled Trials. The search period was from
January first 1990 to April 10th 2013. Main keywords for the search-queries were: pancreatic neoplasm or cancer, pancreas, tumor and fluorodeoxyglucose. Studies, written in English,
SC
which fulfilled our search query and investigated the diagnostic accuracy of 18FDG-PET(/CT) in patients with primary pancreatic cancer were eligible for inclusion. Besides we had to be
M AN U
able to calculate the sensitivity and specificity from these studies. We excluded primarily results and pilot studies. The full search strategy is given in Figure 1. All included studies were screened for cross references.
Data extraction and management
TE D
Two authors (APR, CHJE) independently analyzed results and characteristics of each included study. They were not blinded for information about the journal, author, institution and date of publication, as it has been reported that blinding does not affect the result of meta-
EP
analysis22. Discrepancies were resolved by consensus. Data from each included study were
AC C
extracted using a standardized form.
Quality assessment
Methodological quality of the included studies was assessed with the QUADAS tool23, 24, which was designed to be used in systematic reviews of diagnostic accuracy studies23. In addition, quality of the PET-protocol was assessed with the EANM procedure guidelines for tumour PET imaging15 .
6
ACCEPTED MANUSCRIPT
Statistical analysis The following pooled estimates served as primary outcome measures: sensitivity, specificity,
calculated using a bivariate mixed effect analysis.
RI PT
positive predictive value (PPV), negative predictive value (NPV) and accuracy. These were
Heterogeneity among studies was assessed by means of a meta-regression analysis based on
SC
the methodological quality assessment. In addition, the influence of the covariates year of publication and impact factor on the heterogeneity among studies was analyzed, also by
M AN U
means of a meta-regression analysis. A sensitivity analysis was performed to gain insight into the effect of removing possibly influential studies on the results. First, the pooled outcome variables were calculated using all included studies. Secondly, the pooled outcome variables were calculated after excluding the studies indicated as highly influential by the Cook’s distance. Highly influential was defined as a Cook’s distance of >1/n. The presence of
TE D
publication bias was visually assessed in a funnel plot. A symmetric funnel shape is characteristic for the absence of publication bias. A test for inconsistency was performed to check whether discrepancies between study results were due to fundamental differences
EP
(heterogeneity) or to chance alone (basically homogeneity). Inconsistency was assessed statistically by calculating the I2-statistic An I2-statistic of 25%, 50% and 75% was considered
AC C
as low, moderate and high inconsistency, respectively25.
Statistical significance was set at P0.05). In only one study pancreatic cancer was suspected based on solely clinical findings47 and not
RI PT
on other imaging modalities. Hence, analogue to 18FDG-PET, in the remaining studies the prior chance of pancreatic cancer was possibly increased by other imaging techniques.
However, the pooled accuracy did not differ significantly between the two studies in which
SC
suspicion was based on solely clinical findings and the other studies (P=0.60).
M AN U
None of the covariates was able to explain the heterogeneity among the studies, see Table 2. The funnel plot was slightly asymmetrical, suggesting publication bias, see Supplement 6. The I2-statistic was 32%. Therefore, small to moderate inconsistency was considered present.
18
FDG-PET
TE D
Differentiation between pancreatic cancer and chronic pancreatitis
A total of eight studies evaluated the ability of 18FDG-PET to differentiate between pancreatic
EP
cancer and chronic pancreatitis19, 20, 51-56. The total study population was 454 and the prevalence of pancreatic cancer was 55%. The pooled estimates were: Sensitivity 90% (95%
AC C
CI: 83% to 93%). Specificity 84% (95% CI: 75% to 91%). PPV 89% (95% CI: 82% to 93%). NPV 85% (95% CI: 67% to 93%). Accuracy 87% (84% to 90%). See Figure 3a and Table 1. Accuracy of 18FDG-PET to differentiate between pancreatic cancer and chronic pancreatitis did not differ from the ability to diagnose pancreatic cancer in patients with suspected pancreatic cancer (P=0.60), also after controlling for year of publication, methodological quality, quality of the PET-protocol or impact factor. However, in most other studies on the ability of 18FDG-PET to diagnose pancreatic cancer in patients with suspected pancreatic
11
ACCEPTED MANUSCRIPT
cancer, the most prevalent disease among controls was chronic pancreatitis. This is a possible explanation for the absence of a difference. None of the covariates was able to explain the heterogeneity in the ability of 18FDG-PET to
RI PT
differentiate between pancreatic cancer and chronic pancreatitis, see Table 2. The funnel plot clearly showed asymmetry. See Supplement 5. The I2-statistic was 64.0%. Hence, moderate to
18
SC
high inconsistency was present.
FDG-PET/CT
Only one study evaluated the usefulness of 18FDG-PET/CT to differentiate between
M AN U
pancreatic cancer and chronic pancreatitis52. For the sake of completeness we present the pooled estimates: Sensitivity 96% (95% CI: 80% to 100%). Specificity 17% (95% CI: 0% to 64%). PPV 83% (95% CI: 69% to 97%). NPV 50% (0% to 100%). Accuracy 81% (95% CI: 67% to 95%). The authors themselves of this study did not report the specificity as the
EP
Discussion
TE D
number of histopathological benign findings was too small52.
This meta-analysis revealed that both 18FDG-PET and 18FDG-PET/CT have a pooled
AC C
sensitivity of around 90% and a pooled specificity of around 80% to diagnose pancreatic cancer in patients with suspected pancreatic cancer. As mentioned in the introduction, missing a malignant tumor is a great concern for both physician and patient. Unfortunately, 10% of all patients with a malignant tumor are not classified as such by 18FDG-PET. We also suggested that we are willing to allow four times more false positive results than false negative results. This is not in agreement with the results of our meta-analysis. We think there is still a margin to lower the false negative results by allowing more false positive results.
12
ACCEPTED MANUSCRIPT
The performance of 18FDG-PET to differentiate between pancreatic cancer and chronic pancreatitis did not exceed its diagnostic performance in patients with suspected pancreatic
RI PT
cancer.
Regarding the question how 18FDG-PET and 18FDG-PET/CT compare to CT and MRI we
refer to a meta-analysis published in 2005, which reported a pooled sensitivity and specificity
SC
of 91% and of 85%, respectively, for helical CT and 84% and 82%, respectively for MRI12.
18
M AN U
These values differ little from those we found for 18FDG-PET and 18FDG-PET/CT. However, FDG-PET does not provide anatomical information, which information is crucial for staging
and of utmost importance for determining resectability.
No improvement in diagnostic performance over time
TE D
The diagnostic performance of both 18FDG-PET and 18FDG-PET/CT has not improved over the years, despite ongoing technical innovations. Regarding 18FDG-PET the diagnostic performance did even decrease significantly over the years. A funnel plot suggested moderate
EP
publication bias. Still, neither the methodological quality nor the quality of the PET-protocol could explain this lack of improvement. Regarding
18
FDG-PET/CT none of the covariates
AC C
could explain the absence of improvement in diagnostic performance over the years. Yet, the presence of modest publication bias was reflected in the funnel plot.
Considering all this and also taking into view that the overall diagnostic accuracy between 18
FDG-PET and 18FDG-PET/CT did not differ significantly, it would seem that not the
technique itself, but rather the characteristics of pancreatic cancer and local inflammation are
13
ACCEPTED MANUSCRIPT
the limiting factors for diagnosing pancreatic cancer in patients with suspected pancreatic cancer.
RI PT
Physiology 18FDG uptake and 18FDG-PET image analysis The most frequently used radiotracer is 18FDG. This glucose analogue is imported into the cell via glucose transporters (GLUT) and is than phosphorylated by hexokinase to FDG-6-
SC
phosphate. Unlike glucose, 18FDG is not further metabolized and accumulates inside the cell. The increased accumulation in cancer cells is related to the increased expression of GLUT-1 molecules, the increased activity of hexokinase and reduced levels of gluco-6-phosphatase14,
M AN U
57
. However, infections, inflammatory processes and granulation tissue can result in an
increased 18FDG uptake15. The analysis of the images produced by 18FDG-PET can be performed visual and semiquantitative. For the semiquantitative analysis the standardized uptake value/ratio (SUV or SUR) is commonly used, which is a measure for glucose
TE D
metabolism. The equation of SUV is the activity concentration in the tissue divided by the ratio of injected dose and body size. The SUV is often presented as the mean SUV (SUVmean) or maximal SUV (SUV-max) of the region of interest. The latter is most often used.
EP
The measurement of SUV can be affected by the patients’ blood glucose level15. This should
AC C
be taken into account when performing a 18FDG-PET study15
Methods and Limitations The broad search strategy in all major search engines was aimed not only at minimizing selection bias, but also on analyzing the ability of 18FDG-PET to: (i) Detect metastasis. (ii) Determine resectability. (iii) Determine survival. (iv) Evaluate treatment response. Unfortunately, the studies were too heterogeneous in study design and outcome measures to properly perform meta-analyses on these outcomes. A broad search strategy is
14
ACCEPTED MANUSCRIPT
disadvantageous in that many studies have to be evaluated for eligibility. Cross references were screened to minimize the risk of wrongly excluding studies. Language bias may have been present as only studies in the English language were included. By not excluding earlier
RI PT
published studies we were able to evaluate the influence of year of publication on the pooled diagnostic performance. Data-extraction and management was performed using a pre-set
standardized form designed on the basis of expertise and comparable meta-analyses. Quality
SC
assessment was performed with the QUADAS tool23, 24. The requirements per item were
determined by consensus between the authors after screening the studies. Quality of the PET-
M AN U
protocol was assessed with the EANM procedure guidelines for tumour PET imaging15, thus not with a validated scoring tool. This approach still provided an impression of the quality of the PET-protocol. We chose not to exclude studies based on the quality assessment, but to evaluate the influence of study quality on the pooled diagnostic performance. In this manner we prevented selection bias. Unfortunately, the majority of all studies reported a moderate
TE D
description of their method section and information on variables like tumor stage and size were not reported unambiguously. For this reason we did not perform a meta-regression analysis based on individual parameters, but only on the total scores on both the QUADAS
EP
tool and the quality of the PET-protocol. Furthermore, the available method sections did not enable us to differentiate between the value of 18FDG-PET as a primary diagnostic tool and
AC C
its additional value after a negative CT or MRI scan. The diagnosis pancreatic cancer can for example be hard to make when a double duct sign on CT is absent. The incremental value of 18
FDG-PET is still under debate. Standardization in conducting diagnostic performance
research is highly recommended. Furthermore, multicenter projects could provide better generalization. Important problems and proposed solutions regarding the use of standardized uptake values (SUV) in multicenter trials are addressed by Westerterp et al58.
15
ACCEPTED MANUSCRIPT
The pooled outcome measures were determined with a bivariate mixed effect analysis, which is one of the recommended methods for a diagnostic performance study. It has the advantage of incorporating a possible correlation between sensitivity and specificity, and also accounts
RI PT
for randomly distributed unspecified differences between studies59-63.
In this meta-analysis we did not review other diagnostic modalities like endoscopic
SC
ultrasound (EUS), fine needle aspiration (FNA) and endoscopic retrograde
cholangiopancreatography (ERCP) with biopsy or brush cytology. These modalities are
M AN U
beneficial in case of an uncertain diagnosis after CT64. A comparison with these modalities was not within the scope of our meta-analysis. Still, they must be taken into account when determining the overall value of 18FDG-PET(/CT) in the diagnostic management of pancreatic cancer.
18
TE D
Additional utilities
FDG-PET and 18FDG-PET/CT are also used to detect metastatic disease in patients with
pancreatic cancer but studies and subsequent data are limited. Accuracy to diagnose
EP
metastasis varied per metastatic site. For bone metastasis it ranged from 24%65 to 100%66. Additionally, all studies which compared the detection rate of bone metastasis between FDG-PET (/CT) and CT reported that 18FDG-PET (/CT) was superior over CT65-70. The
AC C
18
detection rate for liver metastasis widely ranged from 0%67 to 90%71. It is harder to detect smaller than larger liver metastases. For example, one study reported that the detection rate of 18
FDG-PET dropped from 88% to 50% when comparing tumors of >1 cm with tumors of
