Exactitud diagnóstica de las diferentes modalidades de imagenología después de la exploración con tomografía computarizada (TC) para la evaluación de la resecabilidad con intención curativa del cáncer pancreático y periampular
Resumen
Antecedentes
El cáncer periampular incluye el cáncer de la cabeza y el cuello del páncreas, el cáncer del extremo distal del conducto biliar, el cáncer de la ampolla de Vater y el cáncer de la segunda parte del duodeno. La resección quirúrgica es el único tratamiento potencialmente curativo para el cáncer pancreático y periampular. Una proporción considerable de pacientes es sometida a una laparotomía innecesaria debido a la subestimación del grado del cáncer en la exploración con tomografía computada (TC). Otros métodos de imagenología que podrían evitar la laparotomía innecesaria como la imagenología por resonancia magnética (RM), la tomografía de emisión de positrones (TEP), la TEP‐TC y la ecografía endoscópica (EE) se han utilizado para detectar la invasión local o las metástasis distantes no visualizadas en la exploración con TC. Ninguna revisión sistemática o metanálisis ha examinado la función de diferentes modalidades de imagenología para evaluar la resecabilidad con intención curativa en pacientes con cáncer pancreático y periampular.
Objetivos
Determinar la exactitud diagnóstica de la RM, la TEP y la EE realizadas como una prueba complementaria o de la TEP‐TC como una prueba de reemplazo a la exploración con TC para detectar la resecabilidad curativa en el cáncer pancreático y periampular.
Métodos de búsqueda
Se hicieron búsquedas en las bases de datos MEDLINE, Embase, Science Citation Index Expanded y Health Technology Assessment (HTA) hasta el 5 noviembre 2015. Dos autores de la revisión, de forma independiente, examinaron las referencias y seleccionaron los estudios para la inclusión. También se buscaron artículos relacionados con los estudios incluidos al aplicar la función "búsqueda relacionada" en MEDLINE (OvidSP) y Embase (OvidSP) y una búsqueda de "referencias citadas" (al buscar los artículos que citan los artículos incluidos).
Criterios de selección
Se incluyeron los estudios de exactitud diagnóstica de la RM, la TEP, la TEP‐TC y la EE en pacientes con cáncer pancreático y periampular potencialmente resecable según la tomografía computarizada. Se aceptó cualquier criterio de resecabilidad utilizado en los estudios. Se incluyeron los estudios independientemente del idioma, estado de publicación o el diseño del estudio (prospectivo o retrospectivo). Se excluyeron los estudios de casos y controles.
Obtención y análisis de los datos
Dos autores de la revisión realizaron de forma independiente la extracción de los datos y la evaluación de la calidad mediante la herramienta QUADAS‐2 (quality assessment of diagnostic accuracy studies ‐ 2). Aunque se planificó utilizar métodos de dos variables para el análisis de las sensibilidades y las especificidades, solo fue posible utilizar los modelos de efectos fijos de una sola variable para la sensibilidad y la especificidad debido a la escasez de los datos. La probabilidad de no resecabilidad en los pacientes que tenían una prueba índice positiva (probabilidad posterior a la prueba de no resecabilidad en pacientes con un resultado positivo de la prueba) y en pacientes con una prueba índice negativa (probabilidad posterior a la prueba de no resecabilidad en pacientes con un resultado positivo de la prueba) se calculó mediante la probabilidad media de no resecabilidad (probabilidad previa a la prueba) a partir de los estudios incluidos y de los cocientes de verosimilitud positivos y negativos obtenidos del modelo. La diferencia entre las probabilidades previas y posteriores a las pruebas generó el valor general agregado de la prueba índice en comparación con la práctica estándar del estadiaje con la exploración con TC sola.
Resultados principales
Solo dos estudios (34 participantes) cumplieron los criterios de inclusión de esta revisión sistemática. Ambos estudios evaluaron la exactitud diagnóstica de la prueba de EE al evaluar la resecabilidad con intención curativa en los cánceres pancreáticos. En ambos estudios hubo escasas preocupaciones con respecto a la aplicabilidad de la mayoría de los dominios. El riesgo general de sesgo fue bajo en un estudio e incierto o alto en el segundo estudio. La probabilidad media de enfermedad no resecable después de la TC en los estudios fue del 60,5% (es decir, 61 de 100 pacientes con cáncer resecable después de la TC presentaron enfermedad no resecable en la laparotomía). La estimación general de la sensibilidad de la EE para la no resecabilidad fue 0,87 (intervalo de confianza [IC] del 95%: 0,54 a 0,97) y la estimación general de la especificidad para la no resecabilidad fue 0,80 (IC del 95%: 0,40 a 0,96). El cociente de verosimilitud positivo y el cociente de verosimilitud negativo fueron 4,3 (IC del 95%: 1,0 a 18,6) y 0,2 (IC del 95%: 0,0 a 0,8), respectivamente. Con una probabilidad media previa a la prueba del 60,5%, la probabilidad posterior a la prueba de enfermedad no resecable en los pacientes con una EE positiva (EE que indicaba no resecabilidad) fue del 86,9% (IC del 95%: 60,9% a 96,6%) y la probabilidad posterior a la prueba de enfermedad no resecable en los pacientes con una EE negativa (EE que indicaba resecabilidad) fue del 20,0% (5,1% a 53,7%). Lo anterior significa que el 13% de los pacientes (IC del 95%: 3% a 39%) con EE positiva presenta un cáncer potencialmente resecable y el 20% (5% a 53%) de los pacientes con EE negativa presenta un cáncer no resecable.
Conclusiones de los autores
Según dos estudios pequeños, hay incertidumbre significativa sobre la utilidad de la EE en los pacientes con cáncer pancreático en los que se encuentra que presentan enfermedad resecable en la tomografía computarizada. Ningún estudio ha evaluado la utilidad de la EE en pacientes con cáncer periampular.
No existe evidencia que indique que se deba realizar de forma sistemática en pacientes con cáncer pancreático o cáncer periampular en los que se encuentra que presentan enfermedad resecable en la tomografía computarizada.
Resumen en términos sencillos
Exactitud diagnóstica de diferentes exploraciones después de una tomografía computarizada para evaluar si el cáncer pancreático y periampular es resecable
Pregunta de la revisión
¿Cuán bien las diferentes exploraciones identifican si el cáncer pancreático y periampular es resecable (se puede eliminar quirúrgicamente) en pacientes con cáncer pancreático en los que la tomografía computarizada (TC) indica que el cáncer se puede eliminar?
La tomografía computarizada incluye una serie de rayos X que se combinan en una computadora para proporcionar imágenes detalladas del área corporal analizada con los rayos.
Antecedentes
El páncreas es un órgano situado en el abdomen cerca de la unión del estómago y el intestino delgado. Secreta jugos digestivos que son necesarios para la digestión de todos los materiales alimentarios. Los jugos digestivos secretados en el páncreas drenan a la parte superior del intestino delgado mediante el conducto pancreático. El conducto biliar es un tubo que drena la bilis del hígado y la vesícula biliar. Los conductos pancreáticos y biliares comparten una vía común justo antes de drenar en el intestino delgado. A esta área se le llama región periampular. La extracción quirúrgica es el único tratamiento potencialmente curativo de los cánceres que surgen en las regiones pancreáticas y periampular. Una proporción considerable de pacientes se somete a una operación abdominal mayor abierta exploratoria innecesaria (laparotomía) porque la tomografía computarizada ha subestimado la diseminación del cáncer. Si el cáncer se ha diseminado dentro del abdomen y se identifica durante la operación mayor abierta, el tratamiento principal es la quimioterapia, que no cura el cáncer pero puede mejorar la supervivencia. Por lo tanto, es posible evitar la operación abdominal mayor abierta con sus riesgos asociados si se conoce la diseminación del cáncer dentro del abdomen antes de la operación mayor. La determinación de la extensión del cáncer se llama "estadiaje" del cáncer. Generalmente la prueba mínima utilizada para el estadiaje es la tomografía computarizada. Sin embargo, la tomografía computarizada puede subestadiar el cáncer, es decir, puede subestimar la diseminación del cáncer. Además de la tomografía computarizada se pueden utilizar otras exploraciones para determinar si el cáncer pancreático es resecable (posible de eliminarse quirúrgicamente). Se incluyen las siguientes pruebas.
1. La resonancia magnética (RM) incluye el uso de un imán potente para producir imágenes de diferentes tejidos del cuerpo.
2. Tomografía de emisión de positrones (TEP): se utiliza una cantidad pequeña de glucosa (azúcar) radiactiva para diferenciar entre tejidos diferentes. Utiliza la propiedad de que las células cancerosas a menudo utilizan más glucosa que las células normales).
3. La ecografía endoscópica (EE) incluye el uso de un endoscopio, una cámara introducida en las cavidades corporales para visualizar el interior del cuerpo. Se utiliza una sonda ecográfica (ondas sonoras de alta energía) al final del endoscopio para diferenciar tejidos diferentes.
Además, se puede realizar una combinación de TEP‐TC en lugar de la TC.
Diferentes estudios informan exactitudes diferentes de estas pruebas para evaluar si es posible eliminar el cáncer. En esta revisión se identificaron todos los estudios y métodos matemáticos apropiados utilizados para identificar la exactitud diagnóstica promedio de estas pruebas para estadiar los cánceres pancreáticos y periampulares considerados resecables después de una tomografía computarizada.
Características de los estudios
En esta revisión, se incluyeron dos estudios con un total de 34 pacientes. Ambos estudios evaluaron la función diagnóstica de la EE. La evidencia está actualizadas hasta el 5 noviembre 2015.
Calidad de la evidencia
De los dos estudios, uno se realizó tan bien como fue posible. La calidad metodológica del otro estudio era deficiente.
Resultados clave
Los dos estudios incluidos mostraron que en los pacientes con cáncer pancreático en los que la TC sola mostró que era posible eliminar el cáncer completamente de forma quirúrgica, el 61% (61 de 100) probaría presentar un cáncer que se diseminó demasiado para posibilitar la eliminación cuando se intentó una laparotomía. Debido al tamaño de la muestra pequeño, hay incertidumbre significativa en la utilidad de la EE en los pacientes con cáncer pancreático en los que se encontró que presentaban enfermedad resecable en la tomografía computarizada. No existe evidencia para indicar que se deba realizar de forma sistemática en los pacientes con cáncer pancreático en los que se encontró que presentaban enfermedad resecable en la tomografía computarizada.
Authors' conclusions
Summary of findings
| Population | People with pancreatic cancer found to resectable on computed tomography (CT) scan | |
| Setting | Secondary or tertiary setting | |
| Index test | Endoscopic ultrasound (EUS) | |
| Reference standard | Laparotomy (surgeon's judgement of unresectability) | |
| Number of studies | 2 studies (38 participants) | |
| Summary sensitivity | 0.87 (95% confidence interval (CI) 0.54 to 0.97) | |
| Summary specificity | 0.80 (95% CI 0.40 to 0.96) | |
| Consistent results | Yes | |
| Overall risk of bias | Moderate to high | |
| Other limitations |
| |
| Pre‐test probability of unresectability from included studies | Post‐test probability of unresectability in people with positive EUS (EUS indicating unresectability) | Post‐test probability of unresectability in people with negative EUS (EUS indicating resectability) |
| Minimum = 53% | 83% (53% to 95%) | 16% (4% to 46%) |
| Mean = 61% | 87% (61% to 97%) | 20% (5% to 54%) |
| Maximum = 67% | 90% (67% to 97%) | 25% (7% to 61%) |
| Interpretation | There is significant uncertainty in the results because of inadequate data | |
| We reported all probabilities in the table as percentages. Abbreviations: CI: confidence interval; EUS: endoscopic ultrasound. | ||
Background
Please see Appendix 1 for a glossary of terms.
Periampullary cancer develops near the ampulla of Vater (National Cancer Institute 2014a). Periampullary cancer includes cancer of the head and neck of the pancreas, cancer of the distal end of the bile duct, cancer of the ampulla of Vater, and cancer of the second part of the duodenum. Pancreatic cancer (pancreatic cancer) is the tenth most common cancer in the USA, the fifth most common cause of cancer‐related mortality in the east and the fourth most common cause of cancer‐related mortality in the west (Parkin 2001; Parkin 2005; Yamamoto 1998). In 2012, 338,000 new patients were diagnosed with pancreatic cancer and there were 330,000 deaths due to pancreatic cancer globally (IARC 2014). There is global variation in the incidence of pancreatic cancers, with an age‐standardised annual incidence rate of 7.2 per 100,000 population in the more developed regions and an age‐standardised annual incidence rate of 2.8 per 100,000 population in the less developed regions (IARC 2014). A similar trend is noted in the age‐standardised annual mortality rates, of 6.8 per 100,000 population in the more developed regions and an age‐standardised annual mortality rate of 2.7 per 100,000 population in the less developed regions due to pancreatic cancer (IARC 2014).
Pancreaticoduodenectomy is the main treatment for cancers that arise in the head of the pancreas, ampulla, and second part of the duodenum. Surgical resection is generally considered the only treatment that can cure pancreatic cancer. However, only 15% to 20% of patients with pancreatic cancers undergo potentially curative resection (Conlon 1996; Engelken 2003; Michelassi 1989; Shahrudin 1997; Smith 2008). The overall five‐year survival after radical resection ranges from 7% to 25% (Cameron 1993; Livingston 1991; Niederhuber 1995; Nitecki 1995; Orr 2010; Trede 1990), with a median survival of 11 to 15 months (British Management Guideline 2005). With adjuvant chemotherapy, the median survival after radical resection varies between 14 and 24 months (Liao 2013). In all other patients, the cancers are not resected because of infiltration of local structures, disseminated disease, or because the patient is deemed unfit to undergo major surgery. Computed tomography (CT) scan is generally used for staging pancreatic and periampullary cancers (National Cancer Institute 2014b). Despite undergoing routine CT scanning to stage the disease, a substantial proportion of patients (approximately 40%) undergo unnecessary laparotomy (opening the abdomen using a large incision) with lack of curative resectability identified only during the laparotomy (Allen 2016). Staging laparoscopy or diagnostic laparoscopy may decrease the proportion of patients that undergo unnecessary laparotomy to approximately 17% (Allen 2016). Tests, such as magnetic resonance imaging (MRI), positron emission tomography (PET) scan, or endoscopic ultrasound (EUS), may be used in addition to CT scan to assess resectability with curative intent and decrease the proportion of patients who undergo unnecessary laparotomy.
Target condition being diagnosed
Inability to perform curative resectability of pancreatic and periampullary cancer ("unresectable" cancers)
Index test(s)
MRI
MRI involves the use of a powerful magnet to produce images of different tissues of the body. This is also called nuclear MRI (NMRI) (National Cancer Institute 2014c). Features, such as extent of the cancer in terms of involvement of adjacent structures and spread of cancer to distant areas (metastases), are taken into account to assess resectability with curative intent. The radiologist usually interprets the images.
PET
PET involves the use of a small amount of radioactive glucose (sugar) to differentiate between different tissues. It utilises the property that cancer cells often use more glucose than normal cells. It is also called PET scan (National Cancer Institute 2014d). This is a form of functional imaging. Cancerous lesions appear as areas of increased uptake. The presence of cancer in different locations and metastases are taken into account to assess resectability with curative intent. The radiologist usually interprets the images.
PET‐CT scan
PET scan can be combined with CT scan (PET‐CT scan), with both tests performed at the same time (National Cancer Institute 2014e). This allows superimposition of the two images by identifying corresponding points of the body in the two scans (coregistration) and allows the combination of the functional imaging (PET scan) with an anatomical imaging (CT scan), which may result in better diagnostic accuracy than either modality alone (National Cancer Institute 2014e). Usually, the radiologist interprets the images.
EUS
EUS involves the use of an endoscope, a camera introduced into the body cavities to view the inside of the body. An ultrasound (high‐energy sound waves) probe at the end of the endoscope is used to differentiate different tissues. This is also called endosonography and EUS (National Cancer Institute 2014f). Local extent and metastases are taken into account to assess resectability with curative intent. The endoscopist usually interprets the images.
Clinical pathway
There is no standard algorithm currently available to assess the resectability of pancreatic and periampullary cancers, with different clinicians following their own algorithms based on either their clinical experience or what they have been taught. Currently, almost all algorithms include a CT scan as one of the tests (National Cancer Institute 2014b). CT may be the only test performed before laparotomy. Other tests, such as diagnostic laparoscopy, PET (PET scan or PET‐CT scan), MRI, or EUS, may be used in addition to CT scan to assess resectability. We have presented the possible clinical pathway in the staging of pancreatic cancers in Figure 1.
Clinical pathway for the staging of pancreatic cancers.
Abbreviations: PET: positron emission tomography; MRI: magnetic resonance imaging; EUS: endoscopic ultrasound; CT: computed tomography.
Prior test(s)
The minimum prior test should be CT and the cancer should be resectable with curative intent on the basis of the CT scan. Other imaging modalities, such as MRI, PET scan, PET‐CT, or EUS, might be used in addition to CT scanning to assess resectability before performing the imaging modality being assessed.
Role of index test(s)
MRI, PET scan, and EUS can be considered as add‐on tests to the CT scan prior to laparotomy done with the intention of performing a potentially curative resection. PET‐CT scan can be considered as a replacement for CT scan prior to laparotomy done with the intention of performing a potentially curative resection. It can also be considered as an add‐on test to the CT scan prior to laparotomy. Although it appears strange to use PET‐CT scan as an add‐on test to CT scan, such an approach is possible if patients are referred to the referral centre with a CT scan. It should be noted that PET and CT scan should be performed simultaneously to allow coregistration. However, the problem with PET‐CT scan as a replacement for CT scan is that PET‐CT has to be performed without contrasts and hence PET‐CT alone may not provide as good an information as PET‐CT along with conventional CT scan.
Alternative test(s)
Diagnostic laparoscopy or laparoscopic ultrasound may be used as an alternative test to these imaging modalities in patients considered to have CT resectable pancreatic and periampullary cancer (Allen 2016; Hariharan 2010). Another Cochrane review has assessed the accuracy of diagnostic laparoscopy (Allen 2016).
Rationale
The different imaging modalities identify the extent of local spread, including invasion of adjacent blood vessels, and may identify distal metastases (MRI, PET scan, PET‐CT scan). If this add‐on test (or replacement test in the case of PET‐CT scan in patients who are referred without a CT scan) can identify unresectable cancers without laparotomy, it might decrease the costs and morbidity associated with unnecessary laparotomy. Currently there is no Cochrane review that has assessed the diagnostic accuracy of these imaging modalities in the assessment of the curative resectability of pancreatic and periampullary cancers.
Objectives
To determine the diagnostic accuracy of MRI, PET scan, and EUS performed as an add‐on test or PET‐CT as an add‐on or replacement test to CT scanning in detecting curative resectability in pancreatic and periampullary cancer.
Secondary objectives
We planned to explore the following sources of heterogeneity.
-
Studies at low risk of bias versus those at unclear or high risk of bias (as assessed by the QUADAS‐2 tool, recommended by the Cochrane Diagnostic Test Accuracy Group) (Whiting 2006; Whiting 2011).
-
Full text publications versus abstracts (this can give a clue about publication bias since there may be an association between the results of the study and the study reaching full publication status) (Eloubeidi 2001).
-
Prospective studies versus retrospective studies.
-
Proportion of patients with pancreatic cancer, ampullary cancer, and duodenal cancers (although classified as periampullary cancers they each have a different prognosis) (Klempnauer 1995). The additional value of the imaging modalities may be different because of the extent of spread in these different types of periampullary cancers.
-
Different definitions for resectable cancer on laparotomy. Different surgeons may consider cancer unresectable differently i.e. different surgeons would have different criteria for unresectability on laparotomy (other than the consensus criteria for resectability). For example, one surgeon may judge that the cancer is unresectable on laparotomy because of the involvement of the local vessels (mainly portal vein and superior mesenteric vein) and consider the reference standard to be positive. This would result in a false negative result for the imaging modality. Another surgeon may judge the same cancer to be resectable despite the involvement of the vessel and proceed with resection. The reference standard would be negative in this situation, which would result in a true negative result for the imaging modality. This might have an intrinsic threshold effect.
-
Additional pre‐tests performed (besides CT scan). This can alter the pre‐test probability of unresectability and can help in the assessment of the additional value of the imaging modality under various situations.
Methods
Criteria for considering studies for this review
Types of studies
We only included studies that provided diagnostic test accuracy data (true positive, false positive, false negative, and true negative) on the different imaging modalities mentioned above in the appropriate patient population (see below) irrespective of language, publication status, or whether data were collected prospectively or retrospectively. However, we excluded case reports which do not provide sufficient diagnostic test accuracy data. We also planned to exclude any identified case‐control studies because case‐control studies are prone to bias (Whiting 2011).
Participants
Adults considered for curative resection of pancreatic or periampullary cancer on the basis of CT findings, who were fit to undergo major surgery. We included patients in this review irrespective of whether they underwent other imaging modalities prior to imaging modality being assessed.
Index tests
MRI, PET scan, PET‐CT scan, or EUS.
Target conditions
The target conditions were unresectable pancreatic and periampullary cancers, that is, we considered the imaging modality a positive test if the pancreatic or periampullary cancer is unresectable with curative intent. In these cancers it is not possible to perform curative resectability. Clinically, it may not be easy to distinguish head of pancreas cancers, ampullary cancers, and cancer of the second part of the duodenum. The treatment for these different cancers is the same, i.e. pancreatoduodenectomy and the final confirmation as to the origin of these cancers may be done after resection without definitive diagnosis of the origin of the cancer, as long as the cancers are resectable. So we considered these cancers together. There are no uniform criteria for resectability of pancreatic and periampullary cancer. Consensus exists for the definition of borderline resectable cancers (Abrams 2009). Therefore, where there is less tissue involvement than in a borderline resectable cancer the tumour can be considered as resectable. We accepted any criteria of resectability used by the study authors and acknowledge that this could potentially create a threshold effect. In general, the cancer will not be resected if liver, peritoneal, or distal nodal metastases were noted, or if the cancer had invaded important adjacent blood vessels that are beyond the criteria for borderline resectable cancers (for example, greater than 180° involvement of the superior mesenteric artery) (Abrams 2009).
Reference standards
Confirmation of liver, peritoneal, or nodal metastatic involvement by histopathological examination of suspicious (liver, peritoneal, or nodal metastatic) lesions obtained at diagnostic laparoscopy or laparotomy. We accepted only paraffin section histology as the reference standard. In clinical practice, depending on the urgency of the results, a frozen section biopsy may be done to obtain immediate results. However, this is always confirmed by subsequent paraffin section histology (which can take several days) because frozen section biopsy is not as reliable as paraffin section histology. We also accepted the surgeon's judgement of unresectability at laparotomy when biopsy confirmation was not possible as an alternate reference standard. For example, if the tumour has invaded the adjacent blood vessels the surgeon may not resect the tumour because of the danger posed by resecting part of a large blood vessel, and so biopsy confirmation cannot be obtained. However, it should be noted that a surgeon's judgement of unresectability at laparotomy is a subjective decision and is a possible source of error in the reference standard. In the absence of an ethical and true gold standard, we accepted this as a reference standard.
Search methods for identification of studies
We included all studies irrespective of the language of publication and publication status. We translated any non‐English articles we found to assess eligibility.
Electronic searches
We searched the following databases on 5 November 2015.
-
MEDLINE (In‐Process & Non‐Indexed Citations) via OvidSP (January 1946 to 5 November 2015; Appendix 2).
-
Embase via OvidSP (January 1947 to 5 November 2015; Appendix 3).
-
Science Citation Index Expanded (including Conference Proceedings Citation Index ‐ Science) via Web of Knowledge (January 1980 to 5 November 2015; Appendix 4).
-
National Insitute for Health Research ‐ Health Technology Assessment (NIHR HTA) (November 2015) through the University of York Centre for Reviews and Dissemination (www.crd.york.ac.uk/CRDWeb/) (Appendix 5).
We included sensitivity maximising diagnostic filters for searching MEDLINE and Embase databases (Haynes 2004; Wilczynski 2005). This is because we retrieved more than 40,000 references when we used the original searches without the filters.
Searching other resources
We searched the references of the included studies to identify additional studies. We also searched for articles related to the included studies by performing the "related search" function in MEDLINE (OvidSP) and Embase (OvidSP) and a "citing reference" search (by searching the articles which cite the included articles) (Sampson 2008) in Science Citation Index Expanded, MEDLINE (OvidSP), and Embase (OvidSP).
Data collection and analysis
Selection of studies
Two review authors (DT and KSG) independently screened the results ofthe search strategy to identify relevant studies. We obtained the full‐text articles of references that at least one of the review authors considered relevant. Two review authors (DT and KSG or DR) independently screened the full‐text papers against the inclusion criteria. We did not have any differences in study selection based on our full‐text article assessments. If the eligibility of the report was unclear, we attempted to contact the study authors to seek clarification. Since we were unable to contact the study authors, we excluded the reports. We listed all excluded studies and their reasons for exclusion in the Characteristics of excluded studies table). Also, we constructed a PRISMA diagram to illustrate the study selection process.
Data extraction and management
Two review authors (DT and KSG) independently extracted the following data from each included study using a data extraction form that KSG designed and piloted. We resolved any differences by discussion.
-
First author.
-
Year of publication.
-
Study design (prospective or retrospective; cross‐sectional studies or randomised controlled trials).
-
Inclusion and exclusion criteria for individual studies.
-
Total number of patients.
-
Number of females.
-
Average age of the participants.
-
Type of cancer (i.e. head and neck of pancreas, body and tail of pancreas, ampullary cancers, duodenal cancer).
-
Criteria for unresectability at the index test and at laparotomy (reference standard).
-
Preoperative tests carried out prior to index test.
-
Description of the index test.
-
Reference standard.
-
Number of true positives, false positives, false negatives, and true negatives.
The unit of analysis was the patient, meaning that if multiple metastases or multiple infiltrations of adjacent structures were found in a patient with a negative index test, we planned to consider the number of false negatives to be one. This is because it is the presence, rather than the number of metastases or the number of infiltrations of adjacent structures, that is important in determining the curative resectability of patients. We planned to consider patients with uninterpretable index test results (no matter the reason given for lack of interpretation) as negative for the test since in clinical practice laparotomy would be carried out on these patients. However, we planned to include such patients in the analysis only if the results of laparotomy were available. We sought further information from the study authors if necessary.
If the same study reported multiple index tests, we planned to extract the number of true positives, false positives, false negatives, and true negatives for each index test. If there was an overlap of participants between multiple reports as suspected by common authors and centres, we planned to contact the study authors to seek clarification about the overlap. If we were unable to contact the authors, we planned to extract the maximum possible information from all the reports. However, we did not any find such reports.
Assessment of methodological quality
Two review authors (DT and KSG) independently assessed study quality using the QUADAS‐2 assessment tool (Whiting 2006; Whiting 2011). We resolved differences through discussion, based on the criteria published in the protocol (Gurusamy 2015). We have presented the criteria that we used to classify the different studies in Table 1. We considered studies which are classified as "low risk of bias" and "low concern" in all the domains as studies with high methodological quality. We planned to present the results in a "Risk of bias" summary and graphs, but because there were only two studies, we have presented the "Risk of bias" summary only.
| Domain 1: patient selection | Patient sampling | Patients with pancreatic and periampullary cancer considered eligible for surgical resection following a computed tomography (CT) scan. |
| Was a consecutive or random sample of patients enrolled? | Yes: if the study included a consecutive sample or a random sample of patients with pancreatic and periampullary cancer eligible for surgical resection after CT scan. | |
| Was a case‐control design avoided? | Yes: if the study assessed a cohort of patients about to undergo surgical resection. | |
| Did the study avoid inappropriate exclusions? | Yes: if the study included all patients with pancreatic and periampullary cancer eligible for surgical resection. | |
| Could the selection of patients have introduced bias? | Low risk of bias: if "yes" classification for all the above 3 questions. High risk of bias: if "no" classification for any of the above 3 questions. Unclear risk of bias: if "unclear" classification for any of the above 3 questions but without a "no" classification for any of the above 3 questions. | |
| Patient characteristics and setting | Yes: we included only patients with pancreatic and periampullary cancer who were considered eligible for surgical resection following a CT scan. So, we anticipated that we would classify all the included studies as "yes". | |
| Are there concerns that the included patients and setting do not match the review question? | Considering the inclusion criteria of this review, as anticipated, we classified all the included studies as "low concern". | |
| Domain 2: index test | Index test(s) | Magnetic resonance imaging (MRI), positron emission tomography (PET), PET‐CT, endoscopic ultrasound (EUS) |
| Were the index test results interpreted without knowledge of the results of the reference standard? | The index test would always be conducted though not interpreted before the reference standard. Yes: if the index test was conducted and interpreted without the knowledge of the results of the reference standard. | |
| If a threshold was used, was it prespecified? | Not applicable. | |
| Could the conduct or interpretation of the index test have introduced bias? | Low risk of bias: if "yes" classification for the only relevant question in this domain. High risk of bias: if "no" classification for the only relevant question in this domain. Unclear risk of bias: if "unclear" classification for the only relevant question in this domain. | |
| Are there concerns that the index test, its conduct, or interpretation differ from the review question? | Low concern: if the criteria for positive index test was clearly stated. High concern: if the criteria for positive index test was not stated. | |
| Domain 3: target condition and reference standard | Target condition and reference standard(s) | Unresectability. The reasons for unresectability include involvement of adjacent structures or distant metastases. There is currently no universal criteria for unresectability. Consensus exists for the definition of borderline resectable cancers (Abrams 2009). Therefore, where there is less tissue involvement than in a borderline resectable cancer the tumour can be considered as resectable. Positive reference standard: confirmation of liver or peritoneal involvement by histopathological examination of suspicious (liver or peritoneal) lesions (irrespective of how the tissues were obtained for histopathological examination). We accepted only paraffin section histology as the reference standard. We also accepted the surgeon's judgement of unresectability on laparotomy when biopsy confirmation was not possible (for example, the surgeon may not resect the tumour if it invaded the adjacent blood vessels but would not obtain a biopsy confirmation of this because of the danger posed by resecting a part of a large blood vessel). However, it should be noted that surgeon's judgement of unresectability at laparotomy is a subjective decision and is a possible source of error in the reference standard but in the absence of an ethical and true gold standard, we accepted this as a reference standard. Negative reference standard: cancer was fully resected i.e. clear resection margins on histology. |
| Is the reference standards likely to correctly classify the target condition? | Yes: if histological confirmation of distant spread or local infiltration of adjacent structures making the cancer unresectable was obtained. The report on the resection margins shows clearly that the cancer had been completely resected. As anticipated, none of the included studies met these criteria because of the danger that biopsy of infiltration of adjacent structures poses. | |
| Were the reference standard results interpreted without knowledge of the results of the index tests? | Yes: if the reference standard was interpreted without the knowledge of the results of the index test. However, the results of the index test are unlikely to influence the results of the reference standard. So, we did not take the answer to this question into account to determine the risk of bias. | |
| Could the reference standard, its conduct, or its interpretation have introduced bias? | We determined the risk of bias as "low" if the answer to the first question was "yes", "high" if the answer to the first question was "no", and "unclear" if the answer to the first question was "unclear". | |
| Are there concerns that the target condition as defined by the reference standard does not match the question? | Considering the inclusion criteria for this review, we classified all the included studies as "low concern", as anticipated. | |
| Domain 4: flow and timing | Flow and timing | The cancer may progress if there is a long time interval between index test and laparotomy. So, we chose an arbitrary time interval of 2 months as an acceptable time interval between index test and laparotomy. |
| Was there an appropriate interval between index test and reference standard? | Yes: if the time interval between index test and laparotomy was less than 2 months. | |
| Did all patients receive the same reference standard? | Yes: if all the patients received the same reference standard (we classified all the included studies as "yes", as anticipated). Unclear: if this information was unclear. | |
| Were all patients included in the analysis? | Yes: if the study included all the patients in the analysis irrespective of whether the results were uninterpretable. | |
| Could the patient flow have introduced bias? | Low risk of bias: if "yes" classification for all the above 3 questions. High risk of bias: if "no" classification for any of the above 3 questions. Unclear risk of bias: if "unclear" classification for any of the above 3 questions but without a "no" classification for any of the above 3 questions. |
Abbreviations: CT: computed tomography; EUS: endoscopic ultrasound; MRI: magnetic resonance imaging; PET: positron emission tomography.
Statistical analysis and data synthesis
We plotted study estimates of sensitivity and specificity on forest plots and in receiver operating characteristic (ROC) space to explore between study variation in the performance of each test. To estimate the summary sensitivity and specificity of each test, we planned to perform the meta‐analysis by fitting the bivariate model (Chu 2006; Reitsma 2005). This model accounts for between‐study variability in estimates of sensitivity and specificity through the inclusion of random effects for the logit sensitivity and logit specificity parameters of the bivariate model. If sparse data results in unreliable estimation of the covariance matrix of the random effects (as indicated by very large variance of logit sensitivity and specificity or if there was lack of convergence), we tried other alternate models including the random‐effects model, ignoring the inverse correlation between sensitivities and specificities in the different studies due to intrinsic threshold effect, and the fixed‐effect model for either sensitivity or specificity or both after visualising the forest plots and summary receiver operating characteristics (SROC) plots (Takwoingi 2015). We based our choice between the different models on the distribution of sensitivities and specificities as noted in the forest plots or ROC space. We also planned to use the model fit as indicated by the −2 log likelihood and planned to consider the model with the lower −2 log likelihood to be the better model.
We planned to compare the diagnostic accuracy of the tests by including covariate terms for test type (MRI, PET, PET‐CT, or EUS) in the bivariate model to estimate differences in the sensitivity and specificity of the tests. We planned to allow both the sensitivity and specificity to vary by covariate. In addition, we also planned to permit the variances of the random effects and their covariance to also depend on test type thus allowing the variances to differ between tests. We planned to use likelihood ratio tests to compare the model with and without covariate (test type). We planned to use a P value of less than 0.05 for the likelihood ratio test to indicate differences in the diagnostic accuracy between the tests. If studies that reported different tests in the same study population were available from at least four studies, we planned to perform a direct head‐to‐head comparison by limiting the test comparison to such studies. We planned to calculate the relative sensitivities and specificities for each pairwise comparison of tests.
We performed the meta‐analysis using the NLMixed command in SAS version 9.3 (SAS Institute Inc, Cary, North Carolina, USA). We created a graph of pre‐test probabilities (using the observed median and range of prevalence from the included studies) against post‐test probabilities. We calculated the post‐test probabilities using these pre‐test probabilities and the summary positive and negative likelihood ratios. We calculated the summary likelihood ratios and their confidence intervals (CIs) from the functions of the parameter estimates from the model that we fitted to estimate the summary sensitivities and specificities. Post‐test probability associated with positive test is the probability of having the target condition (unresectability) on the basis of a positive test result (unresectable disease) and is the same as the term "positive predictive value" used in a single diagnostic accuracy study. Post‐test probability associated with a negative test is the probability of having the target condition (unresectability) on the basis of a negative test result (resectable disease) and is 1 ‐ "negative predictive value". Negative predictive value is the term used in a single diagnostic accuracy study to indicate the chance that the patient has no target condition when the test is negative. We planned to report the summary sensitivity, specificity, positive and negative likelihood ratios, and post‐test probabilities for the median, lower quartile, and upper quartile of the pre‐test probabilities.
Investigations of heterogeneity
We planned to explore heterogeneity by using the different sources of heterogeneity as covariate(s) in the hierarchical summary receiver operating characteristics (HSROC) model. Of the six sources of heterogeneity we listed in the Secondary objectives section, we planned to deal with all items other than proportion of patients with pancreatic cancer, ampullary cancer, and duodenal cancer as categorical covariates. We planned to use the proportion of patients with pancreatic cancer, ampullary cancer, and duodenal cancer as continuous covariates in the regression model. We planned to employ likelihood ratio tests to compare the model with and without covariate. We planned to use a P value of less than 0.05 for the likelihood ratio test to indicate that the covariate was a potential source of heterogeneity.
Sensitivity analyses
We did not plan to perform any sensitivity analyses except when the data available from the studies was ambiguous (for example, the numbers in the text differed from the numbers in the figures), in which case we planned to assess the impact of different data used by a sensitivity analysis.
Assessment of reporting bias
We planned to investigate whether the summary sensitivity and specificity differed between studies published as full texts and those available only as abstracts using the methods we described in the Investigations of heterogeneity section.
Results
Results of the search
We identified a total of 23,346 references through electronic searches of MEDLINE (OvidSP; N = 9763), Embase (OvidSP; N = 8097), Science Citation Index expanded (Web of Knowledge; includes Conference Proceedings Citation Index‐ Science; N = 5412), and HTA (Centre for Reviews and Dissemination; (N = 74). After we removed duplicate references, there were 14,590 articles remaining. We excluded 14,384 clearly irrelevant references through reading abstracts. We retrieved the full‐text publication of 206 references for further detailed assessment. We excluded 204 references for the reasons in the Characteristics of excluded studies section. Two diagnostic accuracy studies (two references) fulfilled the inclusion criteria (see the Characteristics of included studies section). We have presented a study flow diagram in Figure 2.
Study flow diagram.
Included studies
Two studies with small sample sizes met the inclusion criteria. One study was a prospective study (Ahmad 2001), while the other was a retrospective study (Ardengh 2003). These two studies included a total of 38 participants with pancreatic cancer . Ardengh 2003 included 17 participants and Ahmad 2001 included 21 participants. The mean age of the participants in the two trials was 61 years and 64 years respectively (Ahmad 2001; Ardengh 2003). The proportion of females in the two trials was 23.8% and 64.7% respectively (Ahmad 2001; Ardengh 2003). The prevalence of unresectability (pre‐test probability) was 0.529 in Ardengh 2003 and 0.667 in Ahmad 2001.
The tests that participants underwent prior to endoscopic ultrasound (EUS) were cross‐sectional imaging (CT scan in all patients and ultrasound in some patients depending upon the referral centre) in Ahmad 2001, and CT scan and ultrasound in Ardengh 2003 (on people undergoing pancreatic resection after an ultrasound and a CT scan). Both studies evaluated endoscopic ultrasound (EUS) as the index test. The reference standard was surgeon's judgement of unresectability in both studies. In Ahmad 2001, this was vascular invasion during laparotomy, while Ardengh 2003 did not report the criteria that the surgeon used for assessing unresectability during laparotomy.
We have provided the methodological quality of the included studies in the Methodological quality of included studies section.
Excluded studies
We excluded a total of 204 references for the following reasons.
-
Seventeen studies were not primary studies (Barthet 2007; Brugge 1995; Faigel 1996; Fockens 1993; Freeny 2001; García‐Cano 2002; Gaspar 2015; Goh 2006; Lévy 2001; Malfertheiner 2005; Neoptolemos 2005; Pappas 2011; Rösch 1992c; Shin 2013; Snady 1993; Wang 2007a; Wiersema 2000).
-
We excluded 111 studies because participants were not patients with CT resectable pancreatic cancer (Abe 2010; Ahmad 1999; Ahmad 2000a; Ahmad 2000c; Ahmad 2000d; Akahoshi 1998; Anand 2013; Aubertin 1996; Awad 1997; Baarir 1998; Bao 2008; Bettini 2005; Broglia 2001; Burge 2015; Carroll 1999; Catalano 1997; Catalano 1998; Chandler 1999; Chhibber 2006; Chiang 2014; Cieslak 2014; Crippa 2013; Crippa 2014; DeWitt 2004; Egorov 2012; Einersen 2013; Farma 2008a; Fischer 2002; Frohlich 1999; Grenacher 2004; Lopez‐Hänninen 2002; Hochwald 1999; Howard 1997; Hu 2015; Ichikawa 1997; Iglesias‐Garcia 2010; Imazu 2010; Izuishi 2010; Javery 2013; Jemaa 2008; Kala 2007; Karoumpalis 2011; Kim 2001; Kim 2012; Koelblinger 2011; Koranda 2009; Koranda 2010; Kulig 2004; Kysucan 2010; Latronico 2005; Lee 2002; Lee 2010; Lentschig 1996; Makowiec 2000; Maluf‐Filho 2004; Mansfield 2008; McFarland 1996; Megibow 1995; Melzer 1996; Mertz 2000; Motosugi 2011; Mukai 1991; Murakami 1996; Nakamoto 1999; Napolitano 2002; Nishiharu 1999; Palazzo 1993; Park 2009; Patel 2002a; Patel 2002b; Paul 2012; Ramsay 2004; Razzaque 2012; Reiser‐Erkan 2009; Reiser‐Erkan 2010; Ren 2006; Ridtitid 2015; Rivadeneira 2003; Romijn 2000; Rösch 1992a; Rösch 1992b; Schmidt 2004; Schwarz 2001; Seicean 2008; Shami 2011; Sheng 2012; Smedby 1997; Solodinina 2014b; Soriano 2001; Soriano 2004; Strobel 2008; Tapper 2010; Tian 2008a; Tian 2008b; Tian 2008c; Tierney 2001; Tio 1986; Tio 1988; Tio 1990; Tomić 2005; Trede 1997; Turowska 2009; Valinas 2002; Wakabayashi 2008; Wang 2007b; Wang 2015; Warshaw 1990; Woerlein 2002; Younes 1999; Yusoff 2003; Zhong 2005).
-
Thirteen studies had no separate data on patients with pancreatic cancer (Arabul 2012; Buchs 2007; Casneuf 2007; Cieslak 2012; Cieslak 2013; Dewitt 2003; Ho 2008; Lu 2006; Lytras 2005; Pan 2014; Schima 2002; Takaori 2007; Tomazic 2000).
-
Two studies were not conducted in patients undergoing pancreatic resections (Agarwal 2005; Xu 2014).
-
In 61 studies diagnostic accuracy data on unresectability was unavailable (Ahmad 2000b; Arslan 2001; Artifon 2009; Asagi 2013; Aslanian 2005; Baghbanian 2014; Brugge 1996; Buscall 1999; Cahn 1996; Chang 1997; Chen 2001a; Chen 2001b; Chen 2009; Chiang 2012; Costilla 2011; Croome 2010; Czako 2009; Delbeke 1999; Egorov 2013; Einersen 2014; Eloubeidi 2006; Eloubeidi 2007; Erickson 2000; Farma 2008b; Gress 1997; Gress 1999; Harrison 1999; Heinrich 2005; Helmreich 2004; Hemmingsson 1982; Hirokawa 2010; Holzapfel 2011; Kadish 1995; Kim 2015; Lakhtakia 2011; Mehmet 2006; Morris‐Stiff 2011; Prithiviraj 2013; Raj 2013; Rösch 2000; Saif 2008; Shoup 2000; Sironi 1995; Sironi 1996; Skordilis 2002; Snady 1994; Solodinina 2014a; Spencer 1998; Staib 1997; Takayama 2009; Tellez‐Avila 2012; Tio 1996; Wang 2008; Wang 2014; Wee 2012; Yao 2012; Yasuda 1988; Yasuda 1993; Yoneyama 2014; Zhang 2012; Zhang 2015).
Methodological quality of included studies
We have summarised the risk of bias and applicability concerns in the included studies in Figure 3. As shown in Figure 3, there were no applicability concerns in the included studies. However, the risk of bias in the "patient selection" was high in Ardengh 2003 since it excluded pancreatic cancers that were 3 cm or more in diameter. The risk of bias in this domain was low in Ahmad 2001. The risk of bias in the "index test" domain was unclear in Ardengh 2003 since it was unclear whether the index test results were interpreted without knowledge of the results of the reference standard. The risk of bias in this domain was low in Ahmad 2001. As anticipated, both studies used surgeons' judgement on unresectability as the reference standard and so both studies were at unclear risk of bias in the "reference standard" domain. Ardengh 2003 did not report the interval between EUS and surgery and the participant flow. We considered this study to be at unclear risk of bias in the "flow and timing" domain. The risk of bias in this domain was low in Ahmad 2001.
Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study.
Findings
There was no heterogeneity in sensitivity as shown by very good overlap of confidence intervals (CIs) in the forest plots, visualisation of the receiver operating characteristic (ROC) plot, and by the values of sensitivity which were almost identical (0.86 in Ahmad 2001 versus 0.89 in Ardengh 2003) (Figure 4; Figure 5). Although we planned to evaluate the use of univariate random‐effects model for specificity based on the forest plots (there was good overlap of CIs but the difference in point estimate was more with specificity than sensitivity: 0.71 in Ahmad 2001 versus 0.88 in Ardengh 2003) and ROC plot, the only model that converged was univariate fixed‐effect model for both sensitivity and specificity. So, we were unable to choose the best model by comparing the −2 log likelihood.
Pooled sensitivity and specificity of endoscopic ultrasound for assessing the resectability with curative intent in pancreatic and periampullary cancer.
Abbreviations: df: degrees of freedom.
Summary ROC Plot of endoscopic ultrasound for assessing the resectability of pancreatic and periampullary cancer.
The summary estimate of sensitivity for unresectability was 0.87 (95% CI 0.54 to 0.97) and the summary estimate of specificity for unresectability was 0.80 (95% CI 0.40 to 0.96). The positive likelihood ratio and negative likelihood ratio were 4.3 (95% CI 1.0 to 18.6) and 0.2 (95% CI 0.0 to 0.8) respectively. Although we planned to calculate the post‐test probabilities using the median and quartiles of the pre‐test probabilities, we calculated the post‐test probabilities using the mean and range of the pre‐test probabilities because of the inclusion of two studies only. The mean pre‐test probability was 60.5%. At this pre‐test probability, the post‐test probability of unresectable disease for people with a positive EUS (EUS indicating unresectability) was 86.9% (95% CI 60.9% to 96.6%) and the post‐test probability of unresectable disease for people with a negative EUS (EUS indicating resectability) was 20.0% (5.1% to 53.7%). This means that 13% of people (95% CI 3% to 39%) with positive EUS have potentially resectable cancer and 20% (5% to 53%) of people with negative EUS have unresectable cancer. The "Summary of findings" table shows the post‐test probability of unresectable disease at different pre‐test probabilities of unresectable disease (summary of findings Table).
Neither of the included studies reported any complications related to EUS. We did not perform any investigation of heterogeneity because only two studies met the inclusion criteria of this review.
Discussion
Summary of main results
Only two studies (38 participants) that evaluated the diagnostic accuracy of EUS in people with CT‐resectable pancreatic cancers met the inclusion criteria of this review. The summary estimate of sensitivity was 0.87 (95% CI 0.54 to 0.97) and the summary estimate of specificity was 0.81 (95% CI 0.40 to 0.96). The positive likelihood ratio and negative likelihood ratio were 4.3 (95% CI 1.0 to 18.6) and 0.2 (95% CI 0.0 to 0.8) respectively. At the mean pre‐test probability in included studies (60.5%), the post‐test probability of unresectable disease for people with a positive EUS (EUS indicating unresectability) was 86.9% (95% CI 60.9% to 96.6%) and the post‐test probability of unresectable disease for people with a negative EUS (EUS indicating resectability) was 20.0% (5.1% to 53.7%).
Direct laparotomy after CT resulted in approximately 60% of cancers being unresectable, which appears to be higher than the usual unresectability rates after CT scan of around 30% to 40% (Allen 2016). We are unable to identify why the pre‐test probability of unresectability was high in these centres which are specialist centres, considering that they have facilities to perform EUS. When the EUS indicates that the pancreatic cancer is not resectable although CT scan shows that pancreatic cancer is resectable (EUS positive in CT resectable pancreatic cancer), approximately 13% of people (95% CI 3% to 39%) had resectable pancreatic cancer. Since pancreatic resection is the only potentially curative option for pancreatic cancer, omission of laparotomy and resection in these people can have a major negative impact on their survival.
We were unable to assess the diagnostic accuracy of MRI, PET, PET‐CT and compare their diagnostic accuracy with EUS since none of the studies on MRI or PET were on CT resectable pancreatic cancers and none of the studies on PET‐CT indicated the added value of PET clearly.
Strengths and weaknesses of the review
We used formal search strategies and reported this, so that it is possible to independently verify our results. Two review authors independently identified studies and extracted data, thereby minimising human error in the selection of studies and data extraction. We reached agreement based on the information available in the protocol of this review (Gurusamy 2015). The methodological quality in one included study was as good as can be achieved ethically (Ahmad 2001), and the methodological quality in the second included study was mostly unclear (Ardengh 2003). There were no concerns about applicability in either study. There was no heterogeneity in the diagnostic test accuracy between the studies as indicated by the almost identical sensitivities and good overlap of CIs for specificities. These are the major strengths of this review.
The major limitation of this review was the paucity of data: only two studies met the inclusion criteria and both these studies were on EUS. We had to use univariate fixed‐effect models for both sensitivity and specificity since this was the only model that converged. Such models may give reliable and stable results if used in the appropriate situation (Takwoingi 2015). Although we would have liked to compare the model fit of the univariate fixed‐effect models that we performed with the model fit of univariate random‐effects model for at least specificity, this was not possible because convergence was obtained only for univariate fixed‐effect models for both sensitivity and specificity. However, our decision is vindicated to a certain extent by the almost identical sensitivity and good overlap of CIs for specificity and the I² statistic values of 0% for both sensitivity and specificity. The alternative was to present the results of studies individually, which would have negated the advantage of meta‐analysis, i.e. improved precision, particularly when there was no heterogeneity in the results between the two studies.
Another limitation of this review is that we included sensitivity maximising diagnostic filters for searching MEDLINE and Embase databases (Haynes 2004; Wilczynski 2005). This is because the original searches without the filters retrieved more than 40,000 references. We had to balance the possibility of missing some studies against the risk of being unable to complete the review. We decided that it would be more useful to have evidence from major studies rather than having no information at all. Notably, the diagnostic filters we used have a sensitivity of 98.6% for MEDLINE and 100% for Embase. So, the chances of us missing some relevant diagnostic studies are extremely low. We reduced this further by performing a "related search" and "citing reference search" in which we did not find any studies that we could include in this review.
This is the first systematic review on the topic. EUS is not routinely performed to assess resectability of pancreatic cancers in most centres and the findings from our review would suggest that there is insufficient evidence of clinical benefit to justify its inclusion in the standard diagnostic algorithm.
Applicability of findings to the review question
The findings of this review are applicable only to people with pancreatic cancer who were found to be resectable after a CT scan. In addition, all the participants included in this review underwent laparotomy; so the findings of this review are applicable only in those who are fit to withstand major surgery. This review assessed the diagnostic accuracy of EUS in assessing the resectability of pancreatic cancer and does not provide the diagnostic accuracy of EUS in diagnosis of pancreatic cancer or finding the tumour, node, and metastasis (TNM) staging of pancreatic cancer.
Clinical pathway for the staging of pancreatic cancers.
Abbreviations: PET: positron emission tomography; MRI: magnetic resonance imaging; EUS: endoscopic ultrasound; CT: computed tomography.
Study flow diagram.
Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study.
Pooled sensitivity and specificity of endoscopic ultrasound for assessing the resectability with curative intent in pancreatic and periampullary cancer.
Abbreviations: df: degrees of freedom.
Summary ROC Plot of endoscopic ultrasound for assessing the resectability of pancreatic and periampullary cancer.
| Population | People with pancreatic cancer found to resectable on computed tomography (CT) scan | |
| Setting | Secondary or tertiary setting | |
| Index test | Endoscopic ultrasound (EUS) | |
| Reference standard | Laparotomy (surgeon's judgement of unresectability) | |
| Number of studies | 2 studies (38 participants) | |
| Summary sensitivity | 0.87 (95% confidence interval (CI) 0.54 to 0.97) | |
| Summary specificity | 0.80 (95% CI 0.40 to 0.96) | |
| Consistent results | Yes | |
| Overall risk of bias | Moderate to high | |
| Other limitations |
| |
| Pre‐test probability of unresectability from included studies | Post‐test probability of unresectability in people with positive EUS (EUS indicating unresectability) | Post‐test probability of unresectability in people with negative EUS (EUS indicating resectability) |
| Minimum = 53% | 83% (53% to 95%) | 16% (4% to 46%) |
| Mean = 61% | 87% (61% to 97%) | 20% (5% to 54%) |
| Maximum = 67% | 90% (67% to 97%) | 25% (7% to 61%) |
| Interpretation | There is significant uncertainty in the results because of inadequate data | |
| We reported all probabilities in the table as percentages. Abbreviations: CI: confidence interval; EUS: endoscopic ultrasound. | ||
| Domain 1: patient selection | Patient sampling | Patients with pancreatic and periampullary cancer considered eligible for surgical resection following a computed tomography (CT) scan. |
| Was a consecutive or random sample of patients enrolled? | Yes: if the study included a consecutive sample or a random sample of patients with pancreatic and periampullary cancer eligible for surgical resection after CT scan. | |
| Was a case‐control design avoided? | Yes: if the study assessed a cohort of patients about to undergo surgical resection. | |
| Did the study avoid inappropriate exclusions? | Yes: if the study included all patients with pancreatic and periampullary cancer eligible for surgical resection. | |
| Could the selection of patients have introduced bias? | Low risk of bias: if "yes" classification for all the above 3 questions. High risk of bias: if "no" classification for any of the above 3 questions. Unclear risk of bias: if "unclear" classification for any of the above 3 questions but without a "no" classification for any of the above 3 questions. | |
| Patient characteristics and setting | Yes: we included only patients with pancreatic and periampullary cancer who were considered eligible for surgical resection following a CT scan. So, we anticipated that we would classify all the included studies as "yes". | |
| Are there concerns that the included patients and setting do not match the review question? | Considering the inclusion criteria of this review, as anticipated, we classified all the included studies as "low concern". | |
| Domain 2: index test | Index test(s) | Magnetic resonance imaging (MRI), positron emission tomography (PET), PET‐CT, endoscopic ultrasound (EUS) |
| Were the index test results interpreted without knowledge of the results of the reference standard? | The index test would always be conducted though not interpreted before the reference standard. Yes: if the index test was conducted and interpreted without the knowledge of the results of the reference standard. | |
| If a threshold was used, was it prespecified? | Not applicable. | |
| Could the conduct or interpretation of the index test have introduced bias? | Low risk of bias: if "yes" classification for the only relevant question in this domain. High risk of bias: if "no" classification for the only relevant question in this domain. Unclear risk of bias: if "unclear" classification for the only relevant question in this domain. | |
| Are there concerns that the index test, its conduct, or interpretation differ from the review question? | Low concern: if the criteria for positive index test was clearly stated. High concern: if the criteria for positive index test was not stated. | |
| Domain 3: target condition and reference standard | Target condition and reference standard(s) | Unresectability. The reasons for unresectability include involvement of adjacent structures or distant metastases. There is currently no universal criteria for unresectability. Consensus exists for the definition of borderline resectable cancers (Abrams 2009). Therefore, where there is less tissue involvement than in a borderline resectable cancer the tumour can be considered as resectable. Positive reference standard: confirmation of liver or peritoneal involvement by histopathological examination of suspicious (liver or peritoneal) lesions (irrespective of how the tissues were obtained for histopathological examination). We accepted only paraffin section histology as the reference standard. We also accepted the surgeon's judgement of unresectability on laparotomy when biopsy confirmation was not possible (for example, the surgeon may not resect the tumour if it invaded the adjacent blood vessels but would not obtain a biopsy confirmation of this because of the danger posed by resecting a part of a large blood vessel). However, it should be noted that surgeon's judgement of unresectability at laparotomy is a subjective decision and is a possible source of error in the reference standard but in the absence of an ethical and true gold standard, we accepted this as a reference standard. Negative reference standard: cancer was fully resected i.e. clear resection margins on histology. |
| Is the reference standards likely to correctly classify the target condition? | Yes: if histological confirmation of distant spread or local infiltration of adjacent structures making the cancer unresectable was obtained. The report on the resection margins shows clearly that the cancer had been completely resected. As anticipated, none of the included studies met these criteria because of the danger that biopsy of infiltration of adjacent structures poses. | |
| Were the reference standard results interpreted without knowledge of the results of the index tests? | Yes: if the reference standard was interpreted without the knowledge of the results of the index test. However, the results of the index test are unlikely to influence the results of the reference standard. So, we did not take the answer to this question into account to determine the risk of bias. | |
| Could the reference standard, its conduct, or its interpretation have introduced bias? | We determined the risk of bias as "low" if the answer to the first question was "yes", "high" if the answer to the first question was "no", and "unclear" if the answer to the first question was "unclear". | |
| Are there concerns that the target condition as defined by the reference standard does not match the question? | Considering the inclusion criteria for this review, we classified all the included studies as "low concern", as anticipated. | |
| Domain 4: flow and timing | Flow and timing | The cancer may progress if there is a long time interval between index test and laparotomy. So, we chose an arbitrary time interval of 2 months as an acceptable time interval between index test and laparotomy. |
| Was there an appropriate interval between index test and reference standard? | Yes: if the time interval between index test and laparotomy was less than 2 months. | |
| Did all patients receive the same reference standard? | Yes: if all the patients received the same reference standard (we classified all the included studies as "yes", as anticipated). Unclear: if this information was unclear. | |
| Were all patients included in the analysis? | Yes: if the study included all the patients in the analysis irrespective of whether the results were uninterpretable. | |
| Could the patient flow have introduced bias? | Low risk of bias: if "yes" classification for all the above 3 questions. High risk of bias: if "no" classification for any of the above 3 questions. Unclear risk of bias: if "unclear" classification for any of the above 3 questions but without a "no" classification for any of the above 3 questions. | |
| Abbreviations: CT: computed tomography; EUS: endoscopic ultrasound; MRI: magnetic resonance imaging; PET: positron emission tomography. | ||
| Test | No. of studies | No. of participants |
| 1 EUS Show forest plot | 2 | 38 |
