The review concluded that positron emission tomography/computed tomography was the most accurate whole-body imaging modality, closely followed by positron emission tomography, for detection of recurrent disease in colorectal cancer patients with suspected recurrence in follow-up. Limitations in the analyses and potential for error and bias in the review process suggest that this conclusion should be interpreted with some caution.

To compare the diagnostic performance of whole-body imaging modalities for detection of local and/or distant recurrent disease in colorectal cancer patients with suspected recurrent disease: positron emission tomography (PET), PET/computed tomography (CT), CT and magnetic resonance imaging (MRI).

PubMed and EMBASE were searched up to May 2010 without language restriction; search terms were reported. Reference lists were checked for additional relevant studies.

Eligible studies were focused on detection of metastasis and/or recurrence in patients with suspected recurrence during follow-up for colorectal cancer. Studies needed to include more than 20 patients with colorectal cancer and report a 2x2 contingency table (or include sufficient details to construct one). The reference standard was specified as combined histology and follow-up. Case reports, studies that only evaluated local staging or detection of liver metastases, studies that evaluated whole-body imaging at primary staging or for patients with known hepatic metastases and studies that only evaluated response to therapy were excluded.

All studies used histopathological or a combination of histopathology, clinical and radiological follow-up, conventional diagnostic modalities (X-ray, endoscopy, ultrasound) and surgical exploration as reference standards. In all studies, indication for whole-body imaging was based on clinical symptoms, rise in carcinoembryonic antigen (CEA) levels, endoscopy findings or findings from other imaging methods for all or most patients. Between 46% and 71% of patients were men. Mean or median age ranged from 58 to 68 years.

Two reviewers independently selected studies for inclusion in the review. Any disagreements were resolved by consensus.

The methodological quality of the included studies was assessed using the QUADAS checklist. The authors did not state how many reviewers performed the quality assessment.

Data to populate 2x2 contingency tables (numbers of true positive, true negative, false positive and false negative) were extracted and used to calculate sensitivity and specificity values for each study. The authors extracted and used per lesion data where reported; otherwise per patient data was used.

It was not clear how many reviewers were involved in the data extraction.

Pooled estimates of sensitivity, specificity and diagnostic odds ratios for each imaging modality were calculated using a random-effects model. Performance of imaging modalities was compared using relative diagnostic odds ratios. Summary receiver operating characteristic (SROC) curves were constructed (Moses and Littenberg model) and the area under the curve was calculated for each modality.

Statistical heterogeneity was assessed using the Cochran Q test. Subgroup analyses were performed according to the presence or absence of specific study characteristics (patient versus lesion, prevalence of patients with malignant disease, blinding to clinical information, study design and year of publication).

Fourteen observational studies were included in the review (861 patients, range 24 to 114). There were 12 studies on the performance of PET, seven on the performance of CT, five for PET/CT and one for MRI. Three studies compared a single modality and 11 studies compared two or three modalities.

All the included studies had an adequate reference standard. The reference standard was not part of the index test in three studies and was blinded in 11 studies. The index test was blinded in two studies. Withdrawals were reported in only three studies.

Sensitivity estimates ranged from 74% to 100% for PET (12 studies; seven patient-by-patient based analyses and six lesion-by-lesion based analyses), 69% to 91% for CT (five studies; four patient-by-patient based analyses and one lesion-by-lesion based analysis), 75% to 95% for PET/CT (five studies; two patient-by-patient based analyses and three lesion-by-lesion based analyses) and 73% for MRI (one lesion-by-lesion based analysis). Specificity estimates ranged from 56% to 100% for PET, 50% to 93% for CT, 67% to 98% for PET/CT and 93% for MRI. Pooled estimates were not reported.

Pooled diagnostic odds ratios were 55.2 for PET (95% CI 23.2 to 131.2; 12 studies), 55.3 for PET/CT (95% CI 15.9 to 191.8; five studies) and 9.8 for CT (95% CI 4.2 to 22.8; five studies). The diagnostic odds ratio for the single MRI study was 35.1 (95% CI 13.5 to 90.4).

The area under the SROC curve for recurrence detection was 0.94 (95% CI 0.90 to 0.97) for PET, 0.94 (95% CI 0.87 to 0.98) for PET/CT and 0.83 (95% CI 0.72 to 0.90) for CT. CT had a significantly lower diagnostic performance than PET/CT (p=0.021). No significant differences were found between CT and PET or PET and PET/CT.

The Cochran Q test indicated significant heterogeneity between study results for each imaging modality. In all subgroups CT remained the modality with the lowest diagnostic performance.

That PET/CT was the most accurate whole-body imaging modality, closely followed by PET, for detection of recurrent disease in colorectal cancer patients with suspected recurrence during follow-up. CT had the lowest diagnostic performance.

The review stated a clear objective. The literature search was unrestricted by language. No attempt was made to identify unpublished studies so publication bias was possible. Measures were taken to minimise potential for errors and/or bias during study selection; it was unclear whether similar measures were taken during data extraction and validity assessment. The methodological quality of included studies was assessed using appropriate criteria and details of the assessment were reported.

The authors appeared to include per patient and per lesion data in the same meta-analyses and did not distinguish between direct comparison data and comparisons of imaging modalities between studies. There was some exploration of heterogeneity but given the sort of comparison between modalities it would have been better to have used a bivariate or hierarchical model where heterogeneity was found.

The authors conclusion should be viewed with some caution.

Practice: The authors stated that implementation of this diagnostic strategy in clinical practice would depend on cost-effectiveness.

The authors did not state any implications for further research.

None.