This review concluded that magnetic resonance imaging had good diagnostic performance for detection of lymph node metastases in patients with head and neck squamous cell carcinoma. A limited number of small studies suggested that diffusion weighted imaging was superior to conventional imaging. Limitation to studies in English and poor reporting of analyses mean that conclusions should be interpreted with caution.

To assess the diagnostic accuracy of magnetic resonance imaging (MRI) for the detection of lymph node metastases in patients with head and neck squamous cell carcinoma.

MEDLINE and EMBASE were searched from January 1990 to January 2011 and search terms were reported. CANCERLIT and The Cochrane Library were also searched. The bibliographies of included studies were screened for additional articles. Only articles published in English were included.

Studies that assessed MRI for the evaluation of cervical lymph node metastases, in five or more patients with head and neck squamous cell carcinoma, were eligible for inclusion. Included studies had to report sufficient data to calculate the absolute numbers of true positive, false negative, false positive and true negative MRI results, per lesion. Included studies were required to use histopathological analysis from surgical or biopsy samples, and/or clinical follow-up as the reference standard to confirm diagnosis.

The age of study participants ranged from 24 to 87 years and most participants were male. The median prevalence of lymph node metastases was 43%. Eight studies used 1.5T MRI, one study used 1.0T and one study used 0.6T. Six studies used intra-venous contrast agents (four used gadolinium-based vascular contrast agent, and two used super paramagnetic iron oxide intra-cellular contrast). Two studies used diffusion weighted imaging.

Two reviewers independently assessed studies for inclusion. Reviewers were blinded to study authors and journal.

The methodological quality of included studies was assessed using the 14-item QUADAS tool; only studies that passed more than nine items were included.

Methodological quality was independently assessed by two reviewers and any disagreements were resolved by discussion with a third reviewer.

Data were extracted on the numbers of true positive, false negative, false positive and true negative MRI results and these were used to calculate estimates of per lesion sensitivity and specificity, with 95% confidence intervals (CIs). Where studies reported data for other imaging modalities (positron emission tomography, computed tomography and ultrasound), these were also extracted.

Two reviewers independently extracted data and any disagreements were resolved by discussion with a third reviewer. Study authors were contacted for additional data where necessary.

Pooled estimates of sensitivity and specificity, with 95% confidence intervals, were calculated. A summary receiver operating characteristic (SROC) curve was also presented. Positive and negative likelihood ratios and the diagnostic odds ratio were derived from pooled estimates of sensitivity and specificity. A correction factor of 0.5 was added to all zero values.

Between-study heterogeneity was assessed using the Χ² test and, where significant heterogeneity was observed, a random-effects model was used to generate summary estimates of sensitivity and specificity. Meta-regression analysis and subgroup analyses were used to explore sources of heterogeneity. Publication bias was assessed using Deeks' funnel plot asymmetry test.

Sixteen studies, with a total of 878 participants (range seven to 213), were included in the review. The most common methodological issues were possible differential verification bias (eight studies) and lack of blinding in interpretation of the index test and reference standard (six studies). Eleven studies did not report the time interval between the index test and reference standard.

The pooled estimate of sensitivity was 76% (95% CI 70 to 82) and the pooled estimate of specificity was 86% (95% CI 73 to 93). There was significant between study heterogeneity in both estimates; sensitivity estimates ranged from 33 to 100% and specificity estimates ranged from 40 to 100%.

Meta-regression analysis indicated that a lack of clearly reported selection criteria was associated with higher estimates of sensitivity and differential and partial verification biases were associated with higher estimates of specificity. The results of subgroup analyses were also presented; the diffusion weighted imaging subgroup had the highest sensitivity, 86% (95% CI 78 to 92).

Comparisons of MRI with positron emission tomography (seven studies), computed tomography (ten studies) and ultrasound (seven studies), showed no significant difference in diagnostic performance between MRI and other imaging modalities.

No evidence of publication bias was found.

MRI had good diagnostic performance for the detection of lymph node metastases in patients with head and neck squamous cell carcinoma. A limited number of small studies suggested that diffusion weighted imaging was superior to conventional imaging.

The review provided a clearly stated objective and defined inclusion criteria. Several sources were searched for relevant studies, but searches were restricted to studies in English, which raised the possibility of language bias and omission of relevant studies. Measures to minimise error and bias were applied throughout the review and the methodological quality of included studies was assessed and incorporated in the analyses. The model used to generate pooled estimates of sensitivity and specificity was not completely clear; use of a simple random-effects model would be of questionable validity given the considerable between-study heterogeneity observed. Overall, the authors' conclusions reflect the data presented, but should be interpreted with caution given the limitation to English language studies and weaknesses in the reporting of the analyses.

Practice: The authors did not specify any implications for clinical practice.

Research: The authors stated that high quality prospective studies of MRI for the nodal staging of head and neck squamous cell carcinoma were still needed.

Shanghai Leading Academic Discipline Project; Shanghai Jiaotong University Leading Academic Discipline Project.