Skip to main content

Table 5 Study design recommendations for future analyses of EMT marker diagnostic accuracy and prediction increment

From: Diagnostic accuracy and prediction increment of markers of epithelial-mesenchymal transition to assess cancer cell detachment from primary tumors

1. Use Bayesian estimation of diagnostic accuracy latent class models, as the use of priors avoids the identifiability problem of three diagnostic tests (lymph node evaluation, radiologic imaging, and EMT marker expression in primary tumor cancer cells)

2. Estimate EMT marker prediction increment for both tumor stage and full panel of predictors of a given outcome. Models for stage would consist of tumor stage as the base predictor (either overall TNM stage or component T-stage, N-stage, and M-stage) to which EMT marker expression is added as a new predictor. Models for the full panel could include, in addition to tumor stage, other predictors of the outcome—such as age, tumor grade, or tumor subtype—as appropriate.

3. Estimate EMT marker prediction increment using a variety of outcomes, e.g. time to all-cause mortality, time to cancer-specific mortality, time to recurrence or metastasis-free survival, response to therapy.

4. When interpreting prediction increment results, give greater emphasis to risk reclassification statistics than ROC curves or measures of association. For reclassification, examine the event NRI and non-event NRI separately. Do not rely exclusively on the overall NRI that sums the event NRI and non-event NRI together.

5. Sample primary tumors systematically so that EMT marker expression can be measured in, for example, both the invasive front and the center of every tumor. Present portion-specific estimates in manuscripts, for example, diagnostic accuracy and prediction increment for EMT marker expression at the invasive front, and separately, diagnostic accuracy and prediction increment for EMT marker expression at the center of the tumor.

6. Measure multiple EMT markers in the primary tumors, preferably at least one for which expression goes down during EMT (epithelial markers) and at least one for which expression increases during EMT (mesenchymal markers and/or EMT inducers).

7. Measure multiple forms of EMT marker data (e.g. protein expression, RNA expression, gene methylation) in the same set of tumors to evaluate which type of data has the best diagnostic accuracy or prediction increment.

8. Measure EMT marker expression as continuous data whenever possible. Starting from continuous EMT marker expression data, create as many dichotomous EMT marker status variables defined by different cut points as the data permit, then evaluate the diagnostic accuracy and prediction increment of dichotomous EMT marker status for each cut point.

  1. EMT epithelial-mesenchymal transition, NRI net reclassification index, ROC receiver operating characteristic, TNM tumor, node, metastasis