Table 4 Methodological considerations relevant to determination of the molecular profile of tumors.

From: Mucoepidermoid carcinoma (MEC) and adenosquamous carcinoma (ASC), the same or different entities?

A. Application of an adequate study design

Study design

Outcome measures

Pros

Cons

Diagnostic randomized controlled trials: adequately controlled with hypotheses stated in advance and evaluated according to a standardized protocol

Sensitivity, specificity, likelihood ratios, accuracy

Properly conducted, randomized controlled trials are the gold standard to determine accuracy, safety and effectiveness of diagnostic tests. Permit analysis by “intention-to-test” and control of biases, such as context and clinical review bias

Needs more resources, sample size is relevant, and interdisciplinary teamwork required.

Diagnostic cohort studies: allow assessment of the characteristics of a diagnostic test, with control group

Sensitivity and specificity, PPV and NPV, likelihood ratios, diagnostic odd ratios and accuracy can be calculated

Relatively inexpensive, simple to perform, well accepted among the medical research community.

Not directly tied to patient outcomes, risk of bias inherent to study design and difficult to control for confounders and interactions.

Prospective case series: large, consecutive selected, prospective case series

 

Inexpensive, simple to perform, well accepted among clinical community.

Allows only weak inferences and high likelihood of bias associated.

B. Potential confounders to take into account during the statistical analysis

Related to

To be taken into account for adjustment

Determination of the molecular alteration: laboratory methods and technics

Heterogeneity in techniques, different cut offs, variation in measurements, not comparable values provided, geographical differences in determination methods, fast advancing technology with differences over time in determinations

Study population: method of ample selection and sample size

Potential study participants with confounders known to influence experimental test accuracy excluded from study

Test performance: indeterminate tests/Not performed test

Performance of a diagnostic test may vary in different settings (each setting, different mix of patients)

C. Main biases to avoid

Type of bias

Step in research in which bias control can be applied

Selection bias

Study design

Reporting bias

Analysis and dissemination

Context bias:

Study design

Clinical review bias:

Study design