Research Design, Biostatistics, and Literature Evaluation

Typically novel or rare patient population or intervention

Challenges with observational studies

Missing data are frequently a problem within observational research and may be classified as

2012;175:210-7).

Missing completely at random data are truly random and case-dependent and are at decreased

risk for introduction of bias

ii.

Missing at random indicates data that are absent and the absence is related to other patient data

(e.g., correlation with age), and therefore may increase risk of bias.

iii.

Missing not at random indicates data that are absent, but the absence is not related to one of

the above.

There are multiple ways to handle missing data (e.g., imputation), but the method should be defined

a priori given that significant amounts of missing data may introduce bias.

Confounding variables must be handled in a manner that can be controlled during analysis (i.e.,

regression models).

research reports. Outcomes from a meta-analysis may include a more precise estimate of the treatment effect

or risk factor for disease than the individual studies that contributed to the pooled analysis.

Provides examination of heterogeneity or variability in responses

studied, research methods, study subjects, and other factors that may affect the overall findings

Many meta-analyses combine results into a best estimate with statistical confidence bounds meant to

summarize what is known about the clinical problem in question.

Pooled results may incorporate the biases of individual studies and embody new sources of bias (e.g.,

publication bias).

increase flexibility in clinical trials by allowing for trial modification. Prespecified rules dictate modifications

upon scheduled interim evaluations of the data as the trial is ongoing (BMC Med 2018;16:29).

Examples of adaptive designs include: Continual reassessment method; group-sequential method;

sample size re-estimation; multi-arm, multi-stage population enrichment; biomarker adaptive; adaptive

randomization; adaptive dose-ranging; and seamless phase I/II or II/II trials.

harm to patients.

Challenges to adaptive trials include the complexity of statistical interpretation, lack of knowledge of the

scientific community about these designs, and difficulty in communication of the results.

Randomize groups (clusters) of subjects to either control or intervention groups

Useful in system or group-level intervention (i.e., medication diluent change for entire ICU), when

individual randomization is not possible, and to prevent contamination, or the phenomenon which occurs

when providers or subjects learn about the intervention during the study period and start adopting it as

standard of care, whether within the treatment group or not

May result in recruitment bias, more baseline imbalance between groups, loss of entire clusters, or