Table 1 A summary of special considerations in genetic prognostic studies

Characteristics/challenge

Possible solutions

Potential benefits of the solution in genetic prognostic studies

Correlation among genetic polymorphisms

(i) Utilization of the linkage disequilibrium (LD) information and investigating the tagging single nucleotide polymorphisms (tagSNPs) instead can prevent this issue [23]

(i) reduces the redundancy among variables and simplify the analysis while also reducing the genotyping cost and efforts [23]

(ii) Once an association is found with a genetic polymorphism, this genomic region (usually within the same LD block) may be investigated in detail to identify the nearby ‘true’ prognostic factor that modifies the prognosis in patients

(ii) may identify the prognostic factor biologically linked to variable prognosis in patients

Genetic polymorphisms as confounders

Some of the genetic polymorphisms confounding the relationship between the prognostic factor and the outcome are likely to be in close vicinity and can be identified by investigating the genomic region in detail

Genetic confounders can be identified

Hardy-Weinberg equilibrium (HWE) testing in case-only cohorts

Whether appropriate or not remains to be established

Estimating the correct genetic model

Visual inspection of Kaplan-Meier curves for the codominant genetic model may reveal the best suitable genetic model for investigation of each polymorphism in multivariable models

Provides a logical and comprehensive solution while also reduces the number of tests to be performed

Minor allele frequency (MAF) of genetic polymorphisms

Excluding the rare polymorphisms (for example, MAF <5%) from the analysis is a common practice

Prevents unstable model construction and by reducing the multiple testing burden and increasing the events/variables ratio also improves the study power

Population stratification due to variable frequencies of genetic polymorphisms in different ethnicities

Detecting and controlling for the population substructure in the cohort eliminates this problem (for example, outlier samples may be eliminated from the analysis or ethnicity can be used as a covariate in the analysis)

Prevents biased estimations and increases the study power

Multiple testing issue due to the investigation of large numbers of polymorphisms

Correction for multiple testing using a variety of methods such as Bonferroni or false discovery rate (FDR) methods [42]

Reduces the false-positive rate (however, ironically may also increase the false-negative rate)

Use of genomic material extracted from archived specimen

Use of new technologies with high rates of successful genotyping [48, 49]

Reduces bias and increases study power by allowing the construction of models with a higher number of patients

Use of tumor versus non-tumor DNA in the same study

Using one type (either tumor or non-tumor) depending on the objectives of the study in the cohort or checking the correlation of genotype data obtained from both tumor and non-tumor DNA samples in a set of patients to see whether they are comparable with each other (for example, the tumor DNA may not be a good surrogate for non-tumor DNA all the time)

Prevents bias in study results created by alterations in tumor tissue DNA (that is, different genotypes in tumor DNA compared to non-tumor DNA)