Statistical Terms

StatisticSummary of sample and estimation of unknown population parameterNPV is estimated from sample
ParameterNumber summarizing populationNPV is determined for test in population
H0Specific statement about parameters of populationH0 is NPVPET/CT of less than 90%
HABroad statement that pairs with, yet is mutually exclusive from, H0HA is NPVPET/CT of at least 90%
Test statisticSummary of information from sampleWhen comparing 2 means assuming normal distribution, with z as test statistic, z follows standard normal distribution
P valueProbability of obtaining sample statistic at least as extreme as test statistic in direction of HA if H0 were truez of 2.26 (calculated from comparing 154 patients with observed FNR of 15% to 154 patients with observed FNR of 7%) corresponds to P value of 0.0238
Type I error (α)Probability of rejecting H0 when truePhase 3 superiority trials are commonly designed with 1-sided type I error of 0.025
Type II error (β)Probability of failing to reject H0 when false (i.e., HA holds)When clinical trials are designed, type II error is set priori, with β of 0.05–0.20 commonly used
Statistical power (1 − β)Probability of rejecting H0 when HA is trueClinical trials are commonly designed with 80%–95% power
CIRange of possible values of true parameter based on specified level of confidencePathologic analysis of SLNs by routine hematoxylin and eosin revealed NPV of 0.94, with 95% CI of 0.88–0.98 (26).
Familywise error rate controlControl of probability of at least one type I errorBonferroni adjustment divides type I error by number of tests
False-discovery rate controlControl of proportion of significant results that are actually false-positivesHochberg step-down procedure orders P values to compare with adjusted α
  • NPV = negative predictive value; NaF PET/CT = sodium fluoride positron emission tomography.