Miller's Stats Review Flashcards
Ortho Stats Review
Observational study designs
Retrospective (case control or case series), single (cross-sectional survey or case report) or prospective (cohort)
Experimental study designs
Prospective RCT
Only study type you can use to calculate incidence of disease
Cohort study
Difference between systematic review and meta-analysis
Meta-analysis is a systematic review with addition of combining data and forest plots
Phases of FDA research
1: pharmacology, 2: initial efficacy, 3: confirm efficacy, 4: post-market analysis
5 levels of evidence
1: RCT, 2: Cohort, 3: Case-control, 4: Case series, 5: Expert opinion
What makes a randomized controlled trial a level 2 study
Heterogeneous results, follow-up less than 80%, poor randomization
What determines the level of evidence in a systematic review
The lowest quality study used in the review
Bias vs confounding
Bias = systematic error in methodology, confounding = something that wasn’t accounted for that may influence results
Types of bias before the study
Selection, channeling, chronology
Types of bias during the study
Detection, recall, interviewer/assessor, performance and Hawthorne affect
Selection bias
Patient factors are not equal between groups, outpatient total knees have lower infection than inpatient because they are healthier to begin
Detection bias
Inconsistency in outcome assessment, tendency to look more closely for an outcome in one group vs the other
Recall bias
Knowledge of disease alters the patients recall of specific prior events
Assessor bias
Person doing the assessment is not blinded
Performance bias
Unintended difference in quality of care provided to two different groups in the study
Hawthorne effect
Subjects alter behavior when they know they’re being observed
What study is used to determine bias in meta-analysis
Funnel plot
Incidence vs. prevalence
Incidence = new cases over period of time, Prevalence = total number of current cases
Relative risk vs. odds ratio
RR can only be calculated in cohort study = incidence in exposed/incidence in unexposed.
How to calculate prevalence
Cross-sectional study
Categorical data
Nominal and ordinal, chi square for larger numbers, Fisher exact if smaller numbers
Continuous data
Intervals and ratios
Abnormally distributed continuous data analysis
Mann-Whitney U-test
Normally distributed data analysis
T-test if there are only two groups, ANOVA if 3+ groups, Pearson correlation coefficient and regression analysis
Sensitivity
Ability to detect disease (True Positives). TP/TP+FN
Specificity
Ability to detect healthy (True Negatives). TN/TN+FP
PPV
Probability they have disease given a positive test = TP/TP+FP
NPV
Probability they are healthy given negative test = TN/TN+FN
Type I error
Alpha error = false positive result, minimize risk so this happens < 5% of the time, hence p < 0.05
Type II error
Beta error = false negative; minimize risk so this happens <20% of the time, hence 1 – B = 1-20% = 80% power
Confidence interval distributions
95% = 1.96 distribution
Variance
Distribution of your bell curve, large sample size with narrow your variance
