Epi

Statistical Terms

search

Statistical Terms, Statistical Significance, Statistically Significant, P-Value, Confidence Interval, Clinical Significance, Clinically Significant, Study Heterogeneity, Clinical Heterogeneity, Statistical Heterogeneity

  • See Also
  1. Research Study
  2. Research GRADE
  • Approach
  • Statistical Topics
  1. General
    1. Systems Evaluation
    2. Treatment Effect
  2. Epidemiology Rates
    1. Disease Attack Rate
    2. Disease Incidence Rate (Incidence)
    3. Disease Prevalence Rate (Prevalence)
    4. Mortality Statistics (Leading Causes of Death, Mortality Vital Statistics)
  3. Risks
    1. Odds Ratio
    2. Relative Risk
    3. Weighted Mean Difference
  4. Decision Analysis (Decision Tree, Chance Graph)
    1. Incremental Cost Effectiveness Ratio (ICER)
    2. Expected Value Theory (Expected Utility Theory, Time Trade Off, Standard Gamble, Visual Analogue)
    3. Quality Adjusted Life Year ( QALY)
  5. Screening Test
    1. Contingency Grid or Cross Tab (includes Statistics Example)
    2. Bayes Theorem (Bayesian Statistics)
    3. Fagan Nomogram
    4. Experimental Error (Experimental Bias)
    5. Lead-Time Bias
    6. Length Bias
    7. Selection Bias (Screening Bias)
    8. Likelihood Ratio (Positive Likelihood Ratio, Negative Likelihood Ratio)
    9. Number Needed to Screen (Number Needed to Treat, Absolute Risk Reduction, Relative Risk Reduction)
    10. Negative Predictive Value
    11. Positive Predictive Value
    12. Pre-Test Odds or Post-Test Odds
    13. Receiver Operating Characteristic
    14. Test Sensitivity (False Negative Rate)
    15. Test Specificity (False Positive Rate)
    16. U.S. Preventive Services Task Force Recommendations
  • Definitions
  • Statistical Significance
  1. Statistical Significance is the probability that study findings are due to chance
    1. The risk of rejecting the null hypothesis when it is actually true
    2. In planning a study, a level of risk is chosen (e.g. 5% risk or P-Value of 0.05)
  2. P-Value (a-value, level of significance)
    1. P-Value < 0.05: <5% that findings due to chance
    2. Reflects reproducibility of the study findings only
    3. Does not predict individual patient's effect
  3. Power
    1. Probability of detecting a difference of effect when one truly exists (true positive)
      1. Probability of rejecting the null hypothesis when it is truly false
    2. Probability of avoiding a Type 2 Error (B-error)
      1. Power = 1 - B
      2. Power = 80% (power is typically set at 80% in medical studies)
    3. Power is affected by mutiple factors
      1. Sample Size (key factor in increasing Power)
      2. Test Sensitivity
      3. Effect size of the data
      4. Level of significance (P-Value)
  4. Confidence Interval
    1. Provides a range of possible outcomes within which a true result of a statistical estimate will fall
      1. Example: Number Needed to Treat ranges from 20 to 100
    2. Confidence Intervals demonstrate the precision and accuracy of an outcome
    3. Confidence Intervals are typically set at 95% in medical studies
      1. If a study were to be repeated 100 times, the study results would fall in the given range 95% of the time
    4. Confidence Intervals identify factors that are most clinically relevant
      1. Risk ratios (RR), Odds Ratio (OR) and Hazard Ratio (HR) are positive if >1 and negative if <1
      2. Confidence Interval for RR, OR or HR that crosses 1.0 would be a statistically insignificant result
  5. Statistical Significance does not mean clinically useful
    1. See Clinical Significance below
  • Definitions
  • Clinical Significance
  1. Reflects how much of an effect a patient sees
  2. Example:
    1. Study shows drug x significantly improves Hair Growth
    2. Reality: Even the patient cannot see the difference
  • Definitions
  • Heterogeneity
  1. Step 1: Clinical Heterogeneity (study similarity)
    1. Assess similarity of who and what was evaluated across pooled, meta-analyzed studies
    2. Evaluates similarity between studies (systematic review, meta-analysis)
    3. Combined studies in meta-analysis should have similar protocols (e.g. inclusion and exclusion criteria)
  2. Step 2: Statistical Heterogeneity
    1. Assess similarity among study results (were they consistent)
    2. Assign a P-Value to a combined group of studies that reflects the difference in their results and the likelihood that this is due to random chance
    3. Large spread of data across more than one study (i.e. greater heterogeneity) suggests the studies should not be combined in meta-analysis
  3. References
    1. Newman in Herbert (2013) EM:Rap 13(7): 7