Survival Guide to Clinical Research (2018) - Dr. Fang

I. General Comments

Understanding the vocabulary of clinical research, including statistical terms, and also the pitfalls of research is critical to the practice of medicine.

Clinical research can certainly be as complex as proverbial rocket science, as it combines the biological, physical, and chemical sciences, as well as the element of human behavior. People are inherently biased and try to skew results in their expectations. Thus, the best research is conducted when there is sufficient uncertainty in the outcomes by both study scientists and patients that bias is balanced in both a positive and negative direction. When skew is out of balance, study results may not be generalizable to the larger patient population. This leads to wasted time and money in the pursuit of inappropriate interventions. Complicating matters further is the need for pivotal research to establish safety and efficacy for regulators. Keep in mind that in the USA, the approval standards for procedures and devices are far less stringent than for pharmaceuticals at this time.

II. Vocabulary

  1. Phase
  2. Prospective
  3. Retrospective
  4. Case Studies
  5. Inclusion / Exclusoin Criteria
  6. Adverse Event
  7. Adverse Drug Reaction
  8. Humanitarian Device Exemption
  9. Investigational New Drug
  10. Intention-to-Treat Analysis
  11. Meta-analysis
  12. New Drug Application
  13. Primary outcome measure
  14. Dose-response curve
  15. Non-inferiority study
  16. Equipoise
  17. FDA (510k) guidance

III. Statistics

  1. Parametric
  2. Non-parametric
  3. Normal distribution
  4. Odds ration (Hazard ratio)
  5. Hosmer-Lemeshow goodness of fit test for Survey data
  6. Jaccard Index - This is simply a ration of the intersecting portion of two sets vs. the total of the sets
  7. Correlation vs. Regression
  8. Cronbach Alpha - A method of measuring internal consistency within a confined population. Basically, larger sample sizes and lower variance results in higher alpha. High alpha translates to greater internal consistency. Think of this as a coefficient of consistency within a sample.
  9. Bootstrapping - A method of assessing sampling error or bias by resampling from the original sample. This technique can be used multiple times. For bootstrapping to be statistically valid, the samples must be independant.

IV. Pitfalls

  1. Unqualified investigators
  2. Bias
  3. Poor study design
    1. Irrelevant question
    2. Too restrictive
    3. Unrealistic expectations
    4. Lack of adequate controls
  4. Failure to account for natural history of the disease
  5. Faiilure to account for placebo effects
  6. Failure to account for comoridities
    1. Inadequate sample
    2. Inadequate binding (or masking), where applicable
    3. Lack of independance between carriables (eg. studying the effect of ange on mortality)
  7. Unexpected changes in disease status