biostatistics and research design 6350 n.
Skip this Video
Loading SlideShow in 5 Seconds..
Biostatistics and Research Design, #6350 PowerPoint Presentation
Download Presentation
Biostatistics and Research Design, #6350

Biostatistics and Research Design, #6350

147 Vues Download Presentation
Télécharger la présentation

Biostatistics and Research Design, #6350

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

  1. Biostatistics and Research Design, #6350 Confounding Study Design

  2. Thought for the Day: “Worry is a thin stream of fear trickling through the mind. If encouraged, it cuts a channel into which all other thoughts are drained” Arthur Somers Roche

  3. Confounding • The mixing of an effect of an extraneous variable with the effect of an independent variable on the outcome or dependent variable (Paul. DeLand, PhD, former SCO Consultant Statistician)

  4. Confounding Example: OD IncomeFrom a survey of ODs, the following average annual income information was reported: * Example from Paul DeLand, PhD

  5. Confounding Example: OD Income • Question: is years of experience a variable that is confounded with gender to affect income? • Answer: there is a strong correlation between income and years of experience, so the income difference between men and women is explained in part by the difference in years of experience • That is, the effect of gender on income is confounded with the effect of years of experience on income

  6. Confounding Example: • Nightlights and childhood myopia: • Quinn et al., Myopia and ambient lighting at night. Nature; 399:113-114, 1999. • Parents of children in university pediatric clinic (questionnaires, n = 479) • Prevalence of myopia: 10% for those in darkness, 34% with nightlight, 55% for those with room lights on • No controls for selection bias or confounding

  7. Confounding: Myopia Example Confounder Parental Myopia “Exposure”, or, Independent Variable Night Lights Children’s Myopia Study Disease of Interest-Dependent Variable

  8. Handling Confounders • Two accepted methods: • match subjects on the basis of the confounder, or limit the number of confounding subjects • E.g., in night light example: • Ensure that subjects are matched by parents, half hyperopic, half myopic OR • Limit to children of near-emmetropic parents • evaluate confounding in the analysis: via multiple regression analysis

  9. Regression Analysis: Example • Basically asking the question: “How important are certain variables?” • E.g., Tear Film Thickness Study (K Azartash, J Kwan, et al., 2009): n = 22 dry eyes Normals = 3.05 ± 0.20 µm vs. Drys = 2.48 ± 0.32 µm Regression Equation: Tear Film Thickness = 4.93 – 0.547 log Age – 0.155 Female – 0.252 CL – 0.303 Dry Subject • Log age p = 0.000 • Female p = 0.051 • CL p =0.040 • Dry Subj p = 0.001 • Interpretation: Even after adjusting for age, gender, CL wear, (the confounders) the average TFT for drys is estimated at 0.303 m less than normals Overview of Biostatistics

  10. Study Design • Core Competency #1: Design and Conduct a Good Quality Student Project • Core Competency #2: Life-long Learning • Learning Objectives: • given a scenario, identify the study design • contrast advantages, disadvantages of study design • given a question, suggest the appropriate design

  11. Research Designs: Overview • Many study designs; review Dawson and Trapp, Chapter 2 (2004) • Major Types: 3 • Observational studies • Experimental studies • Meta-analyses

  12. Overview of Study Designs Observational(no intervention) Experimental(intervention) Randomized Clinical Trial/Randomized Controlled trial “Descriptive” “Epidemiological”(risk factors, natural history) RCT Example:“Phase 3 1% Azithromycin vs. Tobramycin”; Protzko et al., IOVS 2007. Case Report, Case Series Case-Control Study Clinical Series Prospective Cohort Study RCT Example:FDA Phase III Trial Prospective Cohort Example:Age-Related Eye Disease Study“AREDs” Abstract, AREDS Res. Group,Control Clin Trials, 1999 Retrospective Cohort Study RCT Example:Many Student Projects Cross Sectional Study

  13. Research Designs: Observational* • Case-Series (descriptive; how many investigators get their start) • Case-control (retrospective) • Cross-sectional (prevalence) • Cohort (prospective) • Historical cohort *exposure is not manipulated -- naturally occurring

  14. Observational Study Design: Case-Series • Case-Series: (or case reports): • descriptive; some intriguing aspect of patients • frequently lead to the generation of hypotheses that may be subsequently investigated in case-control, cross-sectional or cohort studies • generally no controls

  15. Observational Study Design: Case Series • Advantages: • very simple; clinical observation only • minimal informed consent, protocol • inexpensive • Disadvantages: • no hypothesis testing • open to investigator bias • usually no controls

  16. Observational Study Design: Case-Control • Case-Control: • backward-looking to try to find causes or risk factors • Involves Cases (disease) and Controls (no disease) • History or previous events are analyzed • Also called “longitudinal” due to time component • Case-Control studies ask “What Happened?”

  17. Onset of Study Case Control Studies:Q: How did they get sick or stay well? + exposure cases - exposure + exposure controls - exposure Direction of Inquiry

  18. Case-Control Studies: Example • Eosinophilia-myalgia Syndrome (EMS): Greenberg et al., Chapter 9, 2001 • Physicians noticed association between EMS and L-tryptophan ingestion (October ‘89) • CDC, health departments quickly involved • Case-Control studies; strong association found • L-tryptophan products removed from market (November, ‘89) • Importance of Clinical Observation!!!

  19. Case-Control Studies:Summary • ALWAYS done retrospectively • Used to evaluate association or risk • Advantages: • The best design for rare diseases • quick and inexpensive • Disadvantages: • subject to recall bias or missing information

  20. Observational Study Design:Cross-Sectional • Cross-sectional: • surveys, epidemiologic, prevalence studies • aims: describe or stage disease • data collected at one time (vs over time) • e.g., “Prevalence of Meibomian Gland Dysfunction” (Hom et al., Optom Vis Sci 67:710-712, 1990.)

  21. + exposure + outcome sample Cross Sectional Study:Q: What IS Happening? - exposure + outcome + exposure - outcome - exposure -outcome one point in time

  22. Cross-Sectional Study:Student Example • Prevalence of MGD • Hom, Martinson, Knapp, Paugh, OVS 1990;67:710-712 • N = 398 patients evaluated during primary care exams • N = 155 exhibited MGD (defined as absent, or inspissated expression w/digital pressure) • Significant correlation with age: p < 0.0001

  23. Prevalence of MGD by Decade* * Hom et al., Optom Vis Sci; 67:710-712, 1990. Percent MGD Positive n = 11 n = 33 n = 165 n = 50 n = 43 n = 35 n = 61 Age by Decade

  24. Cross Sectional Studies:Summary Measure, classify, and compare all at one point in time • Advantages: • quick, easy, inexpensive • can be used to establish population rates of disease (prevalence) • Disadvantages: • subject to bias from confounders • not considered strong evidence

  25. Observational Study Design: Cohort Studies • A “cohort” is a group of people who have something in common and who remain part of a group over an extended time; e.g., • 2nd year Optometry students! • the Framingham study of cardiovascular disease (>6000 citizens of Framingham Mass participated) • Cohort studies can be prospective or retrospective • Also called “longitudinal” due to time component • Aims: risk factors, natural history, incidence of disease

  26. + outcome Prospective Cohort Study:Q: What WILL Happen? exposed eligible subjects - outcome unexposed + outcome - outcome forward in time

  27. Prospective Cohort Studies • start by identifying the groupe.g. geographic, occupational, age • follow FORWARD in time • look for occurrence of disease and risk factors • good for determining risk and disease rates: prevalence and incidence

  28. Prospective Cohort Studies • Advantages: • good for multiple outcomes, study lots of things at once • e.g. CLEERE, CLEK at SCCO: NIH/NEI Studies (see supplemental lecture handout) • Disadvantages: • bad for rare diseases -- may not develop • takes a long time, loss to follow up • expensive

  29. Observational Study Design: Historical or Retrospective Cohort Studies • Historical Cohort: • looks at group treated similarly to determine outcomes • records must be complete and adequately detailed.

  30. Historical Cohort Design:Q: How Did They Progress? withoutcome Subjects(exposed; e.g.radiation Eligible Subjects or Records(prostate cancer) withoutoutcome Subjects(unexposed; e.g., noradiation withoutcome withoutoutcome Direction of Inquiry  Onset of Study

  31. Historical Cohort Studies • start by identifying the groupe.g. geographic, occupational, age • look at current disease status • review history for risk factors • Advantages: • good for determining risk factors and associations • quicker and cheaper than prospective • Disadvantages: possibly bias, confounders

  32. Case-Control vs. Historical Cohort: What’s the Difference? Main Differences are: • Direction of inquiry : • Case-Control: Looking backward in time • Historical Cohort: Looking from time of treatment (forward looking) • The Cohort Aspect : • Case-Control: no cohort • Historical Cohort: is a cohort • Research query • Case-Control: How did they get sick or stay well? • Historical Cohort: How did they progress? The historical subjects were a cohort, that received differing treatments

  33. Onset of Study Case Control Studies:Q: How did they get sick or stay well? + exposure cases - exposure + exposure controls - exposure Direction of Inquiry

  34. Historical Cohort Design:Q: How Did They Progress? withoutcome Subjects(exposed; e.g.radiation Eligible Subjects or Records(prostate cancer) withoutoutcome Subjects(unexposed; e.g., noradiation withoutcome withoutoutcome Direction of Inquiry  Onset of Study

  35. Observational Studies: Time Relationships Direction of Inquiry: Survey Case-Control Cohort TODAY HistoricalCohort

  36. Study Designs: Experimental Studies* • Controlled clinical trials: the “gold standard” according to many investigators • an experimental drug or procedure is compared to another drug or procedure, either a placebo or the previously accepted treatment • Uncontrolled Clinical Trial: the new treatment is described, but not compared to anything *subject is manipulated

  37. + outcome Randomized Controlled Trial or Randomized Clinical Trial experimental treatment eligible subjects randomize - outcome placebo or accepted therapy + outcome - outcome forward in time

  38. Randomized Controlled Trial or Randomized Clinical Trial Ethical concerns • should not compare to an inferior treatment • should not withhold treatment known to be effective • patients need to have informed consent to participate -- statement of risks

  39. Study Designs: Experimental Studies, Concepts • Concurrent Controls: groups treated alike in similar time frame (possibly a “sham” procedure if surgery) • “Hawthorne effect” subjects change their behavior simply because they are in a study • Masking:double is best; neither subject nor investigator knows treatment • Randomization: eliminates selection bias

  40. Study Designs: Meta-Analysis • “Meta” meaning later and more highly organized • Uses published data from several studies to permit an overall conclusion • Similar to review articles, but requires quantitation • NB: review articles; summarize known literature

  41. Study Designs: Summary • Many study designs • Objectives: • to be familiar with study design terminology • to be able to identify the major study designs