1 / 33

Lecture 8: Handling uncertainty with research designs

Lecture 8: Handling uncertainty with research designs. Week 9. Aims. illustrate common uncertainties Consider how we might use probabilities to describe those uncertainties Examine what probabilities and uncertainty look like in their diagnostic, prognostic and treatment/intervention forms

edita
Télécharger la présentation

Lecture 8: Handling uncertainty with research designs

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Lecture 8: Handling uncertainty with research designs Week 9 Dr Carl Thompson, University of York

  2. Aims • illustrate common uncertainties • Consider how we might use probabilities to describe those uncertainties • Examine what probabilities and uncertainty look like in their diagnostic, prognostic and treatment/intervention forms • Understand why particular study designs help address particular uncertainties

  3. A healthcare example: what do you do? • You are an ER nurse • Needle stick injury from female heroin OD • Patient absconds • (3 per 1000) chance of HIV+|needle stick • If HIV+ then antiviral 80% chance of preventing most infections • Don’t know HIV status of patient but IV drug user and so could be • CDC guidelines suggest case by case decision on prophylaxis taking into account of severity and epidemiologic likelihood of exposure

  4. Shaw NJ & Dear PR (1990) Arch.Dis.Child., 65, 520-3

  5. Paling J. Up to your armpits in alligators: how to sort out what risks are worth worrying about. Gainesville, FL: Risk Communication and Environmental Institute, 1997

  6. Types of probability

  7. The probability that a woman of age 40 has breast cancer is about 1%. If she has breast cancer, the probability that she tests positive on a screening mammogram is 90%. If she does not have breast cancer, the probability that she nevertheless tests positive is 9%. What are the chances that a woman who tests positive actually has breast cancer?

  8. Think of 100 women. One has breast cancer, and she will probably test positive. Of the 99 who do not have breast cancer, 9 will also test positive. Thus, a total of 10 women will test positive. How many of those who test positive will actually have breast cancer?

  9. Diagnostic uncertainty • Diagnosis (as a process) is uncertain (Fleming 1997, Eddy 1984) • Differential diagnoses recognise this uncertainty • Diagnostician needs to know all possibilities and their relative frequency in a population • Diagnostic probabilities are expression of uncertainty about the differential Dx list. • E.g. the chance that an IV drug user has a blood-borne infection such as HIV, Hep B or Hep C.

  10. Rules! Comprehensiveness  of all probabilities of all possibilities = 1 (summation) One and only one possibility must be true (mutual exclusivity) E.g. the patient either has HIV or not P(HIV) + P(not HIV)=1 0.25 + 0.75 = 1 Great for single Dx but what about multiple diseases? Summation (single disease)

  11. Summation (multiple disease) • Single disease more common (low likelihood of 2 illnesses with similar signs occurring simultaneously) • Not in the elderly or those with chronic diseases • Multiple diseases need explicit representation • E.g. HIV and hep B being considered then consider each alone, both or neither: P(HIV only) + P(hep B only) + P(HIV & Hep B) + P(neither) =1 General rule P(possibility 1) + P(possibility 2) + … +P(possibility n) = 1

  12. Conditional probabilities • The probability that event E occurs given that event F has occurred (P E | F) • P(HIV|IV drug use) • Probability of HIV given IV drug use • Independence • P(HIV | female) = P(HIV | male) = P(HIV)

  13. Where do we get diagnostic probabilities from • Experience (subjective probabilities) • Empirical data (objective probabilities) • Ideal study = consecutive series (or random sample) of people with the clinical presentation of interest applying a comprehensive Dx workup with adequate follow up of those initially undiagnosed. Appraise against Dx critical appraisal criteria (Jaesche et al. 1994) • First principles or background knowledge • NB THE ‘RULES’ REMAIN THE SAME WHATEVER THE SOURCE

  14. The example • Need to know: P(HIV|IV drug user) • Ideal: recent serological survey of representative sample of IV drug users in York UK (not very likely) • Alternative: data on prevalence of HIV in non-London UK population and sub groups (IV drug users amongst them). (DoH 2002)

  15. Prognostic uncertainty • Uncertainty about future heath states rather than current health states. • the chance that someone surviving an MI will have another • Risk of liver failure in someone with Hep C • Prognosis = probability over time • 5 year survival rate for women diagnosed with breast cancer • Altered by prognostic factors (clinical stage, CD4 cell count, viral load, treatment) • Incidence (% per year) (but rates change) • Survival curves provide best description over time

  16. median 5 year survival ©Van der Walt, 2000

  17. chance tree P(death in first 5 years) Year 1 Year 2 Year 3 Year 4 Year 5

  18. median 5 year survival ©Van der Walt, 2000

  19. chance tree P(death in first 3 years) P (death yr 1) P (alive yr 0) – P (alive yr1) 100% - 83% 17% or 0.17 Year 1 Year 2 Year 3

  20. Where to get information on prognosis and prognostic factors? • MEDLINE (no quality filter) • Pubmed clinical queries • Best evidence • ACP journal club • Evidence Based Journals • National guidelines and associated resources (e.g. NELH heart disease category)

  21. Needlestick seroconversion • The overall rate of HIV transmission from a single percutaneous exposure to HIV infected blood is of the order of 0.3%. (of a thousand people, 3 will seroconvert) • Scottish executive (2002)

  22. Treatment/intervention uncertainty • Hardly any treatments are ‘miracle’ cures with few adverse effects • Diseases fluctuate and remit • E.g. the vast majority of people with hypertension will not have a stroke (prognostic uncertainty) • Lowering BP doesn’t prevent all strokes (treatment uncertainty) • So – effects if imperfect treatments must be weighed against harms

  23. Needlestick treatment uncertainty • Case control study on the use of Zidovudine suggests that 81% of seroconversions could be prevented (Cardo et al. NEJM, 337, 21:1485-1490) • (less direct – vertical transmission) but high quality evidence from RCTs of Zidovudine suggests 67% reduction. • Some evidence of efficacy but don’t know precisely how effective Zidovudine alone is.

  24. OR 0.7 (0.3 - 1.4) adj OR 0.19 (0.06 - 0.52)

  25. Combining probabilities Dr develops HIV HIV Patient has HIV 0.005 Dr does not develop HIV 0.15 No prophylaxis NO HIV 0.995 Patient does not Have HIV Dr develops HIV HIV 0 Dr does not develop HIV 0.85 NO HIV 1

  26. Conditional probabilities Levine et al. 1995

  27. averaging out 0.23 HIV+ 0.84 Hep B+ P(HIV+ | HBV+) HIV- P(HBV+) IV drug user P(HIV- | HBV+) HIV+ Hep B- P(HIV+ | HBV-) HIV- P(HBV-) P(HIV- | HBV-) P HBV+ AND HIV+ (0.84 x 0.23 = 0.19)

  28. summary • Verbal expressions of uncertainty lead to: • Variation in interpretation • Different uncertainties cannot be combined as no unified understanding • Using probabilities • More precise • Well defined rules for combining them • 3 types of uncertainty (at least) • Dx (what is wrong with this patient?) • Prognostic (what might happen with this patient?) • Rx (what are the benefits and harms of options?) • Uncertainties can be represented as chance or Decision trees

More Related