400863 lecture 12 understanding results 3 2014

Upload: saja-qasem

Post on 02-Jun-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    1/40

    Foundations of Research and Evidence- Based Practice

    Lecture 12

    Understanding Results 3Nerida L. Klupp

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    2/40

    Where are we up to.

    Step 1. Is the result from data that is

    Continuous

    Categorical

    Step 2. Is the result

    Descriptive Inferential

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    3/40

    Where are we up to

    Step 3. If inferential, is the result about

    Statistical Significance Clinical Significance

    We will continue on with

    clinical significance..

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    4/40

    Minimum Important Difference

    Minimum Important Difference (MID) is the

    smallest worthwhile difference

    (improvement) expected by a patient toproceed with a treatment

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    5/40

    MID Example

    Pain scale 0-10 (O is no pain, 10 is worst painimaginable)

    You have heel pain and have marked this as 8

    on pain scale Treatment options are

    A) $10 comfy cushioned heel pad

    B) $300 custom orthoses, daily strengthening andstretching

    C) $5000 surgery, 6 weeks functional interruption,post-operative discomfort and some risk of harm

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    6/40

    MID Example

    What is the smallest improvement (points on apain scale) you would expect to make itworthwhile to have each treatment?

    Treatment MID

    A

    B

    C

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    7/40

    What do you do with it?

    You compare the treatment effect size with theminimal important difference. E.g. The amount ofimprovement in pain from having surgery for

    spinal pathology

    MID for having surgery= 80% improvement

    95% Confidence Interval = Between 50% and 60%This means the best evidence suggests average improvement

    in pain following this procedure is between 50% and 60%. The

    patient does not proceed because their minimum

    improvement wanted to undergo surgery is at least 80%.

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    8/40

    Minimal important difference

    At this stage you need to know!

    What MID means

    How it is used on a tree plot to determineclinical significance

    To set an MID for a patient or populationevidence question requires further clinicalreasoning skills. You are not expected tosuggest MID values until third year.

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    9/40

    So what is a tree plot?

    The box represents the point estimate andthe line represents the confidence interval

    Confidence Interval = 6-10kg

    Mean = 8kg

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    10/40

    treatment is not

    worthwhile

    Smallest clinicallyworthwhile effect or MID

    treatment is

    worthwhile

    no effect

    2 4 6 8 10 12 14 16-10 -8 -6 -4 -2 0

    Tree Plot

    Line of no effect

    (no difference between groups)

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    11/40

    treatment is not

    worthwhile

    Smallest clinically worthwhile

    effect or MID: 5 kg weight loss

    treatment is

    worthwhile

    no effect

    Treatment effect = 8 kg weight loss following resistancetraining (95% CI 6 to 10) in favour of treatment group

    2 4 6 8 10 12 14 16-10 -8 -6 -4 -2 0

    Tree Plot

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    12/40

    Clinical Significance

    Statistical significance is that p-value < 0.05

    Clinical significance is when the treatmenteffect (confidence interval) is equal or

    more than the MID.

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    13/40

    treatment is not

    worthwhile

    Smallest clinically worthwhile

    effect or MID: 5 kg weight loss

    treatment is

    worthwhile

    no effect

    2 4 6 8 10 12 14 16-10 -8 -6 -4 -2 0

    Tree Plot

    Is this result clinically significant?

    Why?

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    14/40

    If you put it all together

    Difference in means =

    The result that matters is the difference in weightloss between the groups (not within a group)

    8kg (95%CI 6kg-10kg) p

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    15/40

    But this is for continuousdata .

    what do we do if the data is

    categorical?

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    16/40

    Revision

    Continuous data involves an actualmeasurement e.g. average student mark onassignment = 62 marks

    Categorical data involves a proportion of eventse.g. how many fails compared to passes = 5 failgrades compared to 10 pass grades

    Both examples could provide information aboutstudent performance in an assignment

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    17/40

    Categorical results

    Categorical data are presented asproportions experiencing an event oroutcome of interest.

    Risk ratio is a comparison of risks

    Also called relative risk

    Risk does not mean a harmful thing

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    18/40

    TITANIC

    SURVIVAL

    ALIVE DEAD

    FEMALE 308 154

    MALE 142 709

    Risk of death on Titanic

    TOTAL

    462

    Females

    851Males

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    19/40

    Risk of death on Titanic

    The risk ratio can compare the probability ofdeath according to sex

    The probability of a woman dying was 154

    (women who died) out of 462 (total women)(154/462=0.33) or 33%.

    The probability of a man dying was 709 (menwho died) out of 851 (total men)(709/851=0.8331) or 83%.

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    20/40

    Risk of death on Titanic

    The risk ratio can compare the probability ofdeath according to sex

    83% (male death proportion) compared to 33%

    (female death proportion)= (0.8331/0.3333).

    = There is a 2.5 greater probability of death

    for males than for females.

    Risk ratio = 2.5

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    21/40

    Categorical results

    Minimal important differences (MIDs) are alsoset in exactly the same way. How much morelikely do you want an event to occur to make anintervention worthwhile?

    Confidence intervals are also the same e.g. RR =3 (95%CI 2.2- 3.8)

    In the actual research the outcome was three timesmore likely, but in the population we can estimate with95% certainty that the outcome will be between 2.2and 3.8 times more likely

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    22/40

    Example

    The box represents the point estimate andthe line represents the confidence interval

    Confidence Interval = 2.2 3.8

    Risk Ratio = 3

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    23/40

    No effect

    One thing is different

    Continuous: If there is no difference between

    2 actual measures, the difference = 0 E.g. If both groups lose 5kg then the difference

    between their improvements = 0

    Categorical: If there is no difference betweenrisks, the difference value is 1

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    24/40

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    25/40

    treatment is not

    worthwhile

    Smallest clinically worthwhileeffect or MID: 2.5 times greater risk

    treatment is

    worthwhile

    no effect

    Risk Ratio = 3 (95% CI 2.2 to 3.8) in favour of

    intervention group

    Minimal Important Difference = 2.5

    1 1.5 2 2.5 3 3.5 4 4.5 50.001 0.01 0.1 0.5

    Tree plot for categorical data

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    26/40

    Clinical significance

    A result is clinically significant if

    The point estimate is larger than the MID

    Most of the confidence interval larger than MID

    A narrow confidence interval. If it is narrow, it is

    called precise.

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    27/40

    treatment is not

    worthwhile

    Smallest clinically worthwhile

    effect or MID: 5 kg weight loss

    treatment is

    worthwhile

    no effect

    Mean treatment effect= 2 kg in weight loss followingresistance training(95% CI -4 to 0) in favour of control group

    group

    2 4 6 8 10 12 14 16-10 -8 -6 -4 -2 0

    Study A

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    28/40

    treatment is not

    worthwhile

    Smallest clinically worthwhile

    effect or MID: 5 kg weight loss

    treatment is

    worthwhile

    no effect

    Mean treatment effect= 8 kg in weight loss followingresistance training(95% CI -2 to 18) in favour of treatment group

    2 4 6 8 10 12 14 16-10 -8 -6 -4 -2 0

    Study B

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    29/40

    Activity

    From the Study A and B on the previous two slides

    1) Which study found the treatment effect wasclinically significant ? (the intervention works)

    2) Which study found the treatment effect was not

    clinically significant? (the control group did better-the intervention did not work)

    3) Which study has the most precise results?

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    30/40

    treatment is not

    worthwhile

    smallest clinically

    worthwhile effect or

    MIDvery harmful

    treatment

    very effective

    treatment

    treatment is

    worthwhile

    no effect

    Study A

    Study B

    Study C

    Tree plot of effect size

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    31/40

    Systematic reviews

    The results of systematic reviews arecalled forest plots

    They plot multiple tree plots.

    One tree plot = one clinical trial

    Results will be either mean differences orrisk/odds ratios depending on type of data

    A meta-analysis combines or pools thedata of the trials

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    32/40

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    33/40

    Music for pain relief: Categorical

    Categorical Forest Plot: Risk Ratio

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    34/40

    Music for pain relief: Continuous

    Continuous Forest Plot: Mean Difference

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    35/40

    Tips on Reading Results

    Dont skip the results section!!

    Decide on what is your primary research question

    Decide on your minimal important difference Find the point estimate and confidence intervals

    for that question

    Make sure they are for group comparisons

    And ignore all the less relevant numbers!

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    36/40

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    37/40

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    38/40

    Other studies

    These lectures on results have onlydiscussed evidence on interventionswhatabout studies about diagnosis, prognosis,

    and aetiology? They might use risk ratios

    You will learn some other common results

    for these types of studies next year.

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    39/40

    Still to come

    Final lecture

    Lecture 14 Revision session on challenging topics

    Exam advice

    Bring any questions before your exam

  • 8/10/2019 400863 Lecture 12 Understanding Results 3 2014

    40/40

    Thank you!