evidence based dentistry p162 Flashcards
PICO
Participants - who is the review interested in studying
Intervention (Exposure) - what is the intervention(s) of interest
Comparison - what will the interventions be compared to
Outcome - which outcomes will tell you which intervention is most effective
what is CASP
Critical Appraisal Skills Programme
for RCTs and separate checklist of systematic review and meta analysis
section A in CASP
Are the results of the study valid?
- Did the review ask a clearly focused question
- Did the authors look for the right type of papers
- Did you think all the important relevant studies were included
- Did the review’s authors do enough to assess quality of the included studies
- If the results of the review have been combined was it reasonable to do so
Section B in CASP
What are the results?
- What are the results of the review
- How precise are the results
Section C in CASP
Will the results help locally?
- Can the results be applied to the local population
- Were all the important outcomes considered
- Are the benefits worth the harms and costs
prevalence
related to a specific point in time
incidence
is the rate something in a period of time
risk
what are the chances
outcome - something that might happen
risk =
number of events of interest / total number of observations
e.g. 24 people ski down a slope, 6 fall
6/24
risk of falling = 0.25 -> 25%
odds =
number of events of interest / number without the event
e.g. 24 people ski down a slope, 6 fall
6 fall/18 didnt fall = 0.33 odds of falling (usually not %)
why be wary of risk reduction
e.g new drug reduces prevalence of disease from 4.9% to 0.8%
Small difference
but as a risk reduction that is 84% reduction (false view given)
RELATIVE RISKS THAT SEEM LARGE MIGHT NOT MEAN MUCH IF STARTING RISK IS SMALL
absolute risk difference
Difference between groups
- Risk of group 1 = 63%
- Risk of group 2 = 18%
- ARD = 63-18 = 45%
If no benefit then ARD = 0
ARD = RISK 1 (%) - RISK 2 (%)
number needed to treat
is the number of patients youd need to treat to prevent one from developing disease/condition/outcome
NNT = 1/ARD
risk ratio
risk 1 / risk 2
e.g 63% and 18%
63/18 = 3.42
so person 3.42x more likely to get pain relief than someone in the placebo group
Can’t get 0 value for this, so if the answer comes to 1 then there is no difference in risk ratio between the two criteria
relevant risk reduction
RRR = RISK1 - RISK2 / RISK 1
e.g 63% and 18%
63-18/63
45/63
= 71% risk reduction
odds ratio
Ratio of odds of pain relief in the both groups (usually intervention/control studies)
Work out odds for each group, Then divide odds over one another
OR = ODDS 1/ ODDS 2
e.g
O1 -> 40/23 = 1.74
O2 -> 5/22 = 0.23
1.74/0.23 = 7.56
*Remember odds not in %!
confidence intervals
quantify level of uncertainty of risk ratios etc
For a ratio where value of no difference is 1
- If the confidence interval (contains/overlaps/straddles 1)
- Insufficient evidence = that there is a difference between e.g paracetomol and placebo
- if this interval does not (contain/overlaps/straddles 1)
- Sufficient evidence to say there is a difference
- i.e if value > 1
CI straddles 1
Insufficient evidence = that there is a difference between e.g paracetomol and placebo
CI doesn’t straddle 1
Sufficient evidence to say there is a difference
• i.e if value > 1
systemic review process
1. Well formulated question
Using PICO
- How effective is paracetamol (I) for pain relief (O) compared to placebo (C ) after surgical removal of lower wisdom teeth in adults (P)?
2. Comprehensive data search
• Medline
• Reference lists
• Hand Searching
• Multiple languages
• being aware of reporting bias
➡ Publication bias
➡ Time lag bias
➡ Language bias
➡ Citation bias
3. Unbiased selection and abstraction process
• Selection of relevant papers
• Data extraction to a predefined data extraction form
• Process should be conducted independently by >2 reviewers
• Clear reasons described for why papers were excluded
4. Assessment of papers
• Look at methodology
➡ How well the studies have been designed and conducted
• Process should be conducted by at least >2 reviewers
• Results of the reflection should be used in the analysis
5. Synthesis of data
• Pooling process
‣ Qualitative - narrative
‣ Quantitive - meta-analysis
➡ Using statistical methods to combine the results of different studies
➡ Integrate findings
➡ Pool data
➡ Identify overall trend
tools for assessing quality of papers
comoposite scales
component approach
adv of metal analysis
Increase in power
More precise
ability to answer questions not posed by individual studies
disadv of meta analysis
Can be misleading
Should only be done when
✓Minimal differences across studies
✓Same outcome measure
✓Data in each study are available
discontinuous (biniary) data
e.g illness or not/death or not/ birth or not
Odds ratio (OR)
Risk ratio (RR)
Percent risk reduction, or relative risk reduction (RRR)
Risk difference, or absolute risk reduction (RD)
Number needed to treat (NNT) (1/RD)
continous data
e.g Blood pressure, weight, amount of pain
Weighted mean difference
- can only be used when all the outcomes are using the same scale
Standardised mean difference
- For when all the studies are assessing the exact same outcome but do it in a variety of ways
e. g example of depressed patients but using varied psychometric scales to calculate it
forest plot
Vertical line divides the graph - called the line of no effect
LHS - New treatment that has been more beneficial
RHS - No Tx or that a standard Tx has been more beneficial
the further to the left or further to the right it was, the more pronounced the benefit was found to be
If results close to or on the line = no effect
Boxes - size of study - bigger the box the bigger the study
Line through box - confidence interval
- narrower the line the narrower the confidence interval
- the wider the line = less certain
If the CI line touches the line of no effect when crossing the box then it means = no statistical signficance
Diamond = combined studies height = combined study size width = combined confidence interval
clincal heterogenity
- Variation in participants
- Interventions
- Outcomes
- Study design
metholodigical heterogeneity
Variation in methods used in studies e.g quality of allocation concealment
statistical heterogeneity
Excessive variation in the results of studies
- Variation in treatment effects above that expected by chance
*Statistical heterogeneity = poor overlap of confidence intervals - i.e outliers whose horiztontal lines don’t overlap others vertically
If P<0.1 then stats are too heterogenous so should probably not pool together
sub group analyses when
Where it is expected in advance that certain features may alter the effects of an intervention
e.g Gender, age groups, specific disease subtypes
evaluating results of meta analysis (Cochrane)
G - grading of
R - Recommendations
A - Assessment
D - Development and
E - Evaluation
factors that lower quality in a study (5)
High or unclear risk of bias
- Due to design issues or poor conduct of studies
Inconsistency between studies
- Heterogeneity
Indirectness
- PICO- were all studies similar?
Imprecision
- Numbers and CIs
Publication bias
- Likely that negative/null results not published?
summary of findings table
Summary of the key findings from a systematic review
It presents
- Quality of the evidence
- Magnitude of the effect
- Reasons behind judgements
Format
- PICO
- Outcomes
- Results
- No. studies and participants
- Relative and Absolute effects
- Certainty and quality of evidence
- Notes
