Reading an article

Dr Nahid Sherbini
Consultant Internist & Pulmonologist
Certified from Harvard Medical School in Practice of Clinical
Research

Outlines
Contents of a research paper
BASICS OF CLINICAL RESEARCH ARTICLE
• Introduction to Clinical Trials
• Study Questions
• Study Population
• Basic Study Design
• Study Blinding
• The Randomization Process
• Recruitment of Study Participants
BASIC STATISTICS

• Title
• Author’s name and affiliation
(Byline)
• Abstract
• Keywords
• Introduction
• Literature review
• Method
• Results
• Discussion
• Conclusion
• References
• Acknowledgements
• Biographical sketch
• Appendices
3

 Clear and explanatory title
 An incomplete sentence
 Brief and concise
4

 Miniature of the text
 150 to 200 words
 Paragraph or structured
 Descriptive or informative
 Complete concise sentences
 Contents of Abstracts:
 Purpose or scope of the paper
 Methods
 Results, conclusions and recommendations
6

 Identifies subject area of interest
 States the purpose and hypothesis of the study
 Provides a statement of the rationale for approach to the problem
studied.
8

 Establishes context of the study by providing a brief and balanced
review of pertinent published literature available on subject.
 General to specific
 Name-date citations
9

 Study design
 Participants and their characteristics
 When and where study conducted
 Sampling method and size
 Variables measured
 Method of collecting data
 How data analyzed: Statistical procedures used
10

 Objective presentation of key findings without interpretation
 Orderly, logical sequence using text and illustrations (figures/tables)
11

 Answers questions posed in introduction
 Interprets results in comparison to what is already known on the subject.
 Explains new understanding of the subject based on present results
 It tells the readers how present study moved from what was known to
what is new
 Outlines limitations of the study
 Discusses ideas for future research
12

 Summary
 Recommendations
13

 Complete citations for research cited
 References are listed in alphabetical order by the first author’s last name
 Citations according to style manual, e.g., APA, MLA, Chicago, etc.
14

 Included as needed
 Significant help in thinking, designing, implementing, materials supplied
etc.
 Funding agency.
15

Introduction to Clinical Trials

 Phase I
 Normal Volunteers
 Safety
 Dose-Kinetics
 Action
 Phase II
 Patients
 Evidence of Activity
 Dose
 Phase III
 Phase IV
 Post marketing
 Expanded Indications
 Safety Studies

 Phase I
 12 - 18 months, 200 volunteers
 Phase II
 1.5 – 2.5 years, 300 - 500 patients
 Phase III and beyond
 3 – 4 years, 2000 – 3000 patients

 Who
Normal physiology
 Pathophysiology: rare/severe versus common/complex
 What
 Observational: retrospective or prospective
 Interventional: better
Study Questions

 Primary question
 Most relevant question; sample size
 Secondary questions
 Can help to clarify the primary question
 “Exploratory” or “hypothesis generating”
 Insufficient statistical power
 Multiple comparisons penalties (type I error)
 Ancillary questions
 A sub-study within the primary study
 Limited generalizability
Study Questions

 Dependent: outcome
 Independent: intervention, predictor
 Covariate: may influence the relationship between dependent and independent
variables
 Surrogate: more “accessible” outcome, feasible, safe, cost-permissive
Research variables:

Case Report/Case-series
Case Control/Retrospective
Cohort/Prospective
RC
T
Credibility
and
Impact
Complexity
and
Costs

 Children
 Prisoners
 Pregnant Women
 Cognitive Impairment
 Multicenter Trials

 POPULATION:
 Target population: the group of individuals you want to
study
 Accessible population: sub-set of individuals that are
available
 Study population: included in the study, your sample
 Ideally all should have equal characteristics
Study Population

Target
Population
Accessible
Population
Study
population

Validity
 INTERNAL VALIDITY:
 Observed results are unbiased estimates of
the relationship between exposure and
outcome
 Threats: confounding and bias (e.g. selection
bias, recall bias, detection bias)
 May be affected by randomization, allocation
concealment, blinding , study population

 EXTERNAL VALIDITY:
 Unbiased estimate of the relationship between
exposure and outcome in the target population
 Generalizability of your findings
 Population from you draw your sample
(accessible population) is representative of the
target population
 Threats: sampling bias, strict inclusion criteria
 Internal validity has to be guaranteed
Validity

 Homogeneous sample: good for internal, not good for external
validity
 Large sample size: not necessarily good for external validity
 Broad inclusion criteria: good for external validity
 Random sampling: can improve external validity
 Important to define inclusion and exclusion criteria: who is in and
who is not in the target population
INTERNAL AND EXTERNAL VALIDITY:

 Process by which persons from the accessible population become
part of the study population
 Study population = random sample of the accessible population =
representative of the target population
 Threats: sampling variability or sampling error (by chance, not
bias)
SAMPLING:

 Probability sampling: at random
 Simple random sampling
 Systematic sampling
 Stratified random sampling
 Disproportional sampling
 Cluster/multi-stage sampling
 Non-probability sampling: not random, most common
 Convenience sampling
 Snowball sampling
SAMPLING:

 STUDY DESIGN:
 The ultimate goal:
 Eliminate systematic error (bias)
 Minimize random error (chance, variability)
 Ensure the generalizability of study findings
Types of design:
 Parallel
 Cross-over
 Others
Study
Design

32
Other Parallel Groups Designs
R
a
n
d
Plac
Exp
R
a
n
d
Std
Exp
R
a
n
d
Std
Exp
R
a
n
d
Std + Plac
Std + Exp
R
a
n
d
Std
Exp
Exp
Test of Timing
Test of Combination Treatment
Std
Test of Switching

 FACTORIAL DESIGN:
 An example 2 x 2:
 Pros: efficiency, test more than one hyphotesis
 Cons: possibility of interaction
 Threats: if there is interaction you can get misleading results
Placebo Drug A
Placebo Placebo + Placebo Placebo + Drug A
Drug B Drug B + Placebo Drug B + Drug A

 CROSSOVER DESIGN:
 Pros: each subject serves as his/her own control
 Cons: possibility of carry over effect or order effect
 Washout period may be needed
R
a
n
d
Std
Exp
Exp
Std
Period 1 Period 2

 BIAS:
 May be introduced by study personnel and subjects
 If you “know” subject allocation you are susceptible to bias
 Intentional or non-intentional
 Blinding/masking may reduce the risk

 TYPES OF BLINDING:
 Open label: both the patient and the investigator know allocation
 Single blind: patient does not know
 Double blind: both the investigator and the patient do not know
 Triple blind: both the investigator , the patient, and the
sponsor/statistician do not know
 All study personnel should be “blinded”

 TO ENSURE BLINDING:
 When possible make the different study treatments be
identical in appearance, touch, taste, and smell
 Have an physician/healthcare provider delivering the
treatment different from physician examining for efficacy
 Have an physician examining patient for AEs different from
physician examining for efficacy
 Choose an efficacy outcome that is difficult to introduce bias
(e.g. survival)
 The packet with the drug contains a label with just the
patient ID

 Treatment allocation schemes should:
 Remove investigator bias in the allocation of participants
 Produce treatment groups comparable with respect to known and
unknown risk factors
 Guarantee that statistical tests have valid significance levels
 Only randomization achieves all three goals

 Simple/unrestricted randomization: assigns patient to treatment
randomly independently of all other treatment assignments
 Cons: long runs, imbalances
 Pros: easy
 Blocked randomization: block size is a multiple of number of
treatments
 Pros: maintains balance
 Cons: you can guess the 4th and sometimes the 3rd position
 Random block sizes: e.g. 4 or 6 or 8 (for two treatments)
 Pros: maintains balance, impossible to guess next allocation
 Cons: more complicated
Randomization

 Stratified randomization:
 Patients characteristics (age group, gender, disease severity, etc)
 Study center (multicenter trials)
 Use limited number of strata
 Pick the most important categorical variables
 Consider using statistics to “bypass” imbalance of important
variables
Randomization

The Recruitment of Study
Participants

1. Understanding of study inclusion & exclusion criteria
2. Identification of potential participants
3. Screen for eligibility
4. Obtain informed consent
5. Enrollment into study

 Burdens
 Issues that make enrollment difficult
 Barriers
 Issues that will prevent enrollment
 Benefits:
 Medical, financial, altruism
 Risks:
 Medical: Side effects, assignment to placebo, lack of efficacy
 Logistical: Trade off with other time commitments, medical or opportunity
costs
RECRUITMENT:

 Pool of potential candidates shrinks as it goes through
progressive screens
 85-90% of patients in practice will not be eligible for
trial
Total Patients in
Group
Patients found ineligible due
to co-morbidities, lab values
100%
50%
10-15%
Patients Eligible
Fedor, C.; Cola, P. & Pierre, C., D. Responsible Research. p108-9. 2006.
Patients ENROLLED
2-5%

Drop outs or premature withdrawals:
 Threaten statistical power of the trial
 Can introduce bias if drop outs are not randomly
distributed across treatment groups
 Necessitates that sites access AE and mortality
status via alternate mechanisms where appropriate
 Threaten perception of the trial results

 “The extent to which the patient continues the agreed-upon mode of
treatment under limited supervision when faced with conflicting
demands”
 Examples of non-adherence:
 Taking an incomplete/incorrect dose
 Taking medication at wrong time
 Forgetting to take medication
 Stopping the medication before instructed
ADHERENCE
:

 Non-adherence to follow-up: impact data quality and interpretation
 Non-adherence to regimen: impact the power of the study to detect a
clinical effect or potentially effective treatment
 Example: changes in dose amount and timing
 Adds variance to the outcome
 Drug effectiveness may be underestimated
 AE may be underestimated
 Sample size must increase
 Longer/delayed trails
 Increased costs
ADHERENCE
:

Why is Adherence Important?
- Threatens the ability to complete study
- Misinterpretation of data
- Complicates statistical analyses (ITT vs per protocol)
- Links to drug failure are less clear
- Prolongs studies
- Increased sample size
- Inflates costs

 NON-ADHERENCE FACTORS:
 Participants
 Treatment Regimen
 Disease and Study Specific Issues
 ENHANCING ADHERENCE:
 Pre-randomization screening
 Run-in period
 Test dosing

Statistics
 TYPE OF DATA:
 Qualitative/categorical
 Nominal: unordered categories
 Ordinal: ordered categories
 May be more clinical relevant
 Quantitative
 Discrete: magnitude and order matters; only specified values
 Continuous: not restricted to specified values
 Interval scale: zero value is NOT meaningful
 Ratio scale: zero value is meaningful

 MEASURES OF CENTRAL TENDENCY:
 Mean: best for normal distributions, sensitive to outliers
 Median: best for asymmetric distributions (non-normal)
 Mode: best for qualitative data
 MEASURES OF DISPERSION:
 Variance: sum of the squared deviation of each score from the
mean / N-1
 Standard deviation: squared root of the variance
 Range, interquartile range (P25th - P 75th), standard error (SE), 95%
confidence interval (CI)

 TYPES OF TESTS:
 One tailed OR two tailed
 Direction of the difference known?
 Parametric OR non-parametric
 Random sampling?
 Continuous data (interval or ratio scales)?
 Normal distribution?
 Equal variances (homogeneity)?

 Type of data: continuous vs. categorical data
 Data distribution: normal vs. non-normal distribution
 Number of groups: 2 groups vs more than 2 groups
 Type of measures: repeated measures?
 What you are looking for: difference vs association vs
prediction
WHAT DO YOU NEED TO KNOW
:

Continuous data
Categorical /
binary
Normal
distribution
Non-normal
Compare two groups
(independent variable
- binary)
Unpaired and
paired t-test
Mann-Whitney
Wilcoxon
Chi-square
Fisher’s exact
Compare three or
more groups
- categorical)
ANOVA (one
or n-way
ANOVA)
Kruskal-Wallis
or Friedman
test
Chi-square
Fisher’s exact
Association between
two variables
- continuous)
Pearson
correlation
Spearman
correlation
Association between
three or more
variables
- continuous)
Multiple linear
regression
Multiple
logistic
regression

Continuous
(Normal)
Continuous
(Non-normal)
Dichotomous Multi-
Categorical
Failure Time
1
group
1 sample T-test Chi Sq.
(Fisher’s Exact)
Chi Sq.
(Fisher’s
Exact)
2
groups
T-test
Paired T-Test
Wilcoxon Rank
Sum Test
Chi Sq.
(Fisher’s Exact)
McNemar’s
Test
Chi Sq.
(Fisher’s
Exact)
Log-Rank
>2
groups
ANOVA
Linear
Regression
Kruskal Wallis Chi Sq
(Fisher’s Exact)
Logistic Reg.
Chi Sq.
(Fisher’s
Exact)
Logistic Reg.
Log-Rank
Cox Regresion

 EXAMINE NORMALITY:
 Mean = Median = Mode
 Skewness and kurtosis
 Graphical Assessment
 Statistical Tests:
 Kolmogorov-Smirnov
 Shapiro-Wilk
Normal (Gaussian) Distribution
0
0.05
0.1
0.15
0.2
0.25
2 3.6 5.2 6.8 8.4 10 11.6 13.2 14.8 16.4 18
Probability
Density

 When does it matter?
 Continuous outcome
 Small – moderate sample size
 Significantly non-normal data
 Expected effect size is small
 Non-parametric options not available
 What are the risks?
 Erroneously use parametric tests when data is non-parametric?
 False positive results
 Erroneously use non-parametric tests when data is parametric?
 False negative results (unlikely)
Data Classification in Clinical Research & Sample Size Calculation, Munir Boodhwani, MD, MMSc, FRCSC

 ANOVA:
 Parametric test
 Means ANalysis Of VAriance
 Compares more than two means/groups
 If H0 is rejected, then multiple comparisons/ contrasts between groups
are needed
 Post hoc tests: LSD test, Tukey’s HSD, test, Scheffe’s test, Bonferroni
 Assumptions: observations are independent, normally distributed data,
homogeneity of variances

 “OTHERS” ANOVA:
 RM ANOVA: Repeated Measures for each subject
(e.g. longitudinal study)
 Two-way/factorial ANOVA: two independent variables; main effect +
interaction effect (e.g. study treatment interaction)
 ANCOVA: “ANOVA + regression”; comparing means among different
groups while controlling for covariates

 REGRESSION
 Studies the relation between a predictor variable (x) and a response
variable (y)
 Both dependent and independent variable are continuous
 Can the independent(s) variable(s) predict the dependent variable?
Yi = β0 + β1 Xi
+ ε
Outcome for patient
Dependent variable
InterceptSlope
Predictor/regressor
Independent
variable
Error
term
“noise”

 Correlation quantifies the degree that two variables are related.
 Correlation coefficient (r) = how much one variable tends to change when the other
one does.
 Linear regression finds the best line that predicts Y from X.
HMS-CME Principles and Practice of Clinical Research

 CORRELATION:
 Quantifies the degree that two variables are related
 Does not mean causation
 Pearson product-moment correlation (P)
 Spearman's rank correlation (NP)
 Correlation coefficent: measures the relationship [r or rho (ρ)]
 0 - 0.25 - little or no relationship
 0.25 - 0.50 - fair relationship
 0.50 - 0.75 - moderate relationship
 >0.75 - good to excellent relationship
 -1 or 1 - perfect correlation !

 Usually used as secondary analyses
 Correlation coefficient is important to be reported – strength of association
 Does not depend on values expressed by x or y and also order does not matter
 Correlation is different than causation
 Correlation measures linear association (there might be a quadratic association)
 Correlation is sensitive to outliers

 Let us analyze together
 Scenario 1 – you have a large sample size:
 Using central limit theorem, parametric tests can be used
 Differences in p-value are very small when using non-parametric testing
 Scenario 2 – you have a small sample size:
 Central limit theorem does not apply for parametric tests
 Non-parametric tests lack power for small sample sizes
 7 or fewer values, mann-whitney is always >0.05
 5 or fewer pairs, wilcoxon sign rank is always > 0.05

 HYPOTHESIS TESTING:
 Determines the degree to which your study result can be explained by
chance or sampling variability
 A statistically significant difference means it is unlikely due to chance;
not the same as clinical significance
 The hypotheses should be defined a priori
 Null hypothesis: a=b, treatments are equal
 Alternate hypothesis: a ≠ b, treatments are different; difference can be
1-side/direction or 2-side (both directions)

 AFTER RUNNING YOUR TEST:
 Reject null hypothesis: the treatments are different
 p value < level of significance
OR
 Fail to reject the null hypothesis (“accept the null hypothesis): the
treatment are similar
 Attention to the double negative!!!
 p value > level of significance

Unobserved Truth
in the Population
Ha: Coffee prevents DM
H0: No
association
Observed
in the
Sample
Reject H0:
Treatment
are
different
True positive
(1 – β)
False positive
Type I error (α)
Fail to reject
H0:
Treatments
are similar.
False negative
Type II error (β):
True negative
(1- α)

 ERRORS IN HYPOTHESIS TESTING:
 Error type I
 False positive
  = probability of committing an error type I
 Error type II
 False negative
  = probability of committing an error type II
 1-  = power = probability of detecting a difference when one truly exists (true
positive)

 α (alpha)
 aka level of significance
 Usually 0.05 (5%)
 Probability of type I error (false positive) : rejecting null hypothesis
when it is in fact true
 Risk of providing a treatment that actually does not work
 β (beta)
 Probability of type II error (false negative): failing to detect a difference
when it actually exists
 Usually 0.2 (20%)
 Risk of not offering a treatment that in fact works
 p – value
 probability of obtaining a result as extreme or
more extreme than you found in your study by
chance alone

Regardless of the statistical procedure, sample
size formulas rely on the same parameters
1. Significance level (α)
2. Desired power (1 – β)
3. Measure of variation of the outcome variable – Ie:
standard deviation (σ)
4. Anticipated treatment/exposure effect (Ha)

 Screen the published literature: original papers, SR, meta-analysis
 Look for SD, variance, 95% CI, standard error, effect size, mean effect
 Do a pilot study
 Use an analogous study
 Expert judgment (clinical meaningful ES)
 Minimum Clinically Important Difference (MCID)
HOW DO YOU GET THE DATA YOU
NEED:

 How sample size changes in response to changes in parameters: α, β,
effect size, measures of variability
Sample size
(total)
Power Incidence rate
intervention
Incidence rate
control
100 73% 10% 30%
200 97% 10% 30%
300 69% 10% 20%
400 81% 10% 20%
400 32% 10% 15%
250 83% 15% 30%
300 89% 15% 30%
400 26% 15% 20%
600 46% 15% 20%
SENSITIVITY
ANALYSIS

 TRADE-OFFS
Less false positives (type I error) Increase sample size
Less false negatives (type II error) Increase sample size
More power (true positives) Increase sample size
Small exposure effect Increase sample size
More false positives (type I error) Decrease sample size
More false negatives (type II error) Decrease sample size
Less power (true positives) Decrease sample size
Large exposure effect Decrease sample size

 Use primary outcome
 Prefer continuous outcomes (smaller n)
 Have balanced samples (more power)
 Account for drop-outs
 If you have fixed sample sized:
 Calculate power a priori: do you really want to do this?
 Calculate power post hoc: possibility of a false negative?
SAMPLE
SIZE

 Time to event data
 Not just “survival”: time to rescue medication, time to hospital
admission, etc.
 Needs to be defined:
 When do you start counting (time origin)
 How do measure time (scale)
 Define what is your “failure”(event)
 Manages “censored” or “incomplete” data
SURVIVAL ANALYSIS:

 HOW TO DEAL WITH CENSORED DATA:
 Not all individuals are observed until they have the event
 Censored cannot be related to the probability of the event of interest
 Kaplan-Meier Estimation
 Numerical: median, 50th percentile, when K-M crosses 0.5 on the y axis
 Graphically: survival curve

 COMPARE SURVIVAL BETWEEN 2 GROUPS:
 Without censoring: compare mean time to failure with t-test or Wilcoxon
test
 With censoring: log rank test (NP)
 CONTROL FOR COVARIATES:
 Regression model
 Cox Proportional Hazards Regression
 Hazard ratio (similar to OR)

 MISSING DATA:
 Important source of bias; reduces power and precision
 HOW TO HANDLE IT:
 Per protocol (PP), complete-case analysis
 Available-case analysis (different n for each analysis)
 Weighted procedures (according to likelihood of response)
 Intention to treat analysis (ITT)
 Single imputation-based procedures:
 Mean and median imputation (decrease variance)
 Regression imputation (smaller impact on)
 Stochastic regression imputation (increase variance)
 Last observation carried forward (conservative)
 Worst/best case scenario (sensitivity analysis)
Other Issues in
Statistics

 Missing completely at random: Completely independent of observed and non-
observed data.
 Non-missing data constitutes effectively a random sample (example, a rater that becomes sick or
loss of study files)
 Missing at random (less stringent): probability of a value being missing will generally
depend on observed values (NOT MISSING VALUES), so it does not correspond
to the intuitive notion of 'random'.
- Old subjects might drop out a treatment because they have walking difficulties – as they
cannot go to the clinic center – however among older subjects, the likelihood of dropping out does not
relate to the outcome.
 Missing not at random - present when the pattern of missing data are related to
unobserved data - therefore it is impossible to predict data from other values from
the dataset
Demographic data + baseline performance Follow-up performance
OBSERVED DATA NON-OBSERVED DATA

 COVARIATE ADJUSTMENT:
 Despite randomization there might be imbalances between groups:
 e.g. a group might be older and have more females than the other group
[covariates: gender and age]
 Advantages:
 Improves estimates
 Reduces bias
 Increases statistical efficiency
 Common covariates:
 Study center in multicenter studies
 Prognostic factors

 ANOVA – add another variable (ANCOVA)
anova pain_changes treatment
anova pain_changes treatment gender
 Regression – add another variable
regress pain_changes treatment
regress pain_changes treatment gender
 Survival – add another variable (cox proportional hazard models)
 Categorical variable - Mantel-Haenszel

 To find out if whether treatment effect is different in patients with a certain
characteristic
 Advantages: ideas for new studies, finding groups of patients in whom the
treatment works better
 Risks: over interpretation and misleading results; false positives (if p=0.05, 1 in
each 20 comparisons just by chance)
SUB-GROUP ANALYSIS:

 After primary analysis, often want to look at subgroups
 Does effectiveness vary by subgroup?
 If drug effective, is it more effective in some populations?
 If results overall show no effect, does drug work in subgroup of participants?
 Are adverse effects concentrated in some subgroups?

 SUB-GROUP ANALYSIS versus COVARIATE ADJUSTMENT:
 Covariate adjustment: achieve the most accurate p value; increases
precision and power
 Sub-group analysis: different responses to the treatment; interaction
between the treatment and a certain characteristic

1. Specified in study protocol have highest validity
Especially if number is small
2. Implied by study protocol
randomization stratified by age, sex or disease stage
3. Subgroups suggested by other trials
4. (Weakest) Subgroups suggested by the data themselves (“fishing” or “data
dredging”)
Example: children under 14 born in October (“month of October victimized by
poststudy analyses biased by knowledge of results”)
5. (Disastrous) Subgroups based on post-randomization variables

 ?
 HOW TO ADDRESS MULTIPLE COMPARISONS:
 Bonferroni correction
 Fisher’s Least Significance Difference
 Duncan
 Newman-Keuls
 Tukey’s Honestly Significant Difference
 Scheffe’s comparison
 Disclose error rate (findings by chance alone)
 PROBLEM: “overcorrection” increases risk of type II error

 Meta-analysis refers to the analysis of analyses
the statistical analysis of a large collection of analysis results from individual
studies for the purpose of integrating findings. It connotes a rigorous alternative to
the casual, narrative discussions of research studies which typify our attempts to
make sense of the rapidly expanding literature.
GV Glass (1976). Primary, secondary, and meta-analysis of research. Edu Researcher 5:3-8
META-ANALYSIS:

keep up with the enormous amount of research data, judge the quality of
the studies, and integrate findings
Greater precision of effect estimates, and thus reduce probability false
negative results
Consistency of results over different study populations - generalizability
Highlight the limitations of previous studies and contribute to higher
quality of future studies

0. Identification of the need for a review
1. Preparation of a proposal for a review
2. Development of a review protocol
3. Identification of research
4. Selection of studies
5. Study quality assessment
6. Data extraction
7. Data synthesis
8. Report and recommendations
9. Getting evidence into practice
Preparing the review
Reporting/
dissemination

 Larger studies should be emphasized in the analyses
 Weight each ES by its sample size
 Weight each ES by the inverse variance (better)
 Non-quantitative data synthesis: tables
 Quantitive data synthesis: forest plot (OR, ES)
 Measure of effect: odds-ratio, risk-ratio, effect size
META-
ANALYSIS:

 Sensitivity analysis: assess the robustness of the findings
 Exclude from the analyses some studies: e. g. older studies, lower quality
studies, extremes of distribution
 Heterogeneity: assess variation of outcomes between studies
 Cochran’s Q
 Publication bias
 Begg and the Egger test
 Begg funnel plot
META-ANALYSIS:

 RESEARCH QUESTION
 H0: Tight glycemic control will have the same perioperative morbidity
as standard glycemic control
 H1: Tight glycemic control will not have the same perioperative
morbidity as standard glycemic control

STUDY DESIGN
 Interventional
 New treatment strategy versus standard care (no placebo)
 Randomized and controlled
 Multicentric
 Phase II trial

PRIMARY OUTCOME SECONDARY OUTCOMES

 STUDY POPULATION:
 Target population: children 0-36 months undergoing
cardiac surgery with cardiopulmonary bypass
 Acessible population: same as above + receiving surgical
treatment Boston and University of Michigan Children’s
Hospital
 Study population: same as above + admitted between
September 1006 to May 2012 that were eligible and whose
parents/guardians gave consent
 SAMPLING:
 Convenience sampling (non-probability sampling)

 INCLUSION/EXCLUSION CRITERIA:

 BLINDING:
 The patient: YES
 Intraoperative team: YES
 The bed-side clinicians: NO
 Adjudicators of the primary outcome:
YES
 Investigators: YES

 RANDOMIZATION:
 Random blocks, stratified
by center
 Sealed envelope

 ADHERENCE
 To the study protocol:
 Bedside measurements
 Insulin therapy by dosing alogorithm

 STATISTICAL ANALYSIS
 Sample size

 Primary outcome
 ITT
 Regression
 Logistic regression (secondary
analysis)

 Secondary outcomes:
 Categorical: Fisher’s exact test with adjustment (NP)
 Continuous: Wilcoxon rank-sum with adjustment (NP)

 Sub-group analysis:
 Definied a priori
 Based on categorical variables

 Multiple comparisons:
 No adjustment
 Risk of type I error

 RESULTS
 Primary and secondary outcomes: no differences
between groups
 Sub-group analyses: no differences
 Factors associated with infection risk:
 High surgical risk
 30 days or younger
 Hyperglicemia
 Prolonged stay ICU
 Post-op glucocorticoid therapy
 Not significant when controlling for prolonged ICU
stay

 INTERNAL VALIDITY:
 Threats: convenience sampling, differences in protocol, target population?
 Strengths: random allocation, blinding,
 No placebo (would be ethical?)
 Glucose control in standard care not specified by the protocol
 EXTERNAL VALIDITY:
 Relative broad inclusion criteria
 Sample does no represent all critically ill pediatric patients
 Highly trained centers: lower infections and mortality rates anyway

Reading an article

More Related Content

What's hot (20)

Similar to Reading an article (20)

More from Nahid Sherbini (20)

Recently uploaded (20)

Reading an article