This document discusses different types of clinical studies used in evidence-based medicine, including case reports/series, ecological studies, cross-sectional studies, case-control studies, cohort studies, randomized clinical trials, systematic reviews, and meta-analyses. It provides details on study designs, strengths and limitations, and how to interpret results including risk ratios, odds ratios, confidence intervals, and p-values. Key concepts covered include biases, confounding factors, prevalence versus incidence, and how study size influences precision.

1. Types of clinical studies
Samir Haffar M.D.
Assistant Professor of Gastroenterology

2. Glasziou P, Del Mar C & Salisbury J. Evidence based medicine Workbook.
BMJ Publishing Group – First edition – London – 2003.
Clinical epidemiology
Evidence-based medicine
Evidence-based practice

3. The 3 components of EBP
“EBM is the integration of best research evidence
with clinical expertise & patient values”
- David Sackett
Sackett et al. BMJ 1996; 312:72-3.

4. 1- Ask
PICO
2- Acquire
electonic database 4- Apply
5- Assess
Patient
dilemma
Evidence alone does not decide
Combine with other knowledge & values
3- Appraise
Principles of EBP: the 5 A

5. What is the best evidence?
• The best evidence is the evidence most likely to
provide an unbiased view of the truth
• Bias is difference between study results & truth
• Of course, we can never know the truth, but we can
try to come as close as possible by performing & using
well-designed & well executed studies

6. Interventional
RCT
Observational
Descriptive
Cross-sectional study
Case-control study
Cohort study
Analytic
Case report/series
Ecological study
Types of clinical studies
RCT: randomized controlled trial

7. Types of clinical studies
• Case report/case series
• Ecological study
• Cross-sectional study
• Case control study
• Cohort study
• Randomized clinical trial
Primary research
• Systematic review
• Meta-analysis
Secondary research

8. McGovern D, Summerskill W, Valori R, Levi M. Key topics in EBM.
BIOS Scientific Publishers, 1st Edition, Oxford, 2001.
Evidence pyramid
Increase in
evidence level
Decrease in
bias risk

9. The amount of medical literature
0
500000
1000000
1500000
2000000
2500000
Biomedical MEDLINE Trials Diagnostic
MedicalArticlesperYear
5,000
per day
1,500
per day
95 per
day
55 per
day
Glasziou P, Del Mar C. Evidence based practice workbook.
Blackwell Publishing, 2nd edition, 2007.
So much evidence, so little time

10. • Interface to MEDLINE (NLM*) & easiest way to use it
• From 1950 to date
• 19 million articles as of November 2009
• Growing at rate of 700 000 articles/year
• ≈ 5 000 indexed journals
• > 70 million search done each month
• Search terms by topics, authors or journal
Free on the Internet since the mid-1990s
* US NLM: United States National Library of Medicine

11. Good!
You acquired the article
Now you need to appraise and apply

12. 50,000 articles/yr
from 120 journals
~3,000 articles/yr
meet critical appraisal
& content criteria
(94% noise reduction)
McMaster PLUS project – First level
Critical appraisal filters
 Valid
 Ready for clinical attention
Health Information Research Unit – McMaster University – Canada

13. High quality/relevant data – Pearls
Glasziou P, Del Mar C. Evidence based practice workbook.
Blackwell Publishing, 2nd edition, 2007.
Finding high-quality evidence like searching for ‘rare pearls’

14. High quality/relevant data
Pearls
If not valid No value
If not relevant No value

15. Look at things CRITICALLY

16. All published studies should be critically appraised

17. McGovern D, Summerskill W, Valori R, Levi M. Key topics in EBM.
BIOS Scientific Publishers, 1st Edition, Oxford, 2001.
Evidence pyramid

19. McGovern D, Summerskill W, Valori R, Levi M. Key topics in EBM.
BIOS Scientific Publishers, 1st Edition, Oxford, 2001.
Evidence pyramid

20. Case report & case series
25% of published papers in clinical journals*
• Case report describes medical history of a single patient
• Case series is essentially the same of case report but
with more than one patient to illustrate an aspect of:
- Condition
- Treatment
- Adverse reaction to therapy (most commonly these days)
* Rough search of MEDLINE database from 1997 to 2000

21. Role of case reports/series
Roles Examples
Describe a new phenotype or
genotype of disease
First case report of sickle cell disease (1910)
Describe a new pathogen
(microbe, virus or
environmental exposure)
Discovery of AIDS was an observation of a
patient with immunodeficiency-related
diseases who otherwise had no reason to be
immunodeficient (1981)
Describe unknown adverse
effect of an existing drug
Reye syndrome and aspirin in children (1963)
Thalidomide & limbs defect in pregnancy(1962)
Describe a novel treatment for a
known condition
Colchicine for treatment of FMF (1972)
To remind or educate Case records in different journals for
postgraduate education ‘Care records of MJH’
Quality improvement ‘Lesson of the week’ published in BMJ
Do not make the same mistake as I did
Murad MH et al. BMJ EBM 2018;23:60-63.

22. Thalidomide & limb defect
• Classic example published in 1961 as a letter to the
editor by an obstetrician in Sydney
• 3 newborn infants had same rare limb defect over 6 wk
Suspected link between thalidomide & limb defect
• > 10,000 affected children born worldwide before this
association confirmed & drug removed from the market
McBride WG. Lancet 1961 ; 278 : 1358.

23. Case Records of the Massachusetts General Hospital
NEJM
N Engl J Med 2016;375:2082-92.

24. Lesson of the week
BMJ
BMJ 2012;344:e468

25. Advantages & disadvantages of case series
• Advantages
Useful for hypothesis generation
Informative for rare disease with few established risk factors
Characterizes averages for disorders
• Disadvantages
Prone to bias
Cannot study cause and effect relationships
Cannot assess disease frequency

26. • Between different groups during same period of time
• In the same population at different point of time
Ecological (correlation) studies
Look for associations between exposure and
outcomes in population rather than in individuals
Association between exposure and outcome:
Pearson correlation coefficient “r”

27. Ecological (correlation) study
Armstrong B.K, Doll R. Envoronmental factors and cancer incidence and moratality in different
countries, with special reference to dietary practices. Int.J.Cancer. 1975;15:617.

28. Limitations
 Inability to link exposure with disease in particular
individuals
 Lack to study effects of potential confounding factors
 Data represent average exposure levels rather than
actual individual values
Ecological studies

29. Confounding factors
• If a study demonstrates that men who drink more alcohol
have increased risk to develop lung cancer
• This is not a causal relationship:
Drinking alcohol is confounder to risk factor & outcome
Men who drink more also smoke more
• Confounding factor corrected by
Data stratification - OR for each strata

30. Confounding factor
Systematic error due to influence of a third variable
Risk factor
Smoking
Outcome
Lung cancer
Confounder
Drinking alcohol
Glasser SP. Essentials of clinical research. Springer , 1st Edition, 2008.
Drinking more alcohol is confounder to both
the risk factor (smoking) & the outcome (lung cancer)
Association of smoking & lung cancer

31. McGovern D, Summerskill W, Valori R, Levi M. Key topics in EBM.
BIOS Scientific Publishers, 1st Edition, Oxford, 2001.
Evidence pyramid

32. Cross-sectional study design
Prevalence study
At one point of time
eg: prevalence of coronary heart disease in smokers

33. Means of determining exposure
• Questionnaires (e.g., age, sex, smoking history)
• Laboratory tests (e.g., cholesterol, hemoglobin)
• Physical measurements (e.g., blood pressure, height)
• Special procedures (e.g., electrocardiogram, x-rays)
• Medical records

34. Role of cross-sectional survey
• Determines presence or absence of a disease at one
point of time (prevalence) in a sample population
• Determines association between exposure & disease
e.g.: ischemic heart disease in smokers
• Exposure and disease determined simultaneously
It is not possible often to establish a causal relationship
Hypothesis generating rather than hypothesis testing

35. Incidence & prevalence
• Incidence:
Number of new cases of a disease per year
• Prevalence:
Overall proportion of population suffering from a disease
Prevalence = Pre-test probability

36. Incidence and prevalence
Prevalence = Incidence x mean duration of disease

37. • Qualitative data (e.g. disease present or absent):
Express results into “2 by 2 contingency table”
Risk Ratio (RR) and Odds Ratio (OR) with 95% CI
• Quantitative data (e.g. body weight):
First look at data in a scatter diagram
Correlation “r” with p value
Simple linear regression “R2” with p value
Multiple logistic regression
Analytical cross-sectional study
Measurement of association between variables
CSS: cross-sectional study

38. “2 by 2” table in qualitative data
Exposure
(smoking)
Disease (lung cancer) Total
Disease No disease
Exposed a b a + b
Unexposed c d c + d
Total a + c b + d a + b + c + d

39. Risk & Relative Risk (RR)
Number of patients fulfill criteria for a given end point
divided by total number of patients
Risk in patients (diarrhea during antibiotic tt): 4/10 = 0.4
Risk in controls (diarrhea in control group): 1/10 = 0.1
• Risk
Risk of patients / risk of controls
RR: 0.4 / 0.1 = 4
• Relative Risk

40. Odds & Odds Ratio (OR)
Number of patients fulfill criteria for given endpoint
divided by number of patients who do not
Odds of patients (diarrhea during antibiotic tt): 4/6 = 0.66
Odds of controls (diarrhea in control group): 1/9 = 0.11
• Odds
Odds of patients / odds of controls
OR = 0.66 / 0.11 = 6
• Odds Ratio

41. Risk & Odds
a
a + b
Risk
a
b
Odds

42. Interpretation of RR & OR
RR or OR should be accompanied by their CIs
RR or OR > 1
Increased likelihood of outcome in exposed group
RR or OR < 1
Decreased likelihood of outcome in exposed group
RR or OR = 1
No outcome difference between exposed & control groups
CI: confidence interval

43. Odds ratio or relative risk?
OR will be close to RR if endpoint occurs infrequently (<15%)
If outcome is more common, OR will differ increasingly from RR
Altman DG et all. Systematic reviews in health care: Meta-analysis in context.
BMJ Publishing Group, London, 2nd edition, 2001.

44. Significance of CI
• When we test a new Crohn’s disease drug on randomly
selected sample of patients, the treatment effect we will
get will be an estimate of the ‘‘true’’ treatment effect for
the whole population of patients with CD in the country
• 95% CI of estimate will be range within which we are
95% confident the true population treatment effect will lie
C.I.: confidence interval
CD: Crohn’s disease

45. Confidence intervals
Value 95 % CI are commonly used
90 or 99% CI are sometimes used
Width of CI Indicates precision of the estimate
Wider the interval, less the precision
CI includes 1 No statistically significant difference
CI doesn’t include 1 Statistically significant difference

46. CI: statistical significance & precision
(a) Statistically significant , low precision
(b) Statistically significant, high precision
(c) Not statistically significant, low precision
(d) Not statistically significant, high precision
CI: confidence interval
Glasziou P et al. Evidence based practice workbook. Blackwell, 2nd edition, 2007.

47. Influence of sample size on CI precision
Width of CI (precision of the estimate)
decreases with increasing sample size
CI: confidence interval
Peat JK, et al. Health science research. Allen & Unwin, Australia, 1st ed, 2001.

48. “2 by 2” table in qualitative data
Exposure
(smoking)
Disease (lung cancer) Total
Disease No disease
Exposed a b a + b
Unexposed c d c + d
Total a + c b + d a + b + c + d

49. “2 by 2 table” in qualitative data
Exposure
(smoking)
Disease (hypertension) Total
Disease No disease
Exposed (smoking) 120 280 400
Unexposed (no smoking) 30 570 600
Total 150 850 1000
Prevalence of HTA in smokers: a/a + b = 120/400 = 0.3
Prevalence of HTA in non-smokers: c/c + d = 30/600 = 0.05
Risk Ratio (RR): 0.3/0.05 = 6
Odds of HTA in smokers a/b = 120/280 = 0.43
Odds of HTA in non-smokers c/d = 30/570 = 0.053
Odds Ratio (OR) (a/b) / (c/d) = 8.11

50. Scatter diagram
Relationship between urinary cortisol creatinine (UCC) ratio
and asthma control (N = 62)
Each dot represents one patient
AbuRuz S et al. Br J Clin Pharmacol 2006;1: 110-115.

51. Correlation “r”
AbuRuz S et al. Br J Clin Pharmacol 2006;1: 110-115.
(N = 62)
“r” expressed by Pearson correlation coefficient

52. Interpretation of correlation coefficient “r”
lies between – 1 & + 1
• Positive r One variable increases as the other increases
• Negative r One variable decreases as the other increases
• r = 0 No linear association between 2 variables
• 0 < r < 0.25 Weak association between 2 variables
• 0.25 ≤ r < 0.75 Intermediate association between 2 variables
• 0.75 ≤ r < 1 Strong association between 2 variables
• r = 1 Perfect association between 2 variables – rare
Correlation does not necessarily imply causation

53. When not to calculate “r” ?
 Non-linear relationship
 Presence of outliers
 Data comprise subgroups of individuals

54. Probability value (p value)
• p value is probability that observed difference between
2 groups might occur by chance
• Many use p value of 0.05 as cut off for significance
p < 0.05 Observed difference between groups is so
unlikely to have occurred by chance
Considered as statistically significant
p > 0.05 Observed difference between groups might
have occurred by chance
Considered as not statistically significant

55. • p > 0.05 Statistically insignificant
• p < 0.05 Statistically significant
Probability value (p value)
Statistically
significant
Clinically
significant
Doesn't
mean

56. Statistical versus clinical significance
• Pentoxifylline vs placebo in PAD* (1992)
40 patients randomized to pentoxifylline or placebo
Maximum pain-free walking distance longer in
pentoxifylline group than in placebo group (p < 0.001)
Conclusion: pentoxiphylline clinically effective
• Close examination of data:
Difference in maximum walking distance: 3.5 feet
Doctors & patients consider it not clinically significant
* PAD: peripheral arterial disease
McGovern D et al. Key topics in EBM. BIOS Scientific Publishers, Oxford, 2001.

57. Confidence interval or p value?
• Authors of articles could report both p values & CIs
• CI convey more useful information than p values
• If only one is to be reported, then it should be the CI
• p value is less important & can be deduced from CI

58. R-squared in linear regression

59. R-square in linear regression
• R-squared is always between 0 and 100%
• 0% indicates that the model explains none of the
variability of the response data around its mean
• 100% indicates that the model explains all the
variability of the response data around its mean.
Regression does not necessarily imply causation

60. Multiple logistic regression
Statistically significant difference between black & white mothers
Large CI: imprecise result
Larger study needed to generate more precise estimate of effect
832 pregnant women, 22 (2.7%) develop preeclampsia

61. Hill’s criteria for causal relationship
Hill AB. Proceedings of the Royal Society of Medicine. 1965;58:295-300.
Hill’s criteria Criteria definition
1. Strength
effect size
A small association does not mean there is not a causal effect,
though the larger the association, the more likely it is causal
2. Consistency
reproductibilty
Consistent findings observed by different persons in different
places with different samples strengthens likelihood of an effect
3. Specificity Causation is likely if there is very specific population at specific
site and disease with no other likely explanation
4. Temporality The effect has to occur after the cause
5. Biological
gradient
dose-response
Greater exposure should generally lead to greater incidence of
the effect
6. Plausibility Plausible mechanism between cause and effect is helpful
7. Coherence Coherence between epidemiological & laboratory findings
increases likelihood of effect
8. Experiment Occasionally it is possible to appeal to experimental evidence
9. Analogy Effect of similar factors may be considered

62. Naranjo probability score of adverse drug reaction
Items Yes No Don’t
know
1. Are there previous conclusive reports on this reaction?
2. Did adverse event appear after the drug was administered?
3. Did adverse reaction improve when the drug was discontinued
or a specific antagonist was administered?
4. Did the adverse reaction reappear with drug readministiond?
5. Are there alternative causes that could caused the reaction?
6. Did the reaction reappear when a placebo was given?
7. Was the drug detected in the blood (or other fluids) in
concentrations known to be toxic?
8. Was the reaction more severe when the dose was increased or
less severe when the dose was decreased?
9. Did the patient have similar reaction to the same or similar
drug in any previous exposure?
10. Was the adverse event confirmed by any objective evidence?
+1
+2
+1
+2
–1
–1
+1
+1
+1
+1
0
–1
0
–1
+2
+1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Adverse drug reaction probability: definite 9, probable 5-8, possible 1-4, doubtful < 0
Naranjo CA et al. Clin Pharmacol Ther 1981;30:239-245.

63. Advantages & disadvantages of
cross-sectional study
Advantages Disadvantages
Provide estimate of prevalence Dos not provide incidence
Easy, quick and cheap Prone for biases
Useful for chronic conditions
with low case fatality
Rare diseases, short duration,
high case fatality not detected
Provide wealth of data for
further research
Minimal information of natural
history of disease

64. Diagnostic study
Suspected target
condition
Guyatt G et all. Users’ guides to medical literature: manual for EBP.
McGraw-Hill, New York, USA, 2nd edition, 2008.
Accuracy of diagnostic test compared to gold standard
Gold standard test
Positive
Negative
Diagnostic test
Positive
Negative

65. Accuracy of a diagnostic test
• Dichotomous test (only 2 results)
Sensibility (Sn) & Specificity (Sp)
Positive Predictive Value (PPV)
Negative Predictive Value (NPV)
Likelihood Ratios + & – (LRs)
Diagnostic Odds Ratio (OR)
• Multilevel test (> 2 results)
Receiver Operating Characteristic (ROC)
Newman TB & Kohn MA. Evidence-based diagnosis.
Cambridge University Press, Cambridge, UK, 1st edition, 2009.
with 95%CIs

66. Key components of your clinical question
Concept of PICO
P Patient Elderly patient with IDA
I Intervention Bone marrow aspirates
C Comparaison Ferritin
O Outcome Accuracy (Sn – Sp – PPV – NPV – LR)
* IDA: Iron Deficiency Anemia

67. Diagnosis of IDA in elderly patients
2 x 2 contingency table
Gold standard test
Bone marrow aspirates Row totals
Disease present Disease absent
Diagnostic test
Ferritin
Positive
Negative
Column totals
Disease: Iron deficiency anemia (IDA)
Gold standard test: Bone marrow aspirates
Diagnostic test: Serum ferritin

68. http://www.cebm.net/index
Disease: iron deficiency anemia
Gold standard test: bone marrow aspirate
Diagnostic test: serum ferritin

69. ROC curve
Disease: Iron deficiency anemia
Gold standard test: Bone marrow aspirate
Diagnostic test: Serum ferritin
AUC ROC: area under the curve of receiver operating characteristic

70. McGovern D, Summerskill W, Valori R, Levi M. Key topics in EBM.
BIOS Scientific Publishers, 1st Edition, Oxford, 2001.
Evidence pyramid

71. • Investigates association between disease/suspected causes
• People with disease identified & their medical history
examined retrospectively to identify the risk factors
• Matched control group free from disease identified
& data collected from them in identical fashion
• 2 groups well matched to avoid confounding factors
Case-control Study

72. Case-control study design
Investigate etiology or outcome of disease

73. “2 by 2” table in qualitative data
Exposure Disease Total
Disease No disease
Exposed a b a + b
Unexposed c d c + d
Total a + c b + d a + b + c + d
Odds of disease in exposed group: a/b
Odds of disease in unexposed group: c/d
Odds ratio (OR): (a/b) / (c/d)

74. Advantages & disadvantages of
case-control study
Advantages Disadvantages
Cheap Cannot establish prevalence
Quick & easy to conduct Retrospective/more prone to bias
Good for disease with long
latency periods
Can only assess one disease
Can assess multiple exposures Probably not generalizable
Good for rare diseases

75. McGovern D, Summerskill W, Valori R, Levi M. Key topics in EBM.
BIOS Scientific Publishers, 1st Edition, Oxford, 2001.
Evidence pyramid

76. What is cohort?
Ancient Roman military unit, a band of warriors
Persons banded together
Group of persons with a common statistical characteristic

77. Cohort study
• Investigates link between a hypothetical cause and
a defined outcome
• Subjects exposed to suspected risk factor (cohort) &
similar subjects not exposed (control) are identified &
followed prospectively over a period of time (years or
decades) to identify incidence of outcome in both group
• Can be retrospective (if clear point of 1st exposure)
• 2 groups well matched to avoid confounding factors

78. Cohort study
Investigate etiology or outcome of disease

79. • Express results into “ 2 by 2 table”
Qualitative data: p value (chi-square test) & 95%CI
Quantitative data: grouping – correlation calculations
• Confounding factors
Data stratification & OR for each strata
Correlation & regression
• Association does not prove a causal relationship
Number of questions need to be answered
Cohort Study

80. British doctors study
One of the most famous cohort study
• 34 440 British male & female doctors recruited in 1951
4 groups: nonsmokers, light, moderate & heavy smokers
• Publication of 10-year interim results in 1964
Substantial excess in lung cancer mortality & all cause
mortality in smokers, with dose-response relationship
• Publication of 20-, 40- and 50-year results: same results
• Follow up: 94% of those recruited in 1951 & not died
Doll R et al. BMJ 1964;i:1399–1467.
Doll R et al. BMJ 1976;Ii:1525–1536.
Doll R et al. BMJ 1994;309:901–911.
Doll R et al. BMJ 2004;328:1519–1528.

81. “2 by 2” table in qualitative data
Exposure Disease Total
Lung cancer No lung cancer
Smoking a b a+b
No smoking c d c+d
Total a + c b + d a + b + c + d

82. Incidence rate in cohort study
Exposure Disease Total
Lung cancer No lung cancer
Smoking 70 a 6 930 b 7 000 a+b
No smoking 3 c 2997 d 3 000 c+d
Total 73 a+c 9927 b+d 10 000 a+b+c+d
IR of lung cancer exp a/a+b = 70/7 000 = 10 per thousand
IR of lung cancer non-exp c/c+d = 3/3 000 = 1 per thousand
Relative risk (RR): IR exp / IR unexp = 10 / 1 = 10
Attributable Risk: (IR exp– IR unexp ) / IR exp . 100 = 90%
IR: incidence rate

83. Framingham Heart Study
Framingham city, MA (1948)
• Original cohort  5,000 of 30,000 residents in Framingham
30 to 59 years of age
without established coronary disease
• Exposures Smoking, obesity, HTA, high cholesterol,
physical activity, and others
• Outcomes Development of CHD, stroke, gout, & others
Examining study population every 2 years
• Duration Now on its third generation of participants
• Publications Over 1,000 published papers
CHD: coronary heart disease

84. Milestones from Framingham Heart Study
• 1948 Start of Framingham Heart Study „(FHS)
• 1960 Cigarette smoking increased risk of heart disease
• 1961 Cholesterol & high BP increased risk of heart disease „„
• 1967 Physical inactivity/obesity increased risk of heart disease
• 1970 High blood pressure increased risk of stroke
• 1974 DM associated with cardiovascular disease
• 1988 Elevated HDL cholesterol reduced risk of heart disease
• 1998 Framingham risk score (10-year CV risk w/o CV disease)
CV: cardiovascular
www.framingham.com/heart/timeline.htm

85. Survival curve
Informative way to depict results of a prognosis study
Number of events over time
or conversely
Chance of being free of these events over time

86. Overall survival after age 35 among cigarette
smokers and non- smokers
Doll R et al. BMJ 1994;309:901–911.
British Doctors Study

87. Survival curve
Patients with low grade follicular lymphomas
Earlier follow-up periods include results from more patients:
loss to follow-up – patients not enrolled in study at the same time
Survival curves more precise in earlier periods (narrower CIs)
Wood LA et al. Cancer 1999;85(6):1361-1368.

88. Survival curve after acute MI
RCTs also address issues of prognosis
Each arm of a RCT represents a cohort study
Experimental group: prognosis in patients receiving therapy
Control group: prognosis in patients who did not receive therapy
Mortality higher shortly after MI: initial steep downward slope
ISIS-2 (Second International Study of Infarct Survival) Collaborative Group.
Lancet 1988;2(8607):349-360.

89. Advantages & disadvantages of cohort study
Advantages Disadvantages
Used if no possible randomization Long time to complete
Prospective (less prone to bias) Expensive
Incidence can be calculated Not good for rare disease
Provide direct estimate of risk Loss to follow up bias
Can establish cause and effect Study may alter people’s behavior

90. The cohort study is the gold-standard of
analytical epidemiology

91. Case-control & cohort studies
• Case-control & cohort studies are used to demonstrate
association between suspected causes & disease
• Despite important differences between these 2 studies,
many of the rules regarding design & interpretation
of the results of the studies are applicable to both
McGovern DPB et al. Key topics in evidence-based medicine.
Bios Scientific Publishers, Oxford, UK, 1st Edition, 2001

92. Exposure Outcome
Cohort
Cross-sectional
Time is key
Case-control

93. Critical appraisal of observational studies
Check-list

94. Are the results of the study valid?
Did the authors choose the appropriate study design?
What were the criteria used for the disease/condition?
Was the population adequately defined?
Was the sample volume adequate?
Are there biases that the investigator did not address?
Critical appraisal of observational studies
What were the results?
How large was the point estimate?
How precise was the point estimate? (95% CI)
Can I apply the results to my population?
Were the patients similar to those in my practice?
Was the follow-up sufficiently long?

95. Biases in observational studies
Biases Explanation
Selection bias Select participants into exposed and not exposed
based on characteristics that may affect outcome
Information bias Collect different quality and extent of information
from exposed and not exposed groups
Recall bias Recall is better among cases than controls
because of the presence of the disease
Misclassification bias e.g.: same cut off level of weight for male
and female to determine malnourishment

96. Controlling confounding
• Study design
Randomization
Restriction: limiting type of individuals participating in study
Matching: making like-to-like comparisons (eg: age, sex, ..)
• Data analysis
Stratification: subset analyses of homogenous subgroups
Mathematical modeling: Multiple regression model
Cox proportional hazard

97. McGovern D, Summerskill W, Valori R, Levi M. Key topics in EBM.
BIOS Scientific Publishers, 1st Edition, Oxford, 2001.
Evidence pyramid

98. Perhaps the first large-scale clinical trial using
a properly designed randomized schema

99. Sir Austin Bradford Hill (1897 – 1991)
British epidemiologist & statistician
The father of modern RCTs

100. Number of RCTs published in PubMed

101. Basic structure of a RCT
Akobeng AK. Arch Dis Child 2005 ; 90 : 840 – 844.
Parallel trial is the most frequently used design

102. McGovern D, Summerskill W, Valori R, Levi M. Key topics in EBM.
BIOS Scientific Publishers, 1st Edition, Oxford, 2001.
Evidence pyramid

103. Systematic review & meta-analysis
Systematic reviews
(SR)
Meta-analyses
(MA)
MA may, or may not, include a SR
Egger M et all. Systematic reviews in health care: Meta-analysis in context.
BMJ Publishing Group, London, 2nd edition, 2001.

104. Number of SR/Ma published in PubMed
Harris JD et al. Am J Sports Med 2014; 42:2761-2768

105. Antibiotic prophylaxis & pancreatic necrosis
Bai Y et al. Am J Gastroenterol 2008 ; 103 : 104 – 110.
Forest plot
Reporting the results of meta-analysis

106. Bai Y et al. Am J Gastroenterol 2008 ; 103 : 104 – 110.
Antibiotic prophylaxis & pancreatic necrosis
Diamond shows the overall result of a meta-analysis

107. There can be overlap of evidence at any level
A well designed RCT can be more useful than
mediocre meta-analysis

108. Network meta-analysis

109. Direct and indirect evidence
Riley et al. BMJ 2017;358:j3932
interventions A and B have been compared to placebo (head to head)
No trials have compared A to B directly
Effect of A versus B: Subtract A versus placebo from B versus placebo

110. Number of published network meta-analysis
Bafeta A et al. BMJ 2013;347:f3675.

111. Network meta-analysis of RCTs
Intra-gastric balloons for treatment of obesity
Network of included studies with direct comparisons
Node size reflects number of studies evaluating each treatment
Line thickness reflects number of included patients
Bazerbachi F et al. Obes Surg. 2018 Apr 16.

112. Results at 6 months after balloon placement
% total body weight loss (%TBWL)
Orbera
0.01 (-7.24,7.45) Heliosphere
2.72 (0.97,4.48) 2.72 (-4.93,10.34) ReShape
3.42 (1.89,4.96) 3.41 (-4.18,11.00) 0.70 (-0.95,2.35) Obalon
6.72 (5.55,7.89) 6.71 (-0.82,14.23) 4.00 (2.69,5.31) 3.30 (2.30,4.30) Control
Bazerbachi F et al. Obes Surg. 2018 Apr 16.
Comparisons should be read from left to right
Results are expressed in OR & 95% CI
Bold numbers are statistically significant

113. SUCRA in network meta-analysis
SUCRA: surface under the cumulative ranking curves
Outcome: percentage of total body weight loss (% TBWL)
Bazerbachi F et al. Obes Surg. 2018 Apr 16.

114. Question type & study design
Study DesignQuestion
Intervention RCT
Incidence & prognosis Cohort study
Prevalence Cross-sectional study
Etiology & risk factors Cohort or case-control
Diagnosis Cross-sectional study
In each case, SR of all available studies better than individual study

115. Improving quality of reports
Type of study Tool
Case report/series CARE guidelines
Riley et al. J Clin Epidemiol 2017;89:218-235.
Diagnostic study STARD statment
Bossuyt PM et al. BMJ 2003;326:41–44.
Observational study STROBE statement
https://www.strobe-statement.org
RCTs CONSORT statement
Moher et al. BMJ 2010;340:c869.
SR/MA PRISMA statement
https://www.prisma-statement.org
CARE: case report – CONSORT: consolidated standards of reporting trials – PRISMA: preferred
reporting items of SR & MA – STARD: standards for reporting of diagnostic accuracy –
STROBE: strengthening the reporting of observational studies in epidemiology

116. Limitations of evidence
• Evidence simply doesn’t exist
• Some of what is available
Not easily accessible
Not clinically relevant
Not of good quality
Not applicable to your patient
It takes time to develop high quality evidence
& more time to get it to publication

117. New evidence pyramid
Murad MH et al. Evid Based Med 2016;21 (4):125-127.
Traditional pyramid
Revised pyramid
2 modifications
1. wavy lines between studies
2. remove SR/MA from top
use them as lens to see other
types of studies
Revising the pyramid

118. Conclusion
• Hierarchy of evidence serves to guide clinicians in an
ordered manner to the best available evidence
• Evidence from case reports becomes more significant in
case of absence of higher evidence
• There can be overlap of evidence at any level
• A well designed RCT more useful than a mediocre MA

119. Learning through play
• Try all “buttons”
• Make lots of “mistakes”
• Have fun
• Self-learning

120. Mc Graw Hill
2008
References
BMJ Publishing Group
2003
Elsevier
2009

121. End
Thank You
ThankYou