TRANSLATIONAL RESEARCH IN PUBLIC HEALTH

Causality

MASTER IN TRANSLATIONAL MEDICINE

2021 – 2022 Academic year
1
 WHAT IS A CAUSAL RELATIONSHIP?

That association that exists between two categories of

events when a change in the frequency or quality of one
is followed by a change in the frequency or quality of the
other

McMahon & Trichopoulos. Epidemiology: principles and methods,2nd ed.

Philadelphia: Lippincott Williams & Wilkins, 2001.

 DIAGRAM TO ESTABLISH A POSSIBLE
RELATIONSHIP OF CAUSALITY

POSSIBLE EXPLANATION RESULT

ASSOCIATION

Is there selection or information

bias? YES NO

Results may explained by chance? LIKELY UNLIKELY

Is there confounding? YES NO

Is there interaction?
CAUSALITY
The sample size must be appropriate:
Sample size calculation for a Case-Control study
2
1 p (1 − p0 )
zα (1 + )�p(1 − 𝑝𝑝)̅ + zβ p1 (1 − p1 ) + 0
𝑚𝑚 𝑚𝑚
n= 2
p1 − p0

where,
Zα = Z score for the desired level of significance
Zβ = Z score for the desired power
m = the number of controls per exposed case
p0 = the expected proportion of exposure among controls
p1 = the expected proportion of exposure among cases
p0 OR
p1 =
1 + p0 OR − 1
1
p� = (p1 + p0 )
2

Marrugat J et al. Med Clin (Barc) 1998; 111:267-76

The sample size must be appropriate:
Sample size calculation for a cohort study

2
1 p1 (1 − p1 )
zα 1+ p� (1 − p� ) + zβ + p2 (1 − p2 )
m m
n= 2
p1 + p2

were,
Zα = Z score for the desired level of significance
Zβ = Z score for the desired power
m = number of unexposed participants / number of exposed participants
p1 = the probability of event in unexposed participants
p2 = the probability of event in exposed participants
p2
RR =  p2 = RR × p1
p1
� = (𝐦𝐦𝐩𝐩𝟏𝟏 + 𝐩𝐩𝟐𝟐 )/(𝐦𝐦 + 𝟏𝟏)
𝐩𝐩
Type I and Type II errors in epidemiological studies

-Type I error (or alpha error): Declaring that a difference exists when in fact it
does not

-Type II error (or beta error): Declaring that a difference does not exist when in
fact it does. Statistical power 1-β

CONCLUSION OF THE STUDY

TRUE A<B A=B A>B
A<B ----- error β -----
A=B error α ----- error α
A>B ----- error β -----
Table of sample size estimation for a case-control study

OR
Table of sample size estimation for a cohort study
α=0.05 (two-sided) β=0.20
RR p0 0.01 0.02 0.05 0.1 0.15 0.2 0.4 0.5 0.6 0.8 0.85 0.9 0.95 0.99

0.1 1059 526 207 100 65 47 20 15 11 7 6 5 4 4

0.2 1462 726 285 138 89 64 27 20 15 9 8 7 6 5
0.3 2068 1027 402 194 125 90 38 27 20 12 10 9 8 7
0.4 3030 1504 588 283 181 131 54 39 29 16 14 12 10 9
0.5 4673 2319 906 435 278 199 82 58 42 23 19 16 14 11
0.6 7785 3861 1506 721 460 329 133 93 67 35 29 24 19 16
0.7 14699 7286 2838 1356 862 615 244 170 120 59 48 38 29 23
0.8 35001 17341 6745 3213 2036 1447 564 388 270 123 97 74 53 38
0.9 147711 73147 28408 13495 8525 6039 2311 1565 1068 447 337 240 152 89
1.1 163095 80682 31234 14751 9257 6510 2389 1565 1016 329 208 100
1.2 42693 21109 8158 3841 2402 1683 604 388 244 64
1.3 19827 9798 3780 1774 1106 772 270 170 103
1.4 11631 5745 2213 1035 643 447 152 93 54
1.5 7750 3826 1471 686 424 294 97 58 32
1.6 5594 2760 1059 492 304 209 67 39 20
1.7 4266 2103 806 373 229 157 49 27
1.8 3385 1668 638 295 180 123 37 20
1.9 2769 1364 521 240 146 99 29 15
2 2319 1141 435 199 121 82 23 11
2.5 1199 588 222 100 59 39 8
3 769 376 141 62 36 23
3.5 552 269 100 43 24 15
4 424 206 76 32 17 10
5 285 138 49 20 10 5
6 211 102 36 14 6
7 167 80 27 10
8 137 65 22 7
9 116 54 18 5
10 100 47 15 4
15 58 26 7
20 40 17 3
 DIAGRAM TO ESTABLISH A POSSIBLE
RELATIONSHIP OF CAUSALITY

POSSIBLE EXPLANATION RESULT

ASSOCIATION

Is there selection or information

bias? YES NO

Results may explained by chance? LIKELY UNLIKELY

Is there confounding? YES NO

Is there interaction?
CAUSALITY
 CONFOUNDING

- A situation in which a noncausal association between

a given exposure and an outcome is observed as a
result of a third variable

“confounding variable” or “confounder”

TYPES OF ASSOCIATION

A. Causal B. Due to Confounding

Characteristic Characteristic

Observed Association
Under Study Under Study
Observed Association

Factor X

11 Disease Disease
CONFOUNDING

Principal Coffee
question: consumption
Independent
cause?
Pancreatic
(Association)
cancer

Confounded
Smoking
by:

Is coffee consumption a risk factor for pancreatic cancer or is an association

due to the fact that coffee consumption is associated to smoking?
Association between increasing coffee drinking and
increased risk of pancreatic cancer

A. Causal B. Due to Confounding

Increased Increased
Coffee Drinking Coffee Drinking

Observed Association
Observed Association

Smoking

Increased Risk of Increased Risk of

13 Pancreatic Cancer Pancreatic Cancer
GENERAL RULE FOR DEFINING THE PRESENCE
OF CONFOUNDING

-The confounding variable is Casually associated with the outcome

-and Noncasually or Casually associated with the exposure

-But is not an intermediate variable in the general pathway between

exposure and outcome

It is important to know “a priori” about possible causes of the outcome of

interest
Be careful An intermediate (or intervening variable)
IS NOT a confounding factor:
Adjusting for dehydration will reduce the strength of
association between diarrhoea and death

Principal
Diarrhoea
question:
Independent
cause

(Association) Death

Intervening
variable
Confounded
Dehydration?
by:
 CONTROL OF CONFOUNDING

During the design phase of the study:

• Matching
• Restriction

During the analysis phase of the study:

• Stratification
• Adjustment
• Stratified analysis (Mantel-Haenszel RR or OR)
• Unconditional Logistic Regression
• Conditional logistic regression for matched studies
Number of sexual partners and hepatitis C virus infection
Crude (OR) and adjusted (aOR) odds ratio

Cases Controls
No. of sexual partners n=43 n=172 OR (95%CI) aOR (95%CI)a
No. (%) No. (%)

1 23 (54) 145 (84) 1.00 1.00

3.41 2.81
2-4 13 (30) 24 (14)
(14.41-8.22) (1.14-6.89)

14.71 8.69
>4 7 (16) 3 (2)
(3.10-78.39) (1.32-57.19)

ªOR adjusted for injection-drug use, transfusion, social class and marital status by multiple logistic regression analysis.
Salleras, J Med Virol 1997; 52: 164-7
 DIAGRAM TO ESTABLISH A POSSIBLE
RELATIONSHIP OF CAUSALITY

POSSIBLE EXPLANATION RESULT

ASSOCIATION

Is there selection or information

bias? YES NO

Results may explained by chance? LIKELY UNLIKELY

Is there confounding? YES NO

Is there interaction?
CAUSALITY
INTERACTION (or Effect modification)

-When the incidence rate of disease in the presence of two or more risk
factors differs from the incidence rate expected to result from their
individual effects

-The joint effect can be greater than what we would expect (positive
interaction or synergism) or less than we would expect (negative
interaction or antagonism)
QUESTION TO ASK REGARDING POSSIBLE INTERACTION

Is there an association equally strong in strata formed on the basis of a

third variable?

NO YES

Interaction Interaction Not

Present Present
 INTERACTION

Individual effects A + Z

Expected joint effect

Observed joint effect A+Z

No interaction

Observed joint effect A+Z

(Synergism)

Observed joint effect A+Z

(Antagonism)
SYNERGISTIC AND ANTAGONISTIC INTERACTION

- Synergistic interaction The effect modifier strengths

the effect of the exposure of interest

- Antagonistic interaction The effect modifier

diminishes or eliminates the effect of the exposure of interest
EXAMPLE of INTERACTION

23
 Different strategies about confounding and interaction

Confounding To remove the influence of the confounding factor (adjusted

RR or OR)

Interaction To detect and describe interaction in the greatest possible

details
CONTROL OF INTERACTION

The way that interaction, if present, does not invalidate the

results is when the different strata are analysed separately

Therefore, it is very important to detect the presence of

interaction
 INTERACTION vs. CONFOUNDING

Crude OR/ RR in OR/ RR in Adjusted

OR/ RR Stratum 1 Stratum 2 OR/ RR

Example 1 No confounding 3.0 3.0 3.0 3.0

Example 2 Confounding 3.0 2.0 2.0 2.0

Example 3 Confounding 3.0 2.2 1.9 2.1

Should not be
Example 4 Interaction 3.0 0.8 5.5
calculated
CRUDE AND ADJUSTED OR. Example
Effectiveness of pneumococcal polysaccharide vaccination effectiveness
if there is interaction*
Vaccinated Vaccinated
Adjusted OR
cases/N controls/N Crude OR (95%CI)
(95%CI)
(%) (%)

1.23
All* 973/1895 (51.3%) 896/1895 (47.3%) -
(1.07 – 1.43)

NO Medical Risk 79/193 90/193 0.77 0.62

Conditions (40.9%) (46.6%) (0.51 – 1.18) (0.39 – 0.99)

≥1 Medical Risk 1.31 1.28

894/1702 (52.5%) 806/1702 (47.4%)
Conditions (1.12 – 1.53) (1.08 – 1.51)

*There is interaction between the variables vaccination and medical risk conditions

Domínguez A et al. Plos One 2017

 DIAGRAM TO ESTABLISH A POSSIBLE
RELATIONSHIP OF CAUSALITY

POSSIBLE EXPLANATION RESULT

ASSOCIATION

Is there selection or information

bias? YES NO

Results may explained by chance? LIKELY UNLIKELY

Is there confounding? YES NO

Is there interaction?
CAUSALITY
FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

-Temporal relationship
-Strength of the association
-Dose-response relationship
-Consistency
-Biologic plausability
-Experimental evidence
-Specificity
-Coherence
-Analogy
FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)
Temporal relationship

Necessity that the cause precedes the effect in time

This criterion is unargueable; when the temporal sequence of cause and effect is
erroneous we cannot conclude about causality
In some instances it can be difficult to decide whether an exposure really did occur
before the true origin of the disease (example: cancer is present for may years before
diagnosis)
Temporal bias (or reverse causality); Example: Association between estrogen
replacement therapy (ERT) and edometrial cancer. The true sequence is undiagnosed
cancer symptoms ERT
Temporal relationship: Mean concentration of airborne
particles and the daily death London, 1952

600 3000

Particulates (μg/m3)
Deaths per Day 500

400

300

200 600

100
0 5 10 15
DAY

Schwartz J. Environ Res 1994; 64: 36-52

FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

Strength of the relationship

Strong associations are more likely to be causal than weak associations

Weak associations are more easily explained by undetected biases

The fact than an association is weak does not rule out a causal connection
FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

Dose-response relationship (Biologic gradient)

Refers to the presence of a unidirectional dose-response curve

Such an expectation is not always present

Associations that do show a unidirectional trend in disease frequency with

increasing levels of exposure are not necessarily causal (relation between
birth rank and Down syndrome merely reflects the progressive relation between
maternal age and Down syndrome)
 DOSE-RESPONSE
Age-standardized death rates due to bronchogenic carcinoma (exclusive of
adenocarcinoma) by current amount of smoking
250
217.3
Mortality rate per 100,000 person-

200

150 143.9
years

100

59.3
51.4
50

3.4
0
Never smoked <1/2 Pack/Day 1/2-1 Pack/Day 1-2 Packs/Day 2+ Packs/Day

Hammond EC et al. JAMA 1958; 166:1294-1308

Dose-response relationship in different situations

Intraocular pressure/glaucoma Smoking/lung cancer

Anemia/symptoms Radiation/cancer
Alcohol/traffic accident (?)
Risk

Risk
Exposure Exposure

Maternal age / Down syndrome Weight/mortality

Osteoporosis / fracture Blood pressure/symptoms
Blood pressure / heart disease
Risk

Exposure Risk Exposure

FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

Consistency

Results of repeated observations of an association in different populations under

different circumstances are similar

Apparently consistent results across studies may result from publication bias
(positive results are more likely to be published than null ones)

Lack of consistency does not rule out a causal association beause some effects are
produced by their causes only under unusual circumstances
FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

Biologic plausability

The association is coherent with pathobiological processes

Plausibility is not based on logic or data, but only on prior beliefs

This is not to say that biological knowledge should be discounted when

evaluating a new hypothesis, but only to point out the difficulty to assess
that knowledge
FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

Experimental evidence

An experimental study that demonstrates that the outcome can be altered

after an intervention is often neither feasible nor ethically acceptable

The result of removal of some harmful exposure in an intervention or

prevention program (reversible association) can be considered
experimental evidence
 REVERSIBLE ASSOCIATION

Declining mortality from lung cancer in ex-cigarette smokers

The data exclude people who stopped smoking after getting cancer

20
Ratio of mortality rate of ex-smokers:

15.8
15
people who never smoked

10.7
10

5.9
4.7
5
2.0

0
0 <5 5-9 10-14 15+
Years since stopped smoking

Doll R et al. Br Med J 1976;2:1525-1536

FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

-Temporal relationship
-Strength of the relationship
-Dose-response relationship
-Consistency
-Biologic plausability
-Experimental evidence
-Specificity ??
-Coherence ??
-Analogy ??
FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

Specificity

A cause leads to a single effect

This criterion is wholly invalid: the existence of one effect does not detract
from the possibility that another effect exists
FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

Coherence

A cause and effect association does not conflict with what is known of the
natural history and biology of disease

Presence of conflicting information may indeed undermine a hypothesis

It’s not an useful criteria. It is hard to distinguish from biological plausability

FEATURES THAT SHOULD BE SOUGHT WHEN
DECIDING ABOUT CAUSALITY (Bradford Hill, 1965)

Analogy

Cause and effect relationship already established for a similar exposure or

disease

Absence of such analogy only reflects lack of imagination or experience, not

falsity of the hypothesis

It is handicapped by the “inventive imagination” of scientists who can find

analogies everywhere (Rothman KJ, Greenland S. Am J Public Health 2005)
CLASSIFICATION OF OBSERVATIONAL STUDIES BY
USES AND CAUSAL INFERENCE
Uses of the study
Etiological research Planning

Problems and Causal

Name of the study needs
Hypothesis Hypothesis identification.
Assessment of
preventive and
inference
generation testing Priority curative interventions
establishment

Fairly
Case-control X X X
suggestive

Highly
Historical cohort X X
suggestive

Prospective cohort X X Fairly firm

LEVELS OF EVIDENCE AND SUGGESTIONS FOR PRACTICE
(US PREVENTIVE SERVICES TASK FORCE)
GRADE LEVEL OF EVIDENCE LEVEL OF CERTAINTY ABOUT BENEFITS / HARMS
-Systematic review of RCT
-Individual RCT
A -Natural experiment High certainty of substantial net benefit
-Systematic review of cohort studies
-Individual cohort studies
-Outcomes based on existing records
High certainty of moderate benefit or moderate
B -Systematic review of case-control studies
certainty of moderate benefit
-Individual case-control studies

-Temporal series with controls and cohort

C studies Moderate certainty that the net benefit is small
-Temporal series without controls

-Expert opinion without critical appraisal or not Moderate or high certainty of no net benefit or that
D based on logical deductions harms outweigh benefits

Insufficient evidence to assess the balance of

I -Evidence is lacking or poor
benefits and harms of the service
GRADES OF EVIDENCE, TRANSLATION AND IMPLEMENTATION
(US PREVENTIVE SERVICES TASK FORCE)

GRADE OF RECOMMENDATION ON THE

LEVEL OF CERTAINTY EXPECTED NET BENEFIT
EVIDENCE IMPLEMENTATION OF THE INTERVENTION

Intervention, program or policy should be

A High Substantial
implemented
Intervention, program or policy should be
B High Moderate-Substantial implemented

Offer for selected patients depending on

C Moderate or high Small
individual circumstances*

D Moderate or high None Discourage implementation

Evidence is lacking or The service may be offered on individual

I poor evidence
Unknown
basis**

*The balance of benefits and harms should be explained to eligible individuals

**Patients should understand uncertainty about the balance of benefits and harms
Case Study: TOXIC SHOCK SYNDROME (TSS)

-Staphylococcal TSS was described among a cohort of young women in the USA in the 1980s
and was associated with the use of super-absorbent tampon material

-Bacterial toxins act as highly virulent superantigens that trigger massive immune cell activation
and cytokine release resulting in shock and multiorgan failure

-In 95% of menstrual cases of TSS, staphylococcal TSST-1(toxic shock syndrome toxin-1) is the
causative agent

-Host antibody deficiency is also recognised as an important factor in the development of

fulminant TSS

McDermott C. Case Reports in Critical Care, 2015

TSS: CDC 2011 CASE DEFINITION
1. Fever≥38.9º C
2. Rash-diffuse macular erythroderma
3. Desquamation, especially of palms and soles, 1-2 week after onset of illness.
4. Systolic blood pressure <90 mm Hg in adults or less than the fifth percentile by age for
children <16 yr.
5. Multisystemic involvement: presenting 3 or more of the following :
a) Gastrointestinal-vomiting or diarrhea;
b) Muscular-severe myalgia or elevated creatine phosphokinase;
c) Mucous membranes [vaginal, oropharyngeal or conjunctival] hyperemia;
d) Renal-blood urea nitrogen or creatinine twice upper limit of normal or urinary sediment
with pyuria in absence of urinary infection;
e) Hepatic-serum bilirubin, alanine aminotransferase or aspartate aminotransferase twice
upper limit of normal;
f) Hematologic platelet count < 100,000/mm3;
g) Disorientation or alteration in consciousness without focal neurologic signs in absence of
fever and hypotension
TSS: CDC 2011 CASE DEFINITION

LABORATORY CRITERIA: Negative results on blood, throat or CSF

cultures (may be positive for S. aureus) and on serology of Rocky
Mountain Spotted Fever, Leptospirosis or Measles

PROBABLE CASE: Meets the laboratory criteria and 4 of the 5 clinical

criteria

CONFIRMED CASE: Meets the laboratory criteria and all 5 clinical criteria
TSS: EPIDEMIOLOGY

-TSS is a fulminant Gram (+) infection, typically due to Staphylococcus aureus or Streptococcus
pyogenes (group A streptococci)

-The annual incidence of staphylococcal TSS with passive surveillance is about 0.5/100,000 and
about 0.4/100,000 for streptococcal TSS, though local rates may vary. In UK and Ireland an
annual incidence of 0.3/100,000 has been reported in children

-With active surveillance the incidence increased from 0.9 to 3.4/ 100,000 from 2000 to 2003 in
Minneapolis-St Paul, Minnessotta

-Case fatality is below 5% for menstrual staphylococcal TSS, 5-22% for non-menstrual
staphylococcal TSS and 30-70% for streptococcal TSS
51
TSS ASSOCIATED WITH MENSTRUAL CUP

- A 37-year-old woman presented a two-day history of fever, conjunctival hyperemia, abdominal

cramps, myalgia, vaginal discharge and diffuse erythroderma on thorax, thighs and perineum

- Ten days before, she began using a cup for menstrual blood collection (a reusable alternative to
conventional tampons)

- She used appropriate hygiene when handling and changing the cup, but reported causing a
small abrasion during one of her initial insertions

- Tampons TSS is explained by: a) accumulation of blood in the polyester foam cubes; b)
increase of vaginal pH in menstruation; c) existence of both oxygen and carbon dioxide in the
vagina

- Menstrual cups are made of silicone or rubber. These materials do no support

microbiological growth, but menstrual blood in the uterine environment is sufficient to promote
the growth of S. aureus

Mitchell MA. Can J Infect Dis Med Microbiol, 2015.

 RECURRENT TSS

-A recurrent menstrual TSS was reported in a 14-year-old girl

-With treatment her condition improved within 24 h and was discharged home 48 h
after admission on treatment of amoxicillin and clavulanic acid

-One month later she was diagnosed of recurrent menstrual TSS and clindamycin and
rifampicin for 10 days were offered to clear a presumptive staphylococcal carrier state

Possible explanation:
The production of interleukin-1 (IL-1), the protein responsible for the promotion of
fever and sepsis is inversely related to levels of progesterone and estradiol
TSST-1 is an inducer of IL-1: when progesterone and estradiol are at the lowest levels,
TSST-1 is able to induce IL-1
Dixit S. Australas J Dermatol, 2015
54
Legal note:

Given the nature and the exclusive educational and eminently

illustrative purpose of this lectures’ explanations, the author resorts
to the Article 32 of the Intellectual Property Act regarding the partial
use of other works such as images, graphics, or other material
included in the various slides.

All images presented have been included as necessary for

explanatory purposes of the lecture.
Table of sample size estimation for a case-control study

OR