Meilia P, Herkutanto H, Freeman M, Zeegers M A Review of Causal Inference in Forensic Medicine , Medicine and Pathology 2020, in press

Scienfitic concepts of causal analyses in disease and injury

Prof. dr. Maurice Zeegers [email protected]

LAW AND Opiate Allergy Case

49 year old man with a prior history of Coronary Artery Disease and 9 stents undergoes knee replacement surgery

Hx states “codeine allergy

In recovery complains of severe pain, and surgeon rx morphine

Nurse reminds doc of allergy

Patient dies within 20 minutes of injection

Defense – he died due to bad heart

1 20 1 [email protected] Query US National Inpatient Sample (NIS)

Pull all cases 2000-2009 coded for prior opiate allergy in which there was an allergic reaction to an opiate

Look for death rate

Pull all cases 2000-2009 of knee replacement surgery with a history of CAD and stent placement

Look for cardiac-associated death rate

1 20 2 [email protected] Results

Opiate – 43/10,000 exposures (1 in 233)

Cardiac – 4/10,000 surgeries (1 in 2500)

This yields an Comparative Risk of 10.8 or Probability of Causation of 91%

Hazard period

The time the two competing causes overlap There was 20 minutes between injection and death

The cardiac deaths took place over a week

There are 504 20 minute periods in a week

1 20 3 [email protected] Results

Thus, the competing base rate of death secondary to the heart problems during the 20 minute hazard period was (504 X 2500) = 1,260,000

The CRR of causation attributable to the morphine = 1 in 233/ 1 in 1,260,000

5,408 to 1 (>99.9% Probability of Causation)

1 20 4 [email protected] The Epidemiological Probability

Number of people (cases) with the injury of interest Total number of the people in the population from which these cases arise

1 20 6 [email protected] Forensic Epidemiology

Epidemiology is a Forensic Epidemiology provide legal Medical Science that Fact-finders with evidence regarding investigates Probabilities a causal relation between an action and Determinants of and a medically observed harmful health outcomes in outcome human populations

1 20 5 [email protected] Probabilistic Causation

Attribution (probability that the outcome would not have occurred if it were not for a particular exposure) How likely it is that a plaintiff’s whiplash is caused by the accident, taking into account her disease history, age and sex, and the specific circumstances of the accident?

Forecasting (probability of an outcome given a particular exposure) Although X-ray use predicts broken bones very well, there is no causal relationship between the two Given the car crash circumstances what is the probability that the defendant was wearing a seatbelt.

1 20 7 [email protected] Examples of Investigative Questions Addressed by Forensic Epidemiologic Methods

What is likelihood that the asbestos exposure that Mr X experienced during his employment at company Z caused his lung cancer?

[email protected] 1 20 8 Examples of Investigative Questions Addressed by Forensic Epidemiologic Methods

Could you estimate the probability that the leg amputation of Mrs Y could have been prevented if the delay in diagnosis would not have occurred?

[email protected] 1 20 10 Examples of Investigative Questions Addressed by Forensic Epidemiologic Methods

How likely is it that the heart failure of Mrs Y was indeed caused by the side effect of this drug?

[email protected] 1 20 11 Examples of Investigative Questions Addressed by Forensic Epidemiologic Methods

How likely is it that the bladder cancer of Mrs Y was caused by passive smoking during her imprisonment given the fact that she was an ex-smoker herself?

[email protected] 1 20 15 Examples of Investigative Questions Addressed by Forensic Epidemiologic Methods

What is the chance that the death that followed the administration of the opiate by 20 minutes was due to the drug and not to other (unknown) factors?

[email protected] 1 20 14 2

Theory Health Risk Estimation Overview: Research Methods for Health Risk Estimation

The Language of Epidemiologic Study Designs Randomized Clinical Trial Cohort Study Case control Study

Factual Probabilities Probability of Disease and Injury Probability of Death Case Fatality Rate and Survival Rate

Linking a Potential Causal Factor to Injury Risk Difference (RD) (RR) Comparative risk (CRR) Probability of Causation (PC) Attributable Proportion (AP)

Sampling Error Probability Belief Probability

2 24 1 [email protected] Study designs

2 24 2 The cohort study

2 24 4 Hierarchy of study designs

2 24 6 Probability of disease and injury

Number of people (cases) with the injury of interest Total number of the people in the population from which these cases arise

2 24 7 [email protected] Prevalence

Total number of existing cases in a defined time period Total number of persons in the population to which these cases belong

2 24 8 [email protected] Incidence

Total number of new cases in a defined time period Total number of persons in the population at the beginning of this period

2 24 9 [email protected] Relationship between incidence and prevalence

incidence

prevalence healing/mortality coffee

2 24 10 Probability of Death

Total number of new deaths in a defined time period Total number of persons in the population at the beginning of this period

2 24 11 [email protected] 2009 Dutch medical negligence court case (NL 105.001.264)

Pregnant plaintiff lacked to receive antihypertensive drugs

which may have led to her subsequent stroke

Probability of Causation: 75%

The absolute risk of obtaining the stroke increased from 1 in 13,937 to 1 in 11,315

Court concluded that on the basis of such a low probability no liability could be inferred

2 24 13 [email protected] Linking a Potential Causal Factor to Injury

Negative Positive Health Outcome Health Outcome

Potential cause 10 (a) 20 (b) present (exposed) Potential cause 7 (c) 34 (d) absent (not exposed)

[email protected] 2 24 14 The 2x2 contingency table

The incidence of the negative health outcome when the potential

cause is present (I1) equals a/(a+b) 10/30=0.30

The incidence of the negative outcome when the potential cause is

absent (I0) equals c/(c+d) 7/41=0.17

Negative Positive Health Outcome Health Outcome

Potential cause present (exposed) 10 (a) 20 (b)

Potential cause absent (not exposed) 7 (c) 34 (d)

[email protected] 2 24 15 The Risk Difference

I1-I0=0.30-0.17=0.13 or 13%

Negative Positive Health Outcome Health Outcome

Potential cause present (exposed) 10 (a) 20 (b)

Potential cause absent (not exposed) 7 (c) 34 (d)

[email protected] 2 24 16 The Relative Risk

I1/I0=0.30/0.17=1.76

Negative Positive Health Outcome Health Outcome

Potential cause present (exposed) 10 (a) 20 (b)

Potential cause absent (not exposed) 7 (c) 34 (d)

[email protected] 2 24 17 The Comparative Risk

a derivative of the RR

aims to compare individual probabilities instead of group risks

I1/I0 -> P1/P0

[email protected] 2 24 18 A Comparative Risk where RR equals the CRR

frequency of serious injury in 100 randomly selected unrestrained drivers exposed to a 20 mph frontal collision, eg. 15%

frequency of serious injury in 100 randomly selected restrained drivers exposed to the same collision severity and type, eg. 5%

CRR is 0.15/0.05=3

[email protected] 2 24 19 A Comparative Risk where RR does not equal the CRR:

Example when the numerator of the CRR is a per-event risk, and the denominator is a per-time risk

Patient with a history of deep venous thrombosis (DVT)

Pulmonary embolism (PE) occurs a week after a patient sustained a lower extremity fracture in a crash

What caused the PE?

Compare Probability of PE given a lower extremity fracture (a per event rate)] 1-week risk of PE in a patient of with DVT (a time dependent probability

2 24 20 [email protected] The Attributable Proportion under the Exposed (APe) or the Probability of Causation (PC)

(I1/I0)/I1 = (RR-1)/RR = 0.43%

Negative Positive Health Outcome Health Outcome

Potential cause present (exposed) 10 (a) 20 (b)

Potential cause absent (not exposed) 7 (c) 34 (d)

[email protected] 2 24 22 Higher court toxic tort (CO4/303HR)

Employer was held liable for the lung cancer of an employee

Employee who was potentially exposed to asbestos but also a life-long smoker

The APe for asbestos exposure was 55% The court therefore concluded that the employer was for 55% liable

The court ignored that the APe for smoking is 80%

2 24 23 [email protected] 3

Examples Health Risk Estimation Health Risk Estimation Example: Car crash

35 The case facts

An unrestrained 35 year-old man was traveling on a highway on a winter evening in a 2008 Nissan Altima sedan. His vehicle struck the rear of a slow moving large truck at highway speed in the right lane, causing the airbags in the vehicle to deploy and disabling the vehicle. Within approximately 30 seconds following the first collision the sedan was struck from behind by a semi- tractor trailer traveling at highway speed. The man was pronounced dead at the scene, and later examination revealed extensive skull fractures and brain and spinal cord disruptions, along with severe chest, abdomen, and spine injuries.

3 20 2 [email protected] Health Risk Estimation Example: Car crash

37 The question for the Forensic Epidemiologist

Which of the 2 crashes was the cause of the death? How certain can you be of this?

3 20 4 [email protected] The Comparative Risk Estimate

p (death | first crash) CRR = p (death | second crash)

[email protected] 3 20 5 The Comparative Risk Estimate

p (death | first crash) CRR = 1

[email protected] 3 20 6 P(serious injury) | first crash

In a 17-year span of investigated crashes (NASS- CDS: 1995-2011) there were an estimated 209,760 unrestrained drivers exposed to frontal collisions with an airbag deployment in the range reconstructed for the first collision (24-48 km/h) The number of drivers with serious and greater injury was 11,108, or 5.3% (1 in 19)

3 20 7 [email protected] P(serious injury) | first crash

1 in 19 19 CRR = = 1 1

[email protected] 3 20 8 P(death) | first crash

In a 17-year span of investigated crashes (NASS- CDS: 1995-2011) there were an estimated 209,760 unrestrained drivers exposed to frontal collisions with an airbag deployment in the range reconstructed for the first collision (24-48 km/h) The number of deaths was 181, or 0.09% (1 in 1,159)

3 20 9 [email protected] 4

Theory Specific Causality Overview: Specific Causality

Methods to determine specific causation Spurious methods and fallacies Hill criteria

Multiple competing causes Moderation

4 17 1 [email protected] Spurious methods and fallacies

Consensus panel

Appeal to authority

Post hoc ergo propter hoc fallacy

Fallacy of the transposed conditional

4 17 4 [email protected] Fallacy of the transposed conditional

The vehicle damage was minimal, and therefore the injury risk from the crash was corresponding low, and therefore it is unlikely the claimed injuries resulted from the crash

4 17 5 [email protected] Hill Criteria

Strength of association Consistency Specificity Temporality Biological gradient Biological plausibility Coherence Experiment Analogy

4 17 6 [email protected] Strength of Association

Most important determinant of causation

Often operationalized as an RR, CRR or PC ‘doubling of the risk’ -> RR>2 ‘on the balance of probabilities’ -> CRR # to 1 ‘more likely then not’ -> PC>50% ‘beyond the benefit of the doubt’ -> PC>90%

Closely associated with proximity, either temporal or spacial

4 17 7 [email protected] Consistency

The repetitive observation of a causal relationship in different circumstances (itrio) Different investigators Different settings and circumstances Different populations and subgroups Different study designs

4 17 8 [email protected] Specificity

The degree to which an outcome is associated with a particular exposure Seatbelt mark in Mesothelioma is only associated with asbestos exposure GSW to head

Rare attribute, as most exposures can cause various diseases or injuries

4 17 9 [email protected] Temporality

Sequence: Exposure must precede disease (cohort study is good for this)

4 17 10 [email protected] Temporality

Temporal plausibility: The outcome may not occur before or after the effect range of the hazard

Some foodborne illnesses (i.e. campylo- bacteriosis) only manifest after a matter of hours or days of incubation

Death of a patient occurring 3 days after receiving an injection of a short- acting opiate (i.e. hydro- morphone) is not plausibly related to the injection

Lung cancer that occurs within 2 years after asbestos exposure is less likely to be caused by it (latency period is 10-20 years)

4 17 11 [email protected] Temporality

Temporal latency: the latency time between the exposure and the first indication of disease or injury

A death in a hospital patient that occurs within 20 min of an injection of hydromorphone is much more likely to be associated with the injection than one that occurs 3 h later

4 17 12 [email protected] Biological gradient

The outcome increases monotonically with increase dose of exposure (dose-response) intro

Most relevance in specific causation assessments of adverse drug reactions and exposure to toxic substances

4 17 13 [email protected] Biological plausibility

The observed association can be plausibly explained by known scientific principles

A common error with plausibility assessments is to transpose low pre-event probability of injury and implausibility of injury

4 17 14 [email protected] Coherence

A causal conclusion should not fundamentally contradict present substantive knowledge

It should ‘make sense’ given current knowledge

Extension of plausibility concept

4 17 15 [email protected] Experiment

Evidence from randomized experiments on animals or humans

Absence of experimental evidence of an injury or disease mechanism should not be confused with evidence against an investigated causal relationship

4 17 16 [email protected] Analogy

An analogous exposure and outcome may be translatable to the circumstances of a previously unexplored causal investigation

4 17 17 [email protected] 5

Examples Specific Causality Multiple competing causes

Confounding Moderation

5 9 5 [email protected] Confounding

5 9 6 [email protected] Confounding example

Dietary energy intake is a confounder in the relation between exercise and diabetes mellitus risk

Mileage of use is a confounder in the relation between type of jogging shoe and shock absorbance.

5 9 7 [email protected] Moderation (effect modification, interaction)

5 9 8 [email protected] Moderation examples

Seat belt use is an effect modifier of the relation between driving speed and amount of injury of the driver after a car crash.

For asbestos-related occupational diseases such as asbestosis and mesothelioma, smoking can be an effect modifier. The effect of asbestos exposure is stronger when paired with smoking exposure.

5 9 9 [email protected] 8

Concluding remarks Take home message

Risks can be quantified by epidemiologic study

Recognize the improper use of absolute risk as a measure of causation

Understand that only Comparative Risk assesses causation

Customize direct and cross to incorporate these concepts

8 1 1 [email protected]