Introduction: Causality assessment is a critical process in pharmacovigilance aimed at determining whether a specific drug is responsible for an observed adverse drug reaction (ADR). It helps in identifying the strength of the relationship between drug exposure and the adverse event. Causality assessment is complex, as multiple factors like the patient’s condition, co-administered drugs, or underlying diseases can contribute to the event. The primary goal of causality assessment is to ensure accurate drug safety monitoring and informed decision-making in clinical practice and regulatory affairs.
Several established methods are used to assess causality, including both qualitative and quantitative approaches. Each method has strengths and limitations, depending on the clinical context and available data.
WHO-UMC (World Health Organization-Uppsala Monitoring Centre) Causality Assessment System
The WHO-UMC Causality Assessment System is a globally recognized framework used to determine the likelihood that a drug caused an adverse drug reaction (ADR). It is an essential tool in pharmacovigilance, helping regulatory agencies, healthcare professionals, and pharmaceutical companies assess and monitor drug safety.
Causality Categories and Criteria
1. Certain: A reaction is classified as “Certain” when there is strong evidence linking the adverse event to the drug, with no other plausible explanations.
Criteria:
- A plausible time relationship between drug intake and ADR.
- The reaction cannot be explained by another disease or drug.
- A positive dechallenge (improvement after stopping the drug).
- A positive rechallenge (reaction reoccurs upon re-administration, if ethically permissible).
Example: A patient taking penicillin develops anaphylaxis (severe allergic reaction).
- Symptoms appear within minutes of administration.
- Symptoms resolve after stopping the drug.
- The patient had a previous documented allergic reaction to penicillin.
- No other drugs or conditions explain the reaction.
Causality: Certain
2. Probable / Likely: A reaction is classified as “Probable/Likely” when the drug is the most likely cause, but rechallenge is not necessary.
Criteria:
- A reasonable time relationship to drug intake.
- The reaction is unlikely to be explained by another disease or drug.
- A positive dechallenge (improvement after stopping the drug).
- Rechallenge is not required.
Example: A patient taking ibuprofen for pain relief develops gastric ulcers after a few weeks.
- Ulcer symptoms improve after stopping ibuprofen.
- The patient had no prior history of ulcers.
- No other medications or medical conditions explain the ulcer.
Causality: Probable/Likely
3. Possible: A reaction is classified as “Possible” when a drug-related cause cannot be ruled out, but there are alternative explanations.
Criteria:
- A reasonable time relationship to drug intake.
- Could also be explained by another disease or drug.
- Information on drug withdrawal is unclear or lacking.
Example: A patient on metformin develops nausea and vomiting.
- Symptoms start after metformin use.
- The patient also has gastroenteritis (infection causing nausea and vomiting).
- Symptoms improve, but it is unclear if stopping metformin helped.
Causality: Possible
4. Unlikely: A reaction is classified as “Unlikely” when there is little evidence to support a causal relationship.
Criteria:
- An inconsistent time relationship with drug intake.
- A more likely alternative explanation exists.
Example: A patient on paracetamol (acetaminophen) develops a skin rash two weeks after stopping the drug.
- The time gap is too long to be linked to paracetamol.
- The patient had started another new drug recently.
- A more probable cause (e.g., an allergy to the new drug) exists.
Causality: Unlikely
5. Conditional / Unclassified: A reaction is classified as “Conditional/Unclassified” when there is some evidence, but additional data is required before making a conclusion.
Criteria:
- More data is needed to confirm or reject a causal relationship.
Example: A patient taking a new experimental drug in a clinical trial develops severe headaches.
- The drug has not been studied extensively for this effect.
- No other causes are identified, but more research is needed.
Causality: Conditional/Unclassified
6. Unassessable / Unclassifiable: A reaction is classified as “Unassessable/Unclassifiable” when the available information is insufficient or contradictory.
Criteria:
- Incomplete or inconclusive medical records.
- Conflicting evidence makes assessment impossible.
Example: A patient on multiple medications reports fatigue but does not provide details about dosage or duration.
- No clear pattern is found.
- Other conditions could explain the fatigue.
- Insufficient data to assess causality.
Causality: Unassessable/Unclassifiable
Importance of WHO-UMC Causality Assessment in Pharmacovigilance
Standardization: Ensures a uniform approach for evaluating ADRs globally.
Decision Making: Helps regulatory bodies assess drug safety.
Signal Detection: Identifies emerging safety concerns about drugs.
Patient Safety: Assists healthcare professionals in optimizing medication use.
Strengths:
- Simple and easy to use.
- Provides clear categories for classification.
- Widely accepted and used globally by regulatory bodies.
Limitations:
- Subjective, as it relies on clinical judgment.
- Does not incorporate a quantitative probability score.
2. Naranjo Algorithm for Causality Assessment of Adverse Drug Reactions (ADR)
The Naranjo Algorithm, developed in 1981 by Naranjo et al., is a structured, questionnaire-based tool used to assess the probability that a drug caused an adverse drug reaction (ADR). It is widely used in pharmacovigilance and clinical research to standardize causality assessment.
Structure of the Naranjo Algorithm
The Naranjo algorithm consists of 10 questions, each with a score of +1, 0, or -1 based on objective criteria. The total score determines the likelihood of causality.
| Question | Yes (+1) | No (0) | Don’t know (0) |
| 1. Are there previous conclusive reports on this reaction? | +1 | 0 | 0 |
| 2. Did the ADR appear after the suspected drug was administered? | +2 | -1 | 0 |
| 3. Did the ADR improve when the drug was discontinued (dechallenge)? | +1 | 0 | 0 |
| 4. Did the ADR reappear when the drug was readministered (rechallenge)? | +2 | -1 | 0 |
| 5. Are there alternative causes that could have caused the reaction? | -1 | +2 | 0 |
| 6. Did the ADR appear with a placebo? | -1 | 0 | 0 |
| 7. Was the drug detected in blood (therapeutic levels)? | +1 | 0 | 0 |
| 8. Was the ADR dose-dependent (i.e., higher dose = stronger reaction)? | +1 | 0 | 0 |
| 9. Did the patient have a similar reaction to this drug before? | +1 | 0 | 0 |
| 10. Was the ADR confirmed by objective evidence (e.g., biopsy, lab test)? | +1 | 0 | 0 |
Causality Categories and Scoring
The total Naranjo Score is used to classify the ADR into one of four categories:
| Total Score | Causality Classification | Interpretation |
| ≥ 9 | Definite | Strong evidence that the drug caused the ADR. |
| 5 – 8 | Probable | ADR is likely due to the drug, but other causes are possible. |
| 1 – 4 | Possible | The ADR may be due to the drug, but alternative explanations exist. |
| ≤ 0 | Doubtful | No strong evidence linking the ADR to the drug. |
Example Cases Using the Naranjo Algorithm
Example 1: Penicillin-Induced Anaphylaxis
Revised Example 1: Penicillin-Induced Anaphylaxis
A 30-year-old patient is given penicillin and develops severe anaphylaxis within minutes.
| Naranjo Question | Response | Score |
| 1. Are there previous conclusive reports on this reaction? | Yes | +1 |
| 2. Did the ADR appear after the suspected drug was administered? | Yes | +2 |
| 3. Did the ADR improve when the drug was discontinued (dechallenge)? | Yes | +1 |
| 4. Did the ADR reappear when the drug was readministered (rechallenge)? | Yes | +2 |
| 5. Are there alternative causes (other than the drug) that could have caused the reaction? | No | +2 |
| 6. Did the ADR appear with a placebo? | No | 0 |
| 7. Was the drug detected in blood (evidence of drug presence)? | Not tested | 0 |
| 8. Was the ADR dose-dependent (higher dose = stronger reaction)? | No | 0 |
| 9. Did the patient have a similar reaction to this drug before? | Yes | +1 |
| 10. Was the ADR confirmed by objective evidence? | Yes | +1 |
Total Score = 1 + 2 + 1 + 2 + 2 + 0 + 0 + 0 + 1 + 1 = 10
Correct Classification: Definite ADR (Score ≥9)
Example 2: Ibuprofen-Induced Gastric Ulcer
A 50-year-old patient develops gastric ulcers after 2 months on ibuprofen for arthritis.
| Naranjo Question | Response | Score |
| 1. Are there previous conclusive reports on this reaction? | Yes | +1 |
| 2. Did the ADR appear after the suspected drug was administered? | Yes | +2 |
| 3. Did the ADR improve when the drug was discontinued (dechallenge)? | Yes | +1 |
| 4. Did the ADR reappear when the drug was readministered (rechallenge)? | Not tested | 0 |
| 5. Are there alternative causes (e.g., alcohol, smoking) that could have caused the reaction? | Yes (partially) | -1 |
| 6. Did the ADR appear with a placebo? | No | 0 |
| 7. Was the drug detected in blood (evidence of drug presence)? | Not tested | 0 |
| 8. Was the ADR dose-dependent (higher dose = stronger reaction)? | Yes | +1 |
| 9. Did the patient have a similar reaction to this drug before? | Yes | +1 |
| 10. Was the ADR confirmed by objective evidence (e.g., endoscopy)? | Yes | +1 |
Total Score = 1 + 2 + 1 + 0 1 + 0 + 0 + 1 + 1 + 1 = 6
Correct Classification: Probable ADR (Score 5–8)
Example 3: Metformin-Induced Nausea
A 55-year-old patient starts metformin for type 2 diabetes and develops nausea within 2 days.
| Naranjo Question | Response | Score |
| 1. Are there previous conclusive reports on this reaction? | Yes | +1 |
| 2. Did the ADR appear after the suspected drug was administered? | Yes | +2 |
| 3. Did the ADR improve when the drug was discontinued (dechallenge)? | Not tested | 0 |
| 4. Did the ADR reappear when the drug was readministered (rechallenge)? | Not tested | 0 |
| 5. Are there alternative causes (e.g., diet, infection) that could have caused the reaction? | Yes (partially) | -1 |
| 6. Did the ADR appear with a placebo? | No | 0 |
| 7. Was the drug detected in blood (evidence of drug presence)? | Not tested | 0 |
| 8. Was the ADR dose-dependent (higher dose = stronger reaction)? | Not tested | 0 |
| 9. Did the patient have a similar reaction to this drug before? | No | 0 |
| 10. Was the ADR confirmed by objective evidence? | No | 0 |
Total Score = 1 + 2 + 0 + 0 1 + 0 + 0 + 0 + 0 + 0 = 3
Correct Classification: Possible ADR (Score 1–4)
Advantages of the Naranjo Algorithm
Standardized approach: Reduces subjectivity in ADR assessment.
Easy to use: Simple scoring system with objective questions.
Widely accepted: Used by regulatory authorities (FDA, WHO, EMA).
Limitations of the Naranjo Algorithm
Does not apply well to drug-drug interactions. Cannot assess delayed ADRs (e.g., cancer due to long-term drug use). Relies on patient history and rechallenge, which may not be ethical.
Comparison: Naranjo Algorithm vs. WHO-UMC Causality System
| Feature | Naranjo Algorithm | WHO-UMC System |
| Approach | Questionnaire-based | Expert judgment-based |
| Scoring | Numeric (0–13) | Category-based (Certain, Probable, Possible, etc.) |
| Subjectivity | Less subjective | More subjective |
| Rechallenge Importance | Essential for high scores | Not always required |
| Use Case | Clinical trials, case reports | Regulatory pharmacovigilance |
Conclusion
The Naranjo Algorithm is a useful, structured method for determining whether a drug caused an adverse reaction. It is particularly effective for clinical case evaluations and research, whereas the WHO-UMC system is more suited for regulatory decision-making.
Strengths:
- Standardized scoring system makes it objective.
- Easy to apply, especially in clinical settings.
- Provides a probability score, aiding in decision-making.
Limitations:
- Not specific for all drug types and conditions.
- Lacks sensitivity in detecting complex interactions, such as those involving multiple drugs (polypharmacy).
- Relies on rechallenge and dechallenge, which may not be ethical or feasible in all situations.
3. Bradford Hill Criteria
The Bradford Hill Criteria are a set of nine principles that are used to determine whether an observed association between an exposure and an outcome is causal. These criteria are essential in epidemiology, particularly in the context of determining causality in public health studies. Proposed by Sir Austin Bradford Hill in 1965, these guidelines help to provide a structured approach for evaluating causal relationships in epidemiological research.
Here are the nine Bradford Hill Criteria:
1. Strength of the Association: This refers to how strongly the exposure is connected to the outcome. The stronger the connection, the more likely it is that the exposure causes the outcome. A larger effect makes us more sure that the relationship is causal.
Example: Smoking and lung cancer have a strong association. Studies show that people who smoke are significantly more likely to develop lung cancer compared to non-smokers. This strong association supports a causal relationship.
2. Consistency (Reproducibility): The connection should be the same when studied in different groups of people, by different researchers, and in various situations. If similar results are seen in several studies, the connection is more likely to be causal.
Example: The association between smoking and lung cancer has been observed consistently across different countries, in both men and women, and in various age groups, supporting a causal relationship.
3. Specificity: The exposure should be connected to a particular disease or outcome, not many different conditions. Although specificity isn’t always required, the more specific the connection, the more likely it is to be causal.
Example: The relationship between Helicobacter pylori infection and peptic ulcer disease is highly specific. H. pylori is primarily linked to this condition, making it easier to argue for a causal relationship.
4. Temporality: This is likely the most important factor. The cause must come before the effect. In other words, the exposure should happen before the outcome develops.
Example: In studies of HIV and AIDS, the temporal sequence is clear: exposure to HIV precedes the development of AIDS, which supports the causal relationship.
5. Biological Gradient (Dose-Response Relationship): A dose-response relationship is often an indicator of causality. If increasing exposure leads to an increasing incidence or severity of the disease, the association is more likely to be causal.
Example: The more a person smokes, the higher the risk of developing lung cancer. A dose-response relationship exists, where heavy smokers are at a significantly higher risk than light smokers, providing evidence of causality.
6. Plausibility: There should be a biologically plausible mechanism to explain how the exposure leads to the outcome. This is supported by existing knowledge of biology, physiology, and pathology.
Example: The association between asbestos (Asbestos is a group of naturally occurring silicate minerals composed of long, thin fibres, with different shapes and colours.) exposure and lung cancer is biologically plausible, as asbestos fibers can cause lung tissue irritation and inflammation, leading to cancerous mutations in cells.
7. Coherence: The association should not conflict with the known facts of biology, natural history, and epidemiology. The findings should fit into the established body of scientific knowledge.
Example: The link between alcohol consumption and liver cirrhosis is coherent with biological knowledge. Alcohol is metabolized in the liver and excessive consumption leads to liver damage, supporting the causal relationship.
8. Experiment: If a study can be conducted experimentally, and the exposure can be manipulated to observe its effects, this adds evidence to support causality. While not always feasible, randomized controlled trials (RCTs) are the gold standard for experimental evidence.
Example: The use of randomized controlled trials to assess the effectiveness of aspirin in preventing heart attacks supports a causal relationship between aspirin use and a reduction in cardiovascular events.
9. Analogy: If a similar relationship has been established with another exposure or condition, it can lend support to the hypothesis of causality. The analogy is not proof but can strengthen the case for a causal relationship.
Example: The relationship between thalidomide (a sedative) and birth defects is analogous to other teratogens like alcohol or rubella, which have also been linked to birth defects. This analogy strengthens the case for a causal relationship between thalidomide and birth defects.
It is important to note that these criteria are not rigid rules but are intended as guidelines to help researchers make informed judgments about causality. They are often used together to provide a comprehensive picture of the relationship between an exposure and an outcome.
Strengths:
- Comprehensive and well-established in medical research.
- Encourages consideration of multiple factors when assessing causality.
Limitations:
- Not all criteria are necessary or sufficient for establishing causality.
- Can be challenging to apply in individual patient cases (often used in population-level assessments).
4. RUCAM (Roussel Uclaf Causality Assessment Method) for Liver Injury
The Roussel Uclaf Causality Assessment Method (RUCAM) is a structured and widely used tool for assessing the likelihood that a drug or herbal product caused liver injury. It is primarily applied in cases of drug-induced liver injury (DILI) and herb-induced liver injury (HILI).
Components of RUCAM
RUCAM assigns scores based on different clinical and laboratory criteria, with a total score determining the probability of causality. The key components include:
- Time to Onset (Latency)
- Time from drug intake to liver injury onset.
- Shorter latency periods (5-90 days) get higher scores.
- Course of Alanine Aminotransferase (ALT) or Alkaline Phosphatase (ALP) After Drug Withdrawal
- If ALT or ALP levels decrease by 50% within a set time, it supports causality.
- Risk Factors
- Age (>55 years)
- Alcohol use (>2 drinks/day)
- Concomitant Drugs
- Other drugs that might contribute to liver injury.
- If another drug is more likely to cause liver injury, it lowers the causality score.
- Non-Drug Causes
- Hepatitis A, B, C, and other liver diseases.
- Exclusion of these conditions strengthens causality.
- Previous Information on the Drug
- Whether the drug has a known hepatotoxicity profile.
- Response to Re-Challenge
- If liver injury recurs upon re-administration, it strongly supports causality.
RUCAM Scoring System
The total score determines the causality category:
- ≤0: Excluded
- 1–2: Unlikely
- 3–5: Possible
- 6–8: Probable
- ≥9: Highly probable
Importance of RUCAM
- Provides a standardized method for evaluating hepatotoxicity.
- Helps distinguish DILI/HILI from other liver diseases.
- Used in clinical trials, regulatory assessments, and pharmacovigilance.
Strengths:
- Specifically designed for liver injury, making it highly relevant for hepatotoxic drugs.
- Quantitative scoring system helps in objectivity.
Limitations:
- Only applicable to drug-induced liver injury (DILI), not other types of ADRs.
- Requires detailed liver function test data, which may not always be available.
5. Bayesian Approaches
Bayesian methods offer a probabilistic framework for causality assessment by incorporating prior knowledge, new evidence, and updating beliefs using Bayes’ theorem. This approach is particularly useful in drug-induced liver injury (DILI), herb-induced liver injury (HILI), and pharmacovigilance.
1. Bayesian Framework for Causality Assessment
Bayesian approaches assess causality by computing the posterior probability that a drug caused an adverse event, given observed data and prior knowledge.
Bayes’ Theorem
Where:
- P(H∣D) = Posterior probability (Updated probability that the drug caused the event)
- P(D∣H) = Likelihood (Probability of the observed data given the hypothesis)
- P(H) = Prior probability (Previous belief about causality before new evidence)
- P(D) = Total probability of the observed data (Normalizing factor)
2. Key Components in Bayesian Causality Assessment
(A) Prior Probability P(H)
- Derived from previous knowledge (clinical trials, epidemiological data, past reports).
- Example: If a drug has been linked to liver injury in past studies, its prior probability is higher.
(B) Likelihood P(D∣H)
- Based on observed data, such as:
- Temporal relationship (Latency period)
- De-challenge and re-challenge results
- Biochemical markers (ALT, AST, ALP)
- Alternative explanations (viral hepatitis, alcohol, comorbidities)
- Computed using probabilistic models (e.g., logistic regression, Markov models).
(C) Posterior Probability P(H∣D)
- Represents the updated belief in causality after considering new evidence.
- Helps categorize cases as:
- Definite
- Probable
- Possible
- Unlikely
3. Applications of Bayesian Methods in Causality Assessment
(A) Bayesian RUCAM (B-RUCAM)
- Combines RUCAM scores with Bayesian probabilities.
- Improves sensitivity by integrating real-world data and uncertainty quantification.
(B) Bayesian Networks
- Graphical models that show relationships between multiple risk factors.
- Example: A Bayesian network for DILI can model drug exposure, genetic susceptibility, liver function tests, and comorbidities.
(C) Pharmacovigilance & Signal Detection
- Bayesian data mining techniques (e.g., Bayesian Confidence Propagation Neural Networks BCPNN) analyze adverse event databases (FAERS, VigiBase).
- Helps identify new safety signals for hepatotoxic drugs.
(D) Personalized Medicine
- Bayesian models adjust causality assessment based on individual patient risk factors (e.g., genetics, metabolic profile).
- Helps predict high-risk populations for drug toxicity.
4. Advantages of Bayesian Causality Assessment
Quantifies Uncertainty: Provides probability estimates rather than binary decisions.
Incorporates Prior Knowledge: Uses past data and real-world evidence.
Handles Missing Data: Can integrate incomplete datasets.
Dynamic Updating: Adjusts causality assessment as new evidence emerges.
5. Limitations & Challenges
Requires high-quality prior data (poor priors can bias results). Computationally complex compared to rule-based methods (e.g., RUCAM). Depends on accurate probability estimations for different causality factors.
6. Modified Karch and Lasagna Algorithm
The Modified Karch and Lasagna Algorithm is another structured approach used for assessing drug causality. It evaluates several factors that contribute to the likelihood of causality, including timing, biological plausibility, and alternative explanations.
Key Factors:
- Timing of drug administration relative to ADR onset.
- Biological plausibility of the ADR.
- Known adverse effect profile of the drug.
- Presence of confounding factors (e.g., other diseases or drugs).
- Outcome after drug discontinuation (dechallenge).
Strengths:
- Structured and systematic approach.
- Widely used in clinical and research settings.
Limitations:
- Relies heavily on clinical judgment, making it less objective than purely quantitative methods.
- Does not always account for genetic or patient-specific factors.
7. French Imputability Method
The French Imputability Method combines chronological criteria, semiological (clinical) criteria, and bibliographical criteria to assess the likelihood of a drug causing an ADR.
Key Aspects:
Chronological Imputability: Evaluates the timing of the drug in relation to the ADR.
Semiological Imputability: Assesses the clinical presentation of the ADR and whether it fits with known drug reactions.
Bibliographical Imputability: Looks at existing literature on the ADR to support or refute causality.
Strengths:
- Comprehensive, incorporating multiple dimensions of causality.
- Used widely in European pharmacovigilance systems.
Limitations:
- Some aspects, such as semiological criteria, may be subjective.
- Time-consuming due to the detailed review required.
8. WHO’s Causality Assessment in Clinical Trials
For causality assessment in clinical trials, the WHO recommends a causality assessment framework that considers factors such as:
- Temporal relationship between drug administration and ADR onset.
- The probability of alternative explanations (comorbidities, concomitant drugs).
- The consistency of the ADR with known effects of the drug or class of drugs.
- Biological plausibility and mechanism of action.
This approach emphasizes rigorous, protocol-driven assessment within the controlled environment of clinical trials.
Strengths:
- High degree of control over variables, making causality easier to assess.
- Detailed data available from trial protocols, including dosing, patient demographics, and monitoring.
Limitations:
- Limited applicability to real-world settings where patient populations are more diverse.
- Challenges in capturing long-term ADRs due to limited trial duration.
Conclusion
Causality assessment methods vary from subjective clinical judgment to more objective, structured algorithms. Each method has its advantages and limitations, depending on the ADR, drug, patient, and clinical context. The choice of method depends on factors such as the type of reaction, available data, and the healthcare setting. While no single method is perfect, using a combination of methods improves the accuracy and reliability of causality assessment, contributing to safer drug use and better patient outcomes.