5.3 Epidemiological Study Designs and Measures of Association

Epidemiological Study Designs and Measures of Association

Occupational epidemiology is the study of the distribution and determinants of disease, injury, and mortality in working populations. While industrial hygiene focuses on identifying, evaluating, and controlling workplace hazards, occupational epidemiology provides the empirical human evidence used to establish Occupational Exposure Limits (OELs), validate toxicological dose-response models, and substantiate causal relationships between workplace exposures and chronic occupational illnesses.

---

1. Fundamental Epidemiological Study Designs

Epidemiological studies are categorized into experimental designs (e.g., randomized controlled trials, rarely ethical in occupational exposure settings) and observational designs, which observe workers in their natural employment environments.

EPIDEMIOLOGICAL STUDY DESIGNS IN OCCUPATIONAL HEALTH
│
├── Cohort Studies (Exposed vs. Unexposed → Followed over Time → Disease Incidence)
│   ├── Prospective (Concurrent) Cohort: Followed forward from present into future
│   └── Retrospective (Historical) Cohort: Reconstructed backward using employment records
│
├── Case-Control Studies (Diseased [Cases] vs. Non-Diseased [Controls] → Past Exposure History)
│
└── Cross-Sectional Studies (Snapshot in Time → Simultaneous Exposure & Disease Prevalence)

Comparison of Major Observational Study Designs

---

2. Measures of Association: Mathematical Formulations

Epidemiological measures of association quantify the strength of the statistical relationship between an occupational exposure and a health outcome. All calculations derive from the standard 2 × 2 contingency table.

The Standard 2 × 2 Contingency Table

---

Relative Risk (Risk Ratio, RR)

Used strictly in cohort studies, the Relative Risk (RR) is the ratio of the cumulative disease incidence in the exposed group (Ie) to the cumulative disease incidence in the unexposed group (Iu):

$I_e = \frac{a}{a + b}, \qquad I_u = \frac{c}{c + d}$

$\mathbf{RR = \frac{I_e}{I_u} = \frac{\frac{a}{a + b}}{\frac{c}{c + d}} = \frac{a \cdot (c + d)}{c \cdot (a + b)}}$

• Interpretation:
  • RR = 1.0: No association between exposure and disease.
  • RR > 1.0: Positive association (exposure increases disease risk; e.g., RR = 2.5 indicates exposed workers have 2.5 times the risk of unexposed workers).
  • RR < 1.0: Negative/protective association (exposure reduces disease risk).

Attributable Risk (AR) & Attributable Fraction (AR%)

• Attributable Risk (Risk Difference, AR): The excess disease incidence directly attributable to the occupational exposure: $\mathbf{AR = I_e - I_u = \frac{a}{a + b} - \frac{c}{c + d}}$
• Attributable Risk Percent (Etiologic Fraction in Exposed, AR%): The proportion of disease cases among the exposed that would be eliminated if the occupational exposure were completely removed: $\mathbf{AR\% = \left( \frac{I_e - I_u}{I_e} \right) \times 100\% = \left( \frac{RR - 1}{RR} \right) \times 100\%}$

---

Odds Ratio (OR)

In case-control studies, total population counts (a+b and c+d) are unknown because the investigator chooses the number of cases and controls. Therefore, true incidence cannot be computed. Instead, the Odds Ratio (OR) is calculated:

$\text{Odds of exposure in Cases} = \frac{a}{c}, \qquad \text{Odds of exposure in Controls} = \frac{b}{d}$

$\mathbf{OR = \frac{a / c}{b / d} = \frac{a \cdot d}{b \cdot c}}$

The Rare Disease Assumption

When the disease under investigation is biologically rare in both exposed and unexposed populations (a << b and c << d, such that a + b ≈ b and c + d ≈ d), the Odds Ratio mathematically converges to the Relative Risk:

$RR = \frac{a / (a + b)}{c / (c + d)} \approx \frac{a / b}{c / d} = \frac{a \cdot d}{b \cdot c} = OR$

Rule for the CIH Exam: If a disease has a cumulative incidence of less than ≈ 5% to 10% in the study population, the Odds Ratio (OR) derived from a case-control study is a valid mathematical surrogate for the Relative Risk (RR).

---

3. Standardized Mortality Ratio (SMR) and Indirect Standardization

In occupational cohort studies, researchers frequently compare the observed mortality experience of an industrial workforce against the mortality rates of the general population (e.g., national or state vital statistics).

Because the general population includes infants, the elderly, and disabled individuals, simple crude mortality comparisons are invalid. Indirect standardization adjusts for demographic differences in age, sex, race, and calendar time.

Calculating the Standardized Mortality Ratio (SMR)

$\mathbf{SMR = \left( \frac{\text{Observed Number of Deaths } (O)}{\text{Expected Number of Deaths } (E)} \right) \times 100}$

Where the Expected Number of Deaths (E) is calculated by summing across age- and calendar-strata:

$E = \sum_{i=1}^{k} \left( \text{Person-Years of Follow-Up in Stratum } i \right) \times \left( \text{Standard Mortality Rate in Stratum } i \right)$

• Interpretation of SMR:
  • SMR = 100: Observed cohort mortality matches general population expectation.
  • SMR > 100: Excess mortality in the cohort (e.g., SMR = 165 indicates a 65% excess mortality risk above general population baseline).
  • SMR < 100: Cohort experienced lower mortality than the general population.

(Note: When evaluating disease incidence rather than mortality, the exact same mathematical formulation is called the Standardized Incidence Ratio, SIR).

---

4. Threats to Study Validity: Confounding vs. Effect Modification

Epidemiological investigations must distinguish between artifactual statistical distortions and genuine biological phenomena.

CONFOUNDING TRIANGLE
         Confounder (Z)
        /              \
       / (Associated)   \ (Independent Risk Factor)
      v                  v
Exposure (E) --------> Disease (D)
       (True Causal Path?)

Confounding Variables

Confounding is a mixing of effects where the apparent association between an occupational exposure (E) and a disease (D) is distorted by an extraneous third variable (Z).

Three Strict Criteria for a Variable to Be a Confounder:

1. It must be an independent risk factor for the disease outcome among unexposed individuals.
2. It must be statistically associated with the exposure in the study population.
3. It must NOT be an intermediate step on the causal biological pathway between exposure and disease (E → Z → D).

Classic Occupational Example: In an epidemiological study evaluating lung cancer risk among asbestos-exposed insulators, cigarette smoking is a major confounder. Smoking is an independent cause of lung cancer and may be correlated with employment history, distorting the crude risk estimate if not controlled.

Methods to Control Confounding:

• In Study Design: Restriction (e.g., studying only non-smokers), Matching (matching cases and controls on age/smoking status).
• In Data Analysis: Stratification (Cochran-Mantel-Haenszel pooling across strata), Multivariable Regression (multiple logistic regression, Cox proportional hazards modeling).

Effect Modification (Interaction)

Effect Modification (Interaction) is a real biological phenomenon where the magnitude of the exposure-disease association differs across strata of a third variable.

• Confounding vs. Effect Modification: Confounding is a bias that the researcher must eliminate or adjust for. Effect modification is a biological reality that must be reported and described.
• Classic Occupational Example: The synergistic, supra-multiplicative interaction between asbestos exposure and cigarette smoking on bronchogenic lung cancer:
  • Baseline unexposed non-smokers: RR = 1.0
  • Asbestos exposure alone (non-smokers): RR ≈ 5.0
  • Cigarette smoking alone (unexposed to asbestos): RR ≈ 10.0
  • Combined asbestos exposure AND cigarette smoking: RR ≈ 50.0 (multiplicative synergistic effect, not merely additive 5 + 10 = 15).

---

5. Systematic Biases in Occupational Epidemiology

Bias is any systematic error in the design, conduct, or analysis of a study that results in an erroneous estimate of the exposure-disease association.

1. The Healthy Worker Effect (HWE)

The Healthy Worker Effect is the single most pervasive selection bias in occupational health studies.

• Mechanism: To enter and maintain active employment, individuals must be relatively healthy. The general population—used as the comparison group in SMR calculations—contains chronically ill, severely disabled, institutionalized, and unemployed individuals with high baseline mortality rates.
• Impact: Overall all-cause mortality in an active workforce is typically 70% - 90% of the general population (SMR = 70 - 90).
• Consequence: The HWE can mask true occupational hazards. An observed all-cause SMR = 100 in a working cohort does not indicate safety—it often conceals an underlying excess occupational mortality that is counterbalanced by the workforce's baseline health advantage.
• Mitigation: Compare the exposed worker cohort against an internal unexposed or low-exposure occupational control group (e.g., administrative plant workers) rather than the general population.

2. Information and Observation Biases

• Recall Bias: Differential recollection of past chemical exposures between diseased cases and healthy controls in retrospective case-control studies. Parents of children with birth defects or workers diagnosed with leukemia scrutinize their past chemical exposures far more intensely than healthy controls, artificially inflating the calculated Odds Ratio.
• Interviewer / Observer Bias: Systematic error introduced when the interviewer knows the subject's disease or exposure status and probes more aggressively.
• Misclassification Bias:
  • Non-Differential Misclassification: Exposure or disease measurement errors occur equally across all groups (random error). Always biases the measure of association toward the null (RR → 1.0, OR → 1.0), diluting true causal signals.
  • Differential Misclassification: Measurement errors differ systematically between groups. Biases the effect estimate unpredictably (either toward or away from the null).

---

6. The Bradford Hill Criteria for Causal Inference

In 1965, Sir Austin Bradford Hill articulated nine classic criteria to help scientists determine whether an observed statistical association reflects true biological causality:

---

7. Worked Step-by-Step Calculation Examples

Worked Example 4.3.1: Relative Risk & Attributable Risk in a Cohort Study

Problem: An occupational health department tracks a cohort of 1,000 cadmium-exposed electroplating workers and an unexposed comparison cohort of 2,000 warehouse workers for 20 years. Over the follow-up period, 40 cadmium workers and 20 warehouse workers develop chronic kidney disease (proteinuria/nephropathy).

1. Construct the 2 × 2 contingency table.
2. Calculate cumulative incidence in exposed (Ie) and unexposed (Iu).
3. Calculate the Relative Risk (RR), Attributable Risk (AR), and Attributable Risk Percent (AR%).

Solution Steps:

1. Construct the 2 × 2 Table:

   • Exposed (E+): a = 40 (diseased), b = 1,000 - 40 = 960 (non-diseased); Total = 1,000
   • Unexposed (E-): c = 20 (diseased), d = 2,000 - 20 = 1,980 (non-diseased); Total = 2,000

2. Calculate cumulative incidence: $I_e = \frac{a}{a + b} = \frac{40}{1,000} = 0.040 \quad (4.0\% \text{ or } 40 \text{ per } 1,000)$ $I_u = \frac{c}{c + d} = \frac{20}{2,000} = 0.010 \quad (1.0\% \text{ or } 10 \text{ per } 1,000)$

3. Calculate measures of association:

   • Relative Risk (RR): $\mathbf{RR = \frac{I_e}{I_u} = \frac{0.040}{0.010} = 4.0}$ (Cadmium-exposed workers have 4 times the risk of kidney disease compared to unexposed workers).
   • Attributable Risk (AR): $\mathbf{AR = I_e - I_u = 0.040 - 0.010 = 0.030} \quad (3.0\% \text{ or } 30 \text{ excess cases per } 1,000)$
   • Attributable Risk Percent (AR%): $\mathbf{AR\% = \left(\frac{RR - 1}{RR}\right) \times 100\% = \left(\frac{4.0 - 1}{4.0}\right) \times 100\% = \mathbf{75.0\%}}$ (75% of the kidney disease cases among electroplating workers are directly attributable to occupational cadmium exposure).

---

Worked Example 4.3.2: Odds Ratio Calculation in a Case-Control Study

Problem: A hospital-based case-control study evaluates occupational trichloroethylene (TCE) degreasing exposure and renal cell carcinoma. The study enrolls 150 confirmed renal cancer cases and 300 matched controls. Retrospective occupational hygiene exposure reconstruction reveals that 60 of the cases and 45 of the controls had significant past TCE degreaser exposure.

1. Construct the 2 × 2 table.
2. Calculate the Odds Ratio (OR).
3. State the epidemiological interpretation under the rare disease assumption.

Solution Steps:

1. Construct the 2 × 2 Table:

   • Cases: a = 60 (exposed), c = 150 - 60 = 90 (unexposed); Total Cases = 150
   • Controls: b = 45 (exposed), d = 300 - 45 = 255 (unexposed); Total Controls = 300

2. Calculate Odds Ratio (OR): $\mathbf{OR = \frac{a \cdot d}{b \cdot c} = \frac{60 \times 255}{45 \times 90} = \frac{15,300}{4,050} = \mathbf{3.78}}$

3. Interpretation: The odds of prior occupational TCE exposure among renal cell carcinoma cases are 3.78 times higher than among healthy controls. Because renal cell carcinoma is a rare cancer in the general population, this Odds Ratio (OR = 3.78) serves as an accurate estimate of the Relative Risk (RR ≈ 3.78).