Exposure Assessment: From Understanding the Workplace to Making a Decision

01 Understand the Workplace Before Choosing a Sampler

Basic characterization, similar exposure groups, the exposure profile and where professional judgment fits.

Basic characterization collects and organizes information about the workplace, workforce, agents and exposure conditions. Record materials, processes, tasks, durations, controls, routes of exposure, maintenance, abnormal operations and existing measurements.

A similar exposure group (SEG) groups workers expected to have similar exposure profiles because of shared exposure determinants. A common job title is not enough: tasks, materials, controls and working patterns may differ.

An exposure profile describes the distribution and variability of exposures for a defined agent, group and averaging period. It includes typical exposure, upper-tail exposure and the probability of exceeding a relevant benchmark. The range alone is not a sufficient measure of variability.

Professional judgment combines education, experience, observations, inference and analogy. Document its basis and uncertainty; it is not a substitute for evidence when uncertainty could change the decision.

02 Define the Question and Benchmark

Match the evidence to the question, and use the right OEL form, units, averaging period and jurisdiction.

Use the correct OEL, chemical form, units, averaging period and jurisdiction. An 8-hour TWA, STEL and ceiling answer different questions. A population's long-term arithmetic mean is also different from the upper tail of its daily TWA distribution.

Exposure below an OEL is not a guarantee of zero risk. Consider sensitization, skin exposure, mixtures, unusual susceptibility and the limitations of the benchmark.

03 Combine Qualitative and Quantitative Assessment

Each approach has a limit; combined, monitoring tests and updates the first assessment.

Quantitative assessment does not necessarily mean laboratory analysis: a suitable direct-reading instrument or exposure model can provide numerical evidence. Laboratory competence and validated analytical methods matter when laboratory testing is used.

04 Choose the Sampling Dimensions Separately

Personal or area, grab or integrated, whole air or media, active or passive: one sample can be several of these.

These labels are not mutually exclusive. One sample can be personal, active and integrated.

Whole-air collection can be short or time-integrated. Integrated sampling is not limited to adsorption tubes: filters and other methods also integrate exposure. Passive samplers require an appropriate uptake rate and operating conditions.

Source: NIOSH Manual of Analytical Methods.

05 What Direct-Reading Instruments Actually Measure

Detector tubes, PIDs, electrochemical sensors and infrared analyzers, with the limit that matters for each.

Do not describe every electrochemical sensor as an identical polarographic cell. Do not assume a PID response identifies a particular solvent. OSHA Technical Manual—Direct-Reading Instruments. This is technical guidance, not a Canadian legal requirement.

06 Design Sampling for the Decision

Represent the population of interest; there is no universal number of samples.

Sampling should represent the exposure population of interest: relevant workers, days, tasks, operating conditions and variability. Random or stratified selection supports distributional inference. Deliberately high-exposure sampling is useful for other questions, but should not silently be treated as a random sample of routine exposure.

Record task duration, materials, ventilation, unusual events and departures from procedure. A result without context can be misleading.

There is no universal “6–10 samples makes a robust dataset” rule. AIHA's guidance describes six to ten baseline personal samples as an initial step for some similar exposure groups, not as a guarantee of a robust dataset. Sample requirements depend on variability, representativeness, confidence goals, proximity to the OEL and the decision method. Small datasets can be informative while still producing wide uncertainty. More samples from an unrepresentative condition do not repair a poor design.

AIHA emphasizes upper-tail exposure judgments and their uncertainty. AIHA—Tips and Tools for Improving Exposure Judgments.

07 Validate the Sample Before Calculating Statistics

Calibration, duration, media, backup sections, interruptions and detection limits come first.

Accuracy concerns agreement with a reference value; precision concerns agreement among repeated measurements under specified conditions. A precise method can still be biased.

A backup-section result above a single universal percentage is not an all-method invalidity rule. Check the specified method, including whether the criterion is a percentage of the front section or the total. Similarly, apply the method's flow acceptance criteria. Document deviations and their consequences rather than automatically accepting or discarding every affected sample.

08 Describe the Data Before Fitting a Distribution

Check, plot and describe the data, then choose and test a distribution.

1. Confirm units, averaging periods, sample validity and treatment of nondetects.
2. Plot results in time order and by task, worker and control condition.
3. Look for changing processes, clusters and possible SEG misclassification.
4. Calculate appropriate descriptive statistics.
5. Assess whether the selected distribution and independence assumptions are reasonable.
6. Estimate the decision statistic and its uncertainty.

Lognormality is common in some exposure datasets, not guaranteed. Graphical checks and suitable tests such as Shapiro–Wilk on log values can help, but small samples have limited power. Do not pool distinct exposure populations merely to obtain a larger sample size.

09 Use the Lognormal Formulas Correctly

GM, GSD and the 95th percentile, with a worked example and the exact meaning of GSD.

For positive observations, let $y_i=\ln(x_i)$, with log-data mean $\bar y$ and sample standard deviation $s_y$.

For a fitted lognormal distribution:

The population GM equals the population median under this model. GSD corresponds to approximately X₈₄.₁₃/X₅₀, not exactly X₈₅/X₅₀. More generally, $X_p = GM \times GSD^{z_p}$.

Sources: NIST—Geometric Standard Deviation; NIST—Lognormal Distribution.

The equation card writes GM with base-10 logarithms; natural logarithms give the same GM and GSD as long as the same base is used for the back-transformation.

Equation cards: Geometric mean and Geometric standard deviation.

Worked example

Assume a suitable fitted lognormal model with GM = 0.20 mg/m³, GSD = 2.0, and OEL = 1.0 mg/m³:

The point estimate is about 63% of the OEL. This does not establish the confidence of the decision; uncertainty must also be considered.

10 Separate the Percentile From Confidence in It

An estimated X95 is not a 95/95 upper tolerance limit.

The 95th percentile describes variation in the exposure population. A confidence bound describes uncertainty in an estimated parameter.

A one-sided 95/95 upper tolerance limit is designed to cover at least 95% of the population with 95% confidence, subject to its assumptions. It is not the same as the point estimate of X₉₅ or the upper confidence limit of the arithmetic mean.

A small dataset may yield a moderate X₉₅ estimate but a high upper tolerance limit. That gap signals uncertainty; it does not prove that every exposure is high. Use an appropriate validated statistical procedure rather than substituting a normal-distribution z-value for a tolerance factor.

11 Categorize Exposure Without Automatic Rules

The AIHA 0 to 4 categories describe X95 against the OEL; they are not legal thresholds.

AIHA's common 0–4 framework places the estimated 95th percentile relative to the OEL:

Use the chosen procedure's explicit endpoint convention at exact boundaries. Document uncertainty alongside the category. This table concerns X₉₅, not the arithmetic mean. Source: AIHA—Principles of Good Practice.

If the modeled population X₉₅ exceeds the OEL, the modeled exceedance probability exceeds 5% for a continuous distribution. That does not grant permission for 5% of legal exceedances.

Acceptable means the evidence supports the assessment's defined protective criteria—not zero risk. Uncertain means additional evidence or precautionary control may be needed. Unacceptable means action is needed under the applicable decision criteria.

Do not assign “no action,” mandatory medical surveillance or a specific respirator solely from this category number. Existing controls, legal duties, agent-specific surveillance requirements and acute hazards still matter. Respirator APFs are not universal toxicity or shutdown thresholds. Some frameworks add a fifth category or an automatic action ladder; those belong to the specific framework that defines them and are not part of the AIHA 0 to 4 categories described here.

12 When an OEL Is Missing: Banding Systems

Control banding and the HSE COSHH A to E groups are frameworks, not OELs.

Control banding combines hazard information with exposure potential to select a control approach. It does not manufacture a legal OEL or replace all specialist assessment.

The ranges below are those of the HSE COSHH Essentials technical framework:

The HSE framework also has a Hazard Group S for skin and eye contact, which does not use these airborne ranges. These are framework-specific bands, not universally safe concentrations. Selection also considers factors such as quantity, volatility/dustiness and task. Do not interchange the A–E labels with NIOSH occupational exposure banding or the AIHA exposure categories above. Consult the source for exact boundaries and applicability. HSE—COSHH Essentials Technical Basis.

13 Use Models and Bayesian Analysis Transparently

Priors, likelihood and posterior, and what the common tools are for.

Bayesian analysis combines prior information with the likelihood of the monitoring data to produce a posterior distribution. It can express probabilities that a profile falls within exposure categories. Document prior assumptions and avoid counting the same data twice as independent evidence.

Tool versions and capabilities differ. Record inputs, assumptions and limitations; a spreadsheet output is not independent validation of its inputs. AIHA—Exposure Assessment Tools Overview.

14 Close the Loop

Controls, communication, reassessment and a defensible decision.

Translate findings into controls, communicate the reasoning and reassess after relevant changes. Preserve uncertainty and unresolved conditions in the record.

The useful outcome is a defensible exposure decision—not merely a sample result, a category number or a completed spreadsheet.