Assessment of Exposure-Response Functions for Rocket-Emission Toxicants (1998)

As noted in Chapter 1, the LATRA model is used to define human health risks associated with rocket-launch scenarios (i.e., normal launches with different rocket-fuel types and accident scenarios). Coupling source characteristics of the toxic agents under consideration—HCl, NO₂, and HNO₃ —with real-time meteorological data, a dispersion model (REEDM) is used to simulate exposures (i.e., to predict concentration-time profiles at receptor locations). The exposure-response functions (ERFs) in LATRA translate exposure estimates from REEDM into probabilities of health effects in specified severity categories in the human population. At present, separate ERFs are developed for two segments of the population: "sensitive" and "normal" populations. Within each segment, the model incorporates separate ERFs for "mild" and "serious'' health effects. The ERFs included in LATRA at present are lognormal for noncarcinogenic substances and linear, passing through the origin, for carcinogenic substances. When sufficient data are available to support a nonlinear ERF for a carcinogen, the Air Force should consider modeling such data.

At each receptor location modeled by REEDM, the ERFs are applied to the number of individuals estimated from census data to be present in both population subgroups at that location. For each severity category, the ERF is the probability, P_E, per individual of an effect, Y , exceeding a given severity category, S (mild or serious) given an exposure concentration and duration. That is,

P_E ( S,C,T ) = P ( Y ≥ S/C,T ),

where P_E (S,C,T ) is the ERF for an exposure characterized by concentration, C, and time or duration of exposure, T , and is equal to the probability of the severity equaling or exceeding a given severity category, S, at a specified exposure concentration, C, and duration, T .

As indicated in Chapter 1, the task of the subcommittee was to review and provide recommendations on several issues surrounding the exposure-response components of the LATRA model. Further description of the model and issues surrounding those components is provided below.

The toxicity reference values originally used by the Air Force in the LATRA-ERF model to represent a 1%-effect level for HCl, NO₂, and HNO₃, had been established by other groups almost a decade earlier (e.g., NRC 1987,1991). The toxicity data for those substances needed to be reevaluated with respect to sensitive populations, severity of effect, and the availability of dose-response information. Those reevaluations are presented in Appendices D, E, and F, respectively. Use of that information to derive ERFs for LATRA is explored in Chapter 6.

The LATRA-ERF model divides potentially exposed populations into at least two population subgroups, sensitive and normal, and develops separate ERFs for each subgroup. Sensitive populations are defined as children (less than 15 years of age), the elderly (more than 64 years of age), and all persons with bronchitis, asthma, or other physiological stress, especially upper-respiratory ailments (Gene Killan, U.S. Air Force Space Command, personal commun., May 6,1996). Under LATRA, the remainder of the population is considered "normal" and is assumed to be composed of healthy adults. Census data are used to determine the

locations of the sensitive subgroups around a launch site (e.g., in nursing homes and schools). Because the ERFs for sensitive individuals are applied only to such locations identified by census data and only to the proportion of the population considered sensitive at these sites, they do not not protect sensitive individuals within the larger community. Sensitive subgroups are assumed to respond to the rocket-emission toxicants at lower concentrations than the normal population and to exhibit less variation in response, showing a steeper increase in incidence of response with increasing exposure concentration than does the normal population. How the ERFs are actually quantified for LATRA is discussed later in this chapter in the section Quantification of the Exposure-Response Functions.

Although this approach requires quantifying separate ERFs for each population subgroup, the Air Force pointed out that it allows them to apply the ERF for healthy adults to all locations where sensitive subgroups are not found. The more conservative ERFs for sensitive individuals need be applied only to those locations where the census data indicate that there are sensitive subgroups and only to the proportion of the population considered sensitive at those locations.

Based on information supplied to the subcommittee, it appears that the Air Force adopted the elderly age cutoff of more than 64 years from an EPA definition that was used by CDC (1993) investigators when estimating populations at risk in communities that have not attained one or more National Ambient Air Quality Standards in the United States (Poitrast 1993).

The subcommittee considered several questions when evaluating the sensitivity component of the LATRA-ERF model: What characteristics are likely to make an individual more sensitive to the specific rocket emission toxicants? What is the magnitude of the difference in sensitivity between the more-sensitive members and the more-average members of the population? Is the difference in sensitivity between those groups reflected in the concentration representing a threshold for response, the severity of response for a given concentration, the variability in response, or some combination of those attributes? What considerations for hypersensitive individuals might be appropriate? The subcommittee's evaluation, conclusions, and recommendations regarding these questions are provided in Chapter 3.

The LATRA model has three health-effect severity levels, as listed below:

Mild: no damage to body organs; temporary irritation.

Significant (or severe or serious): damage to body organs; treatment required.

Fatal (considered unacceptable) (Philipson et al. 1996).

The LATRA considers two categories of severity—mild and worse and serious and worse—with separate ERFs for each.

The Air Force asked an inter-agency advisory panel, the Rocket Emissions Working Group (REWG) (see Appendix A), to identify what signs or symptoms would be expected to accompany mild and severe effects. For an acute (minutes to 1 hr) exposure to irritant gases that are relatively soluble in aqueous solution (e.g., HCl), REWG identified a mild response to be a transient irritation of the eyes, skin, and upper airways (nasopharyngeal and upper tracheobronchial tree) (Gene Killan, U.S. Air Force Space Command, personal commun., May 6, 1996). REWG identified likely responses as sneezing, nasal catarrh (i.e., inflammation of mucous membrane), unpleasant smell or taste, throat soreness, smarting of the eyes, and lacrimation (tearing). REWG identified a severe response to be a reversible or irreversible response that might require medical intervention, especially when the central airways of the tracheobronchial tree are involved. REWG identified signs and symptoms of severe effects as coughing, sputum, pain, chest constriction, bronchospasm, shortness of breath, and wheezing.

The subcommittee evaluated the approach of categorizing health effects by severity and of developing separate ERFs for each severity category. Also, because the Air Force stated that a category reflecting moderate effects would assist in making launch decisions, the subcommittee evaluated how to define three levels of effect severity—mild, moderate, and severe—in a way that is both scientifically sound and meaningful to an Air Force commander. The subcommittee's evaluation, conclusions, and recommendations concerning the approach and definition of severity categories are presented in Chapter 4.

The ERFs included in LATRA at present are lognormal for noncarcinogenic substances and linear, passing through the origin, for carcinogenic substances. (Note: ERFs have not been developed for any carcinogenic substances to date (Philipson 1996)). ERFs for noncarcinogenic substances are actually represented by a log-probit model applied to either the maximum 1-hr time-weighted-average (TWA) concentration (C_max, 1-hr) or to the ceiling concentration (C_max). Because the log-probit model is equivalent to a cumulative lognormal distribution function, the LATRA ERF can be characterized generically as a model for predicting

P_E(S,C,T ) = P(Y ≥ S/C,T ) = lognormal_cml [C_max,1-hr,µ₁(S),s₁(S)],

where lognormal_cml [C_max,1-hr,µ1(S),s₁(S)] is the cumulative lognormal distribution function with mean natural logarithm of concentration, 1n(C), equal to µ₁(S) and standard deviation in ln(C) equal to s₁( S). When a ceiling concentration is used instead of the 1-hr TWA, then the term "C_max, 1-hr" is replaced by "C_max," representing the ceiling concentration.

This distribution expresses the probability that a randomly selected individual within a population experiences an effect of at least severity S when the distribution of likelihood of response in that population is such that half the population will experience an effect at

C_max, l-hr = exp(µ₁(S)),

and where the geometric standard deviation (GSD) of this distribution of effects of severity S is given by

GSD(S) = exp(s₁(S)).

The GSD expresses the implicit variation in the probability of response per individual based on the assumed set points for concentrations corresponding to the 1% and 99% incidence rates.

The log probit has a sigmoidal shape. Special graph paper can be used that converts incidence (proportions or percentages) to probits so that a straight line plot of probits versus log dose is obtained. The slope

of that line is the reciprocal of the geometric standard deviation. Specific examples are illustrated in Chapter 6.

Separate ERFs are used for the 1-hr TWA and the ceiling values; the higher probability of effect from the two curves is then used in computing the number of people likely to be affected. The REEDM dispersion model contained within LATRA does not reliably predict instantaneous peak concentrations, however. Averaging times of at least 30 min are considered the most meaningful model output (Stokes 1994), although the model does analyze exposures down to 5-min increments. The documentation supplied to the subcommittee did not specify, however, how maximum peak or TWA exposure concentrations are derived from the REEDM output for comparison with the toxicity values.

The subcommittee considered whether the two different independent variables included in LATRA—a 1-hr TWA value and a ceiling value—are the most appropriate, given what is known about the rocket-mission toxicants and given the likely duration and frequency of exposure. The subcommittee also considered what type of exposure-response model would be most appropriate for the ERFs for the three rocket-emission toxicants under evaluation and for the types of risks that LATRA is attempting to estimate. The subcommittee's evaluation, conclusions, and recommendations regarding those issues are provided in Chapter 5.

The lognormal ERFs for noncarcinogenic rocket emissions are specified at present by two symmetric percentiles: the 1 and 99 percentiles. Specifically, the values µ_l(S) and s₁(S) associated with the GSD are determined by assigning to the 1% effect level of the distribution a safe exposure concentration and by assigning to the 99 % effect level an assumed ED₁₀₀ value (dose assumed to cause an effect in 100% of the population). Thus,

P_E(S,C,T ) = P_E(S,C₀₁,1-hr) = lognormal_cml [C_max = C₀₁,1-hr,µ₁(S), s₁(S)],

where

C₀₁,1-hr = 1-hr SPEGL or other measure of safe dose,

P_E(S,C,T ) = P_E(S, C₉₉,1-hr) = lognormal_cml[C_max = C₉₉,1-hr,µ₁(S), s₁(S)],

and C₉₉,1-hr = 1-hr ED₁₀₀.

Under those constraints,

µ₁(S)=(C₀₁ × C99)^1/2, and

s₁(S) = [ln(C₀₁)-ln(C₉₉)]/(2 × 2.33).

The lognormal curves are cut off to zero below the 1 percentile to reflect a threshold for noncancer effects and raised to 1.0 above the 99 percentile to be conservative. The assumption is that the 1% effect level represents exposures below which "essentially no one" would experience each specified severity of effect. The 99% effect level represents exposures above which "essentially everyone" would experience the specified severity of effect (Philipson 1996).

The Air Force considered SPEGLs and other established exposure concentrations considered safe for the general public when setting the 1% effect levels for sensitive populations. Similarly, the Air Force considered established exposure concentrations considered safe for workers in setting the 1% effect levels for normal populations. The 99% effect levels were set 5-fold higher than the 1% effect levels for sensitive populations and 10-fold higher than the 1% effect levels for normal populations (Philipson 1996). The documentation provided to the subcommittee did not explain the rationale for those values; however, L. Philipson (ACTA Inc., personal commun., Jan. 15-16,1997) indicated to the subcommittee that the range of variability in response among individuals in a normal subgroup was assumed to be twice the range of variability among individuals in a sensitive subgroup. There also was consideration of NIOSH (1994) immediately dangerous to life and health (IDLH) values in developing the tier limits considered dangerous and consideration of the tier limits in identifying appropriate toxicity values for the 99% incidence values (see Chapter 1 and Appendix A). Table 2-1 lists the exposure concentrations associated with the 1% and 99% incidence values for sensitive and normal populations and mild and serious effects for the three rocket-emission toxicants.

The subcommittee evaluated the appropriateness of this approach to quantifying the chemical-specific ERFs, including the appropriateness of using "safe" levels to represent the 1% effect level, the approach for identifying a 99% effect level, and the assumed difference in slope of the log-probit function between sensitive and normal populations. The subcommittee also considered alternative analytic models to the logprobit model and what types of chemical-specific information or data are most appropriate to use for each approach. Those evaluations, and the subcommittee's conclusions and recommendations, are also presented in Chapter 5.

The ERFs represent a deterministic relationship between exposure concentration and the proportion of an exposed population that would exhibit an effect at a specified severity level. At each receptor location where a concentration for each rocket-emission is estimated, the number of people affected in the population at that location is calculated from the probability P_E using a binomial distribution. That is, the number of individuals in each subgroup who suffer at least an effect of severity S at the given receptor location is estimated as

n(N,S,C,T ) = sum(i = 1, 2,... N ){i × P_bnml[i,N,P_E (S,C,T )]},

where n is the expected number of individuals with effects of at least severity S in a population of size N, exposed to concentration C for duration T , and P_bnml is the binomial distribution function that expresses the probability of observing i effects in a population of size N, when the probability of effect per individual is P_E(S,C,T ). In the current LATRA model, all of those binomial distributions are combined by adding their means and variances to obtain the mean and variance of the total number of people suffering effects of a given severity at all receptor locations. To derive a complete risk profile, the binomials are usually approximated by Poissons to obtain a total Poisson distribution for all possible numbers of affected individuals.

LATRA estimates the total number of people at risk from the launch on the basis of: (1) the risks associated with a normal launch, (2) the probability of a normal launch, (3) the risks associated with a cata-

strophic abort, and (4) the probability of a catastrophic abort. Depending on the toxicity criteria used, as well as the quantity of propellants onboard, meteorological conditions, proximity of population centers, and so forth, far greater health risks generally are expected for catastrophic aborts than for normal launches. As a consequence, the total risk estimate produced by LATRA generally reflects the risks estimated for a catastrophic abort because the risks estimated for a normal launch usually are much lower.

The potential for combined effects of exposure to more than one compound at the same time is estimated after the risk profiles for individual compounds are estimated by "developing joint probabilities of effect from the individual toxics' probabilities of effect (assuming their independence)" (Philipson et al. 1996). This assumption was considered conservative because of the high level of correlation expected among exposures to the individual toxic emissions. The mode of action of the various toxicants needs to be considered. Such an investigation might provide support for response additivity or dose additivity of the constituents in a mixture.

The subcommittee considered the relative importance of the various uncertainties associated with the LATRA-ERF model, evaluating how well the important uncertainties are represented in the model and how they should be used to qualify the risk estimates. Those evaluations and the subcommittee's conclusions and recommendations also are provided in Chapter 5.