Subjects rated each symptom on a ratio scale from 0 (barely detectable/no sensation) to 100 (strongest imaginable) (Green et al. 1996). We chose symptoms based on previous literature assessing the health effects of indoor air mixtures (Molhave et al. 1986; Prah et al. 1998). These symptoms included the cognitive and physical effects expected of VOC mixtures (Hudnell et al. 1992; Molhave et al. 1986), anxiety symptoms associated with the odor of exposure, eye irritation, upper respiratory and lower respiratory symptoms associated with O₃ and the secondary products generated from the reactions of O₃ with VOCs (Fan et al. 2003), and somatic symptoms not typically associated with VOC mixtures (Appendix 1) (Dalton et al. 1997).

This computerized divided-attention test of cognitive performance, performance on-line (POL) (Mills et al. 1996) offered five different levels of complexity. The test was validated in alcohol dosing trials and was developed explicitly for use in repeated-measures studies of alcohol and drug effects. POL included a central task in which the subject was presented with two lanes of traffic, divided by a double yellow line. Four conditions of “headlights” and “tail lights” appeared on any one trial. The subject was instructed to press the space bar only when a “safe” condition (i.e., left lane, white headlights, and right lane, red tail lights) existed. The peripheral task required the subject to respond with one of four arrow keys (up, down, left, right) in the direction of the critical stimulus (red octagon among other shapes). Task difficulty increased by increasing the number of distracting stimuli in the peripheral display to a random assortment of different colored circles, squares, and triangles. For the divided-attention display, the subject responded to both central and peripheral critical stimuli. After practice, 10 trials of 45 displays were presented at the most complex level. A composite performance score, composed of seven component scores, including hits, misses, false positives, response latency, and responses to targets at varying levels of visual angle, was the performance variable measured (Badiani et al. 1995).

An extensive literature documents the significant (r ≥0.90) association between salivary and plasma cortisol (Kirschbaum and Hellhammer 1989, 1994). Salivary flow, which may be affected by anxiety, is not documented to affect the concentration of salivary cortisol (Dirks et al. 1988; Kahn et al. 1988), although circadian rhythm affects cortisol production (Walker et al. 1984). Thus, we tested subjects at the same time of day to control for the well-known circadian effect on cortisol production. Kirschbaum and Hellhammer (1989, 1994) also reported that, although a highly significant correlation is shown between salivary and plasma cortisol, absolute values vary significantly. However, in our study, we evaluated relative change scores rather than absolute values. Because hormonal fluctuations also affect cortisol levels, we measured salivary estradiol at baseline before each exposure session to account for any effects of ovulation on salivary cortisol.

Consistent with Kirschbaum and Helhammer (1989), we collected samples using the Salivette (Sarstedt Inc., Rommelsdorf, Germany) method. We asked the subject to chew on a cotton swab for 60 sec, then place the swab into a Salivette holder and affix the cap. The samples were centrifuged at 3,000 rpm for 10 min, which produced 0.5–1 mL of saliva. The saliva was frozen. Samples were analyzed, blind to exposure condition and subject characteristics, in one of the co-investigators’ laboratories (J.O.). Samples were run in duplicate, and equal numbers of samples from each group were run in the same assays.

With the subjects in standing position with nose clips attached, spirometry was performed using a Multispiro SX spirometry system (Creative Biomedics, Inc., San Clemente, CA) that was calibrated daily. We used the highest value for each parameter from any one of three reproducible tracings before exposure for comparison with the highest value for each parameter postexposure. Forced vital capacity (FVC), forced expiratory volume in 1 sec (FEV₁), and forced expiratory flow 5–75% (FEF_25–75) were parameters of interest.

After a 4-min silent preparation period, subjects constructed and delivered a 4-min speech on the following controversial scenarios: a) construct a presentation on the causes of gray hair from a Reader’s Digest article; b) present a position on whether homosexual men should be allowed in the military and be given special civil rights protection; c) defend herself against a shoplifting charge in front of three authority figures (al’Absi et al. 1997). In a direct comparison al’Absi et al. (1997) demonstrated that these three different topics caused a similar level of stress (cortisol response). We randomized the order of the three scenarios across subjects such that each subject delivered a speech on a different topic in each of her three exposure conditions. After the preparation period, subjects were told by the experimenter to stand, face the video camera, and begin giving the speech. To enhance the stressfulness of the procedure, we told each subject that the research technician was evaluating the speech as it was given and that three staff members would evaluate a videotape of the speech later; the subject was promised an additional stipend of $10 if she performed well. After completion of the three experimental sessions, all subjects received the added stipend regardless of performance.

The Environmental and Occupational Health Sciences Institute Controlled Environment Facility (CEF) is a stainless steel room 2.2 m × high × 4.1 m wide × 2.7 m deep with a total volume of 25 m³. The air supply is treated in a series of conditioning processes, which include air cooling/heating, humidification/dehumidification, and filtration through carbon and high-efficiency particulate air (HEPA) filters. All parameter controls are computer interfaced to maintain constant conditions in the CEF. To simulate a typically ventilated office building and to allow sufficient time for the formation of O₃–VOC reaction products, the air flow rate through the facility was controlled at 1.8 ± 0.2 air changes/hr for all the exposures conducted in this study. The air supply enters the facility through two diffusers in the ceiling and exits through the perforated stainless steel floor to the exhaust vents. Small brushless fans (to prevent unwanted particle generation from brush degradation) were used in the CEF to ensure that the air was well mixed. During exposure sessions, the relative humidity and temperature were controlled to a range of between 24 and 49% and 73–82°F (23–28°C), respectively. A Teflon partition separated the subjects’ work stations, which were equipped with a computer and typing stand on a stainless steel table. This configuration allowed exposure of two subjects at a time. The technician communicated separately with each subject through headphones and could view subjects at all times through a two-way window. For subject safety, an electrocardiogram (ECG) signal was collected using disposable electrodes and sampled at 996 Hz through the Flexcomp Biomonitoring system 1.51 B (Thought Technology Ltd., Montreal, Quebec, Canada). This allowed continuous monitoring of heart rate during the exposure period.

The composition and relative weight of the VOC mixture used were similar to those in previous indoor air studies (Kjaergaard et al. 1991; Molhave et al. 1986; Otto et al. 1992) and are reported in Table 1 (Fan et al. 2003). d-Limonene, the most frequently identified terpene in indoor air, was also added to the mixture. A flask containing the liquid mixture of these 23 compounds was heated to 250°C by a hot plate, flash evaporated, and injected into the clean air stream that delivered the VOC mixture into the CEF. Constant concentrations of chemical compounds can be maintained in the CEF by injection of the chemicals continuously into the air supply, which flows through the CEF without recirculation. The desired concentration of VOCs in the indoor environment was achieved by adjusting the flow rate of the delivering air using a mass flow controller.

Total VOC concentration in the air of the indoor environment was approximately 26 mg/m³, with the concentration of each compound below the threshold limit value recommended for occupational exposure by the American Conference of Governmental Industrial Hygienists (1992). MCA was generated by introducing a 1-min spike of the VOC mixture at approximately 10% of the exposure concentration for the VOC exposure condition. The maximum total VOC concentration in the CEF during the MCA condition was approximately 2.5 mg/m³. Given the air exchange rate used in the experiments, it took about 90 min for the masking concentration to decay to 0.25 mg/m³.

The background O₃ concentration is < 3 ppb in the CEF. O₃ generated in situ by an O₃ generator was delivered into the CEF and was maintained at a steady-state concentration of 40 ppb. This was consistent with the intention to examine the effect of products of O₃–VOC reactions rather than O₃ itself. Table 2 compares the concentrations of byproducts generated in the VOC + O₃ condition relative to the VOC condition. Exposure generation and characterization were described in detail in a previous publication (Fan et al. 2003).

We analyzed the effects of exposure, stress, and time on both presence/absence of symptoms and symptom severity. For presence/absence, if a subject reported any symptom at all, then a “yes” was recorded; otherwise, a “no” was recorded. For presence/absence, a hierarchical logistic regression (Davidian and Giltinan 1995; McCulloch and Searle 2000) modeled the log-odds of no symptoms being reported for assessment at minute 15 (baseline) through minute 185 (after removal from exposure facility). For symptom severity, data were analyzed using a hierarchical Poisson regression model. In both cases generalized estimating equations were used that accounted for correlations between repeated measurements on the same individual (Liang and Zeger 1986). Tests of the exposure effects were conducted using type 3 score tests (Liang and Zeger 1986) of the interaction between exposure and time. Time was entered into the model as a categorical variable. We used contrasts to test whether individual changes in symptoms from baseline (minute 15) to each subsequent time point differed between exposures. The mean odds of reporting symptoms or the mean severity of symptoms at baseline was assumed to be the same for all three exposures. This analysis was first completed for the total symptoms and then for each classification of symptoms: VOC physical, VOC cognitive, eye irritation, anxiety, upper respiratory, lower respiratory, and somatic control symptoms. Results were based on the 130 subjects who received all three exposures, including the MCA condition. Uncorrected α-values are reported, with the αlevel after Bonferroni correction noted for each group of multiple comparisons.

Cortisol is normally elevated during ovulation, potentially attenuating a woman’s cortisol response to a stressor at that time. Therefore, cortisol data were analyzed both with and without data from any specific exposure session in which the subject’s estradiol level was ≥5 ng/mL. Based on this cutoff, 13 of the 130 (10%) subjects were ovulating during their MCA exposure, 13 of 130 (10%) were ovulating during the VOC exposure, and 15 of 130 (11.5%) were ovulating during the VOC + O₃ exposure. On comparison of the cortisol data analyses with and without ovulation sessions, no differences in outcome were noted. Therefore, sessions in which subjects were ovulating were included in the final analyses of cortisol.

Given the continuous response, SAS PROC MIXED (SAS Institute Inc. 1999) was used to analyze the fixed-factor effects. The cortisol response was right skewed and therefore, for the linear models, was transformed using a square root transformation to better satisfy the normality assumption required for the mixed linear model.

A mixed linear model was used to test the effect of exposure on the composite score from the neurobehavioral task and on indicators of lung function.

After controlling for baseline symptoms (minute 15), the overall test of the exposure main effect on total symptom severity did not confirm hypothesis 1 (Figure 2). However, marginal effects for presence/absence of the subscales of VOC physical symptoms (chi squared = 18.01; df = 10; p < 0.05) and severity of lower respiratory symptoms (chi squared = 16.92; df = 10; p < 0.08) were observed; no effects were observed for the other categories of symptoms.

There was no significant main effect of exposure on neurobehavioral performance (F = 1.00, df = 6, 387; p = 0.4273), salivary cortisol (F = 0.40; df = 6, 316; p = 0.888), or lung function (data not shown for neurobehavioral or cortisol measures; see Table 3 for lung function).

Hypothesis 2 was not confirmed for presence/absence or for total symptom severity. However, regardless of exposure condition, subjects who were in the stress condition reported significantly greater severity on the anxiety sub-scale (chi squared = 22.73; df = 5; p < 0.0004) than those who were not. Specifically, relative to baseline either before (minute 15) or after exposure onset (minutes 60 and 90), symptoms of anxiety were significantly more severe after the stressor at minutes 110 and 125 (Figure 3). No other symptom subscale was significantly affected by stress or by the exposure × stressor interaction.

The hypothesized interaction effect of exposure × stress on cortisol was not significant, but the main effect of stress on cortisol was significant (F = 4.90; df = 3, 347; p = 0.0024). The significance of the main effect of stress was due to changes in cortisol levels from before the stressor (minute 90) to after the stressor (minute 125) (t = 4.00; p < 0.0001). When examining changes from minutes 110 to 125, on average, the cortisol levels decreased (from 0.178 to 0.154) for the no-stressor condition and increased slightly (from 0.140 to 0.147) for thestressor condition (t = 3.63; p = 0.0003) (Figure 4).