Opinion for the Court filed by Senior Circuit Judge McGOWAN.
McGOWAN, Senior Circuit Judge:In these consolidated cases, we review the Occupational Safety and Health Administration’s rule limiting exposure to ethylene oxide, a chemical widely used in manufacturing and in hospital instrument sterilization. The Administration’s rule sets a long-term exposure limit but does not set a short-term limit. We find substantial evidence in the record to support ■ OSHA’s long-term exposure limit. We thus affirm that aspect of the rule. We also find, however, that OSHA’s decision not to issue a short-term exposure limit is not supported by the record, and we remand for further consideration on that point. Since we remand for further proceedings, we do not reach the constitutional questions raised by several petitioners and amici.
I. The Statutory and Factual Background
Congress enacted the Occupational Safety and Health Act (OSH Act) “to assure ... every working man and woman in the Nation safe and healthful working conditions.” 29 U.S.C. § 651(b) (1982). To achieve this goal, the Secretary of Labor may hold hearings, id. § 655(b)(3), promulgate standards, id. §§ 655(b)(2), 655(b)(4), conduct inspections, id. § 657, issue citations, id. § 658, and pursue enforcement proceedings, id. §§ 659, 662, 666. The Occupational Safety and Health Administration (OSHA) assists the Secretary in accomplishing these tasks.
In this case, the litigants challenge a standard issued by the Secretary regulating permissible exposure to ethylene oxide (EtO), a chemical of major significance in American industry. Occupational Exposure to Ethylene Oxide, Final Standard, 49 *256Fed.Reg. 25,734, 25,734-35 (1984). The bulk of EtO usage occurs in manufacturing, and a small fraction is consumed in the sterilization of hospital instruments. EtO is a highly reactive gas. Thus, in manufacturing processes employing EtO, tightly closed and highly automated systems are used. Nevertheless, workers can be exposed to the chemical during both shipping and equipment maintenance. While hospital sterilization accounts for only a small portion of the nation’s EtO usage, it is far more hazardous to workers. Workers must enter areas where EtO is, or has been, present. For example, instruments are typically placed in a sterilization chamber which is then flooded with EtO. These sterilization chambers must be purged of the gas before workers can enter. The process is imperfect, and workers risk exposure to the chemical. Id.
In 1968, the American Conference of Governmental Industrial Hygenists (ACGIH) recommended a long-term permissible exposure limit (PEL) of 50 parts EtO per million parts air (ppm) as a time-weighted average (TWA) exposure level over eight hours. ACGIH based this recommendation on two types of studies. Six-month animal inhalation studies found ill effects only at levels at or above 50 ppm. A study of employees exposed for at least a decade to levels of 5 to 10 ppm reported no adverse effects. At the time ACGIH recommended the 50 ppm PEL, researchers had not yet discovered the carcinogenic properties of EtO. Id. at 25,736. In 1971, pursuant to 29 U.S.C. § 655(a) (1982), which permits the promulgation of “national consensus standards,” the Secretary of Labor adopted the 50 ppm PEL as a limit on workers’ exposure.
Soon, however, the body of knowledge surrounding EtO began to grow, and researchers discovered more alarming health effects. In 1977, the National Institute for Occupational Safety and Health (NIOSH)1 concluded that a 75 ppm short-term exposure limit (STEL), in addition to the long-term 50 ppm PEL, was necessary to protect workers. A STEL regulates the maximum average exposure over a shorter period — usually fifteen minutes. NIOSH based its conclusion on studies showing that EtO exposure alters cell genetic material. In 1979, ACGIH began considering a PEL of 10 ppm, based on a number of more recent studies, including a study of actual worker exposure to EtO. These studies reinforced the belief that EtO exposure effects cell genetics. Moreover, a study of workers exposed to the chemical revealed a possible connection between EtO and leukemia, a cancer of the blood.
In 1981, ACGIH lowered its recommended PEL to 10 ppm. Moreover, the Conference designated EtO as a suspected carcinogen and proposed an even lower PEL of 5 ppm. Id. ACGIH based this reconsideration of the PEL on a two-year study of rats exposed to EtO. NIOSH also responded in 1981 to the new data on EtO by issuing a bulletin to employers and employees about the potential carcinogenic hazards posed by EtO. NIOSH recommended a revision in the then-current OSHA standard. Id. at 25,737. In June of 1982, ACGIH recommended a PEL of 1 ppm, to take effect in 1984.
In January of 1982, OSHA issued an advance notice of proposed rulemaking, inviting interested individuals to submit data or comments on revising the OSHA EtO standard. See Advance Notice of Proposed Rulemaking, 47 Fed.Reg. 3566 (1982). After some delay, petitioner Public Citizen Health Research Group (Public Citizen) objected to OSHA’s lack of progress on promulgating an EtO standard. A year of litigation over OSHA’s progress ensued, and this court ordered OSHA to propose a rule by April, 1983. Public Citizen Health Research Group v. Auckter, 702 F.2d 1150 *257(D.C.Cir.1983) (per curiam). The court also expected that the agency would “bring this matter to a close within a year.” Id. at 1154 n. 12. OSHA complied by publishing a proposed rule, suggesting a PEL of 1 ppm (replacing the 1971 standard of 50 ppm) and a short-term exposure limit (STEL) “ranging from 5 to 50 ppm for 30 minutes or less.” Occupational Exposure to Ethylene Oxide, Proposed Rule, 48 Fed. Reg. 17,283, 17,284 (1983).
OSHA held public hearings on the proposed rule in July of 1983. Each aspect of the proposed rule was supported and opposed by a wide range of witnesses. The hearing produced 1600 pages of transcript and over 300 exhibits. 49 Fed.Reg. at 25,-737. Despite this court’s ruling, however, OSHA did not issue a final rule by March of 1984. After some prodding by Public Citizen, OSHA stipulated that it would complete the rulemaking by June 15, 1984. On June 14, OSHA had a final rule in hand. It provided for a 1 ppm PEL eight-hour TWA and a 10 ppm STEL fifteen-minute TWA. See Proposed Final Standard, reprinted in Joint Appendix (J.A.) at 12. OSHA did not publish that final rule.
Pursuant to Executive Order No. 12,291, 3 C.F.R. 127 (1981), reprinted in 5 U.S.C. § 601 note, at 431 (1982), OSHA transmitted its final rule to the Office of Management and Budget (OMB) for review. Executive Order 12,291 provides that to the extent permissible by law, regulations of agencies within the executive branch must comply with certain substantive requirements. The Order requires agencies to consider cost/benefit analysis, cost effectiveness, maximization of benefit to society, condition of the regulated industries, and conditions of the national economy. Exec. Order No. 12,291, § 2. Moreover, the Order requires agencies to prepare a “Regulatory Impact Analysis,” a document evaluating the proposed regulations in light of the above substantive requirements. This analysis must be sent to OMB. Id. § 3. Finally, agencies must defer final action until they have responded to OMB’s view. Id. § 3(f)(1). In this case, OMB questioned several aspects of OSHA’s final rule, especially the benefits that the EtO rule would provide to society and its cost-effectiveness. These objections applied to both the PEL and the STEL. OMB objected to the STEL in particular as unsupported by any reasonable risk assessment or inference from the available scientific evidence. Letter from Christopher De-Muth, OMB, to Francis Lilly, Dep’t of Labor (June 14, 1984), reprinted in J.A. at 415.
Notwithstanding OMB’s objections, OSHA published as a final rule the long-term limits embodied in the PEL. The final rule requires employers to ensure that their employees are not exposed to an airborne concentration of EtO in excess of 1 ppm as an eight-hour TWA. Occupational Exposure to Ethylene Oxide, Final Standard, 49 Fed.Reg. 25,734, 25,796 (1984) (codified at 29 C.F.R. § 1910.1047 (1985)). Employers must monitor exposure levels regularly, depending on the concentration of EtO in the air and changes in work practices. The rule sets an “action level” of 0.5 ppm, which if detected, requires employers to engage in regular employee monitoring. The rule also requires employers to inform employees of monitoring results, institute engineering controls to reduce the level of exposure, and, if necessary, provide respiratory protective equipment for employees. If the EtO exposure level is above the 1 ppm PEL, the employer is required to establish a written compliance program to achieve an exposure level below 1 ppm. Employers must institute a medical surveillance program for employees exposed to EtO at or above the action level, and the employer must provide medical examinations to certain employees. Finally, the employer must embark upon an employee information program that includes signs, labels, and training programs. All of these requirements are enforced through specific recordkeeping requirements. 49 Fed.Reg. at 25,796-800.
While OSHA did not heed OMB’s objections with regard to the PEL, it did reserve judgment on the STEL. Rather than issue the STEL as part of the final EtO standard, *258OSHA reopened the record for comments on the desirability of a STEL. Commentary was once again divided, and on January 2, 1985, OSHA issued its final rule declining to impose a STEL. Occupational Exposure to Ethylene Oxide, Supplemental Statement of Reasons, 50 Fed.Reg. 64 (1985). This action contrasted with OSHA’s earlier decision, prior to receiving OMB’s comments, to issue a 10 ppm STEL.
Petitioner Public Citizen launches a two-pronged attack on OSHA’s decision not to issue the STEL. First, Public Citizen claims that the decision not to issue a STEL is unsupported on the record. The Ethylene Oxide Industry Council (EOIC) appears as amicus to support OSHA’s decision not to issue a STEL. Second, Public Citizen argues that OMB’s role in these proceedings was unlawful. In the latter argument, Public Citizen is joined by several members of the House of Representatives as amici.
The Association of Ethylene Oxide Users (AEOU) also petitions this court for review, arguing that OSHA’s decision to issue the PEL is unsupported on the record.
Our opinion treats four basic aspects of this case. First, we set out the appropriate substantive legal requirements applicable to an OSHA standard, as well as the standard of review courts apply to OSHA rule-making. Second, we review the evidence supporting OSHA’s decision to issue the 1 ppm PEL. Third, we consider the evidence supporting OSHA’s decision not to issue a 10 ppm STEL. Finally, because we. find OSHA’s decision not to issue a STEL unlawful, we decline to pass on the legality of OMB’s participation in the rulemaking process.
II. Legal Aspects op OSHA Standard-Setting
A. Legal Requirements Applicable to OSHA Standards
Although OSHA enjoys a broad congressional mandate to ensure safe working conditions, the agency is not unbridled in its ability to regulate industry. The Secretary may promulgate “occupational safety and health standards,” defined as “standard[s] which require[ ] conditions, or the adoption or use of one or more practices, means, methods, operations, or processes, reasonably necessary or appropriate to provide safe or healthful employment and places of employment.” 29 U.S.C. § 652(8) (1982). If the Secretary chooses to regulate “toxic materials or harmful physical agents,” he must:
set the standard which most adequately assures, to the extent feasible, on the basis of the best available evidence, that no employee will suffer material impairment of health or functional capacity even if such employee has regular exposure to the hazard dealt with by such standard for the period of his working life. Development of standards under this subsection shall be based upon research, demonstrations, experiments, and such other information as may be appropriate. In addition to the attainment of the highest degree of health and safety protection for the employee, other considerations shall be the latest available scientific information in the field, the feasibility of the standards, and experience gained under this and other health and safety laws.
29 U.S.C. § 655(b)(5) (1982).
In Industrial Union Department, AFL-CIO v. American Petroleum Institute, 448 U.S. 607, 100 S.Ct. 2844, 65 L.Ed.2d 1010 (1980) {Benzene), the Supreme Court endeavored to define more precisely the specific contours of the OSH Act. In that case, OSHA promulgated a standard regulating worker exposure to benzene, a proven carcinogen at high exposure levels. At that time, OSHA’s regulations (also referred to as its “Cancer Policy”) allowed the Secretary to assume that there is no safe level of exposure to a known carcinogen; thus OSHA set the PEL at the lowest economically and technically feasible level. Id. at 624, 100 S.Ct. at 2854. OSHA made no attempt whatsoever to estimate the significance of the risks to workers exposed at *259level equal to the lowest feasible PEL. Id. at 631-38, 100 S.Ct. at 2858-02.
The Supreme Court, in a plurality opinion, struck down the benzene standard, reasoning that “[i]n the end OSHA’s rationale for lowering the permissible exposure limit to 1 ppm was based, not on any finding that leukemia has ever been caused by exposure to 10 ppm of benzene and that it will not be caused by exposure to 1 ppm, but rather on a series of assumptions indicating that some leukemias might result from exposure to 10 ppm and that the number of cases might be reduced by reducing the exposure level to 1 ppm.” Id. at 634, 100 S.Ct. at 2860 (emphasis in original). Focusing on the “reasonably necessary and feasible” language of the OSH Act, a plurality of four Justices held that the Act does not allow the Secretary “to require employers to provide absolutely risk-free workplaces whenever it is technologically feasible to do so, so long as the cost is not great enough to destroy an entire industry.” Id. at 641, 100 S.Ct. at 2863. Rather, the OSH Act requires the Secretary to make a threshold finding that the health risk posed by the substance is significant. Id. at 639-40, 641, 642, 100 S.Ct. at 2862-63, 2864.
Once the threshold finding of significant risk has been made, OSHA’s standard must be reasonably necessary and appropriate to remedy that risk. Id. at 639, 642, 100 S.Ct. at 2862, 2864 (plurality opinion); id. at 665, 100 S.Ct. at 2875 (opinion of Powell, J.)
B. The Scope of Judicial Review
The statutory directive to courts reviewing standards adopted pursuant to the OSH Act is clear: the Secretary’s determinations shall be sustained if they are “supported by substantial evidence in the record considered as a whole.” 29 U.S.C. § 655(f) (1982). The substantial evidence test is a familiar approach for courts reviewing agency action. The Supreme Court has defined substantial evidence as “such relevant evidence as a reasonable mind might accept as adequate to support a conclusion.” Universal Camera Corp. v. NLRB, 340 U.S. 474, 477, 71 S.Ct. 456, 459, 95 L.Ed. 456 (1951) (quoting Consolidated Edison Co. v. NLRB, 305 U.S. 197, 229, 59 S.Ct. 206, 217, 83 L.Ed. 126 (1938)). The appellate court therefore does not review the evidence de novo. Even if the evidence is susceptible of two inconsistent inferences, the agency must be upheld if a reasonable person could come to either conclusion on that evidence. See Consola v. Federal Maritime Commission, 383 U.S. 607, 620, 86 S.Ct. 1018, 1026, 16 L.Ed.2d 131 (1966).
Statutes like the OSH Act, however, create a peculiar problem for reviewing courts. The statute instructs the reviewing court to apply the substantial evidence test, which is normally reserved for formal adjudications of fact. OSHA rulemaking, however, is a hybrid, combining formal and informal aspects of decisionmaking and including essentially legislative tasks as well. See Industrial Union Department, AFL-CIO v. Hodgson, 499 F.2d 467, 472-74 (D.C.Cir.1974). As we noted in United Steelworkers v. Marshall, 647 F.2d 1189, 1206-07 (D.C.Cir.1980), cert. denied, 453 U.S. 913, 101 S.Ct. 3148, 69 L.Ed.2d 997 (1981):
The peculiar problem of reviewing the rules of agencies like OSHA lies in applying the substantial evidence test to regulations which are essentially legislative and rooted in inferences from complex scientific and factual data, and which often necessarily involve highly speculative projections of technological development in areas wholly lacking in scientific and economic certainty.
Because of this hybrid approach, courts view the substantial evidence test in this context as requiring OSHA “to identify relevant factual evidence, to explain the logic and the policies underlying any legislative choice, to state candidly any assumptions on which it relies, and to present its reasons for rejecting any significant contrary evidence and argument.” Id. at 1207.
This deferential standard in no way conflicts with the substantive legal requirements established by the Benzene case. Benzdne established that the burden of proving that the subject of the proposed *260regulation presents a significant health risk remains with the agency. 448 U.S. at 653, 100 S.Ct. at 2869. Three of the Justices in the plurality in the Benzene case, however, were careful to preserve courts’ traditional deference to agency decision-making. Thus, “the requirement that a ‘significant’ risk be identified is not a mathematical straitjacket. It is the Agency’s responsibility to determine, in the first instance, what it considers to be a ‘significant’ risk.” Id. at 655, 100 S.Ct. at 2870. The plurality opinion further stated that “OSHA is not required to support its finding that a significant risk exists with anything approaching scientific certainty.” Id. at 656, 100 S.Ct. at 2871. Noting that the statute allows the Secretary to regulate on the basis of the “best available evidence,” 29 U.S.C. § 655(b)(5) (1982), the opinion stated that “a reviewing court [must] give OSHA some leeway where its findings must be made on the frontiers of scientific knowledge.” Id.
With these considerations in mind, we turn to the evidence adduced in the proceedings below.
III. Evidence of EtO’s Adverse Health Consequences
OSHA first determined whether exposure to EtO at the then-current PEL of 50 ppm presented adverse health effects, a necessary prerequisite to the finding of significant risk required by the Benzene case. AEOU challenges OSHA’s findings, arguing that the Agency lacks substantial evidence for its conclusions. We have reviewed the evidence in the record and AEOU’s challenges to that evidence. We find OSHA’s findings on the adverse health consequences of EtO exposure supported by substantial evidence in the record as a whole. In this part, we treat only the substantial evidence question. In later parts of this opinion, we review OSHA’s assessment of the magnitude of these health risks. See infra pp. 1493-99.
A. Evidence of Carcinogenicity.
OSHA concluded that there is a relationship between exposure to EtO and a significant increase in the risk of death from cancer. 49 Fed.Reg. at 25,738. The agency based its finding on two types of evidence: epidemiological studies and experimental studies.
1. Epidemiological Evidence.
OSHA relied on a study by Morgan and two studies by Hogstedt for evidence of “a possible association between occupational exposure to ethylene oxide and death from leukemia.” 49 Fed.Reg. at 25,740.
Morgan studied the mortality levels of workers employed at an EtO chemical plant over the course of twenty-two years. Morgan, Claxton, Divine, Kaplan & Harris, Mortality Among Ethylene Oxide Workers, 23 J. Occupational Med. 767 (1981), reprinted in J.A. at 1024. Measurements taken at the plant in 1977 showed no detectable EtO levels in most of the production areas. Measurements elsewhere at the plant showed levels well below 50 ppm. Id. at 768, J.A! at 1025. The mortality rate among the workers was less than expected from such a population sample and no incidence of leukemia was reported. The study did find, however, a significant increase in pancreatic cancer and Hodgkin’s disease. Id. at 769, J.A. at 1026.
Hogstedt I revealed three cases of leukemia occurring between 1972 and 1977 among 230 workers at a Swedish factory where EtO was employed to sterilize hospital equipment. Hogstedt, Malmqvist & Wadman, Leukemia in Workers Exposed to Ethylene Oxide, 241 J.A.M.A. 1132 (1979), reprinted in J.A. at 577. Only 0.2 deaths were expected in such a population during that time span. In 1977, the EtO level in the area where two of the employees who died had worked was estimated at 10-30 ppm. Id. at 1133, J.A. at 578.
Hogstedt II examined 243 workers who had worked at a Swedish EtO production facility for at least one year. Hogstedt, Rohlen, Berndtsson, Axelson & Ehrenberg, A cohort study of mortality and cancer incidence in ethylene oxide production workers, 36 Brit.J.Indus.Med. 276 (1979), *261reprinted in J.A. at 642. The workers were tracked for 16 years, and the study revealed a significant excess mortality rate among employees who worked full-time in EtO production areas. The authors found no excess mortality among part-time or intermittently exposed workers. Id. at 278, J.A. at 644.
OSHA admitted that each of these studies was flawed in some way. 49 Fed.Reg. at 25,738-40. AEOU renews here its arguments against the probative value of these studies, urging us to fault the agency for relying on them at all. See Petitioner AEOU Brief at 24-27. OSHA, however, did not blindly rely on these studies; the agency recognized and accounted for the methodological weaknesses inherent in the studies.
The agency heard conflicting testimony on the Morgan study, including a statement from Morgan himself that the study failed to support a causal link between EtO and leukemia and thus should be considered as negative evidence on EtO’s carcinogenic properties. The study clearly demonstrates, however, a link between EtO and pancreatic cancer and Hodgkin’s disease. As some commenters suggested, the study underestimated the cancer risk, see Hearing Transcript (Tr.) at 792, J.A. at 1796, the study was inconclusive rather than negative on leukemia risk, Tr. at 341, J.A. at 1789, and the study was “a strong piece of evidence indicating that even in very small cohorts, with exposures well below the current OSHA standard, excess cancer risk ... was detected.” 49 Fed.Reg. at 25,739 (quoting Submission of Dr. Jeanne Stellman at 8, J.A. at 1001).
Commenters criticized Hogstedt I because of the small sample size, see, e.g., Tr. 634, J.A. at 1794, and the fact that the subject workers .were exposed to methyl for-mate along with EtO, see, e.g., Health Industry Manufacturers Association (HIMA) Comments at 25, reprinted in J.A. at 1151, 1182; EOIC Hazard Assessment of Ethylene Oxide at 112, reprinted in J.A. at 1443, 1464. OSHA, however, noted that it had no evidence that methyl formate was carcinogenic. Hogstedt II faced similar criticism: workers were exposed to multiple chemical combinations along with EtO. See, e.g., HIMA Comments at 25, J.A. at 1182. Despite these criticisms, only two commenters concluded that the evidence did not establish an increased risk of cancer. Even one of those commenters stated that the studies “may not provide as much reassurance [that EtO is not a carcinogen] as some would like.” 49 Fed.Reg. at 25,-740 (quoting Submission of Robert W. Morgan). In contrast, NIOSH concluded that the Hogstedt studies provide evidence of a possible relationship between EtO exposure and leukemia. See Tr. at 341, J.A. at 1789; NIOSH Current Intelligence Bulletin 5 (May 22, 1981), reprinted in J.A. at 619, 625.
We cannot accept AEOU’s proposition that these studies are rendered totally valueless by their methodological flaws. While each study does suffer from defects, OSHA did not rely on the epidemiological studies as conclusive evidence of the carcinogenic nature of EtO. Rather, OSHA found “that the epidemiological evidence, although not by itself conclusive, is supportive of EtO’s potential carcinogenic ... effects.” 49 Fed.Reg. at 25,740. Moreover, OSHA did not reach this position in the face of unified, contrary scientific thought. Indeed, the record demonstrates that OSHA’s acceptance of these studies for some, but not conclusive, evidence of EtO’s potential carcinogenic effects was supported by a substantial portion of the scientists who attended the rulemaking hearings.
These epidemiological studies, while of some value by themselves, take on added significance when viewed in light of the experimental evidence in the record.
2. Experimental Evidence.
OSHA relied on two studies examining the effects of EtO exposure on animals for further evidence of the chemical’s carcinogenicity: the Bushy Run rat study and the NIOSH rat and monkey study.
*262In the Bushy Run study, researchers at the Bushy Run Research Center in Pittsburgh exposed rats to EtO at concentrations of 100, 33, and 10 ppm for 6 hours per day, 5 days per week. Snellings, Weil & Maronpot, Final Report, Ethylene Oxide Two-Year Inhalation Study on Rats at v (unpublished manuscript Jan. 28, 1981), reprinted in J.A. at 579, 583. This study produced a number of significant results, indicating that EtO exposure was related to development of various types of cancers. See 49 Fed.Reg. at 25,740-41. Many commenters objected to four particular aspects of the Bushy Run study.
First, the rats suffered a viral infection during the course of the study. Some commenters suggested that the infection could have adversely affected the immune systems of the rats, thus invalidating the results. See id. at 25,741 (noting criticism). OSHA concluded, however, based on testimony of the director of the study and an expert pathologist from the National Cancer Institute, that the outbreak had no substantial effect on the outcome of the study. Id.
Second, other commenters suggested that the type of leukemia produced in the Bushy Run rats had no counterpart in other strains of rats, in mice, or in humans. Id. OSHA specifically rejected this assertion, relying on expert testimony that recent studies had described similar leukemia in humans. Id. (citing testimony of Jerrold M. Ward).
Third, EOIC suggested that the only types of tumors appearing in the test rats were types that occur in rats spontaneously. This fact indicated that ETO might promote tumor growth, rather than initiate it. Id. OSHA heard expert testimony that tumor-promoters can simultaneously be tumor-initiators, and thus carcinogenic, as well. Id. (citing testimony of Jerrold M. Ward). The same expert stated that tumor-promoters might be as dangerous in ultimately causing cancer as chemicals that are not categorized as promoters. Id. at 25,742.2
Finally, OSHA rejected an argument that the statistical analyses performed on the Bushy Run data were misleading. See, e.g., HIMA Comments App. H at 5-7, J.A. at 1293-95. OSHA confirmed the Bushy Run researchers’ statistical methodology and specifically reanalyzed the evidence under a “worst case” theory, concluding that the results were essentially accurate. Thus, OSHA found that the Bushy Run study demonstrates that exposure to EtO significantly increases mortality and the incidence of tumors among rats. 49 Fed. Reg. at 25,742.
The second experimental study was a two-year study of rats and monkeys conducted by NIOSH. Chronic Inhalation Toxicity of Ethylene Oxide and Propylene Oxide in Rats and Monkeys — A Preliminary Report (Feb. 17, 1982), reprinted in J.A. at 1131. These animals were exposed to EtO at concentrations of 50 and 100 ppm. During the study, however, the rats contracted an infection that killed a large portion of the group. Once the infection had dispersed, researchers resumed the study. 49 Fed.Reg. at 25,742.
The NIOSH study revealed an increased incidence of leukemia in rats exposed to 50 ppm EtO, but not in rats exposed to 100 ppm EtO. See J.A. at 1144. The study did reveal a relationship in rats between tumor production and EtO exposure. The results paralleled those of the Bushy Run study. See 49 Fed.Reg. at 25,743. None of the monkeys demonstrated any evidence of leukemia, but there was some evidence of central nervous system damage. Id.
Significantly, “[t]he overwhelming majority of comments on the NIOSH study agreed with OSHA’s conclusions that these preliminary results provide additional evi*263dence of EtO’s carcinogenicity in experimental animals.” Id. Two commenters stated that taken together, the Bushy Run and NIOSH studies produced consistent, disturbing results with regard to tumor production. Id. (statements of Leon Goldberg and Jerrold Ward).3
3. OSHA’s Conclusions.
Taking the epidemiological and experimental studies together, OSHA found that EtO causes cancer in laboratory animals and poses a significant cancer risk for humans. While each study individually may not be a model of textbook scientific inquiry, the cumulative evidence is compelling. OSHA takes precisely this view: “[Although these studies do not provide definitive evidence of carcinogenicity, they are suggestive of an association between occupational exposure to EtO and cancer (leukemia) mortality.” 49 Fed.Reg. at 25,-744.
This court’s role is not to review the evidence de novo to arrive at our own estimate of the risks; rather, we look for substantial evidence supporting OSHA’s finding of EtO carcinogenicity. 29 U.S.C. § 655(f) (1982). A reasonable person could draw from this evidence the conclusion that exposure to EtO presents a risk of cancer. Thus, the substantial evidence test is met. Even if a reasonable person could also draw the opposite conclusion, we must uphold the agency’s findings. See Consolo v. Federal Maritime Commission, 383 U.S. 607, 620, 86 S.Ct. 1018, 1026, 16 L.Ed.2d 131 (1966).
B. Noncarcinogenic Health Effects
OSHA also found that EtO exposure is related to cell genetic damage and to reproductive problems. We review these conclusions in turn.
1. Mutagenic and Cytogenic Effects.
In its preamble to the proposed EtO rule, OSHA cited evidence that exposure to the chemical could cause mutations in genetic material serious enough to damage health. According to NIOSH, mutagenicity “is widely accepted as an indicator that the chemical may have carcinogenic potential.” NIOSH Current Intelligence Bulletin 6 (May 22, 1981), reprinted in J.A. at 626. OSHA also reviewed evidence regarding EtO’s cytogenicity. “Cytogenic effects include alterations in a cell’s genetic material that, while detectable, do not necessarily constitute health damage.” Respondent’s Brief at 28 n. 19. The adverse risks associated with cytogenic effects, if any, are not yet determinable. Id. We review both the experimental and epidemiological evidence supporting OSHA’s conclusion that EtO is both mutagenic and cytogenic.
a. Experimental Evidence
OSHA examined five experimental studies and concluded that the accumulated data from these studies strongly implies that EtO exposure may cause genetic damage and cytogenic effects, at least in rats, mice, monkeys, and rabbits.4 AEOU does *264not directly challenge the scientific value or methodologies of these studies. See supra pp. 1489-90, infra pp. 1493-94.
b. Epidemiological Evidence.
OSHA also reviewed the evidence of mutagenic and cytogenic effects in humans. At least seven separate investigations examined this relationship, but OSHA relied most heavily on four studies.
In Pero I, researchers studied female employees at a Swedish plant manufacturing disposable surgical instruments. Pero, Widegren, Hogstedt & Mitelman, In Vivo and In Vitro Ethylene Oxide Exposure of Human Lymphocytes Assessed by Chemical Stimulation of Unscheduled DNA Synthesis, 83 Mutation Research 217 (1981), reprinted in J.A. at 1100. The study indicated that EtO exposure was related to increased chromosomal aberration and inhibited DNA repair. Pero II examined male workers exposed to EtO and found a significant decrease in DNA repair efficiency related to EtO exposure. Pero, Bryngelsson, Widegren, Hogstedt & Welinder, A reduced capacity for unscheduled DNA synthesis in lymphocytes from individuals exposed to propylene oxide and ethylene oxide, 104 Mutation Research 193 (1982), reprinted in J.A. at 1092.
Johnson & Johnson, Inc. initiated a study of workers exposed to EtO in three of its plants. See J.A. at 971; id. at 1646. The workers were exposed to levels below 1 ppm, between 1 and 10 ppm, and between 5 and 200 ppm. The study revealed that excess chromosomal aberrations occurred as a result of exposure, and the aberration level was related to the exposure level. See J.A. at 983.
The Johnson & Johnson result was confirmed by the Yager study. Yager, Sister Chromatid Exchanges as a Biological Monitor for Workplace Exposure: A Comparative Study of Exposure-Response to Ethylene Oxide (unpublished manuscript, March 1982), reprinted in J.A. at 798. Yager’s examination of sterilizer operators in two hospitals revealed chromosomal aberrations related to EtO cumulative doses. See also Exposure to Ethylene Oxide at Work Increases Sister Chromatid Exchanges in Human Peripheral Lymphocytes, 219 Science 1221 (1983).
AEOU does not directly challenge the methodology of these studies. See Petitioner AEOU Brief at 28 & n. 22. To the extent, however, that AEOU generally challenges OSHA’s evidentiary basis for concluding that EtO is dangerous, we evaluate the usefulness of these particular studies. The studies were challenged by other participants, and OSHA responded to the criticisms, finding the studies useful evidence of EtO’s adverse effects. See 49 Fed.Reg. at 25,745.
As with the studies of carcinogenicity, these studies are not models of scientific investigation. The cumulative impact of these studies, however, is quite compelling. Each study corroborates the others to some extent. While we might be hesitant to accept a single study as substantial evidence for an inference that EtO exposure causes chromosomal aberrations, see, e.g., Gulf South Insulation v. United States CPSC, 701 F.2d 1137, 1146 (5th Cir.1983) (single study rejected in face of significant contrary evidence), the combination of experimental and epidemiological evidence constitutes sufficient authority for such a finding. We decline to hold that this evidence cumulatively has so little scientific value that OSHA could not rely on it.
c. Health Effects of Mutagenicity and Cytogenicity.
AEOU’s main attack on this evidence falls not on the studies themselves, but on their relevance. AEOU urges this court to *265reject the studies because, AEOU claims, the weight of reputable scientific thought does not support the proposition that genetic changes are either unhealthful or indicative of carcinogenic properties. See Petitioner AEOU Brief at 28-30.
OSHA found that EtO’s demonstrated effects on chromosomes supported its conclusion that the chemical is carcinogenic. 49 Fed.Reg. at 25,748. OSHA supported its conclusion that EtO’s mutagenetic and eytogenic properties were unhealthful with substantial evidence. An OSHA expert testified that chemicals that cause genetic damage may generate “serious adverse health outcomes.” Id. (statement of Marvin S. Legator). OSHA admitted that it could not establish quantitative assessments of these health effects given the current state of knowledge in the area, but concluded that the available information was cause for concern, especially in light of the other demonstrated dangers of EtO. Id.
Other commenters cited a consensus in the scientific community that no relationship between chromosomal damage and adverse health effects had been established in general and with regard to EtO in particular. At least one expert, however, stated that many chemicals known to cause chromosomal aberrations are known carcinogens. Id. at 25,747-48 (statement of Janice Yager). Moreover, an OSHA scientist stated that the relationship between genetic damage and adverse health effects is stronger in EtO’s case because of the chemical’s established effect in animals and its known mutagenic and eytogenic effects in humans. Id. (statement of Marvin S. Legator). Thus, although the evidence on the link between genetic damage and adverse health effects was equivocal, OSHA found that these data on genetic damage supported its finding that EtO is carcinogenic and produces adverse health consequences.
AEOU attacks this conclusion as both unsupported and outdated. AEOU cites testimony that indicates that chromosomal damage has not yet been connected to adverse health effects. This testimony, AEOU argues, directly contradicts OSHA’s conclusion. See Petitioner AEOU Brief at 28-30.
We find this view of the evidence unduly restrictive. The evidence hardly falls into directly opposing camps. Some experts testified that no connection had yet been established between chromosomal damage and health effects. OSHA’s expert stated that the studies of EtO exposure on animals, coupled with the proven reactive nature of EtO in the human body, provided a “unique situation.” Id. at 25,748 (statement of Marvin S. Legator). This testimony, together with Yager’s conclusion that mutagenic and eytogenic chemicals are carcinogenic, 49 Fed.Reg. at 25,747-48, supports an inference that EtO may indeed cause harmful health effects. OSHA recognized the equivocal nature of this evidence. The agency never stated that this particular evidence conclusively establishes the harmful health effects of EtO. Rather, the agency stated that the evidence clearly establishes eytogenic effects (from which it inferred harmful health consequences) and supported its conclusion that EtO is a carcinogen.
AEOU also argues that OSHA’s expert testimony is “clearly” outdated. Petitioner AEOU Brief at 29. AEOU cites the Interdisciplinary Panel on Carcinogenicity, an international group of scientists, for the proposition that genetic mutation tests are no longer seen as accurate predictors of carcinogenicity. We quote the panel’s observation:
In the main these tests are effective at measuring their intended genetic end points and, when used in batteries, are effective for identifying genetic effects of chemical toxins. It is less clear how well these tests identify chemical carcinogens or how they should be used in the absence of corroborative data on carcinogenicity.
The initially high correlation observed between genetic change and carcinogenicity has decreased with the enlargement of the set of chemicals tested and with the *266separation of test development from test deployment.
Interdisciplinary Panel on Carcinogenicity, Criteria for Evidence of Chemical Carcinogenicity, 225 Science 682, 684 (1982) (emphasis added).
Contrary to the implication of petitioner’s argument, the Panel has not determined that genetic change tests have no value. The Panel merely notes that the tests are now deemed less predictive than once thought. The Panel nowhere states that any reliance on such tests is outdated. Rather, the Panel states what OSHA candidly admitted: genetic mutation data alone does not establish that EtO is harmful.
Moreover, OSHA simply did not rely on the genetic damage data alone for the conclusion that EtO is harmful. The agency was careful to state that the data on cytogenic and mutagenic changes “support” the agency's previous conclusion that EtO is carcinogenic. The Interdisciplinary Panel’s Criteria cautions against uncorroborated reliance on mutation tests. OSHA’s approach comports with that admonition. OSHA’s broad-based inquiry into the deleterious effects of EtO had already established EtO’s carcinogenic potential before the agency considered the genetic damage data.
2. Reproductive Effects.
OSHA determined that in addition to its carcinogenic, mutagenic, and cytogenic effects, EtO poses reproductive hazards. The agency relied on both experimental and epidemiological studies for this result. We summarize here the evidence itself, AEOU’s claims that the evidence is unreliable, and our resolution of those claims.
a. Experimental Studies.
Four studies analyzed the effect of EtO exposure on reproduction in rodents, and two analyzed the effects on reproduction in rabbits. The rabbit studies were inconclusive. 49 Fed.Reg. at 25,748. The rodent studies demonstrated that at high doses, EtO exposure during gestation significantly increases the number of fetal deaths and malformed offspring. Pregestation exposure of both parents followed by gestation exposure of the female also produced fetotoxicity.5 In addition, one study indicated that EtO exposure leads to reduced male rat fertility. Two experimental studies examining mutagenicity and cytogenicity also demonstrated both genetic damage to male reproductive organs and reduced sperm counts. 49 Fed.Reg. at 25,748; see supra n. 4. Moreover, OSHA analyzed the Bushy Run cancer study and discovered that “adverse reproductive responses could be correlated with specific EtO exposure levels.” 49 Fed.Reg. at 25,748. AEOU does not challenge this aspect of these studies.
b. Epidemiological Studies.
The primary study regarding the effect of EtO exposure on human reproduction is the Hemminki study of female hospital employees. Hemminki, Mutanen, Saloniemi, Niemi & Vainio, Spontaneous abortions in hospital staff engaged in sterilising instruments with chemical agents, 285 Brit. *267Med.J. 1461 (1982), reprinted in J.A. at 1031. We discuss the Hemminki study in some detail because AEOU was just one of many commenters that objected to various aspects of the study.
Researchers mailed questionnaires to 80 supervising nurses throughout Finland. The survey asked the nurses to identify all female personnel involved in instrument sterilization activities. Six months later, researchers sent questionnaires to women involved in instrument sterilization and to a control group of unexposed female hospital workers. Id. The researchers attempted to confirm the survey data by examining hospital discharge registers, which were available for all Finnish patients between 1973 and 1979. Id.
The questionnaires yielded some significant results. Hemminki studied the connection between exposure to sterilizing agents in general6 and spontaneous abortion, a severe reproductive abnormality. Hemminki reported no significant difference in spontaneous abortion rates between women exposed to sterilizing agents and the control group. The researchers discovered a significant increase in spontaneous abortions among women exposed to sterilizing agents while pregnant, in contrast to women exposed to the agents while not pregnant. Id. at 1462, J.A. at 1032 (Table I).
The Hemminki study also refined the analysis by comparing women exposed to EtO alone with those exposed to other sterilizing agents. The evidence suggested that EtO exposure, either alone or combined with other chemicals, was associated with increased spontaneous abortions. The study revealed no increase in spontaneous abortions due to exposure to chemicals other than EtO, with one apparent exception. Id. (Table II).7
The Hemminki researchers thought that the study might be flawed because the data revealed an increasing level of spontaneous abortions throughout the population from 1950 through 1981. This increase was attributed to either the increasing age of the subject population or the failure of the subject to recall spontaneous abortions that occurred twenty to thirty years earlier. Id. at 1462, 1462-63, J.A. at 1032, 1032-33. Also, workers exposed to EtO when not pregnant exhibited a lower rate of spontaneous abortion than did the control group over the three decades. Id. at 1462, J.A. at 1032 (Table I). This fact, taken alone, would suggest that there is no relationship between EtO exposure and spontaneous abortion. OSHA theorized, however, that the lower rate might have been due to the employment histories of unexposed workers, many of whom were not yet employed by the hospital during their initial pregnancies. 49 Fed.Reg. at 25,749.
To corroborate these findings, Hemminki compared the questionnaire results with the discharge registers of Finnish hospitals for the period 1973-1979. This subsequent investigation confirmed the finding that EtO exposure is related to spontaneous abortions. 285 Brit.J.Med. at 1462, J.A. at 1032 (Table III).8
Several commenters sharply disputed the value of this study during the rulemaking, and AEOU renews those criticisms here. AEOU Brief at 30-33. OSHA considered criticisms that the subject knew the purpose of the study (thus biasing the selection of the group and invalidating the results), that the study lacked controls on extraneous causes of spontaneous abortions, and that the study lacked data on levels of EtO exposure level and duration. OSHA responded to each of these concerns.
*268Several commenters pointed out possible biases in the Hemminki study. See 49 Fed. Reg. at 25,750-52. If the subjects knew of the study’s purpose, they might have been more likely to report a spontaneous abortion. At least two observers, however, stated that any bias effect was minimal, especially given the corroboration evidenced by the hospital registers. Id. at 25,750 (citing Letters of Jennie Kline and Bernard Pasternak). Other commenters postulated that the study might be biased because some of the sterilization employees who were not exposed during pregnancy were not employed when they became pregnant for the first time. In response to this problem, Hemminki reanalyzed the data from the questionnaires, eliminating pregnancies that did not occur during hospital employment. The correlation between EtO exposure and spontaneous abortion remained. Id. Finally, one commenter supposed that if the supervising nurses knew the object of the study, they could identify the subject nurses who had been pregnant at some time during the study, with an eye toward skewing the results. See Austin, Supplemental Comments Relating to the OSHA Hearings on Ethylene Oxide 2-3 (Aug. 24, 1983), reprinted in J.A. at 1692, 1694-95. Hemminki, joined by several commenters, rejected this notion, stating that the requirements for participation in the study were narrowly drawn and that the hospital staffs were sufficiently large that the supervising nurses would not have adequate knowledge of all the nurses under them. 49 Fed.Reg. 25,750-51. With these opposing arguments, OSHA could reasonably conclude that the bias potential, while real, was minimal.
Some commenters suggested that the Hemminki study was flawed because other sources of spontaneous abortion (i.e., age, habits, prior history of spontaneous abortion) were not fully controlled. At least three commenters found serious flaws in the lack of controls, see 49 Fed.Reg. at 25,751, but another stated that the impact of the lack of controls on the study’s findings was uncertain, id. Moreover, OSHA found that at least some uncontrolled factors, such as alcohol, coffee, and cigarette consumption, have an effect only later in pregnancies. Finally, OSHA assumed that the uncontrolled factors were distributed randomly among all subjects of the study, including the control groups. This would minimize any confounding effect. Id.
Commenters also criticized the unit of examination used in the Hemminki study. The Hemminki researchers examined pregnancies, rather than women, as the unit of analysis. Commenters suggested that since women who have had one spontaneous abortion suffer an increased risk of subsequent spontaneous abortions, the study might unfairly report the number of spontaneous abortions attributable to EtO. See, e.g., Austin, Supplemental Comments, supra, at 4, J.A. at 1696. OSHA acknowledged the difficulty in resolving this issue. 49 Fed.Reg. at 25,751-52. Despite this criticism, several commenters found the Hemminki results convincing. As OSHA found, “[i]n spite of these analytical problems, many commenters agreed that the results of the Hemminki study, taken in combination with animal data that demonstrate the fetotoxic potential of EtO, suggest that occupational exposure to EtO results in an increased risk of spontaneous abortion.” Id. at 25,752.
AEOU argues that the Hemminki report cannot be substantial evidence because of its methodological flaws. We note, however, that the Hemminki study is just one piece of evidence in the EtO puzzle. OSHA did not ignore the study’s shortcomings; indeed, to the agency’s credit, it pointed out the flaws in the study and answered them where possible. The mere fact that the study is challenged by some, or even a mathematical majority, of commenters, does not render it scientifically or legally useless.
C. Conclusion
AEOU’s attack on the evidence in this rulemaking fundamentally misconstrues the roles that OSHA and reviewing courts *269each must play in regulating toxin exposures. AEOU attacks each piece of evidence, suggesting that no individual piece proves a relationship between EtO exposure and various adverse health effects. This approach disregards the marginal contribution that each piece of evidence makes to the total picture. While some of OSHA’s evidence suffers from shortcomings, such incomplete proof is inevitable when the Agency regulates on the frontiers of scientific knowledge.
We have addressed the problem of cumulative evidence before:
Contrary to the apparent suggestion of some of the petitioners, we need not seek a single dispositive study that fully supports the Administrator’s determination. Science does not work that way; nor, for that matter, does adjudicatory factfinding. Rather, the Administrator’s decision may be fully supportable if it is based, as it is, on the inconclusive but suggestive results of numerous studies. By its nature, scientific evidence is cumulative: the more supporting, albeit inconclusive, evidence available, the more likely the accuracy of the conclusion____ Thus, after considering the inferences that can be drawn from the studies supporting the Administrator, and those opposing him, we must decide whether the cumulative effect of all this evidence, and not the effect of any single bit of it, presents a rational basis for the ... regulations.
Ethyl Corp. v. EPA, 541 F.2d 1, 37-38 (D.C.Cir.) (en banc) (footnote omitted), cert. denied, 426 U.S. 941, 96 S.Ct. 2663, 49 L.Ed.2d 394 (1976). This excerpt demonstrates why OSHA need not “prove” its assertions in the manner AEOU demands, and indeed, why it cannot. Rather, OSHA need only gather evidence from which it can reasonably draw the conclusion it has reached. We in no way denigrate the “searching and careful” nature of our inquiry. See Citizens to Preserve Overton Park v. Volpe, 401 U.S. 402, 416, 91 S.Ct. 814, 824, 28 L.Ed.2d 136 (1971) (referring to more deferential “arbitrary and capricious” standard of review). Our function, however, is only to search for substantial evidence, not proof positive. Furthermore, we do not reweigh the evidence and come to our own conclusion; rather, we assess the reasonableness of OSHA’s conclusion.
Moreover, in cases such as these, “[i]t is not infrequent that the available data do not settle a regulatory issue, and the agency must then exercise its judgment in moving from facts and probabilities on the record to a policy conclusion.” Motor Vehicle Manufacturers Association v. State Farm Mutual Automobile Insurance Co., 463 U.S. 29, 52, 103 S.Ct. 2856, 2871, 77 L.Ed.2d 443 (1983). OSHA faces just such a problem in regulating EtO exposure. The scientific evidence in the instant case is incomplete but what evidence we have paints a striking portrait of serious danger to workers exposed to the chemical. When the evidence can be reasonably interpreted as supporting the need for regulation, we must affirm the agency’s conclusion, despite the fact that the same evidence is susceptible of another interpretation. Our expertise does not lie in technical matters. See Weyerhaeuser Co. v. Costle, 590 F.2d 1011, 1026 (D.C.Cir.1978) (there are “obvious limitations upon the capacity of courts to deal meaningfully with knowledge of this kind”); Industrial Union Department, AFL-CIO v. Hodgson, 499 F.2d 467, 474-75 n. 18 (D.C.Cir.1974) (“[wjhere existing methodology or research in a new area of regulation is deficient, the agency necessarily enjoys broad discretion to attempt to formulate a solution to the best of its ability on the basis of available information”).
We find that OSHA has met its duty “to identify relevant factual evidence, ... to state candidly any assumptions on which it relies, and to present its reasons for rejecting any significant contrary evidence or argument.” United Steelworkers v. Marshall, 647 F.2d 1189, 1207 (D.C.Cir.1980), cert. denied, 453 U.S. 913, 101 S.Ct. 3148, 69 L.Ed.2d 997 (1981). We thus uphold OSHA’s finding that EtO is carcino*270genic, cytogenic, mutagenic, and hazardous to human reproduction.9
IY. Quantitative Risk Assessment
A. Legal Requirements
Once OSHA determined that EtO exposure presents human health hazards, it set out to determine whether the health hazard is significant, as required by Benzene, 448 U.S. at 639, 100 S.Ct. at 2862 (plurality opinion). If the agency finds the risk significant, it must then design a standard reasonably necessary and appropriate to remedying that risk, if possible. See 29 U.S.C. § 652(8) (1982). To make both of these determinations, OSHA endeavored to quantify the risk allowed by the current OSHA EtO standard and the risk allowed by the proposed standard. OSHA employs a mathematical model to accomplish these tasks.
B. OSHA’s Mathematical Model
To calculate the risk facing employees exposed to EtO under both the current and proposed standards, OSHA applied its mathematical model to an experimental database — the Bushy Run rat study. The experimental data-base allows OSHA to estimate the number of humans likely to be adversely affected by EtO exposure at experimental levels, once the animal data is converted into human equivalents. The mathematical model then extrapolates that number into areas where experimental data do not exist.
AEOU challenges both the mathematical model and the conversion of the database from animal results to human equivalents. We first review the database to which the model was applied, and we then examine the validity of the model itself.
1. Use of Animal Studies for the Database.
In order to construct a set of data for its mathematical model, OSHA reviewed the results of the study that produced the most data: The Bushy Run study. In that study, researchers exposed rats to EtO gas at levels of 100, 33, and 10 ppm. OSHA plotted the levels of health effects to rats at each of these exposure rates and derived a dose-response curve. The agency used this curve to extrapolate expected health effects resulting from exposure to levels of EtO that had not been tested in the experiment.
AEOU asserts that any animal study of EtO effects is unreliable for extrapolation to human risk. Petitioner AEOU Brief at 19. AEOU relies on the conclusion of EPA’s Carcinogen Assessment Group (CAG), which in examining the health effects of EtO, noted: “[G]iven the limited data available from animal bioassays, especially at the high dosage levels required for testing, almost nothing can be known about the true shape of the dose-response curve at low environmental levels.” Environmental Protection Agency, Health Assessment Document for Ethylene Oxide, Final Report 9-145 (1985) [hereinafter cited as EPA HAD], AEOU asserts that OSHA’s reliance on animal data contravenes the weight of scientific thought, which recognizes the multiple uncertainties involved in translating animal data to human terms. See Interdisciplinary Panel on Carcinogenicity, Criteria for Evidence of Chemical Carcinogenicity, 225 Science 682, 68&-86 (1984).
At the outset, we note that the validity of EPA’s assessment is not before us. Rather, we consider only OSHA’s decisions and the evidence supporting them. It is nevertheless instructive to review the entirety of EPA’s conclusions:
[T]he uncertainty of present estimations of cancer risks to humans at low levels of exposure should be recognized. The CAG feels that, given the limited data available from animal bioassays, espe*271daily at the high dosage levels required for testing, almost nothing can be known about the true shape of the dose-response curve at low environmental levels. At best, the linear extrapolation model used here provides a rough but plausible estimate of the upper limit of risk; i.e., it is not likely that the true risk is appreciably higher than the estimated risk, but it could very well be considerably lower____ The methods used by the CAG for quantitative assessment are consistently conservative in that they tend to result in high estimates of risk.
Health Assessment Document, supra, at 9-145 (emphasis added). In our view, this added context greatly illuminates the portion selectively quoted by AEOU. EPA’s position appears to be substantially similar to OSHA’s view: while nothing can be known (in the sense of scientific certainty) at this time about the precise biological responses at low exposure levels, estimation techniques can provide a reasonable prediction. That is, estimates based on linear extrapolation are the “best available evidence” that science can currently provide of health risks at low exposure levels.
The Supreme Court plurality in the Benzene case explicitly approved the use of conservative assumptions, 448 U.S. at 656, 100 S.Ct. at 2871 (“so long as they are supported by a body of reputable scientific thought, the Agency is free to use conservative assumptions in interpreting the data”), and implicitly accepted the use of animal studies in the substantial risk determination, 448 U.S. at 657 n. 64, 100 S.Ct. at 2871 n. 64 (“In other proceedings, the Agency has had a good deal of data from animal experiments on which it could base a conclusion of the significance of the risk.”).
Moreover, Congress has impliedly approved estimation techniques when evidence is unavailable to “prove” the harmful health effects posed by toxins. In the OSH Act, Congress expressly allowed OSHA to rely on the “best available evidence,” 29 U.S.C. § 655(b)(5) (1982) (emphasis added) and to consider “the latest available scientifie information in the field,” id. (emphasis added). Indeed, Congress did “not [intend] that the Secretary be paralyzed by debate surrounding diverse medical opinions.” H.R.Rep. No. 91-1291, 91st Cong., 2d Sess. 18 (1970), reprinted in Legislative History of the Occupational Safety and Health Act of 1970 at 848 (1971).
We must grant OSHA some “leeway” when it regulates on the frontiers of current knowledge. See Benzene, 448 U.S. at 656, 100 S.Ct. at 2871 (plurality opinion). We demand no more than that the agency arrive at a reasonable conclusion based on all the evidence before it.
We find that OSHA has developed substantial evidence to support its reliance on the animal duties for the establishment of a database. OSHA did not state that animal studies translate perfectly into human terms. Rather, the agency admitted it was dealing in an area of uncertainty. Moreover, OSHA did not blindly choose to rely on animal data in the face of unified scientific opposition. A number of commenters agreed that in light of the available but incomplete evidence, which suggests that EtO is truly dangerous, the agency could rely only on the animal data. See 49 Fed.Reg. at 24,756 (citing submissions of EOIC and Dr. Kenny S. Crump). Given this record support for using the experimental data, OSHA chose the Bushy Run study because it was the best controlled animal study and the study most susceptible of analysis. We find this a reasonable course to take under the circumstances. AEOU’s position amounts to no more than the argument so clearly rejected by the Supreme Court in Benzene: the agency need not wait until deaths occur to regulate EtO. See Benzene, 448 U.S. at 656 & n. 63, 100 S.Ct. at 2871 & n. 63 (plurality); id. at 690-91, 100 S.Ct. at 2888-89 (Marshall, J., dissenting).
2. The Model’s Assumptions.
OSHA models EtO’s health effects by plotting a curve through known points and extrapolating into areas as yet untested. See generally Proposed Rule and Notice of *272Hearing, 48 Fed.Reg. 17,284, 17,292-96 (1983) (describing OSHA’s risk estimation approach). The model employs several assumptions about the relationship between EtO exposure and biological response. For example, OSHA chose to employ a linear, no-threshold model similar to that used by EPA’s Carcinogen Assessment Group. See EPA HAD, supra, at 9-145. This approach assumes that EtO exposure and biological response vary proportionately, and that there is no threshold level below which EtO exposure produces no adverse health effects.
AEOU challenges OSHA’s model on two grounds. First, AEOU asserts that the model unlawfully assumes that there is no threshold level of EtO exposure. This assumption, AEOU argues, contravenes the teaching of the Benzene case. Second, AEOU claims that the model improperly translates breathing rates for rats into equivalent human terms.
a. Threshold Exposure Level.
In 29 C.F.R. § 1990 (1985) et seq., OSHA provides a framework for rulemaking treatment of occupational carcinogens. The regulations allow OSHA to infer a carcinogenic hazard from one or more positive human or animal studies. 29 C.F.R. § 1990.143 (1985). The regulations also provide that: “No determination will be made that a ‘threshold’ or ‘no-effect’ level of exposure can be established for a human population exposed to carcinogens in general, or to any specific substance.” 29 C.F.R. § 1990.143(h) (1985).10
AEOU presents a two-pronged attack on the threshold issue. First, AEOU charges that the no-threshold assumption violates the Benzene rule because it improperly assumes that EtO is harmful at low doses. Second, AEOU charges that OSHA improperly ignored evidence in the record that demonstrates that EtO does indeed have a threshold level. We treat these contentions in turn.
1. The Significance of Benzene for the No-Threshold Model.
In the Benzene case, 448 U.S. 607, 100 S.Ct. 2844, 65 L.Ed.2d 1010 (1980), the Supreme Court rejected OSHA’s benzene standard on two related grounds. The Court first ruled that the OSH Act requires the agency to find that its proposed standard will remedy a “significant” occupational risk. See supra pp. 1482-1484. The Court then held that no substantial evidence supported OSHA’s assumption that benzene presents a risk of “significant” harm at all possible exposure levels. In fact, “OSHA did not even attempt to carry its burden of proof.” 448 U.S. at 653, 100 S.Ct. at 2870 (plurality opinion). OSHA had made no effort whatsoever to estimate the risk presented by low levels of benzene exposure. See, e.g., id. at 631, 632 n. 33, 100 S.Ct. at 2858, 2859 n. 33. In fact, the agency did not even attempt to construct a dose-responsive curve. The Court simply could not find substantial evidence for a bald assertion devoid of record support. See United Steelworkers v. Marshall, 647 F.2d 1189, 1246-47 (D.C.Cir.1980), cert. denied, 453 U.S. 913, 101 S.Ct. 3148, 69 L.Ed.2d 997 (1981).
AEOU asserts that OSHA’s threshold position in this case is equivalent to OSHA’s position on significant risk in the Benzene case, which the Supreme Court invalidated. In particular, AEOU argues that “[t]he Supreme Court squarely rejected OSHA standard-setting based on ‘probable’ or ‘suspected’ risks.” Petitioner AEOU Brief at 44. AEOU’s implied assertion is that OSHA must unequivocally prove the scientific validity of each and every assumption it employs.
To the extent this argument asserts that OSHA cannot make any assumptions, even if they are supported by scientific thought, *273Congress has clearly come to the opposite conclusion. If Congress had intended to require the agency to “prove” all of its assumptions, Congress would not have allowed the agency to rely on the “best available evidence” and the “latest available scientific information.” 29 U.S.C. § 655(b)(5) (1982).
Moreover, AEOU simply misconstrues the Benzene opinion. In that case, the Court carefully explained that, although OSHA may not act without some record basis, the agency also must be given leeway when regulatory subject matter is not subject to strict proof one way or the other. 448 U.S. at 656, 100 S.Ct. at 2871. The reason for this approach is clear: requiring strict proof would fatally cripple all of OSHA’s regulatory efforts and run counter to the legislative branch’s express delegation of hybrid rulemaking power to OSHA. See Industrial Union Department, AFL-CIO v. Hodgson, 499 F.2d 467, 472-74 (D.C.Cir.1974). Accordingly, in a passage critical to our inquiry here, the Benzene plurality stated: “[S]o long as they are supported by a body of reputable scientific thought, the Agency is free to use conservative assumptions in interpreting the data with respect to carcinogens, risking error on the side of overprotection rather than underprotection.” Id.
In the context of this case, Benzene only requires that OSHA find that EtO exposure presents a substantial risk of harm on the basis of record evidence. In the instant case, OSHA has done exactly what the Supreme Court chastised the agency for not doing in Benzene:
OSHA did not discuss whether it was possible to make a rough estimate, based on the more complete epidemiological and animal studies done at higher exposure levels, of the significance of the risks attributable to those levels, nor did it discuss whether it was possible to extrapolate from such estimates to derive a risk estimate for low-level exposures.
448 U.S. at 632 n. 33, 100 S.Ct. at 2859 n. 33. OSHA has most assuredly rectified that failure in the instant case. The agency has gone to great lengths to calculate, within the bounds of available scientific data, the significance of the risk presented by EtO. The EtO proceeding thus stands in stark contrast to the agency’s actions in regulating benzene exposure.
We think it clear from the above-quoted portion of the Benzene case that the Supreme Court intended to permit the very estimates that AEOU so vigorously attacks. Indeed, it would be anomalous for the Court to have required the agency to provide more than it has provided in regulating EtO. See American Textile Manufacturers Institute, Inc. v. Donovan, 452 U.S. 490, 528 n. 52, 101 S.Ct. 2478, 2500 n. 52, 69 L.Ed.2d 185 n. 52 (1981) (“the agency’s candor in confessing its own inability to achieve a more precise estimate should not precipitate a judicial review that nonetheless demands what the congressionally delegated ‘expert’ says it cannot provide”).
Benzene’s basic teaching is that courts must weigh OSHA’s scientific assumptions under the substantial evidence issue. AEOU attacks OSHA on this score as well, arguing that there is no substantial evidence for OSHA’s decision to use a no-threshold model.
2. Evidentiary Support for the No-Threshold Model.
AEOU points to one commenter who supports an EtO threshold. Dr. Thomas Darby theorized that “[wjith ethylene oxide, like most substances, there appears to be a ‘no effect dose’____” J.A. at 1614. Dr. Darby further assumed that human metabolism can detoxify many chemicals by removing them from the body. J.A. at 1624. According to Darby, the human body can rapidly clear itself of EtO. Building on these assumptions, Darby reached the rather tepid conclusion that “it seems reasonable to believe that a safe level for ethylene oxide exists.” J.A. at 1625.
AEOU argues that the Ethylene Oxide Industry Council (EOIC) supported Darby’s position. EOIC stated that “no adverse health effect has been readily demonstrated in man following exposure to 10 *274ppm____” EOIC Hazard Assessment at v, J.A. at 1445. AEOU also notes that OSHA admitted that the record contains no “direct evidence of an excess risk of cancer at chronic exposure levels below 14 ppm____” Petitioner AEOU Brief at 15 (quoting Memorandum of Points and Authorities in Support of Defendants’ Motion for Summary Judgment 19, Public Citizen Health Research Group v. Auchter, 554 F.Supp. 242 (D.D.C.1983)).11
We find neither Dr. Darby’s nor EOIC’s comments sufficient to refute OSHA’s no-threshold model. Although Dr. Darby advocated the threshold concept, his conclusion was less than forceful. Moreover, contrary to AEOU’s assertion, EOIC’s cited comments do not unhesitatingly support Dr. Darby. EOIC stated that no effects have been reliably demonstrated at levels below 10 ppm. The context of this statement, however, reveals EOIC’s uncertainty with an EtO threshold. In discussing the possible methods of quantifying the risk presented by EtO exposure, EOIC hypothesized three exposure zones. The first zone presents “increased probability” of adverse health effects in man. This zone “perhaps extend[s] down to levels above 10 ppm— even though no adverse health effect has been reliably demonstrated in man following exposure to 10 ppm [EtO].” EOIC Hazard Assessment at v, J.A. at 1445. The second zone “is a zone of uncertain consequences, which may exist for [exposures] within the range of 5-10 ppm, 3-10 ppm, or perhaps 1-10 ppm.” Id. The third zone “is considered to involve insignificant exposure, with no apparent hazard being presented by levels at or below 1 ppm.” This discussion supports no more than the proposition that there may be a threshold exposure level at or near 1 ppm. EOIC’s comments hardly support an attack on a PEL that sets the exposure limit at this threshold because, by definition, exposure above the threshold presents a health risk.
Indeed, a number of participants (including OSHA) joined EOIC in calculating the risk presented by low EtO exposures. Many participants assessed health risks down to exposure levels at 1 ppm and below.12 See 49 Fed.Reg. at 25,756-63.
By citing only the Darby and EOIC comments, AEOU presents only one side of the debate. Moreover, as we have just discussed, even the side of the story presented by AEOU is incomplete. We find AEOU’s citation to contrary authority insufficient to assail OSHA’s position. The exercise of a dispute in the scientific community does not allow this court to choose a particular side as the “right” one. We have no special skills to aid in the resolution of these technical questions. Rather, our role is only to demand that OSHA review all sides of the issue and reasonably resolve the matter. OSHA has the expertise we lack and it has exercised that expertise by carefully reviewing the scientific data. Unlike the Benzene case, OSHA *275here expressly found risk at the 1 ppm level based on evidence submitted by a significant portion of the scientific community.13 We find the no-threshold assumption to be supported by substantial evidence. See also Asarco, Inc. v. OSHA, 746 F.2d 483, 492-93 (9th Cir.1984) (no-threshold assumption upheld).
b. Animal Equivalence.
AEOU challenges another aspect of the model, arguing that OSHA improperly converted animal data into equivalent human responses. The source of the controversy is the Bushy Run rat experiment, the main database for the quantitative risk assessment performed by OSHA. All parties agree that the data from the Bushy Run experiment required recalculation to account for differences between human and rat inhalation rates. Differences in inhalation rates between species are significant because, with the equivalent concentrations of EtO in the atmosphere and the identical exposure times, total exposure for each species depends on its inhalation rate.
OSHA chose to account for the difference in rat and human inhalation rates by converting the Bushy Run dosages into milligrams per kilogram of body weight per day. OSHA chose this measure because one commenter (Dr. Crump) had specifically studied the problem and concluded that the above measure was the most appropriate conversion. 49 Fed.Reg. at 25,760-61.
AEOU does not specifically attack the study on which OSHA relied in making its conversion factor determination. Rather, AEOU attacks a particular statement made by the study’s author at the hearing. We reject this attack.14 Absent some substan*276tial indicia that the underlying study on breathing rates was seriously flawed, we conclude that OSHA could reasonably rely on it as substantial evidence for the conversion factor actually used.
V. Significance of the Risk
A. Agency Findings
Having established the basis for a quantitative risk assessment, OSHA set out to quantify the level of risk faced by workers exposed to EtO. OSHA then analyzed that risk to determine whether it met the test of significance set out in the Benzene case.
Applying its model to the Bushy Run database, OSHA calculated the excess15 risk of death to workers exposed to EtO at various levels. At 50 ppm, OSHA estimated that EtO exposure would cause 634 to 1093 excess deaths per 10,000 workers. 49 Fed.Reg. at 25,762. At 1 ppm, OSHA estimated 12-23 excess deaths per 10,000 workers exposed to EtO. Id.
OSHA found that 634 to 1093 excess deaths at 50 ppm exposure levels is a significant risk within the meaning of the Benzene case. Id. at 25,764. The agency further found that reducing the PEL from 50 ppm to 1 ppm is reasonably necessary and appropriate to remedying the risk at 50 ppm. Id. These particular findings (significant risk and necessity) are not directly challenged on appeal.16
OSHA further found that EtO exposure at 1 ppm (12-23 excess deaths estimated) poses a significant risk itself. The agency set the PEL at 1 ppm, however, not because no excess deaths would occur at that level, but because it could not show that any lower long-term limit would be feasible. Id. at 25,772. AEOU does not challenge OSHA’s finding that the 1 ppm PEL is the lowest feasible limit.
AEOU instead charges that OSHA’s estimates inflate the number of excess deaths attributable to EtO exposure. AEOU prof*277fers two related reasons for the allegedly inflated figures. First, AEOU echoes OMB’s objection that OSHA mistakenly assessed the current exposure patterns of hospital workers. In calculating the current risk to workers, OSHA assumed that 20% of all hospital workers are exposed to at least 20 ppm EtO. AEOU agrees with OMB’s contention that the number of employees exposed at this level is more likely zero to 1.2%. This lower number, if accurate, would necessarily lower the number of excess deaths at the current PEL. OMB puts it more bluntly: “This one, unjustified assumption that 20 percent of hospital workers are exposed at 20 ppm accounts for 156 cancers avoided of OSHA’s lower estimate of 457 cancers avoided in the final [regulatory impact analysis required by Executive Order 12,291] — nearly one-third of the total estimated health benefits of the entire rule.” Petitioner AEOU Brief at 34 (quoting Letter from Christopher DeMuth, OMB to Francis Lilly, Dep’t of Labor, App. at 2-3 (June 14, 1984) (emphasis deleted), reprinted in J.A. at 415, 425-26). AEOU repeats a second OMB objection as well. OMB charges that OSHA erred in assuming that workers are exposed for eight hours per day, five days per week. A more realistic exposure pattern postulates only intermittent exposure. Had OSHA assumed the latter exposure pattern, the 1 ppm PEL would avoid only 125 cancer deaths, rather than OSHA’s final estimate of 634 to 1093 detailed above. Id.
Assuming arguendo that these two challenges have merit, they do not undermine the validity of OSHA’s final standard. OSHA expressly found that the 12-23 excess deaths still caused by EtO exposure at the 1 ppm PEL is a significant risk within the meaning of the Benzene case. That finding is unchallenged here. Assuming 12 deaths are significant, 125 deaths are a fortiori significant. Thus, even if AEOU correctly argues that the current standard allows only 125 excess deaths, the argument is irrelevant. We need not decide, therefore, whether OSHA erred in its risk calculation. Even if it did, the error is insufficient to affect the validity of the outcome.
B. Conclusion
We find that OSHA has complied with the relevant legal standards in promulgating the 1 ppm PEL. While the individual pieces of evidence of EtO’s deleterious health effects are equivocal, the cumulative evidence is compelling. OSHA has surveyed all of this evidence exhaustively. The agency has considered the shortcomings of each submission and responded where possible. Furthermore, the agency has identified the considerations it has found relevant in reaching each of its conclusions. See United Steelworkers v. Marshall, 647 F.2d 1189, 1206-07 (D.C.Cir.1980), cert. denied, 453 U.S. 913, 101 S.Ct. 3148, 69 L.Ed.2d 997 (1981). Given all of this evidence, the agency has estimated the actual risk posed by EtO, as required by the Benzene case. OSHA found that risk significant and found that a 1 ppm PEL is reasonably necessary and appropriate to remedying it. In short, in formulating this part of the EtO standard, OSHA has completed the difficult task of regulating occupational exposure to suspected carcinogens with a thorough and professional approach. “It is difficult to imagine what else the agency could do to comply with [the Supreme] Court’s decision in [the Benzene case].” American Textile Manufacturers Institute v. Donovan, 452 U.S. 490, 506 n. 25, 101 S.Ct. 2478, 2489 n. 25, 69 L.Ed.2d 185 (1981).
VI. 10 PPM Short-Term Exposure Limit (STEL)
Although the notice of proposed rule-making contained a 10 ppm STEL proposal, OSHA’s final EtO standard omitted the STEL. OSHA based its rejection of its requirement on three findings: (1) there is no evidence of a dose-rate effect from EtO exposure (as opposed to an effect based on cumulative dose), (2) the 1 ppm PEL itself will reduce a significant health risk, and (3) compliance with the PEL will necessarily result in control of short-term exposures. *278See generally Occupational Exposure to Ethylene Oxide, Supplemental Statement of Reasons, 50 Fed.Reg. 64 (1985). In this Part, we review these findings. We first examine OSHA’s conclusion that EtO has no demonstrated dose-rate effect. We then examine the relationship between the PEL, the STEL, and cumulative dose.
A. Dose-Rate Effects
The central dispute in the supplemental rulemaking was whether exposure to a dose of EtO over a short period of time produces greater health effects than exposure to the same dose over a longer period. If response depends on the rate at which exposure occurs, then EtO has a dose-rate effect. If response only depends on the amount of dose (irrespective of exposure time), then EtO has only cumulative-dose dependent effects. OSHA noted that “if the facts from exposure can be shown to be greater when the total dose is received in a short period of time than when it is spread over a longer period, [a PEL] alone might not be adequate to reduce the risks.” 50 Fed.Reg. at 73.
After reviewing all of the evidence, the agency found that no existing studies support a conclusion that EtO has dose-rate effects. Petitioner Public Citizen vigorously attacks this conclusion. Based on the record presented to us, we conclude that OSHA has indeed accumulated substantial evidence that EtO presents no dose-rate effects. We thus uphold this finding.
As of June 14, 1983, the agency had apparently concluded that the record supported the issuance of a STEL. See Draft Final EtO Standard at 166-67, J.A. at 177-78. After receiving OMB’s objections to the STEL, however, OSHA decided to have a supplemental proceeding. Throughout the supplemental proceeding, OSHA’s scientific staff supported the issuance of a STEL. J.A. at 529. OSHA also admits that no new experimental or epidemiological studies were submitted during the subsequent rulemaking. 50 Fed.Reg. at 73.
The agency did receive, however, extensive new analyses of the studies submitted in the original rulemaking. These new analyses point up the shortcomings in the evidence on dose-rate effects. The agency found that the major epidemiological studies could not be used for dose-rate analysis because the workers in those studies had been exposed to both short-term bursts of EtO and long-term continuous exposures. Id. at 74. For example, the Yager study, see supra p. 17, demonstrated chromosomal effects from a single exposure of 82 ppm for 3.5 minutes per day (equivalent to 0.6 ppm per 8 hours). OSHA determined, however, that the study provides no evidence of a dose-rate effect because it did not examine the effects of an equivalent dose delivered less rapidly. One study did find dose-rate effects on in vitro cultures, id. at 75, but the authors themselves discounted the results because of the difficulty in translating in vitro results to in vivo results.
Based on this evidence, OSHA concluded that no dose-rate effect had been established. Id. Troubled by its findings on EtO in general and the equivocal nature of the dose-rate evidence, however, the agency specifically requested NIOSH to fund research in this area for the express purpose of determining the dose-rate relationship so that OSHA could again “evaluate whether the question of a short term limit for EtO should be reexamined.” Id.
Public Citizen challenges OSHA’s conclusion that no dose-rate effect has been established. Public Citizen argues that analyses done by OSHA’s EtO team and NIOSH suggest that EtO does have certain dose-rate effects. See J.A. at 532 (OSHA’s EtO team); id. at 513 (NIOSH). These statements, however, do not amount to a scientific certainty binding on the agency. Just as we do not choose among competing technical opinions in OSHA’s PEL proceeding, we decline to do so here. We reiterate that the very nature of a scientific inquiry on the frontiers of scientific knowledge will rarely allow a court to compel an agency to adopt a particular hypothesis. Naturally, the further an agency’s regulations stray from the frontiers of knowledge, the less deference *279courts will extend. But as long as Congress delegates power to an agency to regulate on the borders of the unknown, courts cannot interfere with reasonable interpretations of equivocal evidence. We uphold OSHA’s finding that the evidence on this record does not establish a dose-rate relationship.17
B. The Necessity of a STEL
OSHA also concluded that a STEL is unnecessary because the PEL will reduce a significant health risk and simultaneously control short-term exposures. While we acknowledge our deference to the agency’s expertise in most cases, we cannot defer when the agency simply has not exercised its expertise. We conclude that the proffered reasons for rejecting the STEL are insufficient to warrant affirmance of this aspect of the OSHA EtO standard.
While OSHA’s statement that the PEL will reduce a significant risk supports that aspect of the standard, it is inconclusive with regard to a STEL. As noted above, OSHA set the PEL at 1 ppm because of feasibility constraints. See 49 Fed.Reg. at 25,772. The agency recognized that EtO exposures at 1 ppm still allowed a significant health risk. See supra p. 1499.
If in fact a STEL would further reduce a significant health risk and is feasible to implement, then the OSH Act compels the agency to adopt it (barring alternative avenues to the same result). OSHA “shall set the standard which most adequately assures, to the extent feasible, on the basis of best available evidence, that no employee will suffer material impairment of health.” 29 U.S.C. § 655(b)(5) (1982). Since OSHA has found that a significant health hazard remains even with the 1 ppm PEL, the agency must find either that a STEL would have no effect on that risk or that a STEL is not feasible. OSHA has never made the latter contention; we thus address the former.
OSHA’s reasoning in declining to require a STEL follows the following lines. First, “the need to control short-term exposures is only important insofar as it relates to the resultant reduction of total dose.” 50 Fed. Reg. at 76. Second, “the control of short-term excursions are [sic] a necessary part of the overall compliance program for the 1 ppm [PEL].” Id. That is, “the compliance program designed to maintain exposure at or below the 1 ppm ... limit ... will also substantially reduce the magnitude of short-term exposures.” Id. OSHA apparently concluded from these premises that a STEL is unnecessary.
Such a conclusion, however, also requires a third, unstated premise: in attempting to meet the 1 ppm PEL, employers will in every case reduce short-term exposures to a point at or below the 10 ppm STEL. On this petition for review, OSHA argues that “the record clearly indicates that for a number of reasons the benefits that would be achieved by a short-term limit will be *280almost entirely achieved by the PEL.” Respondent’s Brief at 55.
We disagree. OSHA has not established this point with any evidence in the record. OSHA has apparently come to this conclusion because employers will necessarily control short-term exposures in order to meet the PEL. The evidence in this record, however, does not demonstrate that employers will necessarily reduce short-term exposures below 10 ppm in order to meet the PEL. For example, an employer might measure a very low background level of EtO exposure, a level significantly below the 1 ppm PEL. Conceivably, such an employer could allow short-term exposures to exceed 10 ppm over a fifteen-minute period but still have cumulative exposure fall below the 1 ppm PEL, which is an eight-hour average.
We assume that a 10 ppm STEL is feasible because no party has contested the issue at this point. A 10 ppm STEL could reduce total exposure below that demanded by the 1 ppm PEL alone.18 That reduction could reduce a significant risk facing employees, given OSHA’s no-threshold model and express finding that a significant risk of 12-23 excess deaths still exists at 1 ppm. Thus, OSHA has failed to support its decision not to issue a STEL.
We cannot dismiss this justification for the STEL as a theory not brought to OSHA’s attention during the rulemaking. OSHA had identified this theory in the original proceeding19 and it was offered by several participants. The record is replete with suggestions that EtO exposure follows the profile outlined above: low background levels and high intermittent exposures. For example, the National Union of Hospital and Health Care Employees stated that “most exposures to EtO are short term” and that workers often engaged in activities in which exposure to EtO totals “1-2 hours a day, during which exposure levels can soar to 480 ppm for 1 minute or 96 ppm for 5 minutes everyday, and yet total exposures during the day will still be within the 1 ppm [PEL].” Exhibit 189-39 (cited in Petitioner Public Citizen Brief at 49). AFSCME reported at least two instances in which workers could smell EtO (700 ppm threshold of human olfactory sensation) while the long-term average exposure was at or below 1 ppm. Exhibit 44 at 27-28 (cited in Petitioner Public Citizen Brief at 49). Other submissions before OSHA characterized typical EtO exposure as occurring at high concentrations over short intervals.20 Although these submis*281sions did not measure background levels, the short-term exposures would have allowed significant background levels of EtO.
OSHA responded to these arguments, stating that “there are low level EtO background concentration levels in the workplaces where EtO is being used, and these background concentrations must be taken into account in calculating an employeee’s [sic] daily dose of EtO.” 50 Fed.Reg. at 76. There remains a significant gap between an assumption of no background levels of EtO and an assumption of background levels of EtO close to 1 ppm (which would allow virtually no short-term exposures at levels above 1 ppm). OSHA has made no findings at all on patterns of exposure. Thus, we hold that the agency improperly assumed that exposure scenarios will eliminate the need for a STEL.
Since OSHA has “entirely failed to consider an important aspect of the problem,” Motor Vehicle Manufacturers Association, Inc. v. State Farm Mutual Automobile Insurance Co., 463 U.S. 29, 43, 103 S.Ct. 2856, 2867, 77 L.Ed.2d 443 (1983), we cannot find substantial evidence on the record as a whole for OSHA’s refusal to adopt the STEL. See 29 U.S.C. § 655(f) (1982). We thus remand for further consideration on this point. On remand, we expect the agency to ventilate the issues on this point thoroughly and either adopt a STEL or explain why empirical or expert evidence on exposure patterns makes a STEL irrelevant to controlling long-term average exposures.
VII. The Legality of OMB’s Actions
All parties to this action dispute vigorously the legality of OMB’s participation in the rulemaking. At least one lower court has decided the issue, Environmental Defense Fund v. Thomas, 627 F.Supp. 566 (D.D.C.1986), and it has been recently debated in the legal literature, DeMuth & Ginsburg, White House Review of Agency Rulemaking, 99 Harv.L.Rev. 1075 (1986); Morrison, OMB Interference With Agency Rulemaking: The Wrong Way to Write a Regulation, 99 Harv.L.Rev. 1059 (1986); Olson, The Quiet Shift of Power: Office of Management & Budget Supervision of Environmental Protection Agency Rule-making Under Executive Order 12292, 4 Va.J.Nat.Res.L. 1 (1984); see also Bruff, Presidential Power and Administrative Rulemaking, 88 Yale L.J. 451 (1979).
OMB’s participation in the EtO rulemaking presents difficult constitutional questions concerning the executive’s proper rule in administrative proceedings and the appropriate scope of delegated power from Congress to certain executive agencies. Courts do not reach out to decide such questions. See Ashwander v. Tennessee Valley Authority, 297 U.S. 288, 346, 56 S.Ct. 466, 482, 80 L.Ed. 688 (1936) (BrandEis, J., concurring). Since we have determined that OSHA’s decision on the STEL cannot withstand our statutory review, we have no occasion to reach the difficult constitutional questions presented by OMB’s participation in this episode.
VIII. Conclusion
We hold that the OSHA 1 ppm PEL is supported by substantial evidence on the record as a whole. We are unable to reach the same conclusion with regard to OSHA’s decision on the STEL. OSHA has failed to consider the interrelationship between the PEL and STEL. We find that OSHA has not accumulated substantial evidence to support its decision not to issue the STEL. Because we remand for further consideration on this particular point, we need not reach any constitutional claims presented.
Accordingly, this case is remanded to the agency for further proceedings consistent with this opinion.
It is so ordered.
. 29 U.S.C. § 671 (1982) establishes NIOSH to carry out the duties of the Secretary of Labor with regard to health research and employee training. The Director of NIOSH is empowered to “conduct such research and experimental programs as he determines are necessary for the development of criteria for new and improved occupational safety and health standards," id. § 671(d)(1) (1982) and to make recommendations on new standards, id. § 671(d)(2) (1982).
. EOIC also argued that the Bushy Run results might be misleading because a particular type of tumor spontaneously occurring in rats was absent in the Bushy Run subjects. EOIC Hazard Assessment at 157-58, J.A. at 1509-10. OSHA countered, however, with evidence that these particular tumors are very rare and were therefore unlikely to appear in the rat group. 49 Fed.Reg. at 25,742.
. OSHA reviewed at least five other studies, finding further evidence for a relationship between EtO exposure and tumor production. See 49 Fed.Reg. 25,743. This additional evidence further supports OSHA’s conclusion that EtO exposure poses a risk of cancer.
. We briefly summarize the results of these studies. Embree found that exposing male rats to a single dose of 1000 ppm EtO for four hours produced a significantly greater number of fetal deaths among the rats' offspring. Embree, Lyon & Hiñe, The Mutagenic Potential of Ethylene Oxide Using the Dominant-Lethal Assay in Rats, 40 Toxicology & Applied Pharmacology 261, 263 (1977), reprinted in J.A. at 709, 711. Generoso reported genetic damage capable of being passed from one generation to the next among mice injected with EtO. Generoso, Cain, Krishna, Sheu & Gryder, Heritable Translocation and Dominant-Lethal Mutation Induction with Ethylene Oxide in Mice, 73 Mutation Research 133 (1980), reprinted in J.A. at 716. Cumming found that EtO exposure could either inhibit or induce DNA repair in mice, depending on the dosage. Cumming, Michaud, Lewis & Olson, Inhalation Mutagenesis in Mammals: Patterns of Unscheduled DNA Synthesis Induced in Germ Cells of Male Mice by Exposure to Ethylene Oxide in Air 12-13 (unpublished manuscript), reprinted in J.A. at 726, 737-38. NIOSH reported significant dose-related increases in chromosomal aberrations and reductions in sperm count among monkeys exposed *264to differing doses of EtO. See J.A. at 1023 (memorandum from Director of NIOSH to OSHA, April 2, 1982); J.A. at 1145 (NIOSH rat and monkey study preliminary results). Yager and Benz found a dose-related effect on chromosomal aberrations among rabbits. Yager & Benz, Sister Chromatid Exchanges Induced in Rabbit Lymphocytes by Ethylene Oxide After Inhalation Exposure, 4 Env. Mutagenesis 121, 128-32 (1982), reprinted in J.A. 1359, 1366-70.
. Snellings, Zelenak & Weil, Effects on Reproduction in Fischer 344 Rats Exposed to Ethylene Oxide by Inhalation for One Generation, 63 Toxicology & Applied Pharmacology 382, 385-86 (1982), reprinted in J.A. at 1562, 1565-66, revealed a significant drop in the number of offspring born after EtO exposure. Hackett, Brown, Buschbom, Clark, Miller, Music, Rowe, Schirmer & Sikov, Teratogenic Study of Ethylene and Propylene Oxide and n-Butyl Acetate 49-50 (unpublished manuscript, May 1982), reprinted in J.A. at 1074, 1086-87, demonstrated a significant increase in fetal deaths when pregnant rats were exposed to EtO, both during gestation and pregestation periods, and during the gestation period alone. Snellings, Maronpot, Zelenak & Laffoon, Teratology Study in Fischer 344 Rats Exposed to Ethylene Oxide by Inhalation, 64 Toxicology & Applied Pharmocology 476, 478 (1982), reprinted in J.A. at 1569, 1571, reported fetal weight loss among rats exposed to 100 ppm EtO for six hours each on days six through 15 of gestation. LaBorde & Kimmel, The Teratogenicity of Ethylene Oxide Administered Intravenously to Mice, 56 Toxicology & Applied Pharmacology 16, 19-21 (1980), reprinted in J.A. at 676, 679-81, revealed a relationship between EtO exposure and both fetal weight loss and fetal malformation.
. The subjects were exposed to multiple sterilizing agents, one of which was EtO.
. Table II contains an error, which OSHA took into account before relying on the study. 49 Fed.Reg. at 25,749.
. Between 1976 and 1981, exposures in Finnish hospitals were estimated to be between 0.1 and 0.5 ppm. 285 Brit.J.Med. at 1463, J.A. at 1033. Apparently because several commenters were dissatisfied with the study, Hemminki reanalyzed the questionnaire data and reconfirmed the EtO/spontaneous abortion relationship. See Letters to the Editor, Hemminki response, reprinted in J.A. at 1373-74.
. In addition to the major consequences of EtO exposure reviewed here, OSHA also found that it causes eye and respiratory system irritation, growth depression, damage to selected organs, partial paralysis and decreased perception in the extremities, nausea, convulsions, and loss of consciousness. See 49 Fed.Reg. at 25,752-54.
. This rule, however, is subject to the general provision of § 1990.143, which allows a challenge to application of § 1990.143(h) if certain criteria are met. See, e.g., 29 C.F.R. § 1990.-144(d) (1985) (providing rules for documentary evidence challenging the relevance of animal studies to analysis of human health risks at lower dosages).
. AEOU argues that this previous OSHA statement evidences an unexplained change in position. We reject the argument that OSHA has changed its position without explanation. OSHA's statement that there is no direct evidence of cancer at levels below 14 ppm is not a statement that 14 ppm is a threshold. On the contrary, the statement is merely a candid admission of the inadequacy of current knowledge. It is consistent with OSHA’s decision to extrapolate the available data into areas where research is inconclusive or nonexistent. OSHA responds to AEOU’s charge by stating that the quoted material refers only to epidemiological evidence. Respondent's Brief at 61 n. 47. In either case, we see no unexplained change in the agency's position.
. See, e.g., Submission of Dr. Kenny S. Crump at 11-13, 21, J.A. at 1385-87, 1395; EOIC Hazard Assessment at 141, J.A. at 1493 ("continuous lifetime dose" of 1 ppm); id. at 142, J.A. at 1494 (levels below 1 ppm continuous lifetime dose); id. at 143, J.A. at 1495 (actual exposure level of 0.5 ppm); Sielken Submission, cited in 49 Fed. Reg. at 25,758 (risk calculated at 1 ppm exposure level); Submission of AFSCME, cited in 49 Fed.Reg. at 25,761, 25,762 (risk at 0.5 ppm); Submission of Callentan, cited in 49 Fed.Reg. at 25,762 (using rad-equivalence to determine risk at 1 ppm); HIMA Comments, cited in 49 Fed. Reg. at 25,762 (risk at 1 ppm).
Significantly, EPA's Carcinogen Assessment Group, upon which AEOU relies so heavily in faulting OSHA’s use of animal studies, calculated leukemia risk down to 0.56 ppm exposure levels. See 49 Fed.Reg. at 25,761.
. We express no view on the validity of OSHA’s no-threshold assumption when unsupported by substantial evidence in the record. We find here only that there is substantial evidence on the record as a whole for OSHA’s position that EtO exposure presents health hazards at 1 ppm.
. There are at least two methods of converting animal data into human equivalents. Exposure can be calculated by milligrams of EtO/surface area/day (the EPA method) or by milligram/kilogram of body weight (the OSHA method). At the hearing, OSHA’s witness, Dr. Kenny S. Crump,
calculated [that] ... if OSHA had used milligram of intake per surface area per day, an expression of dose used routinely by the Environmental Protection Agency, instead of milligrams per kilogram of body weight per day, ... this expression of dose would increase the estimates of risk approximately 5- to 6-fold.
Petitioner AEOU Brief at 17 (quoting 49 Fed. Reg. at 25,760). AEOU contends that
Dr. Crump miscalculated when he applied the EPA formula. In contrast, in EPA’s Health Assessment Document for Ethylene Oxide ("HAD”), EPA applied its mg/surface area/day formula to the rat data from the Bushy Run study relied upon by OSHA and found that [rats] exposed to EtO concentrations of 100 ppm for six hours would absorb a dose of 20.24 milligrams/kilogram of body weight, the equivalent dose for a human in milligrams per kilogram of body weight was 2.63, or roughly 13% of the equivalent dose of the rat. In other words, application of EPA’s analysis reveals that a rat absorbs roughly 7.7 times that absorbed by a human at equivalent exposure levels, a result directly opposite to that predicted by Crump. See HAD at 9-137.
Petitioner AEOU Brief at 17-18 (footnote omitted).
Although EPA reached the same numerical conclusions, see EPA HAD at 9-143, it is at polar opposites with AEOU concerning the meaning of these numbers. We quote directly from the EPA HAD:
Extrapolations from animals to humans can also be made on the basis of either relative weight or surface area. The latter approach, which is used by the CAG, has more of a basis in human pharmacological responses; however, at the present time there is some question as to which of the two approaches is more appropriate for use with carcinogens. Given this uncertainty, the CAG has chosen the most generally employed method, which is also the more conservative of the two. In the case of ethylene oxide inhalation studies, the use of extrapolation based on surface area rather than weight increases the unit risk estimates by a factor of 5.5 for the males and 6.5 for the females.
Id. at 9-146 (emphasis added); see also id. at 9-153-155.
AEOU misunderstands EPA’s animal equivalence conversions. The EPA conversion produces a dosage for humans at which humans will experience the same response experienced by rats at the experimental dosages. “This means that a dose of 2.63 mg/kg body weight given to a 70 kg human is assumed to produce an equivalent response to that produced by 20.-24 mg/kg in the male rat." Id. at 9-151. In contrast, according to EPA, OSHA estimates the dosage required to produce equivalent respons*276es in humans at 19.30 to 23.94 mg/kg, much closer to 20.24. Id. at 9-153.
Thus, AEOU's conclusions on Dr. Crump’s testimony are backwards. The EPA method produces risk estimates 5-6 times greater than does the OSHA methodology.
. This estimate is in addition to those expected to die from causes unrelated to EtO exposure.
. While AEOU does not directly challenge the 1 ppm PEL, it does attack an ancillary requirement of the EtO standard. 29 C.F.R. § 1910.-1047(j) (1985) requires employers to post signs warning employees of the hazards presented by EtO exposure. For example:
The employer shall post and maintain legible signs demarcating regulated areas and entrances or accessways to regulated areas that bear the following legend:
DANGER
ETHYLENE OXIDE
CANCER HAZARD AND REPRODUCTIVE HAZARD AUTHORIZED PERSONNEL ONLY RESPIRATORS AND PROTECTIVE CLOTHING MAY BE REQUIRED TO BE WORN IN THIS AREA
Id. § 1910.1047(j)( l)(i).
AEOU attacks this requirement on two different grounds. First, AEOU contends that Benzene requires that the agency find that each and every aspect of its standard eliminates a significant risk faced by employees. Since OSHA did not make such a finding with regard to the sign requirement in particular, the sign requirement must be invalidated. See Petitioner AEOU Brief at 57. Second, AEOU contends that there is no substantial evidence in the record that EtO is carcinogenic and causes reproductive damage. Thus, the argument goes, the agency cannot require signs stating that the chemical has these effects. Id. at 57-58.
We reject both of these arguments. In Benzene, the Supreme Court required that OSHA regulations be “reasonably necessary and appropriate to remedying] a significant risk.” 448 U.S. at 638, 100 S.Ct. at 2863; see 29 U.S.C. § 652(8) (1982). We believe that the agency has satisfied this standard. Labels and signs are specifically contemplated by the OSH Act, see 29 U.S.C. § 655(a)(7) (1982), and there is no doubt that some sort of sign can be part of a "reasonably necessary and appropriate” standard. AEOU's true objection is to the particular wording, rather than to the sign itself, since the Association offers an alternative sign: ”[T]he Association suggests that the record would support required signs and labels which state that EtO is extremely toxic at high levels and long exposures, and which carry appropriate warnings concerning safe use, including a warning not to breathe vapors." Petitioner AEOU Brief at 58. We find no fault in OSHA's wording, however, given our affirmance of the agency’s findings that EtO is indeed carcinogenic, mutagenic, cytogenic, and responsible for reproductive damage. See supra Part III. We therefore uphold the sign requirement.
. Petitioner Public Citizen cites Farmers Union Central Exchange, Inc. v. FERC, 734 F.2d 1486, 1499-1500 (D.C.Cir.1984) and Public Citizen v. Steed, 733 F.2d 93, 99 (D.C.Cir.1984) for the proposition that a "heightened" standard of review applies when an agency abruptly changes its position. Petitioner Public Citizen Brief at 33. Public Citizen alleges that OSHA abruptly changed its position when it received OMB’s comments on the STEL, since it was prepared to issue the STEL prior to that time. The cited cases, however, hold only that an agency must give a reasonable justification for a decision to deviate from an existing regulatory scheme, Farmers Union, 734 F.2d at 1500, or from a settled course of agency behavior, Steed, 733 F.2d at 99. We have some difficulty finding that OSHA’s actions in this case amount to ei-
ther an existing regulatory scheme or a settled course of behavior.
Ultimately, however, we need not decide the point. To the extent that these cases provide for more probing scrutiny of OSHA, we find that the agency has met its burden with regard to the dose-rate issue. The agency has clearly justified its basis for departing from its earlier conclusion: new evidence adduced in an entirely separate proceeding had undermined its original decision.
As far as this "heightened” scrutiny argument applies to OSHA's argument that a PEL necessarily subsumes a STEL, we do not reach the point since we find that OSHA fails the statutorily mandated substantial evidence test. See infra pp. 1501-02.
. For example, a background level of 0.5 ppm would allow exposure of 64 ppm for almost 4 minutes (3.75 minutes) without violating the PEL. A 10 ppm STEL would reduce exposure for that short period to 10 ppm, and the cumulative dose to roughly 0.508 ppm, rather than 1 ppm. Thus, a STEL on this scenario would cut exposure almost in half. By employing a STEL in conjunction with the PEL, the evidence indicates that OSHA could conceivably cut cumulative exposure well below 1 ppm even though 1 ppm is the lowest feasible long-term average. (Admittedly, employers could increase the number of employee short-term exposures if a STEL were introduced. We do not assume, however, that employers will change their business practices in order to expose their employees to the maximum allowable level of EtO.)
. OSHA itself identified this logical relationship between the STEL and the PEL prior to receiving comments from OMB. See Draft Final Standard at 166 ("even with a 1 ppm TWA over 8 hours, one and two minute exposures of hundreds of ppm might be allowed to occur”), J.A. at 177; id. at 216 (“For industry sectors where exposure patterns involve periods of intermittent, burst-type exposures to EtO, the STEL will likely result in TWA exposures well below 1 ppm, although OSHA cannot determine how low such exposures will be."), J.A. at 227; id. at 219 ("the 8-hour TWA and 15-minute STEL work hand-in-hand to achieve effective control over exposures of this nature [hospital sterilization activities”], J.A. at 230. OSHA’s final analysis totally failed to discuss this relationship.
.The Yager study, see supra p. 21, measured worker exposure to sterilizer operations at 82 ppm over 3.5 minutes, which translates to 0.6 ppm over eight hours. Thus, if the hospital’s background EtO level were at or below 0.4 ppm averaged per eight hours, it would be in compliance with the OSHA standard. See Draft Final Standard at 68, J.A. at 79. Public Citizen cites further examples of this type of evidence. See Submission of National Institute of Environmental Health Sciences 1-2, J.A. at 518-19, 1735-36; Submission of Dr. Lawrence Heckler 3 (Exhibit 45) (quoted in Petitioner Public Citizen Brief at 48); NIOSH Response to OMB 3, J.A. at 507, 1774.
*281Our review of the record indicates that several other participants fielded this issue as well. See Letter from Vernon Carter to R. Leonard Vance 5 (July 19, 1984) (Exhibit 178), J.A. at 1759; Letter from R.V. Durham to R. Leonard Vance 2 (July 19, 1984) (Exhibit 179), J.A. at 1761; Statement of American Association of Occupational Health Nurses, Inc. 2, 3 (Exhibit 180), J.A. at 1767, 1768.