• Home
• About ATSDR
• Press Room
• A-Z Index
• Employment
• Contact Us
• CDC

U.S. Department Of Health And Human Services Public Health Service Agency For Toxic Substances And Disease Registry

National Exposure Registry Benzene Subregistry Baseline Through Followup 4 Technical Report February 2001

This report was supported in full by funds from the Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA) trust fund through the Agency for Toxic Substances and Disease Registry, U.S. Department of Health and Human Services.

DISCLAIMER

The use of company or product names is for identification only and does not constitute endorsement by the Agency for Toxic Substances and Disease Registry or the U.S. Department of Health and Human Services.

CONTENTS

Page
DISCLAIMER

ii

EXECUTIVE SUMMARY

This is a report on the baseline activities and results of the statistical analyses of the Baseline through Followup 4 data for the Benzene Subregistry of the National Exposure Registry (NER). The NER was created in response to the congressional mandate contained in the 1980 Comprehensive Environmental Response, Compensation, and Liability Act and reiterated in the Superfund Amendments and Reauthorization Act of 1986. This mandate directed the Agency for Toxic Substances and Disease Registry (ATSDR) to create a registry of people exposed to hazardous substances in the environment. The Benzene Subregistry is one of four existing chemical-specific subregistries in the NER.

As with the other subregistries in the NER, the Benzene Subregistry is a database on people who have been exposed to a specific chemical, in this case benzene. The purpose of the Subregistry is to assess the long-term health consequences, if any, of long-term past exposures to low levels of benzene in drinking water. The Subregistry itself is not a definitive study; cause-effect relationships cannot be established using only Subregistry-based information. However, the Subregistry will furnish the information needed to generate appropriate and valid hypotheses for future activities, such as epidemiologic studies.

The data collected for each member of the Benzene Subregistry include environmental data, demographic information, smoking and occupational histories, and self-reported responses to 25 general health status questions. The data files for each Subregistry are established at the time baseline data are collected. Followup surveys are conducted at the end of the first and second years, then at 2-year intervals, thereafter, to update the data files.

The Benzene Subregistry contains information on 1,143 persons (1,127 living and 16 deceased at the time of baseline data collection) who had documented exposure to benzene in their drinking water and were exposed for at least 30 days. These individuals had resided in Texas. The participation rate for those eligible was 97%.

Reported health outcome rates were calculated for the Benzene Subregistrants and compared with morbidity data from the 1990 National Health Interview Survey (NHIS). When interpreting the statistical results and planning future activities based on these results, the limitations of the Subregistry data files must be kept in mind. For instance, a bias in the rate of reporting could have existed because people were aware of their benzene exposure, had been advised of the potential effect on their health, and might have sought medical care more often than the general population. To moderate this potential bias, the Subregistry data were collected with the restriction that a health care provider must have told persons that they had the condition or have treated them for it. Also, some of the questions in the two surveys were worded differently, making direct comparisons of the reported rates more difficult to interpret. Given the large number of comparisons used in the analyses, there might be some false positive findings. These limitations and restrictions are discussed in this report.

The morbidity data analyses indicated an increased reporting of several health outcomes by Benzene Subregistry registrants. Statistically significant increases (p £ 0.01 significance level) at specific interview periods were observed for anemia and other blood disorders at Baseline through Followup 4; arthritis, rheumatism, and other joint disorders at Followup 1; cancer at Baseline through Followup 3; diabetes at Followup 3; kidney disease at Baseline; liver problems at Followups 1 and 2; respiratory allergies and other problems, such as hay fever at Baseline through Followup 4; skin rashes, eczema, or other skin allergies and Baseline and Followups 1,3, and 4; stroke at Followups 1 through 4; ulcers, gallbladder trouble, or stomach or intestinal problems at Followups 1 and 3; and urinary tract disorders, including prostate trouble, at Baseline through Followup 4. Statistically significant deficits were reported for the Benzene Subregistry population for the following health conditions at various interview periods: hearing and speech impairments, asthma and emphysema, and arthritis. The rates of these conditions might have been affected by the limitation imposed on registrants' reporting, that is, that a health care provider had told them they had the condition or had treated them for it. This limitation was not part of the NHIS.

The findings in this report cannot be used to identify a causal relationship between the health outcomes and benzene exposure. Additionally, some methodological differences in data collection may have biased the reporting rates, resulting in false positive findings. The findings of this report do, however, reinforce the need to continue regular follow-up of this population.

Keeping registrants informed of all current information related to their exposures is another of the stated goals of the NER. Both a registrant report and a one-page fact sheet, written for the general public and containing the findings of this technical report, were prepared and sent to each registrant, and then released to the media. The mailing was followed by a public availability meeting at the site for discussion.

SECTION 1: INTRODUCTION

This is a report on the activities and findings from the analyses of data collected from registrants of the Benzene Subregistry at five (5) time points: Baseline and Followups 1-4. The Benzene Subregistry is part of the National Exposure Registry (NER), which was created and is being maintained by the Agency for Toxic Substance and Disease Registry (ATSDR).

In 1988, the policies and procedures proposed for the NER were reviewed extensively by several committees composed of independent scientists, state representatives, representatives of other federal agencies, and other interested people. The revised policies and procedures were published in the NER Policies and Procedures Manual (1). The Benzene Subregistry was one of the first Subregistries established as part of the NER program. The NER currently contains four chemical-specific Subregistries (trichloroethylene [TCE], trichloroethane [TCA], dioxin, and benzene).

The goals and objectives of the Benzene Subregistry reflect those of the NER; specifically, it will be used to facilitate epidemiologic or health studies and surveillance, and will provide information that can be used to assess the effects of exposure to benzene on a general population. Additionally, the Benzene Subregistry will enable federal, state, and local officials to provide exposed persons with timely, relevant information about benzene exposure, potential adverse effects related to that exposure, preventive measures, or therapeutic advances that were not understood when the Benzene Subregistry was established.

The Policies and Procedures Manual (1) describes all policies, procedures, and operational details pertinent to establishing the Benzene and other Subregistries of the NER. Specific topics from the policies and procedures document are reiterated in this report, where necessary, for clarity.

The objective of this report is to present the results of the statistical analyses comparing the reporting rates of registrants at various time points for specific health outcomes with national rates from the National Health Interview Survey (NHIS). The report is an update on the latest Benzene Subregistry findings and highlights some health outcomes and predictive variables to consider for analysis during future epidemiologic or health studies. Results from the analyses presented in this report can be used to suggest specific hypotheses for future research on the Benzene Subregistry population and potentially other residential populations that have experienced similar exposures to benzene.

Section 2 of the report reviews the rationale for the selection of benzene as a primary contaminant for the NER and describes in details the site meeting the criteria for inclusion in the Benzene Subregistry. Section 2 also provides information on the environmental data and periods of exposure. A discussion of the data collection periods, participation rates, and number of registrants is included.

Section 3 describes the methods used for data analysis. It provides details for the descriptive and statistical comparison of the Benzene Subregistry data with national survey data files for smoking habits and demographic characteristics and reported rates of adverse health outcomes. Section 4 provides an overview of the characteristics and health status of registrants who took part in each of the data collection efforts and the results of the descriptive comparisons between the Benzene Subregistry data and the NHIS data. Section 5 contains the results of the health outcome reporting rate comparisons between the Benzene Subregistry at all five time points (Baseline through Followup 4) and the NHIS. Section 6 summarizes the findings of the report and discusses them in relationship to the published literature. Section 7 states the conclusions of the analysis of the Benzene Subregistry data and outlines future activities related to the Subregistry.

SECTION 2: BACKGROUND OF THE BENZENE SUBREGISTRY




SELECTION OF BENZENE AS A PRIMARY CONTAMINANT

In 1989, benzene was selected as the primary chemical for a Subregistry of the NER (1,2). This report also provides details on the presence of benzene in the environment, as well as a summary of evidence of adverse effects observed in both human and animal studies. In summary, the factors that led to the selection of benzene included the prioritization of benzene on the Hazardous Substance Priority List (3); ubiquitousness of benzene in the environment; published evidence of benzene toxicity in workers and in toxicologic studies; and the paucity of information on low-level, long-term exposures to benzene. Each of these factors suggested that establishing of a Benzene Subregistry could contribute significantly to the detection of adverse human health effects, should they exist, following long-term, low-level exposure to benzene in the environment.

SELECTION AND DESCRIPTION OF BENZENE SITE

The site selection process used to develop subregistries for the NER is described in the Policies and Procedures Manual (1). Selection of the Three Lakes Municipal Utilities District (TLMUD) as the site for the Benzene Subregistry has been previously described (4).

The criteria for selection included documentation of exposure levels and duration, identification and estimated size of the exposed population, identification of susceptible subpopulations, and identification of the number and levels of secondary contaminants (1,5). Although 263 potential sites were identified as having benzene as a contaminant of drinking water, only one site was eligible for inclusion in the Benzene Subregistry.

The TLMUD (Figure 2-1) is a small municipality in Harris County, Texas on Highway 249 (the Tomball Parkway), between Houston and Tomball. The TLMUD received its drinking water from a well located near the community. Benzene was first discovered in the TLMUD water supply in September 1990. The contaminated well was the only well serving the TLMUD and water was not treated before it entered the distribution system. The TLMUD is composed of two subdivisions, Three Lakes and Three Lakes Village, with approximately 1,200 residents. Because the site used a public water system, and because water samples were taken from the end of the distribution system, for the purpose of this report exposures for all registrants were considered to be the same. The duration of exposure considered by the Registry is from January 1, 1979, through October 1990, which represents documented beginning dates and confirmed ending dates of use of contaminated water (see Table 2-1). See Table 2-2 for a complete list of contaminants found in the TLMUD water supply.

DATA COLLECTION

Registrants were identified using three key components to define individual eligibility and exposure: (1) valid information indicated the presence of the contaminant(s) of interest in one or more of the media of interest; (2) evidence, for a given individual, of an appropriate route(s) of exposure;

Figure 2-1.—Location of Three Lakes Municipal Utilities District, Texas.




Table 2-1.—Summary of environmental data.

Benzene Subregistry Site	Year Exposure Began	Exposure Period* (Number of Years)	Maximum Level of Benzene Reported (ppb)†
Three Lakes Municipal	1979	11.75	66.0

*Exposure period is based on best available evidence of when contamination occurred and when exposure ceased following switch to an alternative water source.

†ppb = parts per billion.

and (3) evidence of indicated transmission from the contaminated source to the potential registrant during the period of exposure as verified by that individual. In the case of the Benzene Subregistry, the well water had to have been tested and validated for the presence of benzene. Also, the well water had to have been the sole source of water for drinking, bathing, or cooking for all individuals at the site residential addresses. Finally, a registrant would had to have reported using the benzene-contaminated well water for drinking, cooking, or bathing for at least 30 days during the exposure period.

Table 2-2.—Contaminants found in the Three Lakes Municipal Utilities District tap and well water, 1990.

Compounds	Maximum Levels (ppb)*
	Tap	Well
Benzene	66.0	1,100.0
Ethyl benzene	4.1	66.0
Methyl cyclohexane	23.0	30.0
Cyclohexane	7.0	-
Methyl butane	37.0	-
Methyl propane	10.0	70.0
Total trihalomethanes	0.0	31.0
Dibromochloromethane	<1.0	11.0
Bromodichloromethane	<1.0	3.0
Chloroform	<1.0	<1.0
Bromoform	<1.0	17.0
Propane	-	66.0
Dimethyl cyclopentane	-	4.0
Tetramethylcyclopropane	-	42.0
C3	-	6.0

*ppb - parts per billion

Data collection for the Benzene Subregistry began in 1991 and to date four followups have been completed (Table 2-3). Baseline interviews were conducted face-to-face while followup interviews were conducted via computer-assisted telephone interviews (CATI). Before all interviews, registrants were first sent a mailing that contained information about ATSDR, the NER, the Benzene Subregistry, and the chemical benzene. For the Baseline, a public meeting about the Registry was held in the area prior to the start of data collection.

Table 2-3.—Summary of data collection activities.

Interview Period	Date
Baseline	May 1991-July 1991
Followup 1	June 1992-July 1992
Followup 2	October 1993-September 1994
Followup 3	September 1995-January 1996
Followup 4	September 1997-December 1997

At Baseline, each eligible person or a proxy for that person was administered the NER core questionnaire, which included a set of questions about health conditions that the registrant currently had or had ever had and that had been either confirmed or treated by a health practitioner. Each time the respondent reported the presence of one of these health conditions, a set of follow-up questions was asked about the date of first treatment by a physician, current treatment, prescribed medication, and hospitalization related to the condition.

Information on deceased eligible persons was obtained from a knowledgeable proxy (usually the spouse) and a death certificate was requested from the appropriate state office. Information on cause of death, along with other pertinent information, was extracted from the death certificates and coded as copies of death certificates were obtained from the states. These procedures were the same for the baseline and for all follow-up interviews. Analysis of the mortality data is not included in this report; a separate report on mortality is in progress.

The procedures used for locating registrants for each followup were as follows: four to five weeks before the start of data collection activities, ATSDR began tracing efforts of those registrants known to have moved since the last interview. These tracing cases were usually identified through registrant mailings that were returned to ATSDR as undeliverable. During data collection, cases requiring tracing were identified through attempts made to the telephone number on record. A case was forwarded to a locating specialist if the registrant or proxy had moved (no new telephone number provided), if the telephone number had been disconnected, or if calling the number resulted in "ring no answer" or "busy" (this would be considered a non-working number). The locating specialist would then attempt to locate the registrant by one of the following methods (in the order listed): (1) calling directory assistance; (2) calling contacts provided during the last interview; (3) credit bureau searches; and (4) state departments of motor vehicles searches.

At each followup, and using CATI technology, each registrant or proxy for the registrant was again administered the NER core questionnaire, which includes a set of questions about health practitioner confirmed or treated health conditions that the registrant currently had or had since the last interview. If a respondent reported a health condition, further questions were asked about the date of first treatment, current treatment, prescribed medication, and hospitalization history.

Table 2-4 summarizes the response information from interviews of the exposed registrants at each of the data collection time points and overall. At Baseline, 98% of the eligible people who were contacted and asked to take part in the Benzene Subregistry did participate. Total participation rates were calculated by dividing the number of registrants (living and deceased) who completed interviews by the number of potentially eligible persons who were contacted and asked to participate. By the Followup 4 interviews, 89% of the registrants contacted had agreed to participate. Overall, from Baseline through Followup 4, the Benzene Subregistry has retained about two-thirds (66 %) of the original registrants (including all deceased registrants).

Table 2-4.—Summary of registrant response for the Benzene Subregistry, Baseline through Followup 4, all races.

Outcome	Interview Period
	B*	F†1	F 2	F 3	F 4	Overall
	N (%)§	N (%)	N (%)	N(%)	N(%)	N (%)
Completed interview registrant living	1,127 (96.9)	1,034 (91.7)	950 (91.9)	837 (88.1)	740 (88.4)	740 (63.6)
Completed interview registrant deceased	16 (1.4)	3 (0.3)	4 (0.4)	4 (0.4)	4 (0.5)	31 (2.7)
Refusal	17 (1.5)	23 (2.1)	19 (1.8)	37 (3.9)	46 (5.5)	142 (12.2)
Noninterviewed registrants¶	3 (0.3)	67 (6.0)	61 (5.9)	72 (7.6)	47 (5.6)	250 (21.5)
Total eligible	1,163	1,127	1,034	950	837	1,163

*B = Baseline.

†F = Followup.

§% = percentage of total eligible.

¶Includes unable to locate or contact, unavailable during interview period, language barrier, in litigation, and mentally or physically incapable (with no available proxy).

SECTION 3: METHODS FOR DATA ANALYSES

RATIONALE FOR THE COMPARISON OF THE BENZENE SUBREGISTRY DATA WITH NATIONAL DATA

Benzene Subregistry data were compared with data obtained from the 1989 through 1994 NHIS (6-12) which corresponded to the years in which Benzene Subregistry data were collected. In addition, they most closely correspond to the NER design. In 1995, NHIS radically changed the survey and sampling designs, which precluded comparison of the 1995 NHIS data with NER data. As a result of further changes to the NHIS, subsequent years such as 1996 and 1997 could not be combined with years prior to 1995. The comparison with NHIS data was done to assess differences between the NER and NHIS files in reporting rates for the same or related health effects. These comparisons are consistent with Registry objectives and goals as stated in the Polices and Procedures Manual (1), which are to provide a preliminary assessment of the extent to which Benzene Subregistry members may have an excess, if any, of adverse health conditions and to generate, rather than test, hypotheses about benzene exposure and health outcomes. In addition to a comparison of the subregistry health data with national health data, comparisons of registrant demographic and smoking data with national data were also made to indicate the extent to which Benzene Subregistry members were similar to the general population. These comparisons are important because both demographic characteristics and smoking are known to be correlated with, or are possible causes of, many adverse health conditions.

Subsets of the NHIS data were used in the comparisons with demographic, smoking, and health data components of the Benzene Subregistry. The NHIS is an appropriate comparison population because it is a subset of the residential, noninstitutionalized U.S. population, the population of interest for comparisons of the health status of the NER members. As of 1985, a stratified, multistage cluster sample design was used in the NHIS to obtain a representative sample of the target population; this information was used to create representative national norms. The NHIS, similar to the Registry, consists of self-reported data that were obtained using face-to-face interviews.

Because of the similarity of the data collection instrument and methods used by the NHIS and the NER, the NHIS data were appropriate for the calculation of selected prevalence and incidence statistics and could be used for exploratory comparison with Registry data for health outcomes. The weighting factors (12) provided by the National Center for Health Statistics (NCHS) were applied when using the data. The use of a 6-year composite (1989 through 1994) NHIS rate for the analyses moderated any fluctuations in reporting rates over time. The NHIS files used for selected comparisons in this report contained data from 542,472 respondents.

Members of the Benzene Subregistry reside primarily in the South (Texas), with the remainder located throughout the nation. The influence of regionality on reported disease outcome rates for the Benzene Subregistry, a concern when comparing the subregistry reporting rates with the national rates (reflected by the NHIS numbers) was explored previously (4) and ATSDR's review of the NHIS regional rates for selected outcomes found no definitive evidence indicating that the overall health status of those located in the South differed significantly from that of the general United States population. Therefore, differences between the Benzene Subregistry file and the NHIS file were not expected to be, and did not appear to be, the result of regional differences.

METHODS OF DESCRIPTIVE VARIABLE COMPARISONS

Demographic Characteristics

The NHIS and Benzene Subregistry populations were compared in terms of four demographic characteristics-gender, age, race, and education level-as well as cigarette smoking rates. Each of these variables is a potential correlate of health status.

Gender

The distribution of the male-female ratio was assessed on an age-specific basis. The proportion of males and females in each age category was based on the NHIS data and compared with the corresponding proportions in the Benzene Subregistry. Each age-specific proportion in the Benzene Subregistry was compared with the corresponding proportion in the NHIS by testing that the binomial proportion was equal to a specified theoretical value. No significant differences were found between the two files for this variable at any interview period.

The modeling method used to test differences in reporting rates took into account any effect due to sex. If sex is a significant effect modifier in the Poisson modeling, then comparisons of health rates are made by strata, for males and females. If sex is not significant, the comparison made aggregated over sex is still a sex-adjusted rate. Thus, even if there had been differences in the distributions of males and females in the Benzene Subregistry when compared with the NHIS, the comparisons would not have been affected.

Age

The descriptive comparisons of age used a 10-category measure. The regression analyses in this report involved a regrouping of age categories. An eight-category measure of age (combining the lower two and upper two groups) was used because of the sparsity of positive reports in some of the age strata.

It should be noted that because the health outcome analyses involved summarizing age- and sex-specific comparisons rather than analyzing age-adjusted summaries, whether the age distribution of the NHIS file matched the age distribution of the Benzene Subregistry file was not directly relevant unless distribution differed within the age groups.

The age groupings were realigned in the Followup 1 analysis to compensate for the one-year time lapse since Baseline; that is, 0 through 9 years of age became 1 through 10 years; 10 through 17 years of age became 11 through 18 years, and so forth. The age groupings were similarly adjusted for Followup 2 (1 year difference) and Followups 3 and 4 (2 year differences). The realigned ages were also used for the NHIS age groupings for the statistical analyses.

Race

Race is an established correlate of socioeconomic status (13) and health status (14). National data indicate that nonwhites have lower rates for cigarette smoking (15). For these reasons, race is a potential control variable for the comparisons of health status and smoking rates. However, there were too few nonwhites in the Benzene Subregistry to use race as a variable, so all analyses were restricted to registrants responding white to the race question.

Education Level

For education level (the highest level attained as reported by a respondent), the descriptive analyses at Baseline included comparisons in which education level was measured as a four-category ordinal variable (that is, 0 through 11 years, 12 years or the equivalent of a high school diploma, 13 through 15 years or some college, and 16 years or more or the equivalent of a college degree). The information reflects the status at the time of baseline data collection; therefore, education level was not considered in the analyses of any followup data.

Cigarette Smoking

Rates for current and past smoking behavior were compared across sex, age, and education attainment categories. A current smoker ("current rate") was defined as anyone who reported being a smoker at the time of the interview, and who had smoked at least 100 cigarettes in his or her lifetime. Past smoking behavior ("ever rate") was assessed by calculating the rates for people who had ever smoked at least 100 cigarettes during their lifetime. People who had ever smoked included both current and ex-smokers. For the NHIS, only one adult per household was asked the smoking-related questions; all adults were asked about smoking for the NER.

The comparison of Benzene Subregistry population cigarette smoking rates with a national rate was used as a means to assess the general comparability of the Benzene registrants with the U.S. population, as represented by the NHIS population. Adjustments for smoking or the inclusion of a smoking factor in the regression modeling was not possible because this information was not available for each respondent in the NHIS data. The 1990 through 1994 NHIS did solicit some general smoking information; however, smoking information was not asked of each respondent, but only for one person per household. Therefore, direct comparison of Benzene Subregistry and NHIS smoking rates could not be made, and smoking could not be included in any of the statistical models.

METHODS OF COMPARING HEALTH OUTCOMES

Question Comparability

Prior to comparing Benzene Subregistry and NHIS data for health conditions reported by respondents, the comparability of NHIS and Benzene Subregistry health condition questions was assessed. The questions about health conditions in these two surveys differed in three respects: restrictions on the source of diagnosis; the time frame of occurrence or treatment; and, in some cases, the wording of the health condition. A discussion of each potential source of variation in health condition questions was previously reported (4) and is summarized here. The Benzene Subregistry Baseline health-related questions are in Appendix A; the Benzene Subregistry Followup questions are in Appendix B; and the NHIS health-related questions are in Appendix C.

Source of Diagnosis

Benzene Subregistry questions about health conditions specified that the source of diagnosis must be a "physician or other medical provider". This qualification was intended to minimize self-diagnoses or the biased reporting of health problems by registrants, since they might have a greater awareness of health because of their known exposure and publicity related to the exposure. The NHIS questions did not include any type of qualification concerning the source of diagnosis. Therefore, if all other factors were equal or similar, an increased reporting by NHIS respondents when compared with the registrants might be expected. The increases would be expected to be greater for health conditions often self-diagnosed (for example, arthritis, hearing impairment, and some respiratory problems).

Time Frame

Benzene Subregistry baseline questions about health conditions asked about diagnoses of, or treatment for, conditions from the point of birth through the date of the interview ("Has a physician or other medical provider ever told you/SUBJECT that you/he/she/ had or treated you/SUBJECT for CONDITION?"). Only one time frame was addressed: ever had (subject's lifetime). Respondents who reported "yes" to this question were also asked whether the subject was ever treated for the condition, when the subject was first treated for the condition, and whether the subject was currently being treated for the condition.

Health questions in the Benzene Subregistry follow-up interviews similarly asked about diagnoses of, or treatment for, conditions from the date of the last interview through the date of the current interview. (The time interval between interviews was 1 to 2 years). Respondents who reported "yes" to having been told they had, or were treated for, a condition within the stated time frame were then asked if the date of first treatment was since the last interview, and whether they were currently being treated for the condition.

The NHIS questionnaire included questions that focused on three time frames-ever had the condition, had the condition within the last 12 months, or currently had the condition. With the exception of heart diseases, only one time frame was used to create a response rate for any given health condition. A comparison of NHIS and Benzene Subregistry time frames has been previously described (4). Table 3-1 provides a comparison of NHIS and Benzene Subregistry questions in terms of the time frame for each health condition. One NHIS health condition question, the effects of a stroke, was asked and rate calculated in the context "have you ever had." The questions and time frames for the subregistry and NHIS matched on this condition.

Health Conditions

Benzene Subregistry and NHIS questions were also compared in terms of the phrasing of health conditions. The results of this comparison have been reported previously (4). In summary, some health conditions matched exactly, others did not. Only health questions from the Benzene Subregistry data that were considered to be sufficiently similar to the NHIS health conditions were used for the analyses presented in this report.

Table 3-1.—Comparison of time frames for health condition questions.

(Benzene Subregistry Conversion From "ever had")	National Health Interview Survey Time Frame for Condition
	"ever had"	"in the past 12 months"	"now have"
"Ever had" (same)	Stroke
"In the past 12 months" ("Ever had" and "currently have" and/or date of first treatment within past 12 months)		Cancer, rash, anemia, kidney disease, urinary tract disorders, ulcer, liver problems, asthma, respiratory problems and allergies, diabetes, arthritis, hypertension
"Now have" ("Ever had" and "currently have")			Speech impairment, hearing impairment, mental retardation




Calculation of NHIS Rates

Rates of certain health conditions among the general population are calculated, with one difference, as the NCHS directs in their document, Current Health Estimates (16), the companion document to the NHIS data. The difference is, NCHS reports "number of conditions per 1000 people" (a rate), the subregistry calculates "number of people with the condition, per 100 people" (a percentage). Thus, the NCHS rates are not proportions, and can and do count some respondents more than once, if they reported more than one specific condition of the same type. For example, a person may have both stomach ulcer and intestinal ulcers, and each condition reported would be counted in the NCHS method. However, the NER questionnaire used broader categories of conditions and since both conditions are in the same category in our analyses, each person would be counted only once even if they contributed multiple conditions of the same type. Besides collapsing the more numerous and narrowly defined categories of the NHIS data into the broader categories of the NER questionnaire to make comparisons with the subregistry data, the NCHS directions were used to calculate proportions from the NHIS data.

The calculation of the composite or aggregate rate for each category of conditions for the years 1989 through 1994 was also done in the manner directed by NCHS. The person record files and the condition record files are each concatenated across the 6-year period. The total sum, across all six years, of the condition weight times the person weight ("final basic weight") is the numerator of the rate, and the sum across all six years of the person weights is the denominator. (This is done after deleting any condition records that are duplicates for the same person.)

Statistical Analysis of Health Data

The statistical analyses performed treated the NHIS population as a standard population and applied the age- and sex-specific prevalence or period prevalence rates obtained from the NHIS data to the corresponding age- and sex-specific denominators in the Benzene Subregistry. The observed age- and sex-specific numerators for the Benzene Subregistry were compared with the expected numerators based on the NHIS rates.

This one-sample approach ignored sampling variability in the NHIS data because of the large size of the NHIS database relative to the Benzene Subregistry data file. Given that the primary focus of this report is the Benzene Subregistry, treating the NHIS versus Benzene Subregistry comparison as a two-sample problem might have resulted in a dramatic underestimation of the variability associated with the Benzene Subregistry data if any pooled variance estimates were used.

This report used the person-weights (or "final basic weight"[17]) in calculating the health condition rates for NHIS data. To allow for the nonequiprobable sampling in NHIS data, all age- and sex-specific rates that were derived from the NHIS data were weighted by the appropriate person-weights. These weights reflect the complex sampling method used by NCHS in the survey design (17). NCHS does not release with public use data the secondary inclusion (or selection) probabilities that would allow for partial adjustment for the clustering component of the NHIS survey design, so the adjustment was not used in the analyses for this report. The clustering affects only second order (variance and co-variance) estimates, and it is irrelevant because the analyses in this report treat the NHIS data as a population (without sampling error). The age- and sex-specific rates were calculated using only those respondents who were queried about each of the conditions of interest.

Paralleling Poisson regression modeling of standardized mortality ratios, the ratios of the observed-to-expected age- and sex-specific counts were modeled using Poisson regression in the Generalized Linear Interactive Modeling (GLIM) program (18). The Poisson regression approach is described in Breslow and Day (19). Maximum likelihood estimation was used, and likelihood ratio statistics using GLIM. In the Poisson regression analysis, the null model is specified by log(observed) = log(expected) + grand mean. The hypothesis that the rates are the same can be rejected when a confidence interval about the grand mean does not include the value of 1, provided that the null model is adequate. Exact confidence intervals based on the Poisson assumption were used.

By adding terms for age and sex effects to this null model, it was possible to detect structure (confounding) in the ratios of observed-to-expected prevalence rates as a function of these variables.

Statistical Analysis of Cancer Data

For the outcome cancer, incidence data from the 1989 through 1994 NHIS files were used to generate the composite expected numbers, sex- and age- specific, of all cancers. These rates were used for comparison with the observed number in the Benzene Subregistry. The generated expected rates, however, are problematic for Benzene Subregistry comparison purposes. The NHIS rates were asked in the time frame of "in the last 12 months" and are based on the collective information of several specific cancers, not all cancers. Some were queried directly, others ascertained indirectly. The Benzene Subregistry question was asked in the "ever" time frame and for all cancers. To make the Benzene Subregistry database rates comparable to the NHIS rates, the date of first treatment was used as the date of onset and to determine whether onset occurred within "the last 12 months" time frame.

Previously, cancer incidence data from the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) (20) program were used in addition to NHIS data to generate expected numbers of events for comparison with the Benzene Subregistry observed numbers. This report focuses only on NHIS comparisons. The sparsity of the data for specific cancers in each age category, particularly the younger age groups, precludes statistical comparisons for specific cancers or even for specific age groups. In addition to the follow-up information already collected, ATSDR is making a concerted effort to obtain additional information and carry out further relevant comparisons. A separate report on specific cancers reported by registrants at selected sites is in progress.

SECTION 4: RESULTS OF DESCRIPTIVE COMPARISONS




REGISTRANT DESCRIPTIVE DATA

This section provides a discussion of the comparability of descriptive data for the Benzene Subregistry file and the composite NHIS file. The NHIS and Benzene Subregistry samples were compared in terms of four demographic characteristics-sex, age, race, and education level-as well as cigarette smoking rates. The results of this section were used to plan the subsequent analyses of the health outcome data, that is, to determine what variables were appropriate to include as covariates in modeling the health outcome comparisons.

Tables 4-1 and 4-2 contain information about the characteristics of Benzene Subregistry members at all five time points described in this report; that is, Baseline and Followups 1-4. The subregistry has 1,143 members; 1,127 were living at the time of Baseline data collection and 16 were deceased. Only those living at the time of interview are included in the analyses reported in this document. For the deceased registrants, ATSDR is obtaining death certificates, and will publish a separate report on mortality in the Benzene Subregistry population.

Table 4-1 presents sex, age, education and cigarette smoking data for the composite NHIS population (1989 through 1994) and for the Benzene Subregistry population at all five time points. There were slightly more males than females in the Benzene Subregistry at Baseline through Followup 3; Followup 4 had slightly more females than males. In comparison, the composite NHIS population had slightly more females than males.

The Benzene Subregistry had more people in the youngest age group and fewer people in the three oldest age groups compared to the NHIS composite population. This difference in age

Table 4-1.—Descriptive data for living registrants (whites only), Benzene Subregistry and National Health Interview Survey.

Variable	NHIS*		Baseline		Followup 1		Followup 2		Followup 3		Followup 4
	Percentage of Total by Sex
	M	F	M	F	M	F	M	F	M	F	M	F
Total	48.8	51.2	50.7	49.3	50.5	49.5	50.4	49.6	50.2	49.8	48.4	51.6
Age group† (years) 0-9 10-17 18-24 25-34 35-44 45-54 55-64 ³65	14.9 11.1 9.6 17.0 16.1 11.3 8.7 11.4	13.5 10.1 9.3 16.3 15.5 11.1 9.1 15.1	25.4 10.4 5.7 27.6 18.7 7.2 2.6 2.4	20.6 11.6 9.1 30.5 15.1 5.7 3.6 3.8	23.2 10.4 5.5 27.4 18.5 7.1 2.4 2.4	20.5 11.6 9.3 31.3 13.9 6.4 3.3 3.5	27.4 10.8 3.5 26.3 20.1 7.1 2.4 2.4	19.6 12.1 9.0 31.9 14.4 6.1 3.6 3.4	29.7 9.8 4.0 26.2 18.6 6.6 2.8 2.3	21.1 10.9 9.6 32.7 12.9 5.8 3.3 3.6	30.7 10.3 4.1 25.1 18.3 5.9 3.5 2.1	21.3 11.6 8.0 33.8 12.5 6.1 3.3 3.3
Education§ Not high school graduate High school graduate Some college College graduate or more	19.2 35.7 20.5 24.7	19.2 40.2 21.7 19.0	11.9 37.8 29.1 21.2	13.4 43.7 31.7 11.1	10.6 39.1 26.9 23.4	12.2 44.3 31.8 11.6	10.6 36.0 30.0 23.3	9.8 41.4 36.8 12.1	8.1 35.6 31.6 24.7	9.3 43.0 33.7 14.0	7.3 33.0 34.9 24.8	7.2 39.2 38.4 15.2
Cigarette use¶ Current smoker Ever smoked	27.6 59.8	23.3 44.2	32.0 59.4	27.8 46.9	27.7 55.9	23.6 44.9	27.0 57.8	23.4 42.9	27.7 57.6	24.8 44.6	23.2 54.5	24.1 44.1

*Composite rate from 1989-1994 NHIS data.

†Age at baseline.

§³19 years of age.

¶³18 years of age.

Table 4-2.-Descriptive data for living registrants, Benzene Subregistry only.

Variable	Baseline		Followup 1		Followup 2		Followup 3		Followup 4
	Percentage of Total by Sex
	M	F	M	F	M	F	M	F	M	F
Race White Nonwhite	95.9 4.1	93.1 6.9	95.5 4.5	92.9 7.1	96.2 3.8	92.9 7.1	96.1 3.9	92.9 7.1	96.3 3.7	93.0 7.0
Occupational Status (³19 years of age) Currently employed Previously employed Never employed	91.0 9.0 0.0	66.0 32.0 2.0	NA	NA	NA	NA	NA	NA	NA	NA
Type of Interview Subject Proxy	63.4 36.6	66.0 34.0	62.9 37.1	66.9 33.1	61.8 38.2	68.5 31.5	63.0 37.0	70.5 29.5	65.1 34.9	74.0 26.0

distribution between the two populations remained consistent for all time points. Although the current smoking rates are slightly higher for the Benzene Subregistry population compared with the NHIS population, the ever smoked rates are very similar between the two populations, so smoking was not included as a covariate in the models used for analysis in this report.

Among registrants who were 19 years of age or older, 87% had at least a high school diploma at Baseline. This percentage increased to 93% by Followup 4. At all five time points nearly twice as many males than females had completed college or some post-college education.

Table 4-2 provides data on race, occupational status, and type of interview for all living Benzene Subregistry members only. Benzene registrants were overwhelmingly white (93%-96%); therefore no further analyses were performed on the nonwhite groups because of the small numbers and the potential for violating the confidentiality of the respondents. Analyses of data from all five time points were conducted for this report using only respondents who reported their race as white. Data on occupational status was collected only at Baseline because only changes in employment were captured in follow-up interviews. Most (91%) of the male registrants who were 19 years of age or older at Baseline were currently employed either full- or part-time, compared with 66% of the females.

Decedents were not included in any of the analyses presented in this report. Data collected on deceased registrants was limited to the information required to request death certificates from the state of death, and information obtained from death certificates. No lifestyle information has been or will be solicited for decedents. Death certificates have been obtained for all persons identified as deceased through Followup 4 (n=31). A separate report on mortality in the Benzene Subregistry population is in progress.

Demographic Characteristics

Gender

At Baseline (4) the distribution of the male-female ratio in the NHIS data was compared to the Benzene Subregistry. No statistically significant differences were found between the two files for this variable. Because the sex distribution in the Benzene Subregistry remained relatively constant through all time points, the statistical comparison by sex was not repeated for the followup data.

Race

Because of the small number of nonwhite registrants in the Benzene Subregistry sample (6%), all nonwhite subjects from the NHIS and Benzene Subregistry data were excluded from the analysis reported in this report. The small number and diversity of the nonwhite subpopulation (and also the potential for violating confidentiality) precluded conducting any analyses on this subpopulation.

Age

Previous reports (4) showed that there were not significant differences between the Benzene Subregistry and NHIS populations by age group and by sex within age groups. Because the age distribution in the Benzene Subregistry remained relatively stable for all time points, the age statistical comparison with the NHIS data was not repeated for the followup data.

Education Level

A previous report (4) indicated that the Benzene Subregistry participants had attained a higher level of education than the NHIS population. This difference could modify the comparison of health outcome rates in that the Benzene Subregistry population might be expected to have fewer outcomes related to lower socioeconomic factors. However, because of the small sample size of the Benzene Subregistry, education level is not included as a factor in the regression model.

Cigarette Smoking

A more in-depth review of cigarette use in the Benzene Subregistry compared to the NHIS population was previously reported (4). In summary, the smoking rates by age and sex for the NHIS population and the Benzene Subregistry population were generally very similar at all times, so the omission of smoking as a covariate in any of the statistical models used for this report probably does not impact any of the results.

SECTION 5: RESULTS OF HEALTH OUTCOME COMPARISONS

Overall, aggregate reporting rates for health conditions for the Benzene Subregistry population for each of the time points Baseline through Followup 4 and the NHIS composite population are provided Appendix X, Tables 1 through 5. A summary of the Benzene Subregistry population's reporting rates for specific cancers (total population and by sex) at each of the five data collection time points can be found in Appendix Y, Tables 1-5. A summary of the results of the NHIS and Benzene Subregistry file comparisons using Poisson regression analysis can be found in Appendix Z.

The results for each health condition with statistically significant excess reporting by Benzene Subregistry members are presented, according to the appropriate model, in Table 5-4. A discussion of the statistical results for each health outcome follows. Note, the results for some health outcomes have changed from those reported previously (4) due to the use of composite NHIS comparison values instead of individual year results.

For the results of the statistical analyses presented below, the standardized morbidity ratios (SMR) are presented in Table 5-5 and are defined as the number of each health condition reported by the Benzene Subregistry registrants (observed, O) divided by the number expected based on the number of each health condition reported by the NHIS participants (expected, E). The following discussion is based on the results summaries found in Tables 5-4 and 5-5.

Table 5-4.—Summary of statistically significant results* for health outcomes reported in excess at Baseline.

1. Grand Mean, No Age or Sex Effect.

Structure	Health Condition
	Cancer
Overall Summary	2.44

2. Sex Effect Only.

Sex	Health Condition
	Urinary Tract Disorders
Females	5.60

Table 5-4.—Continued.

3. Age Effect Only.

Age (Years)	Health Condition
	Skin Rash	Kidney Disease	Respiratory Allergies
0-9	2.63		3.43
10-17
18-24
25-34
35-44
45-54
55-64		7.33
³65	4.59

*Significance level: p£0.01
4. By Age and Sex.
(a) Males
No significant excesses for Benzene Subregistry compared with NHIS.
(b) Females.
No significant excess for Benzene Subregistry compared with NHIS.


Table 5-5.–Aggregate observed and expected health outcomes using multivariate models at Baseline.

Condition	Observed	Expected	Risk Ratio	99% CI*
Anemia and other blood disorders†	40	13.85	2.89	1.85, 4.29
Arthritis§	46	104.78	0.44	0.29, 0.64
Asthma, emphysema§	56	101.95	0.55	0.38, 0.77
Cancer§	13	5.33	2.44	1.05, 4.79
Diabetes¶	16	14.49	1.10	0.52, 2.03
Hearing impairment§	13	63.18	0.21	0.09, 0.40
Hypertension§	60	60.72	0.99	0.69, 1.37
Kidney disease¶	15	12.13	1.24	0.57, 2.32
Liver problems§	7	2.17	3.23	0.94, 7.90
Mental retardation§	2	6.89	0.29	0.02, 1.35
Respiratory allergies¶	117	108.75	1.08	0.84, 1.36
Skin rashes¶	89	70.00	1.27	0.95, 1.66
Speech impairment§	6	11.33	0.53	0.14, 1.38
Stomach problems, ulcers§	55	62.92	0.87	0.60, 1.23
Stroke§	10	4.50	2.22	0.83, 4.75
Urinary tract disorders**	32	8.73	3.67	2.21, 5.69

*CI- Confidence interval for risk ratio.

†No summary model.

§Summary model: grand mean.

¶Summary model: age effect.

**Summary model: sex effect.

§§Summary model: age and sex effect.

Anemia or Other Blood Disorders

There was a statistically significant overall excess of cases reported by members of the Benzene Subregistry population compared to the NHIS population (SMR = 2.84, 99% CI = 1.81-4.21). However, there was no summary model because the model fit was marginal for anemia, thus comparisons to NHIS are best made within age- and sex-specific categories. No significant differences by age and sex were found, due to the small power of these tests.

Arthritis, Rheumatism, or Other Joint Disorders

The overall SMR for this outcome was statistically significant less than 1, indicating that the reporting rates were generally lower in the Benzene Subregistry Baseline data than expected based on NHIS reporting rates (O/E = 0.44, 99% CI = 0.29-0.64).

Asthma, Emphysema, or Chronic Bronchitis

The overall SMR for this outcome was statistically significant less than 1 (O/E = 0.55, 99% CI = 0.38-0.77) indicating a statistically significantly decreased reporting in this group relative to the NHIS population.

Cancers

The overall SMR was statistically significant (O/E = 2.44, 99% CI = 1.05-4.79) meaning there was greater reporting of cancer by the Benzene Subregistry Baseline population compared to the NHIS population. However, the numbers of reported site-specific cancers were too small to perform site-specific statistical analyses. ATSDR is validating reported cancers, and obtaining state and regional cancer rates using state cancer registries for use in further statistical analyses. Results of these analyses, as well as comparison to SEER rates, will provide better insight in interpreting cancer reporting rates. A detailed report on cancer rates in the Benzene Subregistry is forthcoming.

Diabetes

No cases were reported by Benzene Subregistry members at Baseline in the 0 through 9 years, 18 through 24 years, and 45 through 54 years age groups. Ten of the 16 reported cases were reported by those in the 55 through 64 years (O = 6) and 65 years and older (O = 4) age groups. No statistically significant results were seen in any age group.

Hearing Impairment

For this outcome, there was statistically significantly decreased reporting overall for the Benzene Subregistry population compared with the NHIS population (O/E = 0.21, 99% CI = 0.09-0.40).

High Blood Pressure (Hypertension)

No statistically significant results were observed for the Benzene Subregistry reporting rates when compared with the NHIS population reporting rates. The observed number of cases almost exactly matched the expected, resulting in a SMR = 0.99.

Kidney Disease

The model indicated that age was a significant factor. The only statistically significant result was for registrants aged 55 through 64 years (O/E = 7.33, 99% CI = 1.58-20.75).

Liver Problems

There were no reported cases in many of the age- and sex-specific groups. Based on seven reported cases, the overall SMR (3.23) was elevated but was not statistically significant.

Mental Retardation

While the model was adequate, neither sex nor age were significant terms. Only two cases were reported in the Benzene Subregistry Baseline population, yet the results indicated that there was not a statistically significant under-reporting of this condition.

Other Respiratory Allergies or Problems, Such as Hay Fever

The significant factor in the model was age. Statistically significant results were seen in the 0 through 9 years age group (O/E = 3.43, 99% CI = 2.15-5.17).

Skin Rashes, Eczema, or Other Skin Allergies

The significant factor in the model was age. Statistically significant results were observed in the 0 through 9 years of age group (O/E = 2.63, 99% CI = 1.57-4.11) and the 65 years and older age group (O/E = 4.59, 99% CI = 1.97-9.00).

Speech Impairment

Only six cases of speech impairment were reported by Benzene Subregistry members at Baseline. The overall SMR was less than 1 but was not statistically significant.

Effects of Stroke

Although neither the effects of age nor sex were significant in the model, 7 of the 10 cases of stroke reported in the Benzene Subregistry population at Baseline were male. The overall SMR (2.22) was elevated but not statistically significant.

Ulcers, Gallbladder Trouble, and Stomach or Intestinal Problems

Excess reporting in the two youngest age groups resulted in an age effect observed in the model for this outcome. However, no statistically significant results were seen in any age group.

Urinary Tract Disorders, Including Prostate Trouble

Sex was a significant factor in the model, (O/E = 5.60, 99% CI = 3.17-9.10), driven by excess reports of urinary tract disorders by females. Excesses were seen in females in the following age groups: 0 through 9 years, 10 through 17 years, 25 through 34 years, and 35 through 44 years.

SUMMARY

Statistically significant excesses were observed for the following conditions: anemia or other blood disorders; cancer; kidney disease; skin rashes, eczema, or other skin allergies; other respiratory allergies or problems, such as hay fever; and urinary tract disorders, including prostate trouble. Table 5-4 presents a summary of the statistically significant (p£0.01) risk ratios observed in the Poisson analysis of Baseline data for health outcomes reported in excess. Results of health outcome comparisons, aggregated across age and sex, for all health outcomes reported by Benzene Subregistry members at Baseline can be found in Table 5-5.

SECTION 6: RESULTS OF HEALTH OUTCOME COMPARISONS-FOLLOWUP 1

Overall, aggregate reporting rates for health conditions for the Benzene Subregistry file at Followup 1 and the NHIS file are provided in Table 6-1. A summary of the Followup 1 Benzene Subregistry population's reporting rates for specific cancers (total population and by sex) is shown in Table 6-2 for the "within the last 12 months" time frame. The "within last 12 months" time frame is comparable to the NHIS time frame; the 12-month rates were used in the statistical comparisons of the data files. At the time of the interview, if a registrant reported having been told that he or she had cancer or had been treated for cancer by a health care provider, further questions were asked about the type(s) of cancer. Although multiple types of cancers may have been reported, only one primary type of cancer is assigned to each registrant. The summary table entries and data analysis are based on the reported primary cancers.

CONTROLLING FOR AGE AND SEX

Table 6-3 provides a summary of the results of the NHIS and Followup 1 Benzene Subregistry file comparison using Poisson regression analysis. For each health outcome, the table indicates the likelihood ratio statistics with the associated degrees of freedom and p-values for the effects of age (categorized into eight levels) and sex, based on a model containing age and sex. The residual deviance and the associated degrees of freedom are also given as a global lack-of-fit measure for this model, which specifies multiplicative effects of the age (i) and sex (j) ratios Oij/Eij. For each outcome, the age- and sex-specific numerators Oij were obtained from the Benzene Subregistry data, while the

Table 6-1.—Comparison of Benzene Subregistry and National Health Interview Survey participants reporting health condition at Followup 1.

Health Condition	Benzene Subregistry		NHIS
	Males †	Females †	Males †	Females †
Anemia and other blood disorders*	1.4	6.4	0.4	2.3
Arthritis*	4.5	5.6	13.2	17.2
Asthma, emphysema*	5.9	8.3	8.7	10.9
Cancer*	1.0	2.3	0.9	1.1
Diabetes*	0.0	0.2	2.3	2.5
Hearing impairment¶	0.8	0.0	10.7	7.1
Hypertension*	6.3	5.8	9.2	10.0
Kidney disease*	0.4	1.7	1.0	1.7
Liver problems*	1.0	0.6	0.3	0.2
Mental retardation¶	0.2	0.0	0.7	0.5
Respiratory allergies*	13.2	17.8	10.3	10.9
Skin rashes*	8.9	10.2	5.8	8.0
Speech impairment¶	0.0	0.0	1.2	0.6
Stomach problems/ulcer*	5.3	7.7	6.4	8.9
Stroke§	1.2	0.8	0.9	1.0
Urinary tract disorders*	4.3	7.5	1.4	1.2

*Indicates time frame is last 12 months.
†Percent of total population (white only).
§Indicates time frame is ever.
¶Indicates time frame is now have.

Table 6-2.—Summary of Benzene Subregistry registrants* reporting at least one cancer at Followup 1.

Cancer	Sex				Total
	Male		Female
	N	%	N	%	N	%
None	487	99.0	471	97.7	958	98.4
Lip, oral, pharynx	0	0.0	0	0.0	0	0.0
Digestive system	0	0.0	0	0.0	0	0.0
Respiratory system	1	0.2	0	0.0	1	0.1
Malignant skin	2	0.4	1	0.2	3	0.3
Breast	0	0.0	4	0.8	4	0.4
Genital organs	1	0.2	5	1.0	6	0.6
Urinary organs	0	0.0	0	0.0	0	0.0
Lymphatic tissues	0	0.0	0	0.0	0	0.0
Leukemia	1	0.2	0	0.0	1	0.1
Other†	0	0.0	1	0.2	1	0.1
Total	492	100.0	482	100.0	974	100.0

*White, exposed registrants only; time frame is "last 12 months."
†Unspecified.

Table 6-3.—Summary of Poisson regression modeling for Followup 1.

Condition	Age/Sex		Sex/Age		Residual Deviance (p-value)	df
	LR Stat* (p-value)	df†	LR Stat (p-value)	df
Anemia and other blood disorders	15.85 (0.03)	7	.051 (0.48)	1	3.86 (0.80)	7
Arthritis	19.90 (0.01)	7	0.00 (0.96)	1	5.63 (0.58)	7
Asthma, emphysema	9.82 (0.20)	7	1.31 (0.25)	1	13.77 (0.06)	7
Cancer	10.11 (0.18)	7	0.66 (0.42)	1	7.23 (0.40)	7
Diabetes	22.68 (0.00)	7	0.61 (0.43)	1	8.67 (0.28)	7
Hearing impairment	9.79 (0.20)	7	3.33 (0.07)	1	0.00 (1.00)	7
Hypertension	5.99 (0.54)	7	0.00 (0.95)	1	4.21 (0.76)	7
Kidney disease	8.34 (0.30)	7	0.77 (0.38)	1	6.79 (0.45)	7
Liver problems	13.50 (0.06)	7	0.13 (0.72)	1	3.80 (0.80)	7
Mental retardation	2.10 (0.95)	7	0.84 (0.36)	1	0.00 (1.00)	7
Respiratory allergies	56.18 (0.00)	7	3.88 (0.05)	1	2.88 (0.90)	7
Skin rashes	25.50 (0.00)	7	0.39 (0.53)	1	9.52 (0.22)	7
Speech impairment	0.00 (1.00)	7	0.00 (1.00)	1	0.00 (1.00)	7
Stomach problems/ulcer	19.53 (0.01)	7	0.01 (0.93)	1	12.49 (0.09)	7
Stroke	16.14 (0.02)	7	0.52 (0.47)	1	3.89 (0.79)	7
Urinary tract disorders	13.67 (0.06)	7	0.65 (0.42)	1	3.41 (0.84)	7

*LR Stat = Likelihood Ratio Statistic.

†df = degrees of freedom.

expected numerators Eij were based on the suitably person-weighted age- and sex-specific ratios from the NHIS data. For the purpose of detecting structure in these age- and sex-specific ratios, a significance level of 0.05 was adopted.

As is shown in Table 6-3, the model was adequate (p>0.1) but neither age nor sex was a statistically significant predictor in the models for the health outcomes cancer; hearing impairment; hypertension; kidney disease; liver problems; mental retardation; speech impairment; and urinary tract disorders, including prostate trouble. For the outcomes anemia and other blood disorders; arthritis, rheumatism or other joint disorders; diabetes; skin rashes, eczema, or other skin allergies; and effects of stroke statistically significant variations in the ratios were seen as a function of age. Sex was not a significant factor in the model for any health outcome. Both age and sex were significant factors in the model for other respiratory allergies or problems, such as hay fever. The model fit was marginal (0.01<p<0.1) for the outcomes asthma, emphysema, or chronic bronchitis; and ulcers, gallbladder trouble, and stomach or intestinal problems.

The results for each health condition with statistically significant excess reporting by Benzene Subregistry members are presented, according to the appropriate model, in Table 6-4. A discussion of the statistical results for each health outcome follows.

For the results of the statistical analyses presented below, the standardized morbidity ratios (SMRs) are presented in Table 6-5 and are defined as the number of each health condition reported by the Benzene Subregistry registrants (observed, O) divided by the number expected based on the number of each health condition reported by the NHIS participants (expected, E). The following discussion is based on the results summaries found in Tables 6-4 and 6-5.

Table 6-4.—Summary of statistically significant results* for health outcomes reported in excess at Followup 1.

1. Grand Mean, No Age or Sex Effect.

Structure	Health Condition
	Cancer	Liver Problems	Urinary Tract Disorders
Overall Summary	3.15	3.80	6.89

2. Sex Effect Only.

This model was not applicable for any health condition at Followup 1.

Table 6-4.—Continued.

3. Age Effect Only.

Age (Years)	Health Condition
	Anemia	Arthritis	Skin Rash	Stroke
1-10		7.25	2.93
11-18
19-25
26-35				16.31
36-45
46-55	9.15
56-65	11.68

4. By Age and Sex .

(a) Males.

Age (Years)	Health Condition
	Respiratory Allergies
1-10	3.38

Table 6-4.—Continued.

(b) Females

Age (Years)	Health Condition
	Respiratory Allergies	Stomach Problems, Ulcer
1-10	4.64	4.76
56-65	4.44

*Significance level: p£0.01.

Anemia or Other Blood Disorders

Age was the significant factor in the model for anemia and other blood disorders. Statistically significant results were observed for the age group 46 through 55 years (O/E = 9.15, 99% CI = 2.66-22.38), and the age group 56 through 65 years (O/E = 11.68, 99% CI = 1.96-36.79).

Arthritis, Rheumatism, or Other Joint Disorders

Age had a statistically significant effect in the model due to 4 cases reported in the 1 through 10 years age group (O/E = 7.25, 99% CI = 1.22-22.82). Statistically significant deficits in the Benzene Subregistry Baseline population compared with the NHIS population were seen in the 26 through 35 years and 36 through 45 years of age groups.

Table 6-5.—Summary of observed and expected health outcomes using multivariate models for Followup 1.

Condition	Observed	Expected	Risk Ratio	99% CI*
Anemia and other blood disorders¶	38	12.96	2.93	1.85, 4.39
Arthritis¶	49	99.55	0.49	0.33, 0.70
Asthma, emphysema†	69	95.25	0.72	0.52, 0.98
Cancer§	16	5.08	3.15	1.49, 5.80
Diabetes¶	13	13.74	0.95	0.41, 1.86
Hearing impairment§	4	59.71	0.07	0.01, 0.21
Hypertension§	59	57.50	1.03	0.72, 1.42
Kidney disease§	10	11.31	0.88	0.33, 1.89
Liver problems§	8	2.11	3.80	1.22, 8.83
Mental retardation§	1	6.51	0.15	0.00, 1.14
Respiratory allergies**	151	101.47	1.49	1.20, 1.83
Skin rashes¶	93	65.13	1.43	1.08, 1.86
Speech impairment§	0	10.77	0.00	0.00, 0.43
Stomach problems, ulcers†	63	58.72	1.07	0.76, 1.47
Stroke¶	10	4.24	2.36	0.88, 5.04
Urinary tract disorders§	57	8.27	6.89	4.77, 9.61

*CI - Confidence interval for risk ratio.

†No summary model.

§Summary model: grand mean.

¶Summary model: age effect.

**Summary model: age and sex effect.

Asthma, Emphysema, or Chronic Bronchitis

There was no summary model because the model fit was marginal, thus comparisons to NHIS are best made within age- and sex-specific categories. The only statistically significant result seen for this outcome was a deficit in females aged 11 through 18 years, based on 0 cases reported in the Benzene Subregistry Followup 1 population.

Cancers

The overall SMR (O/E = 3.15) was statistically significant (99% CI = 1.49-5.80) meaning there was greater reporting of cancer by the Benzene Subregistry Followup 1 population compared to the NHIS population. As stated in Section 4, ATSDR is conducting an in-depth analysis of cancers, and will publish the results in a separate report.

Diabetes

Age had a significant effect in the model. Although no statistically significant results were seen in any of the age groups, a slight majority of diabetes cases (7 of 13, or 54%) were reported in the two oldest age groups of the Benzene Subregistry population at Followup 1.

Hearing Impairment

Only four cases of hearing impairment were reported by members of the Benzene Subregistry Followup 1 population. Relative to the NHIS population, there was statistically significantly decreased reporting for the Benzene Subregistry population (O/E = 0.07, 99% CI = 0.01-0.21).

High Blood Pressure (Hypertension)

There were 59 reports of high blood pressure in the Benzene Subregistry Followup 1 population. While the overall SMR was elevated, it was not statistically significant.

Kidney Disease

There were 10 cases of kidney disease reported by the Benzene Subregistry Followup 1 population. No statistically significant results were seen, indicating that reporting of this outcome by Benzene Subregistry members was not very different from that by the NHIS population.

Liver Problems

A statistically significant excess of liver disease was seen in the Benzene Subregistry Followup 1 population compared with the NHIS population (O/E = 3.80, 99% CI = 1.22-8.83). This result is based on 8 cases of liver disease reported by members of the Benzene Subregistry at Followup 1.

Mental Retardation

Only 1 case was reported in the Benzene Subregistry Followup 1 population. The overall SMR was less than 1 but was not statistically significant.

Other Respiratory Allergies or Problems, Such as Hay Fever

Both age and sex were significant factors in the model. Statistically significant results were observed for males aged 1 through 10 years (O/E = 3.38, 99% CI = 1.84-5.66), females aged 1 through 10 years (O/E = 4.64, 99% CI = 2.56-7.68), and females aged 56 through 65 years (O/E = 4.44, 99% CI = 1.29-10.86).

Skin Rashes, Eczema, or Other Skin Allergies

Age was a significant factor in the model. There was an excess reporting of this outcome by the Benzene Subregistry population at Followup1 compared with the NHIS population in the 1 through 10 years of age group (O/E = 2.93, 99% CI = 1.80-4.49).

Speech Impairment

No cases of speech impairment were reported by the Benzene Subregistry population at Followup 1. This result is a statistically significant under-reporting of this condition in the Benzene Subregistry.

Effects of Stroke

The model for effects of stroke indicated that age was a significant factor; statistically significant results were observed in the 26 through 35 years age group (O/E = 16.31, 99% CI = 2.74-51.35).

Ulcers, Gallbladder Trouble, and Stomach or Intestinal Problems

There was no summary model because the model fit was marginal, thus comparisons to NHIS are best made within age- and sex-specific categories. The only statistically significant result was in females aged 1 through 10 years (O/E = 4.76, 99% CI = 1.22-12.41), based on 6 cases. The SMRs were elevated in all age and sex groups up to age 25, and also for males aged 46 through 55 years and females aged 56 through 65 years but none were statistically significant.

Urinary Tract Disorders, Including Prostate Trouble

There was an overall excess of reporting of this outcome in the Benzene Subregistry population at Followup 1 compared with the NHIS population (O/E = 6.89, 99% CI = 4.77-9.61). Although the model was the grand mean, excesses were observed in the following age- and sex-specific groups: males aged 1 through 10 years, females aged 1 through 10 years, females aged 11 through 18 years, females aged 19 through 25 years, males aged 26 through 35 years, females aged 26 through 35 years, males aged 36 through 45 years, and females aged 36 through 45 years.

SUMMARY

Statistically significant excesses were observed for the following conditions: anemia and other blood disorders; arthritis, rheumatism or other joint disorders; cancer; liver problems; skin rashes, eczema, or other skin allergies; other respiratory allergies or problems, such as hay fever; ulcers, gallbladder trouble, and stomach or intestinal problems; effects of stroke; and urinary tract disorders, including prostate trouble. Table 6-4 presents a summary of the statistically significant (p£0.01) risk ratios observed in the Poisson analysis of Followup 1 data for health outcomes reported in excess. Results of health outcome comparisons, aggregated across age and sex, for all health outcomes reported by Benzene Subregistry members at Followup 1 can be found in Table 6-5.

SECTION 7: RESULTS OF HEALTH OUTCOME COMPARISONS-FOLLOWUP 2

Aggregate reporting rates for health conditions for the Benzene Subregistry file at Followup 2 and the NHIS file are provided in Table 7-1. A summary of the Benzene Subregistry Followup 2 population's reporting rates for specific cancers (total population and by sex) is shown in Table 7-2 for the "within the last 12 months" time frame. The "within last 12 months" time frame is comparable to the NHIS time frame; the 12-month rates were used in the statistical comparisons of the data files. At the time of the interview, if registrants reported that a health care provider told them they had cancer or if they had been treated for cancer, further questions were asked about the type(s) of cancer. Although multiple types of cancers may have been reported, only one primary type of cancer is assigned to each registrant. The summary table entries and data analysis are based on the reported primary cancers.

CONTROLLING FOR AGE AND SEX

Table 7-3 summarizes of the results of the NHIS and Benzene Subregistry file comparison using Poisson modeling. For each health outcome, the table indicates the likelihood ratio statistics with the associated degrees of freedom and p-values for the effects of age (categorized into eight levels) and sex, based on a model containing age and sex. The residual deviance and the associated degrees of freedom are also given as a global lack-of-fit measure for this model, which specifies multiplicative effects of the age (i) and sex (j) ratios Oij/Eij. For each outcome, the age- and sex-specific numerators Oij were obtained from the Benzene Subregistry data, while the

Table 7-1.—Comparison of Benzene and National Health Interview Survey participants reporting health condition at Followup 2.

Health condition	Benzene Subregistry		NHIS
	Males †	Females †	Males †	Females †
Anemia and other blood disorders*	1.3	7.4	0.4	2.3
Arthritis*	4.9	6.7	13.2	17.2
Asthma, emphysema*	6.2	10.3	8.7	10.9
Cancer*	1.3	1.6	0.9	1.1
Diabetes*	1.1	2.5	2.3	2.5
Hearing impairment¶	1.1	1.4	10.7	7.1
Hypertension*	6.8	4.0	9.2	10.0
Kidney disease*	1.1	1.8	1.0	1.7
Liver problems*	1.3	0.2	0.3	0.2
Mental retardation¶	0.0	0.0	0.7	0.5
Respiratory allergies*	18.3	18.7	10.3	10.9
Skin rashes*	5.7	9.7	5.8	8.0
Speech impairment¶	0.7	0.5	1.2	0.6
Stomach problems/ulcer*	6.2	8.8	6.4	8.9
Stroke§	1.1	1.4	0.9	1.0
Urinary tract disorders*	4.0	8.5	1.4	1.2

* Indicates time frame is "last 12 months."

† Indicates percentage of total population (white only).

§ Indicates time frame is "ever."

¶ Indicates time frame is "now have."

Table 7-2.—Summary of Benzene Subregistry registrants reporting at least one cancer at Followup 2.*

Cancer					Total
	Male		Female
	Number	%	Number	%	Number	%
None	447	98.7	438	98.6	885	98.6
Breast	0	0.0	1	0.2	1	0.1
Digestive system	1	0.2	2	0.4	3	0.3
Genital organs	0	0.0	2	0.4	2	0.2
Leukemia	1	0.2	0	0.0	1	0.1
Lip, oral, pharynx	0	0.0	0	0.0	0	0.0
Lymphatic tissues	0	0.0	0	0.0	0	0.0
Malignant skin	4	0.9	2	0.4	6	0.7
Respiratory system	0	0.0	0	0.0	0	0.0
Urinary organs	0	0.0	0	0.0	0	0.0
Other	0	0.0	0	0.0	0	0.0
Total	453	100.0	445	100.0	898	100.0

* White, exposed registrants only; time frame is "last 12 months."

Table 7-3.—Summary of Poisson modeling for Followup 2.

Condition	Age/Sex		Sex/Age		Residual deviance (p-value)	df
	LR stat* (p-value)	df†	LR stat (p-value)	df
Anemia and other blood disorders	13.45 (0.06)	7	0.80 (0.37)	1	1.96 (0.96)	7
Arthritis	17.74 (0.01)	7	0.21 (0.64)	1	2.26 (0.94)	7
Asthma, emphysema	5.98 (0.54)	7	3.21 (0.07)	1	6.40 (0.49)	7
Cancer	11.68 (0.11)	7	0.35 (0.55)	1	3.64 (0.82)	7
Diabetes	13.20 (0.07)	7	1.36 (0.24)	1	9.71 (0.20)	7
Hearing impairment	14.12 (0.05)	7	1.00 (0.32)	1	11.58 (0.12)	7
Hypertension	2.74 (0.91)	7	2.83 (0.09)	1	2.80 (0.90)	7
Kidney disease	10.53 (0.16)	7	0.28 (0.60)	1	7.40 (0.39)	7
Liver problems	7.49 (0.38)	7	1.84 (0.17)	1	4.56 (0.71)	7
Mental retardation	0.00 (1.00)	7	0.00 (1.00)	1	0.00 (1.00)	7
Respiratory allergies	40.44 (0.00)	7	0.18 (0.67)	1	170.68 (0.01)	7
Skin rashes	16.80 (0.02)	7	0.47 (0.49)	1	6.90 (0.44)	7
Speech impairment	6.65 (0.47)	7	0.31 (0.58)	1	0.00 (1.00)	7
Stomach problems/ulcer	13.72 (0.06)	7	0.01 (0.92)	1	10.60 (0.16)	7
Stroke	5.21 (0.63)	7	0.00 (0.98)	1	8.52 (0.29)	7
Urinary tract disorders	8.63 (0.28)	7	2.56 (0.11)	1	5.03 (0.66)	7

* LR Stat = Likelihood Ratio Statistic.

† df = degrees of freedom.

expected numerators Eij were based on the suitably person-weighted age- and sex-specific ratios from the NHIS data. For the purpose of detecting structure in these age- and sex-specific ratios, a significance level of 0.05 was adopted.

As Table 7-3shows, the model was adequate (p > 0.1) but neither age nor sex was a statistically significant predictor in the models for the anemia and other blood disorders; asthma, emphysema, or chronic bronchitis; cancer; diabetes; hypertension; kidney disease; liver problems; mental retardation; speech impairment; ulcers, gallbladder trouble, and stomach or intestinal problems; effects of stroke; and urinary tract disorders, including prostate trouble. Statistically significant variations in the ratios were seen as a function of age arthritis, rheumatism or other joint disorders; hearing impairment; and skin rashes, eczema, or other skin allergies. Sex was not a significant factor in the model for any health outcome. The model fit was marginal (0.01 < p < 0.1) for other respiratory allergies or problems, such as hay fever.

The results for each health condition with statistically significant excess reporting by Benzene subregistry registrants are presented, according to the appropriate model, in Table 7-4. A discussion of the statistical results for each health outcome follows.

For the results of the statistical analyses presented below, the SMRs are in Table 7-5 and are defined as the number of each health condition reported by the Benzene Subregistry registrants (observed, O) divided by the number expected based on the number of each health condition reported by the NHIS participants (expected, E). The following discussion is based on the results summaries in Tables 7-4 and 7-5.

Table 7-4.—Summary of statistically significant results for health outcomes reported in excess for Followup 2.*

1. Grand mean, no age or sex effect.

Structure	Health condition
	Anemia	Cancer