• National Exposure Registry
• Benzene
• Dioxin
• Journal Articles
• Trichloroethylene(TCE)
• Trichloroethane(TCA)
• Registry Data

SECTION 3

Comparison Of TCE Subregistry Followup 1 Data And National Norms Rationale For The Comparison

This section contains comparisons of the Trichloroethylene (TCE) Subregistry Followup 1 reporting rates for health conditions with those from a national survey (1990 National Health Interview Survey [NHIS]) (6). These comparisons are consistent with the objectives and goals of the National Exposure Registry (NER) as stated in The National Exposure Registry Policies and Procedures Manual (Revised) (4); one such objective is to provide a preliminary assessment of the extent to which TCE Subregistry members have an excess of adverse health conditions and to generaterather than testhypotheses about TCE exposure and health outcomes. Essentially the same comparisons were made at Baseline with the 1989 NHIS data files (5). The comparison with national norms, specifically to the appropriate year NHIS file, was done to look for consistency between the NER and NHIS files in reporting rates for the same or related health effects. As more data become available with successive followups, other analyses will be carried out that will focus on trends in reporting rates over time within the TCE Subregistry.

In addition to a comparison of the TCE Subregistry Followup 1 health data reporting rates with national health data norms, this section includes a comparison of registrant demographic data with national data. The comparison of demographic characteristics indicates the extent to which TCE Subregistry members at this first followup were similar to the NHIS population.

NATIONAL COMPARISON POPULATIONS

As discussed in the Baseline report (1), the NHIS is the most appropriate comparison population--available because it is a probability sample of the residential, noninstitutionalized U.S. population, the population of interest when assessing 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 rates and norms. The NHIS, similar to the NER, is composed of self-reported data; however, unlike the TCE Subregistry Followup 1 data, the information was obtained using face-to-face interviews. Appro-priate sex-, age-specific subsets of the NHIS data were selected as the comparison data for the NER components. The weighting factors provided by the National Center for Health Statistics (NCHS) were applied when using the data; otherwise, because of the NHIS sample design, the estimates would have been biased.

As discussed in Section 2 of this report, about 90% of the TCE Subregistry sample is located in the East North Central United States (in Illinois, Indiana, and Michigan), with the remainder located throughout other regions. A review of the 1989 and the 1990 NHIS regional rates for selected outcomes found no definitive evidence indicating that the overall health status of midwesterners differed significantly from that of the general U.S. population. In the NCHS publication Current Estimates (6), the 1990 rates for the geographic regions Northeast, Midwest, South, and West (including all races) for the three health conditions reported most often in the TCE Followup 1--stroke, anemia, and urinary tract disorders--showed only relatively moderate fluctuations. The rates for the TCE Followup 1 were higher than the regional rates including the Midwest region--the region where most of the registrants resided. Therefore, it appears differences in reporting rates between the TCE Subregistry Followup 1 file and the NHIS file were greater than differences that might have been attributable to regionality alone.

METHODS OF DESCRIPTIVE VARIABLE COMPARISONS

Demographic Characteristics

The 1990 NHIS and the TCE Subregistry Followup 1 data were compared in terms of the demographic characteristics sex, age, and race. As discussed in the Baseline report (1), each of these variables, along with education level and cigarette smoking rates, is a potential correlate of health status.

Race

Race is an established correlate of socioeconomic status (7) and health status (8). National data indicate that nonwhites have lower rates for cigarette smoking (9). For these reasons, race is a potential control variable for the comparisons of health status and smoking rates.

In the 1989 NHIS sample, 18% (n = 21,066) of the participants were nonwhite; in the TCE Subregistry Baseline sample, 3% (n = 127) of the participants reported their race as other than white. Similarly, in the 1990 NHIS sample, 18.3% (n = 21,899) of the participants were nonwhite and 3% (n = 102) of the TCE Subregistry Followup 1 participants reported their race as other than white. Given this discrepancy in the proportion of nonwhites and the diversity of races reported among the nonwhites in the TCE Subregistry, all nonwhite subjects from the NHIS and TCE Subregistry data were excluded from the Baseline and Followup 1 analyses; therefore, the statistical analyses reported in remainder of this report include 3,579 living, exposed white TCE registrants.

Sex

As at Baseline, the sex distributions of the NHIS and Followup 1 data files were compared. The numerical distributions and the statistical results of the comparisons of the Followup 1 data and 1990 NHIS data are presented later in this section. The distribution of the male-female ratio was assessed overall and on a sex-, and age-specific basis.

The modeling method used to test differences in reporting rates took into account any effect due to sex. Thus, even if there had been differences in the distributions of males and females in the NER compared with the NHIS, this would not have affected the comparisons.

The distribution of males and females in the TCE Subregistry at Followup 1 was quite close to the distribution in the 1990 NHIS data. Table 3-1 indicates that the proportion of females (likewise for males) did not differ statistically significantly between the TCE and NHIS samples overall, nor for any of the age categories examined (p > 0.10 in all cases).

Table 3-1.--Comparison of the distributions of females in the TCE Subregistry Followup 1 and 1990 National Health Interview Survey (NHIS), by age group.

Age Group (Years)	NHIS*	TCE Followup 1*	Chi-square Value	p value†
1 - 10	0.479	0.487	0.093	0.76
11 - 18	0.483	0.520	2.496	0.11
19 - 25	0.522	0.521	0.003	0.95
26 - 35	0.519	0.532	0.504	0.48
36 - 45	0.510	0.521	0.285	0.59
46 - 55	0.516	0.507	0.115	0.74
56 - 65	0.535	0.545	0.124	0.72
over 65	0.591	0.588	0.008	0.93
Overall	0.517	0.527	1.335	0.25

*Cell values are proportions for a given age group of total females.

†The p-values are for the two-sided hypothesis: TCE proportion = NHIS

proportion.

Age

The age distributions of the two groups were compared, using an eight-category measure and comparison methods similar to those used for the Baseline analyses. The results of the Baseline comparisons are discussed in the Baseline report (1). The numerical distributions and the statistical results of the comparisons for Followup 1 data are presented later in this section.

The Poisson regression analyses used an eight-category measure of age; the NER observed and the expected counts based on NHIS data used in the Poisson regression modeling were age-, sex-categories specific. Thus, just as noted for the factor sex, the results of the statistical comparisons for the health effects did not depend on whether the age distribution of the NHIS file was equal to the age distribution of the TCE Subregistry file.

As expected, and as was found at Baseline, the age percentages were affected by the timing of data collection relative to the exposure period. For example, the percentage of people in the 10 years of age or younger group was less for the TCE Subregistry than for the NHIS sample. For some of the TCE sites, the exposure period ended several years prior to data collection and precluded the eligibility of young children (that is, children born after the exposure period ended). For some sites, exposure was occurring at the time of Baseline data collection, and ensured the potential eligibility of children who were 10 years of age or younger at the time of the interview. However, all known exposure due to drinking contaminated water had ceased by the time of Baseline data collection. As expected, and also as was found at Baseline, there were fewer TCE Subregistry members in the oldest age group. This pattern was consistent with ATSDR knowledge about the demographics of these TCE sites.

Table 3-2 provides a comparison of the age distributions in the TCE Subregistry and NHIS samples. The chi-square value for the difference between the TCE and NHIS proportions were statistically different at p 0.01. The cell contributions to the chi-square values indicate that much of the difference between the two files lay in the two youngest and the oldest age categories, confirming the a priori expectations of differences in the age distributions between the TCE Subregistry and the NHIS data files. As discussed previously, this difference was taken into account in the Poisson regression

Table 3-2.--Comparison of the age distributions in the TCE Subregistry Followup 1 and the 1990 National Health Interview Survey (NHIS) sample.

Age Group (Years)	NHIS		TCE Followup 1
	Number*	% of Total	Number*	% of Total
1-10	14,213	15.2	314	9.0
11-18	10,257	11.0	467	13.4
19-25	8,829	9.4	386	11.1
26-35	16,190	17.3	744	21.4
36-45	14,157	15.1	589	17.0
46-55	9,723	10.4	355	10.2
56-65	8,546	9.1	288	8.3
over 65	11,621	12.4	328	9.4
All	93,536	100.0	3,471	100.0

*Includes only those reporting race as white.

.

Education Level

For education level (the highest level attained as reported by a respondent), the demographic 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 data collection; therefore, education level was not considered in the analyses of Followup 1 data. The distribution of education levels attained by the registrants is shown in Table 2-5.

The demographic analyses at Baseline included comparisons in which education level was measured as a four-category ordinal variable. The results given in the Baseline report (1) showed that the TCE registrants might have had somewhat fewer years of education than those in the NHIS sample. However, by Followup 1 the percentage of remaining registrants who had 12 years or more of education increased from 69% to 72%; the comparable value for the NHIS file was 80.5%. See Table 3-3 for a summary of the education data.

Consistent with the results at Baseline, for most health outcomes the data were too sparse to support the inclusion of a four-level education term in the Poisson regression modeling. For education level to be a meaningful variable, a sample must be restricted to those who are at least 19 years of age; thus, the numbers are further reduced. Recognizing the importance of this confounding factor, the possibly significant difference between the populations and the potential of this difference to impact reporting rates, the education level factor was reduced to a dichotomous variable (yes or no high school diploma, or greater). The models were rerun including this factor. With few minor exceptions (which are noted in the discussion of the results) the inclusion of the education variable in the model did not change the results from the model with the education variable excluded.

Cigarette Smoking

The Baseline rates for current and past smoking behavior were compared with national smoking rates and were found to be very similar. A current smoker 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 was assessed by calculating the rates for people who had ever smoked at least 100 cigarettes during their lifetime. People who were classified as "ever smoked" included both current and ex-smokers categories.

The comparison of the Subregistry population cigarette smoking rates at Baseline with a national rate was used as a means to assess the general comparability of the TCE 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 at Baseline because this information was not available for each respondent in the 1989 NHIS data. The 1990 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 Followup 1 and 1990 NHIS smoking rates could not be made and smoking could not be included in the statistical models.

The distribution of the smoking rates of several types of tobacco products for the registrants at the time of Followup 1 are shown in Table 2-6. Also, Table 3-4 provides comparative information on current and past cigarette use for persons 18 years of age or older. Smoking rates are indicated for TCE Subregistry Followup 1 members, National Household Survey on Drug Abuse (NHSDA)

Table 3-3.--Education level attained for TCE Subregistry Followup 1 and National Health Interview Survey (NHIS) populations (white, 19 years of age or older).

*Percentage of group.

Note: Education data was missing or reported as "don't' know" for 15 registrants.

Table 3-4.--Reported rates of cigarette smoking for TCE Subregistry Followup 1, National Health Interview Survey (NHIS), and National Household Survey on Drug Abuse (NHSDA) populations (white only).

Age Group (Years)	TCE Followup 1 Rates (%)			NHIS Rates (%)				NHSDA
				All		Midwest Region		All		Midwest Region
	N	Current*	Ever†	Current	Ever	Current	Ever	Current	Ever	Current	Ever
18 All Males Females	2,737 1,279 1,458	36.1 39.2 33.3	58.8 65.5 52.9	25.3 27.6 23.3	51.7 59.8 44.2	26.4 29.4 23.7	50.9 60.1 42.8
18-25 All Males Females	433 208 225	34.6 33.6 35.6	48.3 45.7 50.7	26.8 27.8 25.8	38.2 38.6 37.9	27.8 28.9 26.9	37.7 38.3 37.1	35.2 35.6 34.8	74.9 77.7 72.4	45.0	84.8
26-34 All Males Females	668 309 359	47.9 51.8 44.6	62.6 64.4 61.0	29.8 31.0 28.6	48.8 51.4 46.3	30.4 32.6 28.4	48.5 52.6 44.7	37.1 40.7 33.6	80.8 83.2 78.3	45.8	82.3
35 All Males Females	1,636 762 874	31.6 35.6 28.2	60.0 71.4 50.1	23.5 26.3 21.0	55.7 67.9 45.0	24.7 28.4 21.5	54.9 68.0 43.5	27.3 32.2 23.0	79.2 88.2 71.4	23.3	78.9

*Reported currently smoking.

†Reported having smoked more than 100 cigarettes in lifetime.

(9), and select NHIS registrants (10). Data are also provided for the Midwest Region of the United States.

The current smoking rates of the registrants exceed the NHIS Midwest Region rates (36.1% versus 26.4%) and NHSDA rates for ages greater than 26 years (47.9% versus 45.8% for ages 26 through 34 years of age, 31.6% versus 23.3% for those 35 years of age or older). The ever smoking rates do not differ that greatly (for example, the registrant rate is 58.8% versus 51.7% for the NHIS subpopulation) and the NHSDA population ever rates exceed the NHIS rates in all age groups. Smoking is an important confounding factor for most diseases and needs to be considered in comparing health reporting rates whenever possible. It does not appear, however, that the smoking rates differ to the extent of invalidating the comparison of health outcome reporting rates where the factor is not incorporated in the model.

METHODS OF COMPARING HEALTH OUTCOMES

Question Comparability

The reporting rates of health conditions by TCE Subregistry Followup 1 and 1990 NHIS respondents were compared. Although the NER and NHIS data collection instrument questions were quite similar, there were differences for some health questions. 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 follows. The NHIS health-related questions and the TCE Subregistry questionnaire health questions are shown in Appendix A. It is possible that this restriction could produce an underreporting bias.

Source of Diagnosis

TCE 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 the same, an increased reporting by NHIS respondents when compared with the TCE registrants might have been expected.

Time Frame

TCE Subregistry Followup 1 questionnaire health questions asked about diagnoses of or treatment for conditions from the date of the last interview (Baseline) through the date of the Followup 1 interview. (For TCE Followup 1, the time interval from the Baseline interview to the follow-up interview was one year.) Respondents who reported "yes" to having been diagnosed or treated 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. The NHIS questionnaire asked whether respondents "ever had" the health condition for some health outcomes, "had it within the last year" for some, or "currently had the condition" for others. For other health outcomes, composites of these time frames were used (see Table 3-5). To make the NHIS and TCE Subregistry data more directly comparable, the time frames were standardized. Table 3-6 provides a comparison of NHIS and TCE Subregistry questions in terms of the time frame for each health condition.

The NHIS health condition question "the effects of a stroke" was asked in the context "have you ever had." The TCE Followup 1 questionnaire asked whether "since the last interview" the person had been told of the condition; therefore, for comparability, the number of stroke cases was calculated by merging the Followup 1 data with the Baseline data. That is, a stroke case was counted

Table 3-5.--Comparison of time frames for health condition questions.

TCE Subregistry Conversion from:	NHIS Version:
	"ever had"	"in the past 12 months"	"now have"
"ever told"	stroke
"in the past 12 months"		anemia, arthritis, asthma, cancer, diabetes, hypertension, kidney problems, liver problems, rashes, respiratory allergies, ulcers, urinary problems
"now have" ("ever told, since last interview" and "currently treated")	speech impairment, hearing impairment, mental retardation

if someone responded positively at Followup 1 or reported at Baseline that he or she had ever been told he or she had had or was treated for a stroke.

The NHIS health question for hypertension was asked in the "ever had" time frame. However, in the NHIS editing process, the positive responses for hypertension problems were discarded unless there had been a hospital visit, doctor visit, one or more bed days, loss of activity, or some indication of the condition having been present in the preceding year. Thus, the reporting period for hypertension was, in effect, "within the past 12 months" time frame for the NHIS data. The Registry file reporting rate was comparable.

The NHIS data on heart conditions was not comparable because of incompatible time frames. The NHIS rate was a composite rate of the "last 12 months" and "ever had" time frames applied to specific heart conditions (as well as using replies from nine other questions from elsewhere in the questionnaire as restrictors) that could not be duplicated for the TCE Subregistry data.

Eleven of the NHIS questions were asked in the time frame "within the past 12 months." The time frames for the Subregistry and NHIS matched on these health items.

There were three health conditions in the NHIS questionnaire that were queried in the present tense time frame "do you now have." These conditions were speech impairment, hearing impairment, and mental retardation. The response time frame for the comparable TCE Subregistry health conditions was adjusted by counting only registrants who reported that they were "currently receiving treatment" for one of these three conditions.

The wording in the NHIS questionnaire made no reference to confirmation or treatment by a health care provider, only whether the respondent believed he or she had had or now had the condition. The stipulation for the TCE registrants of having health care confirmation or receiving

Table 3-6.--Comparison of Trichloroethylene (TCE) Subregistry Followup 1 and National Health Interview (NHIS) health questions.

Q#*	Wording in TCE Survey	NHIS Definition	NHIS Chronic Recodes†	ICD-9§
Class A¶
6	High blood pressure (hypertension)	Essential hypertension Hypertensive heart disease Hypertensive renal disease Hypertensive renal and heart disease	C508	401-05
8	Kidney disease	Kidney stones Kidney infections Other kidney trouble	C409-11	592 590 581-3 593
10	The effects of stroke	Cerebrovascular disease	C509	430-38
14	Liver problems	Liver disease, including cirrhosis	C302	571-2 573.0, .3-.9
15	Asthma, emphysema, or chronic bronchitis	Same	C601-2 609	490-1 492 493
16	Other respiratory allergies or problems such as hay fever	Hay fever Allergic rhinitis without asthma	C603	477
17	Diabetes	Same	C403	250
22	Hearing impairment	Deaf - both ears Other hearing impairment	C203-4	X05 X06-9
25	Mental retardation	Same	C208	X19
Class B**
3	Cancer	Some cancers queried directly; other ascertained indirectly		140-203
5	Skin rashes, eczema, or other skin allergies	Psoriasis Dermatitis Dry (itching) skin	C112-4	696 690-94 698.9
7	Anemia or other blood disorders	Anemia of any kind	C508	401-05
9	Urinary tract disorders, including prostate trouble	Disorders of the bladder (other than bladder infections) Diseases of prostate	C413-14	594.1 596 600-602 (except 601.4)

Table 3-6.--Continued.

Q#*	Wording in TCE Survey	NHIS Definition	NHIS Chronic Recodes†	ICD-9§
Class B**
13	Ulcers, gall bladder trouble, or stomach or intestinal problems	Gallbladder stones Gastric, duodenal, or peptic ulcer Abdominal hernia Gastritis and duodenitis Disease of esophagus Other functional disorders of stomach or digestive system (not indigestion) Enteritis and colitis Spastic colon Diverticula of intestines Other stomach and intestinal disorders (not constipation)	C301 C303 C303-8 C310-3 C315	574 530-7 550-3 555 556 558 560.562 564.1 569 787
18	Arthritis, rheumatism, or other joint disorders	Arthritis Rheumatism Gout Sciatica (and lumbago) Intervertebral disc disorders Bone spur and tendinitis Disorders of bone or cartilage Bursitis	C101-7 C109	711.0, .9 712.8-.9 714-16 720.0 721 729.0 724,.2-.3 722, 726 727.0, .2-.9 730.0-.3, .9 731.0, .2 732-3
20	Speech impairment	Stammering and stuttering Other speech impairment	C205-6	X10 X11
No Match
19	Rheumatic fever, heart disease, or other heart problems	Rheumatic fever Ischemic heart disease Heart rhythm disorders Congenital heart disease Other select heart diseases	C501-7	390 392-9 410-4 427.0-.6, .8-.9 785.0-.2 745-6 415-7, 420.9 421.0, .9 422.9 423-4 425.0-.5, .9 426, 428 429.0.6, .8-.9

*Question in TCE Subregistry questionnaire.

†Chronic Recodes, NHIS, Public Use Data Tape Documentation (10).

§ICD-9 is the International Classification of Diseases, 9th Revision, World Health Organization (WHO) (11).

¶Class A indicates questions match exactly or closely.

**Class B indicates questions are similar.

treatment was more restrictive. Again, if all other factors were equal or the same, an increased reporting by the NHIS respondents when compared with the TCE registrants would have been expected.

Health Conditions

TCE Subregistry and NHIS questions were also compared in terms of the phrasing of health conditions. As Table 3-6 indicates, the wording of some health conditions matched exactly, others did not. An ATSDR panel of scientists and physicians determined matches for the TCE Subregistry health conditions and specific NHIS conditions (ICD-9 codes [11], or NHIS condition recodes). Nine health conditions in Table 3-6 are termed Class A because they either matched exactly or the TCE Subregistry version was inclusive of the NHIS version. That is, the NHIS wording of the health condition and the NHIS classification of the condition in the recodes were the same as or closely paralleled the corresponding TCE Subregistry item. Class B was comprised of seven health conditions that did not match as closely, but were considered to be sufficiently similar for the purposes of the NHIS and TCE Subregistry comparisons.

The TCE Subregistry questionnaire was used to record information on all types of cancer via an open-ended question. The NHIS questionnaire, however, was used to obtain direct information on only some types of cancers--including skin, stomach, intestinal, colon, rectal, lung, breast, and prostate cancer; information on other cancers was obtained indirectly by querying the respondent on hospital stays, doctor visits, and restricted activity. The NHIS question was worded, "In the last 12 months, did anyone in the family have .... cancer?" The time frame restriction and the possible restriction on types of cancers reported would have made this comparison with the TCE registrant data questionable and the interpretation tenuous. Reporting rates for both the TCE Followup 1 population and for the 1990 NHIS population are given in Appendix B.

Statistical Analyses of Health Data

The statistical analyses performed treated the NHIS population as a standard population and applied the age-, sex-specific period prevalence and prevalence rates obtained from the NHIS data with the corresponding age-, sex-specific denominators in the TCE Subregistry. The observed age-, sex-specific numerators for the TCE 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 TCE Subregistry data file.

This report used the condition list weights and person weights (or "final basic weight") (10) in calculating the health condition rates for the NHIS data. To allow for the nonequiprobable sampling in the NHIS data, all age-, sex-specific period prevalence and prevalence rates that were derived from the NHIS data were weighted by the appropriate person-weights. These weights reflected the complex sampling method used by the NCHS in the survey design (10). The Primary Sampling Unit (PSU) codes (10) contained on the NHIS data file allowed for partial adjustment for the clustering component of the NHIS survey design, but the adjustment was not used in the analyses for this report. The age- and sex-specific NHIS rates were weighted by the condition list weight (or "condition list number") (10) to adjust for the fact that only approximately one-sixth of the NHIS respondents were queried about each of the conditions of interest. All health outcomes were analyzed in the following manner. Taking the NHIS as a standard population, weighted age-, sex-specific period prevalence and prevalence rates were constructed using the condition list and person weights. These "standard" rates were applied to the corresponding TCE Subregistry denominators to obtain expected counts in each age and sex combination.

Paralleling Poisson regression modeling of standardized mortality ratios (SMRs), the ratios of the observed-to-expected age- and sex-specific counts were modeled using Poisson regression in the Generalized Linear Interactive Modeling program (GLIM) (12). The Poisson regression approach is described in Breslow and Day (13). Maximum likelihood estimation was used, and likelihood ratio statistics and Wald confidence intervals (where appropriate) were computed. 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 Wald confidence interval about the grand mean does not include the value of 1, provided that the null model is adequate. 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.

RESULTS OF DATA FILE HEALTH OUTCOME COMPARISONS

The Followup 1 age categories were 1 through 10 years, 11 through 18 years, 19 through 25 years, 26 through 35 years, 36 through 45 years, 46 through 55 years, 56 through 65 years, and 65 years of age and older. In order for the Followup 1 analyses to be consistent with the Baseline analyses, the age grouping of the respondents was retained; that is, the age categories used in the Baseline analyses were shifted upward by one year (for example, the 0 through 9 years of age group became the 1 through 10 years of age group), thus grouping the same TCE registrants together. A summary of the response rates for health conditions for the TCE Subregistry Followup 1, and the expected numbers from the 1990 NHIS file are provided in Table 3-7. The overall calculated observed-to-expected ratio (risk ratio or RR) and 99% confidence intervals (CIs) are also shown. A summary of the Poisson regression modeling is shown in Table 3-8.

Table 3-9 provides a summary of the results of the 1990 NHIS and TCE Followup 1 file comparisons. For each health outcome, the tables indicate the p-values for the effects of age (categorized into eight levels) and sex with the associated degrees of freedom, based on a model containing age and sex. For assessing the statistical significance of the contribution of these terms to a model, the 0.01 level of significance was used. This level was used because of the many tests of hypotheses that are implicit in this model building.

The p-values for residual deviance and the associated degrees of freedom are also given as a global lack of fit measure for each model, which specifies multiplicative effects of the age (i) and sex (j) ratios Oij/Eij. The significance level of 0.10 was used to assess the significance of the residual deviance. For each outcome, the age- and sex-specific numerators Oij was obtained from the TCE Subregistry Followup 1 data, while the expected numerators Eij were based on the suitably weighted age- and sex-specific ratios from the NHIS data. For the purpose of detecting structure in these age- and sex-specific ratios, given that the model residual deviance was not statistically significant, a significance level of 0.05 was adopted.

Table 3-7.--Observed (TCE) cases and expected (NHIS) number of health condition and risk ratios for TCE Subregistry Followup 1.

Outcome	Observed	Expected	Risk Ratio	99% CI*
Stroke†	62	35.4	1.749	1.230, 2.407
Hypertension§	310	353.8	0.876	0.753, 1.013
Diabetes§	103	75.5	1.365	1.043. 1.751
Kidney disease§	60	46.3	1.297	0.906, 1.794
Urinary tract disorders§	256	45.9	5.572	4.716, 6.534
Skin rashes§	307	242.7	1.265	1.087, 1.463
Anemia§	103	49.0	2.102	1.606, 2.696
Asthma, emphysema§	289	324.3	0.891	0.762, 1.035
Respiratory allergies§	246	361.4	0.681	0.574, 0.801
Ulcers§	249	259.4	0.960	0.810, 1.128
Liver problems§	23	10.2	2.248	1.224, 3.762
Arthritis§	293	550.3	0.532	0.456, 0.618
Mental retardation¶	7	21.7	0.322	0.094, 0.789
Speech impairment¶	15	27.9	0.538	0.247, 1.010
Hearing impairment¶	37	333.4	0.111	0.070, 0.167

*Confidence interval.

†Indicates time frame is "ever had"; observed cases are cumulative.

§Indicates time frame is "last 12 months.".

¶Indicates time frame is "now have."

Table 3-8.Summary of Poisson regression modeling for TCE Subregistry Followup 1.

Condition	Age/Sex		Sex/Age		Residual Deviance p value	df*
	p value	df	p value	df
Skin rashes	p 0.01	7	p = 0.49	1	p = 0.03	7
Arthritis	p = 0.10	7	p = 0.42	1	p = 0.07	7
Mental retardation	p = 0.79	7	p = 0.92	1	p = 0.61	7
Speech impairment	p = 0.11	7	p = 0.34	1	p = 0.78	7
Hearing impairment	p 0.01	7	p = 0.17	1	p = 0.47	7
Liver problems†	p = 0.13	7	---	1	p 0.01	7
Stomach problems	p = 0.02	7	p = 0.10	1	p = 0.11	7
Anemia and blood disorders	p 0.01	7	p 0.01	1	p 0.01	7
Diabetes	p 0.01	7	p = 0.04	1	p = 0.94	7
Kidney disease	p = 0.08	7	p = 0.20	1	p = 0.23	7
Urinary tract disorders	p 0.01	7	p 0.01	1	p 0.01	7
Hypertension	p 0.01	7	p = 0.02	1	p 0.01	7
Stroke	p = 0.05	7	p = 0.01	1	p 0.01	7
Respiratory allergies	p 0.01	7	p = 0.68	1	p = 0.02	7
Asthma, emphysema	p = 0.29	7	p = 0.93	1	p = 0.26	7

*df - Degrees of freedom.

†Some estimates could not be obtained.

Table 3-9.--Summary results of TCE Subregistry Followup 1-National Health Interview Survey comparison.

Disease Category

Age Groups (Years)

1-10

11-18

19-25

26-35

36-45

46-55

56-65

66+

All

All

M

F

M

F

M

F

M

F

M

F

M

F

M

F

M

F

M

F

Anemia

--

X

X

X

Arthritis

X

R

R

R

R

R

R

R

R

R

R

Diabetes

--

X

X

X

<>>

Hearing impariment

R

R

R

R

R

R

R

R

R

R

R

R

R

R

Hypertension

--

R

R

Liver problems

--

R

R

Mental retardation

R

Skin rashes

X

X

X

Respiratory allergies

X

R

R

R

Urinary tract disorders

--

X

X

X

X

X

X

X

X

X

X

X

X

X = Statistically significant differences, TCE Subregistry rate higher.

R = Statistically significant differences, NHIS rate higher.

-- = Insufficient data.

As is shown in Table 3-9, in the TCE followup data the model was adequate for mental retardation, speech impairment, hearing impairment, stomach problems, diabetes, kidney disease, and asthma and emphysema. The model indicated borderline lack of fit for arthritis (p = 0.07); however, because the lack of fit was not severe, the overall interval estimates are also given. (When the model does not give an adequate fit, the comparisons should be made of the age- and sex-specific ratios.) Skin rashes and respiratory allergies also exhibited lack of fit (p = 0.03 and 0.02, respectively); because the lack of fit was not extreme, the interval estimates are also provided for the statistically significant predictor, which was age. Age was a statistically significant indicator for anemia and blood disorders, urinary tract disorders, and hypertension; because these models had severe lack of fit, the age-, sex-specific interval estimates are provided. (Sex was a statistically significant indicator for anemia and blood disorders and urinary tract disorders.) Finally, the models for both liver problems and stroke exhibited severe lack of fit; thus, for these two conditions it was again more appropriate to consider the age-, sex-specific interval estimates.

The 0.01 significance level was used for all comparison tests. The results of the comparison of the TCE Subregistry Followup 1 data to the NHIS data are summarized in Table 3-9; a discussion of the results follows.

Anemia or Other Blood Disorders

A summary of reporting rates by age- and sex-specific groups for anemia or other blood disorders is shown in Appendix B-1; risk ratios are given in detail in Appendix C-1. Both age and sex were statistically significant in the Poisson regression, however, the modeling did not give an adequate fit; the residual deviation was highly statistically significant. Thus it was more appropriate to consider the age- and sex-specific observed-to-expected or risk ratios. Elevated risk ratios were observed in each age- and sex-specific category; however, only three were statistically significant-- males over 65 years of age (RR = 4.79; 99% CI= 1.03,13.54; O = 5, E = 1.04); females 19 through 25 years of age (RR = 4.53; 99% CI = 2.39,7.75; O = 21, E = 4.64); and females 56 through 65 years of age (RR = 3.78; 99% CI = 1.41,8.09; O = 10, E = 2.64). In two categories, males 19 through 25 and 46 through 55 years of age, risk ratios could not be estimated; there were no occurrences of anemia among the NHIS respondents in these age groups, resulting in a value of 0 for the expected number of occurrences. The inclusion of the education variable in the model did not alter the statistically significant results.

The same modeling results were obtained at the Baseline analysis: the model did not give an adequate fit, and age and sex were statistically significant predictors. At Baseline, the risk ratios were statistically significantly greater than 1 for several age-, sex-groups, two of which matched the Followup 1 data--males 65 years of age or older and females 19 through 25 years of age. The other group statistically significantly elevated at Followup 1, females 56 through 65 years of age, was not elevated at Baseline.

Arthritis, Rheumatism, or Other Joint Disorders

Neither age nor sex was statistically significant as a predictor (p = 0.10 and 0.42, respectively); the overall risk ratio was considered since the model did not show excessive lack of fit. Age- and sex-specific group summary data are shown in Appendix B-1. Except for the risk ratio for males 11 through 18 years of age, all other risk ratios were less than 1. Statistically significantly lower reporting rates were observed among TCE Followup 1 males 26 years of age or older and among females 26 through 35 years of age and females 46 years of age or older. Males 11 through 18 years of age showed a statistically significantly elevated risk ratio (RR = 4.10; 99% CI = 1.20,10.05; O = 7, E = 1.71). Inclusion of the education variable did not alter the results.

It is of interest that this was one of two age-, sex-groups for which the observed-to-expected ratio exceeded 1 (however, not statistically significantly) at Baseline. The age-, sex-specific risk ratios are given in detail in Appendix C-2. In the Baseline analysis, the model fit was adequate (the residual deviance was not statistically significant), and sex was a statistically significant predictor.

Asthma, Emphysema, or Chronic Bronchitis

Because neither age nor sex was a statistically significant predictor (p = 0.29 and 0.93, respectively) and the model fit adequately (residual deviance was not statistically significant), the appropriate summary measure for this condition was the overall ratio of 0.89. There was no indication of any difference in risk (99% CI = 0.76,1.04); however, the calculated risk ratio was close to being statistically significantly less than 1 (a result consistent with Baseline results). The age- and sex-specific risk ratios for asthma ranged from 0.54 through 1.36 (none statistically significantly different from 1). The addition of the education variable to the model did not alter the significant results. The age-, sex-specific risk ratios are given in detail in Appendix C-3; a summary of the rates is in Appendix B-2.

Diabetes

The model fit very well for diabetes (residual deviance p = 0.94), with age and sex both being statistically significant predictors. Since age was the more highly statistically significant predictor, and sex was borderline statistically significant (using the 0.05 level), age-specific risk ratios for diabetes are included in Appendix C-4. (A summary of subgroup rates is given in Appendix B-3.) In the two youngest age groups, the risk ratio was 0; there were no observed cases of diabetes for registrants under 19 years of age in TCE Followup 1. Of the sex-, age categories, three categories exhibited statistically significantly increased risk ratios: females 19 through 25 years of age (RR = 4.90; 99% CI = 1.25,12.78; O = 6, E = 1.22); 46 through 55 years of age (RR = 2.49; 99% CI = 1.21,4.51; O = 17, E = 6.82); and 56 through 65 years of age (RR = 2.23; 99% CI = 1.13,3.92; O = 19, E = 8.5).

This was similar to Baseline results in which the female age groups 18 through 24 years of age and 45 through 54 years of age had statistically significantly higher reporting rates. Age-, sex-specific risk ratios are also given in Appendix C-4. The same three age categories show elevated ratios for females (there are no statistically significant effects among males). When the variable education was introduced, the statitical significance at ages 46 through 55 years was no longer found.

Age was also a statistically significant predictor at Baseline analysis. Similarly, at Baseline several of the risk ratios were statistically significantly increased. At Baseline, the model fit was not adequate, so risk ratios were reported by age-, sex-categories. The females 19 through 25 years of age and 46 through 55 years of age reported statistically significantly higher at both Baseline and Followup 1; the 55 through 65 years of age group was statistically significantly higher at Followup 1 only.

Hearing Impairment

The model was adequate for hearing impairment and age was a statistically significant predictor. Age-specific risk ratios for hearing impairment are given in Appendix C-5 (summary results are shown in Appendix B-3). The time frame of the reporting period for this variable was "currently have." All but the youngest age category, 1 through 10 years of age, showed a statistically significant reduction in the rate of reported hearing impairment. These results were completely consistent with the results obtained at Baseline: the model fit adequately, age was a statistically significant predictor, and the risk ratios were statistically significantly less than 1 for all but the youngest age category. The youngest age group reporting was statistically significantly increased at Baseline.

High Blood Pressure (Hypertension)

The age-, sex-specific risk ratios for hypertension are given in Appendix C-6; summary rates are shown in Appendix B-4. The model did not provide an adequate fit, thus it was most appropriate to consider the age- and sex-specific risk ratios. Of particular note are the elevated risk ratios found for the two youngest age categories (9 observed versus 1.95 expected); the sex-, age-specific ratios, however, did not differ statistically significantly from 1. A risk ratio could not be calculated for females 1 through 10 years of age because there were no cases reported in the 1990 NHIS data. Two age categories had statistically significant underreporting among the TCE registrants--males 36 through 45 years of age (RR = 0.53; 99% CI = 0.26,0.95; O = 18, E = 33.71) and males 46 through 55 years of age (RR = 0.49; 99% CI = 0.24,0.88; O = 17, E = 34.95). When education was entered in the model the results were not changed.

Neither age nor sex was a statistically significant predictor in the Baseline data, and the model did not fit adequately. No significant differences were found.

Kidney Disease

Since neither age nor sex was a statistically significant predictor (p = 0.08 and 0.20, respectively), and since the model gave an adequate fit, the best measure of reporting rate comparisons was the overall risk ratio (RR = 1.30; 99%CI = 0.91,1.79; O =60, E = 46.27). Age-, sex-specific risk ratios for kidney disease, given in Appendix C-7, ranged from 8.16 (for the youngest males) through 0 (for males 11 through 18 years of age); none of the risk ratios for the age-, sex-categories were statistically significantly different from 1. The sparsity of the data percluded including teh education variable in the model.

Followup 1 and Baseline analyses were consistent in that at Baseline all confidence intervals (except for females 55 through 64 years of age) included the value 1, indicating no statistical increase in reporting rates; however, rates for females 55 through 64 years of age were highly elevated at Followup 1. At Baseline analysis, sex was a statistically significant predictor while the model fit was not adequate (residual deviance was statistically significant). Of particular note was the reporting of 5 cases in the 1 through 10 years of age group versus the 1.5 cases expected based on 1990 NHIS data (there was 1 reported at Baseline versus 0.61 expected).

Liver Problems

The model was not an adequate fit for liver problems data; thus, the risk ratios were considered in the age- and sex-specific categories (see Appendix C-8). For persons less than 26 years of age very few cases were expected or observed. Two categories showed statistically significant risk ratios--males 26 through 35 years of age (RR = 6.49; 99% CI = 1.40, 18.37; O = 5, E = 0.77); and females 46 through 55 years of age (RR = 6.64; 99% CI = 1.12, 20.91; O = 4, E = 0.60). The age-, sex-specific summary rates are given in detail in Appendix B-5.

The number of cases of liver problems was quite small (precluding the inclusion of education variable in the model) both at Baseline and at Followup 1. What was noteworthy, however, was the increase in the total number of cases--from 13 to 23; the increase was particularly of note for males 19 through 45 years of age--from 0 cases reported at Baseline to 11 cases reported at Followup 1.

Mental Retardation

There were only seven reported cases of mental retardation in the TCE Subregistry Followup 1 (this condition was asked in the "currently have" time frame). Neither age nor sex was a statistically significant predictor (p = 0.79 and 0.92, respectively); the model did provide an adequate fit. Thus, the appropriate measure of increased reporting was the overall risk ratio, which was 0.32 (99% CI = 0.09,0.79; O = 7, E = 21.72). The age-, sex-specific risk ratios are given in detail in Appendix C-9; summary groups rates are in Appendix B-5.

At Baseline analysis the model could not be fit; there was a lack of convergence because of sparse data. The risk ratios at Followup 1 generally tended to be less than 1; this was the case at Baseline also.

Skin Rashes, Eczema, or Other Skin Allergies

The model for skin rashes and other skin allergies did not give an adequate fit (residual deviance p = 0.03). Thus, the risk ratios were evaluated for the age-, sex-specific categories. These are given in Appendix C-10; summary rates are shown in Appendix B-6. Three age-, sex-specific risk ratios were statistically significant--males 11 through 18 years of age (RR = 2.23; 99% CI = 1.08,4.04; O = 17, E = 7.63); males over 65 years of age (RR = 3.29; 99% CI = 1.77,5.45; O = 15, E = 9.79); and females 19 through 25 years of age (RR = 2.16; 99% CI = 1.20,3.58; O = 24, E = 11.10). Inclusion of the education variable did not alter the results.

The model did provide an adequate fit for the Baseline data (residual deviance was not statistically significant). Neither age nor sex was a statistically significant predictor at Baseline (p = 0.14 and 0.44, respectively). The overall risk ratio at Baseline was 1.28 (99% CI = 1.11,1.48). The age-, sex-specific risk ratios did tend to be slightly greater than the value 1 in the Baseline results, similar to the results observed at Followup 1.

Other Respiratory Allergies or Problems, Such as Hay Fever

The risk ratios for other respiratory allergies or problems, such as hay fever need to be looked at in the age- and sex-specific categories; the model did not give an adequate fit (residual deviance p = 0.02 ). The risk ratios, given in Appendix C-11, were less than 1.0 in all but three age-, sex-categories. The risks were statistically significantly decreased for males 26 through 35 years of age (RR = 0.34; 99% CI = 0.16,0.63; O = 15, E = 44.61) and for males 46 through 55 years of age (RR = 0.29; 99% CI = 0.06,0.83; O = 5, E = 17.01), as well as for females 36 through 45 years of age (RR = 0.62; 99% CI = 0.36,0.99; O = 28, E = 45.08). In contrast, for the youngest group of males the risk ratio (RR = 2.44; 99% CI = 1.15,4.49; O = 16, E = 6.57) was statistically significantly greater than 1. The inclusion of the education variable did not change the results.

Results similar to those at Followup 1 were observed at Baseline--the risk ratios were less than 1. At Baseline, the overall risk ratios both for males and for females were statistically significantly less than 1 (99% CI = 0.37,0.64 for males and 99% CI = 0.56,0.86 for females).

Speech Impairment

Neither age nor sex was a statistically significant predictor for speech impairment (p = 0.11 and p = 0.34, respectively). Furthermore, the model fit was adequate, with the residual deviance not statistically significant. Thus, the overall risk ratio (RR = 0.54; 99% CI = 0.247,1.010; O = 15, E = 27.89) was the most appropriate measure. There were fewer reports of speech impairment in Followup 1 (15 positive reports versus 36 at Baseline). The age-, sex-specific risk ratios are given in detail in Appendix C-12; summary rates are shown in Appendix B-7 (this question was asked in the "currently have" time frame).

Results similar to Followup 1 were observed at Baseline--the highest age-specific risk ratios were found for the youngest age category. However, at Baseline, unlike Followup 1, the observed to expected ratio was statistically greater than 1.

The Effects of a Stroke

The residuals in the modeling of the stroke data were highly statistically significant, indicating that the model did not provide a good fit. Thus, the risk ratios were evaluated for age-, sex-specific categories (Appendix C-13). There were a total of 62 cases reported by the TCE registrants, 30 of these were among registrants who were older than 65 years of age. Summary rate data are shown in Appendix B-3.

The health condition stroke was asked in the "ever" time frame; therefore, if no registrants were lost to followup the number of cases should have remained constant or increased. The number of positive reports decreased by 3 (1 in the Baseline group 10 through 17 years and 2 in the Baseline group 65 years of age and older).

The absence of reported strokes in the younger groups and the presence of reported strokes in the older groups also tended to hold in the NHIS data. There were four age-, sex-groups for which the risk ratio was undefined; there were no cases of stroke for females 11 through 35 years of age and the youngest age male group for the NHIS data.

In the Baseline analysis, the model fit was adequate (residual deviance was not statistically significant, p = 0.36); age was statistically significant (p = 0.04). Just as in Followup 1, the cases in the Baseline were very sparse in all but the oldest age categories. At Baseline, the risk ratios were statistically significantly elevated for the age categories 34 through 44 years, 45 through 54 years, and for those 65 years of age and older. Although not statistically significant at the 0.01 level, if the Followup 1 data were collapsed across sexes, the Followup 1 risk ratios for age groups 25 through 34 and 35 through 44 exceeded the Baseline risk ratios.

Ulcers, Gallbladder Trouble, or Stomach or Intestinal Problems

The model for ulcers, gallbladder trouble, or stomach or intestinal problems was adequate (residual deviance p = 0.11). Neither age nor sex was a statistically significant predictor. The overall risk ratio was not statistically significant (RR = 0.96, 99% CI = 0.81,1.18, O = 249, E = 259.45). The age-, sex-specific risk ratios are given in detail in Appendix C-14. The inclusion of the education variable did not alter the results.

Similarly, the model gave an adequate fit in the Baseline analysis; neither age nor sex was statistically significant as a predictor (p = 0.10 and 0.36, respectively) and none of the sex-, age-specific risk ratios at Baseline were statistically significantly greater than 1.

Urinary Tract Disorders, Including Prostate Trouble

The model did not give an adequate fit, therefore, the risk ratios for the age-, sex-specific categories are given in Appendix C-15. The risk ratios were statistically significantly elevated for all age groups for females--1 through 10 years ( RR = 17.6; 99% CI = 7.24,35.4; O = 12, E = 0.68); 11 through 18 years (RR = 10.2; 99% CI = 4.54,19.6; O = 14, E = 1.37); 19 through 25 years (RR = 23.5; 99% CI = 14.7,35.4; O = 37, E = 1.58); 26 through 35 years (RR = 13.1; 99% CI = 8.18,19.9; O = 36, E = 2.74); 36 through 45 years (RR = 19.0; 99% CI = 12.0,28.5; O = 38, E = 1.99); 46 through 55 years (RR = 8.07; 99% CI = 4.32,13.6; O = 22, E = 2.73); 56 through 65 years (RR = 5.19; 99% CI = 2.46,9.57; O = 16, E = 3.08); and 66 years of age and older (RR = 2.63; 99% CI = 1.33,4.62; O = 19, E = 7.23).

The risk ratios were also statistically significantly elevated for males 11 through 18 years of age (RR = 7.11; 99% CI = 1.20,22.4; O = 4, E = 0.56); 19 through 25 years of age (RR = 14.0; 99% CI = 1.57,51.2; O = 3, E = 0.21); 26 through 35 years of age (RR = 7.15; 99% CI = 3.07,14.0; O = 13, E = 1.82); and 56 through 65 years of age (RR = 3.51; 99% CI = 1.50,6.88; O = 13, E = 3.71). When the education variable was included, the above results were consistent across the education levels.

At Baseline, the model did not give an adequate fit (residual deviance was statistically significant), just as at Followup 1. The risk ratios at Baseline were generally elevated, also as at Followup 1. At Baseline, there were 145 reported cases; at Followup 1, there were 256. The increase might have been due to the difference in calculating the rates; an indirect method was used at Baseline (1) while Followup 1 was a direct 12-month rate. The reporting rates increased for all age groups, except those less than 10 years of age at Baseline.

Cancers

Appendix B-2 shows the results for the "all cancer" outcome using the NHIS all cancer 12-month period prevalence rates that included both first and second cancers. The number of specific cancers is shown in Table 3-10. The comparison with NHIS data using age-, sex-specific groups (Appendix C-16) showed that the age group 19 through 25 years had an excess reporting rate (8 versus 0.77 expected; RR = 10.363; 99% CI = 3.33,24.07). These eight cancers were all in the female group (2 skin, 1 each cervical and uterine, and 2 vaginal cancers). In interpreting the results, it should be noted that at Baseline the 1989 NHIS data file rates predicted that this female age group would have 2.8 cancers versus the 0.48 predicted using the 1990 NHIS data file; however, even at this higher rate, the Followup 1 reporting rate was elevated.

Table 3-10.--Summary of TCE Subregistry Followup 1 response rates for cancer (time frame is "last 12 months").

Cancer	Sex				Total
	Male		Female
	N	%	N	%	N	%
None	1,622	98.8	1,795	98.1	3417	98.4
Lip, oral, pharynx	1	0.1	0	0.0	1	<0.1
Digestive system	3	0.2	4	0.2	7	0.2
Respiratory system	3	0.2	1	<0.1	4	0.1
Unspecified site	0	0.0	1	1.0	1	<0.1
Breast	1	0.1	7	0.4	8	0.2
Genital organs	4	0.2	13	0.7	17	0.5
Urinary organs	1	0.1	1	<0.1	2	0.1
Lymphatic tissues	1	0.1	0	0.0	1	<0.1
Leukemia	1	0.1	0	0.0	1	<0.1
Other	4	0.2	8	0.4	12	0.3
Total	1,641	100.0	1,830	100.0	3,471	100.0