Hormone replacement therapy and major risk factors for reproductive cancers, osteoporosis, and cardiovascular diseases: Evidence of confounding by exposure characteristics

] Clin Epidemiol Vol. 50, No. 5, pp. 611-618, Copyright 0 1997 Elsevier Science Inc.

0895-4356/97/$17.00 PII SO895.4356(97)00004-8

1997

ELSEVIER

Hormone Replacement Factors for Reproductive Cardiovascular Diseases: by Exposure Ingemur

Persson,

*DEPARTMENT

* Leif

OF CANCER

Bergkvist,

Therapy and Major Risk Cancers, Osteoporosis, and Evidence of Confounding Characteristics Christina

EPIDEMIOLOGY,

UNIVERSITY

Lindgren, HOSPITAL,

and Jonathan S-751 85

UPPSALA,

Yuen SWEDEN

ABSTRACT.

Observational studies have yielded reports on long-term effects of hormone replacement therapy (HRT) for cardiovascular, osteoporosis-related, and cancer diseases. There is concern that risk estimates may be confounded by complex mechanisms of selection with regard to important risk determinants. In this study, we tested the hypothesis that baseline characteristics of women vary with exposure characteristics, i.e., the choices of complying with prescriptions, using different compounds and regimens, and continuing intake long-term. We analyzed the prevalence of relevant risk factors and their relationships to characteristics of exposure among 11,231 Swedish women who had received prescriptions for HRT. Associations were studied through logistic regression, with comparisons of women with ever-use versus non-compliance, long-term (73+ months) versus short-term use (l-72 months), intake of conjugated estrogens versus estradiol compounds, and intake of estrogens only versus estrogens combined with progestins, respectively. We found that women denying intake or using HRT short-term had higher parity, earlier age at first birth and a lower prevalence of hysterectomy or oophorectomy than those complying or exposed long-term. A high level of education was associated with compliance and long-term exposure, and heavy physical exercise and high intake of food fibers were associated with compliance. Climacteric symptoms were associated with compliance, long-term intake and use of conjugated estrogens, whereas a history of oral contraceptive intake was associated with use of estrogens alone without progestins. We conclude that selection biases in studies of HRT effects are important and complex in that they may vary with the reported exposure. Our findings are important, as they point to the need for improved methods for measuring, in particular, factors linked to lifestyle and health behavior, in order to account more fully for confounding in the analyses of risk relationships. J CLIN EPIDEMIOL 50;5:611-618, 1997. 0 1997 Elsevier Science Inc.

KEY WORDS.

HRT,

risk

factors,

bias, confounding

INTRODUCTION in the epidemiological literature on relations between hormone replacement therapy (HRT) and diseases of major importance for women’s health. Treatment with estrogens has been claimed to be associated with 40-50% protection from coronary heart disease (CHD) (11 and a 20-30% reduction in the risk of hip fractures [2]. Conversely, it has been linked with a lo-fold increase in the risk of endometrial cancer after about 10 years of estrogen-only intake [3], and a moderate increase in the risk of breast cancer after lo-15 years of treatment There

have

been

extensive

reports

‘Address for correspondence: Ingemar Persson, Department of Cancer demiology, University Hospital, S-751 85 Uppsala, Sweden. Accepted for publication on 15 January 1997.

Epi-

with

either

[4,5].

These

estrogens

or estrogen-progestin

risk estimations

have

been

applied

combinations in both

qual-

itative and quantitative benefit-risk assessments, according to which the benefits seem to outweigh the risks [6]. There is concern that risk estimates from observational studies on HRT-disease relationships may reflect non-causal associations [7,8]. Women using HRT are likely to be at different baseline risk compared with non-exposed women of similar age, and in epidemiological studies it may not be possible to measure or adequately adjust for all relevant risk determinants, leading to residual confounding [8]. Confounding may arise from differences between exposed and non-exposed women with regard to awareness of and access to medical facilities; conditions that are indications or contraindications for treatment; the decision to comply

612

I. Person et al.

with the hormone prescription; and the choice of continuing treatment long-term [9]. There is empirical support for the idea that women with HRT differ in various respects from non-exposed women; e.g., that the former group have a higher socioeconomic status or level of education, a lower body mass index (BMI), a lower prevalence of smoking, a higher consumption of alcohol, more screening examinations and health checkups, and higher rates of hysterectomies and oophorectomies [10,11]. It is under debate whether the reported substantial protective effect of HRT on the risk of CHD is inflated by the fact that exposed women have a better cardiovascular risk factor profile at baseline [lo] or more health-oriented behavior [9,11-151. Also, it is argued that the modest increase in the risk of breast cancer associated with HRT may be confounded by the indication for HRT, i.e., climacteric symptoms [8]. Randomized trials are therefore deemed necessary to verify and to accurately quantify HRT risk relationships [8,12]. The availability of data on HRT exposure and numerous risk factors in a large population-based cohort of women in Sweden prescribed HRT has given us the opportunity to explore risk profiles for reproductive cancers, osteoporotic fractures, and CHD. The aim of the present study was to test the hypothesis that there are differences in the potential for confounding depending on compliance, the choice of compounds and regimens, and the duration of treatment.

MATERIAL Study Base

AND METHODS

The data were collected from a population-based cohort of all women in six counties in the central part of Sweden who were prescribed replacement hormones during the threeyear period 1977-1980. Data from their pharmacy records were compiled, resulting in a cohort of 23,246 women [16]. This cohort has been followed up for the incidence of various cancers, notably endometrial [17] and breast cancers [ 181, and of osteoporotic hip fractures [ 191 and of myocardial infarction [20]. In 1987-1988 we conducted a questionnaire survey of all women in a large subset of the cohort to obtain detailed data on their lifetime hormonal exposure and on risk factors to be used in further studies of the cohort. This subcohort was selected from the whole cohort by the following criteria: the women had been born in 1918 or later, i.e., at the time of questionnaire they were 70 years of age or younger; they should have had one or several prescriptions for a mediumpotency estrogen compound, i.e., an estradiol compound or conjugated estrogens (excluding other or weak estriol estrogens); and lastly, women born on days 05 and 08 of the month were excluded, as they had previously been subjected to questionnaire surveys. The rationale of these criteria was that resources should be focused on those women who had

been exposed to estradiol compounds or conjugated gens at any time and on those younger women who be likely to respond well to the questionnaire. Out 23,246 women in the original cohort, 13,925 (60%) eligible for the study and alive at the time of issuing questionnaire in 1987 and 1988.

estrowould of all were of the

The Questionnaire The questionnaire was sent to the women by mail during the period from the autumn of 1987 through the end of 1988. The form contained 50 questions regarding some major risk factors for cancer, osteoporosis, and cardiovascular diseases (except alcohol consumption) (see items in tables). The main part of the questionnaire requested information on previous use of replacement hormones (both estrogens and combined progestins), indications for this treatment, and use of oral contraceptives (OCs). For HRT use, the women were asked to specify the name of the oral, transdermal, or injectable estrogen or progestin, the dose, the date of the start of treatment, the date of termination (unless current), and the type of schedule (cyclic or continuous). Recall of the brand name was facilitated by displaying colored photographs of the packages. The questionnaire was answered by 11,23 1 (81%) of the eligible women. In about one-fourth of all forms, one or several items were incomplete. Telephone calls were made to these women in order to obtain the missing information. When the timing of the treatment episodes could not be recalled, the information was coded as missing, but if the year but not the month was given, the woman was considered to have been exposed for 6 months of that year. Out of all women who reported treatment episodes, information on dates of the start and end of treatment was missing in 2.1%. The database thus included information on the above selected women from the population-based cohort.

Analyses The prevalence of risk factors was compared between groups of women within the cohort who had different types and levels of HRT exposures. Thus, these analyses concern only levels or gradients of exposure among women who had once received a prescription for HRT, i.e., those who adhered versus those who did not, long-term takers versus short-term takers, and so on. Analyses were made through logistic regression, using the HRT exposure group-the group with the expected greatest effect on diseases of interest versus that with the smallest effect-as the dependent variable and the presence of a specified risk factor as the independent variable. The association between the risk factor and exposure characteristics is given as the odds ratio (OR), together with 95% confidence intervals (CI). Exposure groups were defined in four ways (Table 1 ), each meant to be of relevance for assessing the previously

HRT and Major Risk Factors: Evidence of Confounding

TABLE to the

1. Number questionnaire,

(and %) of the 11,231 women classified into four exposure

Compliance” Ever

Never

9,547

1,684

(85%) “Women

(15%)1 who

ever

613

by Exposure

had

in the selected categories

Duration of intake of medium potency estrogens (months)’

Type of estrogens (73+ months)’

73+

l-72

CE

2,195 (23.5)8

2,610 (27.9)

a prescriprion

for an estrogen

recorded.

cohort

who

Combination with progestins (73+ months)’ E2

Ed

E+P

487

1,429

2,027

(5.2)

(15.3)

(21.7)

“Never”

responded

means

rhar

uxake

was denied

954 (10.2) in the

questionnaire. “‘Medium potency” refers to natural estradiol compounds (in dosages l-2 mg) or conjugated estrogens (0.625 or 1.25 mg): women taking only these types of compounds, without added progestins. “‘CT denotes conjugated estrogens, and “E2” escradiol: women taking only one or the other compound. “Number of estrogen only takers is smaller rhan the number in the “73+, medium potency group,” because of missmg data necessary for the classification. ‘Medium potency estrogens combined cyclically (or continuously) with a progestin, for more than 90% of the duration of treatment. Percentages of all women in the selected cohort. ‘Percentages of all rhose reporting intake.

reported biological effects of HRT on cancer and cardiovascular diseases: (1) ever-use versus non-compliance; (2) long-term use (73-t months) versus short-term use (l-72 months) of estradiol compounds or conjugated estrogens (medium-potency estrogens, as distinguished from high-potency synthetic estrogens used in OCs or from low-potency estrogens used for treatment of urogenital atrophy); (3) use of conjugated estrogens versus estradiol compounds for 73 + months (the rationale of this comparison being claims that estradiol exposure is more unfavorable regarding breast cancer risk than conjugated estrogens [ 181); and (4) use of estrogens alone (estradiol and/or conjugated estrogens) versus estrogens taken cyclically or continuously, combined with progestins (for 90% or longer of the duration of the estrogen intake). Risk factors for diseases were categorized as shown in Table 2, usually using the “no” or lowest value category as the reference. The BMI and fat index variables were grouped according to the quartile distributions of all subjects. To estimate exposures to fat- and fiber-rich foods, data were summarized into an index number to reflect exposure to that particular item. The basic strategy of the analyses was to measure the association of HRT with a set of risk factors derived from the questionnaires, one at a time. Confounding of such an association was conceivable. For example, when studying the relation of treatment compliance to the factors hypertension and BMI, parity and smoking were deemed to be possible confounders and included in a multivariate model; when looking at age at first birth, the level of education and parity were incorporated, etc. For each of these listed risk factors individual multivariate models were tested, with risk factors selected on the basis of biological plausibility. The odds ratio estimates, however, remained virtually unchanged in all

these extended models (data not shown), and for the sake of simplicity we chose to give the results only from bivariate, age-adjusted analyses. RESULTS Table 1 shows the numbers of cohort women who were entered into the four sets of analyses with regard to exposure characteristics. Among all responders to the questionnaire, 9547 women (85%) reported that they had ever taken the prescribed replacement hormones (users). Women reporting use of estradiol compounds alone or conjugated estrogens (medium-potency estrogens) alone, without progestins, comprised about half of ail compliers; 24% were users for more than six years. Users of conjugated estrogens alone, for 6 or more years, were fewer than those taking estradiol alone (5% versus 15% of all compliers). The proportion of women using a progestin-combined regimen for six or more years was about half that of women using medium potency estrogens only (10% versus 22%). Ever-Use

Versus

Non-Compliance

Women who ever took their prescribed hormones differed regarding the prevalence of a number of factors from those not reporting intake of the prescribed drug (Table 2). A history of diabetes, though infrequent, was less likely among ever-takers than among non-compliers. Of the studied reproductive factors, parity of one or more and birth of first child at an age below 20 were less likely in users than in non-compliers, and a history of hysterectomy or bilateral oophorectomy was twice as probable in users. The socioeconomic status seemed to be higher among users, as evidenced by their twice as high OR of having vocational or university

614

I. Persson

TABLE 2. Comparison ish women prescribed

of risk factors replacement

between hormones Number

Factor

Ever

Hypertensionh NO Yes Diabete? NO Yes Parity 0

i-4 5+

Age at first birth 15-19 20-24 25-29 30+ Hysterectomy NO Pre-MP Post-MP Oophorectomy (both No Pre-MP Post-Ml’ Use of OCs Never Ever <1 yr l-4 yrs 4+ yrs Smoking Never Current Previous Education’ Elementary school High school Vocational Unwersity Physical exercise’ None Light Moderate Heavy BMI Ql n‘\7

;i; Q4 Vasomotor No

Yes

users

ever

users

of women’ Non-compliers

and

non+compliers

among

Swede

OR for having the factor, if ever user vs. non-complier (OR, 95% CI)

6,893 868

1,189 167

Ref 1.0 (0.9-1.2)

8,893 50

1,502 23

:;f(0.2-0.7)

1,152 1,862 5,864 357

157 343 1,066 108

Ref > 0.8 (0.6-0.9) 0.8 (0.6-0.9) 0.5 (0.4-0.6)

1,035 3,665 2,352 1,020

251 697 400 155

Ref 1.4 (1.2-1.7) 1.7 (1.4-2.0) 2.0 (1.6-2.5)

7,154 1,747 229

1,352 223 22

Ref 1.3 (0.1-1.6) 2.2 (1.4-3.4)

7,782 1,022 262

1,470 97 24

Ref 1.7 (1.4-2.2) 2.2 (1.5-3.4)

6,394 2,729 904 712 1,034

1,190 412 132 109 135

Ref 1.0 1.0 1.0 1.2

5,047 2,183 1,935

872 467 301

Ref 0.7 (0.6-0.8) 1.0 (0.9-1.2)

7,269 489 829 680

1,487 49 ::

Ref 1.6 (1.2-2.2) 2.2 (1.7-2.8) 2.3 (1.7-3.1)

1,272 2,327 2,586 2,993

307 430 466 462

Ref 1.4 (1.1-1.6) 1.5 (1.3-1.8) 1.6 (1.4-1.9)

2,743 2,435 2,199 1,862

397 385

Ref 1.0 (0.9-1.2) 1.0 (0.8-1.1) 0.7 (0.6-0.8)

ovaries)

(0.9-1.1) (0.8-1.2) (0.8-1.2) (1.0-1.4)

symptoms” 1,927 6,052

Ref 2.2 (1.7-2.8)

Fat index’

z-i

Fiber

2:

2,181 2,346 2,367 2,299

468 410 377 404

Ref 1.2 (1.0-1.4) 1.3 (1.1-1.5) 1.2 (1.0-1.4)

50 309 2,260 5,370 1,145

13 103 476 886 162

0.9 0.7 Ref 1.2 1.4

index’

(0.5-1.7) io.5-0.8j (1.1-1.4) (1.1-1.6)

Abbreviations: Cl = confidence interval, MP = medium potency hormones, OCs = oral contraceptives, OR = odds ratio, (21-44 = quartile range. Distributions of women and ORs from logistic regression analyses, and 95% Cls. All examined risk factors shown. “Numbers may not add up equally because of missing information in some instances. *Reported onset before the year when HRT was started. ‘Highest level reported. “Reported as one of several possible reasons for startmg treatment. ‘Index calculated from several food items, categories with increasing use of fibers.

et al.

HRT and Major Risk Factors: Evidence of Confounding

by Exposure

education, which was noted for some 16% of the ever-users. Further, health-oriented lifestyle among users was indicated by their lower OR for current versus never smoking and for obesity (highest quartile of BMI versus the lowest), and their higher OR for moderate to heavy versus no physical exercise and for a high fiber intake (highest versus third category for users compared with non-compliers). As expected, women choosing to use hormones were twice as likely to report climacteric symptoms (noted in 76% of the exposed women) as those reporting no intake.

Long-

Versus

Short-Term

Use

Women who had taken medium-potency hormones (estradiol compounds and conjugated estrogens) for 6 years or longer also proved to have a different pattern of some risk factors, when compared with those exposed for a shorter time (Table 3). Long-term users showed a lower probability of having hypertension, of being multiparous (2 or more children), and of having a first birth at the age of 20 or younger; they were two to three times more likely to have had a pre- or postmenopausal hysterectomy or oophorectomy (performed in 34% and 22% of the long-term users, respectively), and were slightly more likely to have used OCs for four years or longer. A high level of education (vocational and university) was significantly associated with long-term use, whereas obesity was less common in longterm users. Further, extended use was associated with a 60% higher odds ratio for climacteric symptoms. With regard to smoking, physical exercise, fat, and fiber intake, no significant associations were noted.

Conjugated

Estrogens

Versus

Estradiol

Compounds

Among women using these two medium-potency compounds for a long period (more than six years) users of conjugated estrogens showed a 60% higher odds of suffering from climacteric symptoms and a two to three times higher odds of having undergone hysterectomy or oophorectomy, compared with women using estradiol compounds (Table 4). Further, there were indications that the users of conjugated estrogens were less likely to have used OCs for many years and less likely to have a high level of education.

Estrogens Alone Estrogen-Progestin

Versus Combinations

Compared with women taking combinations with progestins, women treated for six or more years with medium-potency estrogens alone showed a 5-10 times higher odds of having had a hysterectomy or oophorectomy (noted in 49% and 31% of the estrogen-only users, respectively), the odds depending on the timing of the operation in relation to menopause (Table 5). Further, estrogen-only users less often reported use of oral contraceptives, with a 50% lower OR

615

TABLE 3. Comparison of risk factors between long-term (73+ months) and short-term (l-72 months) users of medium-potency estrogen (estradiol compounds and conjugated estrogens) (only those risk factors with significant associations with hormone intake are shown)

Factor Paritv 0 1 2-4 5+ Age at first birth 15-19 20-24 25-29 30+ Hysterectomy No Pre-MP Post-ME’ Oophorectomy No Pre-MP Post-MP Use of OCs Never Ever
Number of women; duration (months) 73+ l-72

OR for the factor, if long-term vs. short-term user (OR, 95% CI)

311 478 1,299 77

270 501 1,708 117

Ref 0.8 (0.7-1.0) 0.6 (0.5-0.8) 0.6 (0.4-0.8)

242 848 522 237

333 1,051 646 294

Ref 1.2 (1.0-1.5) 1.3 (1.1-1.6) 1.3 (1.0-1.7)

1,391 679 75

2,109 414 46

Ref 2.3 (2.0-2.7) 2.8 (1.9-4.1)

1,623 403 90

2,293 215 51

Ref 2.4 (2.0-2.9) 2.7 (1.9-3.8)

1,516 624 178 134 293

1,839 733 273 208 229

Ref 0.9 (0.7-1.0) 0.6 (0.5-0.8) 0.6 (0.5-0.8) 1.3 (1.1-1.6)

1,683 83 221 166

2,153 96 208 123

Ref 1.0 (0.8-1.4) 1.3 (1.1-1.6) 1.6 (1.2-2.0)

690 610 489 376

705 660 637 595

Ref 1.0 (0.8-1.1) 0.8 (0.7-1.0) 0.7 (0.6-0.8)

374 1,575

561 1,635

Ref 1.6 (1.4-1.8)

See footnotes in Tables 1 and 2 for abbreviations and explanations. “Highest level reported. bReported as one of several possible reasons for starting treatment. for ever-use. Notably, estrogen-only users had a lower odds of high school or more extensive school education and were more likely to be obese. DISCUSSION Our data show differences in the prevalence of several risk factors-pertinent to major diseases in women-that de-

I. Persson et al.

616

TABLE 4. Comparison of risk factors ers (more than 6 years) of conjugated estradiol (E2) compounds

between estrogen

Number of women, compounds (taken 73+ months) Factor

Hysterectomy No Pre-MP Post-MP Oophorectomy No Pre-MP Post-MP Use of OCs Never Ever
CE

long-term use (CE) and of

OR for the factor, if long-term user of CE vs. E2 (OR, 95% CI)

E2

227 226 27

986 370 35

Ref 2.9 (2.3-3.7) 3.3 (1.9-5.6)

320 129 28

1,103 224 45

Ref 2.1 (1.7-2.8) 2.1 (1.3-3.5)

373 108 40 27 34

946 438 114 91 223

399 17 41 25

1,069 59 150 118

64 376

0.7 1.0 0.8 0.4

Ref 0.8 (0.5-1.4) 0.7 (0.5-1.1) 0.6 (0.4-0.9)

280 985 -

See footnotes in Tables 1 and 2 for abbreviations “Highest level reported. ‘Reported as one of several possible reasons

(OYO.9) (0.7-1.5) (0.5-1.3) (0.3-0.6)

Ref 1.6 (1.2-2.2) and

explanations.

for starting

treatment.

pended on characteristics of HRT exposure. We noted rather complex patterns associated with the choices of starting treatment when prescribed, of continuing treatment, and of using different compounds and regimens. Non-compliance and short-term use were associated with higher parity and an earlier age at birth, and a lower prevalence of hysterectomy and oophorectomy, as compared with adherence to prescriptions and long-term exposure. Further, we found differences in factors that may be markersof health behavior and lifestyle at large. Notably, a higher level of education was linked to long-term hormone use, and more common practice of moderate to heavy physical exercise and more frequent intake of food fibers were associated with intake compliance. The presence of climacteric symptoms-possibly reflecting relatively low levels of endogenous estrogens [8]-was positively associated with compliance, long-term intake, and exposure to conjugated estrogens; whereas obesity-leading to increased estrogen production [3]-was linked to non-compliance and shortterm exposure. Lastly, exposure to OCs was less common

TABLE 5. Comparison of risk factors between long-term users of medium potency estrogens alone (E) and of estrogens combined with progestogens (E+P) Number of women, by regimens (taken > 73 + months) Factor

Hysterectomy No Pre-MP Post-MP Oophorectomy No Pre-MP Post-MP Use of OCs Never Ever
E

E+P

OR for the factor, if long-term user of E vs. E+P (OR, 95% CI)

621 531 61

838 83 11

Ref 9.7 (7.5-12.6) 7.7 (4.0-14.8)

822 310 66

839 66 12

Ref 5.1 (3.8-6.8) 5.7 (3.1-10.7)

935 276 92 70 103

586 341 90 69 177

0.5 0.7 0.7 0.4

990 36 117 75

671 46 111 108

0.5 (0.!%8) 0.7 (0.5-1.0) 0.5 (0.4-0.7)

371 345 287 223

320 281 182 156

1.0 1.0 (0.8-1.3) 1.3 (1.1-1.7) 1.3 (1.0-1.6)

4 47 326 687 150

2 24 189 574 141

1.1 (0.2-6.1) 1.2 (0.7-2.0) Ref 0.7 (0.6-0.9) 0.6 (0.5-0.9)

Ref (0.4-0.6) (0.5-0.9) (0.5-0.9) (0.3-0.6)

See footnotes in Tables 1 and 2 for abbreviations and explanations. “Highest level reported. ‘Index calculated from several food items, categories with increasing of fibers.

use

in women who had used estrogens alone than in those who had taken estrogen-progestin combinations. Our data imply that reported risk estimates might be confounded by factors related to the particular exposure characteristics being studied. This finding points to the need to recognize and measure a large number of factors, particularly those that may be markers of health behavior. Factors associated with the socioeconomic status and various aspects of lifestyle may be important determinants of both the incidence of and mortality from cancer and cardiovascular diseases [8]. What are the implications for interpretation of reported risk relationships? In previous analyses in this cohort of the risk of relationships between HRT and myocardial in-

HRT and Major

Risk Factors: Evidence of Confounding

by Exposure

farction [20], hip fractures [19], endometrial cancer [17], and breast cancer [18], the observed outcomes were compared with expected outcomes calculated from incidence rates obtained from the background population. A comparison of all women in the cohort, i.e., regardless of compliance or treatment choices, with women from a random sample of the general female population of similar age revealed that the cohort women had on the average more often been hysterectomized or oophorectomized, had a leaner body build, more often were current smokers, and had a higher level of education [21]. The present study-using information on lifetime hormonal exposure and a number of risk factorsextends the comparisons to specific subgroups within the cohort that have been identified as showing important riskaltering exposures with regard to reproductive cancers, cardiovascular diseases and osteoporosis [17-201. Our reported associations between long-term HRT and increased risk of endometrial cancer (only for unopposed estrogen exposure) and of breast cancer may have been confounded by differences in lifestyle factors or indications for HRT, the net effects of which are difficult to predict. As in the case of myocardial infarction and hip fractures, the finding of a higher level of education in HRT users than in non-compliers corroborates the assumption that our reported protective effects [20] may have been overestimated. The problem of residual confounding in HRT-disease relationships is being debated, particularly with regard to effects on coronary heart disease [10,12,13]. Analyses of risk factor data in the Nurses’ Health Study [15,22] and Leisure World Study [23], however, indicated that residual confounding of the protective effect of estrogens on CHD may be small. Adjustment for a number of covariates did not materially alter the risk estimates. Conclusions regarding selection biases in these cohorts may, however, be difficult to generalize, as nurses and members of a retirement community are probably relatively homogeneous with regard to sociodemographic factors and to health-oriented behaviors, as compared with groups of people in a general population. Although the magnitude of the possible protective effect of estrogens against the risk of CHD is uncertain, it is broadly agreed that HRT does have a true, biological, preventive effect [1,15]. This debate also concerns the possibility of residual confounding in the relationships of HRT to osteoporotic fractures and cancer outcomes [8]. When assessing the internal validity of our data, the advantages of a complete, unselected and population-based cohort and a long period of exposure ascertainment should be noted. However, as all data were obtained from a selfadministered questionnaire, there is probably some misclassification of both exposure and risk factor information. In a previous analysis the concordance between self-reported information on hormone intake and data from registered pharmacy records was, however, found to be high [24]. Fur-

617

ther, the external validity needs to be considered. Selective mechanisms are likely to vary between populations and with time. In fact, usage rates have increased in the last decade in Sweden [25], which may imply that new groups of women are becoming exposed. In summary, our data demonstrate that there are patterns of selection that are related not only to a woman’s decision to initiate a prescribed treatment but also to the choice of compounds and regimens and to the decision to continue intake over the long term. We have shown that factors associated with reproduction and health behavior, notably the level of education, habits of physical exercise, and OC use, could be sources of confounding that vary with exposure characteristics. Future epidemiological studies of HRT-related outcomes need to employ improved methods to measure various risk factors, particularly those linked to lifestyle and health behavior, so that they can be taken into account more fully in the analyses. Given the difficulties inherent in observational studies of HRT and long-term outcomes, randomized trials would be necessary when feasible. Supported by grants from the National Swedish

Cancer

Cancer

Institute,

USA,

and the

Society.

References 1. Stampfer MJ, Colditz GA. Estrogen replacement therapy and coronary heart disease: A quantitative assessment of the epidemiologic evidence. Preventive Med 1991; 20: 47-63. 2. Consensus Development Conference: Prophylaxis and treatment of osteoporosis. Am J Med 1991; 90: 107-110. 3. Ziel HK. Estrogen’s role in endometrial cancer. Obstet Gyneco1 1982; 60: 509-515. 4. Brinton LA, Schairer C. Estrogen replacement therapy and breast cancer risk. Epidemiol Rev 1993; 15: 66-79. 5. Stanford JL, Thomas DB. Exogenous progestins and breast cancer. Epidemiol Rev 1993; 15: 98-107. 6. Grady D, Rubin SM, Petitti DB et al. Hormone therapy to prevent disease and prolong life in postmenopausal women. Ann Intern Med 1992; 117: 1016-1037. 7. Hemminki E, Malin M, Topo P. Selection to postmenopausal therapy by women’s characteristics. J Clin Epidemiol 1993; 46: 211-219. 8. Adami HO. Long-term consequences of estrogen and estro, gen-progestin replacement. Cancer Causes & Control 1992; 3: 83-90. 9. Hemminki E, Sihvo S. A review of postmenopausal hormone therapy recommendations: Potential for selection bias. Obstet Gynecol 1993; 82: 1021-1028. 10. Matthews KA, Kuller LH, Wing RR, Meilahn EN, Plantinga P. Prior to use of estrogen replacement therapy, are users healthier than nonusers? Am J Epidemiol 1996; 143: 971978. 11. Barrett-Connor E. Postmenopausal estrogen and prevention bias. Ann Int Med 1991; 115: 455-456. 12. Vandenbroucke JP. Postmenopausal estrogen and cardioprotection (Editorial). Lancet 1991; 337: 833-834. 13. Posthuma WFM, Westendorp RGJ, Vandenbroucke JP. Cardioprotective effect of hormone replacement therapy in

618

I. Persson

postmenopausal women: Is the evidence biased? Br Med 1994; 308: 1268-1269. 14. Grodstein F. Invited commentary: Can selection bias explain the cardiovascular benefits of estrogen replacement therapy?

J

Am J Epidemiol1996;143:979-982. 15. Grodstein F, Stampfer M. The epidemiology of coronary heart disease and estrogen replacement in postmenopausal women. Prog CardiovascDis 1995;38: 199-210. 16. Persson I, Adami HO, Johansson EDB, Lindberg BS, Mane11 P, Westerholm B. A cohort study on estrogen treatment and risk of endometrial cancer: Evaluation of method and its applicability. Eur J Clin Pharmacol1983;25: 625-632. 17. Persson I, Adami HO, Bergkvist L, et al. Risk of endometrial cancer after treatment with oestrogens alone or in conjunction with progestogens: Results of a prospective study. Br Med J 1989; 298: 147-151. 18. Bergkvist L, Adami HO, Persson I, Hoover R, Schairer C. The risk of breast cancer after estrogen and estrogen-progestin replacement. N Engl J Med 1989;321: 293-297. 19. NaessCn T, Persson I, Adami HO, et al. Hormone replacement therapy and the risk for first hip fracture. A prospective, population-based cohort study. Ann Int Med 1990;113:95-103.

et al.

20.

Falkebom M, Persson I, Adami HO, et al. The risk of acute myocardial infarction after oestrogen and oestrogen-progestogen replacement. Br J Obstet &aGynaecol 1992;99: 821-

21.

Bergkvist L, Adami HO, Persson I, Schairer C. Risk factors for breast and endometrial cancer in a cohort of women treated with menopausal estrogen. Int J Epidemiol1988;17:

828.

732-737. 22. Stampfer MJ,

23.

Colditz GA, Willett WC, Manson JE, Rosner B, Speizer FE, Hennekens CH. Postmenopausal estrogen therapy and cardiovascular disease. Ten-year follow-up from the Nurses’ Health Study. N Engl J Med 1991;325: 756-762. Henderson BE, Paganini-Hill A, Ross RK. Decreased mortality in users of estrogen replacement therapy. Arch Int Med

1991;325: 756-762. 24. Persson I, Bergkvist L, Adami

HO. Reliability of women’s histories of climacteric estrogen treatment assessed by prescription forms. Int J Epidemiol1987;16: 222-228. 25. National Pharmaceutical Company. Swedish Drug Statistics 1978-1992, yearly publication, Apoteksbolaget1982-1992.