| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
|
|
|
|||||||
|
|||||||||
a School of Public Health, Saint Louis University, Missouri
b School of Medicine, Saint Louis University, Missouri
Kathleen W. Wyrwich, Saint Louis University School of Public Health, 3663 Lindell Boulevard, St. Louis, MO 63108-3342 E-mail: wyrwichk{at}slu.edu.
William B. Ershler, MD
 |
Abstract |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Methods. We replicated the Iowa approach using the Longitudinal Study on Aging (LSOA), a nationally representative, prospective cohort study. The 3131 community-dwelling women for whom we had complete data for these analyses ranged in age from 70 to 98 years old at baseline in 1984. Using ICD9-CM 174 codes, linked Medicare hospital claims identified 77 women with hospitalizations for breast cancer between 1984 and 1991. Multivariable proportional hazards regression was used to model the risk for this event among disabled, inactive, moderately active, and highly active women.
Results. No significant association between disability in older women and the risk of hospitalization for breast cancer relative to inactive older women was detected (adjusted hazard ratio [AHR]-0.78, 95% CI 0.41–1.5). Highly active older women had a significantly reduced risk of hospitalization for breast cancer (AHR-0.42, 95% CI 0.19–0.95).
Conclusion. The intriguing finding from the Iowa 65+ Rural Health Study that disabled older women's risk for breast cancer was reduced could not be replicated in the LSOA, although power was limited. Highly active older women, however, had a significantly lower risk for breast cancer in both studies.
LAST year, a review of more than 20 studies on the association between physical activity and breast cancer yielded inconsistent conclusions and underscored the need for standardized methods for measuring physical activity in women (1). At the same time, this Journal published an intriguing finding from an investigation of physical activity, physical function, and the risk of breast cancer among women aged 65 years or older in the Iowa 65+ Rural Health Study, one of the four sites of the Established Populations for Epidemiologic Studies of the Elderly (EPESE) (2). In addition to observing that higher levels of physical activity in older women were inversely associated with breast cancer risk, Cerhan and colleagues found a markedly decreased risk of breast cancer among older women with any disability compared with physically capable, but inactive older women (relative risk [RR] = 0.4, 95% confidence interval [CI] 0.2–0.9). Because the researchers had no clear reason for this protective effect exhibited in disabled older women, it was recognized that "this finding needs to be replicated before drawing any conclusions" (2).
Accordingly, this brief report attempts to replicate that intriguing finding by using the Longitudinal Study on Aging (LSOA), a prospective cohort study that enrolled a nationally representative sample of community-dwelling persons ranging in age from 70 to 98 years old in 1984 (3). Medicare hospital claims were used to identify women with hospitalizations for breast cancer between 1984 and 1991, and vital status was taken from the National Death Index (NDI) for the same period. Although the Iowa EPESE and LSOA baseline interview items are not identical, they are sufficiently similar to facilitate a reevaluation of the association between physical activity, disability, and the risk of breast cancer hospitalizations among older women.
|
Methods |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Case Identification
In the Iowa EPESE, incident cases of breast cancer during the 1982–1993 observation window were identified through the State Health Registry of Iowa's cancer database, which is part of the National Cancer Institute's Surveillance, Epidemiology and End Results (SEER) program. In the LSOA, Medicare claims were used to identify breast cancer hospitalization cases between 1984 and 1991. Women with a hospital episode containing an ICD9-CM discharge diagnosis of 174.x in the primary (first) field, or with a discharge diagnosis of 174.x in a secondary (second through fifth) field and surgical codes for both an open breast biopsy and simple mastectomy, were considered breast cancer cases if the discharge date occurred after their baseline interview. In addition, three other women, whose first hospitalizations occurred more than 5 years into the observation window, had either liver or bone secondary metastases as their primary diagnoses, with 174.x as a secondary diagnosis. These three women were also included as cases because breast cancer was the principal cause of their hospitalizations. Measurement of tumor stage (local, regional, and distant) was accomplished through Cooper and colleagues' recently published algorithm using ICD9-CM codes for metastatic diagnoses in recent or concurrent hospitalizations (4). The index event date for cases was coded as the number of days between January 1, 1984, and the date of their breast cancer hospitalization discharge. Women with complete baseline data but no discharge diagnoses indicating a breast cancer hospitalization served as the controls, who were censored at the time of their death or January 1, 1992, whichever came first.
Disability and Physical Activity Level Criteria
In the Iowa EPESE, women were identified as disabled (39%) if they could not perform any one or more of the Rosow-Breslau items without help (i.e., heavy housework, walk up and down a flight of stairs, or walk half a mile) (5). Those not disabled were then classified as inactive (21%), moderately active (30%), or highly active (10%), using their baseline responses to five activity items regarding gardening, housework, walking, and vigorous and moderate exercise (6). This approach, however, is problematic. It assumes that all disabled women are inactive.
Therefore, in the LSOA, we first classified each woman as highly active, moderately active, or inactive. To best approximate the Iowa EPESE method (in the absence of identical measures), we used the women's baseline responses to two items involving having a regular exercise routine (yes/no) and walking a mile or more at least once a week (yes/no). Women answering yes to both items were categorized as highly active, those answering yes to only one item were considered to be moderately active, and those answering no to both items were regarded as inactive. We then determined disability only among the women who were classified as inactive by using three items that asked "Do you have difficulty _____ without help (yes/no)?" The blank was filled in with "doing heavy housework," "walking up 10 steps without rest," and "walking a quarter of a mile." For each yes answer, the women were asked how much difficulty they had (some, a lot, or unable). Those inactive women who were unable to do at least one of these three activities were considered to be disabled. We believe that this approach is better than that used in the Iowa EPESE (6) because without it, 165 LSOA women would have been classified as disabled even though they exercised regularly and/or walked at least a mile or more each week.
Covariates
As in the study by Cerhan and colleagues, a number of potential confounders measured in the LSOA were considered, including age, obtaining a high school education, height, weight, body mass index (kg/m2), and physician visits in the year prior to baseline (a proxy for detection bias). When these covariates were compared between the physical activity/disability categories, statistically significant (p < .05) results were found for all one-way analysis of variance (ANOVA) tests (continuous covariates) and chi-square tests (dichotomous covariates). The Iowa EPESE study also adjusted for both systolic and diastolic blood pressure and five items involving female participants' health history (age at menarche, first pregnancy, menopause, use of hormone replacement therapy, and number of pregnancies). Although these seven variables are not available in the LSOA, we note that none of them had a meaningful impact in the Iowa EPESE study's risk analyses (6).
Statistical Analyses
With SEER data, the Iowa EPESE investigators were able to exclude from their analyses women with prior breast cancer diagnoses (1973–1982). In the LSOA, however, the only way to approximate this was either to exclude all women who self-reported having had cancer of any kind during their baseline interviews or to use that information as a covariate. We chose the latter approach for three reasons: (i) skin cancer is the most common self-reported cancer; (ii) the validity of lifetime self-reports of cancer have been questioned in this age cohort (6); and (iii) we wanted to maintain sample size.
Multivariable proportional hazards regression, with fixed variable entry of the respective covariates, was used to replicate the six risk models developed by Cerhan and colleagues. As in the Iowa EPESE, inactive women served as the referent category. A base model (Model I) included only activity-disability status, age, and prior cancer at any site. Model II added adjustments for education and body mass index, and Model III re-estimated after excluding women with breast cancer hospitalizations during 1984–1985 to evaluate whether "subclinical breast cancer might cause a decrease in physical activity" (2). Model IV proceeded with an adjustment to Model II for physician visits in the year prior to baseline (fully adjusted model). Adjusting only for age and prior cancer at any site, we fit separate models for local (Model V) versus regional and distant stage breast cancer hospitalizations (Model VI). Finally, to capitalize on a possible explanation suggested by the Iowa EPESE study, we reestimated Model IV, after excluding women with a self-reported history of hip fracture or osteoporosis at baseline in Model VII. All analyses were completed using SPSS for Windows (7), with the statistical assumptions assessed using standard methods (8).
|
Results |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Incident hospitalization counts, person-years of follow-up, adjusted hazard ratios (AHRs), and 95% CI for the six risk models are shown in Table 1 . In all models, the disabled older women's risk of hospitalization for breast cancer was less than the null, but not significantly different (p < .05) from that of inactive older women (referent). Highly active older women had the lowest risk of being hospitalized for breast cancer in all models, and this reduced risk was statistically significant in Models I, II, IV, and VI. Excluding the 14 women with breast cancer hospitalizations in 1984–1985 appreciably affected only the risk for highly active women and the precision of that risk (Model II vs Model III), whereas excluding the 311 women with self-reported osteoporosis or broken hips prior to their 1984 interviews moved the disabled older women's risk closer to the null (Model VII).
|
|
Discussion |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Limited support was found for a conjecture offered by the Iowa EPESE investigators to explain their finding of disabled older women's reduced risk. Citing recent reports linking estrogen levels to the reduced risk of breast cancer among women with either lower bone mass or wrist and hip fractures (9)(10)(11)(12), Cerhan and colleagues hypothesized that the presence of these conditions, which are precursors to disability, may have produced the protective effect observed for the disabled women in the Iowa EPESE study. When the 311 LSOA women who self-reported having osteoporosis or a prior hip fracture were excluded from the analysis (Model VII), the AHR for the disabled women (.94) was noticeably altered and became virtually identical to that of the physically inactive women (referent).
Because of the design and data collection differences between the Iowa EPESE and the LSOA, our attempt at replication is not definitive. The most important of these differences involves the classification of disability and activity, case identification, assurance of incident cases, and the power to detect significant effects. The differences in the classification of disability and activity result from two issues: the slightly different underlying items, and the way in which they were combined. In an attempt to address the former, we reestimated all of the models after reclassifying the women by using a subjective item that asked them to rate their activity level relative to that of their peers. The results of those analyses (not reported) found no significant effect for disabled older women's risk. Although we believe our approach to restrict the classification of the disabled to only those women who are physically inactive is conceptually more appropriate than that used in the Iowa EPESE, we also reestimated all of the models by using their approach. Those analyses (not reported) also found no significant effect for disabled older women's risk. Thus, it appears that our failure to replicate the intriguing protective effect of disability on the risk of breast cancer is not an artifact of these classification differences.
Despite the case identification and incident case assurance differences between the two studies, the ratio of cases per person-years of follow-up in our Model III (.0032) is virtually identical to that in the comparative Iowa EPESE model (.0029). Although this aggregate comparison is somewhat reassuring, other recent investigations comparing the SEER registry with Medicare hospitalization claims data have found the latter to be lacking in their sensitivity to detect incident breast cancer cases in older women (13)(14). Therefore, we are unsure whether case identification and incident case assurance differences account for our failure to replicate the intriguing Iowa EPESE finding.
Finally, power may also limit our ability to detect a significant protective effect for disabled women. Among women hospitalized with localized breast cancer (Model V), the AHR for disabled women comes closest to the adjusted risk estimate in the Iowa EPESE study (0.57 vs 0.4), although the 95% CI was much wider (0.23–1.40 vs 0.2–0.9). In this model, 1% of the disabled women were hospitalized for breast cancer, and 2% of the inactive women had this outcome between 1984 and 1991. The model's power to detect this protective effect with statistically significant confidence intervals (
< .05) is only 77%. Interestingly, this is the same level of power that the fully adjusted model (IV) has for finding an effect for disabled women (15). Consequently, the small number of events among LSOA women limits our ability to conclude whether there is a statistically significant effect in the disabled/inactive women comparisons, although our AHR point estimates were protective in all replicate models.
|
Acknowledgments |
|---|
Received April 6, 1999
Accepted December 12, 1999
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. Suzuki, M. Kojima, S. Tokudome, M. Mori, F. Sakauchi, Y. Fujino, K. Wakai, Y. Lin, S. Kikuchi, K. Tamakoshi, et al. Effect of Physical Activity on Breast Cancer Risk: Findings of the Japan Collaborative Cohort Study Cancer Epidemiol. Biomarkers Prev., December 1, 2008; 17(12): 3396 - 3401. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C M Friedenreich and A E Cust Physical activity and breast cancer risk: impact of timing, type and dose of activity and population subgroup effects Br. J. Sports Med., August 1, 2008; 42(8): 636 - 647. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| All GSA journals | The Gerontologist |
| Journals of Gerontology Series B: Psychological Sciences and Social Sciences | |
