Reproductive history and osteoarthritis in the Women's Health Initiative.

The Women’s Health Initiative (WHI) baseline cohort includes 161,808 post-menopausal women aged 50-79 years who were enrolled in 40 clinical centers throughout the US; details have been published previously.( 22 ) Women were enrolled between 1993 and 1998 into either an Observational Study (OS) cohort (93,676 women) or a Clinical Trial (CT) cohort (68,132 women). Women who self-reported rheumatoid arthritis, systemic lupus erythematous, or ulcerative colitis were excluded from the study due to possible overlap of symptoms, which may result in inaccurate self-report of OA. Of the 161,808 women enrolled in the WHI, 151,767 remained after exclusion for these conditions. We did not exclude women who underwent joint replacement as this population likely included women with OA. The final cohort included 147,975 women who were not missing information on parity or type of arthritis, and the parous cohort consisted of 130,331 women ( Figure 1 ). Osteoarthritis prevalence was self-reported on Form 30 of the WHI at baseline. Participants were asked “Did your doctor ever say that you have arthritis?” with the choices as “Yes” or “No.” For participants selecting “Yes,” they were then asked “What kind of arthritis do you have?” with possible responses of “rheumatoid arthritis,” “others/don’t know,” and “missing.” Women who answered no to the first question were classified as not having osteoarthritis. Women were classified as having osteoarthritis if they answered yes to the first question and “Others/don’t know” to the second question. This method has been previously used in a WHI study on osteoarthritis, ethnicity, and BMI.( 23 ) Parity was assessed based on women’s response to the questions “Have you ever been pregnant?” and “How many live births did you have?”. Women who answered that they had never been pregnant or that they had no live births were classified as nulliparous. Women who reported at least one live birth were classified as parous. Age at menarche was self-reported through questionnaires. For women with no oophorectomy, age at menopause was self-reported, defined by the age at which a woman last had any menstrual bleeding or began using menopausal HT. For hysterectomy without bilateral oophorectomy, the age at menopause was self-reported based on the age at which a woman either began using HT or first had vasomotor symptoms. For women who had a hysterectomy without bilateral oophorectomy at age 50 years or older but no use of HT or symptoms, the age at menopause was defined as the age at hysterectomy. If the algorithm defined an age at menopause as older than 60 years, it was recorded as 60 years.( 24 ) Breastfeeding period was self-reported based on question “Thinking about all the children you breastfed, how many months total did you breastfeed?” We employed logistic regression to estimate the univariate association between osteoarthritis and reproductive factors that may influence estrogen exposure over a lifetime. The associations of the above reproductive factors were further explored in a single multivariable model while adjusting for potential confounders. We investigated multicollinearity between the predictors prior to fitting the multivariable models. Of note, we were only able to study associations rather than causality as this was a cohort study, and there was not an a priori experiment set up to investigate this relationship. We adjusted for potential confounders that may affect the relationship between osteoarthritis and reproductive factors based on literature. Confounders included the following( 25 ): parity, baseline HT use and duration in years, OCP use and duration in years (oral contraceptive pills which may be combined estrogen + progestin, estrogen only, or progestin-only), age at WHI baseline, ethnicity (American Indian, black, Hispanic, other, Pacific Islander, white), income ($100,000), insurance status (yes/no), current health care provider (yes, no), smoking (never, past, current), alcohol use (never, past, current), body mass index (BMI defined as kg/m 2 ), history of diabetes mellitus (yes, no), physical activity, and enrollment into OS versus CT. Physical activity was assessed through baseline questionnaires about intensity and duration of exercise, which was then converted into a measure of metabolic equivalent (MET)-min/week (=1200). Among parous women, we then analyzed odds ratios of osteoarthritis in relation to age at first birth, age at first pregnancy, number of pregnancies, number of live births, and breastfeeding duration in a multivariable model adjusted for the potential confounders listed above. All confounders were ascertained from baseline questionnaires, which were given at enrollment between 1993 and 1998. Due to missing self-reported information of some potential confounders at baseline in the WHI cohort, we used multiple imputation to allow us to include all women in the models. The number of women missing each variable is displayed in Table 1 . For multiple imputation, we used the R package Multivariate Imputation by Chained Equations (MICE)( 26 ), which imputes each missing value with a simulated plausible value until all missing values are imputed. We created 5 imputed datasets, ran our models on each dataset, and then pooled the estimates from the imputed datasets. Additional details on the missing data and multiple imputation approach are included in Supplement 1 . We expect there to be differences from a multiple imputation-based approach and a complete-case approach, as validity for the latter relies on an assumption that the data are missing completely at random, an unrealistic assumption for most studies in medicine. The validity for the former (our approach) relies on an assumption that the data are missing at random, a more practical and flexible assumption that missingness is related to observed variables only. Analyses were conducted in R version 3.2.3. All statistical tests were two-sided and performed at the alpha = 0.05 level. P-values were not adjusted for multiple comparisons given single outcome and should be interpreted accordingly.

The baseline characteristics in our cohort are displayed in Table 1A , stratified by parity status. In our cohort, there were 130,331 (88.1%) parous women and 17,644 (11.9%) nulliparous women. Most characteristics were statistically significantly different between the two groups due to large sample size, though many values were similar. The mean age in the parous cohort was 63.2 years versus 62.6 years in the nulliparous cohort. Both cohorts were primarily Caucasian. Reproductive characteristics for the parous cohort are displayed in Table 1B . In the entire cohort, 64,868 (43.8%) women reported osteoarthritis. Table 2 displays the odds ratios of self-report of OA in relation to reproductive factors for the entire cohort. No significant clinical associations were found between measures of reproductive history and osteoarthritis due to small effect size. Younger age at menarche was associated with statistically increased (but clinically insignificant) likelihood of OA (aOR 0.992, 95% CI 0.984-0.999 for age 12; aOR 0.996, 95% CI 0.988-1.003 for age 13; and aOR 0.998, 95% CI 0.990-1.007 for age >13 compared to age <12; global p<0.001) in our adjusted analysis, while age at menopause was not associated with OA (aOR 1.000 per year, 95% CI 0.997-1.002, p=0.851). History of parity was associated with OA, aOR 1.017 (95% CI 1.009-1.026, p<0.001 compared to non-parous). OA had clinically insignificant associations with both OCP use (aOR 1.008 95% CI 1.001-1.016, p<0.01 compared to never users), and current use of HT [reference current users, 0.951 (95% CI 0.943-0.959) for never and aOR 0.981 (95% CI 0.972-0.989) for past users, global p<0.001]. Odds of OA did not show a clinically significant association with duration of OCPs (aOR 0.998 per year, 95% CI 0.997-0.999 p<0.001) or HT (aOR 1.001 per year, 95% CI 1.000-1.002 p=0.012). Women who had a history of hysterectomy had self-report OA aOR 1.013 (95% CI 1.004-1.022) compared to no hysterectomy. In addition, women with unilateral oophorectomy had risk of OA aOR 1.015 (95% CI 1.004-1.026, p<0.01) compared to no oophorectomy; however, women with history of bilateral oophorectomy were found to have decreased OA risk (aOR 0.987, 95% CI 0.978-0.995, p = 0.007 compared to oophorectomy). Again, it is important to note all associations were clinically insignificant. We also separately investigated the relationship between osteoarthritis and additional reproductive factors among parous women in Table 3 . Age at first birth and pregnancy were not associated with OA. The number of pregnancies and live births in relation was associated with OA (global p<0.001 for both); however, no clear pattern was observed with the number of pregnancies or births in relation to OA. Similarly, no clear pattern was observed with breastfeeding duration and OA.

The relationship between hormonal factors and OA is complex. Reproductive factors are known to affect estradiol levels, and some prior studies have reported that decreased estradiol levels are associated with increased risk of knee OA.( 27 , 28 ) Estradiol has been demonstrated to promote the health of skeletal muscle through multiple pathways, including transforming growth factor-β and insulin-like growth factor-1 and 2.( 8 – 10 ) OA is more common in women than men and its incidence also rises sharply starting around the time of menopause, which also suggests possible contribution of hormonal factors in its pathogenesis.( 5 , 6 ) However, despite these studies, the mechanisms underlying the relationship between estrogen (and other hormones) with OA remain heterogeneous and unclear and further understanding of clinical associations is warranted. Other reproductive factors such as parity and pregnancy may also affect OA development due to weight gain, joint loading, and local and systemic inflammation.( 12 – 14 ) Previous studies on the relationship between OA and reproductive factors have reported conflicting findings, likely due to heterogeneity in sample size, methodology, populations and demographics, and different methods of assessing OA (including self report, radiograph assessment, and proxies such as joint replacement). Multiple studies have reported the relationship of OA with parity and age of menarche/menopause. A study of 4.6 million Danish patients revealed that parity was positively associated with OA hospitalizations, the risk of OA hospitalization (particularly for knee) increased with number of children, and all subtypes of OA hospitalization were positively associated with time after the most recent childbirth.( 15 ) The Million Women Study prospectively analyzed the association between reproductive factors and the risk of primary hip and knee replacement for OA among 1.3 million women in England and Scotland (1996-2001). Younger age at menarche (<12) was associated with greater risk of hip and knee replacement for OA (OR1.09-1.15 depending on age category), while menopausal status and the age of menopause were not significantly associated with hip or knee replacement for OA.( 16 ) In another study, increasing age of menarche was associated with reduced risk of total knee replacement due to primary OA.( 29 ) A separate study also reported that younger age at menarche was associated with risk of radiographic hand OA.( 30 ) However, another study found that with regard to radiographic OA, parous women had higher odds of both joint space narrowing and osteophytes compared to nulliparous women, though neither reached statistical significance.( 17 ) Several other studies have reported no significant association between the age of menarche or menopause and the risk of hip OA.( 18 , 19 ) Though the existing literature is conflicting as noted above, our analysis is consistent with prior large-scale studies in finding that younger age at menarche and history of parity are associated with statistically increased risk of OA,( 15 , 16 , 31 , 32 ) though not all studies have reported these relationships.( 18 , 21 , 33 ) However, our findings were not clinically significant as compared to other studies that found effect sizes ranging from 5-20% as previously cited in this paragraph. Age at menarche and parity both affect lifetime exposure to estrogen; in addition, pregnancy and childbirth can also cause physiologic changes such as increased weight on joints which may impact OA risk. However, in our study no clear trend existed for the number of pregnancies or live births in relation to OA. We also found that a number of factors were not associated with self-reported OA in the WHI, including age at menopause, first birth, and first pregnancy; the literature findings on these associations have been inconsistent as above. Additionally, a systematic review of 16 studies found that there was no association between female hormonal aspects and OA of the hand, hip, and knee.( 20 ) In our study, we did not find any clinically significant association with most other studies with odds ratios very close to 1. The definition of clinical significance varies depending on the field and variables of interest, but is typically closer to the range of an effect size of at least 5-10%, which is larger than the findings in our study. Other articles cited on this subject (see studies cited in Discussion ) have typically reported effect sizes of greater than 5%. This may be due to our large sample size as well as our ability to control for a comprehensive set of confounders and reproductive factors, including hysterectomy and oophorectomy status. The literature on hysterectomy and oophorectomy in relation to OA is very limited and our study is one of the first to study this association, finding a statistically but not clinically significant increased effect. Hysterectomy may affect OA due to complex hormonal effects; studies have shown that hysterectomy itself, even with ovarian conservation, may affect hormonal function and lead to earlier menopause and ovarian failure.( 34 ) Limited studies have been mixed on the relationship between oophorectomy, hysterectomy, and OA ( 33 , 35 , 36 ); however, it is unclear if oophorectomy and HT were included as confounders in all studies. These complex hormonal relationships of hysterectomy and oophorectomy in relation to OA warrant further investigation. Our study also found that use of OCPs and current use of HT were both found to be associated with increased OA self-report in WHI, though findings were not clinically significant. The literature on HT and OA is conflicting. The Million Women Study of 1.3 million patients also reported that current use of postmenopausal HT was associated with significant increase in hip and knee replacement incidence, while OCPs were not associated with OA; mechanisms of these findings were unclear.( 16 ) A cross-sectional study of 489 women also found that parity (but not HT or OC use) was independently associated with greater patellar cartilage defects (OR 2.87).( 17 ) However, several studies have not found significant associations or even reductions of OA in relation to HT or OC use.( 37 ).( 36 ) ( 38 ) The Women’s Health Initiative has previously reported that women receiving estrogen-only therapy had significantly lower rates of any arthroplasty (as a proxy for severe OA), though associations were not significant for either hip or knee arthroplasty when examined separately. No association was found for estrogen-plus-progestin replacement and arthroplasty, suggesting a possible effect of unopposed estrogen in relation to bone health.( 39 ) This is in contrast to our findings of HT being associated with increased self-report of OA. This may be due to the fact that we studied any self-report of OA (as compared to OA that was clinically significant enough to result in arthroplasty). In addition, HT use may signify lower levels of estrogen due to hypoestrogenism symptoms and lower estradiol levels at baseline, which may reflect differences in the underlying populations rather than the effect of HT itself (as the timeline of when the OA developed in relation to the hormone use is unknown in our cohort). Overall, the relationship between HT and OA is very conflicting in literature and warrants further study, including on durations and types of HT. In regards to breastfeeding, we found no clinically significant associations with odds ratios close to 1. Lactation is a low estrogen state, and the limited evidence is mixed on breastfeeding and osteoarthritis. A cross-sectional study of 348 women reported that ever breastfeeding was found to be associated with a 63% decrease in clinically verified carpometacarpal joints (CMC) OA. However, this association was not found to be consistent across sites or severity measures and may reflect false positive association.( 21 ) This is in contrast to an NHANES study which found that women who breastfed for one month or longer had a statistically significant 21% increased self-reported OA risk.( 40 ) Overall the literature on the effect of reproductive events on OA is inconsistent and of unclear clinical significance. The strengths of our study include the large sample size and richness of the dataset for reproductive variables of interest. Additionally, we were able to adjust for not only reproductive factors, but also known confounders of OA including BMI, diabetes, as well as proxies of access of health care. We also examined OA directly rather than proxies of OA such as hospitalization or joint replacement. We were also able to include information on hysterectomy status and oophorectomy status, which has rarely been studied in literature. Overall, very few studies have included such a comprehensive set of reproductive and non-reproductive confounders in relation to OA in the same cohort. A major limitation of our study was that OA was self-reported, which is subject to recall bias and inaccuracy. A prior WHI study reported that validation of self-reported data collection for OA has not been widely investigated, and another study found that a rheumatologist could confirm 81% of self-reported OA cases.( 23 , 41 ) We attempted to adjust for this by excluding other conditions which may be confused with OA, including rheumatoid arthritis, SLE, and Crohn’s Disease. The literature definitions of OA vary widely and range from self-report to radiographic verification to joint replacement; compared to our study, objective measures such as radiographic OA or joint replacement may be more valid and objective, though may not capture less severe OA cases or cases that were not radiographically verified. It is important to note that our findings can only be interpreted for self-report of any OA. In addition, our cohort was mostly Caucasian, limiting the generalizability of the analysis. Another limitation was that the study was a retrospective cross-sectional format and therefore we do not have information on the timing of OA development in relation to different reproductive factors, which makes interpreting associations more difficult. For reproductive factors occurring late in life (e.g. age at menopause), it is possible that OA developed prior to menopause and we cannot be certain of the temporal relationship between the exposure and the outcome. However, the incidence of OA dramatically increases with age, so we would expect most diagnoses to have occurred when women were older. Another limitation is that we only had report of overall OA and not site-specific OA or information about OA severity; it is important to note that OA risk factors likely differ by site. The findings may be diluted as a result. In conclusion, in the large, multi-ethnic Women’s Health Initiative, we did not find reproductive factors to be clinically significant in relation to self-report of OA. The existing literature on reproductive factors in relation to OA is conflicting and heterogeneous; our study adds to the body of literature suggesting that reproductive factors are likely not to have large clinical significance on the development of OA after adjustment for comprehensive risk factors. Given the large economic and medical burden of OA in older women, this subject warrants additional investigation in large prospective cohorts with control for all relevant reproductive and hormonal factors. In particular, the relationships between hysterectomy, oophorectomy, and HT in relation to OA have been rarely studied in literature. Additionally, other areas for future study include reproductive history in relation to biological mechanisms of OA, specific OA subtypes, degrees of clinical severity of OA, ethnic differences in OA, and timing of OA development.

Osteoarthritis (OA) is the most prevalent form of musculoskeletal disease and is a major cause of impairment in an elderly population, with over 10% of the world’s population over age 60 years reporting clinical problems associated with OA( 1 , 2 ) including severe pain and disability.( 3 , 4 ) The prevalence of OA among women is greater than men, with rates of 14.9% versus 7.9% (ages 60-69), and 16.5 versus 10.2 % (ages 70-79).( 5 , 6 ) Sex hormones, particularly estrogen, are hypothesized to play an important role in the pathophysiology of OA in women( 7 ) through multiple pathways.( 8 – 11 ) Additionally, other reproductive factors such as parity and pregnancy may affect risk for OA development due to factors such as weight gain, joint loading, and local and systemic inflammation.( 12 – 14 ) The relationship between reproductive factors and OA is complex and prior studies have reported conflicting findings, which may be due to heterogeneity in sample size, methodology, populations, and demographics. Two large scale studies reported that parity and younger age at menarche were both positively associated with either OA or proxies of OA.( 15 ).( 16 ) However, other studies have reported a non-significant relationship between parity and radiographic OA( 17 ), as well as the age of menarche and menopause and the risk of hip OA.( 18 , 19 ) Additionally, a review of 16 studies found that there was no association between female hormonal aspects and OA of the hand, hip, and knee.( 20 ) The literature on breastfeeding and OA is limited, but one study suggested that breastfeeding may be protective for carpometacarpal joint (CMC) OA.( 21 ) The existing literature is conflicting, and most studies have included only a subset of relevant reproductive factors. We aimed to study associations of reproductive factors with self-reported OA using the large, multi-ethnic Women’s Health Initiative (WHI) in a retrospective cross-sectional format, using the WHI’s rich dataset on reproductive history and exogenous estrogen use.