By using a selected cohort of 1,068 women, we built a FXPOI prediction model based on FMR1
CGG repeat size, menopausal age of first degree relatives, and the environmental factor smoking, together with a correction for ascertainment location. We found that this model can predict the risk for menopause at a certain age, as indicated by the acceptable fit, the small confidence intervals and the good results of the internal validation procedure (i.e., small amounts of optimism in R2
and concordance index c
and a shrinkage factor close to 1). To our knowledge, this is the first model developed to help predict menopausal age in FMR1
premutation carriers. Other reproductive factors such as menarche and hormone use were not significantly associated with menopausal age in the premutation carriers, and thus not included in the prediction model. Van Noord et al.12
reported that the impact of environmental factors on menopausal age is limited, which seems to be true for premutation carriers as well. A prominent association with smoking (hazard ratio of 1.34) and menopausal age in premutation carriers was found, in accordance with data previously reported by Allen et al.13
The toxic effect of smoking could give rise to destruction of the primordial oocytes at the level of the ovary 28
and could, therefore, accelerate the onset of FXPOI.
Many studies dealing with relationship between various factors, such as menarche, OC use, BMI and parity, and early menopause (including POI and EM) have turned out to be controversial.12,29-32
In our set of premutation carriers no significant association with menopausal age was seen for these factors.
In the final model the FMR1
CGG repeat size, represented by ‘risk index by repeat size (RIR)’, is the most influential predictor. This was expected since the repeat size has been strongly associated with FXPOI before.13-16
To accommodate the non-linear association of repeat size and risk for FXPOI and to avoid the somewhat arbitrary classification of repeat groups, we used an HR-based approach and incorporated these risks relative to a reference repeat group into the prediction model. In this way the fast rise in risk after 60 CGG repeats and the slower decline in risk after 100 CGG repeats can be modelled. The HR pattern for repeat size indicates that a significant high risk for FXPOI starts earlier and continues longer than what was previously reported.13,14
Due to the low HRs from 55-65 repeats, the higher risk at the end of the previously defined 55-79 premutation range might not have been evident before. The high risk around 105-110 repeats has never been described before and may be attributed to small numbers (three women with a repeat size of 110 and one with a repeat size of 105 and a menopause at a relatively young age). An isolated high risk for FXPOI based on a few repeats is considered unlikely. Despite a small sample size, a significant risk was also seen for 160 repeats. Thus, an elevated risk above 130 repeats could not be excluded and requires further assessment. Discrepancies exist about risk for ovarian insufficiency among intermediate CGG repeat size carriers. Our analysis confirms that intermediate sized FMR1
CGG repeats should not be considered a high risk factor for ovarian insufficiency, as recently reported by Bennett et al.33
In addition, we did not observe a higher risk for menopause at an early age among intermediate repeat alleles. We examined this association using the defined genotypes by Gleicher and his colleagues, where they defined a normal allele as those with 26-34 repeats and abnormal alleles as those outside of this range.34,35
None of the women in our study who were either heterozygous or homozygous abnormal had menopausal age before the age of 40. Most had menopausal age above 45 years. Additional studies need to be conducted to further evaluate the association of these genotypes on phenotypes associated with ovarian function.36-39
We observed a significant effect of ascertainment location of our study population. Several possible explanations for this effect have been explored. We excluded a difference in distribution of CGG repeat sizes and the proportion of premutation carriers in both groups as a basis for the observed ascertainment location effect. Also, a difference in genetic background of the Dutch and American populations seems unlikely because of the inclusion of Caucasians only. Most Caucasian Americans are from European descent. One explanation could be the manner in which the interviews were taken. In the Atlanta study, the reproductive questionnaire was part of a large assessment, often including neuropsychological, medical and neurological assessments as well. The Nijmegen study was set up for reproductive evaluation and the study information included a description of FXPOI, which could result in participants reporting menopause at an earlier age than when it actually occurred.
A limitation of all cross-sectional studies is the use of time-dependent variables such as smoking, BMI, parity, and hormone use. Such variables are preferably analyzed by an extended Cox proportional hazards model. However, exact information of these variables per subject at different time points between 25 years of age and the start of menopause were lacking. Therefore, these variables were applied as if they were fixed at the age of 25 years. Even though this is a raw approximation of the reality, results based on this approach are at least indicative for the overall effect of these variables on menopausal age. In addition, age at menopause is a transitory state and subject to reporting and recall bias. A previous study on recall of menopausal age showed a bias towards the mean as women deviated from the final menstrual period.40
In our case, recall and memory bias might have led to an earlier menopausal age because of awareness for FXPOI.
Furthermore, we have used age at menopause as a surrogate for severity of FXPOI, but in general POI should not be considered a natural menopause, because of intermittent ovarian function and a 5-10% chance of pregnancy after diagnosis.41
We used the definition of amenorrhea for at least 1 year due to menopause as a uniform criterion that applies to older and younger women.
Taken together, our data strongly indicate that a prediction model for FXPOI is feasible, when taking ascertainment site into account. This effect clearly confirms the importance of data collection and ascertainment protocols, as they could bias results. Nevertheless, we consider our model as a first step in developing clinically applicable risk estimates for FXPOI, which could facilitate counseling. To further develop and individualize risk estimates for FXPOI, a predictor based on endocrine markers, such as AMH, could be included. AMH is an early marker of ovarian reserve and a very promising predictor for menopause and POI in general, and for premutation carriers in particular.42
The decline in ovarian reserve with time in premutation carriers with FXPOI, however, is not fully comprehended yet. Development of reference values of AMH by age for premutation carriers might allow early identification of women at high risk for FXPOI. In addition, examination of different ethnic/racial groups might be important in these prediction models. Studies have shown that the frequency distribution of FMR1 alleles differs among groups,43,44
however, no studies have examined whether the affect of the FMR1 alleles differs by ethnic/racial group. Using our current model based on RIR, mean menopausal age of first degree relatives with the same mutation status and smoking, premutation carriers with a high risk for FXPOI might benefit from monitoring of ovarian reserve regularly and consider vitrification, i.e., cryopreservation of oocytes.