To the Editor:
Since the progesterone levels are related to serum estradiol levels, the use of the P/E ratio was proposed for the diagnosis of early luteinization, because it provided the best indicators in the ROC curve and logistics regression models. The poor response of the isolated determination of progeste-rone in these models is due to the fact that only excessive levels are associated to a detriment in the pregnancy rate, while in the rest of the spectrum both pregnancy rate and progesterone levels are associated positively. By contrast, the P/E ratio maintains a nearly constant negative relationship between its values and the pregnancy rate.
In relation to the pregnancy rate, a high level of the P/E ration identifies both excessive progestero-ne levels and low estradiol levels. Several authors, such as Younis (1) have proven this fact, howe-ver, erroneously interpreted as an association of low ovarian response to hyperestimulation with PL. In fact, the cut-off points suggested normally for progesterone or P/E, even when they provide simi-lar pregnancy rates, are far from identifying the same subjects.
With this objective, we studied 434 consecutive IVF/ICSI cycles with GnRH analogues, conducted in our center during 2009. If the analysis included only those cycles with an appropriate ovarian response, the predictive ability of progesterone > 1.50 or a P/E > 0.50 are similar. Whereas if we analyze the low ovarian response cycles, none of these cut-off points provide significant differences in pregnancy rate. If we analyze the whole sample, the cut-off point is optimized with a progesrone > 1.35 ng/mL (pregnancy rate of 49.1%, vs 22.6%, p= 0.004) or a P/E ration > 0.85 (pregnancy rate 49.0% vs 19.2%, p= 0.003), but while the prevalence with both parameters is similar (7.1% vs 6.0%), the consistency between both is a mere 1.8%.
We fully agree with Martinez (2) that comparison of progesterone values must be made between cycles with similar ovarian response to stimulation, a target that has been sufficiently proven not to meet the use of the P/E ratio (in our case 2.20 vs 1.09 ng/mL estradiol, p<0.001). Strict inclusion criteria, such as Elgindy’s (3,4) prevent us from extrapolating their results into clinical practice. We believe that the appropriate strategy is to find a formula that identifies extremely high progesterone cases for each value of estradiol given the quadratic regression relation established between both values (R2: 0.194, p<0.001), and to this end we suggest the formula P3/E. With a cut-off point of 0.75, P3/E identifies the group whose values of progesterone are above one standard deviation from the average curve of estradiol-progesterone. This cut-off point is much more in line with the actual distribution of both hormones, and it is independent of the degree of ovarian stimulation (2.12 vs 2.23 ng/mL estradiol, p = 0.50). We strongly encourage all groups working in the diagnosis of PL to test this new formula in their data.
Joseandres Guijarro Ponce, M.D.
Rocío Núñez Calonge, B.D.
1. Younis JS, Matilsky M, Radin O, Ben-Ami M. Increased progesterone/estradiol ratio in the late follicular phase could be related to low ovarian reserve in in vitro fertilization-embyo transfer cycles with a long gonadotropin-releasing hormone agonist. Fertil Steril 2001;76:294-9 doi:10.1016/S0015-0282(01)01918-5
2. Martínez F. Progesterone level and progesterone/estradiol ratio on the day of hCG administration: detrimental cutoff levels and new treatment strategy?. Fertil Steril 2011; doi:10.1016/j.fertnstert.2011.04.013
3. Elgindy EA. Progesterone level and progesterone/estradiol ratio on the day of hCG administra-tion: detrimental cutoff levels and new treatment strategy. Fertil Steril 2011;95:1639–44. doi:10.1016/j.fertnstert.2010.12.065
4. Elgindy EA. Reply of the Author. Fertil Steril 2011; doi:10.1016/j.fertnstert.2011.04.014
Published online in Fertility and Sterility doi:10.1016/j.fertnstert.2011.05.066
The Authors Respond:
We thank Dr. Guijarro Ponce for interest in our article (1). However, there are some major differences between our study and their study that should be highlighted and may provide potential reasons for discrepancies between the results:
1) Our study was prospective in nature while theirs is retrospective, which could bring multiple potential biases. For example, there is no detail in their presentation of the results about baseline characteristics, or other confounders that could skew results. Potential confounders not reported include type of analogue, ovarian reserve, ICSI cycle characteristics and day of ET. If their patient pool is significantly different from ours then why should we expect the results to be similar?
2) Dr Guijarro Ponce applied our cutoffs on patients with low ovarian response cycles and reported that none of these cutoffs provide significant differences in pregnancy rates. Importantly, Venetis et al. (2), in a systematic review and meta-analysis have suggested that, the use of ROC curve is a more appropriate method to analyze any possible detrimental effect. Moreover, they emphasized the importance of including a homogenous group of patients with comparable ovarian responses. Actually, premature progesterone rise in women with poor ovarian response is different from its rise in women with good response. So, their suggested cutoff for the whole sample with variable responses appears inappropriate.
According to their results, when they predicted response by P4 >1.35 and E/P ratio >0.85, both gave nearly the same pregnancy rates but with varying values within pregnant women and non-pregnant women. This means low validity of the predictor. Although consistency analysis is a considerable tool to validate diagnostic/prognostic markers, the authors used all their cases, not good responders only, which make application to our markers unsuitable.
3) Dr. Guijarro Ponce suggested that our study’s strict inclusion criteria prevent extrapolating the results into clinical practice. Actually, this is an issue of generalizability and does not affect the core of research. Results of this efficacy study now need to be tested in an effectiveness study design with more lax inclusion criteria. Even so, lack of generalizability is not a risk of bias or methodological weakness and most high-quality evidence comes from rigorous study designs that use strict inclusion criteria (e.g., RCTs).
4) As for the use of quadratic regression, this relates back to the data type and the relationship between the studied variables. We used the best available model in relation to our data. To judge, we must 1st know the details of their data and to cross validate their equation.
Eman Elgindy, M.D.
Zagazig University School of Medicine
Al-Banoon Fertility Center
1. Elgindy EA. Progesterone level and progesterone/estradiol ratio on the day of hCG administration:detrimental cutoff levels and new treatment strategy. Fertil Steril 2011;95:1639–44.
2. Venetis CA, Kolibianakis EM, Papanikolaou E, Bontis J, Devroey P, Tarlatzis BC. Is progesterone elevation on the day of human chorionic gonadotrophin administration associated with the probability of pregnancy in in vitro fertilization? A systematic review and meta-analysis. Hum Reprod Update 2007;13:343–55.
Published online in Fertility and Sterility doi:10.1016/j.fertnstert.2011.05.065