Does the effectiveness of IUI in couples with unexplained subfertility depend on their prognosis of natural conception? A replication of the H2Oil study.

Couples who have been trying to conceive for at least 12 months and whose fertility workup fails to reveal any abnormalities are considered to have unexplained subfertility ( Aboulghar et al. , 2009 ; Brandes et al. , 2010 ). In several countries, IUI used as first-line treatment in these couples, especially in combination with ovarian stimulation (OS), since IUI is less invasive and less costly than IVF ( Tjon-Kon-Fat et al. , 2015 ), despite the lack of evidence from randomized controlled trials (RCTs) regarding the effectiveness of IUI-OS ( Wang et al. , 2019 ). The two trials that compared IUI-OS to expectant management used different thresholds for the prognosis of natural conception as inclusion criteria ( Steures et al. , 2006 , Farquhar et al. , 2018 ). In women with an intermediate prognosis to conceive naturally, i.e. an estimated probability between 30% and 40% to conceive within 12 months leading to live birth, IUI-OS was no more effective than expectant management ( Steures et al. , 2006 ). In women with a poor prognosis, i.e. <30% over 12 months, IUI-OS did result in more live births than expectant management ( Farquhar et al. , 2018 ). In a previous study, we found that the different outcomes of these two trials might be explained by the difference in the prognosis of natural conception ( van Eekelen et al. , 2019 ). In a Dutch cohort of 1896 couples, we found that couples with lower prognoses of natural conception had more benefit from IUI-OS in terms of a relative and absolute difference in the chance of conception compared to expectant management. Due to sample size limitations, it was difficult to identify a fixed threshold for prognoses at which point IUI-OS becomes effective. For a prognosis below 25% over 1 year, IUI-OS seemed effective, leading to higher chances of ongoing pregnancy in 6 months compared to expectant management. For higher prognosis than 40% over 1 year, IUI-OS was not effective and led to similar chances of ongoing pregnancy compared to expectant management. Between these thresholds, it was uncertain whether IUI-OS was effective. Replication of research findings is an essential part of medical research as many findings unfortunately cannot be reproduced in further studies ( Ioannidis, 2005 ). Addressing the same research question with different methodological approaches, such as trial and observational data, provides more evidence that a single result is not a chance finding ( Lawlor et al. , 2016 ; Munafò and Smith, 2018 ). More knowledge on who does and who does not benefit from IUI-OS can guide clinical practice and inform evidence-based shared decision-making on when to start treatment. Because these thresholds hold great importance to patients, they should be based on solid evidence. The aim of this study was to replicate the previous result (i.e. that the benefit of IUI-OS compared to expectant management for couples with unexplained subfertility depends on the prognosis of natural conception) in an independent data source derived from an RCT on contrast fluid used for hysterosalpingography (HSG). This is a different approach to see if the effectiveness of IUI-OS here also depends on the prognosis of natural conception, as well as to validate the thresholds of 25% and 40%.

The H2Oil study (NTR 3270) was an investigator-initiated study that was funded by the two academic institutions (AMC and VUmc) of the Amsterdam UMC. The follow-up study (NTR 6577) was also an investigator-initiated study with funding by Guerbet, France. The funders had no role in study design, collection, analysis and interpretation of the data. B.W.M. is supported by an Investigator grant (GNT1176437) from the Australian National Health and Medical Research Council (NHMRC).

Out of 1119 couples included in the H2Oil trial, we selected 975 for analysis after excluding couples with other diagnoses than unexplained subfertility, couples who conceived before HSG and couples with missing outcome data. Of these 975, 587 couples (60%) received 2386 IUI cycles after HSG, of whom 221 couples conceived after IUI leading to ongoing pregnancy (rate: 0.74 per couple per year over a median follow-up for IUI of 5 months). Out of 388 couples (40%) followed up for 18 months of expectant management after HSG, 299 conceived naturally leading to ongoing pregnancy (rate: 0.56 per couple per year over a median follow-up of 4 months). In total, 62 couples out of 587 (11%) who underwent IUI-OS started treatment directly after HSG and the remaining 525 (89%) first had a period of expectant management. The median period between HSG and starting IUI-OS was 4 months. A total of 1723 (72%) IUI cycles used OS. Forty-two couples (4%) received IVF as their first treatment, with a median period of expectant management of 9 months between completion of the fertility workup and the start of IVF. We depicted the number of couples followed under expectant management or followed under IUI-OS over time in Fig. 1 . Until approximately 6 months of follow-up, the number of couples who were currently in an IUI-OS treatment pathway kept increasing, after which this number declined again. Number of couples who, over follow-up, are currently on expectant management or receiving IUI-OS. IUI-OS, IUI with ovarian stimulation. The baseline characteristics for couples who eventually received at least one cycle of IUI-OS within 18 months after HSG or who remained untreated are summarized in Table I . Treated couples more often had primary subfertility (73% vs 60%) compared to couples that were not treated. Female age, median duration of subfertility, total motile sperm count, one-sided tubal pathology and referral status were similar between groups. Baseline characteristics of 975 patients just before receiving hysterosalpingography. Data are mean with the 5th–95th percentile in brackets (unless median is specified) or n (%). OS, ovarian stimulation. Assessed by hysterosalpingography or in addition, a later laparoscopy. In the 13 weighted mimicked trial datasets, the standardized mean differences between treated and untreated couples were below 0.10 for all characteristics, indicating that the two groups were well balanced in terms of prognostic factors after weighting. The mean weight used in the pooled dataset was 1.00, indicating that weights are stable and do not artificially inflate sample size. Starting IUI-OS was associated with increased ongoing pregnancy rates compared to expectant management, as shown by an estimated hazard ratio of 1.50 (95% CI: 1.19–1.89), pooling all 13 mimicked, weighted trial datasets running over 6 months. The predicted probability that a couple would conceive over the course of 6 months of expectant management after HSG was 29% (95% CI: 25–32%). If the couple started IUI-OS directly after HSG, their estimated probability of conception in the next 6 months was 40% (32–47%). Judging by a decrease of at least 2 points in terms of AIC, the relative effect of IUI-OS did not depend on the prognosis of natural conception (AIC decreased by 1 point, P  = 0.17). The relative effect of IUI-OS was dependent on how long after the HSG treatment it was started (AIC decreased by 2.8 points, P  = 0.08). The relations between prognosis, the start of treatment and the estimated treatment effect are visualized in Fig. 2 . The figure shows the 6-month cumulative probabilities of conception with and without starting IUI-OS for three different example couples with a prognosis to conceive naturally at completion of the fertility workup over the next year of 40% ( Fig. 2 , left panel), 32% ( Fig. 2 , middle panel) or 25% ( Fig. 2 , right panel), which were updated over time when these couples fail consecutive natural cycles and start treatment later. The association between the predicted prognosis of natural conception and the estimated benefit of starting IUI-OS at different time points. This association is shown as cumulative probabilities over 6 months ( y -axis) when starting IUI-OS, or not, at different time points after completion of the fertility workup ( x -axis) for three example couples that have three different prognoses at time of hysterosalpingography (HSG): 40% (left), 32% (middle) or 25% (right). The prognosis was calculated over 1 year and updated after additional failed natural cycles. Grey bands represent 95% CIs. Left panel: Couple A is referred by their general practitioner, where the female partner is nulliparous and 32 years old, the couple has 1 year of subfertility at the time of completion of the fertility workup and the semen analysis showed 37% progressively motile sperm. Middle panel: Couple B has the same characteristics as Couple A except for 2 year duration of subfertility at the completion of the fertility workup. Right panel: Couple C has the same characteristics as Couple A but for 3.5 year duration of subfertility. In Fig. 2 , the absolute chance to conceive over 6 months decreased over time for expectant management, but not for IUI-OS, of which the absolute chance seemed much less variable between couples and timing of treatment start, at around 37%, 34% and 30% for the three couples. However, CIs were wide, especially for pregnancy chances after IUI-OS. The decrease in chances for expectant management over time led to a larger treatment benefit as IUI-OS was started later. It follows from Fig. 2 that the prognosis for a couple does not have a large influence on the expected benefit in terms of the absolute difference between the IUI-OS and expectant management line, as there was always a difference of 5% or more. This was different in the previous study, in which the benefit of IUI-OS was dependent on prognosis. However, the later that treatment was started, the larger the expected benefit of treatment. This was the same finding as in the previous study.

K.D. reports receiving travel and speaker fees from Guerbet. B.W.M. reports consultancy for ObsEva, Merck, Merck KGaA, iGenomix and Guerbet. V.M. reports receiving travel- and speaker fees as well as research grants from Guerbet.

The H2Oil study was a multicentre RCT to compare ongoing pregnancy rates in subfertile women who underwent HSG with oil-based versus water-based contrast (Netherlands Trial Register number, NTR3270). The study was approved by the Institutional Review Board of the Amsterdam University Medical Centre-Academic Medical Centre (reference 2008.362, dated 12 February 2009). The study details and results have been published previously ( Dreyer et al. , 2017 ). The H2Oil follow-up study (NTR 6577) assessed the long-term outcomes until 3–5 years after the H2Oil study and was approved by the Institutional Review Board of the Amsterdam University Medical Centre, location VU University Medical Centre (reference 2017.221, dated 14 June 2017). Study details and results have been published elsewhere ( van Rijswijk et al. , 2020 ). Included women were between 18 and 39 years of age, were having ovulatory cycles and had a low perceived risk of tubal pathology based on their medical history. They had tried to conceive unsuccessfully for at least 1 year and had an indication for tubal patency testing. Exclusion criteria were known endocrine disorders and having a partner with severe male subfertility (defined as a total motile sperm count after sperm wash of less than 3 million per millilitre). Women were randomized for an HSG with oil-based contrast or an HSG with water-based contrast. Data regarding fertility treatments and pregnancies were collected until 3–5 years after randomization. For the follow-up of selected couples, we distinguished between time spent pursuing expectant management and time spent receiving IUI-OS cycles. The start of the IUI period was defined as the first day of menstruation before the first IUI cycle. The end of the IUI period was defined as the first day of menstruation before the last IUI cycle. All pregnancies in the IUI period thus resulted from IUI. Follow-up for expectant management started 14 days before they received HSG and ended at the last date of contact, first day of last menstruation before starting IUI or IVF or, in case they conceived naturally, the first day of the last menstruation before conceiving. The endpoint was ongoing pregnancy, defined as the presence of foetal cardiac activity at transvaginal sonography at a gestational age of at least 12 weeks ( Dreyer et al. , 2017 ). Couples who miscarried before 12 weeks were not censored since they could still achieve ongoing pregnancy in subsequent cycles after their miscarriage. If no ongoing pregnancy occurred, we censored follow-up at the end of expectant management or, if treated, at the end of the IUI period. We used the same statistical approach as in our previous study ( van Eekelen et al. , 2019 ). In short, we used the sequential Cox approach to compare multiple cycles of treated and untreated couples, not only directly after completion of the fertility workup but also if they started later ( Gran et al. , 2010 ). In this approach, we derived multiple datasets from the cohort in which couples started IUI-OS at approximately the same point in time and compared them to couples undergoing expectant management at that time, ‘mimicking’ hypothetical RCTs ( Gran et al. , 2010 ). At completion of the fertility workup and each consecutive month thereafter, named the landmark time points, we constructed such a mimicked trial from our data in which we included all couples who remained in the cohort, i.e. couples who had not conceived, had not started treatment and were not lost to follow-up before that landmark time point. In these ‘trial’ datasets spanning 6 months, we considered couples as treated if they started IUI-OS early, i.e. within 1 month after the landmark time point. Couples who did not start IUI-OS within the first month were used as controls. Couples who started IUI-OS within the 6 months window of a trial, but later than 1 month after the landmark time point, were counted as controls during their untreated period and ‘artificially censored’ at the time of starting IUI-OS. This way, couples were not included in a single group throughout the study. Instead, couples who at some point started IUI-OS were analysed as controls (under expectant management) in the ‘mimicked’ trial datasets preceding the month in which they started IUI-OS. When they started IUI-OS, their following treatment cycles were analysed as part of the treated (IUI-OS) group in the mimicked trial dataset that started that month. In order to compare results to the previous study, we restricted our data to a maximum of 18 months of follow-up. In our data, couples were not randomized to either expectant management or IUI-OS. Thus, patients starting IUI could differ from those who did not in terms of important predictors of conception such as female age or duration of subfertility. In order to achieve groups that are on average similar, we opted for a statistical technique called applied iterative inverse probability weighting ( Austin, 2011 ; van der Wal, 2011 ; Austin and Stuart, 2015 ). By reweighting patients’ contribution to the data, these characteristics are balanced. Details on how we derived the weights to adjust for these differences are given in the Supplementary Data . We chose to balance for the same patient characteristics as in the previous study with the exception of fertility clinic, as that would lead to very unbalanced weights: female age, duration of subfertility, primary or secondary subfertility, total motile sperm count, referral status and the presence of one-sided tubal pathology ( Hunault et al. , 2004 ; van Eekelen et al. , 2017a ). We calculated the mean weight to assess potential inflation of the effective sample size induced by the weighting, which is ideally around 1 ( Cole and Hernan, 2008 ). We assessed the degree of balance in patient characteristics before and after weighting using the standardized mean difference between the treated and untreated group in each of the mimicked trial dataset. A lower standardized mean difference between groups represents better balance and a value below 0.10 generally indicates no important difference ( Austin, 2011 ; Austin and Stuart, 2015 ). We analysed the weighted mimicked trial datasets using a pooled Cox proportional hazards model with IUI-OS or expectant management as a treatment covariate. We calculated an overall hazard ratio by stratifying on the 13 mimicked trials. We used a robust sandwich variance estimator to adjust precision measures since couples can be included in multiple mimicked trial datasets ( Wei et al. , 1989 ). To address whether the effect of starting IUI-OS depends on the decreasing prognosis of natural conception of the individual couple, we added the prognosis and a treatment-by-prognosis interaction term to the model. We calculated a time-updated prognosis of natural conception over the next six cycles at the start of each mimicked trial dataset by using an existing dynamic prediction model that comprises female age, duration of subfertility, primary or secondary subfertility, percentage of progressive motile sperm, referral by a general practitioner or specialist, and the unsuccessful number of menstrual cycles since the fertility workup ( van Eekelen et al. , 2017a ). The prognosis for a couple that we used is thus not one fixed value throughout the study but decreases after consecutive failed natural cycles. We transformed the updated prognosis by taking the complementary log–log of its value such that it is linear on the log-hazard scale used by the Cox model ( Prentice and Gloeckler, 1978 ). We included the complementary log–log of this updated prognosis as a main effect, the main effect for treatment and the treatment-by-prognosis interaction effect in the pooled Cox model. The weighting procedure was adjusted slightly for this analysis because the difference in prognosis between groups was adjusted for by adding it to the model as a main effect ( VanderWeele, 2009 ) (see also the Supplementary Data ). For three hypothetical couples, we visually depicted the relationship between their worsening prognoses and the accompanying 6-month cumulative predicted probability of conception following expectant management or starting IUI-OS, as treatment is initiated later. The first example is a couple referred by their general practitioner, where the female partner is nulliparous and 32 years old, the couple has 1 year of subfertility at the time of completion of the fertility workup and the semen analysis showed 37% progressively motile sperm. In this case, the estimated prognosis of natural conception over the first six cycles is 25%. A second couple with the same characteristics except for a 2-year duration of subfertility at the completion of the fertility workup has a prognosis of 20%, while a third couple with the same characteristics but for a 3.5-year duration of subfertility has a prognosis of 15%. At the time of the completion of their fertility workup, these couples have prognoses of 25%, 20% and 15%, respectively over six cycles, which translates to approximately 40%, 32% and 25%, respectively over 13 cycles i.e. 1 year ( van Eekelen et al. , 2017a ). The chances of natural conception for these three hypothetical couples decrease over time based on the number of unsuccessful menstrual cycles between the diagnosis/HSG and the start of a landmark. Estimated cumulative probabilities of ongoing pregnancy from this model are derived from the separate mimicked trials that all have different observed conception rates, thus although predictions are expected to decrease over time, our estimates may fluctuate. We considered an absolute difference of more than five percentage points between estimates of the cumulative ongoing pregnancy rates, estimated at completion of the fertility workup, to indicate a benefit of IUI-OS. In addition to modelling the impact of prognosis and consecutive failed natural cycles on the effect of treatment, we assessed if the effect of IUI-OS depends on the time of initiation of treatment by adding an additional interaction between treatment and landmark time point to the pooled Cox model already including treatment, prognosis and the treatment-by-prognosis interaction. If the interaction between prognosis and treatment yielded a better fit, we also added a three-way interaction between treatment, prognosis and landmark time point to the previous model to see if the effect modification of prognosis on IUI-OS changed over mimicked trials i.e. when starting treatment later. We used Akaike’s Information Criterion (AIC) (at least two points difference) and Wald tests for the interaction terms to determine whether including the interactions resulted in a better fit of the model to the data ( Akaike, 1974 ). Data were missing on duration of subfertility (n = 3), referral status (n = 2), primary or secondary subfertility (n = 1), the percentage of progressive motile sperm (n = 81) and total motile sperm count (n = 93) and was accounted for using single imputation. All statistical analyses were performed using R version 3.3.2 ( R Core Team (2017) . http://www.R-project.org/ ) using the survival , dynpred , mice and CreateTableOne packages.

N.v.G., B.W.M., V.M., M.v.W. and R.v.E. conceived the study. N.v.W., K.D. and K.R. collected and cleaned the data. R.v.E., N.v.G. and M.J.E. wrote the statistical analysis plan. R.v.E. conducted the statistical analyses. R.v.E., N.v.G., N.v.W. and K.R. drafted the manuscript. All authors contributed critical revision to the paper and approved the final manuscript.

We replicated the finding that on average, IUI-OS increases the chance of an ongoing pregnancy compared to expectant management and that when IUI-OS is started later, the expected benefit in terms of the absolute and relative difference with expectant management was larger. We did not replicate the finding that the benefit of IUI-OS depends on the prognosis of natural conception. Couples with unexplained subfertility still have good chances of natural conception at the time of diagnosis, and treatment is thus not always necessary. Clinicians should counsel couples on the option to prolong expectant management before commencing with IUI-OS.

We replicated the finding that in couples with unexplained subfertility, starting IUI-OS within 18 months after completion of the fertility workup was associated with increased ongoing pregnancy rates over 6 months compared to expectant management. However, the estimated benefit of treatment did not depend on the prognosis of natural conception but did depend on when treatment was started after diagnosis. We replicated the finding that starting IUI-OS later yields a larger absolute and relative benefit of treatment. The main strength of this study was the use of trial data from the H2Oil study with a follow-up of 3–5 years, low loss-to-follow-up and few missing data. For the purpose of triangulation, i.e. the use of multiple approaches to address the same question ( Lawlor et al. , 2016 ; Munafò and Smith, 2018 ), we now have data from two RCTs comparing IUI-OS to expectant management in different patient selection, cohort data and data from an RCT in which couples could receive IUI-OS during follow-up. Workup and treatment protocols differed between these three data sources. Via triangulation, our confidence in two findings has been strengthened: namely that starting IUI-OS later yields a larger absolute and relative benefit of treatment and that the absolute chance of an ongoing pregnancy after IUI-OS is less variable between couples than the chance of natural conception. We did not find significant evidence that the effect of IUI-OS depends on the prognosis for natural conception of a couple. Weaknesses are that the H2Oil trial was not designed with this secondary question in mind. In addition, the sample size was moderate, which led to wide CIs and perhaps limited the power to show an interaction with prognosis. We observed that the chances of an ongoing pregnancy after 6 months of expectant management in the present study were much higher than what was found in previous studies, at 29% instead of approximately 18% ( Hunault et al. , 2004 ; van der Steeg et al. , 2007 ). This could be due to the fact that in the H2Oil study, all couples received HSG during the diagnostic work-up, which might increase their chances, especially when using oil-based contrast medium ( Dreyer et al. , 2017 , 2019 ). The pooled, i.e. overall effect of IUI-OS versus expectant management was less strong in the current study, with a point estimate for the hazard ratio of 1.50 compared to 1.96 that we found previously. The interaction that showed the dependency of the effect of IUI-OS on the prognosis for natural conception was in the same direction as the previous study, with lower prognoses having more benefit of IUI-OS, but this did not reach statistical significance in terms of AIC or P -value. It could be that we lacked the power, owing to a less strong main treatment effect and sample size restrictions, to show this dependency as we did in the previous study, which had a larger sample size of 1896 with 800 couples treated with IUI ( van Eekelen et al. , 2019 ). It remains unknown why couples with better prognoses would benefit less from IUI-OS. A possible mechanism is that there is a ‘ceiling’ to the chance of conception for subfertile couples in terms of a maximum and that some unexplained subfertile couples with good prognoses remain around this ceiling whereas couples with different indications are further below their fertility potential, which is increased by IUI-OS ( Moreau et al. , 2019 ). Not all couples who received IUI-OS conceived or continued IUI-OS over the 6 months follow-up of a ‘mimicked’ trial dataset. In the previous study, there was no follow-up of natural conception after IUI-OS dropout. We repeated this approach in the current study. In contrast to the previous study, we found that the effect of IUI-OS depended on when treatment was started independent of the decreasing prognosis of natural conception after failed cycles. This might be due to additional selection over time that is not explained by the dynamic prediction model ( van Eekelen et al. , 2017a , b , 2019 ). We found that when accounting for this time effect, the absolute chances after IUI-OS were much more stable over time than chances after expectant management, as the latter clearly decreased over time whereas the former did not. This provides more evidence that the chance after IUI-OS is less dependent on individual factors, i.e. that couples’ chances become more similar when receiving IUI-OS. This can be important for counselling couples, as it suggests that further expectant management will not come at a great loss in terms of a decreased chance of pregnancy when receiving IUI-OS later.

Supplementary data are available at Human Reproduction Open online.

Click here for additional data file.