Kimiko A. Broeze

Although ovarian reserve tests (ORTs) are frequently used prior to IVF treatment for outcome prediction, their added predictive value is unclear. We assessed the added value of ORTs to patient characteristics in the prediction of IVF outcome. An individual patient data (IPD) meta-analysis from published studies was performed. Studies on FSH, anti-Müllerian hormone (AMH) or antral follicle count (AFC) in women undergoing IVF were identified and authors were contacted. Using random intercept logistic regression models, we estimated the added predictive value of ORTs for poor response and ongoing pregnancy after IVF, relative to patient characteristics. We were able to collect 28 study databases, comprising 5705 women undergoing IVF. The area under the receiver-operating characteristic curve (AUC) for female age in predicting poor response was 0.61. AFC and AMH each significantly improved the model fit (P-value P = 0.19) of poor response. In predicting ongoing pregnancy after IVF, age was the best single predictor (AUC 0.57), and none of the ORTs added any value. This IPD meta-analysis demonstrates that AFC and AMH clearly add to age in predicting poor response. As single tests, AFC and AMH both fully cover the prediction of poor ovarian response. In contrast, none of the ORTs add any information to the limited capacity of female age to predict ongoing pregnancy after IVF. The clinical usefulness of ORTs prior to IVF will be limited to the prediction of ovarian response.

Tubal patency tests are routinely performed in the diagnostic work-up of subfertile patients, but it is unknown whether these diagnostic tests add value beyond the information obtained by medical history taking and findings at physical examination. We used individual patient data meta-analysis to assess this question. We approached authors of primary studies for data sets containing information on patient characteristics and results from tubal patency tests, such as Chlamydia antibody test (CAT), hysterosalpingography (HSG) and laparoscopy. We used logistic regression to create models that predict tubal pathology from medical history and physical examination alone, as well as models in which the results of tubal patency tests are integrated in the patient characteristics model. Laparoscopy was considered to be the reference test. We obtained data from four studies reporting on 4883 women. The duration of subfertility, number of previous pregnancies and a history of previous pelvic inflammatory disease (PID), pelvic surgery or Chlamydia infection qualified for the patient characteristics model. This model showed an area under the receiver operating characteristic curve (AUC) of 0.63 [95% confidence interval (CI) 0.61–0.65]. For any tubal pathology, the addition of HSG significantly improved the predictive performance to an AUC of 0.74 (95% CI 0.73–0.76) (P < 0.001). For bilateral tubal pathology, the addition of both CAT and HSG increased the predictive performance to an AUC of 0.76 (95% CI 0.74–0.79). In the work-up for subfertile couples, the combination of patient characteristics with CAT and HSG results gives the best diagnostic performance for the diagnosis of bilateral tubal pathology.

BACKGROUND The Chlamydia IgG antibody test (CAT) shows considerable variations in reported estimates of test accuracy, partly because of the use of different assays and cut-off values. The aim of this study was to reassess the accuracy of CAT in diagnosing tubal pathology by individual patient data (IPD) meta-analysis for three different CAT assays. METHODS We approached authors of primary studies that used micro-immunofluorescence tests (MIF), immunofluorescence tests (IF) or enzyme-linked immunosorbent assay tests (ELISA). Using the obtained IPD, we performed pooled receiver operator characteristics analysis and logistic regression analysis with a random effects model to compare the three assays. Tubal pathology was defined as either any tubal obstruction or bilateral tubal obstruction. RESULTS We acquired data of 14 primary studies containing data of 6191 women, of which data of 3453 women were available for analysis. The areas under the curve for ELISA, IF and MIF were 0.64, 0.65 and 0.75, respectively (P-value < 0.001) for any tubal pathology and 0.66, 0.66 and 0.77, respectively (P-value = 0.01) for bilateral tubal pathology. CONCLUSIONS In Chlamydia antibody testing, MIF is superior in the assessment of tubal pathology. In the initial screen for tubal pathology MIF should therefore be the test of first choice.

BACKGROUND Conventional meta-analysis has estimated the sensitivity and specificity of hysterosalpingography (HSG) to be 65% and 83%. The impact of patient characteristics on the accuracy of HSG is unknown. The aim of this study was to assess by individual patient data meta-analysis whether the accuracy of HSG is associated with different patient characteristics. METHODS We approached authors of primary studies reporting on the accuracy of HSG using findings at laparoscopy as the reference. We assessed whether patient characteristics such as female age, duration of subfertility and a clinical history without risk factors for tubal pathology were associated with the accuracy of HSG, using a random intercept logistic regression model. RESULTS We acquired data of seven primary studies containing data of 4521 women. Pooled sensitivity and specificity of HSG were 53% and 87% for any tubal pathology and 46% and 95% for bilateral tubal pathology. In women without risk factors, the sensitivity of HSG was 38% for any tubal pathology, compared with 61% in women with risk factors (P = 0.005). For bilateral tubal pathology, these rates were 13% versus 47% (P = 0.01). For bilateral tubal pathology, the sensitivity of HSG decreased with age [factor 0.93 per year (P = 0.05)]. The specificity of HSG was very stable across all subgroups. CONCLUSIONS The accuracy of HSG in detecting tubal pathology was similar in all subgroups, except for women without risk factors in whom sensitivity was lower, possibly due to false-positive results at laparoscopy. HSG is a useful tubal patency screening test for all infertile couples.

Systematic reviews and accompanying meta-analyses are the cornerstones of evidence-based medicine. Systematic reviews summarize clinical evidence; meta-analyses provide summary estimates of the treatment effect or the diagnostic test accuracy. Although deemed to provide the highest level of evidence, their clinical value is limited as they can only summarize aggregated data. In these meta-analyses the true variability of the treatment effects cannot be explored to the desired extent, because the meta-analyses cannot distinguish between patients with different clinical profiles. Systematic reviews and meta-analyses based on individual patient data (IPD), described as the ‘gold standard’ for systematic reviews are a promising approach that might overcome these limitations. IPD meta-analyses allow treatment effects and diagnostic accuracy to be estimated at the level of relevant patient subgroups. This enables researchers to investigate the effectiveness of treatment in patients with different profiles. In this article, we address the opportunities of systematic reviews and meta-analyses using IPD in reproductive medicine. We discuss its potential based on three clinical examples: single versus double embryo transfer in IVF, the diagnosis of tubal pathology and the prognostic value of ovarian reserve tests. We propose to show potential advantages of IPD systematic reviews and meta-analyses in providing stratified clinical evidence, which could improve medical care.

Introduction In IVF cycles, it is common practice to confirm pregnancy through measurement of serum hCG levels 10-12 days after blastocyst transfer or 12-14 days after cleavage stage embryo transfer. It has been shown that early observation of rising serum hCG concentration in the peri-implantation period of spontaneous pregnancies is predictive of pregnancy outcome. The current study seeks to determine if hCG serum concentration measured 5 days after blastocyst transfer is predictive of IVF cycle outcome. Material and Methods Following ovarian stimulation with gonadotrophins and pituitary suppression with GnRH antagonist, oocytes were retrieved and inseminated by intracytoplasmic sperm injection. Embryos were cultured to the blastocyst stage in sequential media, and transferred on the first day that good expanded blastocysts appeared (day 5 or day 6). Serum hCG was measured at 3, 5, 7, and 10 days after blastocyst transfer. Inclusion criteria required that exactly two blastocysts were transferred in non-donor IVF. This criterion included enough multiple pregnancies for meaningful analysis. Pregnancy was defined by serum hCG titers rising above 5 IU/l by 10 days after blastocyst transfer, while intrauterine pregnancy was defined by the ultrasonic observation of a gestational sac in the uterus at approximately 5 weeks. Multiple pregnancy was defined by the presence of multiple fetal cardiac activity at 6-7 weeks. Ongoing pregnancy was defined by the observation of intrauterine fetal heart motion at approximately 10 weeks gestation. Spontaneous abortion (SAB) was defined as an intrauterine pregnancy confirmed on ultrasound at 5-6 weeks that was lost before 10 weeks. Logistic regression was used to assess relationships between nominal outcomes and serum hCG level. Fisher's exact test was used to assess the significance of hCG thresholds for predicting nominal outcome. IRB approval was obtained for this retrospective study. Results There were 767 qualifying fresh autologous blastocyst transfers in the 6-year study period starting 1 January, 2004. Patients averaged 32.7 ± 4.7 years of age, and 69.6 ± 17.2 kg of body weight. Serum hCG level at 5 days following blastocyst transfer was predictive of pregnancy (P < 0.0001). A serum level threshold of 5 IU/l hCG yielded 74% sensitivity and 97% specificity for predicting pregnancy (P < 0.0001, odds ratio (OR) 103.2, 95% CI 44.9 to 237.5). Among cycles achieving pregnancy, day 5 serum hCG level was predictive of ongoing pregnancy (P < 0.0001), ectopic pregnancy (P < 0.0001), and multiple pregnancy (P < 0.0001). A serum level threshold of 5 IU/l hCG yielded 92% sensitivity and 65% specificity for predicting ongoing pregnancy (P < 0.0001, OR 22.1, 95% CI 13.5 to 36.2), 80% sensitivity and 72% specificity for predicting ectopic pregnancy (P < 0.0001, OR 10.1, 95% CI 3.94 to 25.8), and 97% sensitivity and 36% specificity for predicting multiple pregnancy (P < 0.0001, OR 17.7, 95% CI 7.1 to 44.1). Among cycles achieving intrauterine pregnancy, day 5 hCG level was predictive of spontaneous abortion (P < 0.0001). A serum level threshold of 5 IU/l hCG yielded 31% sensitivity and 89% specificity for predicting subsequent SAB (P = 0.0005, OR 4.2, 95% CI 2.0 to 9.0). Conclusions The serum hCG level 5 days after blastocyst transfer is predictive of cycle outcome and of multiple pregnancy. Patients failing to achieve a serum hCG level of at least 5 IU/l are at increased risk of no pregnancy. Those who achieve pregnancy after failing that day 5 criterion are still at increased risk for early pregnancy loss and ectopic pregnancy. Those who achieve intrauterine pregnancy after failing that day 5 criterion have increased risk for SAB. The significant relationships observed in this study indicate that conditions during the peri-implantation period, as reflected in serum hCG levels, are critical to IVF cycle outcome.

In clinical practice a diagnosis is based on a combination of clinical history, physical examination and additional diagnostic tests. At present, studies on diagnostic research often report the accuracy of tests without taking into account the information already known from history and examination. Due to this lack of information, together with variations in design and quality of studies, conventional meta-analyses based on these studies will not show the accuracy of the tests in real practice. By using individual patient data (IPD) to perform meta-analyses, the accuracy of tests can be assessed in relation to other patient characteristics and allows the development or evaluation of diagnostic algorithms for individual patients.