“Women hoping to have a baby through fertility treatment can from today use an online calculator to show them how likely they are to succeed,” reported The Guardian today. The Mirror called it the ‘world’s most accurate IVF test’.
The newspapers reported on a study in which researchers made refinements to an existing prediction model for determining the chance of success with IVF. Using data from more than 144,000 IVF cycles performed in the UK, they found several additional factors for predicting successful IVF.
Early tests using the data with which the prediction model was created, found that it was an improvement. However, it now needs to be tested in an independent sample before its true accuracy will be known. To do this, the researchers have developed an online version and a Smartphone app through which they can collect real-life data.
Importantly, the tool is designed for people who have already had fertility treatment and who have had the cause of their infertility investigated. People who have been unsuccessfully trying for a baby, but have not yet sought medical help, would not have the necessary information to complete the tool in its current state.
The study was carried out by researchers from the University of Glasgow and the University of Bristol. Funding was provided by the UK Medical Research Council and the University of Bristol. The study was published in the peer-reviewed journal PLoS Medicine.
Many of the news reports are premature in announcing the accuracy of this tool as it is still in development and its validity has not yet been established. It is too early to call this the ‘world’s most accurate IVF test’, as the Daily Mirror has done.
IVF is reportedly successful in about a third of women under the age of 35 and in 5-10% of women over 40 years old. This research was aimed at developing a model that could predict the chances of a successful pregnancy according to a list of contributing factors. The researchers say that such a tool could support patient counselling, decision making and the allocation of resources.
The data was obtained from the Human Fertilisation and Embryology Authority (HFEA), which regulates IVF treatment in the UK. The data included information on couples who had been given IVF, the number of cycles they had, particular treatments, subsequent pregnancy-related factors and complications, and live birth rates.
The aim was to construct a model that could identify couple- and treatment-specific factors, giving couples a reliable indication of their likelihood of success with IVF. There is already a model used to predict success that was developed by Templeton and colleagues in 1996. In this study, the researchers wanted to refine that model so it could predict live births by including a number of extra infertility characteristics and treatment-related factors, which the earlier model had not been able to consider.
The study used all treatment cycles and outcomes registered on the HFEA database between 2003 and 2007. During this time, 163,425 IVF cycles were completed in the UK, with 23.4% resulting in at least one live birth. A total of 144,018 of these cycles (88%) had full data available to enable the researchers to develop a type of statistical prediction model for the likely outcomes of IVF.
Information put into the model included the duration and cause of infertility, prior IVF attempts, prior successes and live births, and numerous maternal characteristics. The main outcome of success in the prediction model was considered to be at least one live birth that survived to at least one month.
As secondary outcomes, the researchers also considered the likelihood of other complications related to the pregnancy or birth such as prematurity and low birthweight.
This new model differs from the previous Templeton model in that it includes four additional details: the source of the egg (donor or patient’s own), the type of hormonal preparation used, whether or not intracytoplasmic sperm injection was used (ICSI, injecting a single sperm directly into the egg cell), and how many IVF cycles had been tried previously.
In particular, the consideration of intracytoplasmic sperm injection (ICSI) in the newer model is an important factor. The Templeton model used HFEA data obtained between 1991 and 1994 prior to the introduction of ICSI. As ICSI has now been widely adopted for male factor infertility, inclusion of this would be expected to improve the accuracy of their prediction model for today’s couples.
The newer model also considers wider causes of infertility. The Templeton model classified infertility into only two categories: tubal causes and all other causes.
The data indicated there was less chance of a live birth with the following factors:
There was a higher chance of success if the woman had had a previous live birth resulting from IVF or if ICSI was used.
Regarding secondary outcomes, the model predicted that there was an increased chance of a premature birth or low birth weight baby if donor eggs were used, or if ICSI was not used. Older mothers had a greater risk of a baby with excessive birthweight (macrosomia) as did women who had previous live births. Prematurity and a low or high birth weight baby were all associated with infertility that was the result of cervical problems. All of these factors went into refining the existing Templeton model.
When the researchers compared their prediction tool with the Templeton model, they found that theirs was better able to predict the correct number of live births in the sample. However, they say that, in part, this is likely to be because it was tested with the same data that was used to develop the model.
The researchers say their model can estimate the affect that specific factors have on the chances of a baby from IVF, and of having other pregnancy- and birth-related complications.
They say that, pending external validation, the results show that couple- and treatment-related factors can be used to provide infertile couples with an accurate assessment of whether they have low or high chance of success following IVF.
This well conducted study has refined an existing prediction model for likely success with IVF. The new model has strengths in that the refinements were based on patterns in a large amount of data from more than 144,000 IVF cycles performed in the UK.
The model also uses data on the use of intracytoplasmic sperm injection (ICSI) – information that was not available when the previous model was constructed. As ICSI has now been widely adopted for male factor infertility, its inclusion improves the accuracy of their prediction model for IVF in the present day.
The new model appears to be very accurate when retested in the sample from which it was developed. The researchers also acknowledge some limitations to their model, including that they could not use data from 12% of their available cohort.
The researchers openly acknowledge that the tool is still in development, and before it could be used to guide clinical decisions and to counsel patients, it needs to be validated outside of this test population. Therefore, the researchers have made a free web-based prediction tool and iPhone app (IVFpredict). They say that users will be told about the current lack of external validation and will be asked if their anonymous data can be used to test the model. As such the newspapers, in particular the Daily Mirror , have been premature in hailing this as the ‘world’s most accurate IVF test’.
The tool is aimed at people who are already seeking fertility treatment (and may or may not have already undergone previous IVF cycles) and who have had the cause of their infertility investigated. People who have been unsuccessfully trying for a baby, but who have not yet sought medical help, would not have the necessary information to complete the tool in its current state. The model is not designed with the aim of helping people decide whether or not they should consult medical help for fertility problems.
The tool could potentially aid the discussion and decision process, but is still likely to be best used alongside medical care and advice, and not purely as an online substitute.