Celmatix, a New York-based biotechnology firm, will launch a randomized prospective clinical trial this month to test Polaris, a web-based clinical decision support system that the company developed to help physicians predict women's potential of getting pregnant with or without the use of fertility treatments such as in vitro fertilization and make decisions about appropriate courses of action.
During the year-long trial, clinics and hospitals in the company's partner network will test Polaris on hundreds of patients, Piraye Yurttas Beim, the company’s founder and CEO, told BioInform this week. The list of testing facilities includes the Reproductive Medicine Associates (RMA) of New York, RMA of Texas, RMA of Philadelphia, the Center for Advanced Reproductive Services at University of Connecticut, and others.
Celmatix has been working on a non-invasive diagnostic assay for female infertility that takes into account genetics and the influences of phenotypic factors such as hormone levels and age. In addition to analyzing clinical data collected from patients with unexplained infertility, the company has also been analyzing whole-genome sequence data to find biomarkers that would be predictive of a patient's success with various infertility treatment options.
"For a complex disorder like infertility, there are many reasons why someone might or might not succeed and only a subset of those will be genetic" and "we wanted to make sure that … we were really doing a good job upfront in understanding the heterogeneity," Beim said. That meant looking at various metrics, for example how many cycles of failed IVF treatments a patient would need to undergo before they and their physician could be reasonably sure that the treatment would not work for them and when they might start to consider a genetic root.
"In the spirit of doing a better, more cost-effective biomarker discovery, … what we did was build a data science team and start to ask those [sorts of] questions," Beim said. In examining the number of failed IVF treatments metric, for instance, "[we found] that there's not a one-size solution for everybody … but rather based on their personal metrics, that timeline looked different for everybody," she said.
Some studies suggest that patients who've had between two to six cycles of failed IVF treatments — calculated based on "different averages and theoretical assumptions" — are not likely get pregnant using this method. "But that’s a pretty wide distribution for someone doing biomarker discovery," according to Beim. "So we wanted to create personalized algorithms that would, on a personalized level, help us determine when somebody has entered into the 'unlikely to get pregnant' [category] and then if they have tried that many cycles and haven’t gotten pregnant, then they are a very interesting candidate for biomarker discovery."
The company had developed and was using these algorithms to create its infertility diagnostic when it began receiving requests from physicians in its partner hospitals for a tool that would enable them use clinical information they were collecting from their patients to provide more personalized treatment recommendations to maximize their chances of becoming pregnant.
This is especially important after a particular round of treatment fails. In most states in the country, insurance companies don’t cover the cost of infertility treatments and patients usually pay for their care with personal funds. When a treatment doesn’t result in a viable pregnancy, physicians and patients have to decide whether or not to try again, a decision that is both emotionally and financially dear — fertility treatments can run north of $12,000 per round depending on the state, according to Beim — as well as whether or not to continue the same type of treatment or to try something else.
Furthermore, without any guidance about their chances of success should they continue treatment, many patients drop out of treatment much earlier than they should, Beim said. In its studies, Celmatix has found that one out of four people or 25 percent of patients discontinue treatment while they still have a high chance of success. The immediate assumption is usually that the high cost of treatment is the primary deciding factor, but according to Beim, patients typically report "a lack of clarity of whether there is any point in continuing" as their main reason for choosing to stop treatment.
Yet in a number of cases, Celmatix found that continued treatment would have ultimately paid off, Beim said. In one study which Celmatix presented at a meeting of the American Society for Reproductive Medicine, the company's algorithms predicted that in a cohort of 100 patients, if all of the patients had done one more month of treatment after a failed round, 40 percent of them would have had a baby; and if they'd all done two more months of treatment, 60 percent of them would have achieved a viable pregnancy, she said.
With the genetic assay still some years from completion, "we realized that there is a huge potential here to make an [early] impact on the field … with informatics tools that help doctors and patients interact with this personalized information," Beim said. And that’s how Polaris was born. Underlying the platform are a series of machine learning algorithms that look for patterns in data that they can use to bin people into categories based on particular outcomes.
Named for the North Star, which helps sailors navigate, Polaris similarly helps guide treatment decisions. For example, instead of telling a patient aged 35 that she has a 40 percent chance of getting pregnant, a physician could say based on Polaris analysis of the patient's metrics, "we know that for every 100 couples that we see in the clinic like you, this is what their treatment journey ends up looking like," Beim said. "In the first month, 40 out of 100 have achieved a viable pregnancy. The next month, of that original hundred, 60 have achieved a viable pregnancy" and so on.
An added benefit of the software is that it shifts the focus away from reporting the success rates of things like IVF treatment based on just the age of the patient and the number of cycles, Beim said. "Its important to move away from these age-based, cycle-level statistics because there is richer information available through electronic medical records and with the right analytics applied to them, you can actually give patients a much more personalized success rate than they would get from just their age alone" or how many treatment cycles other patients like them went through.
With this extra information, Celmatix can make more accurate success rate calculations and it can do so over time, she said. For example, "we are able to tell patients that … you have an 80 percent chance of getting pregnant with IVF but you may have to try it three or four times to maximize your personal chance of success with this particular method."
Furthermore, Celmatix algorithms are able to predict the effect of multiple types of treatment. It lets doctors look at cases where individuals with similar profiles to their patients tried methods other than IVF and what their success rates where. For example, a doctor might find that in a cohort of 100 patients similar to his or her own, 18 couples who tried timed intercourse over several months eventually had a viable pregnancy. However, the data might also show doing a fertility treatment such as IVF would double the patient's chance of success. Other uses of the system could be to look at success rates for similar patients who used their own eggs for IVF versus a donor egg, Beim said. Also couples planning to have children later in life can look for similar patients who had successful pregnancies at their target age. This ability to provide a broad suite of treatment options other than IVF is something Celmatix believes sets it apart from competing systems in the space.
In developing the software, Celmatix "started off with over 200 variables that are normally collected during the course of treatment," Beim said. That included self-reported information such as age; information collected by the hospital such as height and weight; medical history including information on how long they've been trying to conceive; and the results of diagnostic workups such as information on the woman's fallopian tube function, eggs, hormone profile, and the man's sperm count. Then "we distilled it down to the smallest number of inputs that would give us the most predictive value" and that shortlist includes things like hormone level, age, and sperm quality metrics, she said.
Next, "we validated Polaris using retrospective data and we know that the accuracy levels are within the range that our physicians find clinically promising," she said. "Starting this month, the algorithm and platform will be available through our partner clinics as a clinical study. And at the conclusion of that study which will be this time next year, Polaris will be more broadly available commercially to clinics." The company will also sort out pricing prior to next year's launch.
Celmatix plans to have its genetic assay ready for the market sometime in the next two years. Beim said that the company has completed the biomarker discovery phase of the development and has moved on to validating the markers based on retrospective data. This step is expected to last for about a year. Once it's complete, Celmatix will then validate the biomarkers in a prospective study — this part of the validation process is slated to start in spring 2015. Meanwhile, the company is developing an algorithm that can be incorporated into Polaris' infrastructure that will enable the system make use of genetic markers in its calculations of patients' outcomes.
The algorithm for the genetic markers will be incorporated into a second version of the company's CDSS which will be named Polaris X, Beim said. That version of the platform will move into prospective testing at the same time as the biomarker validation study next year, she said. Celmatix plans to launch Polaris X and its genetic test in 2016 and offer both products as a bundle. The idea, Beim said, would be to feed the test results directly into the system so that it can take the information into account when it makes its decisions. However, these plans might change.