A “deep level of rigor” has been placed on the three key inputs, or pillars, of test development, says Kennedy, namely the clinical cohorts, genomic science, and machine learning. “Superimposed on all that is our processes for validating that the test can be used in a clinical setting and then ... [ensuring it] can help as many patients as possible, be covered by insurance, and widely adopted” by being included in clinical practice guidelines.
To fit the science around the highly defined clinical question required cohorts of samples, which in an initial PNKO01 exploratory clinical study—conducted at Mayo, Columbia University, and Houston Methodist—came from 236 women aged 18 and older scheduled to undergo a hysterectomy due to positive biopsies, pre-malignant conditions, or benign issues. Results showed that molecular signals associated with endometrial cancer could be detected and classified using a simple vaginal swab without accessing the uterus (International Journal of Gynecological Cancer, DOI: 10.1016/j.ijgc.2026.104546)
This stands in sharp contrast to the simple vaginal swab test from PinkDx. In the PNKO01 study, the vaginal swabs provided sufficient RNA signals to reliably detect endometrial cancer.
The second clinical trial, PNKO02, is now underway known as the PROACTION (Prospective Accrual to Investigate Reproductive Cancers in Women) study. It seeks to validate these findings in broader populations and is being conducted at 16 sites purposefully chosen to build diversity into the trial. Beyond the three sites in the PINKO01 trial, these include Washington University in St. Louis, the University of Alabama Birmingham, Meharry Medical College, the University of Tennessee Health Science Center, Summa Health, AdventHealth Research Institute, Ob-Gyne Associates of Lake Forest, Sutter Health, The Ohio State University, Avera Research Institute, University of Texas Health Science Center at San Antonio, Baylor College of Medicine, and Indiana University.
The study is enrolling females aged 45 and older with abnormal uterine or postmenopausal bleeding—exactly the kind of patients being targeted once the test is on the market, says Anderson. A robust set of clinical data is being collected on the women, including multiple samples and pathology diagnoses that will slot them into a classification of cancer or no cancer.
For those determined not to have cancer, the next step is to discover what they do have going on (e.g., polyps or fibroids) that is causing the abnormal bleeding, she says. If cancer is detected, the follow-up questions include what type and what stage.
Database lock on this trial will happen over the next six months, after which PinkDx will continue gathering information from the 3,000-patient cohort at three- and five-year follow-ups.
Deep RNA sequencing, coupled with machine learning, will generate a “robust biorepository” that will answer questions regarding recurrence rates as well as which cancers recur more, and which treatments do and don’t work, she says. It’ll be “a gift that keeps giving for years to come.”
The vaginal swab is of value both to women presenting with symptoms who have cancer and those who don’t, says Anderson. The overall workup process can sometimes improve more by moving non-cancer cases “out of the way” for the patients with cancer who can benefit most from an early diagnosis.
The scenario today is that somewhere between 1.3 million and 2 million women present for a medical workup with abnormal and often post-menopausal bleeding each year and upwards of 7% of them will be diagnosed with cancer, Anderson continues. Most of those women will get an endometrial biopsy.
Test developers generally focus solely on identifying the cancer, the proverbial “needle in the haystack,” she says. “When you’re approaching an indication to rule out [cancer] you are going to take a very different approach” because high sensitivity and high negative predictive value are needed.
This is where Kennedy comes in. The vaginal swab test is being developed to be very good at picking up the positive cases, using machine learning models developed from whole transcriptome analysis; this approach has been shown to be sensitive for detecting women with the condition, while minimizing false negatives.
“The specificity is also very important, but it is less important than the sensitivity,” says Kennedy. “Therefore, the models are chosen, selected, modified, tuned and developed over and over to be more and more sensitive.”
Standard machine learning model architectures—e.g., support vector machines, random forest, neural net, and logistic regression—are all at play, she says. But the real “secret sauce” is knowing how to select the right number and type of features to wrap an algorithm around so that it works. And that comes from knowledge gained from many years of using RNA transcriptome data, because “if you don’t know how to work with it you may end up concentrating on the noise and not on the signal.”
“The models are only as good as the data that you present it with, and so it will learn the truth from what you tell it is the truth,” continues Kennedy. “We develop strong models because we have clinical truth obtained by exhaustive combing, scrubbing, and verification of clinical records as well as samples that are comprehensive across the whole spectrum of disease.”
The black-box classifiers of the past are gaining acceptance as people become more comfortable with algorithms, such as natural language processing , where the features can be words (as with ChatGPT) or algorithms like support vector machines, which can use genes, gene transcripts, and RNA, Kennedy says. “At the end of the day, all these algorithms are extremely similar in concept to what we’ve been doing for many, many years. There’s noise and then there’s signal and your job is to train on enough samples with enough good truth, gold standard if you can, to create a model that approximates the truth against which the algorithm has been trained.”
Building the Evidence Base
Women could safely and accurately perform vaginal swabbing on their own at home for diagnostic purposes, and this might one day be how samples are collected for detecting endometrial cancer, says Anderson. “But out the door we want to have the patient there with the doctor.”
When PinkDx started, only four years ago, there was no evidence that a vaginal swab could match, let alone surpass, the sensitivity of an endometrial biopsy. When Kennedy perused the literature, all she could find was “a few spotty references to using genomics in endometrial cancer,” she says. “But there was nothing on a vaginal swab that could be used in conjunction with something like a whole transcriptome analysis ... what I like to call the living genome, because it aggregates signals from everything in the area of the tumor.”
The PNK001 trial sought a signal distal to the actual tumor up in the uterus, says Kennedy, “using the least invasive sample that we could. But we didn’t know if the signal of endometrial cancer would propagate ... all the way down through the cervix and into the vaginal canal.” That was demonstrated by taking four samples from each study participant—one each from the malignant or benign tumor, endocervix, ectocervix, and the vagina—with each woman serving as her own positive control.
PNK002, the PROACTION study, is picking up whatever signals are detectable via the vaginal swab, adds Anderson, which includes molecular signals of ovarian cancer and fibroids as well as endometrial cancer. The long-term possibility is a multiplex test that helps guide what happens next to patients. If that test is beneficial, it will also be a very cost-effective approach that payers will cover in lieu of women undergoing invasive costly procedures of limited value.
Part of the reason PinkDx chose endometrial cancer as its starting point is that it fit the “Rubik’s cube” the company was trying to solve: a large unmet clinical need where a minimally invasive test could demonstrate clinical utility, give physicians better data, and offer patients a more comfortable diagnostic experience, Anderson says. To do this while solving a big unmet need means the company stands to build a sizeable business around the vaginal swab, enabling it to “pursue the next pipeline dream product down the road.”
From a biological standpoint alone, it makes sense that “the closer you can get to the tumor ... the better the signal you’re going to get,” Anderson says. “We’ve experienced this in other indications that we’ve developed products for, but our number one goal was first to see if we could pick this up in a vaginal swab.”
Most of the current activity in cancer diagnostics is around liquid biopsies, she adds. “But who wants to have a needle put in their arm if you can on the first visit to the gynecologist have the sample taken [with a swab] that is going to answer that question?”
The vaginal swab will come to market as a laboratory developed test while PinkDx raises the capital to support the business through the usual early stage of insurance noncoverage, Anderson says. She and Kennedy plan to quickly amass the “compelling evidence ... answering a question that really matters” to abbreviate that pre-coverage period.