FDA Clears Natural Cycles' Hybrid Machine Learning Fertility Algorithm

The U.S. Food and Drug Administration has cleared an updated version of Natural Cycles, the Swedish company's app for preventing or planning pregnancy, built on a new algorithm that adds machine learning to its existing statistical model. The company announced the clearance, its sixth from the FDA, on July 27 in a release distributed by PR Newswire.

The FDA's decision letter and 510(k) summary for K253694 is dated July 17, 2026. The agency found the updated app substantially equivalent to a previous Natural Cycles version cleared under K231274. It is a Class II device under the regulation for software applications for contraception, and the indications for use are unchanged.

How the new algorithm works

According to the FDA summary, the updated algorithm has three parts. All data first pass through the statistical algorithm used in earlier versions of the app. A machine learning model then uses the data the user logged, along with the statistical algorithm's outputs, to estimate the probabilities of ovulation and menstruation. A decision engine combines both outputs and determines which to use for a given user and cycle.

The summary describes the machine learning model as locked, meaning it gives the same output for the same input and does not change with use. It was developed with data from more than 900,000 app users who consented to research use, split into separate training, tuning and test sets.

The company said the update is meant to give many users more "Green Days," the days the app indicates that pregnancy protection is not needed, and to improve its handling of irregular cycles, lower-quality temperature data and cycles that change over time. Natural Cycles also said it has filed a patent application covering parts of the algorithm.

What the validation showed

The FDA summary describes a comparison of the new hybrid algorithm with the previous statistical algorithm, version 3.6.0, on a held-out test set of 111,446 users and 1,031,706 complete cycles. Ovulation dates derived from users' logged luteinizing hormone tests served as the reference, and those test results were withheld from both algorithms.

On each prespecified measure the new algorithm passed a non-inferiority test, according to the summary. The average share of green days per cycle was 48.41%, against 48.03% for the earlier algorithm. Positive percentage agreement, the share of true fertile days the app marked as high risk, was 99.42%, against 99.26%. Ovulation was detected in 98.23% of cycles with LH data, against 97.59%.

Natural Cycles said the app remains 98% effective with perfect use and 93% effective with typical use, figures the company attributes to peer-reviewed research and real-world evidence. The FDA summary does not report those rates.

Six clearances

The company's earlier U.S. authorizations include a 2024 510(k) clearance with a predetermined change control plan for adding hardware, and a 2025 clearance that allows the app to be offered over the counter as well as by prescription, according to the release. The change control plan has since been used to add the company's NC° Band wearable and a Garmin integration, the company said.

"One of the unique advantages of a software-based contraceptive is that it can continue to evolve," Elina Berglund Scherwitzl, Natural Cycles' chief executive and co-founder, said in the release.