What's Next

One patient, one consumer ring, every night since January 8, 2026. I had an allogeneic stem-cell transplant in November 2023. On March 16, 2026 I started ruxolitinib (Jakavi) for chronic graft-versus-host disease. The pipeline behind this site re-runs 30 reports on the ring data every morning; this page was generated from data through October 6, 2026, day 205 on the drug. Claude Code wrote the code. I described what I needed in Norwegian and checked what came back.

What it is not: a medical device, a clinical trial, or evidence about anyone but me. Every page carries that caveat because it is true. How this was built lists who did what; Every number, checked is the pipeline auditing its own pages.

Two things happened 2 days apart. Ruxolitinib started March 16, 2026. Hepatitis E was diagnosed March 18, 2026. At day 8 this page said the two could not be separated, and set a test:

“Time. HEV is acute. It resolves in weeks. Ruxolitinib is sustained. If the signals persist and strengthen at day 28 (~April 13), it is ruxolitinib. If they fade, it was HEV. We cannot know before then.”

This is the table the page showed then, day 8 on ruxolitinib, kept exactly as published:

The reading in March: the resting heart rate had gone up, which fit acute hepatitis E better than a JAK-inhibitor response; the temperature drop could be either; the HRV change was within noise.

Data through October 6, 2026. Every number is a live query from the pipeline's own JSON.

Did the signals fade? No.

All nights before treatment against all nights on treatment, the split the March page used:

These are descriptions, not evidence. The pooled Mann-Whitney test behind them fires at 100% of 20 random pre-treatment dates in this pipeline's own placebo calibration, so its p-values are not shown here.

Was it ruxolitinib? The March test could not see the second medicine.

A beta-blocker was added on April 8, 2026, 23 days after ruxolitinib. Three phases: before treatment (67 nights), ruxolitinib alone (22 nights), ruxolitinib plus beta-blocker (181 nights). The piecewise interrupted time series fits a step at each medicine start with AR(1) errors; Tau-U is the trend-corrected single-case effect size between phases.

By the test this page set itself, the signals did not fade. But the change the March page was waiting for did not arrive in the 22 nights on ruxolitinib alone: HRV moved from 10.0 to 10.9 ms, and the phase-corrected model finds no step there (-1.5 ms, p=0.71). It arrived after April 8, 2026: HRV averaged 28.1 ms on the two medicines together, with a step of +19.1 ms at the beta-blocker (p<0.001). The twin, trained only on the nights before treatment, projected HRV around 10 ms without it: 1 of 23 nights on ruxolitinib alone sat above that range, 173 of 182 nights since the beta-blocker did.

So the honest answer at day 205: the signal persisted, and by the pipeline's own phase analysis it sits with the second medicine, or with the two together. Whether that is a beta-blocker acting on heart rate, a delayed ruxolitinib effect, or both, is the open question. The Research Synthesis page holds the two-hit hypothesis; the Piecewise ITS, Tau-U and Sequential CI pages hold the models.

What time cannot settle: this is one person and an observational before-and-after. Hepatitis E was diagnosed two days after ruxolitinib started and resolved in the same window. The causal report runs the single-split methods and shows their own placebo check; read that before reading any paragraph here as proof.

The shortcut named in March still stands: Oura's own cohort contains users on ruxolitinib without hepatitis E, and users on beta-blockers without either. That comparison is impossible with N=1 and trivial at cohort scale.

This pipeline needs a config file and an OAuth token. It runs end to end in 5 minutes on one CPU. If your cohort contains even 50 users on ruxolitinib, you can validate or falsify every finding on this site in an afternoon. The code is MIT-licensed. The tables above say exactly which comparisons resolve which questions.