Does Tracking Your Bloodwork Actually Improve It? A 20,000-User Study, Read Honestly

What the Study Actually Measured

The paper is titled "Improvements in blood and fitness tracker biomarkers in a longitudinal real-world cohort of digital health platform users." It was published in PLOS Digital Health on March 24, 2026 (DOI 10.1371/journal.pdig.0001271, PMID 41875148) and announced by InsideTracker in a press release on April 2, 2026.

It is a retrospective longitudinal analysis of 20,342 users of a digital health platform that combines blood biomarkers, polygenic risk scores, and fitness tracker data with personalized lifestyle recommendations. To be included, a user needed at least two blood tests separated by 90 days or more. The median gap between the baseline draw and the follow-up draw was 260 days. The cohort was 64.2 percent male and 84.3 percent white, with a mean age near 46. A subset of users with five or more repeat draws was followed for an average of 4.2 years, which is unusually long for real-world consumer health data.

The paper describes an investigation of 39 clinical blood biomarkers, selected for relevance to age-related disease risk and overall health status. Fitness tracker measures, specifically step count and REM sleep percentage, were analyzed as behavioral correlates rather than as primary blood biomarkers.

Two design facts matter more than any individual result. There was no control group. And the analysis was retrospective, meaning the researchers looked backward at data that already existed rather than randomizing anyone to anything. The authors say plainly that this design precludes causal inferences about the relationships between user behaviors and biomarkers.

What Improved, What Resisted, and Where the 43-Biomarker Headline Comes From

Among users who began with suboptimal baseline values, the paper reports statistically significant improvement by Mann-Whitney U test at p below 0.05 for the twelve markers displayed in its primary figure: LDL cholesterol, triglycerides, fasting glucose, cortisol, HbA1c, hsCRP, vitamin B12, vitamin D, folate, iron, testosterone in males, and HDL cholesterol.

The pattern across categories is more informative than the list. Trade coverage of the study in NutraIngredients broke out per-marker improvement rates by the second test. Micronutrients were the most tractable, with iron improving in 63.9 percent of users who started low, folate in 63.2 percent, vitamin D in 57.1 percent, and magnesium in 49.8 percent, though vitamin B12 sat well below that band at 36.1 percent. Lipids were the most resistant: 20.4 percent of users improved an elevated LDL cholesterol and 27.3 percent improved a low HDL cholesterol. A subgroup that began with HbA1c in the diabetic range was reported to decline across five longitudinal tests, with the group average reaching the prediabetic range by the fifth draw.

|---|---|---|

The genetics layer adds a useful nuance. Polygenic risk scores correlated both with baseline biomarker levels and with the degree of change over time. Users with a higher genetic predisposition toward elevated LDL cholesterol showed smaller improvements than users without it, despite the same platform and the same recommendations. That is a reminder that biology sets different starting points and different slopes for different people, which is exactly why comparing your own trend to your own history is more meaningful than comparing your number to someone else's.

Now the headline. The company announcement and the trade coverage that followed both state that 43 biomarkers showed statistically significant improvement at p below 0.05, and that figure is now the most-repeated number about this study. Reading the paper itself, the count does not line up cleanly. The introduction describes an investigation of 39 clinical blood biomarkers. The results section states that the biomarker changes displayed in the paper's first main figure, which shows 12 markers, were all statistically significant by Mann-Whitney U test. We could not locate the figure of 43 stated as such in the paper's abstract or main results text.

One plausible explanation is that the 43 count aggregates blood biomarkers with fitness tracker measures and with subgroup or repeat-test comparisons reported in the paper's supplements. That would make it a defensible tally rather than an invented one. But it is a company-communications number, not a sentence you can point to in the results section, and it is worth saying so rather than repeating it as though the two were the same thing. If you plan to cite this study in client materials, cite the design and the named markers, which are unambiguous, rather than leaning on a headline count you would struggle to source to a specific line.

The Action Plan, the Steps, and the Sleep

Here is the result that deserves more attention than it has received. Among users with two or more blood tests, 72 percent created at least one action plan during the window between their baseline and follow-up draws. Users who created targeted action plans showed greater biomarker improvement than users who remained inactive on the platform.

The comparison inside the study is therefore not really tested versus untested. Everyone in the cohort tested, at least twice. The comparison is between people who tested and then built a structured, individualized plan, and people who tested and did not. Access to the data was constant. Structured engagement with the data was the variable that tracked with divergent results. That is a meaningful reframe for anyone who assumes the panel itself is the intervention. It is also, honestly, the reframe that most flatters coaches and practitioners, which is a reason to hold it to a higher evidentiary bar rather than a lower one.

The wearable correlates are specific enough to be interesting and soft enough to be easy to overstate.

Read that carefully. This is an association observed after the fact in people who were already doing many other things at once, including changing diet and supplementation on the platform's recommendations. It is not evidence that adding a thousand steps moves cholesterol, and it is certainly not a prescription. People who are sleeping better and moving more are usually also eating differently, drinking less, and under less stress, and none of that was randomized or controlled. If you want to track wearable data alongside labs, the practical value is context for your own trend line, which is the same logic behind [pairing wearable data with bloodwork](/blog/wearable-bloodwork-peptide-tracking-2026), not a causal lever you can pull.

Reading It Honestly: Healthy-Adherer Bias and the Rest

The single most important thing to understand about this study is healthy-adherer bias. In observational research, the people who stick with a program tend to do better than the people who do not, and they tend to do better on outcomes the program never touched. Adherers are systematically different: more motivated, more health-literate, better resourced, more likely to be doing five other beneficial things at the same time. A comparison of engaged users to unengaged users measures adherence as much as it measures the intervention. Every finding in this paper about action plans, steps, and sleep is exposed to that problem.

Four other limits belong alongside it.

None of this means the study is worthless. Twenty thousand people with repeat labs, and a subset followed for over four years, is a real dataset and a genuine contribution. The honest version of the claim is still worth something: among motivated people who test repeatedly, suboptimal markers tend to move toward range over time, and structured engagement is associated with more movement than passive access. What it is not is proof that buying a panel changes your physiology.

|---|---|

What to Track: The Markers This Study Followed

If you want a personal record that mirrors what this research measured, these are the markers involved. This is a description of what was studied, not a recommendation to test or to change anything. Which tests are appropriate for you is a clinical question.

The mechanics matter as much as the list. Repeat draws are only comparable if the conditions are comparable, which is the whole subject of [comparing lab results over time](/blog/how-to-compare-lab-results-over-time), and deciding what belongs on a repeatable panel in the first place is covered in [building a longevity blood panel](/blog/how-to-build-longevity-blood-panel). Browse plain-language explanations of each test in the full [biomarker library](/biomarkers).

[Keep your lab history, your wearable data, and the plan you actually followed in one place with MyProtocolStack.](/auth/login?mode=signup)

What This Means If You Coach People

For practitioners and coaches, the usable finding is narrow and specific, which is what makes it credible.

The thing that tracked with divergent outcomes in this cohort was not owning the data. Everyone owned the data. It was whether the user built and worked from an individualized plan. That is the coach's function described in a peer-reviewed paper rather than asserted in a sales deck, and it is a stronger position than the vague claim that testing helps.

The way to use it without overclaiming is to pair it with the caveat, in the same breath. Something like: in a 2026 PLOS Digital Health analysis of more than 20,000 platform users, people who built individualized action plans improved more than people who did not, though the study was observational and vendor-funded, so it shows an association rather than proof that the plan caused the change. That sentence survives scrutiny from a skeptical client, a compliance reviewer, and a physician. The version that drops the second half does not.

It is also worth being straight with clients about what a platform does and does not do. A tracking tool stores, organizes, and visualizes. It does not interpret and it does not decide. The differences between bundled lab subscriptions and open tracking platforms are laid out in our [comparison of MyProtocolStack and InsideTracker](/blog/myprotocolstack-vs-insidetracker), and none of those products, including ours, has evidence that owning the software changes anyone's physiology.

Frequently Asked Questions

Does tracking your bloodwork actually improve your biomarkers?

No study has shown that tracking causes biomarker improvement, because no randomized controlled trial has tested it. The best available evidence is a March 2026 retrospective analysis in PLOS Digital Health of 20,342 digital health platform users, which found statistically significant improvement in markers including LDL cholesterol, HbA1c, fasting glucose, vitamin D, hsCRP, and testosterone, concentrated among users who started with suboptimal values. Because the study had no control group and drew on self-selected paying subscribers, it demonstrates an association, not causation. Healthy-adherer bias, meaning the tendency of motivated people who stick with any program to do better regardless of the program, is the leading alternative explanation.

What did the 20,000-user PLOS Digital Health study actually find?

The study analyzed 20,342 users with at least two blood tests separated by 90 days or more, with a median gap of 260 days, and a subset with five or more draws followed for an average of 4.2 years. It investigated 39 clinical blood biomarkers and reported statistically significant improvement by Mann-Whitney U test at p below 0.05 for the markers in its primary figure, including LDL cholesterol, triglycerides, fasting glucose, cortisol, HbA1c, hsCRP, vitamin B12, vitamin D, folate, iron, testosterone in males, and HDL cholesterol. Improvements were largest among users with suboptimal baselines and were sustained or increased across repeat tests. Micronutrients moved most readily and lipids least. The widely quoted figure of 43 improved biomarkers comes from the company announcement rather than a sentence in the paper's results section.

Why can't this study prove that tracking caused the improvement?

Because it was retrospective and observational with no control group. Nobody was randomized, so there is no comparison group of similar people who did not track. The authors state directly that their design precludes causal inference about relationships between user behaviors and biomarkers. They partially addressed this by comparing against users with optimal baseline levels who received no biomarker-targeted guidance, but they note that this comparator is not a counterfactual and remains subject to confounding. Regression to the mean is an additional issue the authors raise themselves, since a group selected for extreme baseline readings will tend to test closer to average on repeat draws through normal variability alone.

What is healthy-adherer bias and why does it matter here?

Healthy-adherer bias is the well-documented pattern in observational research where people who consistently stick with a program do better than people who do not, including on outcomes the program never addressed. Adherers tend to be more motivated, more health-literate, better resourced, and more likely to be improving several habits at once. In this study, the comparisons between engaged and unengaged users, and between users who did or did not increase their step counts, are all exposed to it. That does not make the results false, but it means the improvement cannot be attributed to the platform rather than to the kind of person who uses a platform consistently.

Did users who made an action plan really improve more than those who did not?

That is what the study reports. Among users with two or more blood tests, 72 percent created at least one action plan during the window between baseline and follow-up, and those who created targeted plans showed greater biomarker improvement than inactive users. It is a genuinely interesting finding because access to the data was constant across the cohort while structured engagement with it was not. It is still an observational comparison, though, and the paper had no structured adherence tracking, so people who build and follow a plan differ systematically from people who do not. The honest framing is that planning was associated with more improvement rather than that planning produced it.

Sources

1. PLOS Digital Health, "Improvements in blood and fitness tracker biomarkers in a longitudinal real-world cohort of digital health platform users," DOI 10.1371/journal.pdig.0001271, March 24, 2026. https://journals.plos.org/digitalhealth/article?id=10.1371%2Fjournal.pdig.0001271

2. PubMed record for the same study, PMID 41875148. https://pubmed.ncbi.nlm.nih.gov/41875148/

3. PR Newswire, InsideTracker announcement of the study, April 2, 2026. https://www.prnewswire.com/news-releases/landmark-study-of-20-000-users-provides-peer-reviewed-proof-insidetrackers-ai-health-platform-linked-to-significant-sustained-health-biomarker-improvements-302732131.html

4. NutraIngredients, "InsideTracker study highlights rise of biomarker-driven supplement personalization," May 12, 2026. https://www.nutraingredients.com/Article/2026/05/12/insidetracker-study-highlights-rise-of-biomarker-driven-supplement-personalization/

*MyProtocolStack is a tracking and education tool, not medical advice, diagnosis, or treatment, and you should always consult a qualified healthcare professional before making any changes to your health protocol.*