New survey-weighted analysis of 6,100 US adults. A blood-pressure reading and a category assigned to that reading are not the same thing. We compared the first of three consecutive measurements in NHANES August 2021–August 2023 with their three-reading average. 16.41% of adults were assigned a different US blood-pressure category (95% CI 15.36–17.45%). This is within-visit classification disagreement, not a rate of incorrectly diagnosed hypertension.
Key finding: 10.90% moved to a lower US category when their readings were averaged, while 5.51% moved higher. A separate analysis of crossings of the 130/80 threshold found 6.85%, not 16.41%.
Research question
How often do an individual’s first office blood-pressure measurement and the average of three consecutive measurements place that person in different categories? How much does the answer depend on the classification scheme and comparison rule?
What the data show
NHANES employed a standardised protocol: participants were seated and rested for five minutes, after which an automated Omron HEM-907XL device recorded up to three readings, spaced approximately 60 seconds apart. These are measurements within one research-centre visit. The study does not contain paired home or ambulatory blood pressure, and no diagnosis is adjudicated from this series.
| Classification comparison | Survey-weighted estimate | 95% CI | Changed / complete adults |
|---|---|---|---|
| US 2025, four categories: first vs mean of three | 16.41% | 15.36–17.45% | 990 / 6,100 |
| US 2025: first vs third | 25.26% | 23.80–26.72% | 1,565 / 6,100 |
| US 2025: first vs mean of second and third | 22.43% | 21.18–23.69% | 1,386 / 6,100 |
| ESC 2024, three categories: first vs mean of three | 11.04% | 9.82–12.26% | 674 / 6,100 |
| Crossing the 130/80 threshold: first vs mean of three | 6.85% | 6.05–7.65% | 408 / 6,100 |
| Crossing the 140/90 threshold: first vs mean of three | 4.65% | 3.89–5.41% | 297 / 6,100 |
All percentages use NHANES examination weights and the same complete-case adult sample; numerators are raw sample counts. Confidence intervals account for 15 masked strata and 30 primary sampling units (Taylor linearisation; 15 design degrees of freedom).
Where do categories change?
When using the four US categories, approximately 94% of recorded category changes occurred in participants whose first readings were sufficiently close to an effective threshold that a shift of no more than 5 mmHg could change their category. Threshold proximity helps explain why a small numerical change sometimes leads to a large-sounding diagnostic label change. This is mostly a property of classifying continuous measurements, not a newly identified physiological mechanism.
What if the first reading is already elevated? Conditional on the initial US category, the survey-weighted share moving down after averaging all three readings was 23.18% for Elevated (to Normal), 19.69% for Stage 1 (to Elevated or Normal), and 18.88% for Stage 2 (to any lower category). These row-conditional figures differ from the population-wide 16.41% reclassification rate. They are descriptive within-visit transitions, not individual probabilities of hypertension being ruled out. Percentages are calculated from the full-precision weighted transition table; subgroup confidence intervals are not supplied here.
Weighted transition table, US four-category rule
Percentage of the entire complete-case adult population. Rows show the first-reading category, columns the three-reading mean. Diagonal cells retain the same category.
| First \ Mean | Normal | Elevated | Stage 1 | Stage 2 |
|---|---|---|---|---|
| Normal | 43.57% | 2.02% | 0.74% | 0.03% |
| Elevated | 3.17% | 9.14% | 1.35% | 0.00% |
| Stage 1 | 1.71% | 2.77% | 16.90% | 1.37% |
| Stage 2 | 0.11% | 0.14% | 3.01% | 13.98% |
Exploratory extension: how much asymmetry survives randomised measurement order?
Supplementary analysis, 13 October 2026 (not part of the archived v1.0 DOI files). We independently reanalysed the public CDC-derived CSV mirrors linked by SEQN (n = 6,100; examination weights; 15 strata and 30 PSUs), reproducing the 16.409% headline result. Between readings three and one, survey-weighted mean systolic pressure changed by −0.22 mmHg (95% design CI −0.47 to +0.03), and diastolic pressure by −0.81 mmHg (−0.96 to −0.67). The weighted standard deviations of these paired differences were 6.99 and 4.64 mmHg, respectively. Small average changes coexist with substantial within-person variation.
Order-symmetrisation test. Holding each participant’s three observed reading pairs and their three-reading mean fixed, we gave each of the three pairs equal probability of being designated the first reading. Under this descriptive counterfactual, the weighted share moving to a lower US category was 9.28%, versus 5.59% moving higher (gap 3.69 percentage points, 95% design CI 3.08–4.30). In the actual order the shares were 10.90% versus 5.51% (gap 5.39 points, 95% CI 4.06–6.72). The actual-minus-symmetrised gap was 1.70 points (95% CI 0.74–2.67). About 68% of the observed directional gap therefore persists when within-person measurement order is ignored. This decomposition does not identify anxiety, regression to the mean or device error as separate causes, and does not quantify their causal contributions. The supplementary estimates come from public derivative CSV mirrors, not from an independently authenticated fresh CDC XPT download; direct source-byte verification remains outstanding for this extension.
Verification update · 14 October 2026
The headline result has been independently recalculated at participant level using two public NHANES 2021–2023 CSV derivatives, rather than the published summary matrix. In 6,100 adults with complete readings, the survey-weighted first-reading versus three-reading-average category difference was 16.408655% (95% design CI 15.36235–17.45496%). The downward and upward category changes were 10.9008% and 5.5079%, respectively; the first-versus-mean-of-last-two comparison was 22.4334%. All match the reported figures after rounding. Survey weighting used 15 strata and 30 PSUs.
Verification boundary: the CSV copies were obtained from public third-party repositories; direct byte-level authentication against freshly downloaded CDC XPT originals was not part of this independent check. See the standalone reproduction instructions and the original CDC source-file links below. This does not measure clinical diagnostic error or establish a physiological mechanism.
Methods and reproducibility
We linked the August 2021–August 2023 BPXO_L.xpt and DEMO_L.xpt files by SEQN. The demographic file contained 11,933 people; blood-pressure examination data contained 7,801 records. Of 6,337 examined adults aged 18 or above, 6,100 (96.26%) had three complete systolic–diastolic pairs. We used WTMEC2YR examination weights, masked SDMVSTRA and SDMVPSU, and Taylor-linearised variance for a survey-weighted proportion. For the primary result the design-based SE was 0.491 percentage points.
The first reading was classified directly. For the comparator, we calculated each person’s mean systolic pressure and mean diastolic pressure across all three readings, then reapplied the same classification. The two US thresholds of 130/80 and 140/90 were analysed separately as binary status changes. ESC office-based categories were applied to the same US participants, not to a European sample.
The two provided original-format XPT inputs were processed together. Re-running the pipeline reproduced all six headline estimates and the 25 cells across US and ESC weighted transition matrices. An earlier error in counting examined adults was traced to SAS missing-value encoding and corrected by explicitly requiring RIDSTATR = 2; this did not change the weighted estimates. File digests and the source audits are included in the local reproducibility package retained by the Institute.
Interpretation and limitations
- The mean of three readings is a comparison reference, not a verified gold-standard diagnosis.
- All readings are from one research-centre visit. We cannot attribute the effect to anxiety, arm choice, acclimatisation or white-coat hypertension.
- Participants missing one or more readings (237 examined adults) are excluded, so residual selection bias is possible.
- US and ESC categories are different definitions applied to the same data; cross-country comparisons of disease prevalence cannot be inferred.
- Survey confidence intervals describe sampling uncertainty; they do not correct systematic measurement errors.
- This is a descriptive public-data reanalysis, not a peer-reviewed clinical intervention study.
Prior work and novelty
Handler, Zhao and Egan (2012) previously examined the number of blood-pressure measurements and classification using NHANES 1999–2008. Our contribution is the updated August 2021–August 2023 data, current US/European rules, transparent weighted transition estimates and explicit separation of changes in category from clinically important threshold crossings. We do not claim discovery of the repeat-measurement effect.
Context and competing evidence. Handler, Zhao and Egan (2012; DOI) studied repeated readings in historical NHANES data. Tang et al. (2021; DOI) compared simplified screening protocols using NHANES and other datasets. Furthermore, the NHANES 2017–March 2020 pre-pandemic BP module already contains repeated oscillometric Omron readings. A historical-versus-current analysis must account for methods, weights, population differences and the change from auscultatory measurements; this report does not claim that such a comparison has been completed.
Reproduce the estimate
Copy the tested Python reproduction script and download both original CDC source files. The public method reproduces the weighted 16.408655% estimate and its 95% confidence interval to six decimal places.
Primary sources
- CDC/NCHS, Oscillometric Blood Pressure, NHANES August 2021–August 2023, including BPXO_L.xpt.
- CDC/NCHS, Demographics, NHANES August 2021–August 2023, including DEMO_L.xpt.
- AHA/ACC 2025 Blood Pressure Guideline.
- ESC 2024 Guidelines.
- Handler J, Zhao Y, Egan BM. Impact of the number of blood pressure measurements on blood pressure classification in US adults: NHANES 1999–2008. 2012.
Suggested citation: Nikitin, I. (2026). How Often Does One Blood Pressure Reading Change the Category? A Survey-Weighted NHANES 2021–2023 Reanalysis. Version 1.0. The Null Institute. https://doi.org/10.5281/zenodo.23342882. The analysis has not been peer-reviewed.
DOI and archived files: Zenodo version 1.0 (report, code and tables) · All versions (Concept DOI).