A Line Painted on a Slope — An Introduction to Reference Ranges and Diagnostic Thresholds

A Single Asterisk on Your Test Report

When you receive a health checkup report, one or two lines among roughly twenty numbers often have a mark like H or L next to them. That mark means the number fell outside the reference range printed on the report. The small differences that laboratories flag as low, high, or abnormal can cause patients needless worry, and results flagged outside the "normal range" this way are far more common than clinically meaningful abnormalities (PMID 40268322). The reasons lie first in how those ranges are made, and are added to by different testing methods across laboratories, variations by age, sex, race, and season, and random fluctuations (PMID 40268322).

The lines you encounter on test reports and in the exam room fall largely into three categories. The first is the range printed next to the report's numbers, which shows where the values of healthy people cluster. The second is the diagnostic threshold, which defines the value at and above which a condition is called a disease. The third is the treatment threshold, which defines the value at and above which treatment begins. These three lines are made from different ingredients, so even when they sit on top of the same number, their meanings differ. Blood pressure is a good example. Hypertension is a state of persistently high blood pressure, and which value counts as high is a line drawn by guidelines.

There is one picture this article will use all the way through. The risk of disease is often closer to a slope that rises gradually as a value increases, rather than a staircase that suddenly leaps at a certain value. Diagnostic thresholds and treatment thresholds are lines painted somewhere on that slope. How well a single test screens for disease, and how missing people and false alarms are traded off when moving a threshold, was covered in "What a Test Actually Tells You — An Introduction to Sensitivity, Specificity, and Pretest Probability". Here, we will not redo those calculations, but instead follow what ingredients the thresholds themselves are made of.

Reference Ranges — 95% of Healthy People

The standard procedure for creating a reference range starts with gathering at least 120 healthy people. Values are measured in a group matched for age group, sex, race, and testing method, and the middle 95% of them is used as the "normal range" (PMID 40268322). The reference range is the span where the middle 95% falls, after collecting values from healthy people and trimming off the bottom and top 2.5% each. By definition, this method marks 5% of healthy people as having an abnormal result (PMID 40268322).

This 5% is a story about a single test, so if you do multiple tests, the fraction grows. Let us do a hypothetical calculation. Assuming that test results are completely independent of each other and a healthy person receives twenty tests, the probability that all twenty fall within the range is 0.95 multiplied twenty times, which is roughly 36%. The remaining roughly 64% receive a flag on at least one line. Actual test items are intertwined, so it is not exactly this number. Still, the point remains that the more tests you have, the larger the fraction of healthy people who get flagged. A single asterisk on a result sheet cannot tell you that you are sick.

A reference range changes depending on whom you use as a reference population. Normal values must come from healthy people under the same conditions as the person being tested (PMID 8796934). Therefore, the procedure for establishing a reference range includes a step to decide in advance whether to divide the group by criteria such as age and sex (PMID 18852866). This is because values vary depending on the laboratory's method, age, sex, race, and season, even for the exact same test (PMID 40268322). Children and the elderly are still cited as groups for which it is difficult to establish reference ranges (PMID 26981015). If you change the reference population, the line moves too. The 2017 U.S. pediatric hypertension guidelines recreated the blood pressure reference charts for children using data from normal-weight children (PMID 28827377).

A value that falls within the range is not a guarantee that there is no disease either. Spirometry tests view only low values as a problem, so they use just a single lower line. The lower limit of normal is the line that separates the bottom 5% among healthy people of the same age, height, and sex. This line is set at the 5th percentile of healthy non-smokers (PMID 22222132). If you use a single fixed ratio for everyone as a rule of thumb, young and tall people are underdiagnosed, while older and short people are overdiagnosed (PMID 22222132). Near the lower limit of normal, low values in healthy people overlap with values in early disease. Therefore, for someone with a high pretest probability of disease, such as a patient referred to a pulmonary function laboratory, a value higher than the lower limit of normal may still be more likely to indicate disease than not (PMID 22222132). Still, this review concluded that high-quality spirometry remains the best tool for finding undiagnosed, distinct lung disease (PMID 22222132). Where pretest probability comes from was covered in "Where Does Pretest Probability Come From? — History, Physical Examination, and Clinical Prediction Rules".

The reference range describes where healthy people stand, not where disease begins. There is also a suggestion that it could be helpful if result sheets included the clinical meaning of the result in plain words along with the range, such as "Nothing to worry about right now," "Discuss with your doctor at your next visit," or "Contact your doctor for action" (PMID 40268322).

Decision Limits — A Line Drawn Backwards from Outcomes

The diagnostic cutoff for type 2 diabetes, a glycated hemoglobin of 6.5%, is a value determined by looking at how common retinopathy is among people who did not know they had diabetes (PMID 38615570). Glycated hemoglobin is the proportion of hemoglobin inside red blood cells that has glucose stuck to it. And diabetic retinopathy is a disease where the tiny blood vessels of the retina—the light-receiving membrane at the back of the eye—are damaged. The material used to make this line is different from the distribution of healthy people. Instead of asking where healthy people are clustered, it is a line obtained by asking above which value complications increase.

A decision limit is a line determined on the basis that the risk of disease or the pros and cons of treatment change with that value as the boundary. Diagnostic thresholds and treatment thresholds fall into this category. An old review noted that the line of normality is often determined by looking at "the consequences of being called abnormal" (PMID 8796934). Even numbers written in the same format on a test report may differ in which side's line they were made for, and confusing reference ranges with decision limits is still cited as a problem in laboratory medicine today (PMID 26981015).

Such lines are tested again when outcome data accumulates. Among 19,897 Japanese adults who underwent a health checkup in 2006 and were tested again 3 years later, those with a glycated hemoglobin of 6.5–6.9% had a statistically significantly higher risk of developing new retinopathy within 3 years than those with 5.0–5.4%, while no evidence of increased risk was seen in those with 5.5–6.4% (PMID 22891221). The authors concluded that this result supports the validity of the 6.5% line. Among 2,920 Qatari adults with a glycated hemoglobin of 5.6% or higher who re-evaluated the line based on urinary albumin, a sign of kidney damage, no value below 7.0% emerged as the best line for predicting kidney damage (PMID 38615570). The authors considered that the 6.5% based on retinopathy did not need to be changed. Each of the two studies represents the sample of one country.

In many cases, the risks above and below the line continue like a slope. An observational study is a study where people's lives are not interfered with, leaving them as they originally live, and watching what happens afterwards. In an observational study following British patients who already had type 2 diabetes, the lower the glycated hemoglobin by 1 percentage point, the lower the risk of diabetes-related outcomes by 21% and the risk of microvascular complications by 37%, and no threshold where risk jumps suddenly was seen in any outcome (PMID 10938048). In this study, the group with the lowest risk was people whose glycated hemoglobin was below the normal range of 6.0%. In places where risk continues like this, the line becomes a mark of decision painted on a slope. However, this study is data from people who already have diabetes. Near the diagnostic line drawn for people without diabetes, results have also emerged showing that there may be a threshold of retinopathy risk between a glycated hemoglobin of 6.0% and 7.0%, as in the Japanese study above (PMID 22891221).

When you paint a line on a slope, some of the information about the slope disappears. Dichotomization is the act of cutting a continuous value with a single line to divide it into two groups, above and below. Clinical practice requires decisions such as "it is a disease or not" and "treat or do not treat," which is why lines are drawn. When values are cut in two like this in research, a price is paid. You largely lose the power to find differences, the influence of other factors remains less filtered out, and finding and cutting the "best" line within the data severely skews the results (PMID 16217841). The authors of this paper concluded that dichotomization is unnecessary in statistical analysis, and in particular should not be used for explanatory variables in regression analysis. As a hypothetical example, if you draw a line at 130 for systolic blood pressure, 129 and 131 fall into different boxes, and 131 and 179 fall into the same box. All the line leaves behind is "above or below," and the slope of the hill is discarded.

If you paint one more line on the slope, a pre-stage of the disease is created. Prediabetes is a range where blood sugar is higher than normal but does not reach the line for diabetes. Among people with prediabetes, 5 to 10% cross over to diabetes every year, and a similar share return to normal blood sugar (PMID 22683128). The same review regarded prediabetes as a state of high diabetes risk, cited evidence that an association with the risk of early kidney disease, retinopathy, and large blood vessel disease was seen in observational studies, and that lifestyle improvements reduced the risk of diabetes by 40 to 70% (PMID 22683128). Whether to view this section as a disease in its own right is a matter of debate, and one review argues based on changes seen in multiple organs that it should be regarded as a separate clinical state (PMID 30935247). Where each organization draws the line for this section was covered in "What Do Blood Sugar Tests Measure? — Three Numbers, and the Lines Each Organization Drew".

The Year the Line Moved — Blood Pressure in the United States in 2017

In 2017, the blood pressure diagnostic threshold in the United States dropped from 140/90 to 130/80 mmHg. The guideline issued that year by the American College of Cardiology and the American Heart Association (hereafter the 2017 US guideline) established that blood pressure consistently at or above 130/80 should be diagnosed as hypertension (PMID 31521481), whereas the threshold used previously in the US (JNC7) was 140/90 (PMID 33384395). The blood pressure threshold has been moving from the beginning. Tracing back 32 hypertension guidelines issued by various international organizations and academic societies, the threshold started at 160/95 in the late 1970s, gradually shifted downward, and varied depending on an individual's cardiovascular risk (PMID 42029627). Over the past decade, the directions have even split. The World Health Organization, the International Society of Hypertension, and the UK National Institute for Health and Care Excellence (NICE) maintained 140/90 for the general population, while European and American societies lowered both the threshold and treatment targets further (PMID 42029627). Looking solely at the diagnostic threshold, as of 2019, European guidelines began diagnosing hypertension at 140/90, while US and Canadian guidelines started at 130/80 (PMID 31030859).

Applying both thresholds to the same groups of people reveals what happens. When the blood pressure of 9,623 participants in the 2011–2014 US National Health and Nutrition Examination Survey was scaled to the entire US adult population, the proportion of adults with hypertension was 31.9% under the JNC7 criteria and 45.6% under the 2017 US guideline (PMID 29133599). In the same comparison, the proportion for whom blood pressure medication was recommended increased from 34.3% to 36.2%. The number of people receiving the label of hypertension increased by nearly 14 percentage points, whereas the number recommended medication increased by less than 2 percentage points (1.9 percentage points). Under the 2017 US guideline, 9.4% of US adults had hypertension but were recommended lifestyle modifications instead of medication (PMID 29133599). Among those already taking blood pressure medication, the proportion whose blood pressure remained above the treatment target was 39.0% by JNC7 criteria and 53.4% by the 2017 US guideline, and the authors concluded that many people would end up lowering their blood pressure more aggressively.

This difference occurred because the 2017 US guideline separated the diagnostic threshold from the treatment threshold. The range of 130–139/80–89 is termed stage 1 hypertension, and most people in this range are not candidates to start medication immediately (PMID 31521481). For this range, the guideline for the first time calculates the absolute risk of future cardiovascular events to determine who is at high risk. The 2017 US guideline recommended starting medication at 130/80 or higher for those at high risk and at 140/90 or higher for those not at high risk, while making exceptions to start at 140/90 even for high-risk individuals, such as for the secondary prevention of stroke in people who have never used medication (PMID 31521481). The diagnostic threshold is a label indicating that "this person's blood pressure is on the high side," whereas the treatment threshold is a judgment that "the risk is high enough that the benefits of medication outweigh the harms." Why absolute risk matters in treatment decisions was covered in What Does "50% Lower Risk" Mean in People? — Reading Relative Risk, Absolute Risk, and NNT.

Applying the same threshold to other countries also moves the numbers. When the 2017 US guideline was applied to a national sample of 11,822 Chinese adults aged 45–74 from 2011–2012, it was estimated that compared to previous criteria, the proportion with hypertension would increase from 44.1% to 56.4%, and the proportion recommended medication would rise from 41.6% to 49.1% (PMID 32521545). The authors of this study suggested that this guideline has the potential to reduce risk by selectively recommending medication to those who will benefit, adding that the benefit depends on whether the guideline is adopted and can be followed. In a 2015 survey of 180,335 Indian adults (mean age 40.6 years), the proportion with hypertension more than doubled, jumping from 29.7% under the JNC8 criteria using 140/90 to 63.8% under the 2017 US guideline (PMID 33465185). When you move the line, people's blood pressure stays the same, but the number of patients changes.

Real risk indeed exists even for those who climbed above the new line. In an observational study following roughly 100,000 Chinese adults without a history of blood pressure medication use, myocardial infarction, stroke, or cancer for a median of about 11 years, individuals with stage 1 hypertension had a 20% higher risk of the combined outcome of myocardial infarction, stroke, and death compared to those with normal blood pressure (PMID 37319412). Although observational data from a single country, this shows that the 130–139 range does not sit on a flat plateau of a slope.

The trial that one review called "landmark" as the basis for the lower target is SPRINT (PMID 31030859). A total of 9,361 participants with high cardiovascular risk but without diabetes were randomized into two groups, with one group treated to a target systolic blood pressure of less than 120 mmHg and the other to less than 140 mmHg (PMID 26551272). The trial was stopped early after a median of 3.26 years because the combined outcome of myocardial infarction, stroke, heart failure, and cardiovascular death occurred at a rate of 1.65% per year in the group targeting less than 120 and 2.19% per year in the group targeting less than 140, meaning the risk was 25% lower in the group targeting less than 120, and the risk of all-cause mortality was also 27% lower (PMID 26551272). In exchange, serious adverse events such as hypotension, syncope, electrolyte abnormalities, and acute kidney injury were more frequent in the group targeting less than 120, while injurious falls did not increase. In the final report, which completed adjudication of events and added post-trial follow-up, the patterns of benefit and adverse effects were similar, except that the difference between the two groups disappeared for heart failure alone (PMID 34010531).

In an analysis pooling 80,220 participants from six trials including SPRINT, the group with more intensively controlled blood pressure reduced cardiovascular events by about 1.7 per 100 people and increased adverse effects such as hypotension and syncope by about 1.8 per 100 people over a median of 3.2 years (PMID 40902616). Asians accounted for 82.6% of the participants. The authors concluded that there was a net benefit after weighting and comparing each event. When the number of reduced events and increased adverse effects is similar, which side carries more weight depends on the weights assigned to the events. Noting SPRINT and various meta-analyses as the rationale for lower targets, one review summarized that this must be weighed against concerns that low-risk individuals may be harmed by over-treatment (PMID 31030859). How to interpret treatment targets such as target blood pressure was covered in Better Numbers, Better Patients? — An Introduction to Surrogate Endpoints.

A Slope That Varies by Age and Population — Kidneys and Body Size

Let us imagine two people receive the exact same kidney test numbers. One person is thirty years old, and the other is eighty years old. The representative number on a kidney test result is eGFR. There are many tiny clusters of blood vessels called glomeruli in the kidneys that filter blood, and the estimated glomerular filtration rate is the volume of water from the blood that the kidney's glomeruli filter out in one minute, serving as a representative number indicating how hard the kidneys are working. Measuring this value directly is cumbersome, so it is estimated through a calculation. eGFR is a value estimating the glomerular filtration rate using a formula that incorporates information such as age and gender into the creatinine concentration in the blood. Creatinine is a waste product mainly produced by muscles. The basics of reading this number were covered in "Reading Your Kidney Test Results — A Sieve You Cannot See, What Sieve Stays Behind and What Leaks Through".

Chronic kidney disease is a condition where the kidneys are damaged or kidney function is impaired for three months or longer. The definition by the international kidney guidelines (KDIGO) uses an eGFR below 60 regardless of age as the threshold for chronic kidney disease, and this threshold has been widely used as a standard for a long time (PMID 38411155, PMID 40674065). This threshold has been supported by epidemiological studies showing that when eGFR drops below 60, the risk of death or end-stage kidney disease increases (PMID 27057075). A meta-analysis pooling several general population samples worldwide also supported the threshold of 60, adding that the range of 60 to 74 could represent early-stage chronic kidney disease (PMID 22038337). In this analysis, the relationship between eGFR and mortality was J-shaped, showing elevated risk both below 75 and above 120, and this pattern was similar in those under 65 and those 65 and older. The authors added that this conclusion should be re-examined in eGFR ranges higher than normal using calculation formulas with less bias. The high end of the J-shape carries the caveat that it may be a limitation of the calculation formula.

As people age, the glomerular filtration rate naturally declines (PMID 27057075). Those who argue against a single threshold point out that in older adults with an eGFR slightly below 60, the risk of death and end-stage kidney disease is minimal, whereas in younger individuals with an eGFR slightly above 60, the risk is clearly elevated (PMID 27057075). Some studies report that the eGFR at which risk begins to rise is lower in older adults and higher in younger adults (PMID 38411155). Therefore, age-specific thresholds have been proposed: 75 for those under 40, 60 for those aged 40 to 65, and 45 for those over 65 without urinary protein (PMID 27057075).

When adults in Alberta, Canada, whose decline in kidney function was newly identified between 2009 and 2017 are divided by two different thresholds, the difference becomes apparent (PMID 34459844). The number of people newly classified with chronic kidney disease per 100,000 population per year was 537 using the single threshold of 60, and 343 using the age-specific thresholds. Among the 72,703 people who became patients solely due to the single threshold, 75% were aged 65 or older with an eGFR of 45 to 59 and normal or mildly elevated urine albumin. Their 5-year risk of kidney failure was 0.12% or lower, which was similar to a comparison group without chronic kidney disease, and their mortality risk was 69 to 935 times greater than their risk of kidney failure, depending on their age group. The authors concluded that a threshold independent of age may inflate the burden of chronic kidney disease in older adults and lead to overdiagnosis and unnecessary interventions. Overdiagnosis is the labeling of a condition with a disease name that would never have caused symptoms or harm in a person's lifetime. The issue where some patients increased by lowering or expanding the threshold are cases of overdiagnosis will be continued when the next installment of this series covers overdiagnosis.

There is also evidence on the opposing side. In data from approximately 2.5 million people across 7 European countries, the proportion of individuals over 65 with a low eGFR was 45% using the international criterion of 60, and 19% using the age-specific threshold (PMID 40674065). The authors concluded that before incorporating such thresholds into the definition of chronic kidney disease, their relationship with outcomes must first be confirmed. In a study following 4,952 individuals from the general population for up to 15 years, age-specific thresholds did not improve the prediction of rapid kidney function decline and mortality (PMID 39990905). In this study, 12 out of 55 individuals over 65 who were reclassified as not having chronic kidney disease died during follow-up, and there were no adverse outcomes among 24 individuals under 40 newly classified with chronic kidney disease. Conversely, in data from 17,686 community-dwelling older adults, the risks of disability-free survival or major adverse cardiovascular events mainly appeared below an eGFR of 45 (PMID 42027555). Age-specific thresholds remain a proposal without a settled conclusion.

Thresholds for body size also represent differing risks across populations. Body mass index is a value obtained by dividing body weight (kg) by the square of height (m). A 2004 World Health Organization expert consultation concluded that among Asian populations, the proportion of individuals at high risk for type 2 diabetes and cardiovascular disease is substantial even below 25, the conventional overweight threshold (PMID 14726171). The threshold where risk begins to appear ranged from 22 to 25 across different Asian populations. Therefore, instead of redefining thresholds for each population and abandoning the international classification, the consultation added public health action points such as 23.0 and 27.5. In a study tracking about 1.47 million people using English primary care records, the values reaching the same incidence rate of type 2 diabetes as a BMI of 30 in white individuals were 23.9 for South Asians, 26.6 for Arabs, 26.9 for Chinese, and 28.1 for Black individuals (PMID 33989535). The authors concluded that race-specific thresholds should be revised.

There is also evidence where age acted to erase a threshold. In an analysis pooling 358,707 participants from 51 trials, the proportion of major cardiovascular events reduced by lowering systolic blood pressure by 5 mmHg was 18% in those under 55, around 9% in each of the three age groups spanning 55 to 84, and showed no clear difference in those aged 85 and older (PMID 34461040). There was no evidence that this proportion varied by initial blood pressure category, down to individuals with a baseline blood pressure below 120/70, and the absolute number of events prevented was greater in the older cohorts. The authors concluded that international guidelines should remove age-dependent blood pressure thresholds. Even with the exact same numbers, the risk varies depending on whose slope you are standing on, and how to capture those differences in a threshold is still in motion.

The People Who Draw the Line — Panels, Evidence, and Conflicts of Interest

Out of 16 American guidelines covering the definitions of 14 common diseases, 10 recommended broadening the definitions, 1 recommended narrowing them, and 5 had unclear impacts; these are documents published between 2000 and 2013 (PMID 23966841). Expanding a disease definition means changing the diagnostic criteria to increase the share of people classified as having that condition. There were three ways they were broadened: creating a new "pre-disease" stage, lowering the diagnostic threshold, and diagnosing at an earlier point or through a different method. The reasons put forward included standardizing diagnostic criteria and new evidence regarding the risk in people previously considered disease-free. However, not a single document rigorously examined the harms that might arise when definitions were broadened.

The same study also looked at the relationships between panel members and the industry. A conflict of interest is a situation where a person making a judgment has a relationship through which they can gain money or other benefits depending on that judgment. Among 14 panels that disclosed relationships, the share of members who had ties with pharmaceutical or medical device companies averaged 75%, and the chairs of 12 panels had such ties (PMID 23966841). The authors noted the limitation that only disclosed relationships could be seen. This study showed that ties with the industry are common, but it did not show whether those ties caused the definitions to be broadened. The rules dealing with conflicts of interest are also uneven. In a study reviewing 57 guideline development manuals, all 54 manuals that included conflict-of-interest regulations required the disclosure of relationships, but only 35.2% defined what a conflict of interest actually was (PMID 42740742).

Tools to reduce these problems have also emerged. A 13-member task force including affiliates of the Guidelines International Network and the World Health Organization published an 8-item checklist for panels to consider before changing a disease definition in 2017 (PMID 28505266). These are how the definition changes, how many people are affected, what triggered the call for change, how well the new definition predicts future outcomes, how accurate the definition is, the potential benefits, the potential harms, and the balance between benefits and harms. This task force suggested that guideline groups test and validate this checklist moving forward.

How solid the evidence supporting a line is is also noted separately. The level of evidence is a grade assigned based on what kind of study supports a given recommendation. American cardiology guidelines assign an A if supported by multiple randomized trials or one large randomized trial, a B if supported by observational studies or a single randomized trial, and a C if based only on expert opinion (PMID 30874755). Among 2,930 recommendations included in 26 American cardiology guidelines from 2008 to 2018, 8.5% were A, 50.0% were B, and 41.5% were C; among 3,399 recommendations in 25 European cardiology guidelines, 14.2% were A and 54.8% were C (PMID 30874755). The authors concluded that this pattern did not clearly improve between 2008 and 2018. These numbers count all recommendations together, including diagnostic thresholds.

How trustworthy the evidence is and how strongly a recommendation is made are different questions. The strength of recommendation is a grade assigned to how strongly a recommendation should be made by weighing the balance between benefits and harms, how reliable the evidence is, and costs together. The GRADE system, which grades these two separately, was created to harmonize the grading methods that varied across different guidelines, and it is adopted by institutions in multiple countries (PMID 15205295, PMID 18436948). A review tracking back 32 hypertension guidelines noted that while the methods for evaluating evidence improved significantly over time, situations where conflicts could arise were visible where strong recommendations were made on weak evidence (PMID 42029627). The line is co-created by the share determined by evidence and the share determined by people.

Common Misunderstandings — and One Person's Blood Pressure

The first misunderstanding is "if it's outside the range, it's a disease." By definition, reference ranges leave 5% of healthy people outside them (PMID 40268322). The second misunderstanding is the opposite: "if it's inside the range, it's fine." Just like the example of a spirometry test, someone who is sent to a pulmonary function testing laboratory and has a high pre-test probability of disease may still have a disease even if they score above the lower limit of normal (PMID 22222132). Both misunderstandings arise from reading reference ranges as the boundaries of disease.

The third misunderstanding is "lines are drawn by people, so they are meaningless." Although judgment goes into where a line is placed, the slope the line rests on is real. In an observational study from China, the risk in the stage 1 hypertension range was higher than normal blood pressure (PMID 37319412), a lower target in SPRINT reduced cardiovascular events and death (PMID 26551272), and the 6.5% hemoglobin A1c line stands on top of retinopathy data (PMID 22891221). Because side effects also increased in the same SPRINT study, lowering the target comes with both benefits and costs.

The fourth misunderstanding is "the line was lowered because of pharmaceutical companies." While it is true that panel members writing definitions for common diseases disclosed ties averaging 75%, causality that those ties produced the decisions was not confirmed, and the reasons put forward by panels also included new risk evidence (PMID 23966841). The fifth misunderstanding is "when the line goes down, the number of people newly recommended medication increases by that much." In an estimate applying JNC7 and the 2017 US guideline to the same US survey data from 2011–2014, the share with hypertension rose from 31.9% to 45.6%, but the share recommended medication rose only slightly from 34.3% to 36.2% (PMID 29133599). There was also a shift on the medication side. In the same estimate, among people already taking medication, the share whose blood pressure remained above the treatment target increased from 39.0% to 53.4%, leading the authors to conclude that many people would lower their blood pressure more aggressively. In an estimation applying the 2017 US guideline to a 2011–2012 sample of Chinese adults aged 45–74, the share recommended medication also rose by 7.5 percentage points, from 41.6% to 49.1% (PMID 32521545).

Now let us follow a hypothetical individual. Suppose a 55-year-old office worker has an office blood pressure reading of 134/84 (this is a hypothetical number). By the JNC7 line, this was not hypertension, and by the 2017 US guideline line, it is stage 1 hypertension. By the European guideline diagnostic line of 140/90 current as of 2019, they do not receive a diagnosis of hypertension (PMID 31030859). The same blood pressure receives a different name depending on which guideline you open. Even within the 2017 US guideline, whether to start medication depended on this person's absolute cardiovascular risk. That guideline stipulated that people without high risk should start medication at 140/90 or higher, while those aged 65 or older or with diabetes, chronic kidney disease, or previously diagnosed cardiovascular disease were considered high risk and should start medication at 130/80 or higher, with exceptions made for preventing recurrent stroke (PMID 31521481). The commentary for this guideline also notes that most errors in the blood pressure measurement process tilt toward raising the values, leading to overdiagnosis of hypertension (PMID 31521481). How blood pressure changes depending on where it is measured was covered in "What Blood Pressure Measures — Two Numbers, and an Answer That Changes With Where You Measure". And in 2025, a new US guideline replaced the 2017 US guideline (PMID 40811497). Lines continue to be redrawn.

What this person will take away from this article are questions to ask in the clinic. The judgment of whether their blood pressure is fine comes after finding the answers to those questions together with their medical team. Questions for identifying reliable guidelines include whether the strength of the recommendation matches the certainty of the evidence, whether conflicts of interest could have influenced the recommendation, and whether all outcomes important to patients were considered (PMID 37386667). Adapted to the content of this article, these become the following questions:

• Was this line derived from the distribution of healthy people, or from the relationship with outcomes such as complications or death?
• Does this line imply the same risk for my age and the group I belong to?
• What is my absolute risk, and what do I gain and endure if I treat it?
• How strong is this recommendation, and how reliable is the evidence behind it?

When you know what a line is made of, instead of judging your test results alone, you see what to ask your healthcare provider. In weak or conditional recommendations, weighing the patient's circumstances, values, and preferences is decisive (PMID 37386667).

Where to Look Further

It has not yet been confirmed whether the age-specific eGFR cutoffs filter out future outcomes any better (PMID 40674065), and research has also emerged suggesting that establishing race- and ethnicity-tailored body mass index obesity classifications is urgent (PMID 39223976). Regarding the problem of blood pressure guidelines drawing different lines, a proposal has been made that standardizing the methods for evaluating and interpreting evidence can reduce those differences (PMID 42029627).

Articles to read next in this series are as follows. The performance of a single test and the trade-offs when shifting a cutoff are in "What a Test Actually Tells You — An Introduction to Sensitivity, Specificity, and Pretest Probability", the pretest probability is in "Where Does Pretest Probability Come From? — History, Physical Examination, and Clinical Prediction Rules", treatment targets are in "Better Numbers, Better Patients? — An Introduction to Surrogate Endpoints", and how to read the pros and cons of treatment using absolute risk is in "What Does "50% Lower Risk" Mean in People? — Reading Relative Risk, Absolute Risk, and NNT". Why some of the diagnoses that increased by lowering the cutoff are harmful is covered in the next part of this series on overdiagnosis.

Source Papers

• "Interpretating Normal Values and Reference Ranges for Laboratory Tests." (2025) — PMID 40268322 — Defines the reference range using the middle 95% of at least 120 healthy individuals, which by definition flags 5% of healthy people as abnormal, meaning out-of-range results are far more common than clinical abnormalities, and proposes writing clinical meanings in plain language on lab reports.
• "The range of normal." (1996) — PMID 8796934 — Normal limits are often determined by looking at results when they were called abnormal, and the reference population must be healthy people under the same conditions as the person being tested.
• "Reference intervals." (2008) — PMID 18852866 — Procedures for establishing reference intervals: principles (middle 95%), data sources, age and sex partitioning, and defining clinical significance.
• "Reference intervals: current status, recent developments and future considerations." (2016) — PMID 26981015 — The confusion between reference intervals and clinical decision limits remains a problem, with pediatric and elderly populations posing major challenges.
• "Clinical Practice Guideline for Screening and Management of High Blood Pressure in Children and Adolescents." (2017) — PMID 28827377 — Revisions including new pediatric blood pressure reference tables created using data from normal-weight children.
• "How should the lower limit of the normal range be defined?" (2012) — PMID 22222132 — The lower limit of normal for lung function is the 5th percentile of healthy non-smokers, fixed-ratio rules underdiagnose young and tall people and overdiagnose older and short people, and if pretest probability is high, results above the lower limit may still likely indicate disease.
• "Using nephropathy as an outcome to determine the HbA1c diagnostic threshold for type 2 diabetes." (2024) — PMID 38615570 — 6.5% is based on the prevalence of retinopathy, and among 2,920 people in Qatar, no threshold below 7.0% was the optimal predictor of kidney disease, meaning 6.5% does not need to be changed.
• "New diabetes diagnostic threshold of hemoglobin A(1c) and the 3-year incidence of retinopathy." (2012) — PMID 22891221 — In 19,897 Japanese individuals, the 3-year incidence risk of retinopathy increased at 6.5–6.9%, while 5.5–6.4% showed no evidence of an increase, supporting the validity of 6.5%.
• "Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study." (2000) — PMID 10938048 — Each 1% reduction in HbA1c was associated with a 21% reduction in diabetes-related outcomes and a 37% reduction in microvascular complications, with no risk threshold for any outcome, and the lowest risk below 6.0%.
• "Dichotomizing continuous predictors in multiple regression: a bad idea." (2006) — PMID 16217841 — Dichotomization results in a loss of statistical power and residual confounding, data-driven optimal cutpoints are severely biased, and it is unnecessary for statistical analysis.
• "Prediabetes: a high-risk state for diabetes development." (2012) — PMID 22683128 — Each year, 5% to 10% of people with prediabetes develop diabetes, while the same proportion return to normal blood sugar.
• "Understanding prediabetes: definition, prevalence, burden and treatment options for an emerging disease." (2019) — PMID 30935247 — Viewing prediabetes as a distinct pathological state is controversial, but based on changes in multiple tissues, it is argued that it should be considered a distinct clinical state.
• "Blood pressure and the new ACC/AHA hypertension guidelines." (2020) — PMID 31521481 — Diagnosis at 130/80 or higher, most stage 1 patients are not immediate candidates for medication, high-risk status is determined by absolute cardiovascular risk (starting medication at 130/80 for high risk and 140/90 for non-high risk), and measurement errors mostly raise values leading to overdiagnosis.
• "Potential US Population Impact of the 2017 ACC/AHA High Blood Pressure Guideline." (2018) — PMID 29133599 — In 9,623 individuals from NHANES 2011–2014, hypertension was 45.6% vs. 31.9% (JNC7), medication recommendations were 36.2% vs. 34.3%, non-pharmacological was 9.4%, and those exceeding the target among medication users were 53.4% vs. 39.0% — a large increase in prevalence, a small increase in medication recommendations, and intensified treatment for those already taking medication.
• "Impact of the 2017 ACC/AHA guideline on the prevalence of elevated blood pressure and hypertension: a cross-sectional analysis of 10 799 individuals." (2020) — PMID 33384395 — Applying JNC-7 (140/90) and 2017 ACC/AHA (130/80) definitions to 10,799 Saudi adults showed a large increase in prevalence and a small increase in medication recommendations.
• "The Evolution of Blood Pressure Thresholds and Targets over Time: A Historical Review." (2026) — PMID 42029627 — Across 32 guidelines, thresholds gradually dropped from 160/95, with recent guidelines diverging between WHO/ISH and NICE at 140/90 versus ESH/ESC and ACC/AHA at 130/80, causing conflict when strong recommendations are based on weak evidence, and proposing standardization of methods.
• "Hypertension Guidelines: Effect of Blood Pressure Targets." (2019) — PMID 31030859 — Diagnosis starts at 140/90 in Europe and 130/80 in the US and Canada, and concerns about overtreating low-risk groups must be weighed against evidence from SPRINT and meta-analyses.
• "Potential Impact of the 2017 High Blood Pressure Guideline Beyond the United States: A Case Study of the People's Republic of China." (2020) — PMID 32521545 — In 11,822 Chinese adults aged 45–74, estimated hypertension increased from 44.1% to 56.4% and medication recommendations from 41.6% to 49.1%, potentially reducing risk by selectively recommending medication to those who will benefit (depending on acceptance and adherence).
• "Impact of the 2017 ACC/AHA guidelines on the prevalence of hypertension among Indian adults: Results from a cross-sectional survey." (2020) — PMID 33465185 — In 180,335 Indian adults, JNC8 was 29.7% vs. 2017 ACC/AHA 63.8%.
• "Association of stage 1 hypertension defined by the 2017 ACC/AHA guideline with cardiovascular events and mortality in Chinese adults." (2024) — PMID 37319412 — Untreated stage 1 hypertension had a composite hazard ratio of 1.20 for myocardial infarction, stroke, and death compared to normal blood pressure (Chinese observational cohort).
• "A Randomized Trial of Intensive versus Standard Blood-Pressure Control." (2015) — PMID 26551272 — In 9,361 participants from SPRINT, a target below 120 reduced composite outcomes (hazard ratio 0.75) and death (0.73) compared to below 140, while increasing hypotension, syncope, electrolyte abnormalities, and acute kidney injury, with no increase in injurious falls.
• "Final Report of a Trial of Intensive versus Standard Blood-Pressure Control." (2021) — PMID 34010531 — The final report and post-trial follow-up showed similar patterns of benefits and adverse effects, while the difference in heart failure disappeared.
• "Benefit-harm trade-offs of intensive blood pressure control versus standard blood pressure control on cardiovascular and renal outcomes: an individual participant data analysis of randomised controlled trials." (2025) — PMID 40902616 — In 80,220 participants across 6 trials (82.6% Asian), cardiovascular events were absolutely reduced by 1.73% and adverse events increased by 1.82%, resulting in a positive net benefit using decision weights.
• "Age-stratified and blood-pressure-stratified effects of blood-pressure-lowering pharmacotherapy for the prevention of cardiovascular disease and death: an individual participant-level data meta-analysis." (2021) — PMID 34461040 — In 358,707 participants across 51 trials, per 5 mmHg hazard ratios were 0.82 for under 55, 0.91 for 55–84, and 0.99 (not significant) for 85 and older, with no evidence of a difference by initial blood pressure category, absolute reductions were greater in older adults, and proposing the removal of age-specific blood pressure thresholds.
• "Has the time come for age-adapted glomerular filtration rate criteria to define chronic kidney disease: how soon is now?" (2024) — PMID 38411155 — The fixed eGFR threshold of below 60 means the eGFR at which risk begins to rise is lower in older adults and higher in young adults, leading to overdiagnosis in older adults and underdiagnosis in young adults.
• "An Age-Calibrated Definition of Chronic Kidney Disease: Rationale and Benefits." (2016) — PMID 27057075 — The threshold of 60 has been defended based on epidemiological evidence, but it fails to reflect age-related decline, proposing age-specific thresholds of 75/60/45.
• "A population-based approach for the definition of chronic kidney disease: the CKD Prognosis Consortium." (2012) — PMID 22038337 — The relationship between eGFR and death is J-shaped (increased risk below 75 and above 120), supporting the threshold of 60 while noting that 60–74 may represent early CKD.
• "Accounting for Age in the Definition of Chronic Kidney Disease." (2021) — PMID 34459844 — In Alberta, new CKD per 100,000 person-years was 537 vs. 343, of the 72,703 individuals identified by the fixed threshold alone, 75% were aged 65 or older with an eGFR of 45–59, and their 5-year kidney failure risk was 0.12% or lower, similar to the control group, raising concerns about overestimation and overdiagnosis.
• "Prevalence of reduced eGFR in European adults using KDIGO and age-adapted eGFR thresholds." (2025) — PMID 40674065 — In roughly 2.5 million European adults, the prevalence for over 65 was 45% with KDIGO vs. 19% with age-adapted thresholds, and the relationship with outcomes prior to incorporation into definitions needs to be confirmed.
• "Impact of Different CKD Definitions on Long-Term Renal Function and Mortality in a Population-Based Cohort Study." (2025) — PMID 39990905 — In 4,952 individuals, age-specific definitions did not improve the prediction of rapid functional decline or mortality.
• "Evaluating Long-Term Outcomes Across eGFR Equations in Older Adults." (2026) — PMID 42027555 — In 17,686 older adults, the risks of reduced disability-free survival, death, and major cardiovascular events were primarily concentrated below an eGFR of 45.
• "Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies." (2004) — PMID 14726171 — WHO expert consultation found that a substantial proportion of high-risk individuals in Asian populations have a BMI below 25, with risk starting points varying across populations from 22 to 25, leading to the decision to maintain international classifications while introducing action points at 23.0/27.5/32.5/37.5.
• "Ethnicity-specific BMI cutoffs for obesity based on type 2 diabetes risk in England: a population-based cohort study." (2021) — PMID 33989535 — In roughly 1.47 million people, the diabetes incidence equivalent to a BMI of 30 in white people was 23.9 for South Asian, 28.1 for Black, 26.9 for Chinese, and 26.6 for Arab individuals, indicating a need to revise thresholds by ethnicity.
• "Comparison of race- and ethnicity-specific BMI cutoffs for categorizing obesity severity: a multicountry prospective cohort study." (2024) — PMID 39223976 — Concluding that obesity severity thresholds based on identical diabetes risk criteria vary by race, making race- and ethnicity-tailored classifications an urgent priority.
• "Expanding disease definitions in guidelines and expert panel ties to industry: a cross-sectional study of common conditions in the United States." (2013) — PMID 23966841 — Out of 16 conditions, 10 definitions were expanded and 1 narrowed, with expansion occurring in 3 distinct ways, without rigorous evaluation of harm, and with an average of 75% of disclosing panel members and 12 committee chairs having industry ties, subject to the limitation that only disclosed relationships were examined.
• "A cross-sectional review of policies on conflicts of interest and funding in the development manuals of practice guidelines." (2025) — PMID 42740742 — Across 57 guideline manuals, all 54 with COI policies required disclosure, while only 35.2% defined COI.
• "Guidance for Modifying the Definition of Diseases: A Checklist." (2017) — PMID 28505266 — An 8-item checklist for modifying disease definitions by a 13-member working group, proposing pilot testing and validation by guideline groups.
• "Levels of Evidence Supporting American College of Cardiology/American Heart Association and European Society of Cardiology Guidelines, 2008-2018." (2019) — PMID 30874755 — Among 2,930 ACC/AHA recommendations, 8.5% were Level of Evidence A, 50.0% were B, and 41.5% were C, while among 3,399 ESC recommendations, 14.2% were A and 54.8% were C, with no improvement between 2008 and 2018.
• "Grading quality of evidence and strength of recommendations." (2004) — PMID 15205295 — Recommendation strength is judged by weighing the balance of benefits and harms, quality of evidence, applicability, certainty of baseline risk, and costs.
• "GRADE: an emerging consensus on rating quality of evidence and strength of recommendations." (2008) — PMID 18436948 — Grading methods varied across guidelines, with GRADE currently being adopted by multiple institutions worldwide.
• "How to recognize a trustworthy clinical practice guideline." (2023) — PMID 37386667 — 6 questions for a trustworthy guideline (consistency between recommendation strength and certainty of evidence, management of conflicts of interest, patient-important outcomes, etc.), noting that weaker recommendations place greater importance on patient values and preferences.
• "2025 AHA/ACC/... Guideline for the Prevention, Detection, Evaluation and Management of High Blood Pressure in Adults." (2025) — PMID 40811497 — The 2025 guideline replacing the 2017 ACC/AHA hypertension guideline.