I teach doctors how to interpret test results – and the labels ‘normal’ and ‘abnormal’ often hide what matters most

Mr. V experienced chest tightness and shallow breathing, leading him to seek emergency care. Doctors initially suspected heart failure due to symptoms like swollen hands, legs, and an enlarged heart on an X-ray. A blood test measuring BNP (B-type natriuretic peptide), a hormone released by the heart under strain, returned a result of 14 picograms per milliliter (pg/mL). While doctors typically consider BNP levels below 100 pg/mL as normal, a value this low (under 50 pg/mL) is uncommon in heart failure, even in overweight patients. The resident treating Mr. V correctly recalled that obesity can falsely lower BNP levels, so he dismissed the normal result and prescribed a diuretic to reduce fluid buildup. However, Mr. V did not improve, and a colleague later diagnosed him with cardiac tamponade, a life-threatening condition where fluid compresses the heart, impairing its ability to pump. The diuretic could have worsened his condition, but timely surgery saved his life. This case highlights how labels like 'normal' can mask critical details in medical test results.

Medical tests often categorize results as 'normal' or 'abnormal' based on predefined cutoffs, but these labels oversimplify complex biological data. For example, the D-dimer test, used to detect blood clots, flags levels above 500 nanograms per milliliter (ng/mL) as abnormal. However, values between 500 and 1,000 ng/mL slightly decrease the likelihood of a blood clot, making them nearly as reassuring as 'normal' results for low-risk patients. Similarly, LDL cholesterol (often called 'bad' cholesterol) is considered normal below 100 milligrams per deciliter (mg/dL), but for patients who have already had a heart attack or stroke, levels above 55 mg/dL are considered too high. These examples show that a single cutoff does not capture the nuanced information a test result provides about a patient's health.

The labels 'normal' and 'abnormal' do not reflect how strongly a test result supports or rules out a diagnosis. For instance, Mr. V's BNP of 14 pg/mL was not just normal—it was so low that it strongly argued against heart failure. Yet the label 'normal' treated it the same as a BNP of 95 pg/mL, obscuring critical diagnostic clues. This issue arises because test results are often reduced to binary categories for simplicity, despite the fact that human biology is far more complex. Clinicians trained to interpret these labels may miss subtle but important information that could change a patient's treatment. The article emphasizes that test results should be viewed as continuous data, not just as 'normal' or 'abnormal'.

To address the limitations of binary cutoffs, researchers developed interval likelihood ratios, a metric that quantifies how much a test result changes the odds of a specific diagnosis. Unlike simple cutoffs, this method divides test results into ranges and compares how often patients with and without a diagnosis fall into each range. For example, a D-dimer level between 500 and 1,000 ng/mL might slightly decrease the likelihood of a blood clot, while a level above 2,000 ng/mL could significantly increase it. A study involving over 600 clinicians across four U.S. academic institutions found that those who saw interval likelihood ratios made correct decisions 93% of the time, compared to 82% for those relying on standard cutoffs. This suggests that providing more detailed information could improve diagnostic accuracy and patient outcomes.

The way test results are displayed can influence clinical decisions and patient safety. A result labeled 'abnormal' may pressure clinicians to order additional tests or treatments, even when their judgment suggests otherwise. Interval likelihood ratios provide an evidence-based rationale for holding off on further interventions, potentially reducing unnecessary procedures and associated risks. The article notes that diagnostic errors remain a leading cause of harm in medicine, and improving how test results are presented could help mitigate these errors. However, the study's findings are based on simulated scenarios, and further research is needed to determine if these benefits hold true in real-world clinical settings across different institutions and patient populations.

The 21st Century Cures Act, signed into law in 2016, gave patients greater access to their medical test results through online portals, often before clinicians have reviewed them. While this increases transparency, it also introduces challenges, such as patients seeing 'abnormal' results and feeling alarmed, or 'normal' results providing false reassurance. Clinicians may feel obligated to act on 'abnormal' labels due to legal concerns, even when their clinical judgment suggests otherwise. Interval likelihood ratios could help address this by providing documented, evidence-based reasons for not pursuing further testing. However, integrating this approach into electronic health records requires access to high-quality research studies to derive these ratios, which is not currently standard practice.