Treatment Thresholds
Pauker and Kassirer's threshold model: when should you treat without testing, test first, or do neither? Set how much treatment helps and harms and how good and how risky the test is, and the model gives the two probabilities that split every patient into the three choices.
Set up the decision
Treatment
Test
The thresholds
Expected harm per 100 patients (lower is better) against pretest probability of disease (across). The shaded line under the three is the best choice. Click or drag on the chart to place your patient.
Data table
Your patient
From a clinical score or your gestalt; PE Workup gives one for suspected PE.
How the thresholds are calculated
Everything is counted in one currency: bad outcomes per 100 patients, where one unit is a death or anything you judge equally bad. B is the net benefit of treating someone who has the disease, H the net harm of treating someone who doesn't, and R the harm of the test itself, which every tested patient bears. For a pretest probability p:
Each threshold is where two of those lines cross:
If the testing threshold comes out at or above the test–treatment threshold, the test line never dips below the other two, testing never pays, and the treatment threshold alone splits the patients.
Reading the picture
- The treatment threshold depends only on benefit and harm. Treat when p·B exceeds (1−p)·H, that is when the odds of disease exceed H/B. A safe, effective drug (meningitis) has a threshold near zero; a toxic one (chemotherapy) needs near certainty.
- A test is worth doing only where it can change what you do. Below the testing threshold even a positive result wouldn't justify its own harm and its false positives; above the test–treatment threshold you'd treat even after a negative one. With a harmless test (R = 0) the thresholds are exactly the pretest probabilities at which a positive result just reaches, or a negative result just falls to, the treatment threshold.
- Raise the test's harm and the window shrinks from both ends. The test line moves straight up. Past the "most harm the test can carry" figure the window closes: treat or don't on clinical grounds alone.
- Sensitivity mostly moves the upper threshold, specificity the lower one. A test that misses disease can't safely rule out a likely diagnosis; a test that raises false alarms can't safely be spent on an unlikely one.
- It's the ratio that matters. Doubling B, H and R together changes nothing. That is why getting patients' and clinicians' values for each outcome right matters more than precision about any one of them.
The model assumes one test, one treatment and a yes-or-no disease; the test's result is acted on; benefits and harms add up linearly; and a test's harm is the same for everyone tested. Preset values are illustrative, chosen to show the shape of each decision, not taken from one study or meant for a real patient.
Sources and further reading
- Pauker SG, Kassirer JP. The threshold approach to clinical decision making. N Engl J Med 1980;302:1109–17.
- Pauker SG, Kassirer JP. Therapeutic decision making: a cost-benefit analysis. N Engl J Med 1975;293:229–34. (The treatment threshold.)
- Hunink MGM, Weinstein MC, Wittenberg E, et al. Decision Making in Health and Medicine: Integrating Evidence and Values. 2nd ed. Cambridge University Press, 2014.
- Kline JA, Mitchell AM, Kabrhel C, Richman PB, Courtney DM. Clinical criteria to prevent unnecessary diagnostic testing in emergency department patients with suspected pulmonary embolism. J Thromb Haemost 2004;2:1247–55. (The PERC rule, built to find patients below a testing threshold of about 2%.)
The same idea, in the other direction: Barriers to Ordering asks whether friction moves clinicians' own testing thresholds in the right patients.