Skip to content

Stopping Rule Application

Why It Matters

Without explicit stopping rules, information gathering continues indefinitely - there's always one more test that could increase confidence. This wastes resources and delays treatment. Stopping rules correct the systematic error of conflating diagnostic certainty with actionable confidence. You don't need certainty; you need sufficient confidence to act.

The Key Move

Before testing begins, specify: "If any hypothesis reaches X% probability, I'll treat for it. If all serious hypotheses fall below Y%, I'll pursue conservative management." Then stick to these thresholds - don't move the goalposts. The thresholds depend on treatment risk and disease severity: high-risk treatments and low-severity diseases require high confidence; low-risk treatments and high-severity diseases permit lower confidence.

Classic Application

Antibiotic treatment for suspected sepsis. Define stopping rule prospectively: "If probability of bacterial infection exceeds 30%, start antibiotics immediately. If below 10% after initial workup, observe without antibiotics." This prevents both under-treatment (waiting for culture confirmation while sepsis progresses) and over-treatment (antibiotics for obvious viral infections). The threshold of 30% reflects that antibiotic risks are low and sepsis mortality is high.

Surprising Application

Hiring decisions. Define stopping rule: "If after 3 interviews, candidate probability of success exceeds 70%, make offer. If below 40%, reject. Between 40-70%, conduct additional interviews (maximum 5 total)." This prevents both premature offers (after one good interview) and indefinite evaluation (perpetual uncertainty). The thresholds reflect hiring cost, opportunity cost of delayed hire, and cost of bad hire.

Failure Modes

Setting thresholds arbitrarily rather than based on decision analysis. The tool requires explicit reasoning about treatment risks and disease consequences. Fails when stopping rules are too rigid - sometimes exceptional cases warrant breaking the rule. Also fails when rules are performative - stated but not followed, which is worse than no rule because it creates false accountability. The worst failure is treating stopping rules as permanent when they should adapt to changing evidence about test accuracy or treatment effectiveness.

Go Deeper

Pauker, S.G. & Kassirer, J.P. (1980). "The Threshold Approach to Clinical Decision Making," New England Journal of Medicine, 302(20), 1109-1117; Wald, A. (1945). "Sequential Tests of Statistical Hypotheses," Annals of Mathematical Statistics, 16(2), 117-186.