Probability revision in general practice: the cases of occult blood in stool in patients with indigestion, and daily smoking in patients who cough

Abstract

Prior calculation of the sex- and age-distributed positive predictive value (PPV) of a symptom in relation to a disease forms the basis for probability revision based on new information about signals observed during the medical encounter: a symptom, a sign, a piece of information based on the medical history, a laboratory test. Probability revision requires estimates of sensitivity and specificity of the signal. Probability revision following positive and negative signals have been made for occult blood in the stool in relation to colorectal cancer in patients presenting in general practice with indigestion, and for daily smoking in relation to lung cancer in general practice patients with coughs. A positive test for occult blood in the stool gives an approximately fivefold rise in PPV for all age groups. For patients 50 years of age and more PPV exceeds 10%, and for patients 70 years old or more PPV is 18.9% for men and 25.0% for women when the test is positive. As expected, the predicted value decreases when the test is negative, but for old persons the predicted value is about 1 % even when the test is negative. Daily smoking in 60- to 69-year-old patients with coughs increases PPV from 6.4% 10 8.5% for men, and from 0.8% to 2.0% for women. The likelihood ratio (LR) also rises. Daily smoking is much more common than coughing in the general population, and daily smoking in itself therefore has a lower PPV for lung cancer than a persistent cough. The change in probabilities after revision is illustrated. The results confirm that a signal conditionally independent of signals used in the calculation of prior probabilities is more useful than a signal that cannot be assumed to be conditionally independent for the patients studied.