account the local prevalence and presentations of tuberculosis. Many different isolation strategies have been reported.

Pegues and colleagues (29) studied the impact at the Massachusetts General Hospital from 1993 through 1994 after the implementation in 1993 of a tuberculosis isolation algorithm. The algorithm includes typical signs and symptoms (chronic cough, fever, weight loss, etc.) and risk factors (HIV infection, homeless, intravenous drug abuse [IVDA], jail, immigration from a country where tuberculosis is endemic, etc.), as well as the CXR. If the patient had a normal CXR, then the patient was not placed in isolation. If the CXR was abnormal, then a risk evaluation was done. If low risk, the patient was placed in a private room until one smear was AFB negative. If the patient was at moderate risk (i.e., had risk factors or a suspicious CXR), the likely degree of infectivity was considered. If the patient was judged to be likely infectious (as determined by the presence of a cavity on CXR or cough/sputum production by history), then the patient was placed in a tuberculosis isolation room. Otherwise the patient was placed in a lesser isolation room until three sputum samples were shown to be AFB negative.

There were 31 case patients with pulmonary tuberculosis over the 2-year study period (out of 58 patients with + AFB smears). All had an abnormal CXR, and 9/31 (29 percent) had cavitary disease. Ages ranged from 7 months to 97 years.

Isolation was initiated within 24 hours of admission in 19/31 (61 percent), 17 in the ED. Of 12 patients not isolated appropriately, 7 were eventually isolated (after 2 to 31 days; median, 9 days), and 5 were never isolated during admission (range, 3 to 28 days; median, 4 days). Reasons for inappropriately not isolating the patients included misclassified risk factors for five patients (three with HIV infection); seven patients had atypical or misinterpreted (but abnormal) CXRs and were not captured by the algorithm because they had no risk factors. No data on the total number of patients isolated are presented.

The 12 patients inappropriately not isolated led to 136 patient-exposure days. Of 11 roommates and 281 employees exposed, no PPD conversions or cases of active tuberculosis were found.

In the discussion the authors note that if the five patients who should have been isolated by the algorithm had been isolated, the sensitivity would be 77 percent. Inclusion of other risk factors (such as end-stage renal disease and residence at a long-term-care facility) would have improved the sensitivity, but at the cost of much more overisolation. Unfortunately, there is no discussion as to whether the new algorithm led to improvements in isolation practices compared with the previous policies.

The results of a survey including isolation practices in 159 Veterans Affairs hospitals (100% response, but not on all questions) were reported by Roy (30). Overall, 1,063 patients/month were isolated (median, 3 per