facility). In 1993, a total of 974 patients were diagnosed with pulmonary tuberculosis (median, 3 per facility). The ratio of patients isolated/patients with pulmonary tuberculosis ranged from 1 to 120 (median, 12). There was no correlation between this ratio and the number of tuberculosis patients at the facility. Unfortunately there are no data presented on health care worker PPD conversion rates. Nevertheless, the variability in the degree of overisolation is striking. The methods used to determine who should be isolated were not discussed.
Columbia University has a renowned medical informatics group, and not surprisingly, an informatics approach to tuberculosis isolation was evaluated there, as reported by Knirsch and colleagues (31).
A clinical protocol for tuberculosis isolation was implemented in 1992 (17). Tuberculosis isolation was to be initiated (and continued until three negative AFB smears were obtained) in patients with a CXR suggesting tuberculosis (e.g., cavitary lesion, or any abnormality on CXR for patients with HIV infection) plus HIV risk factors or homelessness. Overall prompt isolation of tuberculosis patients improved from 51 percent in 1992 to 75 percent in 1993.
An automated protocol of computer screening of records was developed in 1995 using the CXR as the starting point. CXR reports were already automatically parsed at Columbia, so terms suggesting tuberculosis could be checked for. If the CXR was abnormal, immunodeficiency status was checked from other records (e.g., laboratory and pharmacy records). The hospital epidemiologist was notified via a computer generated e-mail to review the record for anyone meeting the preselected criteria. In 1995–1996 the combined clinical and automated protocol correctly isolated 34/43 (79 percent) of patients with tuberculosis. The clinical system alone would have isolated 30/43 (70 percent). The automated alert system flagged the records of 22/43 patients (51 percent). The automated protocol generated 15 alerts for every culture-positive tuberculosis patient, which was thought to be a tolerable number. By its nature, the system failed to detect patients with a normal CXR and patients with an abnormal CXR but no evidence of HIV infection—these accounted for most of the 21 percent not isolated by either system.
An effort to improve the isolation protocol at Grady was reported by Bock and colleagues (32). The charts of 376 patients (12 percent of all medicine admissions) on tuberculosis isolation from October through December 1993 were reviewed shortly after admission. Of these, 53 had pulmonary tuberculosis and 51 (96 percent) had been appropriately isolated. The two patients missed should have been isolated under existing protocols. Thus, 7.4 patients were isolated for every case of tuberculosis (positive predictive value, 14 percent).
A total of 295 of these patients (42 with tuberculosis) agreed to be interviewed. The authors evaluated 15 variables available on admission.