Objective: To evaluate potential clinical predictors of bacteremia in hospitalized geriatric patients and to propose an individual risk score as an alternative to "subjective" clinical judgment for a more efficient approach in early recognition and treatment of bacteremia.
Design: A 16-month prospective study.
Setting: The University Geriatric Hospital of Geneva, Switzerland.
Patients: Four hundred thirty-eight patients aged 62 years or older in whom 558 episodes of bacteremia were suspected.
Measurements: The unit of evaluation was the blood culture episode, which was defined as a 48-hour period beginning with the drawing of the first blood for culture. An extensive precoded protocol, including clinical and biological data, was completed by the resident who requested the blood cultures. For each episode, the resident also provided a subjective assessment of the probability of bacteremia. Odds ratios and their variances were used to estimate the relative risks of potential predictors of bacteremia. The performance of a predictive clinical model based on risk score threshold was evaluated by means of a receiver-operating characteristic analysis.
Results: Of the 558 potentially bacteremic episodes investigated, 46 (8.2%) yielded positive blood cultures. The bacteremia rate was strongly associated with the type of episode: it reached 15.6% among the community-acquired (CA) episodes (those occurring within 48 hours of hospital admission) and 6.0% only among the hospital-acquired (HA) episodes (those occurring after the first two days of hospitalization). Predictors of bacteremia with highest relative risks included: bladder catheter removal, fever (> or = 38.5 degrees C), rigors, shock, total band count > or = 1500/mm3, and lymphocyte count < or = 1000/mm3. When assessed by episode type, it appeared that bladder catheter removal and rigors were good predictors of bacteremia in HA episodes only, whereas fever (> or = 38.5 degrees C) had a good predictive value in CA episodes only. The performance of the clinical model was two times better than the physician's subjective ability to predict bacteremia when the threshold of the risk score was fixed at two or more predictors per episode.
Conclusions: These findings provide means to identify older hospitalized patients at high risk of bacteremia. Although the proposed predictive model will need further validation and more precise evaluation of the potential benefits, it may nevertheless be of some help in early recognition and treatment of bacteremia.