Long-term Survival and Function After Suspected Gram-negative Sepsis

OBJECTIVE: To determine the long-term (> 3 months) survival of septic patients, to develop mathematical models that predict patients likely to survive long-term, and to measure the health and functional status of surviving patients. SETTING: A large tertiary care university hospital and an associated Veterans Affairs Medical Center. DESIGN: From December 1986 to December 1990, a total of 103 patients with suspected gram-negative sepsis entered a double-blind, placebo-controlled efficacy trial of monoclonal antiendotoxin antibody. Of these, we followed up 100 patients for 7667 patient-months. Beginning in May 1992, we reviewed hospital records and contacted all known survivors. We measured the health status of all surviving patients. MAIN OUTCOME MEASURES: The determinants of long-term survival (up to 6 years) were identified through two Cox proportional hazard regression models: one that included patient characteristics identified at the time of sepsis (bedside model) and another that included bedside, infection-related, and treatment characteristics (overall model). RESULTS: Of the 60 patients in the cohort who died at a median interval of 30.5 days after sepsis, 32 died within the first month of the septic episode, seven died within 3 months, and four more died within 6 months. In the bedside multivariate model constructed to predict long-term survival, large hazard ratios (HRs) were associated with severity of underlying illness as classified by McCabe and Jackson criteria (for rapidly fatal disease, HR = 30.4, P < .001; for ultimately fatal disease, HR = 7.6, P < .001) and the use of vasopressors (HR = 2.5; P = .001). In the overall model for long-term survival, severity of underlying illness (rapidly fatal disease, HR = 23.7, P < .001; ultimately fatal disease, HR = 6.5, P < .001), number of active comorbid illnesses (HR = 1.3; P = .04), use of vasopressors at the time of sepsis (HR = 2.0; P = .02), and development of adult respiratory distress syndrome (HR = 2.3; P = .02) predicted patients most likely to die. The Acute Physiology and Chronic Health Evaluation II score was not a significant predictor of outcome when either model included the simpler McCabe and Jackson classification of underlying disease severity. We compared the health status scores with norms for the general population and found that patients with resolved sepsis reported more physical dysfunction (P < .001), including problems with work and activities of daily living (P = .02), and more poorly perceived general health (P < .01). In contrast, patients' scores for perceived emotional health were higher than those in the general population (P = .004). The mean Barthel score of our patients was 85 (100 = total independence) and the mean Eastern Cooperative Oncology Group score was 0.7 (0 = normal, 4 = 100% bedridden), suggesting that the patients' physical function was not normal. CONCLUSIONS: At the onset of suspected gram-negative sepsis, severity of underlying illness and in-hospital use of vasopressors are strong and consistent predictors of short- and long-term survival. Our data validate the McCabe and Jackson severity of illness scoring system for predicting long-term survival after sepsis. Physical dysfunction and more poorly perceived general health occur commonly after sepsis.