OBJECTIVES: Laparoscopic cholecystectomy (LC) has become a popular alternative to open cholecystectomy (OC). Previous studies comparing outcomes in LC and OC used small selected cohorts of patients and did not control for comorbid conditions that might affect outcome. The aims of,this study were to characterize the morbidity, mortality, and costs of LC and OC in a large unselected cohort of patients.METHODS: We used the population-based North Carolina Discharge Abstract Database (NCHDAD) for January 1, 1991, to September 30, 1994 (n = 850,000) to identify patients undergoing OC and LC. We identified the indications for surgery, complications, and type of perioperative biliary imaging used. We compared length of stay, hospital charges, complications, morbidity, and mortality between OC and LC patients. To account for variations in outcomes from differences in age and comorbidity between the OC and LC groups, we used the age-adjusted Charlson Comorbidity Index in regression analyses quantifying the association between type of surgery and outcome.RESULTS: Our cohort consisted of 43,433 patients (19,662 LC and 23,771 OC). The mean age-adjusted Charlson Comorbidity Index score was slightly higher for the OC compared to the LC group (4.3 vs 4.1, p < 0.05). The OC patients had longer hospitalizations, generated more charges ($12,125 vs $9,139, p < 0.05), and required home care more often. The crude risk ratio comparing risk of death in OC to LC was 5.0 (95% CI = 3.9-6.5). After controlling for age, comorbidity, and sex, the odds of dying in the OC group was still 3.3 times (95% CI = 1.4-7.3) greater than in the LC group. In the LC group, the number of patients with acute cholecystitis rose over the study period, whereas the number of patients with chronic cholecystitis declined. In the OC group, the number of patients with acute and chronic cholecystitis declined. The use of intraoperative cholangiography was greater in the OC group but declined in both groups over the study period. The use of ERCP was greater in the LC group and increased in both groups over time.CONCLUSIONS: The introduction of LC has resulted in a change in the management of cholecystitis. Despite a higher proportion of patients with acute cholecystitis, the risk of dying was significantly less in LC than in OC patients, even after controlling for age and comorbidity. Based on lower costs and better outcomes, LC seems to be the treatment of choice for acute and chronic cholecystitis. (Am J Gastroenterol 2002;97:334-340. (C) 2002 by Am. Coll. of Gastroenterology).

Aims of the study: Minimally invasive therapy for erectile dysfunction (ED) has changed the frequency of penile prosthesis surgery. The purpose of this study is to describe the changes in frequency, hospital stay, hospital charges and penile prosthesis type in North Carolina. Materials and Methods: The data source was a statewide hospital discharge database which includes data on hospitalized patients for all 151 hospitals in North Carolina. Results: From 1988–1993, 2354 patients underwent implantation of penile prostheses. The total number of penile prostheses implanted has declined over this six year period. Similarly, hospital stay has declined from an average of 4.03–2.96 d with a 46.6% decrease in total hospital days. Despite this change in hospital stay, hospital charges rose significantly from an average of $7252.48 to $12 842.18 driving total charges from $2 973 516.80 to $3 826 969.60 (1993) representing a 28.7% increase. Conclusions: Minimally invasive therapy and changes in reimbursement have had a major impact on the number of patients undergoing penile prosthesis implantation for ED. This downward trend may continue as more treatment options develop from the marked increase in research in this field. However, this may result in an increase of patients seeking treatment overall.

Background: Accurate data are needed to evaluate outcomes, therapeutics, and quality of care. This study assesses the accuracy of administrative databases in recording information about trauma patients. Methods: Patients with thoracic aorta injury were identified with a state trauma registry, and the medical records were reviewed. Data collected were compared to administrative data on patients with thoracic aorta injuries, at the same hospitals in the same time period. Results: Fifteen patients (16.3%) with thoracic aorta injury were not recorded in the administrative database, and 23 patients (18.7%) were misdiagnosed. Ninety-one patients were found in both data sources. The administrative database significantly (P < .05) underrecorded abdominal injuries (50 vs 35), orthopedic injuries (117 vs 75), and chest injuries (77 vs 48). The number of aortograms (78 vs 8), type of operative procedures (use of graft; 70 vs 30), use of bypass (35 vs 16), and complications (77 vs 33) were underreported (P < .05). The Injury Severity Score was underestimated by the administrative database (38.65 ± 12.41 vs 25.66 ± 9.53; P < .05). Conclusions: Administrative data lack accuracy in the recording of associated injury, injury severity, diagnostic, and procedural data. Whether these data should be used to evaluate treatment or quality of care in trauma is questionable. (Surgery 1999:126:191-7.)

BACKGROUND:The Glasgow Coma Scale (GCS), which is the foundation of the Trauma Score, Trauma and Injury Severity Score, and the Acute Physiology and Chronic Health Evaluation scoring systems, requires a verbal response. In some series, up to 50% of injured patients must be excluded from analysis because of lack of a verbal component for the GCS. The present study extends previous work evaluating derivation of the verbal score from the eye and motor components of the GCS.METHODS:Data were obtained from a state trauma registry for 24,565 unintubated patients. The eye and motor scores were used in a previously published regression model to predict the verbal score: Derived Verbal Score = -0.3756 + Motor Score * (0.5713) + Eye Score * (0.4233). The correlation of the actual and derived verbal and GCS scales were assessed. In addition the ability of the actual and derived GCS to predict patient survival in a logistic regression model were analyzed using the PC SAS system for statistical analysis. The predictive power of the actual and the predicted GCS were compared using the area under the receiver operator characteristic curve and Hosmer-Lemeshow goodness-of-fit testing.RESULTS:A total of 24,085 patients were available for analysis. The mean actual verbal score was 4.4 +/- 1.3 versus a predicted verbal score of 4.3 +/- 1.2 (r = 0.90, p = 0.0001). The actual GCS was 13.6 + 3.5 versus a predicted GCS of 13.7 +/- 3.4 (r = 0.97, p = 0.0001). The results of the comparison of the prediction of survival in patients based on the actual GCS and the derived GCS show that the mean actual GCS was 13.5 + 3.5 versus 13.7 + 3.4 in the regression predicted model. The area under the receiver operator characteristic curve for predicting survival of the two values was similar at 0.868 for the actual GCS compared with 0.850 for the predicted GCS.CONCLUSIONS:The previously derived method of calculating the verbal score from the eye and motor scores is an excellent predictor of the actual verbal score. Furthermore, the derived GCS performed better than the actual GCS by several measures. The present study confirms previous work that a very accurate GCS can be derived in the absence of the verbal component.

Objective: This project is designed to develop and validate a predictive model that is a useful benchmarking and quality of care assessment tool based on International Classification of Diseases, Ninth Revision (ICD-9), diagnoses and procedures. This model, the ICD-9-Based Illness Severity Score (ICISS), was developed from the Agency for Health Care Policy Research's Health Care Utilization Project database and is used to predict hospital survival, hospital length of stay, and hospital charges of injured patients admitted to University of North Carolina Hospitals. The study also compared the outcome predictions of ICISS with those of the long-established diagnosis-related groups (DRG) and the 3M product APR-DRG systems.Methods: We performed a retrospective study of 9,483 trauma patients at University of North Carolina Hospitals. A model was developed to predict survival, length of stay, and hospital charges. The accuracy of the model of survival was assessed using the area under the receiver-operating characteristics curve; the adjusted R-2 statistic was used to judge the proportion of variation described by the models of length of stay and hospital charges.Results: ICISS proved to be superior to both DRG and APR-DRG in predicting survival of trauma patients: the area under the receiver-operating characteristics curve for prediction of hospital survival was 0.957 for ICISS, 0.707 for DRG, and 0.808 for APR-DRG, ICISS also outperformed DRG and APR-DRG in predicting hospital length of stay and hospital charges: the adjusted R-2 for the ICISS length of stay model was 0.57, compared with the DRG length of stay model with adjusted R-2 Of 0.31 and the APR-DRG length of stay model with adjusted R-2 Of 0.35. The adjusted R-2 for the ICISS hospital charges model was 0.67, compared with the DRG and APR-DRG hospital charges model R-2 of 0.46 and 0.51, respectively (p < 0.001 in all cases).Conclusion: This study demonstrates that an ICD-9-based predictive model (ICISS) can markedly outperform both DRG and APR-DRG as a predictor of survival, hospital length of stay, and hospital charges.

s Of Papers To Be Presented At The Eleventh Annual Scientific Session; Eastern Association For The Surgery Of Trauma; January 11-16, 1999; Orlando, Florida

Since their inception, the Injury Severity Score (ISS) and the Trauma and Injury Severity Score (TRISS) have been suggested as measures of the quality of trauma care. In concept, they are designed to accurately assess injury severity and predict expected outcomes. ICISS, an injury severity methodology based on International Classification of Diseases, Ninth Revision, codes, has been demonstrated to be superior to ISS and TRISS. The purpose of the present study was to compare the ability of TRISS to ICISS as predictors of survival and other outcomes of injury (hospital length of stay and hospital charges). It was our hypothesis that ICISS would outperform ISS and TRISS in each of these outcome predictions."Training" data for creation of ICISS predictions were obtained from a state hospital discharge data base. "Test" data were obtained from a state trauma registry. ISS, TRISS, and ICISS were compared as predictors of patient survival. They were also compared as indicators of resource utilization by assessing their ability to predict patient hospital length of stay and hospital charges. Finally, a neural network was trained on the ICISS values and applied to the test data set in an effort to further improve predictive power. The techniques were compared by comparing each patient's outcome as predicted by the model to the actual outcome.Seven thousand seven hundred five patients had complete data available for analysis. The ICISS was far more likely than ISS or TRISS to accurately predict every measure of outcome of injured patients tested, and the neural network further improved predictive power.In addition to predicting mortality, quality tools that can accurately predict resource utilization are necessary for effective trauma center quality-improvement programs. ICISS-derived predictions of survival, hospital charges, and hospital length of stay consistently outperformed those of ISS and TRISS. The neural network-augmented ICISS was even better. This and previous studies demonstrate that TRISS is a limited technique in predicting survival resource utilization. Because of the limitations of TRISS, it should be superseded by ICISS.

OBJECTIVE:This project is designed to develop and validate a predictive model that is a useful benchmarking and quality of care assessment tool based on International Classification of Diseases, Ninth Revision (ICD-9), diagnoses and procedures. This model, the ICD-9-Based Illness Severity Score (ICISS), was developed from the Agency for Health Care Policy Research's Health Care Utilization Project database and is used to predict hospital survival, hospital length of stay, and hospital charges of injured patients admitted to University of North Carolina Hospitals. The study also compared the outcome predictions of ICISS with those of the long-established diagnosis-related groups (DRG) and the 3M product APR-DRG systems. METHODS:We performed a retrospective study of 9,483 trauma patients at University of North Carolina Hospitals. A model was developed to predict survival, length of stay, and hospital charges. The accuracy of the model of survival was assessed using the area under the receiver-operating characteristics curve; the adjusted R2 statistic was used to judge the proportion of variation described by the models of length of stay and hospital charges. RESULTS:ICISS proved to be superior to both DRG and APR-DRG in predicting survival of trauma patients: the area under the receiver-operating characteristics curve for prediction of hospital survival was 0.957 for ICISS, 0.707 for DRG, and 0.808 for APR-DRG. ICISS also outperformed DRG and APR-DRG in predicting hospital length of stay and hospital charges: the adjusted R2 for the ICISS length of stay model was 0.57, compared with the DRG length of stay model with adjusted R2 of 0.31 and the APR-DRG length of stay model with adjusted R2 of 0.35. The adjusted R2 for the ICISS hospital charges model was 0.67, compared with the DRG and APR-DRG hospital charges model R2 of 0.46 and 0.51, respectively (p < 0.001 in all cases). CONCLUSION:This study demonstrates that an ICD-9-based predictive model (ICISS) can markedly outperform both DRG and APR-DRG as a predictor of survival, hospital length of stay, and hospital charges.

s Of Papers To Be Presented At The Joint Meeting Of The American Association For The Surgery Of Trauma (Fifty-Eight Annual Meeting) And The Trauma Association Of Canada, September 24-26, 1998, The Renaissance Harborplace Hotel, Baltimore, Maryland

Introduction: Endoscopic techniques offer an alternative to surgical therapy in the management of common bile duct (CBD) stones.This study tested the hypothesis that the use of ERCP for CBD stones is associated with equivalent survival, shorter length of stay (LOS) and decreased rate of CBD exploration (CBDE).Methods: Data were obtained from a 157-hospital statewide discharge database.Patients with diagnoses of biliary tract disease (ICD-9 codes 574-576.9)were selected.Surgical and endoscopic procedures were identified.Results: From 1988 to 1992, 76,817 patients were admitted for biliary tract diseases (Table ).While admissions for CBD stones remained constant at 10%-11%, the number of CBDE declined from 1,541 in 1988 to 1,143 in 1991.Linear regression analysis predicted a continued decline in CBDE to only 12% of admissions for CBD stones in 1996.Survival remained unchanged.The average LOS for patients with CBD stones declined from 9.2 days in 1988 to 7.7 days in 1992 (p=0.01).The number of ERCP ~erformed increased throughout the study.

s Of Papers To Be Presented At The Joint Meeting Of The American Association For The Surgery Of Trauma (Fifty-Seventh Annual Meeting) And The Japanese Association For Acute Medicine; The Hilton Waikoloa Village, Waikoloa, Hawaii September 24-27, 1997

The low occurrence, nonspecific signs and symptoms, and high rate of associated morbidity and mortality of pulmonary embolus (PE) create major problems in the prevention, diagnosis, and treatment of PE. The purpose of this study was to analyze the frequency and outcome of PE in an entire state's trauma population using a large, population-based, hospital discharge data base. With the inclusion of an entire population, the reported incidence, high risk groups of patients, and specific risk factors regarding PE were assessed. A multivariate, logistic regression model was created from the data to determine predictive power of selected risk factors in patients at risk.Methods: The data source was a statewide, hospital discharge data base that includes data on all hospitalized patients for all of the hospitals in North Carolina. Data were available from 1988 to 1993. Using primary discharge diagnosis and nine additional ICD-9 coded diagnoses from the discharge abstract, patients were selected by presence of diagnostic codes for traumatic injury (800-959.9) and PE (415.1). Statistical analysis was performed using univariate and multivariate analysis to determine significant risk factors and to create a candidate model for the prediction of risk in the study population.Results: Of 318,554 patients, 952 (0.30%) had a recorded diagnosis of PE. The mortality rate for patients with PE (26%) was 10 times higher than the mortality rate in patients without PE (2.6%). In evaluating specific risk factors, age was a significant predictor of the risk of PE: 0.05% for patients under age 55 and 0.7% in those 55 years and over. The rate of PE, 0.3%, was low for the entire study population, but was highest in patients with injuries of the extremities, 0.53%. Increasing Injury Severity Score and Abbreviated Injury Scale score for determined body systems were also found to correlate with an increasing risk of PE.Over the course of the study, the incidence of PE among patients discharged from non-trauma centers showed a significant decrease. There was also a decrease in the mortality in non-trauma centers for PE. This finding cannot be due to coding changes coincident with the advent of diagnosis related groups because it would be associated with more vigorous combing of charts for diagnoses? It may well be that the use of prophylactic measures in injured patients initially used at trauma centers was adopted by the physicians at non-trauma centers over this time with the resultant decline in PE and associated mortality. From the univariate linear regression models, a logistic regression model was created that confirmed age as the most significant risk factor, followed by Injury Severity Score and Abbreviated Injury Scale score for soft tissue, extremity, and chest. The calculated area under the receiver operator characteristic curve was 0.72.Conclusion: Using a large, population-based data base, we were able to determine the reported incidence of PE among trauma patients and establish specific risk factors. The reported incidence of PE in this population is low, 0.30%. The mortality among those with PE, however, is significant at 26%. In this study, age, Injury Severity Score, and injury to specific body regions (soft tissue, extremity, chest) were associated with an increased risk of PE. The investigation of prophylaxis of PE and the general management of injured patients may be influenced by the overall low reported frequency of PE and the specific high risk populations described in this study. In light of the low incidence of PE in patients without specific risk factors, prophylactic interventions cannot be routinely recommended unless their benefits clearly outweigh their risks.

INTRODUCTION:Appropriate stratification of injury severity is a critical tool in the assessment of the treatment and the prevention of injury. Since its inception, the Injury Severity Score (ISS) has been the generally recognized "gold standard" for anatomic injury severity assessment. However, there is considerable time and expense involved in the collection of the information required to calculate an accurate ISS. In addition, the predictive power of the ISS has been shown to be limited. Previous work has demonstrated that the anatomic information about injury contained in the International Classification of Diseases Version 9 (ICD-9) can be a significant predictor of survival in trauma patients. The goal of this study was to utilize the San Diego County Trauma Registry (SDTR), one of the nation's leading trauma registries, to compare the predictive power of the ISS with the predictive power of the information contained in the injured patients' ICD-9 diagnoses codes. It was our primary hypothesis that survival risk ratios derived from patients' ICD-9 diagnoses codes would be equal or better predictors of survival than the Injury Severity Score. The implications of such a finding would have the potential for significant cost savings in the care of injured patients.METHODS:Data for the test population were obtained from the SDTR, which contains data from 1985 through 1993 from five participating hospitals. Four data sources were utilized to estimate the expected survival rate/mortality rate for each ICD-9 code in the SDTR. These were (1) the SDTR patients themselves, (2) the North Carolina State Hospital Discharge Database, (3) the North Carolina Trauma Registry Database, and (4) the Agency for Health Care Policy Research's Health Care Utilization Project Database. Each of these data sources was separately utilized to develop a survival risk ratio (SRR) for each ICD-9 diagnoses code. The SRR was calculated by dividing the number of survivors for patients with each ICD-9 code by the total number of all patients with the particular ICD-9 diagnoses code. The four groups of SRRs derived from our four data sources were used as predictors of survival and the ability of the SRRs to predict survival was compared with the predictive power of the ISS using measures of accuracy, sensitivity, specificity, and receiver operator characteristic curves.RESULTS:During the years 1985 through 1993, complete data were available for analysis on 44,032 patients. Of these, 2,848 patients died during their hospitalization (6%). Survival risk ratios were calculated for each of the diagnoses in the data base. Logistic regression, using the SAS System for statistical analysis, was used to assess the relative predictive power of the ISS and the survival risk ratios derived from the ICD-9 diagnoses codes from each of the four data bases. The analyses demonstrated that the regression models using the SRRs were generally as good or better than ISS as predictors of survival. The predictive power of the SRRs derived from the SDTR data, the North Carolina Trauma Registry data and the Health Care Utilization Report data were the best. In a subsequent analysis, the SRR values and the ISS were added to the patient's age and the revised Trauma Scores to create new predictive models in the mode of TRISS methodology. The analyses again indicated that the models using SRRs had as good or better predictive power than the model using the ISS.CONCLUSIONS:The present study confirms previous work showing that survival risk ratios derived from injured patients' ICD-9 diagnoses codes are as good as or better than ISS as predictors of survival.

BACKGROUND:Trauma registries are an essential but expensive tool for monitoring trauma system performance. The time required to catalog patients' injuries is the source of much of this expense. Typically, 15 minutes of chart review per patient are required, which in a busy trauma center may represent 25% of a full-time employee. We hypothesized that International Classification of Disease-Ninth Revision (ICD-9) codes generated by the hospital information system (HI) would be similar to those coded by a dedicated trauma registrar (TR) and would be as accurate as TR ICD-9 codes in predicting outcome.METHODS:One thousand eight hundred twelve patients admitted to a Level I trauma center during 2 years had International Classification of Disease Injury Severity Scores (ICISS) calculated based on HI and TR ICD-9 codes. The relative predictive powers of these two ICISSs were then compared for every patient using Receiver Operator Characteristic Curve Area (ROC) and Hosmer Lemeshow Statistics.RESULTS:Eighty-nine percent of patients (1,608 of 1,812) had identical HI and TR ICISSs. Eleven patients' ICISSs differed by >0.1, and only two patients' scores differed by >0.2. ICISS proved to be a powerful predictor of outcome whether derived from HI (ROC = 0.884; 95% confidence interval (CI) = 0.850-0.917) or TR (ROC = 0.872; 95% CI = 0.837-0.908). Although these predictive powers were not significantly different (p = 0.076), the trend was for HI to perform better than TR. ISS calculated for the same data set using the MacKenzie dictionary proved significantly less predictive of outcome than either ICISS (ROC(MacKenzie) = 0.843; 95% CI = 0.792-0.884; p = 0.034).CONCLUSION:We conclude that in our hospital TR data on individual injuries can be replaced by HI data without loss of predictive power. ISS based on the MacKenzie dictionary should be abandoned because it is much less predictive of outcome than ICISS.

s Of Papers To Be Presented At The Joint Meeting Of The American Association For The Surgery Of Trauma (Fifty-Seventh Annual Meeting) And The Japanese Association For Acute Medicine; The Hilton Waikoloa Village, Waikoloa, Hawaii September 24-27, 1997

Background Trauma registries are an essential but expensive tool for monitoring trauma system performance. The time required to catalog patients' injuries is the source of much of this expense. Typically, 15 minutes of chart review per patient are required, which in a busy trauma center may represent 25% of a full-time employee. We hypothesized that International Classification of Disease-Ninth Revision (ICD-9) codes generated by the hospital information system (HI) would be similar to those coded by a dedicated trauma registrar (TR) and would be as accurate as TR ICD-9 codes in predicting outcome. Methods One thousand eight hundred twelve patients admitted to a Level I trauma center during 2 years had International Classification of Disease Injury Severity Scores (ICISS) calculated based on HI and TR ICD-9 codes. The relative predictive powers of these two ICISSs were then compared for every patient using Receiver Operator Characteristic Curve Area (ROC) and Hosmer Lemeshow Statistics. Results Eighty-nine percent of patients (1,608 of 1,812) had identical HI and TR ICISSs. Eleven patients' ICISSs differed by >0.1, and only two patients' scores differed by >0.2. ICISS proved to be a powerful predictor of outcome whether derived from HI (ROC = 0.884; 95% confidence interval (CI) = 0.850-0.917) or TR (ROC = 0.872; 95% CI = 0.837-0.908). Although these predictive powers were not significantly different (p = 0.076), the trend was for HI to perform better than TR. ISS calculated for the same data set using the MacKenzie dictionary proved significantly less predictive of outcome than either ICISS (ROCMacKenzie = 0.843; 95% CI = 0.792-0.884; p = 0.034). Conclusion We conclude that in our hospital TR data on individual injuries can be replaced by HI data without loss of predictive power. ISS based on the MacKenzie dictionary should be abandoned because it is much less predictive of outcome than ICISS.

BACKGROUND: For more than 40 years carotid endarterectomy (CE) has been used in the treatment of extracranial carotid disease for the prevention of stroke, Recent prospective clinical trials have confirmed the benefit of CE for both symptomatic and asymptomatic patients, Our purpose was to examine statewide trends in the numbers of CE over a 6-year time period and to evaluate outcomes,METHODS: Using data from the North Carolina Medical Database Commission (NCMDC) all CE procedures from 1988 to 1993 were identified, Numbers of CE were compared with the population and hospital admissions, Variables of length of stay, hospital charges, discharge disposition, and occurrence of stroke and death were analyzed.RESULTS: A total of 11,973 CE were performed in 6 years, Compared by admissions, population, and the proportion of elderly, the number of CE increased yearly, The stroke rate was 1.7% and the death rate 1.2% for an overall in-hospital stroke plus mortality rate of only 2.7%.CONCLUSIONS: From a diverse group of hospitals and a large number of surgeons and patients, this hospital-based study documents the acceptance and safety of CE in the treatment of extracranial carotid disease. (C) 1997 by Excerpta Medica, Inc.

While the number of patients listed for liver transplant has increased, the pool of donor organs has remained constant. Questions have arisen regarding equitable access to organs. The purpose of this study was to analyze factors associated with access to liver transplantation (LT) using a large, population-based, hospital discharge database. The primary hypothesis was that a variety of factors other than medical need could be associated with access to LT. The rate of LT was defined as the number of liver transplants per admission for liver disease. The data sources were selected to allow a population-based, time-series analysis of all patients admitted with liver disease and those receiving liver transplants in all 157 nonfederal hospitals in North Carolina from 1988 to 1993. The hypotheses of this study were that age, gender, payment source, type of liver disease, distance from the transplant center, and rural county of residence were associated with patients' likelihood of access to LT. During the six years studied, 56,803 patients were admitted with liver disease and 126 underwent liver transplantation (LT). The rate of LT increased from 0.07% to 0.27%. Age, gender, source of payment, type of liver disease, rural county of residence, and distance of residence from the transplant center were associated with rates of transplantation. In the multivariate model, source of payment appeared to have the strongest association with the likelihood of LT. These findings raise important questions associated with equitable access to health care, need for physician education, and transplant center regionalization.

More reliable prediction of outcome would be helpful for clinicians who treat severely head-injured patients. To determine if neural network modeling would improve outcome prediction compared with standard logistic regression analysis and to determine if data available 24 h after severe head injury allows better prediction than data obtained within 6 h, we tested the ability of both techniques at these two times to predict outcome (dead versus alive) at 6 months. One thousand sixty-six consecutive patients with Glasgow Coma Scale scores of 8 or less during the first 24 h after injury were randomly divided into two groups. Data from the first group (n = 799) were used to develop the models; data from the second group (n = 267) were used to test the accuracy, sensitivity, and specificity of the models by comparing predicted and actual outcomes. The 6-month mortality rate was 63.5%. Our findings confirm the importance of age, Glasgow Coma Scale scores, and hypotension in predicting outcome. Using data available at 24 h improved the predictive power of both models compared with admission data; at both time points, however, the differences in the results obtained with the two models were negligible. We conclude that outcome (dead versus alive) at 6 months after severe head injury can be predicted with logistic regression or neural network models based on data available at 24 h. Critical therapeutic decisions, such as cessation of therapy, should be based on the patient's status 1 day after injury and only rarely on admission status alone.