ABSTRACT
International quality improvement initiatives such as Fast-Hug bring a focus on improving the delivery of early enteral nutrition to critically ill patients, however surveys demonstrate current practice remains variable. One way to reduce variability in practice is to provide strong evidence to convince clinicians to change. The purpose of this overview was to identify current best evidence supporting the delivery of early enteral nutrition in critical illness. We sought high-quality evidence in the form of systematic reviews containing meta-analyses of randomised controlled trials. Two authors independently identified studies and assessed methodological quality. Data sources included Medline, EMBASE and hand-searching of guideline reference lists. The literature search identified five systematic reviews that summarised 30 clinical trials. These systematic reviews focused on acutely hospitalised patients, critical illness, burns, elective intestinal surgery and pancreatitis. Early enteral nutrition significantly reduced mortality in elective intestinal surgery patients (relative risk 0.41, 95% confidence interval 0.18 to 0.93, P = 0.03, I2 = 0.0%) and significantly reduced infectious complications in acutely ill hospitalised patients (relative risk 0.45, 95% confidence interval 0.3 to 0.66, P = 0.00006, heterogeneity P = 0.049). Four of five identified systematic reviews had key methodological quality deficiencies. The results of this overview highlight the variability in the evidence regarding the benefits of early enteral nutrition in critically ill patient populations. The inconsistent delivery to critically ill patients may be explained by the lack of convincing evidence. Better evidence may be needed to reduce the irregularity in the provision of early enteral nutrition to critically ill patients.
Subject(s)
Critical Care/statistics & numerical data , Critical Illness , Enteral Nutrition/statistics & numerical data , Evidence-Based Medicine , Critical Illness/mortality , Humans , Infections/epidemiology , Length of Stay , Meta-Analysis as Topic , Postoperative Complications/epidemiology , Postoperative Complications/mortality , Randomized Controlled Trials as Topic , Treatment OutcomeABSTRACT
EBM represents a more formal process for considering the merits of published research in the context of clinical decision making. By combining the application of formal rules of evidence in evaluating the clinical literature (critical appraisal) with the ability to conduct efficient and effective literature searches, any clinician is able to keep current on topics of interest. Learning the core skills that enable the practice of EBM is easier than expected. The EBM Working Group published a series of Users' Guides to enable clinicians to learn critical appraisal of the primary literature in a systematic fashion. The complete series of Users' Guides is available free of charge over the web (www. CCHE.net). Similarly, PubMed, which contains a number of easy to use search filters (Clinical Queries) that can identify high-quality papers published in key veterinary journals, is also available free of charge over the web (www.PubMed.org). How ever one obtains the core skills, an efficient, structured approach to the practice of EBM can actually reduce the amount of time one requires to keep current by helping the user locate and identify methodologically rigorous research. Although the entire process can be self-taught, as with any new technique, other effective ways to bring these new skills into your practice setting include attending a conference workshop or even hiring a new graduate who has already been trained in the process.
Subject(s)
Evidence-Based Medicine , Veterinary Medicine/trends , Animals , HumansABSTRACT
The aim of this study was to examine the cellular distribution of osteopontin (OPN) protein [by immunohistochemical (IHC) analysis] and mRNA [by in situ hybridization (ISH)] in the primary tumors of lymph node negative (LNN) breast cancer patients and to determine whether the level of immunodetectable OPN may be associated with tumor aggressiveness. We examined OPN levels in tumors from 154 patients with LNN breast cancer who were followed for a median of 7 years (range 1.7-16.3 years). IHC staining for OPN was seen in tumor infiltrating macrophages and lymphocytes in 70% of these tumors, and in the carcinoma cells themselves in 26%. ISH was performed to determine cellular distribution of OPN mRNA expression in sections from selected tumors. OPN mRNA was detected in groups of tumor cells, individual tumor cells and tumor infiltrating macrophages and lymphocytes. Matched sections showed that some tumor cells with IHC staining for OPN protein were also positive for OPN mRNA by ISH, in contrast with previous studies which have shown OPN mRNA expression only in tumor infiltrating inflammatory cells. Our results thus indicate that OPN protein can be produced by breast cancer cells in vivo and suggest that it may also be taken up from the environment (i.e., secreted by inflammatory cells or other tumor cells). Tumor cell IHC staining intensity was then assessed using a semiquantitative scoring system. Univariate analysis showed tumor cell OPN positivity above an optimized cutpoint to be significantly associated with decreased disease-free survival (DFS) and overall survival (OS). The results of this pilot study thus suggest that the ability of breast cancer cells to either synthesize OPN or to bind and sequester OPN from the microenvironment may be associated with tumor aggressiveness and poor prognosis.
Subject(s)
Breast Neoplasms/chemistry , Gene Expression , Lymph Nodes/pathology , Sialoglycoproteins/analysis , Adult , Aged , Aged, 80 and over , Breast Neoplasms/pathology , Disease-Free Survival , Female , Humans , Immunohistochemistry , In Situ Hybridization , Menopause , Middle Aged , Osteopontin , Prognosis , RNA, Messenger/analysis , Sialoglycoproteins/geneticsABSTRACT
In 1754, aboard HMS Salisbury, James Lind conducted a simple, controlled clinical trial. He took 12 patients with "pale and bloated skin, listlessness, an aversion to exercise, swollen gums, halitosis, ecchymotic mucous membranes, and limb edema" and allocated them to receive treatment with one of six different therapies. Since the patients receiving two of his six chosen interventions had such a dramatic recovery, he felt ethically obligated to end his trial and administer these treatments to all the remaining sailors. Today we fully recognize the impact that the controlled clinical trial can have on the development of new interventions. Unfortunately, very few of these interventions are likely to have as dramatic an impact on outcomes as lemons and oranges did on scurvy. Because the interventions we study tend to have relatively small treatment effects, and because the design and reporting of published RCTs has consistently been documented to be less than perfect, there is a real need for us to develop critical appraisal skills. This article is by no means the only approach to critical appraisal, but hopefully it serves as an adequate starting point for the journey.
Subject(s)
Randomized Controlled Trials as Topic/standards , Algorithms , Data Interpretation, Statistical , Guidelines as Topic , Humans , Randomized Controlled Trials as Topic/methodsABSTRACT
OBJECTIVE: To test the hypothesis that adult sheep pretreated with polymyxin-dextran and then made septic by cecal ligation and perforation would have fewer changes in microvascular integrity and cellular architecture in extrapulmonary organs. DESIGN: Prospective, randomized, double-blind, placebo-controlled animal study. SETTING: An animal research facility in a university-affiliated hospital. SUBJECTS: Mature, male Suffolk sheep (32 to 67 kg). INTERVENTIONS: Animals with chronic indwelling catheters were pretreated with polymyxin B-dextran (6 mg/kg) or placebo (dextran) and an intra-abdominal focus of infection was then produced by cecal ligation and perforation. Treatment (polymyxin B or placebo) was continued every 8 hrs for 48 hrs. MEASUREMENTS AND MAIN RESULTS: Forty-eight hours after randomization, the polymyxin B-dextran group manifested significantly less pyrexia (p = .04), higher mean arterial pressures (p = .02), less variable serum albumin concentrations (p = .05), and a trend toward decreased lactate concentrations (p = .10). Qualitative morphometry and semiquantitative scoring of tissue from gastrocnemius muscle demonstrated that polymyxin B-dextran-treated sheep had significantly increased total capillary (p = .04) and capillary luminal areas (p = .038) and less mitochondrial swelling and damage (p = .03) compared with the placebo sheep. CONCLUSIONS: Pretreatment of sheep in a polymicrobial, peritonitis model of sepsis with polymyxin B-dextran resulted in a significant amelioration of sepsis-induced ultrastructural damage. In placebo-treated control animals, these ultrastructural lesions were associated with a greater severity of sepsis, as measured by the presence of pyrexia, increased lactate concentrations, and less stable blood pressures. These findings justify the investigation of the effects of polymyxin B-dextran in a post onset model of sepsis.
Subject(s)
Anti-Bacterial Agents/therapeutic use , Endotoxins/antagonists & inhibitors , Polymyxin B/therapeutic use , Sepsis/prevention & control , Animals , Dextrans/administration & dosage , Hemodynamics/drug effects , Male , Muscles/drug effects , Muscles/pathology , Plasma Substitutes/administration & dosage , Premedication , SheepABSTRACT
OBJECTIVE: To compare the histopathologic features and expression of p53 and c-erb B2 in the tumours detected by mammography only (clinically occult tumours) and the tumours detected by a nurse examiner (clinically palpable tumours). SETTING: London branch of the Ontario Breast Screening Program, which uses both clinical breast examination and mammography as screening methods. INTERVENTIONS: Pathologic review and immunohistochemical staining of all tumours detected between 1990 and 1993. OUTCOME MEASURES: Categorization of tumours by detection method and analysis of tumour size, grade, type, lymph node status and c-erb B2 and p53 expression in each group. RESULTS: From 1990 to 1993, 131 tumours were detected in patients ranging in age from 50 to 85 years (median 63 years). Sixty-seven occult tumours and 64 palpable lesions were detected. The occult tumours were significantly smaller (1.34 cm v. 2.29 cm, p < 0.0001) than the palpable ones and included a higher proportion of special-type lesions and ductal carcinoma in situ (43.3% v. 10.9%, p < 0.0001). Occult invasive carcinomas were of lower grade than palpable carcinomas (68.4% grade 1, 21.1% grade 2, 10.5% grade 3 v. 32.8% grade 1, 36.1% grade 2, 31.1% grade 3, p < 0.0001). Fewer occult lesions showed axillary nodal metastases (19.6% v. 40.6%, p = 0.02). No statistically significant differences were found for p53 or c-erb B2 positivity between the 2 groups. CONCLUSION: Tumours detected by different screening methods in a screening program have different pathologic characteristics.
Subject(s)
Breast Neoplasms/diagnosis , Mammography , Palpation , Aged , Aged, 80 and over , Breast Neoplasms/chemistry , Breast Neoplasms/pathology , Carcinoma/chemistry , Carcinoma/diagnosis , Carcinoma/pathology , Carcinoma in Situ/chemistry , Carcinoma in Situ/diagnosis , Carcinoma in Situ/pathology , Carcinoma, Ductal, Breast/chemistry , Carcinoma, Ductal, Breast/diagnosis , Carcinoma, Ductal, Breast/pathology , Female , Humans , Immunohistochemistry , Lymphatic Metastasis , Middle Aged , Pilot Projects , Receptor, ErbB-2/analysis , Tumor Suppressor Protein p53/analysisABSTRACT
OBJECTIVES: To determine the ability of the current literature to supply appropriate data for benchmarking admission practice to a multidisciplinary critical care unit. DESIGN: Retrospective review of data collected prospectively on a cohort of 614 patients and a systematic review of the literature. SETTING: A 30-bed multidisciplinary critical care unit at a university teaching hospital. PATIENTS: Consecutive admissions to the critical care unit over a 6-month period. INTERVENTIONS: None. MEASUREMENTS AND RESULTS: For each patient, demographic data and admitting diagnosis were recorded on admission. Information necessary to calculate the Acute Physiology and Chronic Health Evaluation II and Therapeutic Intervention Scoring System (TISS) scores were collected daily. TISS variables were categorized as "active" or "non-active" treatment variables. Patients were then identified on a daily basis as receiving or not receiving active treatment. A review of the literature, using MEDLINE and the search term "Therapeutic Intervention Scoring Index" (as a textword), was conducted to identify studies that had similarly divided their patients. Using the method of benchmarking, the proportion of patients admitted who received active treatment during their stay in the critical care units was compared between the index critical care unit and those in the literature. A greater proportion of the patients admitted to our unit received active treatment (97.7%) when compared to other studies in the literature (20-66%). However, a number of potential confounding factors were present, such as the availability of intermediate care units, overnight recovery room ventilation, and critical care bed availability between the index critical care unit and those described in the literature. CONCLUSIONS: The current literature does not provide adequate data on critical care unit admission practices to allow useful application of the method of benchmarking. There is a need for publicly accessible large databases to allow individual critical care units to determine their level of efficiency when compared to similar institutions.
Subject(s)
Efficiency, Organizational/statistics & numerical data , Patient Admission/standards , Patient Selection , Trauma Centers/statistics & numerical data , APACHE , Adolescent , Adult , Aged , Aged, 80 and over , Chi-Square Distribution , Cohort Studies , Critical Care/statistics & numerical data , Critical Illness/epidemiology , Databases, Factual , Diagnosis-Related Groups , Efficiency, Organizational/standards , Female , Guidelines as Topic , Humans , Male , Middle Aged , Retrospective Studies , Review Literature as Topic , TriageABSTRACT
OBJECTIVE: To determine whether the time that red blood cells are stored in citrate phosphate dextrose adenine-1 solution before transfusion alters the ability to improve tissue oxygenation. DESIGN: Prospective, randomized, controlled study. SETTING: University research institute laboratory. SUBJECTS: Male Sprague-Dawley rats (350 to 450 g). INTERVENTIONS: Twenty-four hours after randomization to sham laparotomy (n = 21) or cecal ligation and perforation (n = 16)1 supply-dependency of systemic oxygen uptake (VO2) was induced in rats by isovolemic hemodilution. Rats were then re-randomized to receive either rat red blood cells stored in citrate phosphate dextrose adenine-1 for 3 days ("fresh" n = 17) or rat red blood cells stored in citrate phosphate dextrose adenine-1 for 28 days ("old" n = 20). MEASUREMENTS AND MAIN RESULTS: Changes in systemic VO2 were measured for 90 mins to determine the efficiacy of the treatment. Statistical analysis included a fully factorial repeated-measures, generalized linear model. No significant interaction was found between cecal ligation and perforation or sham animals and transfusion with fresh or old red blood cells. However, comparing the combined groups of animals receiving either fresh or old red blood cells, we found that after the transfusion of old red blood cells, systemic VO2 was not significantly improved (after hemodilution 1.68 +/- 0.27 mL/100 g/min, after transfusion 1.86 +/- 0.17 mL/100 g/min; p > .05). In contrast, transfusion with fresh red blood cells acutely increased systemic VO2 (after hemodilution 1.62 +/- 0.06 mL/100 g/min, after transfusion 2.10 +/- 0.09 mL/100 g/min; p = .049). CONCLUSION: Storage of rat red blood cells for 28 days in citrate phosphate dextrose adenine-1 impaired their ability to improve tissue oxygenation when transfused into either control or septic rats placed into supply dependency of systemic VO2.
Subject(s)
Blood Preservation/adverse effects , Erythrocyte Transfusion , Oxygen Consumption , Adenine , Animals , Anticoagulants , Blood Preservation/methods , Citrates , Glucose , Hemodilution , Hemodynamics , Ligation , Male , Random Allocation , Rats , Rats, Sprague-Dawley , Sepsis/etiology , Time FactorsABSTRACT
Health technology refers to the instruments, equipment, drugs and procedures used in health care delivery, as well as the organizations supporting it. Health technology assessment, which is the process of conducting investigations to establish the criteria for efficacious, effective and efficient patient care, is becoming increasingly important in an era of diminishing resources. This survey of 39 community hospitals in southwestern Ontario found that improved purchasing strategies can result in substantial cost savings which can in turn be used to improve patient care. The study shows that optimizing the price of basic hospital commodities could save an average community hospital as much as $625,000 per year.
Subject(s)
Hospitals, Community/statistics & numerical data , Purchasing, Hospital/statistics & numerical data , Technology Assessment, Biomedical/statistics & numerical data , Cost Control/methods , Costs and Cost Analysis , Data Collection , Equipment and Supplies, Hospital/economics , Equipment and Supplies, Hospital/statistics & numerical data , Financial Management, Hospital/methods , Health Care Surveys , Hospitals, Community/economics , OntarioABSTRACT
Tumor angiogenesis is associated with metastasis in several types of solid tumors, including melanoma, breast, prostate, lung, bladder, and oral-cavity tumors. The purpose of this study was to determine whether tumor angiogenesis could predict recurrence following curative surgery for colorectal cancer. Thirty-five patients were studied, including 13 patients with recurrent tumor and 22 without. Representative formalin-fixed, paraffin-embedded sections of invasive colorectal cancers from these patients were sectioned. The endothelial cells of microvessels within the tumors were highlighted by immunohistochemical staining for CD31. The most active areas were identified and the microvessels counted in a x 400 field (0.152 mm2) by two observers in a blinded fashion. Tumor microvessel count (p = 0.0062). Dukes' staging (p = 0.0004), vascular invasion (p = 0.0280), and tumor grade (p = 0.0559) were all significantly associated with tumor recurrence. Tumor microvessel counts > or = 65 per x 400 field were associated with tumor recurrence (p = 0.0035, relative risk [RR] = 11.3). Controlling for Dukes' stage, a multivariate logistic regression model revealed that a tumor microvessel count > or = 65 is an important predictor of tumor recurrence (p = 0.0783, RR = 6.0). A backwards elimination proportional hazards model revealed that a microvessel count > or = 65 shows a trend toward independent prediction of time to tumor recurrence (p = 0.1203, RR = 2.967) when controlled for Dukes' staging (p = 0.0029, RR = 9.089). Despite the small number of patients studied, these results suggest that the number of microvessels in sections of invasive colorectal adenocarcinoma immunohistochemically stained with CD31 may be an important independent predictor of tumor recurrence and time to recurrence.
Subject(s)
Adenocarcinoma/blood supply , Adenocarcinoma/pathology , Colorectal Neoplasms/blood supply , Colorectal Neoplasms/pathology , Neoplasm Recurrence, Local/pathology , Neovascularization, Pathologic/pathology , Adenocarcinoma/surgery , Aged , Aged, 80 and over , Chi-Square Distribution , Colorectal Neoplasms/surgery , Female , Humans , Incidence , Male , Microcirculation/pathology , Middle Aged , Multivariate Analysis , Neoplasm Invasiveness , Neoplasm Recurrence, Local/epidemiology , Neoplasm Staging , Proportional Hazards Models , Retrospective Studies , Survival AnalysisABSTRACT
Evidence-based approaches to assessing the clinical literature are used increasingly in issues relating to critical care medicine. As we discussed previously, this approach attempts to provide a logical and convenient framework from which the quality and relevance of clinical studies may be assessed in an unbiased manner. An evidence-based approach also allows the reader to differentiate between solid evidence and evidence that is based on a presumed mechanism, standard practice, or conventional wisdom. Evidence-based medicine that deemphasizes intuition, unsystemic clinical experience, and pathophysiologic rationale is sufficient grounds for clinical decision making and stresses the examination of evidence from clinical research. Importantly, it is the objective nature by which the evidence-based medicine paradigm approaches the questions of "What are we doing" and "How can we do better," that causes health care providers and funding agencies to increasingly adopt this paradigm as a primary principle. The role of evidence-based medicine, therefore, is not to discount expert opinion but, wherever possible, to require that recommendations be based on the results of rigorous and controlled scientific study. We introduced this article by highlighting the growing imbalance between resources and patient needs in the critical care environment. At the level of diagnostic technology and therapeutic care plans, critical care professionals increasingly are asked to identify strategies to improve efficiencies-approaches with rigorous costs that at the same time promote better patient care. Formalized technology assessment is one of the mechanisms to accomplish this. Using critical appraisal within the context of evidence-based medicine is one of the mechanisms by which data can be synthesized to describe technology assessment. In this example, we used controversies surrounding use of pulmonary artery catheterization in critically ill patients to highlight some of the principles of evidence-based medicine at the bedside. We introduced how a particular case example can be used to ask the question, "What is the evidence that a particular diagnostic technology or therapeutic modality improves outcome?" In the case of PACs, no one would argue that a diagnostic technology a priori leads to improved outcome; rather, we have to presume that given good data, the care provider will apply the most appropriate management strategy. In this case example, however, we demonstrate how the critical appraisal process should identify a search strategy to find the most appropriate evidence to support the questioning process. From this, the evidence can be critically appraised and tabulated. In the case of the PAC, Table 13 demonstrates that there are few data to identify a grade A recommendation that the PAC should be used as part of the care process in critically ill patients. Finding that there is little evidence to support the use of pulmonary artery catheterization in the clinical literature does not mean that this diagnostic technology is neither efficacious nor effective. It might well be that information provided by PACs is important in the care process. However, what this exercise has taught us is that there is little objective evidence to support this conclusion. The challenge to critical care practitioners is not only to apply the evidence-based processes more frequently to our environment but also to use the information to separate out clearly what is fact versus opinion. Where there is little evidence to support a particular clinical practice, as we have demonstrated with the PAC review earlier, the challenge to the clinician should be the design and conduct of clinical trials clarifying debate between opinion and evidence.
Subject(s)
Catheterization, Swan-Ganz , Critical Care , Evidence-Based Medicine , Aged , Catheterization, Swan-Ganz/methods , Female , Hemodynamics , Humans , Oximetry , Treatment OutcomeABSTRACT
Of 49 patients with nonconvulsive seizures studied with continuous EEG monitoring, the overall mortality was 33% (16/49). Of the 23 patients with nonconvulsive status epilepticus (NCSE), 13 died (mortality 57%). Individual variables significantly associated with mortality were age, presence of NCSE, seizure duration, hospital and NICU length of stay, and delay to diagnosis and etiology (acute illness versus remote symptomatic). With multivariate logistic regression, only seizure duration (p = 0.0057, OR = 1.131/hour) and delay to diagnosis (p = 0.0351, OR = 1.039/hour) were associated with increased mortality. Acute symptomatic cases could not be adequately classified as either absence, simple, or complex partial status epilepticus when the impairment of consciousness arose form the initial illness. Current classifications of status epilepticus are inadequate for such cases.
Subject(s)
Electroencephalography , Seizures/physiopathology , Adult , Aged , Aged, 80 and over , Female , Humans , Intensive Care Units , Male , Middle Aged , Monitoring, Physiologic , Seizures/mortalityABSTRACT
The objectives of this study were to determine if a standardized gentamicin dosing protocol would improve clinical effectiveness, yield higher peak serum concentrations, and improve the success rate of attaining peaks in the desired range when compared with empiric dosing practices used by prescribers. The study was conducted as a before-after program effectiveness evaluation in non-critically ill patients, aged 16-65 y with stable renal function, who were prescribed gentamicin. A standardized dose of 2 mg/kg (ideal or adjusted weight) was administered intravenously every 12 h in the intervention phase. Response to therapy (time to defervescence, white cell count, reinstitution of antibiotic therapy), serum concentrations (peaks > 10 mumol/L (5.6 mg/L) and troughs < 4 mumol/L (2.2 mg/L)), and toxicity were monitored in both groups. Thirty-four consecutive patients were enrolled into the control phase and an equal number into the intervention phase. Surgical patients comprised the majority of the study population. Desired peak concentrations were attained in 97% of intervention vs. 59% of control patients (p < 0.001). Mean peak serum concentrations were higher in the intervention phase than in the control phase, 16.1 mumol/L vs. 11.2 mumol/L (p < 0.001), respectively. Median time to become afebrile trended toward a statistical decrease in the intervention as compared to the control group, 3 vs. 5 d (p = 0.076), respectively. There was no significant difference in clinical effectiveness nor in the occurrence of nephro- or ototoxicity. Continued evaluation of this dosing protocol is warranted.
Subject(s)
Drug Administration Schedule , Gentamicins/blood , Adolescent , Adult , Aged , Body Weight , Creatinine/blood , Female , Fever/metabolism , Gentamicins/administration & dosage , Gentamicins/toxicity , Humans , Leukocytes/metabolism , Male , Middle AgedABSTRACT
BACKGROUND AND PURPOSE: The role of genetics in cerebrovascular disease remains controversial. The purpose of this study was to assess the influence of family history on atherothrombotic infarction or transient ischemic attack. METHODS: Ninety patients with stroke or transient ischemic attack and 90 age- and sex-matched community control subjects were studied prospectively. Medical and family histories were obtained from all subjects, and a complete physical examination was performed. RESULTS: Eighty-five patients and 86 control subjects knew their family history for ischemic heart disease and stroke. A positive history for ischemic heart disease was present in 62 (73%) of the patients and 46 (53%) of the control subjects (P = .019), and a positive family history for stroke was present in 38 (47%) of the patients and 21 (24%) of the control subjects (P = .014). CONCLUSIONS: Although a positive vascular family history was not an independent risk factor in a multivariate analysis, it was an excellent marker of the presence of other established vascular risk factors. Personal histories of ischemic heart disease, hypertension, and hyperlipidemia were found to be significant independent risk factors for stroke.
Subject(s)
Cerebrovascular Disorders/genetics , Adult , Aged , Aged, 80 and over , Cerebral Infarction/epidemiology , Cerebral Infarction/genetics , Cerebrovascular Disorders/epidemiology , Coronary Disease/epidemiology , Coronary Disease/genetics , Cross-Sectional Studies , Female , Humans , Hyperlipidemias/epidemiology , Hyperlipidemias/genetics , Ischemic Attack, Transient/epidemiology , Ischemic Attack, Transient/genetics , Logistic Models , Male , Middle Aged , Ontario/epidemiology , Prospective Studies , Risk FactorsABSTRACT
The objective of this study was to compare and contrast two techniques of modeling mortality in a 30 bed multi-disciplinary ICU; neural networks and logistic regression. Fifteen physiological variables were recorded on day 3 for 422 consecutive patients whose duration of stay was over 72 hours. Two separate models were built using each technique. First, logistic and neural network models were constructed on the complete 422 patient dataset and discrimination was compared. Second, the database was randomly divided into a 284 patient developmental dataset and a 138 patient validation dataset. The developmental dataset was used to construct logistic and neural net models and the predictive power of these models was verified on the validation dataset. On the complete dataset, the neural network clearly outperformed the logistic model (sensitivity and specificity of 1 and .997 vs. .525 and .966, area under ROC curve .9993 vs. .9259), while both performed equally well on the validation dataset (area under ROC of .82). The excellent performance of the neural net on the complete dataset reveals that the problem is classifiable. Since our dataset only contained 40 mortality events, it is highly likely that the validation dataset was not representative of the developmental dataset, which led to a decreased predictive performance by both the neural net and the logistic regression models. Theoretically, given an extensive dataset, the neural network should be able to perform mortality prediction with a sensitivity and a specificity approaching 95%. Clinically, this would be an extremely important achievement.(ABSTRACT TRUNCATED AT 250 WORDS)
Subject(s)
Hospital Mortality , Intensive Care Units/statistics & numerical data , Logistic Models , Neural Networks, Computer , Humans , Middle Aged , Multivariate Analysis , Prognosis , Severity of Illness Index , SoftwareABSTRACT
The purpose of this study was to examine the litter size and stillbirth rate associated with cesarean-derived litters and to examine the relationship between the number of previous cesarean sections a sow had experienced and litter size. The records of 856 cesarean sections during a ten-year period from 1978 to 1988 were examined. The mean litter size was 10.6 +/- 3.2 and the mean number of stillborn pigs per litter was 0.2 +/- 0.8. The low level of stillbirth observed in this study provides supportive evidence to many earlier publications suggesting that the majority of stillborn pigs die during the birth process itself.A negative correlation between the number of previous cesarean sections a sow had experienced and litter size (r(2) = 0.015, p < 0.001) was observed. The slope of the regression line was -0.55, suggesting that litter size is reduced by approximately one-half a piglet for each cesarean section the sow has previously experienced. There are many factors which influence the variation in litter size. The small correlation coefficient (r(2) = 0.015) observed in this study indicates that only 1.5% of the variation in litter size can be explained by the number of previous cesarean sections that the sow has experienced.