Surveillance of microbial resistance in European Intensive Care Units: a first report from the Care-ICU programme for improved infection control.

PURPOSE: To report initial results from a European ICU surveillance programme focussing on antibiotic consumption, microbial resistance and infection control. METHODS: Thirty-five ICUs participated during 2005. Microbial resistance, antibiotic consumption and infection control stewardship measures were entered locally into a web-application. Results were validated locally, aggregated by project leaders and fed back to support local audit and benchmarking. RESULTS: Median (range) antibiotic consumption was 1,254 (range 348-4,992) DDD per 1,000 occupied bed days. The proportion of MRSA was median 11.6% (range 0-100), for ESBL phenotype of E. coli and K. pneumoniae 3.9% (0-80) and 14.3% (0-77.8) respectively, and for carbapenem-resistant P. aeruginosa 22.5% (0-100). Screening on admission for alert pathogens was commonly omitted, and there was a lack of single rooms for isolation. CONCLUSIONS: The surveillance programme demonstrated wide variation in antibiotic consumption, microbial resistance and infection control measures. The programme may, by providing rapid access to aggregated results, promote local and regional audit and benchmarking of antibiotic use and infection control practices.

Non-compliance with a postmastectomy radiotherapy guideline: decision tree and cause analysis.

BACKGROUND: The guideline for postmastectomy radiotherapy (PMRT), which is prescribed to reduce recurrence of breast cancer in the chest wall and improve overall survival, is not always followed. Identifying and extracting important patterns of non-compliance are crucial in maintaining the quality of care in Oncology. METHODS: Analysis of 759 patients with malignant breast cancer using decision tree induction (DTI) found patterns of non-compliance with the guideline. The PMRT guideline was used to separate cases according to the recommendation to receive or not receive PMRT. The two groups of patients were analyzed separately. Resulting patterns were transformed into rules that were then compared with the reasons that were extracted by manual inspection of records for the non-compliant cases. RESULTS: Analyzing patients in the group who should receive PMRT according to the guideline did not result in a robust decision tree. However, classification of the other group, patients who should not receive PMRT treatment according to the guideline, resulted in a tree with nine leaves and three of them were representing non-compliance with the guideline. In a comparison between rules resulting from these three non-compliant patterns and manual inspection of patient records, the following was found: In the decision tree, presence of perigland growth is the most important variable followed by number of malignantly invaded lymph nodes and level of Progesterone receptor. DNA index, age, size of the tumor and level of Estrogen receptor are also involved but with less importance. From manual inspection of the cases, the most frequent pattern for non-compliance is age above the threshold followed by near cut-off values for risk factors and unknown reasons. CONCLUSION: Comparison of patterns of non-compliance acquired from data mining and manual inspection of patient records demonstrates that not all of the non-compliances are repetitive or important. There are some overlaps between important variables acquired from manual inspection of patient records and data mining but they are not identical. Data mining can highlight non-compliance patterns valuable for guideline authors and for medical audit. Improving guidelines by using feedback from data mining can improve the quality of care in oncology.

Antibiotic susceptibility patterns and clones of Pseudomonas aeruginosa in Swedish ICUs.

Pseudomonas aeruginosa is 1 of the bacteria most adaptive to anti-bacterial treatment. Previous studies have shown nosocomial spread and transmission of clonal strains of P. aeruginosa in European hospitals. In this study we investigated antibiotic susceptibility and clonality in 101 P. aeruginosa isolates from 88 patients admitted to 8 Swedish ICUs during 2002. We also compared phenotypes and genotypes of P. aeruginosa and carried out cluster analysis to determine if phenotypic data can be used for surveillance of clonal spread. All isolates were collected on clinical indication as part of the NPRS II study in Sweden and were subjected to AFLP analysis for genotyping. 68 isolates with unique genotypes were found. Phenotyping was performed using MIC values for 5 anti-pseudomonal agents. Almost 6% of the isolates were multi-drug resistant (MDR), and this figure rose to almost 8% when intermediate isolates were also included. We found probable clonal spread in 9 cases, but none of them was found to be an MDR strain. Phenotypical cluster analysis produced 40 clusters. Comparing partitions did not demonstrate any significant concordance between the typing methods. The conclusion of our study is that cross-transmission and clonal spread of MDR P. aeruginosa does not present a clinical problem in Swedish ICUs, but probable cross-transmission of non-MDR clones indicate a need for improved hygiene routines bedside. The phenotype clusters were not concordant with genotype clusters, and genotyping is still recommended for epidemiological tracking.

A data mining approach to analyze non-compliance with a guideline for the treatment of breast cancer.

Postmastectomy radiotherapy (PMRT) is prescribed in order to reduce the local recurrence of breast cancer and improve overall survival. A guideline supports the trade-off between benefits and adverse effects of PMRT. However, this guideline is not always followed in practice. This study tries to find a method for revealing patterns of non-compliance between the actual treatment and the PMRT guideline. Data from breast cancer patients admitted to Linköping University Hospital between 1990 and 2000 were analyzed in this study. Cases that were not treated in accordance with the guideline were selected and analyzed by decision tree induction (DTI). Thereafter, four resulting rules, as representations for groups of patients, were compared to the guideline. Finding patterns of non-compliance with guidelines by means of rules can be an appropriate alternative to manual methods, i.e. a case-by-case comparison when studying very large datasets. The resulting rules can be used in a knowledge base of a guideline-based decision support system to alert when inconsistencies with the guidelines may appear.

Predicting metastasis in breast cancer: comparing a decision tree with domain experts.

Breast malignancy is the second most common cause of cancer death among women in Western countries. Identifying high-risk patients is vital in order to provide them with specialized treatment. In some situations, such as when access to experienced oncologists is not possible, decision support methods can be helpful in predicting the recurrence of cancer. Three thousand six hundred ninety-nine breast cancer patients admitted in south-east Sweden from 1986 to 1995 were studied. A decision tree was trained with all patients except for 100 cases and tested with those 100 cases. Two domain experts were asked for their opinions about the probability of recurrence of a certain outcome for these 100 patients. ROC curves, area under the ROC curves, and calibration for predictions were computed and compared. After comparing the predictions from a model built by data mining with predictions made by two domain experts, no significant differences were noted. In situations where experienced oncologists are not available, predictive models created with data mining techniques can be used to support physicians in decision making with acceptable accuracy.

Prescription of antibiotic agents in Swedish intensive care units is empiric and precise.

Since the prescription of antibiotics in the hospital setting is often empiric, particularly in the critically ill, and therefore fraught with potential error, we analysed the use of antibiotic agents in Swedish intensive care units (ICUs). We examined indications for antibiotic treatment, agents and dosage prescribed among 393 patients admitted to 23 ICUs at 7 tertiary care centres, 11 secondary hospitals and 5 primary hospitals over a 2-week period in November 2000. Antibiotic consumption was higher among ICU patients in tertiary care centres with a median of 84% (range 58-87%) of patients on antibiotics compared to patients in secondary hospitals (67%, range 35-93%) and in primary hospitals (38%, range 24-80%). Altogether 68% of the patients received antibiotics during the ICU stay compared to 65% on admission. Cefuroxime was the most commonly prescribed antibiotic before and during admission (28% and 24% of prescriptions, respectively). A date for decision to continue or discontinue antibiotic therapy was set in 21% (6/29) of patients receiving prophylaxis, in 8% (16/205) receiving empirical treatment and in 3% (3/88) when culture-based therapy was given. No correlation between antibiotic prescription and laboratory parameters such as CRP levels, leukocyte and thrombocyte counts, was found. The treatment was empirical in 64% and prophylactic in 9% of cases. Microbiological data guided prescription more often in severe sepsis (median 50%, range 40-60% of prescriptions) than in other specified forms of infection (median 32%, range 21-50%). The empirically chosen antibiotic was found to be active in vitro against the pathogens found in 55 of 58 patients (95%) with a positive blood culture. This study showed that a high proportion of ICU patients receive antimicrobial agents and, as expected, empirical-based therapy is more common than culture-based therapy. Antibiotics given were usually active in vitro against the pathogen found in blood cultures. We ascribe this to a relatively modest antibiotic resistance problem in Swedish hospitals.

An approach for generating fuzzy rules from decision trees.

Identifying high-risk breast cancer patients is vital both for clinicians and for patients. Some variables for identifying these patients such as tumor size are good candidates for fuzzification. In this study, Decision Tree Induction (DTI) has been applied to 3949 female breast cancer patients and crisp If-Then rules has been acquired from the resulting tree. After assigning membership functions for each variable in the crisp rules, they were converted into fuzzy rules and a mathematical model was constructed. One hundred randomly selected cases were examined by this model and compared with crisp rules predictions. The outcomes were examined by the area under the ROC curve (AUC). No significant difference was noticed between these two approaches for prediction of recurrence of breast cancer. By soft discretization of variables according to resulting rules from DTI, a predictive model, which is both more robust to noise and more comprehensible for clinicians, can be built.

Canonical correlation analysis of risk factors and clinical outcomes in cardiac surgery.

Assessment of the association between risk factors and outcomes in cardiac surgery is a complex problem. The aim of this study was to explore the relationship between possible risk factors and several clinical outcomes in cardiac surgery by using canonical correlation analysis (CCA). This retrospective study of 2605 consecutive adult patients who underwent cardiac surgery, evaluated 74 potential risk factors and up to 12 outcomes by canonical correlation analysis. For three serious outcomes, sternal wound complications/mediastinitis, cerebral complications, and perioperative myocardial infarctions, CCA was preceded by univariate analyses and backward stepwise multivariate logistic regression analyses. The CCA suggests that the major risk factors for complications in these models are intraoperative and postoperative risk factors. The power of risk prediction models developed with multivariate regression analysis can be enhanced by application of canonical correlation analysis, thereby offering new ways of analyzing and interpreting sets of potential risk factors in relation to sets of clinical outcomes.

Canonical correlation analysis for data reduction in data mining applied to predictive models for breast cancer recurrence.

Data mining methods can be used for extracting specific medical knowledge such as important predictors for recurrence of breast cancer in pertinent data material. However, when there is a huge quantity of variables in the data material it is first necessary to identify and select important variables. In this study we present a preprocessing method for selecting important variables in a dataset prior to building a predictive model.In the dataset, data from 5787 female patients were analysed. To cover more predictors and obtain a better assessment of the outcomes, data were retrieved from three different registers: the regional breast cancer, tumour markers, and cause of death registers. After retrieving information about selected predictors and outcomes from the different registers, the raw data were cleaned by running different logical rules. Thereafter, domain experts selected predictors assumed to be important regarding recurrence of breast cancer. After that, Canonical Correlation Analysis (CCA) was applied as a dimension reduction technique to preserve the character of the original data.Artificial Neural Network (ANN) was applied to the resulting dataset for two different analyses with the same settings. Performance of the predictive models was confirmed by ten-fold cross validation. The results showed an increase in the accuracy of the prediction and reduction of the mean absolute error.

Exploring cancer register data to find risk factors for recurrence of breast cancer--application of Canonical Correlation Analysis.

BACKGROUND: A common approach in exploring register data is to find relationships between outcomes and predictors by using multiple regression analysis (MRA). If there is more than one outcome variable, the analysis must then be repeated, and the results combined in some arbitrary fashion. In contrast, Canonical Correlation Analysis (CCA) has the ability to analyze multiple outcomes at the same time. One essential outcome after breast cancer treatment is recurrence of the disease. It is important to understand the relationship between different predictors and recurrence, including the time interval until recurrence. This study describes the application of CCA to find important predictors for two different outcomes for breast cancer patients, loco-regional recurrence and occurrence of distant metastasis and to decrease the number of variables in the sets of predictors and outcomes without decreasing the predictive strength of the model. METHODS: Data for 637 malignant breast cancer patients admitted in the south-east region of Sweden were analyzed. By using CCA and looking at the structure coefficients (loadings), relationships between tumor specifications and the two outcomes during different time intervals were analyzed and a correlation model was built. RESULTS: The analysis successfully detected known predictors for breast cancer recurrence during the first two years and distant metastasis 2-4 years after diagnosis. Nottingham Histologic Grading (NHG) was the most important predictor, while age of the patient at the time of diagnosis was not an important predictor. CONCLUSION: In cancer registers with high dimensionality, CCA can be used for identifying the importance of risk factors for breast cancer recurrence. This technique can result in a model ready for further processing by data mining methods through reducing the number of variables to important ones.

High antibiotic susceptibility among bacterial pathogens in Swedish ICUs. Report from a nation-wide surveillance program using TA90 as a novel index of susceptibility.

Local infection control measures, antibiotic consumption and patient demographics from 1999-2000 together with bacteriological analyses were investigated in 29 ICUs participating in the ICU-STRAMA programme. The median antibiotic consumption per ICU was 1147 (range 605-2143) daily doses per 1000 occupied bed d (DDD1000). Antibiotics to which > 90% of isolates of an organism were susceptible were defined as treatment alternatives (TA90). The mean number of TA90 was low (1-2 per organism) for Enterococcus faecium (vancomycin:VAN), coagulase negative staphylococci (VAN), Pseudomonas aeruginosa (ceftazidime:CTZ, netilmicin: NET) and Stenotrophomonas maltophilia (CTZ, trimethoprim-sulfamethoxazole: TSU), but higher (3-7) for Acinetobacter spp. (imipenem:IMI, NET, TSU), Enterococcus faecalis (ampicillin:AMP, IMI, VAN), Serratia spp. (ciprofloxacin:CIP, IMI, NET), Enterobacter spp. (CIP, IMI, NET, TSU), E. coli (cefuroxime:CXM, cefotaxime/eftazidime:CTX/CTZ, CIP, IMI, NET, piperacillin-tazobactam:PTZ, TSU), Klebsiella spp. (CTX/CTZ CIP, IMI, NET, PTZ, TSU) and Staphylococcus aureus (clindamycin, fusidic acid, NET, oxacillin, rifampicin, VAN). Of S. aureus isolates 2% were MRSA. Facilities for alcohol hand disinfection at each bed were available in 96% of the ICUs. The numbers of TA90 available were apparently higher than in ICUs in southern Europe and the US, despite a relatively high antibiotic consumption. This may be due to a moderate ecological impact of the used agents and the infection control routines in Swedish ICUs.

Comparison of the Glasgow Coma Scale and the Reaction Level Scale for assessment of cerebral responsiveness in the critically ill.

OBJECTIVE: The Glasgow Coma Scale (GCS) is a well-known source of error in outcome prediction models. We compared assessment of cerebral responsiveness with the GCS and the Reaction Level Scale (RLS) in two otherwise similar outcome prediction models. DESIGN AND SETTING: Prospective, observational study in a general intensive care unit. PATIENTS AND PARTICIPANTS: All admissions of patients with or at risk of developing impaired brain function between 1997 and 1998 ( n=534). MEASUREMENTS AND RESULTS: Admissions were scored by RLS and APACHE II (includes scoring with the GCS). The RLS scores were transformed to APACHE II central nervous system scores according to a predetermined protocol. APACHE II estimated probability of death was calculated conventionally with the GCS and the RLS. Vital status 90 days after admission was secured from a national database. Bias and precision was 0.5% and 16.6%, respectively. The area under receiver operating characteristic curves was slightly but significantly greater with the RLS-based APACHE II model than with the GCS-based model (0.92 vs. 0.90). Discrimination was improved primarily in admissions with low and intermediate probability of death. CONCLUSIONS: Scoring of cerebral responsiveness with the RLS instead of the GCS was associated with minimal bias of the APACHE II probability of death estimate. Assessment of consciousness in critically ill with the RLS deserves further evaluation

Risk factor analysis of early and delayed cerebral complications after cardiac surgery.

OBJECTIVE: To report the incidence, severity, and possible risk factors for early and delayed cerebral complications. DESIGN: Retrospective study. SETTING: Linköping University Hospital, Sweden. PARTICIPANTS: Consecutive patients who underwent cardiac surgery in the period July 1996 through June 2000 (n = 3,282). INTERVENTIONS: A standard cardiopulmonary bypass (CPB) technique was used for most patients. Postoperative anticoagulant treatment included heparin or anti-Xa dalteparin. Patients undergoing coronary artery bypass graft surgery received acetylsalicylic acid, and patients undergoing valve surgery received warfarin. MEASUREMENTS AND MAIN RESULTS: Cerebral complications occurred in 107 patients (3.3%). Of these, 60 (1.8%) were early, and 33 (1.0%) were delayed, and in 14 (0.4%) patients the onset was unknown. There were 37 variables in univariate analysis (p < 0.15) and 14 variables in multivariate analysis (p < 0.05) associated with cerebral complications. Predictors of early cerebral complications were older age, preoperative hypertension, aortic aneurysm surgery, prolonged CPB time, hypotension at CPB completion and soon after CPB, and postoperative arrhythmia and supraventricular tachyarrhythmia. Predictors of delayed cerebral complications were female gender, diabetes, previous cerebrovascular disease, combined valve surgery and coronary artery bypass graft surgery, postoperative supraventricular tachyarrhythmia, and prolonged ventilator support. Early cerebral complications seem to be more serious, with more permanent deficits and a higher overall mortality (35.0% v 18.2%). CONCLUSION: Most cerebral complications had an early onset. The results of this study suggest that aggressive antiarrhythmic treatment and blood pressure control may imfurther prove the cerebral outcome after cardiac surgery.

A variance-based measure of inter-rater agreement in medical databases.

The increasing use of encoded medical data requires flexible tools for data quality assessment. Existing methods are not always adequate, and this paper proposes a new metric for inter-rater agreement of aggregated diagnostic data. The metric, which is applicable in prospective as well as retrospective coding studies, quantifies the variability in the coding scheme, and the variation can be differentiated in categories and in coders. Five alternative definitions were compared in a set of simulated coding situations and in the context of mortality statistics. Two of them were more effective, and the choice between them must be made according to the situation. The metric is more powerful for larger numbers of coded cases, and Type I errors are frequent when coding situations include different numbers of cases. We also show that it is difficult to interpret the meaning of variation when the structures of the compared coding schemes differ.