Risk stratification for in-hospital mortality in spontaneous intracerebral haemorrhage: a Classification and Regression Tree analysis

Clinical and neuroimaging factors associated with 30-day fatality among indigenous West Africans with spontaneous intracerebral hemorrhage

BACKGROUND

Spontaneous intracerebral hemorrhage (ICH) is associated with a high case fatality rate in resource-limited settings. The independent predictors of poor outcome after ICH in sub-Saharan Africa remains to be characterized in large epidemiological studies. We aimed to determine factors associated with 30-day fatality among West African patients with ICH.

METHODS

The Stroke Investigative Research and Educational Network (SIREN) study is a multicentre, case-control study conducted at 15 sites in Nigeria and Ghana. Adults aged ≥18 years with spontaneous ICH confirmed with neuroimaging. Demographic, cardiovascular risk factors, clinical features and neuroimaging markers of severity were assessed. The independent risk factors for 30-day mortality were determined using a multivariate logistic regression analysis with an adjusted odds ratio (OR) and 95% confidence interval (CI).

RESULTS

Among 964 patients with ICH, 590 (61.2%) were males with a mean age (SD) of 54.3(13.6) years and a case fatality of 34.3%. Factors associated with 30-day mortality among ICH patients include: Elevated mean National Institute of Health Stroke Scale(mNIHSS);(OR 1.06; 95% CI 1.02-1.11), aspiration pneumonitis; (OR 7.17; 95% CI 2.82-18.24), ICH volume > 30mls; OR 2.68; 95% CI 1.02-7.00)) low consumption of leafy vegetables (OR 0.36; 95% CI 0.15-0.85).

CONCLUSION

This study identified risk and protective factors associated with 30-day mortality among West Africans with spontaneous ICH. These factors should be further investigated in other populations in Africa to enable the development of ICH mortality predictions models among indigenous Africans.

Were metabolic and other chronic diseases the driven onset epidemic forces of COVID-19 in Mexico?

The underline hypothesis of this study was that SARS-CoV-2 can infect individuals regardless of health condition, sex, and age in opposition to the classical epidemiological assumption of an identifiable susceptible subpopulation for epidemic development. To address this issue, a population cohort with 24.4 million metadata associated with 226,089 official RT-qPCR positive and 283,450 negative cases, including 27,769 deceased, linked putatively to B.1. and B.1.1. SARS-CoV-2 lineages were analyzed. The analysis baseline was to determine the infection and mortality structure of the diseased cohort at the onset-exponential phase of the first epidemic wave in Mexico under the assumption of limited herd immunity. Individuals with nonchronic diseases (NOCDs) were compared with those exhibiting at least one of 10 chronic diseases (CDs) adjusted by age and sex. Risk factors for infection and mortality were estimated with classification and regression tree (CART) and cluster analysis based on Spearman's matrix of rho-values in RStudio®, complemented with two proposed mortality indices. SARS-CoV-2 infection was independent of health condition (52.8% NOCD vs. 47.2% CDs; p = 0.001-0.009) but influenced by age >46 in one risk analysis scenario (p < 0.001). Sex contributed 9.7% to the overall risk. The independent effect was supported by the health structure of negative cases with a similar tendency but a higher proportion of NOCDs (61.4%, p = 0.007). The infection probability in individuals with one CD was determined by the disease type and age, which was higher in those older individuals (≥56 years) exhibiting diabetes (12.3%, cp = 0.0006), hypertension (10.1%, cp < 0.0001), and obesity (7.8%, cp = 0.001). In contrast, the mortality risk was heavily influenced by CD conditioned by sex and age, accounting for 72.3% of total deaths (p = 0.001-0.008). Significant mortality risk (48%) was comprised of women and men (w, m) aged ≥56 years with diabetes (19% w and 27.9% m, cp < 0.0004), hypertension (11.5% w, cp = 0.0001), and CKD (3.5% w and 5.3% m, cp = 0.0009). Older people with diabetes and hypertension comorbidity increased the risk to 60.5% (p = 0.001). Based on a mortality-weighted index, women were more vulnerable to preexisting metabolic or cardiovascular diseases. These findings support our hypothesis and justify the need for surveillance systems at a communitarian level. This is the first study addressing this fundamental epidemiological question.

Preliminary Prognostication for Good Neurological Outcomes in the Early Stage of Post-Cardiac Arrest Care

We investigated prognostic strategies for predicting good outcomes in the early stage of post-cardiac-arrest care using multiple prognostic tests that are available until 24 h after the return of spontaneous circulation (ROSC). A retrospective analysis was conducted on 138 out-of-hospital cardiac-arrest patients who underwent prognostic tests, including the gray-white-matter ratio (GWR-BG), the Glasgow Coma Scale motor (GCS-M) score before sedative administration, and the neuron-specific enolase (NSE) level measured at 24 h after the ROSC. We investigated the prognostic performances of the tests as single predictors and in various combination strategies. Classification and regression-tree analysis were used to provide a reliable model for the risk stratification. Out of all the patients, 55 (44.0%) had good outcomes. The NSE level showed the highest prognostic performance as a single prognostic test and provided improved specificities (>70%) and sensitivities (>98%) when used in combination strategies. Low NSE levels (≤32.1 ng/mL) and high GCS-M (≥4) scores identified good outcomes without misclassification. The overall accuracy for good outcomes was 81.8%. In comatose patients with low NSE levels or high GCS-M scores, the premature withdrawal of life-sustaining therapy should be avoided, thereby complying with the formal prognostication-strategy algorithm after at least 72 h from the ROSC.

Immature granulocyte and other markers in prediction of mortality in spontaneous intracerebral hemorrhage

Aim: This study aims to evaluate immature granulocyte count (IG#) and percentage (IG%) in the prediction of mortality in spontaneous intracerebral hemorrhage (SICH). Material and Method: Demographic characteristics and laboratory test results of patients diagnosed with SICH and admitted to the neurology clinic in a tertiary hospital between January 1, 2020, and January 1, 2022, were recorded. One hundred ten patients were included in the study. While 80 of these patients constituted the group that recovered after treatment, 30 of them formed the group that died despite treatment. IG and other laboratory and clinic parameters were statistically compared in both groups. Results: Of 110 patients, 45 (42.7%) were female, and 65 (57.3%) were male. IG counts were higher in the non-survival group than in the survival group (p=0.001). When the patients were divided according to low IG% (=0.6), 30 patients were in the high IG# group, and 80 patients were in the low IG% group. White blood cell (WBC), neutrophil count (NEUT#), monocyte count (MONO#), IG#, neutrophil-lymphocyte ratio (NLR), and hemorrhage volume (HV) values were statistically significantly higher in the high IG% group than in the low IG% group; Glasgow coma score (GCS) and percentage of lymphocytes (LYMPH%) values were significantly lower too. In addition, the mortality rate in the high IG# group was significantly higher than the mortality rate in the low IG% group (53.23% vs. 17.5%). Conclusion: IG is a new, easily accessible, inexpensive, and promising marker for predicting in-hospital mortality in patients with SICH.

Risk of suicide attempts and self-harm after 1.4 million general medical hospitalizations of men with mental illness

BACKGROUND

The short-term risk of suicide after medical hospital discharge is four times higher among men compared with women. As previous work has identified female-specific antecedents of suicide-related behavior after medical hospitalization of women with serious mental illness, we examined predictors among a similar population of men with multimorbidity.

METHODS

Classification and regression tree (CART) models were developed and validated using electronic health records (EHRs) from 1,423,161 medical (non-psychiatric) hospitalizations of men ≥ 18-years-old with an existing diagnosis of a depressive disorder, bipolar disorder, or chronic psychosis. Hospitalizations occurred between 2009 and 2017. Risk groups were evaluated using an independent testing set. The primary outcome was readmission within one year associated with ICD-9 or -10 code for self-harm or attempt.

RESULTS

The 1-year readmission rate for intentional self-harm and suicide attempt was 3.9% (55,337/1,423,161 hospitalizations). The classification model discriminated risk with area under the curve (AUC) 0.73 (Confidence Interval [95%CI] 0.68-0.74), accuracy 0.82 (95%CI 0.71-0.83), sensitivity 82.6% (95%CI 81.2-84), and specificity 83.1% (95%CI 81.7-84.5). Strongest predictors were medical comorbidity, prior self-harm, age, and prior hospitalization. Men with greater medical comorbidity burden and prior self-harm were at highest risk (Odds Ratio [OR] 3.10, 95%CI 3.02-3.18), as were men < 62-years-old with few medical comorbidities (OR 1.11 95%CI 1.08-1.13).

LIMITATIONS

The study focused on medical hospitalizations for suicide attempt and thus captured only severe attempts resulting in hospitalization.

CONCLUSIONS

After medical hospitalization, men with serious mental illness experienced a high risk of self-harm (1:25 hospitalizations). Risk was particularly elevated among younger patients without prior medical conditions and older patients with medical comorbidity and prior self-harm.

Convenient Decision Criteria for Surgery in Elderly Patients with Oral Squamous Cell Carcinoma

Elderly patients with oral squamous cell carcinoma (OSCC) have difficulty undergoing curative surgical treatment due to various factors besides age. The purpose of the present study was to study the factors determining surgery in elderly patients with OSCC. We designed and implemented a retrospective cohort study. The study sample included elderly patients aged ≥ 70 years with OSCC and they were statistically compared between the surgery and non-surgery groups. The primary outcome variable was selecting surgery as the treatment plan, while the secondary outcome was the prognosis of each group. The sample comprised 76 patients aged ≥ 70 years with OSCC, of whom 52 treated with surgery and 24 patients treated with non-surgery. As decision factors, performance status (PS), clinical stage, serum Alb level, body mass index (BMI), and Geriatric Nutritional Risk Index (GNRI) were significantly associated with the selection of surgery. Logistic multivariate analysis identified three independent predictive factors for selecting surgery: Alb (≥3.5 vs. <3.5), PS (0, 1, 2, 3), and clinical stage. According to the decision tree analysis, curative surgery is the recommended treatment strategy for elderly patients with Alb ≥ 3.5 g/dL, PS 0, and stage I, II. In conclusion, Alb, PS, and clinical stage may be the criteria for selecting surgery in elderly patients.

CLASSIFICATION AND REGRESSION TREE ANALYSIS FOR PREDICTING PROGNOSIS IN WILDLIFE REHABILITATION: A CASE STUDY OF LEPTOSPIROSIS IN CALIFORNIA SEA LIONS (ZALOPHUS CALIFORNIANUS)

The spirochete bacterium Leptospira interrogans serovar Pomona is enzootic to California sea lions (CSL; Zalophus californianus) and causes periodic epizootics. Leptospirosis in CSL is associated with a high fatality rate in rehabilitation. Evidence-based tools for estimating prognosis and guiding early euthanasia of animals with a low probability of survival are critical to reducing the severity and duration of animal suffering. Classification and regression tree (CART) analysis of clinical data was used to predict survival outcomes of CSL with leptospirosis in rehabilitation. Classification tree outputs are binary decision trees that can be readily interpreted and applied by a clinician. Models were trained using data from cases treated from 2017 to 2018 at The Marine Mammal Center in Sausalito, CA, and tested against data from cases treated from 2010 to 2012. Two separate classification tree analyses were performed, one including and one excluding data from euthanized animals. When data from natural deaths and euthanasias were included in model-building, the best classification tree predicted outcomes correctly for 84.7% of cases based on four variables: appetite over the first 3 days in care, and blood urea nitrogen (BUN), creatinine, and sodium at admission. When only natural deaths were included, the best model predicted outcomes correctly for 87.6% of cases based on BUN and creatinine at admission. This study illustrates that CART analysis can be successfully applied to wildlife in rehabilitation to establish evidence-based euthanasia criteria with the goal of minimizing animal suffering. In the context of a large epizootic that challenges the limits of a facility's capacity for care, the models can assist in maximizing allocation of resources to those animals with the highest predicted probability of survival. This technique may be a useful tool for other diseases seen in wildlife rehabilitation.

Application of Artificial Intelligence Algorithms to Estimate the Success Rate in Medically Assisted Procreation

The aim of this study was to build an Artificial Neural Network (ANN) complemented by a decision tree to predict the chance of live birth after an In Vitro Fertilization (IVF)/Intracytoplasmic Sperm Injection (ICSI) treatment, before the first embryo transfer, using demographic and clinical data. Overall, 26 demographic and clinical data from 1193 cycles who underwent an IVF/ICSI treatment at Centro de Infertilidade e Reprodução Medicamente Assistida, between 2012 and 2019, were analyzed. An ANN was constructed by selecting experimentally the input variables which most correlated to the target through Pearson correlation. The final used variables were: woman’s age, total dose of gonadotropin, number of eggs, number of embryos and Antral Follicle Count (AFC). A decision tree was developed considering as an initial set the input variables integrated in the previous model. The ANN model was validated by the holdout method and the decision tree model by the 10-fold cross method. The ANN accuracy was 75.0% and the Area Under the Receiver Operating Characteristic (AUROC) curve was 75.2% (95% Confidence Interval (CI): 72.5–77.5%), whereas the decision tree model reached 75.0% and 74.9% (95% CI: 72.3–77.5%). These results demonstrated that both ANN and decision tree methods are fair for prediction the chance of conceive after an IVF/ICSI cycle.

Prediction models considering psychological factors to identify pain relief in conservative treatment of people with knee osteoarthritis: A multicenter, prospective cohort study

BACKGROUND

Pain-related affective and/or cognitive characteristics such as depressive symptoms, pain catastrophizing, and self-efficacy are known to exacerbate pain in people with knee osteoarthritis. However, no studies have investigated whether these psychological factors can interfere with pain relief during conservative treatment. The object of this study was to assess the prediction models considering psychological factors to predict pain relief in people with knee osteoarthritis receiving conservative treatment.

METHODS

Study design was a multicenter, and prospective cohort study. Data were collected in the department of physical therapy in 1 hospital and 7 orthopedic clinics. Eighty-eight people with knee osteoarthritis participated in this study and were followed for 3 months. The numeric rating scale and the Knee Injury and Osteoarthritis Outcome Score scale were used to evaluate pain relief. Potential predictors for pain relief were depressive symptoms, self-efficacy, and pain catastrophizing. The classification and regression trees methodology was used to develop the model for predicting the presence of pain relief at 1 and 3 months after the start of observation. The prediction accuracy was evaluated using the area under the receiver operating characteristic curves (AUCs).

RESULTS

The model at 1 month after the start of observation included pain intensity at baseline, positive affect, and disease duration. The AUC of this model was 0.793 (95% confidential interval, 0.687-0.898). The model at 3 months after the start of observation included pain catastrophizing and self-efficacy. The AUC of this model was 0.808 (95% confidential interval, 0.682-0.934).

CONCLUSIONS

The accuracy of prediction model considering pain-related affective and/or cognitive characteristics is moderate for pain relief in people with knee osteoarthritis receiving conservative treatment.

Prognosis and futility in neurosurgical emergencies: A review

A patient with a life-threatening intracranial insult presents a difficult situation to the neurosurgeon. In a few short minutes the neurosurgeon must assess the patient's neurologic status, imaging, and medical condition then confer with the patient's proxy regarding treatment. This assessment ideally includes recognition of situations where aggressive care is futile and therefore such treatments should not be offered. The proxy discussion must involve surgical and nonsurgical management options and the impact of these options on survival and residual disability. Surgical decision-making is frequently difficult, even for designated proxies armed with advance directives, as these documents are usually vague with regard to acceptable functional outcomes. To complicate things further, when emergencies are off-hours, housestaff or physician extenders may need to represent the medical team in these discussions so that surgical treatment, if desired, can be arranged expeditiously. These difficulties sometimes lead to the performance of emergent surgical procedures in situations where poor outcome is certain, with deleterious effects to the patient, family, and healthcare system. It is clear then that neurosurgeons as well as their housestaff and extenders should have working knowledge of prognostic information relating to intracranial insults and familiarity with the complex ethical concept of medical futility. In this paper we review the relevant literature and our goal is to juxtapose these topics so as to provide a framework for decision making in that critical time.

Assessment and Comparison of the Four Most Extensively Validated Prognostic Scales for Intracerebral Hemorrhage: Systematic Review with Meta-analysis

BACKGROUND/OBJECTIVE

Intracerebral hemorrhage (ICH) is a devastating disorder, responsible for 10% of all strokes. Several prognostic scores have been developed for this population to predict mortality and functional outcome. The aim of this study was to determine the four most frequently validated and most widely used scores, assess their discrimination for both outcomes by means of a systematic review with meta-analysis, and compare them using meta-regression.

METHODS

PubMed, ISI Web of Knowledge, Scopus, and CENTRAL were searched for studies validating the ICH score, ICH-GS, modified ICH, and the FUNC score in ICH patients. C-statistic was chosen as the measure of discrimination. For each score and outcome, C-statistics were aggregated at four different time points using random effect models, and heterogeneity was evaluated using the I² statistic. Score comparison was undertaken by pooling all C-statistics at different time points using robust variance estimation (RVE) and performing meta-regression, with the score used as the independent variable.

RESULTS

Fifty-three studies were found validating the original ICH score, 14 studies were found validating the ICH-GS, eight studies were found validating the FUNC score, and five studies were found validating the modified ICH score. Most studies attempted outcome prediction at 3 months or earlier. Pooled C-statistics ranged from 0.76 for FUNC functional outcome prediction at discharge to 0.85 for ICH-GS mortality prediction at 3 months, but heterogeneity was high across studies. RVE showed the ICH score retained the highest discrimination for mortality (c = 0.84), whereas the modified ICH score retained the highest discrimination for functional outcome (c = 0.80), but these differences were not statistically significant.

CONCLUSIONS

The ICH score is the most extensively validated score in ICH patients and, in the absence of superior prediction by other scores, should preferably be used. Further studies are needed to validate prognostic scores at longer follow-ups and assess the reasons for heterogeneity in discrimination.

Artificial liver support system therapy in acute-on-chronic hepatitis B liver failure: Classification and regression tree analysis

Artificial liver support systems (ALSS) are widely used to treat patients with hepatitis B virus-related acute-on-chronic liver failure (HBV-ACLF). The aims of the present study were to investigate the subgroups of patients with HBV-ACLF who may benefit from ALSS therapy, and the relevant patient-specific factors. 489 ALSS-treated HBV-ACLF patients were enrolled, and served as derivation and validation cohorts for classification and regression tree (CART) analysis. CART analysis identified three factors prognostic of survival: hepatic encephalopathy (HE), prothrombin time (PT), and total bilirubin (TBil) level; and two distinct risk groups: low (28-day mortality 10.2-39.5%) and high risk (63.8-91.1%). The CART model showed that patients lacking HE and with a PT ≤ 27.8 s and a TBil level ≤455 μmol/L experienced less 28-day mortality after ALSS therapy. For HBV-ACLF patients with HE and a PT > 27.8 s, mortality remained high after such therapy. Patients lacking HE with a PT ≤ 27.8 s and TBil level ≤ 455 μmol/L may benefit markedly from ALSS therapy. For HBV-ACLF patients at high risk, unnecessary ALSS therapy should be avoided. The CART model is a novel user-friendly tool for screening HBV-ACLF patient eligibility for ALSS therapy, and will aid clinicians via ACLF risk stratification and therapeutic guidance.

Prognostic models for intracerebral hemorrhage: systematic review and meta-analysis

Background

Prognostic tools for intracerebral hemorrhage (ICH) patients are potentially useful for ascertaining prognosis and recommended in guidelines to facilitate streamline assessment and communication between providers. In this systematic review with meta-analysis we identified and characterized all existing prognostic tools for this population, performed a methodological evaluation of the conducting and reporting of such studies and compared different methods of prognostic tool derivation in terms of discrimination for mortality and functional outcome prediction.

Methods

PubMed, ISI, Scopus and CENTRAL were searched up to 15th September 2016, with additional studies identified using reference check. Two reviewers independently extracted data regarding the population studied, process of tool derivation, included predictors and discrimination (c statistic) using a predesignated spreadsheet based in the CHARMS checklist. Disagreements were solved by consensus. C statistics were pooled using robust variance estimation and meta-regression was applied for group comparisons using random effect models.

Results

Fifty nine studies were retrieved, including 48,133 patients and reporting on the derivation of 72 prognostic tools. Data on discrimination (c statistic) was available for 53 tools, 38 focusing on mortality and 15 focusing on functional outcome. Discrimination was high for both outcomes, with a pooled c statistic of 0.88 for mortality and 0.87 for functional outcome. Forty three tools were regression based and nine tools were derived using machine learning algorithms, with no differences found between the two methods in terms of discrimination (p = 0.490). Several methodological issues however were identified, relating to handling of missing data, low number of events per variable, insufficient length of follow-up, absence of blinding, infrequent use of internal validation, and underreporting of important model performance measures.

Conclusions

Prognostic tools for ICH discriminated well for mortality and functional outcome in derivation studies but methodological issues require confirmation of these findings in validation studies. Logistic regression based risk scores are particularly promising given their good performance and ease of application.

Electronic supplementary material

The online version of this article (10.1186/s12874-018-0613-8) contains supplementary material, which is available to authorized users.

Classification of Different Degrees of Disability Following Intracerebral Hemorrhage: A Decision Tree Analysis from VISTA-ICH Collaboration

Background and purpose

Prognostication following intracerebral hemorrhage (ICH) has focused on poor outcome at the expense of lumping together mild and moderate disability. We aimed to develop a novel approach at classifying a range of disability following ICH.

Methods

The Virtual International Stroke Trial Archive collaboration database was searched for patients with ICH and known volume of ICH on baseline CT scans. Disability was partitioned into mild [modified Rankin Scale (mRS) at 90 days of 0–2], moderate (mRS = 3–4), and severe disabilities (mRS = 5–6). We used binary and trichotomy decision tree methodology. The data were randomly divided into training (2/3 of data) and validation (1/3 data) datasets. The area under the receiver operating characteristic curve (AUC) was used to calculate the accuracy of the decision tree model.

Results

We identified 957 patients, age 65.9 ± 12.3 years, 63.7% males, and ICH volume 22.6 ± 22.1 ml. The binary tree showed that lower ICH volume (<13.7 ml), age (<66.5 years), serum glucose (<8.95 mmol/l), and systolic blood pressure (<170 mm Hg) discriminate between mild versus moderate-to-severe disabilities with AUC of 0.79 (95% CI 0.73–0.85). Large ICH volume (>27.9 ml), older age (>69.5 years), and low Glasgow Coma Scale (<15) classify severe disability with AUC of 0.80 (95% CI 0.75–0.86). The trichotomy tree showed that ICH volume, age, and serum glucose can separate mild, moderate, and severe disability groups with AUC 0.79 (95% CI 0.71–0.87).

Conclusion

Both the binary and trichotomy methods provide equivalent discrimination of disability outcome after ICH. The trichotomy method can classify three categories at once, whereas this action was not possible with the binary method. The trichotomy method may be of use to clinicians and trialists for classifying a range of disability in ICH.

Classification and regression tree analysis of acute-on-chronic hepatitis B liver failure: Seeing the forest for the trees

At present, there is no ideal model for predicting the short-term outcome of patients with acute-on-chronic hepatitis B liver failure (ACHBLF). This study aimed to establish and validate a prognostic model by using the classification and regression tree (CART) analysis. A total of 1047 patients from two separate medical centres with suspected ACHBLF were screened in the study, which were recognized as derivation cohort and validation cohort, respectively. CART analysis was applied to predict the 3-month mortality of patients with ACHBLF. The accuracy of the CART model was tested using the area under the receiver operating characteristic curve, which was compared with the model for end-stage liver disease (MELD) score and a new logistic regression model. CART analysis identified four variables as prognostic factors of ACHBLF: total bilirubin, age, serum sodium and INR, and three distinct risk groups: low risk (4.2%), intermediate risk (30.2%-53.2%) and high risk (81.4%-96.9%). The new logistic regression model was constructed with four independent factors, including age, total bilirubin, serum sodium and prothrombin activity by multivariate logistic regression analysis. The performances of the CART model (0.896), similar to the logistic regression model (0.914, P=.382), exceeded that of MELD score (0.667, P<.001). The results were confirmed in the validation cohort. We have developed and validated a novel CART model superior to MELD for predicting three-month mortality of patients with ACHBLF. Thus, the CART model could facilitate medical decision-making and provide clinicians with a validated practical bedside tool for ACHBLF risk stratification.

Predictors of 30-day mortality in patients with spontaneous primary intracerebral hemorrhage

BACKGROUND

Intracerebral hemorrhage (ICH) is a life threatening entity, and an early outcome assessment is mandatory for optimizing therapeutic efforts.

METHODS

We retrospectively analyzed data from 342 patients with spontaneous primary ICH to evaluate possible predictors of 30-day mortality considering clinical, radiological, and therapeutical parameters. We also applied three widely accepted outcome grading scoring systems [(ICH score, FUNC score and intracerebral hemorrhage grading scale (ICH-GS)] on our population to evaluate the correlation of these scores with the 30-day mortality in our study. We also applied three widely accepted outcome grading scoring systems [(ICH score, FUNC score and intracerebral hemorrhage grading scale (ICH-GS)] on our population to evaluate the correlation of these scores with the 30-day mortality in our study.

RESULTS

From 342 patients (mean age: 67 years, mean Glasgow Coma Scale [GCS] on admission: 9, mean ICH volume: 62.19 ml, most common hematoma location: basal ganglia [43.9%]), 102 received surgical and 240 conservative treatment. The 30-day mortality was 25.15%. In a multivariate analysis, GCS (Odds ratio [OR] =0.726, 95% confidence interval [CI] =0.661-0.796, P < 0.001), bleeding volume (OR = 1.012 per ml, 95% CI = 1.007 - 1.017, P < 0.001), and infratentorial hematoma location (OR = 5.381, 95% CI = 2.166-13.356, P = 0.009) were significant predictors for the 30-day mortality. After receiver operating characteristics analysis, we defined a "high-risk group" for an unfavorable short-term outcome with GCS <11 and ICH volume >32 ml supratentorially or 21 ml infratentorially. Using Pearson correlation, we found a correlation of 0.986 between ICH score and 30-day mortality (P < 0.001), 0.853 between FUNC score and 30-day mortality (P = 0.001), and 0.924 between ICH-GS and 30-day mortality (P = 0.001).

CONCLUSIONS

GCS score on admission together with the baseline volume and localization of the hemorrhage are strong predictors for 30-day mortality in patients with spontaneous primary intracerebral hemorrhage, and by relying on them it is possible to identify high-risk patients with poor short-term outcome. The ICH score and the ICH-GS accurately predict the 30-day mortality.

Reliability of intracerebral hemorrhage classification systems: A systematic review

BACKGROUND

Accurately distinguishing non-traumatic intracerebral hemorrhage (ICH) subtypes is important since they may have different risk factors, causal pathways, management, and prognosis. We systematically assessed the inter- and intra-rater reliability of ICH classification systems.

METHODS

We sought all available reliability assessments of anatomical and mechanistic ICH classification systems from electronic databases and personal contacts until October 2014. We assessed included studies' characteristics, reporting quality and potential for bias; summarized reliability with kappa value forest plots; and performed meta-analyses of the proportion of cases classified into each subtype.

SUMMARY OF REVIEW

We included 8 of 2152 studies identified. Inter- and intra-rater reliabilities were substantial to perfect for anatomical and mechanistic systems (inter-rater kappa values: anatomical 0.78-0.97 [six studies, 518 cases], mechanistic 0.89-0.93 [three studies, 510 cases]; intra-rater kappas: anatomical 0.80-1 [three studies, 137 cases], mechanistic 0.92-0.93 [two studies, 368 cases]). Reporting quality varied but no study fulfilled all criteria and none was free from potential bias. All reliability studies were performed with experienced raters in specialist centers. Proportions of ICH subtypes were largely consistent with previous reports suggesting that included studies are appropriately representative.

CONCLUSIONS

Reliability of existing classification systems appears excellent but is unknown outside specialist centers with experienced raters. Future reliability comparisons should be facilitated by studies following recently published reporting guidelines.

Prognostic Tools for Early Mortality in Hemorrhagic Stroke: Systematic Review and Meta-Analysis

Background and Purpose

Several risk scores have been developed to predict mortality in intracerebral hemorrhage (ICH). We aimed to systematically determine the performance of published prognostic tools.

Methods

We searched MEDLINE and EMBASE for prognostic models (published between 2004 and April 2014) used in predicting early mortality (<6 months) after ICH. We evaluated the discrimination performance of the tools through a random-effects meta-analysis of the area under the receiver operating characteristic curve (AUC) or c-statistic. We evaluated the following components of the study validity: study design, collection of prognostic variables, treatment pathways, and missing data.

Results

We identified 11 articles (involving 41,555 patients) reporting on the accuracy of 12 different tools for predicting mortality in ICH. Most studies were either retrospective or post-hoc analyses of prospectively collected data; all but one produced validation data. The Hemphill-ICH score had the largest number of validation cohorts (9 studies involving 3,819 patients) within our systematic review and showed good performance in 4 countries, with a pooled AUC of 0.80 [95% confidence interval (CI)=0.77-0.85]. We identified several modified versions of the Hemphill-ICH score, with the ICH-Grading Scale (GS) score appearing to be the most promising variant, with a pooled AUC across four studies of 0.87 (95% CI=0.84-0.90). Subgroup testing found statistically significant differences between the AUCs obtained in studies involving Hemphill-ICH and ICH-GS scores (p=0.01).

Conclusions

Our meta-analysis evaluated the performance of 12 ICH prognostic tools and found greater supporting evidence for 2 models (Hemphill-ICH and ICH-GS), with generally good performance overall.

Could the clinical interpretability of subgroups detected using clustering methods be improved by using a novel two-stage approach?

Background

Recognition of homogeneous subgroups of patients can usefully improve prediction of their outcomes and the targeting of treatment. There are a number of research approaches that have been used to recognise homogeneity in such subgroups and to test their implications. One approach is to use statistical clustering techniques, such as Cluster Analysis or Latent Class Analysis, to detect latent relationships between patient characteristics.

Influential patient characteristics can come from diverse domains of health, such as pain, activity limitation, physical impairment, social role participation, psychological factors, biomarkers and imaging. However, such ‘whole person’ research may result in data-driven subgroups that are complex, difficult to interpret and challenging to recognise clinically.

This paper describes a novel approach to applying statistical clustering techniques that may improve the clinical interpretability of derived subgroups and reduce sample size requirements.

Methods

This approach involves clustering in two sequential stages. The first stage involves clustering within health domains and therefore requires creating as many clustering models as there are health domains in the available data. This first stage produces scoring patterns within each domain. The second stage involves clustering using the scoring patterns from each health domain (from the first stage) to identify subgroups across all domains. We illustrate this using chest pain data from the baseline presentation of 580 patients.

Results

The new two-stage clustering resulted in two subgroups that approximated the classic textbook descriptions of musculoskeletal chest pain and atypical angina chest pain. The traditional single-stage clustering resulted in five clusters that were also clinically recognisable but displayed less distinct differences.

Conclusions

In this paper, a new approach to using clustering techniques to identify clinically useful subgroups of patients is suggested. Research designs, statistical methods and outcome metrics suitable for performing that testing are also described. This approach has potential benefits but requires broad testing, in multiple patient samples, to determine its clinical value. The usefulness of the approach is likely to be context-specific, depending on the characteristics of the available data and the research question being asked of it.

A predictive rule for mortality of inpatients with Staphylococcus aureus bacteraemia: A classification and regression tree analysis

OBJECTIVE

To create a predictive rule to identify risk factors for mortality among patients with Staphylococcus aureus bacteraemia (SAB).

DESIGN, SETTING AND PATIENTS

This was a retrospective cohort study of all adult patients with SAB at a large community hospital in Tokyo, Japan, from April 1, 2004 to March 31, 2011. Baseline data and clinically relevant factors were collected from electronic charts. The primary outcome was in-hospital mortality. All candidate predictors were included in a classification and regression tree (CART) analysis. A receiver operating characteristic (ROC) curve was drawn, and the area under the curve (AUC) was obtained. A cross-validation analysis was performed.

MEASUREMENTS AND MAIN RESULTS

A total of 340 patients had SAB during the study period. Of these, 118 (34.7%) patients died in hospital. Among 41 potential variables, the CART analysis revealed that underlying malignancy, serum blood glucose level, methicillin resistance, and low serum albumin were predictors of mortality. The AUC was 0.73 (95% CI: 0.67-0.79). For validation, the estimated risk was 0.26 (± SE: 0.02) in the resubstitution analysis and 0.33 (± SE: 0.03) in the cross-validation analysis.

CONCLUSION

We propose a predictive model for the mortality of patients with SAB consisting of four predictors: underlying malignancy, low serum albumin, high glucose, and methicillin resistance. This model may facilitate appropriate preventative management for patients with SAB who are at high risk of mortality.

Validation of clinical prediction scores in patients with primary intracerebral hemorrhage

BACKGROUND

Initial reports of the FUNC score suggest that it may accurately identify those patients suffering from intracerebral hemorrhage (ICH) with an ultra low chance of functional neurologic recovery. This study's aim is to validate the FUNC score and determine if it accurately identifies the cohort of patients with an ultra low chance of survival with good neurologic recovery.

METHODS

Retrospective review of 501 consecutive primary ICH patients admitted from the Emergency Department to a large healthcare system. Performance of the FUNC, ICH-GS, and oICH scores was determined by calculating areas under the receiver-operator-characteristic curves. Patients with a predicted 100 % chance of poor neurologic outcome (PNO) (FUNC <4 and ICH-GS >10) scores were evaluated to determine if DNR impacted 90 day survival or rate of survival with a Glasgow Outcome Score of <3.

RESULTS

In 366 cases of primary ICH who presented during the study period, 222(61 %) survived to discharge. Both the FUNC (AUC: 0.873) and ICH-GS (AUC: 0.888) outperformed the oICH (AUC: 0.743) in predicting 90-day mortality (p = <0.001). Of 68 patients with a FUNC score <4, 67 (98.5 %) had PNO at discharge. The presence of DNR was not associated with a significant difference in the rate of PNO at discharge (40/40 = 100 % vs. 27/28 = 96.4 % p = 0.42) or 90-day mortality (40/40 = 100 % vs. 21/28 = 75 %, p = 0.06).

CONCLUSION

The FUNC and ICH-GS appear superior to the oICH in predicting outcome in patients with primary ICH. In addition, the FUNC score appears to accurately identify patients with low chance of functional neurologic recovery at discharge.

Hematoma shape, hematoma size, Glasgow coma scale score and ICH score: which predicts the 30-day mortality better for intracerebral hematoma?

PURPOSE

To investigate the performance of hematoma shape, hematoma size, Glasgow coma scale (GCS) score, and intracerebral hematoma (ICH) score in predicting the 30-day mortality for ICH patients. To examine the influence of the estimation error of hematoma size on the prediction of 30-day mortality.

MATERIALS AND METHODS

This retrospective study, approved by a local institutional review board with written informed consent waived, recruited 106 patients diagnosed as ICH by non-enhanced computed tomography study. The hemorrhagic shape, hematoma size measured by computer-assisted volumetric analysis (CAVA) and estimated by ABC/2 formula, ICH score and GCS score was examined. The predicting performance of 30-day mortality of the aforementioned variables was evaluated. Statistical analysis was performed using Kolmogorov-Smirnov tests, paired t test, nonparametric test, linear regression analysis, and binary logistic regression. The receiver operating characteristics curves were plotted and areas under curve (AUC) were calculated for 30-day mortality. A P value less than 0.05 was considered as statistically significant.

RESULTS

The overall 30-day mortality rate was 15.1% of ICH patients. The hematoma shape, hematoma size, ICH score, and GCS score all significantly predict the 30-day mortality for ICH patients, with an AUC of 0.692 (P = 0.0018), 0.715 (P = 0.0008) (by ABC/2) to 0.738 (P = 0.0002) (by CAVA), 0.877 (P<0.0001) (by ABC/2) to 0.882 (P<0.0001) (by CAVA), and 0.912 (P<0.0001), respectively.

CONCLUSION

Our study shows that hematoma shape, hematoma size, ICH scores and GCS score all significantly predict the 30-day mortality in an increasing order of AUC. The effect of overestimation of hematoma size by ABC/2 formula in predicting the 30-day mortality could be remedied by using ICH score.

Predicting the probability of mortality of gastric cancer patients using decision tree

BACKGROUND

Gastric cancer is the fourth most common cancer worldwide. This reason motivated us to investigate and introduce gastric cancer risk factors utilizing statistical methods.

AIM

The aim of this study was to identify the most important factors influencing the mortality of patients who suffer from gastric cancer disease and to introduce a classification approach according to decision tree model for predicting the probability of mortality from this disease.

METHODS

Data on 216 patients with gastric cancer, who were registered in Taleghani hospital in Tehran,Iran, were analyzed. At first, patients were divided into two groups: the dead and alive. Then, to fit decision tree model to our data, we randomly selected 20% of dataset to the test sample and remaining dataset considered as the training sample. Finally, the validity of the model examined with sensitivity, specificity, diagnosis accuracy and the area under the receiver operating characteristic curve. The CART version 6.0 and SPSS version 19.0 softwares were used for the analysis of the data.

RESULTS

Diabetes, ethnicity, tobacco, tumor size, surgery, pathologic stage, age at diagnosis, exposure to chemical weapons and alcohol consumption were determined as effective factors on mortality of gastric cancer. The sensitivity, specificity and accuracy of decision tree were 0.72, 0.75 and 0.74 respectively.

CONCLUSIONS

The indices of sensitivity, specificity and accuracy represented that the decision tree model has acceptable accuracy to prediction the probability of mortality in gastric cancer patients. So a simple decision tree consisted of factors affecting on mortality of gastric cancer may help clinicians as a reliable and practical tool to predict the probability of mortality in these patients.

Management of non-traumatic intraventricular hemorrhage

Intraventricular hemorrhage (IVH) is defined as the eruption of blood in the cerebral ventricular system and is mostly secondary to spontaneous intracerebral hemorrhage and aneurysmal and arteriovenous malformation rupture. IVH is a proven risk factor of increased mortality and poor functional outcome. Its seriousness is correlated not only with the amount of blood but also with the involvement of the third and fourth ventricles. There are four mechanisms that explain the pathophysiology of this event: acute obstructive hydrocephalus, the mass effect exerted by the blood clot, the toxicity of blood-breaking products on the adjacent brain parenchyma, and, lastly, the development of a chronic hydrocephalus. It is thus obvious that the clearance of blood from the ventricles should be a therapeutic goal. In cases of acute hydrocephalus, external ventricular drainage is a mandatory step, but proven often insufficient. The concomitant use of intraventricular fibrinolytics such as recombinant tissue plasminogen activator or urokinase seems to be beneficial at least in the context of spontaneous intracerebral hemorrhage, in which their use is now accepted but not yet validated by a randomized trial. Given the potential neurotoxicity of these agents, further research is needed in order to identify the best treatment for intraventricular fibrinolysis (IVF). The endoscopic retrieval of intraventricular blood was also described recently and seems to be as efficient as IVF, but its use is limited to specialized centers. IVH represents a therapeutic challenge for neurosurgeons, neurologists, and intensivists. Thus, a better understanding of this dramatic event will help in better tailoring the treatment strategies.

Spontaneous intracerebral hemorrhage: Clinical and computed tomography findings in predicting in-hospital mortality in Central Africans

BACKGROUND AND PURPOSE

Intracerebral hemorrhage (ICH) constitutes now 52% of all strokes. Despite of its deadly pattern, locally there is no clinical grading scale for ICH-related mortality prediction. The first objective of this study was to develop a risk stratification scale (Kinshasa ICH score) by assessing the strength of independent predictors and their association with in-hospital 30-day mortality. The second objective of the study was to create a specific local and African model for ICH prognosis.

MATERIALS AND METHODS

Age, sex, hypertension, type 2 diabetes mellitus (T2DM), smoking, alcohol intake, and neuroimaging data from CT scan (ICH volume, Midline shift) of patients admitted with primary ICH and follow-upped in 33 hospitals of Kinshasa, DR Congo, from 2005 to 2008, were analyzed using logistic regression models.

RESULTS

A total of 185 adults and known hypertensive patients (140 men and 45 women) were examined. 30-day mortality rate was 35% (n=65). ICH volume>25 mL (OR=8 95% CI: 3.1-20.2; P<0.0001), presence of coma (OR=6.8 95% CI 2.6-17.4; P<0.0001) and left hemispheric site of ICH (OR 2.6 95% CI: 1.1-6; P=0.027) were identified as significant and independent predictors of 30-day mortality. Midline shift > 7 mm, a consequence of ICH volume, was also a significant predictor of mortality. The Kinshasa ICH score was the sum of individual points assigned as follows: Presence of coma coded 2 (2 × 2 = 4), absence of coma coded 1 (1 × 2 = 2), ICH volume>25 mL coded 2 (2 × 2=4), ICH volume of ≤25 mL coded 1(1 × 2=2), left hemispheric site of ICH coded 2 (2 × 1=2), and right hemispheric site of hemorrhage coded 1(1 × 1 = 1). All patients with Kinshasa ICH score ≤7 survived and the patients with a score >7 died. In considering sex influence (Model 3), points were allowed as follows: Presence of coma (2 × 3 = 6), absence of coma (1 × 3 = 3), men (2 × 2 = 4), women (1 × 2 = 2), midline shift ≤7 mm (1 × 3 = 3), and midline shift >7 mm (2 × 3 = 6). Patients who died had the Kinshasa ICH score ≥16.

CONCLUSION

In this study, the Kinshasa ICH score seems to be an accurate method for distinguishing those ICH patients who need continuous and special management. It needs to be validated among large African hypertensive populations with a high rate of 30-day in-hospital mortality.

A simple risk score accurately predicts in-hospital mortality, length of stay, and cost in acute upper GI bleeding

BACKGROUND

Although the early use of a risk stratification score in upper GI bleeding is recommended, existing risk scores are not widely used in clinical practice.

OBJECTIVE

We sought to develop and validate an easily calculated bedside risk score, AIMS65, by using data routinely available at initial evaluation.

DESIGN

Data from patients admitted from the emergency department with acute upper GI bleeding were extracted from a database containing information from 187 U.S. hospitals. Recursive partitioning was applied to derive a risk score for in-hospital mortality by using data from 2004 to 2005 in 29,222 patients. The score was validated by using data from 2006 to 2007 in 32,504 patients. Accuracy to predict mortality was assessed by the area under the receiver operating characteristic (AUROC) curve.

MAIN OUTCOME MEASUREMENTS

Mortality, length of stay (LOS), and cost of admission.

RESULTS

The 5 factors present at admission with the best discrimination were albumin less than 3.0 g/dL, international normalized ratio greater than 1.5, altered mental status, systolic blood pressure 90 mm Hg or lower, and age older than 65 years. For those with no risk factors, the mortality rate was 0.3% compared with 31.8% in patients with all 5 (P < .001). The model had a high predictive accuracy (AUROC = 0.80; 95% CI, 0.78-0.81), which was confirmed in the validation cohort (AUROC = 0.77, 95% CI, 0.75-0.79). Longer LOS and increased costs were seen with higher scores (P < .001).

LIMITATIONS

Database data used does not include outcomes such as rebleeding.

CONCLUSIONS

AIMS65 is a simple, accurate risk score that predicts in-hospital mortality, LOS, and cost in patients with acute upper GI bleeding.

Prognostic Value of Intraventricular Bleeding in Spontaneous Intraparenchymal Cerebral Hemorrhage of Small Volume: A Prospective Cohort Study

BACKGROUND:

The literature is controversial on whether intraventricular bleeding has a negative impact on the prognosis of spontaneous intracerebral hemorrhage. Nevertheless, an association between intraventricular bleeding and spontaneous intracerebral hemorrhage volumes has been consistently reported.

OBJECTIVE:

To evaluate the prognostic value of intraventricular bleeding in deep intraparenchymal hypertensive spontaneous hemorrhage with a bleeding volume &lt;30 cm3.

METHODS:

Of the 320 patients initially evaluated, 33 met the inclusion criteria and were enrolled in this prospective study. The volume of intraparenchymal hemorrhage was calculated by brain computed tomography (CT) image analysis, and the volume of intraventricular bleeding was calculated by the LeRoux scale. Clinical data, including neurological complications, were collected daily during hospitalization. Neurological outcome was evaluated 30 days after the event by using the Glasgow outcome scale. Patients were assigned to 1 of 3 groups according to intraventricular bleeding: Control, no intraventricular bleeding; LR 1, intraventricular bleeding with LeRoux scale scores of 1 to 8; or LR 2, intraventricular bleeding with LeRoux scale scores &gt;8.

RESULTS:

There were no significant differences among groups concerning age, mean blood pressure, and time from onset to brain CT scan. Patients with greater intraventricular bleeding presented lower initial Glasgow coma scale scores, increased ventricular index and width of temporal horns, increased number of clinical and neurological complications, and longer hospitalization. Furthermore, their relative risk for unfavorable clinical outcome was 1.9 (95% confidence interval 1.25-2.49).

CONCLUSION:

Intraventricular bleeding with a LeRoux scale score &gt;8 appears to have a negative effect on deep spontaneous intraparenchymal cerebral hemorrhage of small volume.

Risk stratification for hospitalization in acute asthma: the CHOP classification tree

OBJECTIVE

Simple risk stratification rules are limited in acute asthma. We developed and externally validated a classification tree for asthma hospitalization.

METHODS

Data were obtained from 2 large, multicenter studies on acute asthma, the National Emergency Department Safety Study and the Multicenter Airway Research Collaboration cohorts. Both studies involved emergency department (ED) patients aged 18 to 54 years presenting to the ED with acute asthma. Clinical information was obtained from medical record review. The Classification and Regression Tree method was used to generate a simple decision tree. The tree was derived in the National Emergency Department Safety Study cohort and then was validated in the Multicenter Airway Research Collaboration cohort.

RESULTS

There were 1825 patients in the derivation cohort and 1335 in the validation cohort. Admission rates were 18% and 21% in the derivation and validation cohorts, respectively. The Classification and Regression Tree method identified 4 important variables (CHOP): change [C] in peak expiratory flow severity category, ever hospitalization [H] for asthma, oxygen [O] saturation on room air, and initial peak expiratory flow [P]. In a simple 3-step process, the decision rule risk-stratified patients into 7 groups, with a risk of admission ranging from 9% to 48%. The classification tree performed satisfactorily on discrimination in both the derivation and validation cohorts, with an area under the receiver operating characteristic curve of 0.72 and 0.65, respectively.

CONCLUSIONS

We developed and externally validated a novel classification tree for hospitalization among ED patients with acute asthma. Use of this explicit risk stratification rule may aid decision making in the emergency care of acute asthma.

Blood pressure management in acute intracerebral haemorrhage guidelines are poorly implemented in clinical practice

BACKGROUND

Optimal management of blood pressure (BP) in spontaneous intracerebral haemorrhage (ICH) is controversial. We assessed adherence to BP guidelines and its management in ICH in a tertiary Canadian Stroke Centre.

METHODS

We conducted a retrospective analysis of 142 CT confirmed primary ICH patients admitted within 24h of symptoms between 2005 and 2006. Initial practice with respect to BP control was reviewed and compared with current guidelines. This retrospective sample was compared with a prospective cohort participating in a BP lowering trial for the attainment of pre-defined BP targets. We also assessed the effect of BP treatment on hematoma expansion and mortality.

RESULTS

Blood pressure treatment orders were established in 73% of the 142 patients (median age 71 years, 61% male). Only 26% of patients had target orders as advised in the current AHA guidelines. Only 54% achieved BP targets as compared with 83% of the prospective cohort within 1h. Patients with established BP orders were more likely to have repeat brain imaging (70.2%) than those without (39.5%; p=0.001 Mortality rates were 29.8% and 47.4% in those with and without BP targets respectively (p=0.051).

CONCLUSIONS

Management of BP varies considerably and there appears to be little adherence to recommended guidelines. Targets are achieved more rapidly if a BP treatment protocol is utilized.

Symptomatologic versus neuroimaging predictors of in-hospital survival after intracerebral haemorrhage

Symptomatological prediction of Intracerebral haemorrhage (ICH) mortality is a simple and effective method compared to pathological predictors. In this study we considered consciousness level as an easily measurable predictor and compared it to haemorrhage location, intraventricular penetration and haemorrhage size derived from Computerized Tomography (CT) to predict mortality using a parametric survival analysis model. Two hundred and thirty eight ICH patients from a neurology hospital ward were enrolled into this comparative study. Patient history was documented with respect to mortality and a questionnaire outlining background variables and medical history was completed for them. Consciousness level was clinically evaluated by a physician while haemorrhage size and location were determined via computerized tomographic scanning reports. Data were entered into the computer and analyzed according to the Weibull parametric survival analysis model using STATA 8 statistical software. Males constituted 47.1% of the 238 patients, 52.9% were females. The age range of the patients varied from 13 to 88 years, with a mean age of 62.4 +/- 13.6 (Mean +/- SD). Half of the patients survived more than 20 days. Using the Weibull regression model, the only significant independent symptomatological predictor of mortality was found to be the level of consciousness. Cumulative hazard during the 90 days was compared for different levels of consciousness. Application of Weibull to pathological predictors of ICH mortality showed that the two independent predictors were haemorrhage size and intraventricular penetration. Results of statistical modelling didn't provide evidence of priority for pathological predictors of survival compared to easily measurable levels of consciousness as a symptomatological predictor. Easily measurable symptoms of level of consciousness can be used as a survival predictor of stroke due to intra-cerebral haemorrhage when compared to pathological indicators.

Random forest can predict 30-day mortality of spontaneous intracerebral hemorrhage with remarkable discrimination

BACKGROUND AND PURPOSE

Risk-stratification models based on patient and disease characteristics are useful for aiding clinical decisions and for comparing the quality of care between different physicians or hospitals. In addition, prediction of mortality is beneficial for optimizing resource utilization. We evaluated the accuracy and discriminating power of the random forest (RF) to predict 30-day mortality of spontaneous intracerebral hemorrhage (SICH).

METHODS

We retrospectively studied 423 patients admitted to the Taichung Veterans General Hospital who were diagnosed with spontaneous SICH within 24 h of stroke onset. The initial evaluation data of the patients were used to train the RF model. Areas under the receiver operating characteristic curves (AUC) were used to quantify the predictive performance. The performance of the RF model was compared to that of an artificial neural network (ANN), support vector machine (SVM), logistic regression model, and the ICH score.

RESULTS

The RF had an overall accuracy of 78.5% for predicting the mortality of patients with SICH. The sensitivity was 79.0%, and the specificity was 78.4%. The AUCs were as follows: RF, 0.87 (0.84-0.90); ANN, 0.81 (0.77-0.85); SVM, 0.79 (0.75-0.83); logistic regression, 0.78 (0.74-0.82); and ICH score, 0.72 (0.68-0.76). The discriminatory power of RF was superior to that of the other prediction models.

CONCLUSIONS

The RF provided the best predictive performance amongst all of the tested models. We believe that the RF is a suitable tool for clinicians to use in predicting the 30-day mortality of patients after SICH.

Stratification of the severity of critically ill patients with classification trees

Background

Development of three classification trees (CT) based on the CART (Classification and Regression Trees), CHAID (Chi-Square Automatic Interaction Detection) and C4.5 methodologies for the calculation of probability of hospital mortality; the comparison of the results with the APACHE II, SAPS II and MPM II-24 scores, and with a model based on multiple logistic regression (LR).

Methods

Retrospective study of 2864 patients. Random partition (70:30) into a Development Set (DS) n = 1808 and Validation Set (VS) n = 808. Their properties of discrimination are compared with the ROC curve (AUC CI 95%), Percent of correct classification (PCC CI 95%); and the calibration with the Calibration Curve and the Standardized Mortality Ratio (SMR CI 95%).

Results

CTs are produced with a different selection of variables and decision rules: CART (5 variables and 8 decision rules), CHAID (7 variables and 15 rules) and C4.5 (6 variables and 10 rules). The common variables were: inotropic therapy, Glasgow, age, (A-a)O2 gradient and antecedent of chronic illness. In VS: all the models achieved acceptable discrimination with AUC above 0.7. CT: CART (0.75(0.71-0.81)), CHAID (0.76(0.72-0.79)) and C4.5 (0.76(0.73-0.80)). PCC: CART (72(69-75)), CHAID (72(69-75)) and C4.5 (76(73-79)). Calibration (SMR) better in the CT: CART (1.04(0.95-1.31)), CHAID (1.06(0.97-1.15) and C4.5 (1.08(0.98-1.16)).

Conclusion

With different methodologies of CTs, trees are generated with different selection of variables and decision rules. The CTs are easy to interpret, and they stratify the risk of hospital mortality. The CTs should be taken into account for the classification of the prognosis of critically ill patients.

Candidemia on presentation to the hospital: development and validation of a risk score

INTRODUCTION

Candidemia results in substantial morbidity and mortality, especially if initial antifungal therapy is delayed or is inappropriate; however, candidemia is difficult to diagnose because of its nonspecific presentation.

METHODS

To develop a risk score for identifying hospitalized patients with candidemia, we performed a retrospective analysis of a large database of 176 acute-care hospitals in the United States. We studied 64,019 patients with bloodstream infection (BSI) on presentation from 2000 through 2005 (derivation cohort) and 24,685 from 2006 to 2007 (validation cohort). We used recursive partitioning (RPART) to identify the best discriminators for Candida as the cause of BSI. We compared three sets of models (equal-weight, unequal-weight, vs full model with additional variables from logistic regression model) for sensitivity analysis.

RESULTS

The RPART identified 6 variables as the best discriminators: age < 65 years, temperature <or= 98 degrees F or severe altered mental status, cachexia, previous hospitalization within 30 days, admitted from other healthcare facility, and need for mechanical ventilation. The prevalence for patients presented with 0 through 6 risk factors in the derivation cohort was 28.7%, 38.8%, 21.8%, 8.3%, 2.1%, 0.3%, and < 0.1% respectively. The corresponding candidemia rates were 0.4% (69/18,355), 0.8% (196/24,811), 1.6% (229/13,984), 3.2% (168/5,330), 4.2% (58/1,371), 9.6% (15/157), and 27.3% (3/11) respectively (P < 0.0001). Findings were similar in the validation cohort (P < 0.0001). The equal-weight risk score model, which signed 1 point to each risk factor, yielded good discrimination in both cohorts with areas under the receiver operating curve (AUROCs) of 0.70 versus 0.71 (derivation versus validation). AUROC values were similar for the unequal-weight model, which signed different weight to each risk factor based on multivariable logistic regression coefficient, (AUROCs, 0.70-0.72). Both equal-weight and unequal-weight models were well calibrated (all Hosmer-Lemshow P > 0.10, indicating predicted and observed candidemia rates did not differ significant across the 7 risk stratus). The full model with 16 risk factors had slightly higher AUROCs (0.74 versus 0.73 for derivation versus validation); however, 7 variables were no longer significant in the recalibrated model for the validation cohort, indicating that the additional items did not materially enhance the model.

CONCLUSIONS

A simple equal-weight risk score differentiated patients' risk for candidemia in a graded fashion upon hospital presentation.

Development and validation of a risk score for post-infectious irritable bowel syndrome

OBJECTIVES

Acute gastroenteritis (GE) is an important risk factor for the development of irritable bowel syndrome (IBS). We used observational data from the Walkerton Health Study (WHS) to develop and validate a risk score for post-infectious (PI) IBS.

METHODS

Model derivation and validation were based on a split-sample method from a cohort of patients with exposure to GE (n=1,368). Study participants were randomly assigned to the derivation and validation cohorts in a 1:1 ratio. Within the derivation cohort, univariate and multivariable logistic regression were used to identify risk factors associated with IBS. The risk model was then applied to the validation cohort. Overall model performance was assessed using the area under the receiver operating curve (ROC). The risk score was developed using multivariable regression coefficients obtained from the derivation set and validated in the validation set. Classification and regression tree (CART) modeling was used to determine cutoff values for high, intermediate, and low risk based on the total score.

RESULTS

Nine variables were identified as important predictors of IBS (gender, age<60, longer duration of diarrhea, increased stool frequency, abdominal cramping, bloody stools, weight loss, fever, and psychological disorders (anxiety and depression)). The discriminatory power of the risk model based on the area under ROC was 0.70 and was similar in the validation set. The risk score model showed good accuracy in both the derivation and validation sets and was able to distinguish among cohorts at low, intermediate, and high risk for developing PI-IBS. Percentages of patients with PI-IBS in the low, intermediate and high risk were 10, 35, and 60% in the derivation cohort and 17, 36, and 62% in the validation cohort.

CONCLUSIONS

A simple risk tool that uses demographics and symptoms of acute GE can predict which patients with acute GE are at risk of developing PI-IBS. This tool may be used clinically to assess risk and to guide treatment.

Predicting the outcome of patients with subarachnoid hemorrhage using machine learning techniques

BACKGROUND

Outcome prediction for subarachnoid hemorrhage (SAH) helps guide care and compare global management strategies. Logistic regression models for outcome prediction may be cumbersome to apply in clinical practice.

OBJECTIVE

To use machine learning techniques to build a model of outcome prediction that makes the knowledge discovered from the data explicit and communicable to domain experts.

MATERIAL AND METHODS

A derivation cohort (n = 441) of nonselected SAH cases was analyzed using different classification algorithms to generate decision trees and decision rules. Algorithms used were C4.5, fast decision tree learner, partial decision trees, repeated incremental pruning to produce error reduction, nearest neighbor with generalization, and ripple down rule learner. Outcome was dichotomized in favorable [Glasgow outcome scale (GOS) = I-II] and poor (GOS = III-V). An independent cohort (n = 193) was used for validation. An exploratory questionnaire was given to potential users (specialist doctors) to gather their opinion on the classifier and its usability in clinical routine.

RESULTS

The best classifier was obtained with the C4.5 algorithm. It uses only two attributes [World Federation of Neurological Surgeons (WFNS) and Fisher's scale] and leads to a simple decision tree. The accuracy of the classifier [area under the ROC curve (AUC) = 0.84; confidence interval (CI) = 0.80-0.88] is similar to that obtained by a logistic regression model (AUC = 0.86; CI = 0.83-0.89) derived from the same data and is considered better fit for clinical use.