Comparison of the prediction of extremely low birth weight neonatal mortality by regression analysis and by neural networks

Machine learning risk stratification for high-risk infant follow-up of term and late preterm infants

BACKGROUND

Term and late preterm infants are not routinely referred to high-risk infant follow-up programs at neonatal intensive care unit (NICU) discharge. We aimed to identify NICU factors associated with abnormal developmental screening and develop a risk-stratification model using machine learning for high-risk infant follow-up enrollment.

METHODS

We performed a retrospective cohort study identifying abnormal developmental screening prior to 6 years of age in infants born ≥34 weeks gestation admitted to a level IV NICU. Five machine learning models using NICU predictors were developed by classification and regression tree (CART), random forest, gradient boosting TreeNet, multivariate adaptive regression splines (MARS), and regularized logistic regression analysis. Performance metrics included sensitivity, specificity, accuracy, precision, and area under the receiver operating curve (AUC).

RESULTS

Within this cohort, 87% (1183/1355) received developmental screening, and 47% had abnormal results. Common NICU predictors across all models were oral (PO) feeding, follow-up appointments, and medications prescribed at NICU discharge. Each model resulted in an AUC > 0.7, specificity >70%, and sensitivity >60%.

CONCLUSION

Stratification of developmental risk in term and late preterm infants is possible utilizing machine learning. Applying machine learning algorithms allows for targeted expansion of high-risk infant follow-up criteria.

IMPACT

This study addresses the gap in knowledge of developmental outcomes of infants ≥34 weeks gestation requiring neonatal intensive care. Machine learning methodology can be used to stratify early childhood developmental risk for these term and late preterm infants. Applying the classification and regression tree (CART) algorithm described in the study allows for targeted expansion of high-risk infant follow-up enrollment to include those term and late preterm infants who may benefit most.

Status of cognitive frailty in elderly patients with chronic kidney disease and construction of a risk prediction model: a cross-sectional study

OBJECTIVE

To investigate the risk factors of cognitive frailty in elderly patients with chronic kidney disease (CKD), and to establish an artificial neural network (ANN) model.

DESIGN

A cross-sectional design.

SETTING

Two tertiary hospitals in southern China.

PARTICIPANTS

425 elderly patients aged ≥60 years with CKD.

METHODS

Data were collected via questionnaire investigation, anthropometric measurements, laboratory tests and electronic medical records. The 425 samples were randomly divided into a training set, test set and validation set at a ratio of 5:3:2. Variables were screened by univariate and multivariate logistic regression analyses, then an ANN model was constructed. The accuracy, specificity, sensitivity, receiver operating characteristic (ROC) curve and area under the ROC curve (AUC) were used to evaluate the predictive power of the model.

RESULTS

Barthel Index (BI) score, albumin, education level, 15-item Geriatric Depression Scale score and Social Support Rating Scale score were the factors influencing the occurrence of cognitive frailty (p<0.05). Among them, BI score was the most important factor determining cognitive frailty, with an importance index of 0.30. The accuracy, specificity and sensitivity of the ANN model were 86.36%, 88.61% and 80.65%, respectively, and the AUC of the constructed ANN model was 0.913.

CONCLUSION

The ANN model constructed in this study has good predictive ability, and can provide a reference tool for clinical nursing staff in the early prediction of cognitive frailty in a high-risk population.

Machine Learning Models for Predicting Mortality in 7472 Very Low Birth Weight Infants Using Data from a Nationwide Neonatal Network

Statistical and analytical methods using artificial intelligence approaches such as machine learning (ML) are increasingly being applied to the field of pediatrics, particularly to neonatology. This study compared the representative ML analysis and the logistic regression (LR), which is a traditional statistical analysis method, using them to predict mortality of very low birth weight infants (VLBWI). We included 7472 VLBWI data from a nationwide Korean neonatal network. Eleven predictor variables (neonatal factors: male sex, gestational age, 5 min Apgar scores, body temperature, and resuscitation at birth; maternal factors: diabetes mellitus, hypertension, chorioamnionitis, premature rupture of membranes, antenatal steroid, and cesarean delivery) were selected based on clinical impact and statistical analysis. We compared the predicted mortality between ML methods—such as artificial neural network (ANN), random forest (RF), and support vector machine (SVM)—and LR with a randomly selected training set (80%) and a test set (20%). The model performances of area under the receiver operating curve (95% confidence interval) equaled LR 0.841 (0.811−0.872), ANN 0.845 (0.815−0.875), and RF 0.826 (0.795−0.858). The exception was SVM 0.631 (0.578−0.683). No statistically significant differences were observed between the performance of LR, ANN, and RF (i.e., p > 0.05). However, the SVM model was lower (p < 0.01). We suggest that VLBWI mortality prediction using ML methods would yield the same prediction rate as the traditional statistical LR method and may be suitable for predicting mortality. However, low prediction rates are observed in certain ML methods; hence, further research is needed on these limitations and selecting an appropriate method.

Prognostic Models Predicting Mortality in Preterm Infants: Systematic Review and Meta-analysis

CONTEXT

Prediction models can be a valuable tool in performing risk assessment of mortality in preterm infants.

OBJECTIVE

Summarizing prognostic models for predicting mortality in very preterm infants and assessing their quality.

DATA SOURCES

Medline was searched for all articles (up to June 2020).

STUDY SELECTION

All developed or externally validated prognostic models for mortality prediction in liveborn infants born <32 weeks' gestation and/or <1500 g birth weight were included.

DATA EXTRACTION

Data were extracted by 2 independent authors. Risk of bias (ROB) and applicability assessment was performed by 2 independent authors using Prediction model Risk of Bias Assessment Tool.

RESULTS

One hundred forty-two models from 35 studies reporting on model development and 112 models from 33 studies reporting on external validation were included. ROB assessment revealed high ROB in the majority of the models, most often because of inadequate (reporting of) analysis. Internal and external validation was lacking in 41% and 96% of these models. Meta-analyses revealed an average C-statistic of 0.88 (95% confidence interval [CI]: 0.83-0.91) for the Clinical Risk Index for Babies score, 0.87 (95% CI: 0.81-0.92) for the Clinical Risk Index for Babies II score, and 0.86 (95% CI: 0.78-0.92) for the Score for Neonatal Acute Physiology Perinatal Extension II score.

LIMITATIONS

Occasionally, an external validation study was included, but not the development study, because studies developed in the presurfactant era or general NICU population were excluded.

CONCLUSIONS

Instead of developing additional mortality prediction models for preterm infants, the emphasis should be shifted toward external validation and consecutive adaption of the existing prediction models.

Machine Learning Models for Predicting Neonatal Mortality: A Systematic Review

INTRODUCTION

Approximately 7,000 newborns die every day, accounting for almost half of child deaths under 5 years of age. Deciphering which neonates are at increased risk for mortality can have an important global impact. As such, integrating high computational technology (e.g., artificial intelligence [AI]) may help identify the early and potentially modifiable predictors of neonatal mortality. Therefore, the objective of this study was to collate, critically appraise, and analyze neonatal prediction studies that included AI.

METHODS

A literature search was performed in PubMed, Cochrane, OVID, and Google Scholar. We included studies that used AI (e.g., machine learning (ML) and deep learning) to formulate prediction models for neonatal death. We excluded small studies (n < 500 individuals) and studies using only antenatal factors to predict mortality. Two independent investigators screened all articles for inclusion. The data collection consisted of study design, number of models, features used per model, feature importance, internal and/or external validation, and calibration analysis. Our primary outcome was the average area under the receiving characteristic curve (AUC) or sensitivity and specificity for all models included in each study.

RESULTS

Of 434 articles, 11 studies were included. The total number of participants was 1.26 M with gestational ages ranging from 22 weeks to term. Number of features ranged from 3 to 66 with timing of prediction as early as 5 min of life to a maximum of 7 days of age. The average number of models per study was 4, with neural network, random forest, and logistic regression comprising the most used models (58.3%). Five studies (45.5%) reported calibration plots and 2 (18.2%) conducted external validation. Eight studies reported results by AUC and 5 studies reported the sensitivity and specificity. The AUC varied from 58.3% to 97.0%. The mean sensitivities ranged from 63% to 80% and specificities from 78% to 99%. The best overall model was linear discriminant analysis, but it also had a high number of features (n = 17).

DISCUSSION/CONCLUSION

ML models can accurately predict death in neonates. This analysis demonstrates the most commonly used predictors and metrics for AI prediction models for neonatal mortality. Future studies should focus on external validation, calibration, as well as deployment of applications that can be readily accessible to health-care providers.

Comparative Study of Back Propagation Artificial Neural Networks and Logistic Regression Model in Predicting Poor Prognosis after Acute Ischemic Stroke

Objective

To investigate the predictive value of clinical variables on the poor prognosis at 90-day follow-up from acute stroke onset, and compare the diagnostic performance between back propagation artificial neural networks (BP ANNs) and Logistic regression (LR) models in predicting the prognosis.

Methods

We studied the association between clinical variables and the functional recovery of 435 acute ischemic stroke patients. The patients were divided into 2 groups according to modified Rankin Scale scores evaluated on the 90th day after stroke onset. Both BP ANNs and LR models were established for predicting the poor outcome and their diagnostic performance were compared by receiver operating curve.

Results

Age, free fatty acid, homocysteine and alkaline phosphatase were closely related with the poor outcome in acute ischemic stroke patients and finally enrolled in models. The accuracy, sensitivity and specificity of BP ANNs were 80.15%, 75.64% and 82.07% respectively. For the LR model, the accuracy, sensitivity and specificity was 70.61%, 88.46% and 63.04% respectively. The area under the ROC curve of the BP ANNs and LR model was 0.881and 0.809.

Conclusions

Both BP ANNs and LR model were promising for the prediction of poor outcome by combining age, free fatty acid, homocysteine and alkaline phosphatase. However, BP ANNs model showed better performance than LR model in predicting the prognosis.

A machine learning approach to estimating preterm infants survival: development of the Preterm Infants Survival Assessment (PISA) predictor

Estimation of mortality risk of very preterm neonates is carried out in clinical and research settings. We aimed at elaborating a prediction tool using machine learning methods. We developed models on a cohort of 23747 neonates <30 weeks gestational age, or <1501 g birth weight, enrolled in the Italian Neonatal Network in 2008-2014 (development set), using 12 easily collected perinatal variables. We used a cohort from 2015-2016 (N = 5810) as a test set. Among several machine learning methods we chose artificial Neural Networks (NN). The resulting predictor was compared with logistic regression models. In the test cohort, NN had a slightly better discrimination than logistic regression (P < 0.002). The differences were greater in subgroups of neonates (at various gestational age or birth weight intervals, singletons). Using a cutoff of death probability of 0.5, logistic regression misclassified 67/5810 neonates (1.2 percent) more than NN. In conclusion our study - the largest published so far - shows that even in this very simplified scenario, using only limited information available up to 5 minutes after birth, a NN approach had a small but significant advantage over current approaches. The software implementing the predictor is made freely available to the community.

Gestational age and birthweight for risk assessment of neurodevelopmental impairment or death in extremely preterm infants

BACKGROUND

The risk of poor outcomes in preterm infants is primarily determined by birthweight (BW) and gestational age (GA). It is not known whether BW is a better outcome predictor than GA.

OBJECTIVE

To test whether BW is better than GA (measured in days, rather than completed weeks) for prediction of neurodevelopmental impairment (NDI) and death.

DESIGN/METHODS

Extremely preterm infants born at the National Institute of Child Health and Human Development (NICHD) Neonatal Research Network centres between 1998 and 2009 were studied. For the unadjusted analysis, the associations of GA (in days based on best obstetrical estimate) and BW (in grams) with NDI or death were compared using area under the curve (AUC). Adjusted analyses were performed using birth year, sex, race, antenatal steroids, singleton birth, pre-eclampsia, Apgar score at 5 min and small for GA as covariates.

RESULTS

10 652 preterm infants (89%) had outcome data at 18-22 months' corrected age. The mean BW was 678 g (SD: 155) and the mean GA was 173 days (SD: 10) or 24^5/7 weeks (SD: 1^3/7). The AUC for NDI or death was 80% with BW and 79% with GA (p=0.82). Unadjusted and adjusted analyses did not differ. NDI or death rates decreased with increasing GA through 26 weeks (estimated risk reduction with each additional day of gestation: 2.2%).

CONCLUSION

Both BW in grams and GA in days are good predictors of NDI and death in a preterm population selected on the basis of reliable GA.

TRIAL REGISTRATION NUMBER

NCT00009633.

Predicting reintubation, prolonged mechanical ventilation and death in post-coronary artery bypass graft surgery: a comparison between artificial neural networks and logistic regression models

INTRODUCTION

In coronary artery bypass (CABG) surgery, the common complications are the need for reintubation, prolonged mechanical ventilation (PMV) and death. Thus, a reliable model for the prognostic evaluation of those particular outcomes is a worthwhile pursuit. The existence of such a system would lead to better resource planning, cost reductions and an increased ability to guide preventive strategies. The aim of this study was to compare different methods - logistic regression (LR) and artificial neural networks (ANNs) - in accomplishing this goal.

MATERIAL AND METHODS

Subjects undergoing CABG (n = 1315) were divided into training (n = 1053) and validation (n = 262) groups. The set of independent variables consisted of age, gender, weight, height, body mass index, diabetes, creatinine level, cardiopulmonary bypass, presence of preserved ventricular function, moderate and severe ventricular dysfunction and total number of grafts. The PMV was also an input for the prediction of death. The ability of ANN to discriminate outcomes was assessed using receiver-operating characteristic (ROC) analysis and the results were compared using a multivariate LR.

RESULTS

The ROC curve areas for LR and ANN models, respectively, were: for reintubation 0.62 (CI: 0.50-0.75) and 0.65 (CI: 0.53-0.77); for PMV 0.67 (CI: 0.57-0.78) and 0.72 (CI: 0.64-0.81); and for death 0.86 (CI: 0.79-0.93) and 0.85 (CI: 0.80-0.91). No differences were observed between models.

CONCLUSIONS

The ANN has similar discriminating power in predicting reintubation, PMV and death outcomes. Thus, both models may be applicable as a predictor for these outcomes in subjects undergoing CABG.

Comparison of new modeling methods for postnatal weight in ELBW infants using prenatal and postnatal data

OBJECTIVES

Postnatal infant weight curves are used to assess fluid management and evaluate postnatal nutrition and growth. Traditionally, postnatal weight curves are based on birth weight and do not incorporate postnatal clinical information. The aim of the present study was to compare the accuracy of birth weight-based weight curves with weight curves created from individual patient records, including electronic records, using 2 predictive modeling methods, linear regression (LR) and an artificial neural network (NN), which apply mathematical relations between predictor and outcome variables.

METHODS

Perinatal demographic and postnatal nutrition data were collected for extremely-low-birth-weight (ELBW; birth weight <1000 g) infants. Static weight curves were generated using published algorithms. The postnatal predictive models were created using the demographic and nutrition dataset.

RESULTS

Birth weight (861 ± 83 g, mean ± 1 standard deviation [SD]), gestational age (26.2 ± 1.4 weeks), and the first month of nutrition data were collected from individual health records for 92 ELBW infants. The absolute residual (|measured-predicted|) for weight was 84.8 ± 74.4 g for the static weight curves, 60.9 ± 49.1 g for the LR model, and 12.9 ± 9.2 g for the NN model, analysis of variance: both LR and NN P<0.01 versus static curve. NPO (nothing by mouth) infants had greater weight curve discrepancies.

CONCLUSIONS

Compared with birth weight-based and logistic regression-generated weight curves, NN-generated weight curves more closely approximated ELBW infant weight curves, and, using the present electronic health record systems, may produce weight curves better reflective of the patient's status.

Artificial Neural Networks, and Evolutionary Algorithms as a systems biology approach to a data-base on fetal growth restriction

One of the specific aims of systems biology is to model and discover properties of cells, tissues and organisms functioning. A systems biology approach was undertaken to investigate possibly the entire system of intra-uterine growth we had available, to assess the variables of interest, discriminate those which were effectively related with appropriate or restricted intrauterine growth, and achieve an understanding of the systems in these two conditions. The Artificial Adaptive Systems, which include Artificial Neural Networks and Evolutionary Algorithms lead us to the first analyses. These analyses identified the importance of the biochemical variables IL-6, IGF-II and IGFBP-2 protein concentrations in placental lysates, and offered a new insight into placental markers of fetal growth within the IGF and cytokine systems, confirmed they had relationships and offered a critical assessment of studies previously performed.

Predictors of mortality and major morbidities in extremely low birth weight neonates

OBJECTIVE

To determine predictors of mortality and morbidity in extremely low birth weight neonates (ELBW) from a developing country.

STUDY DESIGN

Prospective observational study.

SETTING

Level III neonatal unit in Northern India.

SUBJECTS

Neonates <1000 g born and admitted to intensive care during study period were enrolled. They were analyzed based on survival and development of major morbidity. Multivariable logistic regression model was used to determine independent risk factors.

OUTCOME

Mortality and major morbidity (one or more of the following: Bronchopulmonary dysplasia (BPD), Retinopathy of Prematurity (ROP) requiring laser, grade III or IV intraventricular hemorrhage (IVH), periventricular leukomalacia (PVL) and necrotizing enterocolitis (NEC) stage III) during hospital stay.

RESULTS

Of 255 ELBW neonates born, 149 received optimal care, of which 78 (52%) survived and 57 (39%) developed morbidities. Mean birth weight and gestational age were 29.1±2.6 weeks and 843±108 g. Major causes of mortality were sepsis (46%), birth asphyxia (20%) and pulmonary hemorrhage (19%). Birth weight <800 g [OR (95% CI)-3.51 (1.39-8.89), P=0.008], mechanical ventilation [4.10 (1.64-10.28), P=0.003] and hypotensive shock [10.75 (4.00-28.89), P<0.001] predicted mortality while birth weight <800 g [3.75 (1.47-9.50), P=0.006], lack of antenatal steroids [2.62 (1.00-6.69), P=0.048), asphyxia [4.11 (1.45-11.69), P=0.008], ventilation [4.38 (1.29-14.79), P=0.017] and duration of oxygen therapy [0.004 (1.001-1.006), P=0.002] were the predictors of major morbidities.

CONCLUSIONS

Low birth weight, mechanical ventilation and hypotensive shock predicted mortality in ELBW neonates while low birth weight, lack of antenatal steroids, birth asphyxia, ventilation and duration of oxygen therapy were predictors for major morbidity.

Fatores de risco para mortalidade infantil em municípios do Estado do Paraná, de 1997 a 2008

OBJETIVO: Identificar e analisar os fatores de risco para a mortalidade infantil nos municípios que compõem a 9ª Regional de Saúde do Estado do Paraná, entre os anos de 1997 a 2008. MÉTODOS: Estudo retrospectivo, com dados secundários a partir do Sistema de Informações de Nascidos Vivos (SINASC) e do Sistema de Informação sobre Mortalidade (SIM). Foi considerado variável dependente o óbito antes do primeiro ano de vida e variáveis independentes: sexo, peso ao nascer, duração da gestação, local de ocorrência do óbito, tipo de gravidez, tipo de parto, idade materna e escolaridade da mãe. Os fatores de risco associados ao óbito foram avaliados por meio da análise univariada. RESULTADOS: Entre os anos de 1997 e 2008, foram registrados 92.716 nascimentos pelo SINASC e 1.535 óbitos em crianças menores de um ano pelo SIM. Foram fatores de risco para a mortalidade nascidos vivos do sexo masculino (OR 1,09; IC95% 1,04-1,15), com baixo peso (OR 4,37; IC95% 4,14-4,62), prematuros (OR 4,83; IC95% 4,18-5,58), nascidos vivos de parto vaginal (OR 1,11; IC95% 1,05-1,17), mães adolescentes (OR 1,11; IC95% 1,02-1,22) e com baixa escolaridade (OR 1,97; IC95% 1,84-2,10). CONCLUSÕES: Os dados mostram diminuição da mortalidade infantil e de informações consideradas ignoradas nos bancos de dados e identificam os fatores de risco sugerindo atenção dos profissionais da saúde para o grupo de maior vulnerabilidade desde a assistência no pré-natal.

Prediction of mortality in very premature infants: a systematic review of prediction models

Context

Being born very preterm is associated with elevated risk for neonatal mortality. The aim of this review is to give an overview of prediction models for mortality in very premature infants, assess their quality, identify important predictor variables, and provide recommendations for development of future models.

Methods

Studies were included which reported the predictive performance of a model for mortality in a very preterm or very low birth weight population, and classified as development, validation, or impact studies. For each development study, we recorded the population, variables, aim, predictive performance of the model, and the number of times each model had been validated. Reporting quality criteria and minimum methodological criteria were established and assessed for development studies.

Results

We identified 41 development studies and 18 validation studies. In addition to gestational age and birth weight, eight variables frequently predicted survival: being of average size for gestational age, female gender, non-white ethnicity, absence of serious congenital malformations, use of antenatal steroids, higher 5-minute Apgar score, normal temperature on admission, and better respiratory status. Twelve studies met our methodological criteria, three of which have been externally validated. Low reporting scores were seen in reporting of performance measures, internal and external validation, and handling of missing data.

Conclusions

Multivariate models can predict mortality better than birth weight or gestational age alone in very preterm infants. There are validated prediction models for classification and case-mix adjustment. Additional research is needed in validation and impact studies of existing models, and in prediction of mortality in the clinically important subgroup of infants where age and weight alone give only an equivocal prognosis.

Desenvolvimento motor de crianças prematuras e a termo: uso da Alberta Infant Motor Scale

OBJETIVOS: Comparar o desenvolvimento motor de crianças nascidas pré-termo e a termo aos quatro e seis meses de idade, aplicando a Alberta Infant Motor Scale-AIMS na versão brasileira. MÉTODOS: Estudo longitudinal, comparativo, realizado em Fortaleza-Ceará, entre novembro/ 2009 e maio/2010. amostragem por conveniência, foi constituída por 24 crianças pré-termo e 24 a termo. RESULTADOS: Nas crianças de quatro meses, verificou-se diferença estatisticamente significante na posição em pé (p=0,014) e, nas de seis meses, em todas as posições (prono, supina, sentada, em pé) e escores totais. Quanto ao percentil, aos quatro e seis meses, respectivamente, 37,5% das crianças pré-termo mostraram desempenho excelente e 54,2%, normais. CONCLUSÃO: A análise estatística do desempenho motor grosso entre os grupos de crianças estudadas mostrou diferenças no desenvolvimento e evolução dos percentis da AIMS.

[Assessment of preterm children's early development]

Preterm birth is an event that affects the child's healthy development. Several studies have addressed the evaluation of children born preterm and the influence that multiple risk factors have on the course of their development. This study performed a systematic review of the literature from 2000 to 2005 about the evaluation of the development of children born preterm until the age of 24 months. The biological risk factors were present in every study, with highlights on intraventricular hemorrhage, necrotizing enterocolitis, chronic pulmonary disease, and retardation of intrauterine development as the most studied risks. The child's motor development was the most studied area. In terms of age, the first evaluations focused on the first six months of life. Neonatal risk, low birth weight, baby boys, cerebral injuries, and first-week abnormal spontaneous movements were predicting factors of preterm child development at the age of two years.

Placental determinants of fetal growth: identification of key factors in the insulin-like growth factor and cytokine systems using artificial neural networks

BACKGROUND

Changes and relationships of components of the cytokine and IGF systems have been shown in placenta and cord serum of fetal growth restricted (FGR) compared with normal newborns (AGA). This study aimed to analyse a data set of clinical and biochemical data in FGR and AGA newborns to assess if a mathematical model existed and was capable of identifying these two different conditions in order to identify the variables which had a mathematically consistent biological relevance to fetal growth.

METHODS

Whole villous tissue was collected at birth from FGR (N = 20) and AGA neonates (N = 28). Total RNA was extracted, reverse transcribed and then real-time quantitative (TaqMan) RT-PCR was performed to quantify cDNA for IGF-I, IGF-II, IGFBP-1, IGFBP-2 and IL-6. The corresponding proteins with TNF-alpha in addition were assayed in placental lysates using specific kits. The data were analysed using Artificial Neural Networks (supervised networks), and principal component analysis and connectivity map.

RESULTS

The IGF system and IL-6 allowed to predict FGR in approximately 92% of the cases and AGA in 85% of the cases with a low number of errors. IGF-II, IGFBP-2, and IL-6 content in the placental lysates were the most important factors connected with FGR. The condition of being FGR was connected mainly with the IGF-II placental content, and the latter with IL-6 and IGFBP-2 concentrations in placental lysates.

CONCLUSION

These results suggest that further research in humans should focus on these biochemical data. Furthermore, this study offered a critical revision of previous studies. The understanding of this system biology is relevant to the development of future therapeutical interventions possibly aiming at reducing IL-6 and IGFBP-2 concentrations preserving IGF bioactivity in both placenta and fetus.

Imputation of missing values by integrating neural networks and case-based reasoning

Missing values in a medical database present a problem when trying to develop a prediction model for a broad range of patients, if the data are not missing at random. We present a data imputation approach for physiologic parameters that incorporates individualized case information into the imputed values. We replaced missing values in a neonatal intensive care unit (NICU) database with relevant data by integrating aspects of artificial neural networks (ANNs) and case-based reasoning (CBR).

Predicting clinical outcomes for newborns using two artificial intelligence approaches

Two different approaches, based on artificial neural networks (ANN) and fuzzy logic, were used to predict a number of outcomes of newborns: How they would be delivered, their 5 minute Apgar score, and neonatal mortality. The goal was to assess whether the methods would be comparable or whether they would perform differently for different outcomes. The results were comparable for Correct Classification Rate (CCR) and Specificity (true negative cases). Sensitivity (true positive cases) was slightly higher for the back-propagation feed-forward ANN than using the Fuzzy-Logic Classifier (FLC). Since this is one single database and a very large one, it is possible that the FLC would perform better than the ANN for very small databases, as shown by some of the co-authors in the past. The next step will be to test a small database with both methods to assess strengths and weaknesses with the intent to use both if needed with some medical data in the future.

Predicting high-risk preterm birth using artificial neural networks

A reengineered approach to the early prediction of preterm birth is presented as a complimentary technique to the current procedure of using costly and invasive clinical testing on high-risk maternal populations. Artificial neural networks (ANNs) are employed as a screening tool for preterm birth on a heterogeneous maternal population; risk estimations use obstetrical variables available to physicians before 23 weeks gestation. The objective was to assess if ANNs have a potential use in obstetrical outcome estimations in low-risk maternal populations. The back-propagation feedforward ANN was trained and tested on cases with eight input variables describing the patient's obstetrical history; the output variables were: 1) preterm birth; 2) high-risk preterm birth; and 3) a refined high-risk preterm birth outcome excluding all cases where resuscitation was delivered in the form of free flow oxygen. Artificial training sets were created to increase the distribution of the underrepresented class to 20%. Training on the refined high-risk preterm birth model increased the network's sensitivity to 54.8%, compared to just over 20% for the nonartificially distributed preterm birth model.

Early prediction of poor outcome in extremely low birth weight infants by classification tree analysis

OBJECTIVE

To predict death or neurodevelopmental impairment (NDI) in extremely low birth weight infants by classification trees with recursive partitioning and automatic selection of optimal cut points of variables.

STUDY DESIGN

Data from the Trial of Indomethacin Prophylaxis in Preterms were randomly divided into development (n=784) and validation sets (n=262). Three models were developed for the combined outcome of death (8 days to 18 months) or NDI (cerebral palsy, cognitive delay, deafness, or blindness at 18 months corrected age): antenatal: antenatal data; early neonatal: antenatal+first 3 days data; and first week: antenatal, first 3 days, and 4th to 8th days data. Decision trees were tested on the validation set.

RESULTS

Variables associated with death/NDI in each model were: Antenatal: Gestation<or=25.5 weeks and antenatal steroids<7 days. Early neonatal: Birth weight<or=787 g and fluid intake>01 mL/kg/d. First week: Birth weight<or=787 g: transfusion>3 mL/kg/d. Birth weight>787 g: cranial echodense intraparenchymal lesion and transfusion>1 mL/kg/d. Correct classification rates were 61% to 62% for all models.

CONCLUSIONS

The ability to predict long-term morbidity/death in extremely low birth weight infants does not improve significantly over the first week of life. Effects of different variables depend on age.

Prediction of death for extremely low birth weight neonates

OBJECTIVE

To compare multiple logistic regression and neural network models in predicting death for extremely low birth weight neonates at 5 time points with cumulative data sets, as follows: scenario A, limited prenatal data; scenario B, scenario A plus additional prenatal data; scenario C, scenario B plus data from the first 5 minutes after birth; scenario D, scenario C plus data from the first 24 hours after birth; scenario E, scenario D plus data from the first 1 week after birth.

METHODS

Data for all infants with birth weights of 401 to 1000 g who were born between January 1998 and April 2003 in 19 National Institute of Child Health and Human Development Neonatal Research Network centers were used (n = 8608). Twenty-eight variables were selected for analysis (3 for scenario A, 15 for scenario B, 20 for scenario C, 25 for scenario D, and 28 for scenario E) from those collected routinely. Data sets censored for prior death or missing data were created for each scenario and divided randomly into training (70%) and test (30%) data sets. Logistic regression and neural network models for predicting subsequent death were created with training data sets and evaluated with test data sets. The predictive abilities of the models were evaluated with the area under the curve of the receiver operating characteristic curves.

RESULTS

The data sets for scenarios A, B, and C were similar, and prediction was best with scenario C (area under the curve: 0.85 for regression; 0.84 for neural networks), compared with scenarios A and B. The logistic regression and neural network models performed similarly well for scenarios A, B, D, and E, but the regression model was superior for scenario C.

CONCLUSIONS

Prediction of death is limited even with sophisticated statistical methods such as logistic regression and nonlinear modeling techniques such as neural networks. The difficulty of predicting death should be acknowledged in discussions with families and caregivers about decisions regarding initiation or continuation of care.

Conceptual Framework of Knowledge Management for Ethical Decision-Making Support in Neonatal Intensive Care

This research is built on the belief that artificial intelligence estimations need to be integrated into clinical social context to create value for health-care decisions. In sophisticated neonatal intensive care units (NICUs), decisions to continue or discontinue aggressive treatment are an integral part of clinical practice. High-quality evidence supports clinical decision-making, and a decision-aid tool based on specific outcome information for individual NICU patients will provide significant support for parents and caregivers in making difficult "ethical" treatment decisions. In our approach, information on a newborn patient's likely outcomes is integrated with the physician's interpretation and parents' perspectives into codified knowledge. Context-sensitive content adaptation delivers personalized and customized information to a variety of users, from physicians to parents. The system provides structuralized knowledge translation and exchange between all participants in the decision, facilitating collaborative decision-making that involves parents at every stage on whether to initiate, continue, limit, or terminate intensive care for their infant.

Fatores de risco para mortalidade neonatal em Blumenau, Santa Catarina: linkage entre bancos de dados

OBJETIVOS: investigar os fatores de risco para a mortalidade neonatal no município de Blumenau, Santa Catarina, empregando como possíveis fatores de risco as variáveis apresentadas nas "Declarações de Nascidos Vivos". MÉTODOS: trata-se de um estudo de coorte com apresentação de 3812 nascidos vivos, entre os quais ocorreram 26 óbitos neonatais. Os fatores de risco foram estimados utilizando a regressão logística. RESULTADOS: no modelo final apresentaram associação significativa com o óbito no período neonatal os nascidos com peso menor que 2500 gramas (OR=4,70 IC95%: 1,31-16,87); menor que 36 semanas (OR=4,16; IC95%: 1,22-14,20); Apgar no 5° minuto menor que oito (OR=62,38; IC95%: 22,31-174,39) e presença de anomalia (OR=63,19: IC95%: 15,17-263,15). As variáveis socioeconômicas, médico-assistenciais e de serviços de saúde não apresentaram associação significativa, estando confundidas com os fatores de risco biológicos. CONCLUSÕES: os dados mostram a importância de identificação dos principais fatores de risco de óbito para período neonatal, reafirmando a influência dos fatores biológicos fortemente relacionados ao componente neonatal da mortalidade infantil.

Clinical decision support systems for neonatal care

BACKGROUND

Clinical decision support systems (CDSS) are computer-based information systems used to integrate clinical and patient information to provide support for decision-making in patient care. They may be useful in aiding the diagnostic process, the generation of alerts and reminders, therapy critiquing/planning, information retrieval, and image recognition and interpretation. CDSS for use in adult patients have been evaluated using randomised control trials and their results analysed in systematic reviews. There is as yet no systematic review on CDSS use in neonatal medicine.

OBJECTIVES

To examine whether the use of clinical decision support systems has an effect on 1. the mortality and morbidity of newborn infants and 2. the performance of physicians treating them

SEARCH STRATEGY

The standard search method of the Cochrane Neonatal Review Group was used. Searches were made of the Cochrane Central Register of Controlled Trials (CENTRAL, The Cochrane Library, Issue 1, 2004), MEDLINE (from 1966 to August 2004), EMBASE (1980-2004), CINAHL (1982 to August 2004) and AMED (1985 to August 2004).

SELECTION CRITERIA

Randomised or quasi-randomised controlled trials which compared the effects of CDSS versus no CDSS in the care of newborn infants. Trials which compared CDSS against other CDSS were also considered. The eligible interventions were CDSS for computerised physician order entry, computerised physiological monitoring, diagnostic systems and prognostic systems.

DATA COLLECTION AND ANALYSIS

Studies were assessed for eligibility using a standard pro forma. Methodological quality was assessed independently by the different investigators.

MAIN RESULTS

Two studies fitting the selection criteria were found for computer aided prescribing and one study for computer aided physiological monitoring.Computer-aided prescribing: one study (Cade 1997) examined the effects of computerised prescribing of parenteral nutrition ordering. No significant effects on short-term outcomes were found and longer term outcomes were not studied. The second study (Balaguer 2001) investigated the effects of a database program in aiding the calculation of neonatal drug dosages. It was found that the time taken for calculation was significantly reduced and there was a significant reduction in the number of calculation errors.Computer-aided physiological monitoring: one eligible study (Cunningham 1998) was found which examined the effects of computerised cot side physiological trend monitoring and display. There were no significant effects on mortality, volume of colloid infused, frequency of blood gases sampling (samples per day) or severe (Papile Grade 4) intraventricular haemorrhage. Published data did not permit us to analyse effects on long-term neurodevelopmental outcome.

AUTHORS' CONCLUSIONS

There are very limited data from randomised trials on which to assess the effects of clinical decision support systems in neonatal care. Further evaluation of CDSS using randomised controlled trials is warranted.

Bias in reported neurodevelopmental outcomes among extremely low birth weight survivors

OBJECTIVES

The purpose of this study was to investigate possible bias in the evaluation of neurodevelopment and somatic growth at 18 to 22 months' postmenstrual age among extremely low birth weight (ELBW) survivors (401-1000 g at birth).

METHODS

Data from a cohort of 1483 ELBW infant survivors who were born January 1993 through December 1994 and cared for at centers in the Neonatal Research Network of the National Institute of Child Health and Human Development were examined retrospectively. Children who were compliant with an 18- to 22-month follow-up visit, who visited but were not measured, or who made no visit were compared regarding 4 outcomes: 1) Bayley Scales of Infant Development, 2nd edition, Mental Developmental Index (MDI) <70 and 2) Psychomotor Developmental Index (PDI) <70, 3) presence or absence of cerebral palsy, and 4) weight <10th percentile for age. Logistic regression models were used to predict likelihood of these outcomes for children with no follow-up evaluation, and predicted probability distributions were compared across the groups.

RESULTS

Compared with children who were lost to follow-up, those who were compliant with follow-up were more likely to have been 1 of a multiple birth, to have received postnatal glucocorticoids, and to have had chronic lung disease. These factors were significantly associated with MDI and PDI <70 in the compliant group. Chronic lung disease was associated with increased risk of cerebral palsy (CP). MDI and PDI scores <70 were found in 37% and 29% of children who were evaluated at follow-up, respectively. Prediction models revealed that 34% and 26% of infants in the no-visit group would have had MDI and PDI scores <70. Compliant children tended to have greater incidence of MDI <70 compared with those predicted in the no-visit group but not PDI <70. CP was identified in 17% of the compliant group and predicted for 18% of the no-visit group. Predicted probabilities of having CP were marginally higher among the no-visit infants compared with those who were compliant with follow-up. There were no statistically significant somatic growth differences among the compliant, visit but not measured, and no-visit groups.

CONCLUSION

ELBW infant survivors who weighed 401 to 1000 g at birth and who are compliant with follow-up evaluations may have worse Bayley Scales of Infant Development, 2nd edition, MDI scores than infants with no visit. Thus, follow-up studies based on infants who are compliant with follow-up care may lead to an overestimation of adverse outcomes in ELBW survivors.

Predicting extubation outcome in preterm newborns: a comparison of neural networks with clinical expertise and statistical modeling

Even though ventilator technology and monitoring of premature infants has improved immensely over the past decades, there are still no standards for weaning and determining optimal extubation time for those infants. Approximately 30% of intubated preterm infants will fail attempted extubation, requiring reintubation and resuming of mechanical ventilation. A machine-learning approach using artificial neural networks (ANNs) to aid in extubation decision making is hereby proposed. Using expert opinion, 51 variables were identified as being relevant for the decision of whether to extubate an infant who is on mechanical ventilation. The data on 183 premature infants, born between 1999 and 2002, were collected by review of medical charts. The ANN extubation model was compared with alternative statistical modeling using multivariate logistic regression and also with the clinician's own predictive insight using sensitivity analysis and receiver operating characteristic curves. The optimal ANN model used 13 parameters and achieved an area under the receiver operating characteristic curve of 0.87 (out-of-sample validation), comparing favorably with multivariate logistic regression. It also compared well with the clinician's expertise, which raises the possibility of being useful as an automated alert tool. Because an ANN learns directly from previous data obtained in the institution where it is to be used, this makes it particularly amenable for application to evidence-based medicine. Given the variety of practices and equipment being used in different hospitals, this may be particularly relevant in the context of caring for preterm newborns who are on mechanical ventilation.