Comparison of mortality risk: a score for very low birthweight infants

Hybridized neural networks for non-invasive and continuous mortality risk assessment in neonates

Premature birth is the primary risk factor in neonatal deaths, with the majority of extremely premature babies cared for in neonatal intensive care units (NICUs). Mortality risk prediction in this setting can greatly improve patient outcomes and resource utilization. However, existing schemes often require laborious medical testing and calculation, and are typically only calculated once at admission. In this work, we propose a shallow hybrid neural network for the prediction of mortality risk in 3-day, 7-day, and 14-day risk windows using only birthweight, gestational age, sex, and heart rate (HR) and respiratory rate (RR) information from a 12-h window. As such, this scheme is capable of continuously updating mortality risk assessment, enabling analysis of health trends and responses to treatment. The highest performing scheme was the network that considered mortality risk within 3 days, with this scheme outperforming state-of-the-art works in the literature and achieving an area under the receiver-operator curve (AUROC) of 0.9336 with standard deviation of 0.0337 across 5 folds of cross-validation. As such, we conclude that our proposed scheme could readily be used for continuously-updating mortality risk prediction in NICU environments.

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.

Derivation and validation of a prognostic score for neonatal mortality in Ethiopia: a case-control study

Background

Early warning scores for neonatal mortality have not been designed for low income countries. We developed and validated a score to predict mortality upon admission to a NICU in Ethiopia.

Methods

We conducted a retrospective case-control study at the University of Gondar Hospital, Gondar, Ethiopia. Neonates hospitalized in the NICU between January 1, 2016 to June 31, 2017. Cases were neonates who died and controls were neonates who survived.

Results

Univariate logistic regression identified variables associated with mortality. The final model was developed with stepwise logistic regression. We created the Neonatal Mortality Score, which ranged from 0 to 52, from the model’s coefficients. Bootstrap analysis internally validated the model. The discrimination and calibration were calculated. In the derivation dataset, there were 207 cases and 605 controls. Variables associated with mortality were admission level of consciousness, admission respiratory distress, gestational age, and birthweight. The AUC for neonatal mortality using these variables in aggregate was 0.88 (95% CI 0.85–0.91). The model achieved excellent discrimination (bias-corrected AUC) under internal validation. Using a cut-off of 12, the sensitivity and specificity of the Neonatal Mortality Score was 81 and 80%, respectively. The AUC for the Neonatal Mortality Score was 0.88 (95% CI 0.85–0.91), with similar bias-corrected AUC. In the validation dataset, there were 124 cases and 122 controls, the final model and the Neonatal Mortality Score had similar discrimination and calibration.

Conclusions

We developed, internally validated, and externally validated a score that predicts neonatal mortality upon NICU admission with excellent discrimination and calibration.

Assessment of sickness severity of illness in neonates: review of various neonatal illness scoring systems

Sickness severity scores are widely used for neonates admitted to neonatal intensive care units to predict severity of illness and risk of mortality and long-term outcome. These scores are also used frequently for quality assessment among various neonatal intensive care unit and hospital. Accurate and reliable measures of severity of illness are required for unbiased and reliable comparisons especially for benchmarking or comparative quality improvement care studies. These scores also serve to control for population differences when performing studies such as clinical trials, outcome evaluations, and evaluation of resource utilisation. Although presently there are multiple scores designed for neonates' sickness assessment but none of the score is ideal. Each score has its own advantages and disadvantages. We did literature search for identifying all neonatal sickness severity score and in this review article, we discuss these scores along with their merits and demerits.

Clinical prediction models for bronchopulmonary dysplasia: a systematic review and external validation study

Background

Bronchopulmonary dysplasia (BPD) is a common complication of preterm birth. Very different models using clinical parameters at an early postnatal age to predict BPD have been developed with little extensive quantitative validation. The objective of this study is to review and validate clinical prediction models for BPD.

Methods

We searched the main electronic databases and abstracts from annual meetings. The STROBE instrument was used to assess the methodological quality. External validation of the retrieved models was performed using an individual patient dataset of 3229 patients at risk for BPD. Receiver operating characteristic curves were used to assess discrimination for each model by calculating the area under the curve (AUC). Calibration was assessed for the best discriminating models by visually comparing predicted and observed BPD probabilities.

Results

We identified 26 clinical prediction models for BPD. Although the STROBE instrument judged the quality from moderate to excellent, only four models utilised external validation and none presented calibration of the predictive value. For 19 prediction models with variables matched to our dataset, the AUCs ranged from 0.50 to 0.76 for the outcome BPD. Only two of the five best discriminating models showed good calibration.

Conclusions

External validation demonstrates that, except for two promising models, most existing clinical prediction models are poor to moderate predictors for BPD. To improve the predictive accuracy and identify preterm infants for future intervention studies aiming to reduce the risk of BPD, additional variables are required. Subsequently, that model should be externally validated using a proper impact analysis before its clinical implementation.

The care of preterm infants with birth weight below 1250 g: risk-adjusted quality benchmarking as part of validating a caseload-based management system

BACKGROUND

In Germany, controversy currently surrounds the contention that the quality of care for preterm infants weighing less than 1250 g is best assured by requiring that centers treat a minimum of 30 such cases per year.

METHODS

A risk-adjusted model was developed on the basis of neonatal data from 7405 preterm infants treated in German centers, and the effect of caseload on risk-adjusted mortality was analyzed. In addition, the discriminative ability of the minimal caseload requirement for quality assessment was studied. The authors designate the quality of care in a particular center as above average if the observed mortality is lower than would have been expected from the risk profile of the preterm infants treated there.

RESULTS

Risk-adjusted mortality was found to be significantly higher in smaller centers (those with fewer than 30 cases per year) than in larger ones (odds ratio, 1.34). Even among centers whose caseload exceeded the minimum requirement, there was still marked variability in risk-adjusted mortality (range: 3.5% to 28.6%). Of all the preterm infants treated in larger centers, 56% were treated in centers with above-average quality of care. 44% of the centers with above-average quality of care had caseloads in the range of 14 to 29 cases per year.

CONCLUSION

Because of the marked variability in risk-adjusted mortality, even among larger centers, a caseload of 30 or more cases per year is not a suitable indicator of the quality of care. The neonatal data of external quality assurance should be used to develop an instrument for quality-based coordination of care that takes not just morbidity and mortality, but also the treating centers' competence profiles into account.

Prognostic factors determining mortality in surgical neonates

BACKGROUND

To assess the prognosis of surgical neonates at admission and the factors responsible for mortality in neonates.

MATERIAL AND METHODS

A prospective study was conducted in a tertiary level hospital over 15 months and various clinical and biochemical parameters were collected and analyzed using STATA(®) and SPSS(®).

RESULTS

On multivariate analysis of 165 neonates, early gestational age, respiratory distress and shock at presentation were the factors of poor prognosis in neonates. The factors could be related to poor antenatal care and sepsis acquired before transfer of the baby to the nursery.

CONCLUSION

The improvement in antenatal care and asepsis during transfer and handling the babies is of utmost importance to improve the prognosis of surgical neonates.

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.

Simplified age-weight mortality risk classification for very low birth weight infants in low-resource settings

OBJECTIVE

To identify a valid neonatal mortality risk prediction score feasible for use in developing countries.

STUDY DESIGN

Retrospective study of 467 neonates, < or =1500 g, enrolled in trials during 1998 to 2005 at tertiary care children's hospitals in Dhaka, Bangladesh, and Cairo, Egypt, and a community field site in Sarlahi District, Nepal. We derived simplified mortality risk scores and compared their predictive accuracy with the modified Clinical Risk Index for Babies (CRIB) II. Outcome was death during hospital stay (Dhaka and Cairo) or end of the neonatal period (Nepal).

RESULTS

The area under the curve receiver operating characteristic was 0.62, 0.71, 0.68, and 0.69 on the basis of the (a) CRIB II applied to the Dhaka-Cairo dataset; (b) an 18-category, simplified age, weight, sex score; (c) a binary-risk simplified age-weight (SAW) classification derived from the Dhaka-Cairo dataset; and (d) external validation of the binary-risk SAW classification in the Nepal dataset, respectively. Mortality risk prediction with the SAW classification on the basis of gestational age (< or =29 weeks) or weight (<1000 g) was improved (P = .048) compared with CRIB II.

CONCLUSIONS

The SAW classification is a markedly simplified mortality risk prediction score for use in identifying high-risk, very low birth weight neonates in developing country settings for whom urgent referral is indicated.

Epidemiology of prematurity--how valid are comparisons of neonatal outcomes?

Direct comparisons of neonatal outcomes at any level (unit, regional or international), require detailed validation and standardisation to ensure 'like for like' evaluation. Reported variation in neonatal performance may be either real or the result of one or more artefacts of the data collection. These issues need to be understood in order for an accurate interpretation to be made. Such artefacts are a particular feature of national data collection systems and can lead to serious misinterpretation. For example, very preterm deliveries have a major impact on neonatal mortality rates in developed countries with births before 33 weeks of gestation accounting for between 35% and 70% of neonatal deaths. Variation in the rate of very preterm delivery rates and differing practices regarding registration of these infants can have a major effect on the recorded neonatal mortality rate. At a more local level the validity of neonatal comparisons often depends upon whether the question being raised is appropriately matched to the data obtained to answer it. Problems arise when the question being addressed has been poorly framed or the data used to answer it has been inappropriately chosen. Comparisons using questions based on clearly defined standardised outcome measures and good quality prospective data collection are a much better way to proceed.

A new score for predicting neonatal very low birth weight mortality risk in the NEOCOSUR South American Network

OBJECTIVE

To develop and validate a model for very low birth weight (VLBW) neonatal mortality prediction, based on commonly available data at birth, in 16 neonatal intensive care units (NICUs) from five South American countries.

STUDY DESIGN

Prospectively collected biodemographic data from the Neonatal del Cono Sur (NEOCOSUR) Network between October 2000 and May 2003 in infants with birth weight 500 to 1500 g were employed. A testing sample and crossvalidation techniques were used to validate a statistical model for risk of in-hospital mortality. The new risk score was compared with two existing scores by using area under the receiver operating characteristic curve (AUC).

RESULTS

The new NEOCOSUR score was highly predictive for in-hospital mortality (AUC=0.85) and performed better than the Clinical Risk Index for Babies (CRIB) and the NICHD risk models when used in the NEOCOSUR Network. The new score is also well calibrated - it had good predictive capability for in-hospital mortality at all levels of risk (HL test=11.9, p=0.85). The new score also performed well when used to predict in hospital neurological and respiratory complications.

CONCLUSIONS

A new and relatively simple VLBW mortality risk score had a good prediction performance in a South American network population. This is an important tool for comparison purposes among NICUs. This score may prove to be a better model for application in developing countries.

What is the role of neonatal organ dysfunction and illness severity scores in therapeutic studies in sepsis?

The goal of therapeutic studies in neonatal sepsis is to increase disability-free survival. Mortality may not be an adequate measure of outcome if it is reduced at the cost of increased disability. No neonatal organ dysfunction score has yet been validated as a reliable surrogate for disability-free survival. To validate a score as a surrogate would require randomized trials showing 1) a causal connection between change in the score and change in disability-free survival and 2) that the score fully captured all the effects of treatment on disability-free survival. Neonatal illness severity scores provide a convenient, but imperfect, tool to adjust for risk of sepsis in observational studies. They can help to stratify infants by risk of sepsis at entry to trials, allowing analyses of outcome in predefined subgroups. However, they cannot circumvent the need for randomized trials of adequate size in neonatal sepsis, which address disability-free survival as the primary outcome.

A review of the Apgar score indicated that contextualization was required within the contemporary perinatal and neonatal care framework in different settings

OBJECTIVE

To triangulate the Apgar score by using a crossdisciplinary approach and highlighting the differences that exist between actual everyday practice and accepted standards of scoring in contrasting populations of the world.

STUDY DESIGN AND SETTING

Clinimetrics review of Apgar scoring.

RESULTS

The Apgar scoring has weighting problems, rigid categorization, redundancy and subjectivity in its variables. Poor inter-rater reliability and equivocal validity mark its use in the present milieu. The ceiling and floor effects further hamper the evaluative responsiveness of scoring. Moreover, despite some recent evidence in its favor, the Apgar score has poor calibration when used as an isolated criterion to predict mortality and long-term morbidity, particularly in preterms. Also, the vigor of resuscitation (nature and duration), in essence, is beyond the realm of the Apgar score in contemporary resuscitation guidelines. In developed nations, with rapidly decreasing age of viability, and alternative modes of childbearing, threats to Apgar are more ominous today than before. On the other hand, in developing countries, feasibility problems due to unattended home deliveries and barriers to effective scoring in the overburdened and understaffed hospitals cast doubts about its accuracy as a measure of neonatal well-being.

CONCLUSION

Use of the Apgar score definitely needs to be contextualized within the contemporary perinatal and neonatal care framework in different settings.

Neonatal disease severity scoring systems

Illness severity scores have become widely used in neonatal intensive care. Primarily this has been to adjust the mortality observed in a particular hospital or population for the morbidity of their infants, and hence allow standardised comparisons to be performed. However, although risk correction has become relatively commonplace in relation to audit and research involving groups of infants, the use of such scores in giving prognostic information to parents, about their baby, has been much more limited. The strengths and weaknesses of the existing methods of disease severity correction in the newborn are presented in this review.

The mortality index for neonatal transportation score: a new mortality prediction model for retrieved neonates

OBJECTIVE

To develop a mortality prediction score for retrieved neonates based on the information given at the first telephone contact with a retrieval service.

METHODS

Data from the New South Wales Newborn and Pediatric Emergency Transport Service database were examined. Analysis was performed with the results for 2504 infants (median gestational age: 36 weeks; range: 24-43 weeks) who were <72 hours of age at the time of referral and whose outcome (neonatal death or survival) was known. The study population was divided randomly into 2 halves, the derivation and validation cohorts. Univariate analysis was performed to identify variables in the derivation cohort related to neonatal death. The variables were entered into a multivariate logistic regression analysis with neonatal death as the outcome. Receiver operator characteristic (ROC) curves were constructed with the regression model and data from the derivation cohort and then the validation cohort. The results were used to generate an integer-based score, the Mortality Index for Neonatal Transportation (MINT) score. ROC curves were constructed to assess the ability of the MINT score to predict perinatal and neonatal death.

RESULTS

A 7-variable (Apgar score at 1 minute, birth weight, presence of a congenital anomaly, and infant's age, pH, arterial partial pressure of oxygen, and heart rate at the time of the call) model was constructed that generated areas under ROC curves of 0.82 and 0.83 for the derivation and validation cohorts, respectively. The 7 variables were then used to generate the MINT score, which gave areas under ROC curves of 0.80 for both neonatal and perinatal death.

CONCLUSION

Data collected at the first telephone contact by the referring hospital with a regionalized transport service can identify neonates at the greatest risk of dying.

An illness severity score and neonatal mortality in retrieved neonates

The Clinical Risk Index for Babies (CRIB) score is a simple tool to measure clinical risk and illness severity in very low birth weight infants. The aim of this study was to determine if a modified CRIB score (MCRIB) used at first telephone contact with a transport service differentiated between retrieved infants who did or did not die in the neonatal period and hence might be a useful triage tool. A retrospective cohort study of 2504 infants, median gestational age 36 weeks and birth weight 2782 g, transported by the New South Wales Newborn and Paediatric Emergency Transport Service (NETS) was performed. MCRIB was calculated at four time points during the retrieval process. The MCRIB score at the time of the first call and the change in the MCRIB score over the retrieval process were related to outcome (neonatal death or survival). The mean MCRIB score at the time of first call was higher in those infants who died during the neonatal period (4.37) than in those who survived (2.63), (P < 0.0001). MCRIB performed better (area under the receiver operator characteristic curves of 0.72) with regard to predicting mortality than gestational age (0.56) or birth weight (0.52). The mean MCRIB score fell progressively from the time of first call to admission at the accepting NICU (P < 0.0001); infants whose MCRIB score increased were more likely to die (P < 0.0001).

CONCLUSION

these results suggest an illness severity score, applied at the time of first call to a transport service would be helpful in setting priorities for retrievals.

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

AIMS

To compare the prediction of mortality in individual extremely low birth weight (ELBW) neonates by regression analysis and by artificial neural networks.

STUDY DESIGN

A database of 23 variables on 810 ELBW neonates admitted to a tertiary care center was divided into training, validation, and test sets. Logistic regression and neural network models were developed on the training set, validated, and outcome (mortality) predicted on the test set. Stepwise regression identified significant variables in the full set. Regression models and neural networks were then tested using data sets with only the identified significant variables, and then with variables excluded one at a time.

RESULTS

The area under the curve (AUC) of receiver operating characteristic (ROC) curves for neural networks and regression was similar (AUC 0.87+/-0.03; p=0.31). Birthweight or gestational age and the 5-min Apgar score contributed most to AUC.

CONCLUSIONS

Both neural networks and regression analysis predicted mortality with reasonable accuracy. For both models, analyzing selected variables was superior to full data set analysis. We speculate neural networks may not be superior to regression when no clear non-linear relationships exist.

Pumactant and poractant alfa for treatment of respiratory distress syndrome in neonates born at 25-29 weeks' gestation: a randomised trial

BACKGROUND

Exogenous surfactant preparations vary in their constitution and biophysical properties. Synthetic and animal-derived preparations lower the rate of death compared with controls. No significant differences in mortality or important long-term clinical outcomes have been shown between them in randomised trials. We did a randomised controlled trial to compare pumactant, a synthetic surfactant, with poractant alfa, an animal-derived surfactant, both of which are widely used in the UK.

METHODS

We enrolled 212 neonates born between 25 weeks' and 29 weeks and 6 days' gestation who were intubated for presumed surfactant deficiency and were free from life-threatening malformations. We randomly assigned 105 neonates poractant alfa, and 107 pumactant. The primary outcome was duration of high-dependency care and mortality was a secondary outcome. Analysis was by intention to treat.

FINDINGS

Outcome data were analysed for 199 babies. The trial was stopped on the recommendation of the data and safety monitoring committee because mortality assumed a greater importance than the primary outcome. Predischarge mortality differed significantly between groups, in favour of poractant alfa (14.1 vs 31.0%, p=0.006; odds ratio 0.37 [95% CI 0.18-0.76). This difference was sustained after adjustment for centre, gestation, birthweight, sex, plurality, and use of antenatal steroids.

INTERPRETATION

Mortality was unexpectedly lower among neonates who received poractant alfa than among those who received pumactant, and was independent of all the variables we investigated. Stopping the trial early may have widened the difference between the treatment groups.

Risk adjusted and population based studies of the outcome for high risk infants in Scotland and Australia. International Neonatal Network, Scottish Neonatal Consultants, Nurses Collaborative Study Group

OBJECTIVES

To compare outcomes of care in selected neonatal intensive care units (NICUs) for very low birthweight (VLBW) or preterm infants in Scotland and Australia (study 1) and perinatal care for all VLBW infants in both countries (study 2).

DESIGN

Study 1: risk adjusted cohort study; study 2: population based cohort study.

SUBJECTS

Study 1: all 2621 infants of < 1500 g birth weight or < 31 weeks' gestation admitted to a volunteer sample of hospitals comprising eight of all 17 Scottish NICUs and six of all 12 tertiary NICUs in New South Wales and Queensland in 1993-1994; study 2: all 5986 infants of 500-1499 g birth weight registered as live born in Scotland and Australia in 1993-1994.

MAIN OUTCOMES

Study 1: (a) hospital death; (b) death or cerebral damage, each adjusted for gestation and CRIB (clinical risk index for babies); study 2: neonatal (28 day) mortality.

RESULTS

Study 1. Data were obtained for 1628 admissions in six Australian NICUs, 775 in five Scottish tertiary NICUs, and 148 in three Scottish non-tertiary NICUs. Crude hospital death rates were 13%, 22%, and 22% respectively. Risk adjusted hospital mortality was about 50% higher in Scottish than in Australian NICUs (adjusted mortality ratio 1.46, 95% confidence interval (CI) 1.29 to 1.63, p < 0.001). There was no difference in risk adjusted outcomes between Scottish tertiary and non-tertiary NICUs. After risk adjustment, death or cerebral damage was more common in Scottish than Australian NICUs (odds ratio 1.9, 95% CI 1.5 to 2.5). Both these risk adjusted adverse outcomes remained more common in Scottish than Australian NICUs after excluding all infants < 28 weeks' gestation from the comparison. Study 2. Population based neonatal mortality in infants of 500-1499 g was higher in Scotland (20.3%) than Australia (16.6%) (relative risk 1.22, 95% CI 1.08 to 1. 39, p = 0.002). In a post hoc analysis, neonatal mortality was also higher in England and Wales than in Australia.

CONCLUSIONS

Study 1: outcome was better in the Australian NICUs. Study 2: perinatal outcome was better in Australia. Both results may be consistent, at least in part, with differences in the organisation and implementation of neonatal care.

Predicting chronic lung disease in very low birthweight infants: comparison of 3 scores

AIM

To evaluate 3 scores in their ability to predict Chronic Lung Disease (CLD) in very low birthweight (VLBW) infants.

METHODS

The records of 188 VLBWs admitted to neonatal intensive care within two years were retrospectively reviewed. Two mortality scores -the CRIB (clinical risk index for babies) score and the Berlin admission score- and one morbidity score developed to predict CLD (Sinkin score) were assigned to each infant. Areas (AUC) under receiver operating characteristic (ROC) curves were used for comparison.

RESULTS

The Sinkin score and the Berlin admission score had AUCs of 89.0 and 85.8% to predict CLD ("alive in oxygen at 28 days of life"). The AUCs were 87.7 and 81.2%, respectively, using the CLD definition "alive in oxygen at 36 weeks gestational age", the CRIB had an AUC of 77.0%.

CONCLUSION

To enroll patients in trials aimed at early interference with the course of CLD, the Berlin admission score or the Sinkin score could be used.

Artificial neural network for risk assessment in preterm neonates

AIM

To predict the individual neonatal mortality risk of preterm infants using an artificial neural network "trained" on admission data.

METHODS

A total of 890 preterm neonates (< 32 weeks gestational age and/or < 1500 g birthweight) were enrolled in our retrospective study. The neural network trained on infants born between 1990 and 1993. The predictive value was tested on infants born in the successive three years.

RESULTS

The artificial neural network performed significantly better than a logistic regression model (area under the receiver operator curve 0.95 vs 0.92). Survival was associated with high morbidity if the predicted mortality risk was greater than 0.50. There were no preterm infants with a predicted mortality risk of greater than 0.80. The mortality risks of two non-survivors with birthweights > 2000 g and severe congenital disease had largely been underestimated.

CONCLUSION

An artificial neural network trained on admission data can accurately predict the mortality risk for most preterm infants. However, the significant number of prediction failures renders it unsuitable for individual treatment decisions.