Competing risks model for prediction of small-for-gestational-age neonates from biophysical markers at 19 to 24 weeks' gestation.

BACKGROUND: Antenatal identification of women at high risk to deliver small-for-gestational-age neonates may improve the management of the condition. The traditional but ineffective methods for small-for-gestational-age screening are the use of risk scoring systems based on maternal demographic characteristics and medical history and the measurement of the symphysial-fundal height. Another approach is to use logistic regression models that have higher performance and provide patient-specific risks for different prespecified cutoffs of birthweight percentile and gestational age at delivery. However, such models have led to an arbitrary dichotomization of the condition; different models for different small-for-gestational-age definitions are required and adding new biomarkers or examining other cutoffs requires refitting of the whole model. An alternative approach for the prediction of small-for-gestational-age neonates is to consider small for gestational age as a spectrum disorder whose severity is continuously reflected in both the gestational age at delivery and z score in birthweight for gestational age. OBJECTIVE: This study aimed to develop a new competing risks model for the prediction of small-for-gestational-age neonates based on a combination of maternal demographic characteristics and medical history with sonographic estimated fetal weight, uterine artery pulsatility index, and mean arterial pressure at 19 to 24 weeks' gestation. STUDY DESIGN: This was a prospective observational study of 96,678 women with singleton pregnancies undergoing routine ultrasound examination at 19 to 24 weeks' gestation, which included recording of estimated fetal weight, uterine artery pulsatility index, and mean arterial pressure. The competing risks model for small for gestational age is based on a previous joint distribution of gestational age at delivery and birthweight z score, according to maternal demographic characteristics and medical history. The likelihoods of the estimated fetal weight, uterine artery pulsatility index, and mean arterial pressure were fitted conditionally to both gestational age at delivery and birthweight z score and modified the previous distribution, according to the Bayes theorem, to obtain an individualized posterior distribution for gestational age at delivery and birthweight z score and therefore patient-specific risks for any desired cutoffs for birthweight z score and gestational age at delivery. The model was internally validated by randomly dividing the data into a training data set, to obtain the parameters of the model, and a test data set, to evaluate the model. The discrimination and calibration of the model were also examined. RESULTS: The estimated fetal weight was described using a regression model with an interaction term between gestational age at delivery and birthweight z score. Folded plane regression models were fitted for uterine artery pulsatility index and mean arterial pressure. The prediction of small for gestational age by maternal factors was improved by adding biomarkers for increasing degree of prematurity, higher severity of smallness, and coexistence of preeclampsia. Screening by maternal factors with estimated fetal weight, uterine artery pulsatility index, and mean arterial pressure, predicted 41%, 56%, and 70% of small-for-gestational-age neonates with birthweights of <10th percentile delivered at ≥37, <37, and <32 weeks' gestation, at a 10% false-positive rate. The respective rates for a birthweight of <3rd percentile were 47%, 65%, and 77%. The rates in the presence of preeclampsia were 41%, 72%, and 91% for small-for-gestational-age neonates with birthweights of <10th percentile and 50%, 75%, and 92% for small-for-gestational-age neonates with birthweights of <3rd percentile. Overall, the model was well calibrated. The detection rates and calibration indices were similar in the training and test data sets, demonstrating the internal validity of the model. CONCLUSION: The performance of screening for small-for-gestational-age neonates by a competing risks model that combines maternal factors with estimated fetal weight, uterine artery pulsatility index, and mean arterial pressure was superior to that of screening by maternal characteristics and medical history alone.

Assessment and modeling of routinely used biochemical laboratory data of healthy individuals and end-stage renal failure (ESRF) patients by three different chemometric methods.

BACKGROUND: In recent years, the use of biochemical markers has received increasing attention for purposes of risk assessment and clinical management in renal failure patients. Chemometric methods are often used in medical studies and there are already indications for their specific role as a tool of the medical statistics. METHODS: Three chemometric methods, discriminant analysis (DA), binary logistic regression analysis (BLRA), and cluster analysis (CA), were used for assessment and modeling of routinely used biochemical laboratory data of 18 parameters that were determined from 185 healthy individuals (HIs) and 173 end-stage renal failure (ESRF) patients. RESULTS: The above-mentioned chemometric methods were performed using the data set of 14 parameters since the rest 4 parameters did not present significant difference between healthy and patients. DA created a model using only ALB (Albumin), K (Potassium), TG (Triglyceride), and ALP (Alkaline phosphatase); BLRA model also used the above four parameters; CA classified all the cases into two clusters using the same four parameters and one more parameter, AST (aspartate aminotransferase). CONCLUSIONS: This study provides models for assessment and modeling of routinely used biochemical laboratory data, finding groups of similarity among clinical tests usually determined on HIs and ESRF patients, contributing in data mining and reducing costs.