Design, construction and evaluation of systems to predict risk in obstetrics.

We present a systematic, practical approach to developing risk prediction systems, suitable for use with large databases of medical information. An important part of this approach is a novel feature selection algorithm which uses the area under the receiver operating characteristic (ROC) curve to measure the expected discriminative power of different sets of predictor variables. We describe this algorithm and use it to select variables to predict risk of a specific adverse pregnancy outcome: failure to progress in labour. Neural network, logistic regression and hierarchical Bayesian risk prediction models are constructed, all of which achieve close to the limit of performance attainable on this prediction task. We show that better prediction performance requires more discriminative clinical information rather than improved modelling techniques. It is also shown that better diagnostic criteria in clinical records would greatly assist the development of systems to predict risk in pregnancy.

Artificial neural networks for decision support in clinical medicine.

Connectionist models such as neural networks are alternatives to linear, parametric statistical methods. Neural networks are computer-based pattern recognition methods with loose similarities with the nervous system. Individual variables of the network, usually called 'neurones', can receive inhibitory and excitatory inputs from other neurones. The networks can define relationships among input data that are not apparent when using other approaches, and they can use these relationships to improve accuracy. Thus, neural nets have substantial power to recognize patterns even in complex datasets. Neural network methodology has outperformed classical statistical methods in cases where the input variables are interrelated. Because clinical measurements usually derive from multiple interrelated systems it is evident that neural networks might be more accurate than classical methods in multivariate analysis of clinical data. This paper reviews the use of neural networks in medical decision support. A short introduction to the basics of neural networks is given, and some practical issues in applying the networks are highlighted. The current use of neural networks in image analysis, signal processing and laboratory medicine is reviewed. It is concluded that neural networks have an important role in image analysis and in signal processing. However, further studies are needed to determine the value of neural networks in the analysis of laboratory data.

Excessive secretion of insulin precursors characterizes and predicts gestational diabetes.

Using assays that specifically measure insulin, intact proinsulin, and 32,33 split proinsulin, we examined the beta-cell secretory response to an oral glucose tolerance test (OGTT) in 64 women with gestational diabetes mellitus (GDM) and 154 pregnant normoglycemic control subjects of comparable age and body mass index. Women with GDM were characterized by a lower 30-min insulin increment (40.8 [34.9-47.6] vs. 58.6 [53.6-64] pmol insulin/mmol glucose, P < 0.001; geometric mean [95% confidence interval]) and a higher plasma insulin level at 120 min (702 [610-808] vs. 444 [400-492] pmol/l, P < 0.001). 32,33 split proinsulin levels were elevated in GDM patients in both fasting (9.1 [7.3-11.4] vs. 6.7 [6.0-7.5] pmol/l, P < 0.02) and 120-min (75.2 [62.9-90.0] vs. 52.2 [46.7-58.3] pmol/l, P < 0.001) samples, respectively. Intact proinsulin levels were significantly elevated at 120 min in the women with GDM (21.3 [18.1-25.1] vs. 14.8 [13.4-16.3] pmol/l, P < 0.001). Thus, the qualitative abnormalities of insulin secretion in GDM patients (low 30-min insulin increment, high 120-min plasma insulin, and elevated 32,33 split proinsulin) are similar to those seen in nonpregnant subjects with impaired glucose tolerance. To determine whether measures of proinsulin-like molecules (PLMs) might assist in the prediction of GDM, women who had a 1-h glucose level of > 7.7 mmol/l after a 50-g glucose challenge at 28-32 weeks' gestation had insulin and PLMs measured in the 1-h sample.(ABSTRACT TRUNCATED AT 250 WORDS)

Application of neural networks to the ranking of perinatal variables influencing birthweight.

In this paper we compare Multi-Layer Perceptrons (a neural network type) with Multivariate Linear Regression in predicting birthweight from nine perinatal variables which are thought to be related. Results show, that seven of the nine variables, i.e., gestational age, mother's body-mass index (BMI), sex of the baby, mother's height, smoking, parity and gravidity, are related to birthweight. We found no significant relationship between birthweight and each of the two variables, i.e., maternal age and social class.

Objective measurement of anxiety in hypertensive pregnant women managed in hospital and in the community.

OBJECTIVES: To determine whether pregnant hypertensives women are more anxious when monitored in hospital or at homes. DESIGN: Prospective randomized controlled trial. SETTING: Rosie Maternity Hospital and women's homes. SUBJECTS: Ninety-nine pregnant hypertensive women: 50 had their blood pressure measured telemetrically from home, and 49 had it measured in hospital. MAIN OUTCOME MEASURES: Number of episodes of monitoring, duration of monitoring, mean blood pressure during monitoring, gestational age at delivery, trait and state anxiety levels. RESULTS: There were no significant differences in anxiety levels, or in any other outcome measure, between the home and hospital groups. CONCLUSION: When blood pressure is being monitored serially in pregnant hypertensive women, there is no measurable difference in their anxiety levels, whether they are in hospital or at home.

A comparative clinical study of fetal phono- and movement-sensors from Amsterdam, Cambridge and Edinburgh.

In a European collaboration, a joint project to conduct an experimental and clinical investigation of recently developed sensors from three centres (Amsterdam, Cambridge and Edinburgh) has been carried out. The Amsterdam sensor was based on an inductive principle whereas the Cambridge and Edinburgh transducers used a piezo-electric material (PVDF) as transducing element. Nine patients with varying gestational age (29-38 weeks) were measured in a clinical investigation. Recordings of fetal heart sounds (FHS) and fetal breathing movements (FBM) were made using three sensors; one from each centre. These recordings were digitized directly into a computer using a purpose-built data acquisition system. For each patient 3 min of FBM data, and 1 min of FHS data were recorded by each sensor. The FBM recordings were carried out simultaneously with ultrasound, so as to enable a correlation to be made between both recordings. The FHS recordings were carried out simultaneously with the maternal heart pulse, to discount any maternal heart influences on the resulting signals. Of the nine patients analysed, FHS were recorded in seven patients. On the other hand, it appeared difficult to identify fetal breathing movements in the FBM recordings due to the dominance of the maternal breathing component. The analysis of the FBM signals and its correlation with ultrasound could not be carried out due to the relatively poor quality of the signals detected by the sensors, given the present techniques of analysis. The evaluation of the FHS recordings showed that although there is relatively little difference between the sensors, the inductive sensor performed best.

No difference between home and clinic blood pressure measurements in pregnancy: a computerized telemetric study.

Many decisions in pregnancy are based on one or two casual blood pressure estimations made in the antenatal clinic. No previous study has challenged the validity of this practice or attempted to discover whether there are significant differences between the blood pressures of pregnant women when they are measured in the clinic and when they are measured in their home environment. We measured the blood pressures of 35 healthy pregnant women both at the antenatal clinic and in their own homes. All the women were at between 28 and 32 weeks gestation. At both locations, ten consecutive blood pressure and pulse measurements were made at one minute intervals using a Dinamap 1846P automated blood pressure monitor. The readings taken in the clinic were supervised, whilst at home the patients used our blood pressure telemetry system without direct supervision, the results being transmitted automatically to the hospital computer via the public telephone network. The Dinamap results were analysed in detail. Each ten minute recording session was summarised in terms of the first, last, highest, lowest and mean of the systolic, diastolic and mean arterial blood pressure, and pulse measurements in that session. Also, two measures of the variability within a ten measurement series were considered, the average change in pressure or pulse from one reading to the next, and the difference between the highest and lowest readings of pressure and pulse in the series. All descriptors of the clinic and the home measurements were compared using the paired Student's t-test. Clinic and home blood pressure measurements were not significantly different from each other in any respect.(ABSTRACT TRUNCATED AT 250 WORDS)

Blood pressure measured telemetrically from home throughout pregnancy.

Sixty-eight healthy pregnant women measured their blood pressure at home for 1 day every 4 weeks from 14 weeks' gestation until delivery. Blood pressure measurements were made with the Cambridge blood pressure telemetry system, which was composed of a Dinamap 1846 monitor linked to and controlled by a dedicated microprocessor which communicated transtelephonically with a hospital-based microcomputer. On each measurement day each woman measured her blood pressure five times: on first rising and then at 10 AM, 2 PM, 6 PM and 10 PM. At each measurement session the blood pressure was automatically taken 10 times at 1-minute intervals. A total of 23,790 individual blood pressure measurements were analyzed. Systolic, diastolic, and mean arterial blood pressures were lowest at 18 weeks' gestation, rising slowly thereafter at a rate of 0.4 mm Hg systolic and 0.2 mm Hg diastolic per week. The maternal pulse rate also increased steadily after 18 weeks at a rate of 0.25 beats/min per week. There was a strong correlation between the blood pressures at 18, 26, and 38 weeks. All blood pressure parameters were significantly lower on first rising than at other times of the day, although the differences were small: 5 mm Hg for systolic, 4 mm Hg for mean arterial, and 3 mm Hg for diastolic blood pressure. After 10 AM, however, there was no significant daytime variation in any blood pressure parameter, indicating that for most purposes the time of day at which pressure measurements are made is unimportant.

GLUCOPAGE: a computer program for managing diabetic pregnancy.

The microcomputer is an invaluable aid in the management of diabetes mellitus because of its ability to manipulate, summarise, and produce graphical displays from blood glucose data in such a way that new trends, or the effects of changes in therapy, are more readily apparent to the clinician. Currently available software however, while adequate for most purposes, has not been written to cope with the unique problems presented by diabetic pregnancy in which large amounts of data are generated over short periods of time, insulin requirements may change rapidly, and blood glucose control must be extremely tight. We have produced a new program, GLUCOPAGE, specifically for use in the management of diabetic pregnancy. GLUCOPAGE operates in conjunction with the Ames Memory Glucometer, from which it can download data either directly or telemetrically via a modern into the host IBM PC or compatible microcomputer. Large amounts of data can be summarised in various tabular and graphical formats on screen or in printed form. A single A4-size printout, the Glucopage, provides tabular and graphical summaries of a full week's blood glucose data. Copies of the Glucopage may conveniently be filed in the case notes, given to the patient or distributed to other medical staff.

An 'admission challenge test' to predict severe hypertension in pregnancy?

Although many pregnant women are admitted to hospital for 'high blood pressure' during the third trimester, less than a third of them will subsequently develop severe hypertension. We used a Dinamap 1846 vital signs monitor to measure the blood pressure every 5 min during the first hour in hospital of a group of 51 previously normotensive pregnant women admitted with hypertension. An average systolic pressure at or above 140 mmHg was recorded in 37% of patients and, of these, 68% went on to develop severe hypertension. Conversely, if the average systolic pressure on admission was below 140 mmHg, only 3% of patients later developed severe hypertension. In terms of sensitivity, specificity, and positive and negative predictive values of severe hypertension, the 1-hour automated blood pressure profile was superior to conventional measurements whether they were made prior to admission, on admission, or during the first 24 hours after admission.

Morphometric differences between the placental vasculature of non-smokers, smokers and ex-smokers.

The aim of this study was to determine whether cigarette smoking during pregnancy has an adverse effect upon the placenta's capacity for gaseous exchange. Using recently developed stereological techniques, in conjunction with perfusion fixation, computer-assisted measurements were made on the placentas of 15 non-smokers, 15 moderate smokers, 15 heavy smokers and 13 ex-smokers, 7 of whom stopped smoking during the course of the pregnancy. Compared with the placentas of non-smokers and of those who stopped before pregnancy, it was found that the placentas of smokers and of those who stopped after conception exhibited a reduced capillary volume fraction, and an increased thickness of the villous membrane. Although they must impair gaseous exchange across the placenta, these changes were less severe than suggested by previously published reports. Nonetheless it is clear that in order to prevent these changes women should stop smoking before conception rather than during the course of a pregnancy.

PIP: To learn more about the mechanism of action whereby cigarette smoking leads to a significant reduction in birth weight, computer-assisted measurements were made of the affects of cigarette smoking on placental structure. The study sample included 15 non-smokers, 15 moderate (1-15 cigarettes/day) smokers), 15 heavy (over 15 cigarettes/day) smokers, and 13 ex-smokers, 7 of whom stopped smoking during the index pregnancy. Mean cigarette consumption was 9.3/day in the moderate smoker group and 22.2/day in the heavy smoker group. Pre-conceptional ex-smokers stopped, on average, 13 weeks before conception, while post-conception ex-smokers stopped, on average, 12 weeks after conception. There was no significant difference in birthweight between offspring of non-smokers, moderate smokers, and ex-smokers; however, the infants of heavy smokers were significantly smaller. The terminal villi from all subjects, even women who smoked 30 cigarettes/day, showed large dilated fetal capillaries, with smooth luminal outlines, and no evidence of widespread trophoblastic damage. In terms of placental vasculature, the only significant finding was a slight reduction in the capillary volume fraction in women who had smoked at any stage of pregnancy compared to non-smokers or women who quit smoking before conception. In addition, the villous membrane was significantly thicker among smokers and post- conception stoppers compared to women in the other 2 groups. These findings suggest a less severe effect on the placental vasculature than reported earlier; however, the increased thickness of the villous membrane of smokers observed in this study could compromise gas transfer to the fetus and thus cause growth retardation.

Computerized and telemetric management of diabetic pregnancy in Cambridge.

In order to optimize the quality of our care, we have used the Ames Memory Glucometer in our computerized management of 17 insulin-dependent diabetic pregnancies. Patients measured their own blood glucose levels on 4.5 +/- 1.2 (SD) occasions per day, and six of them transmitted their blood glucose measurements telemetrically from home into a hospital-based computer on a total of 42 occasions. We achieved near-optimal blood glucose levels and normal HbA1 levels throughout most of these 17 pregnancies. Average blood glucose levels were 6.4 +/- 0.9 mmol/l in the first, 5.9 +/- 1.2 mmol/l in the second, 5.4 +/- 1.0 mmol/l in the third trimester. Corresponding values for HbA1 were 8.0 +/- 1.5%, 6.5 +/- 0.8%, and 6.2 +/- 0.4%. Average insulin requirement doubled during pregnancy, but fell to 74% of the pre-pregnancy dosage by one week after delivery. The average gestational age at delivery was 38.9 +/- 1.3 weeks, and we had no perinatal deaths. Two of the babies were macrosomic, even though their mothers had good diabetic control. Four patients were delivered by Caesarean section. However we achieved a vaginal delivery in 93% of those in whom it was planned. Our computerized system of managing diabetic pregnancies was efficient, safe, and popular with our patients.

Computerized home telemetry of maternal blood pressure in hypertensive pregnancy.

We have developed a telemetric technique whereby maternal blood pressure, which is self-measured by pregnant women in their own homes using a Dinamap 1846 automated blood pressure recorder, can then be transmitted over the commercial telephone network into the Rosie Maternity Hospital in Cambridge, where it is computer-processed. The maternal blood pressure is then reviewed by the obstetrician as part of the clinical management protocol. We have used this telemetric technique on 90 occasions, from the homes of 10 pregnant hypertensive women. On almost every occasion, the blood pressure measured at home was lower than that previously measured in the hospital antenatal clinic. This technique offers great promise, both in terms of health economics and also in terms of reducing pregnant women's unhappiness about their being admitted to hospital whenever they exhibit moderate to severe hypertension in the antenatal clinic. Indeed, in the antenatal period, home telemetry should allow the vast majority of hypertensive pregnancies to be managed just as safely at home as in hospital. In the management of high risk pregnancy, home telemetry of maternal blood pressure complements three other home telemetric techniques which have already been described: fetal heart rate, maternal blood glucose and uterine contractions.

Computerized ultrasonic image analysis for placental characterization in normal and hypertensive pregnancies.

A technique has been developed for characterizing ultrasonic images of the human placenta by computerized image analysis. An ultrasonic image data base has been assembled from routine obstetric scans collected from 112 patients. A region within the placenta was manually identified in each image, and a series of parameters which mathematically describe the image texture in the region of interest was calculated. Our pilot study has shown that gestational age at scan, placental position and the presence or absence of hypertension can all be correlated with the mathematically defined textural descriptors of the ultrasonic placental image.

Measurement of compliance of the maternal abdominal wall in pregnancy.

Recent importance in the signal-to-noise ratio of fetal phonocardiograms has been attributed to a 'compliance-matching' between the phono-sensor and the maternal abdominal wall. We have developed a device to measure objectively the compliance of the maternal abdominal wall in pregnancy. Using this we have shown that the compliance varies considerably from patient to patient and that it varies most significantly (and inversely) with gestational age. As a result of this study we are now in a better position to optimize the compliance-matching of phono-sensors and thereby to detect fetal heart sounds with much more fidelity than hitherto.

An experimental rig to simulate fetal heart sounds.

Renewed interest in fetal phonocardiography has brought about new phono-sensor designs. We have developed an experimental rig to simulate generation of fetal heart sounds and also their passage through to the maternal abdominal wall. Using this rig we have investigated the design techniques applied to the new phono-sensors and we have also analysed the sensors themselves. We can now approach an ideal sensor design for the faithful detection of fetal heart sounds.

CEUSPEC--a computerized system for fetal home telemetry.

We have developed a computerized system whereby the fetal heart rate can be recorded telemetrically from patients' homes, transmitted over conventional public telephone lines, and then computer-processed in real time in the obstetric unit.

Ultradian rhythms in human fetal heart rate: a computerised time series analysis.

Even though fetal heart rate recordings are widely used to monitor fetal health, both antenatally and in labour, the underlying physiology is not well understood. For example, it is not known with any certainty whether the oscillations seen in fetal heart rate are highly organised, in reflection of underlying ultradian rhythms, or whether they are entirely random and haphazard. In order to answer this question, therefore, we have used mathematical techniques of time series analysis to look for clear evidence of ultradian rhythms in fetal heart rate recordings. We have found that specific short-term ultradian rhythms are indeed present, and that they can be measured objectively in terms of their frequency, amplitude and phase. Such rhythms have cycle lengths of 10 to 90 s and they can persist for long periods of time. They may also disappear and later reappear, locking back into synchrony again with previous oscillations. Individual rhythms may undergo amplitude change, phase shift, and perhaps even frequency shift.