Assessment of risk of perinatal death in Jamaica

Please download to get full document.

View again

of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Art & Photos

Published:

Views: 2 | Pages: 8

Extension: PDF | Download: 0

Share
Related documents
Description
Assessment of risk of perinatal death in Jamaica
Tags
Transcript
  Paediutric and Perinatal Epidemiology 1994 8 Suppl. 1,166-173 Assessment of risk of perinatal death in Jamaica Rosemary Greenwooda and Deanna Ashleyb stitUte of Child Health, Bristol, UK and bMinistry of Health, ]arnica Summary. Data from the Jamaican Perinatal Mortality Survey had been used to create a statistical model using logistic regression.' From this a simple additive scoring system to predict perinatal death was devised and tested on the 2 cohort months of the study. The score had a theoretical range of &28 points, with the higher the score, the greater the likelihood of a perinatal death. For a cut-point of 7, sensitivity was 43 and specificity 84 . A cut-point of 8 resulted in 27 sensitivity and 94 specificity. Higher cut-points resulted in much reduced sensitivity but enhanced specificity (e.g. cut-point 10 11 ensi- However, it is likely that these estimates are optimistically high, and to achieve unbiased estimates of sensitivity and specificity he score needs to be tested on a sample of the population from which it was not derived before implementation takes place. Meanwhile, the cut-off level for im plementation will depend on appropriate resources available. tivity, 99 specificity). Introduction We have shown elsewhere that quality of available perinatal care, as determined by the availability of consultant obstetric and paediatric services n the mother's parish of residence, has a profound effect on the risk of perinatal mortality in both twin2 and singleton' pregnancies. In a country with limited resources (such as Jamaica), t is not feasible to provide such expertise and facilities throughout the island. Consequently t is important to identlfy mothers at highest risk, and transfer them, while still pregnant, to be cared for in a centre where appropriate care is available. A simple scoring system may be useful in this regard. In this paper we take Address for correspondence: Rosemary Greenwood, University of Bristol, Department of Child Health, Royal Hospital for Sick Children, St Michael's Hill, Bristol BS2 8BJ. 166  Assessment of risk of perinatal death 167 those factors already shown to be independently associated with perinatal mor- tality among singletons, and create a proposed scoring system for this outcome. Twin pregnancies carry such a high risk that once identified all should be referred to specialist care. Material and methods The Jamaican Perinatal Mortality Survey comprised two nterlocking components: (1) a study of all births occurring on the island in the 2-month period of September and October 1986, known as the cohort study, and (2) a study of all perinatal deaths occurring on the island in the 12-month period from 1 September 1986 to 31 August 1987. Comparison of the singleton perinatal deaths in the 12 months with the single- ton survivors from the cohort months has been carried out previously,’ and logistic regression models produced. This process produced a model in which a large number of terms were independently associated with perinatal deaths. It was thought that it would be unreasonable to expect midwives to calculate a risk score for each of their patients which contained information on all these items, especially as several of the items would be subject to change during the pregnancy, and many would contribute very little in terms of predictive power. It was decided, therefore, to concentrate on those items which contributed most to the score in terms of odds ratio. It was arbitrarily decided that a multiplicative increase or decrease in risk of 1.5 represented a clinically significant difference. Thus from the final model, all factors that showed a (1.5-fold) decrease or increase in relative risk across cat- egories were selected. The model was then refitted, and the process repeated until the only terms left in the model were those which strongly affected the likelihood of perinatal death. This cut down the number of clinically significant variables to a more manageable level. It was thought that an additive score would be easier to complete than a multiplicative score, so the log of the odds ratios were used. These were multiplied by 2 and rounded. Other integers (3,4,5 etc.) were used in an attempt to make a more sensitive or more accurate score, but were found to add no improvements to the predictive ability of the score, while increasing its complexity. The scores were calculated using an additive model for all survivors from the cohort months and all perinatal deaths for which there were complete data (6813, and 898 deaths). From this, sensitivities and specificities were calculated for a range In order to calculate he positive predictive value, the sampling fraction must be taken into account, so that it holds true for the population. Thus, to calculate positive predictive value, and negative predictive value, only births from the 2 cohort months have been used (6813 survivors and 214 perinatal deaths). of cut-off points.  168 Results R Greenwood and D. shley In Table 1, the remaining terms with their simplified scores are set out. It can be seen that there are two major categories: past obstetric history (contributing up to points) and maternal disorders present during the pregnancy (contributing up to 17 points). In contrast only one social/environmental variable was used which was the number of children in the household. This variable is highly predictive, and although combining a proxy for parity and past obstetric history, is also thought to include a measure of social supp0rt.l For all births in each study, social, biological and environmental features were elicited by interview with the mother. Antenatal, obstetric and neonatal records were abstracted onto structured proforma. At least 94 of the births were a~certained.~ Although there had been a number of other social variables in the ha1 ogistic regression model,’ the ratio between the lowest and highest odds ratios did not Table 1. Scores ascribed Score Environmental No. of children in household under age 11 0 I 2 3+ Past obstetric history Previous miscamage or termination Previous stillbirth Previous early neonatal death Previous Caesarean section Maternal disorders Syphilis Bleeding < 28 weeks Bleeding 28 + weeks Vaginal discharge/infection untreated Antenatal eclampsia Diabetes First diastolic bp 90 mm + Highest diastolic bp: 100-109 110+ Highest proteinuria: ++ +++ or more Health behavioudantenatal care Not taking iron tablets 2 1 2 1 3 1 2 1 3 1 2  Assessment of risk of perinatal death 169 reach 1.5 for union (marital) status 1.49; maternal employment status 1.48; number of adults in household 1.47; use of toilet 1.36. Although the trimester of start of antenatal care was significantly related to risk of perinatal mortality, the odds ratio was less than 1.5. Other features of health behaviour such as the protective associ- ation of drinking alcohol, were also omitted because of low odds ratio. The 'trying to get pregnant' variable was omitted after the reduced model had been fitted because the odds ratio fell below 1.5 at this stage, as did the odds ratio for parish of residence. The score obviously can increase as the pregnancy develops. However the only way it can decrease (and then only by 1 point) is when the mother starts taking prophylactic iron. The distribution of scores among the cohort survivors and perinatal deaths is summarised in Table 2. It can be seen that, for example, 66 of the deaths had a score of 6 or more compared with 39 of the survivors, implying that the risk of perinatal death for mothers with a score of 6 or more was 3 times that of a pregnancy with a score of less than 6. The sensitivity of the scoring system for the three major types of death: antepar- tu fetal death (APFD), deaths from immaturity (Ih4MAT) and deaths from intrapartum asphyxia (PA) are shown in Table 3 (specificity remains similar o that in Table 2). It can be seen that from cut-points of 6 or higher, the score predicts APFD slightly more efficiently than the other two, but all three types of death are predicted efficiently. Table 2. Performance of score using various cut points, with odds ratio (OR) of death in high-risk group Whole sample Cohort months only Cut-Point Sensitivity Specificity OR Sensitivity Positive Negative predictive predictive value value 4+ 97.8 3.6 1.7 97.7 3.1 98.0 5+ 87.6 25.6 2.6 87.9 3.6 98.5 6+ 65.9 61.1 3.1 65.9 5.1 98.3 7+ 42.9 84.4 3.7 44.4 8.2 98.0 8+ 27.1 94.0 4.7 24.3 11.2 97.5 9+ 16.8 97.8 6.8 11.2 13.7 97.2 10+ 10.6 99.0 9.0 5.1 14.5 97.1 + 7.2 99.6 13.0 4.2 26.5 97.1 12i 4.5 99.8 16.6 2.3 25.0 97.0 13 + 2.0 99.9 22.3 1.4 30.0 97.0 14 + 0.7 100.0 - 0.0 - based on 12 months deaths  17 R Greenwood and D. shley Table 3. Sensitivity for each type of death in the 12 months according to score Cut-point APFD IMMAT IPA 4+ 5+ 6+ 7+ 8+ 9+ 10 + 12 + 13 + 14 + 98.0 87.4 72.4 49.7 33.3 24.5 16.7 13.6 9.2 4.8 O 98.4 90.2 69.9 43.9 25.2 14.6 8.1 5.7 4.1 1.6 0.8 97.5 88.9 63.0 40.1 24.7 13.6 8.1 4.3 2.0 0.5 0.5 The optimum cut-point to choose depends on the use to which the score is put. Some authors use the maximum of Jouden’s index (sensitivity + specificity - loo ), which for the present data gives results of 1, 13,27,27,23,14,10 nd 7 respectively for each of the cut-points shown in Table 2, mplying that cut-points at 6 or 7 would do equally well, but that a cut-point at 8 is also efficient. If the score were to be used to refer pregnant mothers across the island to centres of experi- enced perinatal care, then it might be felt more reasonable to have the cut-point at 8 which, while only identifymg 27 of the deaths .would have only referred 6 of mothers. Discussion In any statistical assessment of risk it is important to test the prediction score on data other than that from which it was derived? It could be argued that a split sample approach could have provided an independent assessment, but others have found that this method reduces the accuracy of estimates by virtue of smaller n~rnbers.~ owever, the sensitivities calculated for the cohort months alone were almost identical to those calculated for the whole sample. It is usual to find that testing a score on the data from which it was derived yields optimistically high estimates of sensitivity and specificity. However, with a data set of this size, the true estimate is unlikely to be very different. If it is to be validated on a separate set of data it is important that data be collected before the score is implemented, as any improvements in mortality due to the score would be in the high-risk groups and, if intervention is beneficial, this would effectively even out the differences that the score is designed to detect. Thus in order to validate the success of the score after implementation, it would be necessary to compare the mortality rates before and after.
Recommended
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x