Using the Birth Record to Develop a Screening Instrument for Infant Mortality and Morbidity
Daniel R. Thompson, Richard S. Hopkins, and Sharon M. Watkins
Beginning April 1, 1992, all infants born in Florida are screened for increased risk of mortality and morbidity. To develop the screening criteria, birth records for 1989 were linked to infant death records. This data was used to select factors from the birth record that are good predictors of post-neonatal death. It was found that a score based on 10 items from the birth record includes 14 percent of the births and 48 percent of the post-neonatal deaths. Infants that screen positive are 6.2 times more likely to die post- neonatally than those that screen negative. This screening criteria is applied to all infants born in Florida since 4/1/92 and those that screen positive are offered enhanced case management services.
Background Information In 1991, the Florida Legislature passed legislation generally referred to as the Healthy Start Initiative (Florida Statutes section 383.14, 1990 supplement). The law specified that an infant screening instrument was to be selected by the Florida Department of Health and Rehabilitative Services and applied to all infants born in Florida to identify those with greater than average risk of having health problems.
The screening criteria selected was developed by the Florida Department of Health and Rehabilitative Services, State Health Office and approved by the Healthy Start Advisory Committee. The committee included representatives of Florida County Public Health Units, two Florida universities, the state legislature and seven representatives from private sector health care.
The purpose of this paper is to describe the development of the screening instrument and present preliminary data regarding its effectiveness.
Summary of Methods To determine which items on the birth certificate are useful predictors of health risk, an outcome measure is needed. For this analysis the outcome measure is infant death at age 28 to 364 days (also known as post-neonatal death). The assumption is that a group of infants that have a high proportion of post-neonatal deaths will include survivors that by definition have the same characteristics as the fatalities. Even though the survivors did not die, it is assumed that since they have the same risk factors as the fatalities, they are at increased risk of morbidity and likely to require more intensive health care than infants without these characteristics.
The data file used for this analysis includes all births to Florida residents in 1989. In cases where the infant died before one year of age, the birth records are linked to the corresponding death records. There were 192,887 births in this file. Of these, 1,250 died in the neonatal period (0 to 27 days) and 521 died post- neonatally (28 to 364 days).
This file was used to calculate the death rates for all of the relevant factors on the birth certificate. For example, there are 1,050 births recorded as having Abnormal Condition number 7 on the birth certificate. This is labeled "Assisted ventilation > or = 30 min." on the birth certificate. Out of the 1,050 births with this condition, 34 died before they were one day of age; 85 died at age one to 27 days; and 24 died at age 28 to 264 days. The postneonatal death rate for infants age 28 to 364 days with this condition is calculated from the above to be 25.78 deaths per 1,000 births. As a comparison, the same rate for all of the births is 2.72. This means that infants recorded as having assisted ventilation > or = 30 minutes were approximately nine times more likely to die at age 28 to 364 days. However, most of the infants (907 out of 931) with this condition did not die. Based on the assumption explained above, these survivors are expected to require more intensive health care than the average child.
This analysis was done for all of the potentially relevant factors on the birth certificate. Many of the factors were too rare for inferences to be made about their associated risk, and some factors were not associated with high risk for post-neonatal death. In general, factors were considered for use in the screening criteria if they occurred in more than 1,000 births and had an associated risk of at least twice the average risk. The selected factors are listed in Table I with their post-neonatal deaths.
It can be seen that the factors in Table I all meet the two conditions for inclusion in the risk criteria, except the mother's education factor. For example, the post-neonatal death rate for infants with "abnormal conditions of newborn" is 22.5 per 1,000 infants versus 2.6 for infants who do not have this factor. The risk ratio is 22.5/2.6 or 8.7. This meets the first condition which is a risk ratio of two or more. In the fourth column of Table I, the number of births positive for the abnormal conditions factor is 1,380. This meets the second condition of 1,000 or more births positive for the factor.
The mother's education factor does not meet the risk ratio condition since the risk ratio is 1.5, but based on the professional judgment of several educators, this factors is included in the risk criteria and given a slightly greater weight (see explanation of the risk scoring algorithm below) than other factors with comparable risk ratios. The basis for this is the assertion that the mother's education is more closely associated with infant morbidity than post neonatal death, so post neonatal death is a biased proxy for morbidity in the case of mother's education. The emphasis given to mother's education is based on an assumed association with post neonatal death. Thus the predictors of death are not necessarily good predictors of other important non medical outcomes.
The screening criteria was then developed by using the factors to classify the infants into several groups based on the level of risk. This is illustrated by Table II. Infants with the factor associated with the highest risk are counted as members of group 1 in Table II. Births with the factor associated with the second highest risk are counted as members of group 2 if they are not already included in group 1. This is done for all of the selected risk factors until the last group has none of the selected risk factors. In cases where births qualify for membership in two or more groups, they are counted only in the highest risk group for which they qualify. Risk rates are then calculated for each group.
The classification scheme used in Table II is an attempt to assess the contribution that each individual risk factor makes to the risk criteria. For example, in Table II the infants in group 1 have one risk factor, birth weight under 2,000 grams. These infants have a very high post-neonatal death rate of 25.7 per 1,000 infants. These infants are therefore classified as high risk.
The question then becomes which of the remaining infants should be classified as high risk. Table II provides information directed at this question in the data presented for group 2. This group includes infants that have one or more abnormal conditions (see Table II), and are not included in group 1. None of the infants in group 2 have a birth weight below 2,000 grams. Infants in group 2 have a post-neonatal death rate of 14.2 per 1000 infants. This is a high death rate compared to the overall rate of 2.7, so these infants are also included in the high risk group.
The next question of the infants not included in group 1 or 2 should be classified as high risk. This question is addressed by the data for group 3 in Table II. Group 3 includes infants, who are not included in groups 1 or 2, who have one or more congenital anomalies. The post- neonatal death rate for these infants is 11.1 per 1,000 infants. Compared to the overall rate of 2.7, this is relatively high, so the births in group 3 are also classified as high risk.
This process is repeated with groups 4 through 11. The infants in groups 1 through 7 are all classified as high risk based on their relatively high rates of post- neonatal death. The infants in groups 8 through 11 do not meet the criteria that qualify them for inclusion in any of the groups 1 through 7, and they have relatively low postneonatal death rates. These infants are therefore not classified as high risk.
Discussion It should be noted that there are various multivariate statistical methods that could be used instead of the hierarchical grouping method described above. However, the above method was used because of its comparative ease of application and the practical focus of the results. Also this method is understandable to a wider audience than multivariate methods. This last advantage can be important when the screening criteria developed is to be used by a large number of practitioners with various levels of training in statistical methods.
The decision regarding the level of risk used to classify infants as high risk is as much a policy decision as it is a scientific one. In Florida, the decision is that infants in groups 1 through 7 all have risk levels high enough to classify them as high risk. According to column 5 in Table II, this includes 13.8% of all infants born in Florida in 1989 and from column 4 in Table II it can be seen that 48.2% of the post-neonatal deaths occurred among this 13.8% of the infants. Incidentally, the 48.2% also represents the sensitivity of this screen when infants in groups 1 through 7 are classified as positive. The significance of being classified high risk is that intensive services (care coordination, home visits, nutrition and parenting education) will be provided for infants in this group. Including group 8 in the high risk group increases the percentage of high risk infants from 13.8% to 22.7% and increases the sensitivity from 48.2% to 58.3%. This is a small gain in sensitivity in exchange for a large increase in workload. Since the infants in group 8 have a relatively low post-neonatal mortality rate of 3.1 per 1,000 births, they are not included in the high risk group. With additional funding and service delivery capacity, it might be decided that infants in group 8 should be added to the high risk category. The point here is that it is a program and policy decision and not a purely quantitative problem.
Classifying infants into the group in Table II is too complicated to be done on a routine basis. To make the risk criteria easier to use operationally, the scoring values shown in the last column of Table I are applied. For example, an infant who has congenital anomalies and whose mother smoked, would be assigned 4 points for the congenital anomalies factor and 1 point for the smoking risk factor, for a total risk score of 5. The scoring values are constructed so that infants in groups 1 through 7 in Table II are given a score of 4 or greater and classified as high risk.
Since April 1, 1992, all infants born in Florida have been scored using this algorithm. Table III shows the postneonatal death rates associated with each score and Graph 1 is a graph of death rates by risk score. Table III and Graph 1 show a steady and substantial increase in risk of post-neonatal death as the score increases. Infants that screen positive (score of 4 or more) are 6.1 times more likely to die post-neonatally than those that screen negative.
In an effort to reduce their health risks, Healthy Start infants are given enhanced services from county public health units and private health care providers. The enhanced services are, at present, mainly case management, home visits, parenting education, and nutritional counseling. The hope is that more educational, psychosocial and economic services will be available in the future.
Conclusion It appears that the data on the birth certificate can be used effectively to identify infants with higher than normal risk of death at age 28 to 364 days. If the assumption is accepted that death is strongly associated with morbidity, then the birth record data can also be used as a screening tool for infant morbidity.
The screening tool will be evaluated for performance and utility as soon as enough data is available to do so and also periodically thereafter. This will ensure that the instrument continues to be effective and will also facilitate continuous improvement of the form based on new data.