Methods

Linked birth and infant death certificate data from the National Center for Health Statistics (NCHS) were analyzed for the nearly 35 million US live births occurring from 1983–91. Methods of data linkage and file characteristics as produced by NCHS are reported elsewhere.¹³

Relative risks of infant injury death (age <1 year) based on injury fatality rates are compared for the unintentional, undetermined, and intentional injury groups by a variety of maternal and infant characteristics. All live births and infant deaths are kept in the analysis to determine rates, relative risks, and assessment of large enough numbers for stable rates.

Characteristics analyzed include mother's age, race, education, and marital status; trimester of pregnancy prenatal care began; gestational age; and live birth order. Variable selection is based on statistical methods described in previous analyses of high risk factors for both intentional and unintentional injuries using the same data files.^{2, 3} Variable selection is limited to items available on the certificates. Many relevant social indicators, such as parental occupation or family structure, are not present from these data sources. For fatalities, the high per cent of missing data on father's characteristics (for example, age 33%, race 30%), necessitated the omission of father's risk factors from the analysis.

Gestational age and birth weight are highly correlated, with similar relative risks, so only gestational age at birth is kept in tables because of implications for timing of interventions during the prenatal period. For example, there are targeted interventions that after 28 weeks of gestation would miss the opportunity for prenatal interventions for infants born before that time. In addition, premature infants who are more susceptible to health problems may put additional stress on the family.

Determination of intent is based on official designation of the underlying cause of death on death certificates, using the International Classification of Diseases, 9th revision (ICD-9), for external causes of death (E codes).¹⁴ Deaths considered unintentional are classified under E codes 800–949, and as intentional under E codes 960–969 (homicide and injury purposely inflicted by other persons). Deaths of undetermined intent are classified under E codes 980–989 (injury undetermined whether accidentally or purposely inflicted).

Results

Of 10 370 total injury deaths, 22.6% are reported as intentional (n=2345), 4.2% as undetermined intent (n=431), and 73.2% as unintentional (n=7594) (table 1). Relative risks of death are consistently higher across all intent categories for infants of mothers with the least education, no prenatal care, young maternal age, and single marital status. The same is true for infants who are third or later born, preterm, black, or American Indian.

Fatalities with undetermined intent have larger relative risks in the highest risk categories than either intentional or unintentional injuries for almost all variables examined. Only for birth order as third or later born is the relative risk for unintentional greater than either intentional or undetermined.

In addition, the risk profile for deaths with undetermined intent more closely resembled profiles for intentional deaths than unintentional. Among deaths with undetermined intent, the risks for infants born to mothers with less than 12 years of education are 9.4 times greater than for infants of mothers with ≥16 years whereas relative risks for intentional injuries and unintentional injuries are 8.3 and 5.8, respectively. Among infants born to mothers without prenatal care, the risks for undetermined intent are 10.3 times greater than for mothers obtaining care in the first trimester. The corresponding relative risks for intentional injuries and unintentional injuries are 8.7 and 4.2, respectively.

More recent data from linked files for 1995 were reviewed to verify these findings. (Linked files for 1992–94 are not available and because of the gap, 1995 data were not included in our study.) The same general trends were shown within broad categories but the number of fatalities were too small for the highest risk categories in cases with undetermined intent. Only 45 total cases were classified with undetermined intent in 1995 resulting in similar but statistically unreliable rates estimates.

Discussion

The striking similarity of levels of risk for suspicious deaths and deaths known to be intentional, suggests that many undetermined deaths may actually be homicides. This assessment agrees with results of earlier studies of under counting of homicides based on careful review of records from multiple sources.^6–9

Even if deaths reported as undetermined for intent were combined with those reported as intentional, the resulting numbers are likely to underestimate the true magnitude of the problem. Reviews of records from multiple sources, after the cause of death had been determined by a medical examiner, showed that substantiated child abuse or neglect may be involved in 7% to 27% of injury deaths reported as unintentional.^6–8 At least 5% of deaths classified as sudden infant death syndrome may be due to child abuse and neglect.^{6, 9}

Another death certificate classification that may include some intentional deaths is “deaths of unknown cause” or ICD code 799. This group includes deaths where no cause could be found or determination of cause was still pending when the database was closed for compilation of statistics (n=6446 for infants in 1983–91). Cases with legal findings still to be reported may be included in this group. Unfortunately, it is not possible to separate these out. Further comparison of specific country practices for determining intent and allocation of ICD code 799 for pending cases and unknown cause of death will enhance our ability to compare injury rates internationally.⁵

Better dissemination of results of child death review investigations of circumstances around fatalities also should improve the official reporting and description of risk factors and thereby facilitate development of better interventions.¹⁵ To improve quality of certificate data sources, physicians, medical examiners, and coroners should be encouraged to promptly file amended death certificates in cases where the cause of death was initially unknown or was changed after further investigation. Otherwise, these changes will not be reflected in available health statistics data.

This analysis of intentionality classification presented separate descriptors of characteristics such as race, maternal age, and marital status. No attempt was made to perform multilevel categorizations of variable cross classifications by intent because of very small numbers in the category strata for undetermined intent which represented 4.2% of all injury fatalities. An earlier multivariate analysis of predictive risk factors for homicides showed that effects of race were diminished, but still significant, when other social factors were accounted for.³ Unfortunately the small numbers and limited variables involved in this study do not permit a more in depth multivariate analysis of race and social status interactions.

It is possible that the assignment of intentional or undetermined intent by medical examiners or coroners is biased toward the higher risk factor profiles shown in this analysis. Another study found that professional health workers may be more likely to identify and report child abuse in lower class and black families than in other families with about half of serious abuse cases unreported.¹⁶ Similar risk factor profiles may cause the certifiers to question intent for lower socioeconomic groups more often than in cases occurring in middle or upper class families. Such considerations may result in some false positive identification of homicide or suspicious deaths consistent with the higher risk factors in this study. Similarly, under-reporting (false negatives) may be prevalent in the higher socioeconomic groups. The effect of these biases on this analysis would be to further the misclassification of fatalities due to violence.

During 1983–91, 2345 infants were reported as intentionally killed and 431 died under suspicious circumstances—almost one a day. Deaths of undetermined intent are likely to represent intentional injuries. In analyses of infant homicide, excluding deaths of undetermined intent or merging them with unintentional injuries may lead to an underestimation of the magnitude of the public health problem of intentional injuries in infants.

Combining deaths reported as intentional and undetermined for intent provides more accurate estimates of the magnitude of violent deaths among infants and young children. The 1983–91 combined rate of intentional and suspicious fatalities is 8.0 per 100 000 births or 26.8% of all infant injury deaths compared with the reported rate of 6.7 per 100 000 (22.6%). Our findings suggest that estimates of infant homicide that incorporate deaths with undetermined (suspicious) intent on birth certificates could be almost 20% greater than reported rates of homicide.

We need better risk profiles to effectively target preventive services to the families where carefully planned interventions are most likely to be effective. Neither accurate risk profiles nor carefully planned interventions can be produced without linkage of information on families and fatality circumstances from social services, police investigators, health care providers, and medical examiners or coroners. Epidemiological studies are needed to describe the interactions among the effects of family structure, parental social support, alcohol, drug abuse, and other contributors. Lack of data and poor specificity in risk factors for violence against young children force us to rely on the crude measures available from existing information systems.

Birth certificates are limited to biological measures such as maternal age or infant's birth order, gestational age, and birth weight. In the US social measures such as parental education or race are available, but more pertinent information on family structure or social support, including parental occupation, are not. These social measures may be markers for, and indicative of, the lack of readily available caretaking and parenting resources that increase the risks of violence among infants and children. But without better information on specific family circumstances, documentation of increased risks for homicide according to the measures available may be the first step toward targeting interventions toward the highest risk groups. At the same time, the many homicides that occur among lower risk social groups may be neglected because of stereotyping. Both social service agencies and child fatality review teams need to organize and support improved information systems to facilitate prevention programs with appropriately targeted interventions.^{15, 17, 18}