Objective. To study the association between maternal/infant characteristics and mortality from injury for children 0 through 4 years of age.
Design. Historical cohort.
Setting. State of Tennessee.
Participants. Children 0 through 4 years of age at any time between January 1, 1985 and December 31, 1994. We linked birth certificates and US census data to obtain information on maternal age, race, education, neighborhood income, parity, use of prenatal care, residence location, infant's gender, and gestational age.
Main Outcome Measures. The outcome was death from injury, as determined from linked death certificates. The incidence density rates for each stratum (defined by maternal/child characteristics) were calculated by dividing the number of injury deaths by child years in the stratum. We used multivariate analysis to assess the independent contribution of each characteristic to risk of injury death.
Results. There were 1 035 504 children 0 through 4 years of age who contributed 3 414 436 child years. There were 803 deaths from injury, ie, 23.5 deaths per 100 000 child years. In the multivariate analysis, children had at least a 50% increased risk of injury mortality if they were born to a mother who had less than a high school education (relative risk [RR] = 2.88; 95% confidence interval [CI]: 1.92–4.34) compared with a college education, was <20 years of age (RR = 2.42; 95% CI: 1.76–3.31) compared with >30 years, or had >2 other children (RR = 2.97; 95% CI: 2.29–3.85) compared with no other children. Neither race nor income was significantly associated with childhood injury mortality in the multivariate analysis. Classification of children by maternal education, age, and parity defined a pronounced risk gradient in which high-risk children had an injury mortality rate >15 times that of low-risk children. The steep risk gradient was present for both infants (24-fold increase for high-risk children) and children 1 through 4 years of age (13-fold increase for high-risk children). If the injury mortality rate for all children were equal to that of the low-risk group, 614/803 (76.3%) of injury deaths would not have occurred.
Conclusions. For young children, maternal education, age, and parity are strongly and independently associated with injury mortality. These factors define a steep gradient of risk, suggesting that many injury deaths could be prevented.
Injuries are a serious and persistent public health problem in early childhood. For children 1 through 4 years of age, injuries are the leading cause of death and now account for nearly four times as many deaths as the next leading cause.1 In 1985, there were 112 000 injury hospitalizations and ∼4 million medically treated injuries in children ages 0 through 4 years,2 with estimated direct medical care costs of $1.3 billion.2Although injuries in young children are recognized as a major public health problem,3-5 progress in prevention has been limited. Among children 1 through 4 years of age, injuries accounted for 10% of deaths in 1930, but accounted for half of all deaths by 1985.6 Recent data suggest that mortality rates for some types of injuries in young children in fact may be increasing.7 This lack of progress for injuries is particularly unfortunate, because there is evidence that with appropriate environmental or behavioral modifications, the majority of injuries in young children can be prevented.8
An important element of injury-prevention programs for young children is identification of populations at high risk. Most epidemiologic studies of fatal childhood injuries have focused on infants and have determined that maternal sociodemographic characteristics such as African-American race, low income, low education level, and young age are associated with increased risk.7,9,10 However, whether each of these factors is independently associated with increased injury mortality is unclear, because these sociodemographic characteristics are often highly intercorrelated, and none of the studies has had complete information on all of these factors. Furthermore, even though injury is the leading cause of death in children 1 through 4 years of age, little is known about the predictors of injuries in this age group. Although maternal characteristics are strong predictors of injury deaths in infants, the extent to which these characteristics identify older children at high risk for fatal injuries is unknown.
The objective of our study was to carefully investigate the association between maternal/infant characteristics and mortality from injury for a very large population of children 0 through 4 years of age. We had three specific goals: 1) to identify maternal/infant characteristics that are independent risk factors for mortality from injury, 2) to study risk factors in children ages 1 through 4 years, and 3) to determine whether maternal/infant characteristics define populations of children at very high risk of injury. The study was conducted in the state of Tennessee over the 10-year period between 1985 and 1994, using a linked database with birth and death certificates and US census data.
The study cohort was identified from birth certificates in the state of Tennessee that have been linked with death certificates11 and US census data. Children (live births) were eligible for the study if 1) their mother was a Tennessee resident at the time of the child's birth, 2) they were 0 through 4 years of age at any time between January 1, 1985 and December 31, 1994, and 3) the key study variables were present in the birth and death certificates. There were 1 128 532 children of ages 0 through 4 during the study period. Of these, 86 693 (7.6%) children were excluded because the mother was an out-of-state resident. An additional 6335 (0.6%) children were excluded because of missing data: infant gender (n = 12), singleton/multiple birth type (n = 46), maternal race (n = 117), marital status (n = 308), maternal age (n = 160), parity (n = 426), maternal education (n = 1934), and gestational age (n = 3332). Thus, the final cohort comprised 1 035 504 children.
For each child in the cohort, we calculated the child years during which the child was included in the study. Follow-up began at the later of the date the study began (January 1, 1985) or the date of the child's birth. Follow-up ended at the earliest of the date the study ended (December 31, 1994), the attainment of age 5, or the date of death.
The birth certificates provided most of the information on the characteristics of the study mothers (age at delivery, race, education, previous live births, use of prenatal care, residence [in a standard metropolitan statistical area]) and children (gender, gestational age, and birth type [single or multiple gestation]). We estimated the number of other children the mother had from the number of previous live births, adjusted for nonsingleton current births. If the gestation age was missing or implausible (13% of certificates), it was estimated from the birth weight, using the race- and calendar year-specific distributions of gestational age in the population for whom both gestation age and birth weight were known. A previous validation study that compared 1989 birth certificates with maternal and infant hospital records12 found high reliability for study data elements routinely present in the hospital chart. Although the validity of information on maternal education could not be assessed, because it was not routinely recorded in the hospital charts, concordance was >95% for date of birth, child's gender, birth type, mother's race, mother's marital status, mother's date of birth, number of living children, and county of residence. Concordance was 82% for prenatal care within the first 4 months and 71% for gestational age, defined as agreement within 1 week.
Because the incomes for individual women were not available, we used neighborhood income as a surrogate. This variable was defined as the block group mean annual per capita income obtained from 1990 US census files.13-16 (Tennessee had 4417 block groups in 1990, with a mean of 1104 residents per group.) Census block groups for women in the study were identified by linking the maternal address from birth certificates with the census files. When we could not identify block group, we used mean income of the census tract (23% of study women) or county (5.9%).
Study outcomes included all deaths from injuries that occurred in the cohort during study follow-up. Deaths were identified from death certificates that had been linked to birth certificates.11 These included deaths for residents of Tennessee whose deaths occurred in other states.
Injury deaths were defined as certificates with the cause of death coded between E800 and E999 (ninth revision of the International Classification of Diseases). Although we did not confirm these diagnoses by review of medical records, the broad category of injury as coded on death certificates has been found to have a positive predictive value of between 83% and 90% in infant and young adult populations.17,18
The study design did not capture out-of-state deaths among children who moved from Tennessee during the study period. We obtained an estimate of out-of-state migration by measuring intercounty (within Tennessee) migration for children who suffered a fatal injury. Of these children, 9.8% resided in a different county at the time of death, which suggests the study loss to follow-up (and corresponding underestimate of mortality rates) should not exceed 10%.
The incidence density rates19 for each stratum defined by the maternal/child characteristics were calculated by dividing the number of injury deaths by child years in the stratum. Univariate incidence density rate ratios were calculated by dividing stratum-specific rates. Given that injury deaths are a rare outcome, the incidence density ratios are good approximations of the relative risk.
Because the maternal/child characteristics were highly intercorrelated, we performed a Poisson regression multivariate analysis20to assess the independent contribution of each of the characteristics to risk of injury death. The multivariate analysis, the model for which included terms for each of the characteristics studied, provided adjusted estimates of the incidence density rate ratios. To assess the effect of secular trends, we also constructed models that included terms for calendar year; however, these results were essentially identical to those presented here.
All analyses were conducted with SAS version 6.11 under Windows NT 3.51.
During the study period, 1 035 504 children 0 through 4 years of age contributed 3 414 436 child years of follow-up. There were 803 deaths from injury, or 23.5 deaths per 100 000 child years. This is consistent with the estimated rate for the United States in 1985 of 25 per 100 000 child years.2 The 803 deaths included 228 from motor vehicle crashes, 180 from fires or burns, 75 from drowning, 121 from suffocation or aspiration, 85 from intentional injuries, 18 from poisoning, and 96 from other causes.
The univariate analysis showed injury death rates varied markedly with maternal and child characteristics (Table1). Factors associated with a ≥50% elevation in risk included maternal education <12 years, >2 other children, neighborhood income <$10 000, African-American maternal race, mother unmarried at delivery, prenatal care delayed beyond month 4 of pregnancy, and gestational age <37 weeks.
However, in the multivariate analysis (Table 1), only three characteristics were associated with an increase in risk ≥50%: maternal education, number of other children, and maternal age. There was a strong inverse association between years of maternal education and rates of deaths from injuries (Table 1). Compared with children whose mothers had a college education, those whose mothers had less than a high school education had a relative risk of death from injury of 2.88 (95% confidence interval [CI]: 1.92–4.34). Children whose mothers' education ended at high school or who had some college had lesser, yet significant, elevations in mortality compared with the children of college-educated mothers.
Injury death rates increased with increasing number of other children and with young maternal age (Table 1). Children whose mothers had >2 other children at the time of delivery had 2.97 (95% CI: 2.29–3.85) times the risk of mortality of children whose mothers had no other children. Children whose mothers were <20 years of age at delivery had a relative risk of death from injury of 2.42 (95% CI: 1.76–3.31) compared with children whose mothers were >30 years old.
Because very young mothers cannot have attained high levels of education, we assessed the interaction between maternal education and age. In a separate analysis stratified by maternal age, maternal education maintained an inverse relationship with the childhood injury mortality rates for each of the age strata.
Other characteristics associated with smaller increases in risk of injury mortality were mother unmarried at delivery, non-SMSA (standard metropolitan statistical area) residence, delayed prenatal care, premature birth, and male gender of the child. Although low family income and African-American race were strongly associated with injury mortality in the univariate analysis, these associations were not present after controlling for other factors.
We conducted a separate analysis for infants and children 1 through 4 years of age (Table 2). Infants contributed 698 460 child years of follow-up and had 248 deaths, resulting in a mortality rate of 35.5 injury deaths per 100 000 child years. Children 1 through 4 years of age contributed 2 715 978 child years of follow-up and had 555 injury deaths, resulting in a mortality rate of 20.4 injury deaths per 100 000 child years.
Similar patterns of association between maternal and child characteristics and injury mortality were present for both age groups (Table 2). For both infants and children 1 through 4, maternal education, number of other children, and age were the strongest predictors of injury mortality. Among infants, unmarried maternal status (relative risk [RR] = 1.78; 95% CI: 1.29–2.44) and premature birth (RR = 1.50; 95% CI: 1.08–2.07) also substantially increased injury mortality rates.
To determine whether groups of children at very high risk of injury could be identified, we explored the effect of the presence of the three characteristics most strongly associated with elevated injury mortality rates. We calculated a risk score based on maternal education (≥16 years, 0 points; 13 to 15 years, 1 point; 12 years, 2 points; <12 years, 3 points), maternal age (>30 years, 0 points; 26 to 30 years, 1 point; 20 to 25 years, 2 points; <20 years, 3 points), and number of other children (none, 0 points, one, 1 point; two, 2 points; three or more, 3 points). Children were separated into seven groups based on the total of this score. Because children with risk scores from 0 to 2 had nearly indistinguishable mortality rates, they were combined to form the low-risk group. Children with scores of >7 comprised the high-risk group.
There was a steep gradient of increased rates of injury deaths with increased risk score. For children of all ages (Fig1), the death rate increased from 5.6 per 100 000 child years for children with a risk score ≤2 to 85.9 per 100 000 for children with a risk score of >7, a 15-fold increase. Children in the upper quartile for risk score (score ≥6) accounted for half of all injury deaths. Although a steep gradient was present for both infants (Fig 2) and children of ages 1 through 4 years (Fig 3), the gradient was most pronounced for infants; those with scores of >7 had an injury mortality rate that was 24 times the rate of infants with scores ≤2.
We calculated the number of potentially preventable injury deaths in the study population by assuming that mortality rate for all children could be reduced to that of the low-risk children (injury risk score ≤2). Under this assumption, 614/803 (76.3%) of injury-related deaths were preventable. Of the excess deaths, 29% were attributable to fires/burns, 20% to motor vehicle trauma, 17% to suffocation or aspiration, 12% to intentional injuries, 10% to drowning, and 12% to other causes.
We identified three maternal characteristics that were strongly and independently associated with increased risk of mortality from injury during early childhood: low education, young age, and increased number of children. The association was similar for both infants and for children 1 through 4 years of age. When combined, these factors defined populations at very high risk: young children whose mothers had <12 years of education, were <20 years of age, and had >2 other children had a rate of injury mortality 15 times higher than for children whose mothers were college graduates, >30 years of age, and who had <3 other children.
The primary limitation of our study was the inability to measure potentially important environmental and behavioral factors associated with injury risk, such as use of child vehicle restraints,21 type of housing,22,23 use of smoke detectors,22 parental supervision,5 and family dynamics.24,25 It is likely that the maternal characteristics we identified are surrogates for one or more of these unmeasured factors. For example, the strong association between injury mortality and number of other children probably reflects the effects of parental supervision. However, because it is time-consuming and expensive to measure these environmental and behavioral factors, these would be difficult to include in investigations of sufficient size to study childhood injury mortality. Our findings suggest that readily determined maternal characteristics are convenient surrogate markers for identification of high-risk populations.
Because income is not reported on birth certificates, we and other researchers7,26,27 have used neighborhood income as a surrogate for maternal income. We were able to use 1990 census data to identify mean per capita income in block groups, which are considerably smaller and thus induce less misclassification than the census tracts, zip code regions, or counties used in previous research. The threefold increased risk present in the univariate analysis for children whose mothers resided in low-income neighborhoods confirmed that this variable identifies populations at high risk; interestingly, the multivariate analysis demonstrated this elevated risk was explained by other maternal characteristics. Nevertheless, because income is likely to vary within neighborhoods, it is possible that low individual income is an independent risk factor for overall injury mortality and that some of the effects associated with the maternal characteristics identified in this study are attributable to residual confounding by individual income. It is also possible that income and other demographic factors have independent associations with certain types of injury mortality; additional study of the predictors of specific types of injury is needed.
Another limitation of our study was the inability to measure changes in maternal characteristics over time. This may explain, in part, our finding that the association between maternal education and number of other children was somewhat stronger for deaths in infants than for deaths in children 1 through 4 years of age. However, despite this potential misclassification, maternal factors were strong predictors of injury death risk even after the first year of life.
Our findings are consistent with those of previous investigations of deaths from injuries among infants. Both Emerick and colleagues9 and Wicklund and colleagues10reported an association between low maternal education and increased childhood injury mortality, but these studies lacked data on family income and other maternal factors. Cummings et al7 reported that injury mortality rates were increased when infants were born to mothers who were <20 years old (RR = 9.0), were African-American (RR = 2.4), had >1 child (RR = 5.0), or lived in a rural environment (RR = 2.0). However, maternal education and income data for block groups were not available. Other studies that included postinfancy deaths have reported associations of African-American maternal race26,28 and low income26,27 with increased injury mortality, but these studies lacked information on maternal age, education, number of other children, and other factors.
Managed care organizations that provide care to Medicaid patients include low-income mothers and children whose sociodemographic characteristics are very similar to those of the high-risk children identified in this study.29 If these high-risk populations have a similar elevation in rates of nonfatal injuries, our findings suggest that these organizations could be attractive settings for implementation of injury-prevention programs.
Unlike other disease processes that have etiologies with strong genetic components, current thinking is that variations in childhood injury rates result from the child's physical and psychosocial environment.5,30,31 Many unsafe practices have been identified and can be addressed by relatively straightforward interventions with demonstrated efficacy, such as using child vehicle restraints21 and smoke detectors,22,32 lowering water temperature thermostats,5 avoiding walkers,5 and improving parental education.5Indeed, a recent metaanalysis suggests that home visitation programs combining some of these elements can reduce substantially injury rates in young children.33
Given this, the mortality rate of low-risk populations represents a reasonable goal for all children. The steep socioeconomic gradient in injury risk identified in this study suggests that a substantial fraction of injury deaths in young children are preventable. If the mortality rate for all children were equal to that of the lowest risk group, our data indicate that injury-related deaths could be reduced by >75%. There is a pressing need for effective programs to prevent serious injuries in high-risk populations.
- Received October 28, 1996.
- Accepted December 30, 1996.
Reprint requests to (S.J.S) 5028 Medical Center East, Vanderbilt University Medical Center, Nashville, TN 37232-8555.
- CI =
- confidence interval •
- RR =
- relative risk
- ↵National Center for Health Statistics. Vital Statistics of the United States, 1984. Hyattsville, MD: US Department of Health and Human Services, Public Health Service, National Center for Health Statistics; 1987
- ↵Rice DP, MacKenzie EJ, and associates. Cost of Injury in the United States, A Report to Congress 1989. San Francisco, CA: Institute for Health and Aging; 1989
- ↵Baker SP, O'Neill B, Ginsburg MJ, Li G. The Injury Fact Book. New York, NY: Oxford University Press; 1992:1
- Emerick SJ,
- Foster LR,
- Campbell DT
- Piper JM,
- Ray WA,
- Griffin MR,
- Fought R,
- Daugherty JR,
- Mitchel E Jr.
- Piper JM,
- Mitchel EF Jr.,
- Snowden M,
- Hall C,
- Adams M,
- Taylor P
- ↵Bureau of the Census. Census of Population and Housing, 1990: Summary Tape File 3 on CD-ROM Tennessee (Abstract). Memphis, TN; 1992
- ↵Bureau of the Census. Census of Population and Housing, Summary. Summary Tape File 3 on CD-ROM Technical Documentation (Abstract). Memphis, TN; 1992
- ↵Bureau of the Census. TIGER/Line Files (TM), 1992 (Abstract). Memphis, TN; 1993
- ↵Bureau of the Census. TIGER/Line Files (TM), 1992 Technical Documentation (Abstract). Memphis, TN; 1993
- Moyer LA,
- Boyle CA,
- Pollock DA
- ↵Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic Research Principles and Quantitative Methods. New York, NY: Van Nostrand Reinhold Company, Inc. 1982;1
- ↵McCullagh P, Nelder JA. Generalized Linear Models. New York, NY: Chapman and Hall; 1989:193
- Roberts I
- Taylor B,
- Wadsworth J,
- Butler R
- Wadsworth J,
- Burnell I,
- Taylor B,
- Butler N
- Roberts I,
- Kramer M,
- Suissa S
- Copyright © 1997 American Academy of Pediatrics