The Changing Face of Race: Risk Factors for Neonatal Hyperbilirubinemia

Setting

This prospective cohort study was conducted at the Henry Ford Health System (HFHS) in Detroit, MI. HFHS includes Henry Ford Hospital, Henry Ford Medical Group, Henry Ford Home Care, Henry Ford Information Services, Henry Ford Corporate Data Stores, and the Health Alliance Plan.

Subjects

We used data from the HFHS Corporate Data Store between January 2001 and October 2002 to enroll newborns born at Henry Ford Hospital and to obtain birth weight and TSB tests done during the birth hospitalization. For each newborn, a research assistant manually abstracted from nursing logs in the newborn nurseries the gestational age, feeding type (breast, formula, or combination of the 2), and whether the newborn had a hospital stay of ≥3 days in the NICU. We excluded newborns with birth weight <2000 g, gestational age <35 weeks, those who stayed ≥3 days in an intensive care nursery, and those with TSB ≥10 mg/dL in the first 24 hours of life. The study cohort covers 3197 study newborns.

Medical Record-Documented Maternal Race

We obtained maternal race from the HFHS Corporate Data Store, which is based on information from medical records. Mothers had been assigned a race, with or without their input, using 7 categories: American Indian, Asian, black or African American, Hispanic, Middle Eastern/Arabic, white/Caucasian, and other. The HFHS data system categories for race can be considered modifications of the 2000 Census categories to suit local circumstances. The 2000 Census uses 5 racial categories: American Indian or Alaska Native, Asian, black or African American, Native Hawaiian or other Pacific Islander, and white. The 2000 Census includes those of Middle Eastern origin as “white.” However, in the HFHS data system, “Middle Eastern/Arabic” is identified separately, because of the large Middle Eastern population in Greater Detroit. The category Native Hawaiian or Pacific Islander was not used in the HFHS data system, because few such individuals live in Detroit. In addition, the 2000 Census treats Hispanic ethnicity as a separate question from race.⁷

Survey Data Collection

We designed a computerized telephone survey to elicit information from mothers of study newborns on 5 content areas: breastfeeding experience and education; experience of care received for their newborn; hyperbilirubinemia detection; interventions and education; demographic data; and racial ancestry for mother, father, and newborn.

The racial ancestry section used in this article came at the end of the interview. To elicit mother's responses on racial ancestry we asked the open ended question, “What is your infant's race?” The initial answer was recorded as N1. Mothers were then asked if their newborn was any other race. That response was recorded as N2, and this question was repeated up to 3 more times to allow for racial identification of up to 3 additional categories, coded N3, N4, and N5. Mother's and father's race were similarly elicited generating up to 5 categories for each (M1–M5, and F1–F5).

When asked about their race, some respondents provided an ethnicity or country of origin, most commonly Hispanic or Latino. Although Hispanic is not a race, per se, it was a common response. It is also treated as race in the literature on predicting hyperbilirubinemia in newborns.^10–12 We accepted Hispanic or Latino as a racial identification in addition to any others the respondent might have given. When the respondent provided a nationality or country of origin for their race, we recoded that response to match the 7 categories used in the HFHS data system as follows: nationalities from Europe were recoded as white/Caucasian; nationalities from Spanish-speaking or South American countries were recoded as Hispanic; nationalities from the Caribbean were recoded as Black/African American (unless they came from a Spanish-speaking country); nationalities from sub-Saharan Africa were recoded as Black/African American; and nationalities from Asia, including India, China, and Japan, were recoded as Asians. Nationalities from the Middle East or Arab-speaking countries were classified as Middle Eastern.

We contracted with the National Research Corporation (NRC) for implementation of our surveys. NRC developed a computerized interactive telephone interview, with 7 callbacks, including calls on nights and weekends. NRC followed industry-standard protocols for placing calls and recording data. In addition, several project-specific protocols were followed: NRC used female interviewers only, Spanish-speaking interviewers for Spanish-speaking mothers, and spoke only with the mother identified in the sample (no surrogates).

The HFHS and Harvard School of Public Health Institutional Review Boards for the protection of human subjects approved this study. Mothers with newborns in the study received a prenotification letter in their birth hospitalization discharge packets. The letter informed new mothers of the survey's sponsors and purpose, provided information necessary for informed consent, including their rights to refuse and persons to call with questions, and told them that they might receive a call and that their newborn's medical records might also be confidentially reviewed. Informed consent was then obtained at the start of the NRC interview.

Definitions

Data Analysis

We used Microsoft Access 2000 (Microsoft, Redmond, WA) for clinical database management and SPSS (SPSS, Chicago, IL) for survey data management. We used SPSS software version 10.0 for Windows (SPSS) and Microsoft Excel (Microsoft) for all of the data analyses. We merged the clinical and survey data files using each newborn's study identification number. Using medical record-documented maternal age and race, feeding type, and gestational age of newborn, we compared the mothers who completed the survey to those who did not. Using the same variables, we also compared those contacted to those who could not be contacted, and, of those contacted, those who responded versus those who declined to respond. For these comparisons, we used independent t tests for continuous variables and the χ² test for categorical variables. Differences with a P < 0.05 were considered significant.

We cross-tabulated medical record-documented maternal race with self-reported maternal race. For the 93 newborns with ≥2 races reported by the mother, we cross-tabulated agreement in race for newborn, mother, and father using the mother's first named for each (N1, M1, and F1) to simulate reporting if they had been given only a single option for race. We identified the total number of multiracial infants based on the following: (1) mother defining the newborn, herself, or the father as of ≥2 races; (2) the newborn was the offspring of an interracial union where the parents' races were not the same; or (3) the newborn race first named by the mother was different from the first-named race for the mother or father. For all of the multiracial infants, we cross-tabulated race concordance between the first race of the infants (N1), fathers (F1), and mothers (M1) to look for patterns in racial identification when only a single racial category is available.

Survey Sample and Response Rate

Of a total of 3197 newborns delivered from January 1, 2001, and October 31, 2002, 135 subjects were disqualified and excluded from the survey process because of ≥1 precall exclusion criteria, including reasons such as speaking a language other than English or Spanish and no access to a telephone. Also, in the first few months of surveying, we applied a quota to a random number-ordered list to limit calls, and 41 newborns were excluded, because of exceeding the quota. The mothers of the remaining 3021 newborns qualified for potential inclusion in the survey.

Of those 3021 subjects, 1773 could not be contacted because of a variety of reasons, for example, a nonworking telephone number and a wrong telephone number, no answer or telephone busy after 7 attempts, telephone number given was a business/government telephone, respondent not available, and caller obtained only an answering machine or voice mail. The failure-to-contact rate of 1773 (58.7%) of 3021 indicates that technical and communication factors accounted for a substantial amount of survey noncompletion.

Among the 3021 mothers whom NRC attempted to contact, there was a statistically significant association between failure to contact and mother's race as recorded in the medical record (P < .0001). The most difficult race to contact were Hispanics: we failed to contact 566 (68%) of 835 Hispanic mothers, 840 (57%) of 1469 black mothers, 119 (55%) of 215 “other race” including Middle Eastern mothers, 38 (54%) of 71 Asian mothers, and 210 (49%) of 431 white mothers. There was also a statistically significant difference (P < .0001) between mother's age and failure to contact, with younger mothers harder to contact: the mean age of those not contacted (n = 1747) was 26.2 years, whereas the mean age of those contacted (n = 1228) was 27.1. We also observed statistically significant differences (P = .001) between contact status and feeding type, with partially breastfeeding mothers most difficult to contact: we failed to contact 598 (63%) of 951 mothers who were partially breastfeeding, 632 (60%) of 1060 mothers who were formula feeding, and 448 (54%) of 830 mothers who were breastfeeding only. There was no difference for contact status in association with newborn's gestational age (P = .33).

For the 1248 mothers with whom telephone call contact was made, 866 successfully completed the survey, and 382 subjects refused, postponed, failed to adhere to a callback appointment, or incorrectly assumed that they had already completed a similar article survey. Among those contacted, there was no significant difference between those who did and did not accept the interview on mother's race, age, feeding type, or newborn's gestational age. With 866 subjects completing the interview, the subject-acceptance rate was 69.4%. Overall, less than one third of the cohort sample did not complete the survey because of human and personal-choice factors.

Medical Record-Documented Versus Self-Reported Maternal Race

Medical record-documented maternal race is the race variable most often used to predict the risk of hyperbilirubinemia. In Table 1, we show how the definition of maternal race differs between medical record-documented race using HFHS data system categories and mother self-reported race using the modified Census 2000 categories. Medical record-documented maternal race was missing for 29 of the 866 mothers, and mother self-reported race missing for 23 of the 866 mothers. For numerous mothers in each category of race, medical record-documented race differs from mother self-reported race. Of mothers documented as white in the medical record, only 64% self-reported as white, whereas 13% self-reported as Hispanic, 3% as Middle Eastern, and 11% as ≥2 races. Of 427 mothers documented as black in the medical record, 70% self-reported as black and 23% as ≥2 races. We also found poor agreement between medical record-documented and mother self-reported race in the Asian and Middle Eastern categories (only 35% and 50% agreement, respectively). Table 1 also shows that for the 126 mothers who named themselves as ≥2 races, they were most likely to be documented in the medical record as being black. If given only 1 choice for race, 89% of mothers documented as black in the medical record chose black for themselves. In addition, >70% of Hispanic mothers defined themselves as Hispanic only without reference to a race in addition to their ethnicity (data not shown in Table 1). Detailed, open-ended categories for first-named race, before collapsing categories, are provided in the footnote to Table 1.

Maternal Reported Race of Newborns, Mothers, and Fathers

For the 866 mothers with a completed survey, Table 2 shows the race as reported by mothers for their newborns, themselves, and the newborns' fathers. Of the 866 mothers, 20 (2.3%) did not answer the race question for their newborns, 23 (2.7%) did not answer for themselves, and 31 (3.6%) did not answer for the fathers. Most (83%) mothers reported only 1 race for themselves: 303 (35%) named themselves as black, 245 (28%) as Hispanic, 108 (13%) as white, 28 (3%) as Middle Eastern, 23 (3%) as Asian, 8 (1%) as American Indian, and 2 as other. Paradoxically, although 126 (15%) of mothers reported ≥2 races for themselves, and 77 (9%) of mothers reported that the father was of ≥2 races, only 93 (11%) of mothers reported ≥2 races for their newborn. Based on our analyses of the data, 24% (205) of the newborns had some indication of multiracial ancestry, defined by the mother as ≥2 races, the mother defining herself or the father as of ≥2 races, the newborn was the offspring of 2 parents whose races were not the same, or the newborn first-named race was different from the first-named race for mother or father (newborn N1 ≠ mother M1 or newborn N1 ≠ father F1).

Relationship Between Parental and Newborn Race for the 93 Newborns Reported to be ≥2 Races

Newborns defined by their mothers as ≥2 races are of special interest, because first-named race of mother, which is often used to predict infant risk for hyperbilirubinemia, does not fully represent the newborn's ancestry. We, therefore, examined racial ancestry for the 93 newborns defined by their mothers as of ≥2 races. Table 3 shows the relationship between newborn race using only the first named (N1), mother's race using only the first named (M1), and father's race using only the first named (F1). For 41% of the 93 newborns, the mother reports that newborn's race (N1) matches both that of mother and father, although when additional choices were given, the mother selected ≥2 races for the newborn. For 23% of the 93 newborns, infant race (N1) is reported to match mother's race (M1) only and for 25% to match father's race (F1) only. Overall, for 63% of newborns said to be of ≥2 races, race (N1) is the same as mother's race (M1), which is used to predict risk for hyperbilirubinemia. However, for 11% of these newborns, race (N1) is not equal to either mother's or father's race (M1 or F1). As the footnote to Table 3 shows, this unexpected result relates in some cases to maternal reports that either or both parents are of ≥2 races.

In addition to the 93 newborns defined by their mothers as ≥2 races, there was a total of 205 newborns who we identified as being of multiracial ancestry, that is, mother defined the newborn, herself, or the father as ≥2 races, the first race reported for mother and father was not the same, or the newborn race first named by the mother was different from the first-named race for the mother or father. For 112 (55%) of the “multiracial” newborns, mothers did not describe their newborn as of ≥2 races, despite supplying evidence that could support that definition. More specifically, for the 70 newborns whose parents did not have the same race, only 45 (64%) were reported by their mothers to be >1 race. The 70 mothers in interracial unions assigned father's race to 54% of their 70 newborns and mother's race to 40% and the remainder to neither parent's first-named race. Selection of father's versus mother's race varied by racial category. The majority of newborns named as black or Middle Eastern were assigned to father's race and the majority of newborns named as Hispanic to mother's race. Newborns named as white were equally assigned to father's or mother's race. The general trend was to assign the infant to a nonwhite race, rather than assigning race of either the mother or the father.

Principal Findings

We found differences between medical record race and maternal self-reported race for all of the racial categories. For the 2 largest racial categories among our mothers (blacks and whites), only 70% of mothers documented as black in the medical record defined themselves as black only, and 64% of mothers defined as white in the medical records self-reported as white. In addition, 14% of all of the mothers defined themselves as of ≥2 races, a fact that itself undermines the concept of using a single category of race as a predictor of hyperbilirubinemia. Of the women who were documented in the medical record as black, 23% self-identified as being >1 race. Depending on their ancestry, their infants' risk for neonatal jaundice may be severely underestimated if their clinicians view them as black without additional ancestry.

We explored racial ancestry among newborns defined by their mothers as of ≥2 races, a group for whom categorization into a single race seemed obviously problematic. We discerned no consistent decision rule that mothers seemed to follow within these groups in assigning newborn race if given only 1 choice. Of 93 newborns reported as of ≥2 races, for 41%, both parents were assigned the same race, and this race was also assigned to the newborn. However, 23% of the 93 newborns were assigned to mother's race and 25% to father's race, with 11% assigned to the race of neither mother nor father.

Also, the 70 mothers who reported interracial unions assigned father's race to 54% of their 70 newborns, mother's race to 40%, and the remainder to neither parent's first-named race. The general trend was to assign the infant to a nonwhite race rather than assigning race of either the mother or the father. When given >1 choice for race, only 45 (64%) of these 70 mothers identified their infant as >1 race.

Strengths and Weaknesses of Our Study

We conducted a detailed exploration of racial ancestry among newborns in the context of neonatal hyperbilirubinemia. Our study population had a higher rate of interracial unions (8%) than is currently reported in the overall US population (4%). We, therefore, had a large group of multiracial newborns for study, which allowed us to analyze their parents' patterns of racial identification. Our study reflects the population balance and pattern of interracial unions in a particular city: none of our mothers were Pacific Islanders or Alaska Natives, and few were Asian, thus limiting the generalizability of our findings.

Less than one third of the initial cohort sample did not complete the survey, and the respondents we failed to contact were more likely to be Hispanic, younger, and to partially breastfeed their infants. The demographic and infant feeding factors are not likely to introduce biases into the study, because we would not expect an association between those factors and racial misclassification. However, one of the exclusion criteria included speaking a language other than Spanish or English. This could potentially remove respondents who were Asian or from French-speaking or Portuguese-speaking countries in the Caribbean and South America. This further limits the applicability of our findings of racial misclassification in these groups.

A limitation of the study was the lack of consistent categories for classifying race/ethnicity data between medical records and self-reporting. Most notably, the medical record does not usually include multiracial categories for patients. This would have the greatest impact on black respondents, who were most likely to report being >1 race. Had there been only a single racial category available for self-report, then 89% of mothers documented as black in the medical record would have self-reported as black.

We did not have information on how measures of race/ethnicity are collected at HFHS. Hospitals have reported collecting these data using a variety of methods that include observation by hospital staff including physicians, nurses, and admitting staff, as well as asking the patients in a variety of settings, such as admitting, the emergency department, and the outpatient clinics.¹⁶ Without knowing the methods used to collect these data and whether they are consistently applied, we are limited in saying whether the accuracy of race measures in this setting is generalizable to other settings.

Strengths and Weaknesses in Relationship to Other Studies

A study by Setia et al¹⁷ supports our concern that mother's race may lead to erroneous estimates of newborn risk for hyperbilirubinemia. They compared diagnoses of hyperbilirubinemia among newborns classified in birth certificate and hospital discharge data as being Asian, white, or Asian combined with white parentage. They found that newborns with a white mother and Asian father had a 32% greater risk than those of white parentage only, whereas no increased risk was observed among newborns with Asian mothers and white fathers. Newborns of mixed white and Asian parentage were at lesser risk than those of Asian parentage only.¹⁷ In other words, if risk for hyperbilirubinemia had been assessed among these interracial unions in terms of mother's race, risk would have been overestimated for Asian mothers and underestimated for white mothers.

Meaning of the Study: Possible Mechanism and Implications for Clinicians or Policymakers

The use of maternal race as a risk factor arose at a time when acknowledged interracial unions were rare, and mother's race was assumed to be a good surrogate for newborn's race. Also, mother's race is more likely to be available in clinical data sources than newborn's race, because newborns in many US hospitals do not have their own medical records or hospital discharge abstract data. Within the context of hyperbilirubinemia, race serves as a gross indicator of underlying genetic polymorphisms that cause jaundice. Some, such as G6PD deficiency and Gilbert's syndrome, are well known, whereas there may be others that have yet to be described. Until universal screening for the actual genetic causes of hyperbilrubinemia is part of medical practice, race will continue to serve as a proxy, albeit inexact, for risk of hyperbilirubinemia.

Our study demonstrates the lack of accuracy in maternal race as recorded in the medical record and undermines the concept of maternal race as single category, as well as its use for predicting infant race and risk for hyperbilirubinemia. For example, a clinician who perceives a newborn as black, without knowledge of its American Indian or Asian heritage, would underestimate the newborn's risk of hyperbilirubinemia.

Most of our respondents reported a single race; however, there is a significant and growing number of people who consider themselves to be >1 race. When ancestry is important, as in the case of hyperbilirubinemia, then having a single category to respond to a query about race is insufficient. Even with multiple responses available, we found that less than half of the infants who could be considered multiracial were identified by their mothers as being so. In addition, we could draw no conclusions to predict who would or would not be identified as multiracial or, if given only a single choice, what race a parent might select.

The only pattern that did emerge is that, when given the option, women who were documented in the medical records as black were most likely to report being ≥2 races. We also noted that among infants of interracial unions whose parents reported them as being only 1 race, they were more likely to be reported a race other than white. These 2 observations seem to be consistent with practices of the 1-drop rule used previously in US racial categorization and racial politics.

Unanswered Questions and Future Research

Our study provides a first look at the issue of multiracial heritage, which limits use of maternal race as a risk factor in predicting hyperbilirubinemia. To provide guidance to clinicians for our increasingly multiracial future, much larger studies in more varied populations will be necessary. In addition, the study provides a direct comparison of various methods for collecting race and ethnicity data, based on an assumption that self-reporting is best, but there are different methods for eliciting a self-report of race.

The importance of race in predicting health outcomes is well described, particularly in the health disparities literature. In most cases, race is a social construct that relates to culture, behaviors, and how one is perceived and treated in society. Although these are important social predictors of health outcomes, there are some cases in medicine where race is an independent predictor of risk, unrelated to social constructs; hyperbilirubinemia is one such case. Whether one is measuring race for social or biological risk, the lack of rigor in measuring race has been criticized recently, and there is a need to establish standards for measuring race.³

Self-report for race is likely the most valid method for obtaining this measure. How that is best achieved in clinical settings has yet to be determined. What is not clear is how deeply one needs to probe into ancestry to accurately measure race. In the case of neonatal jaundice, it seems that probing to grandparentage may be appropriate. However, when race is being measured as a social construct, ancestral lineage may be less important than how one is perceived or perceives oneself in society.