The impact of low-level lead toxicity on school performance among children in the Chicago Public Schools: a population-based retrospective cohort study

This research is a collaborative effort of the Chicago Department of Public Health
(CDPH), CPS and the University of Illinois at Chicago, School of Public Health (SPH).
Existing data sets were linked to examine the association between early childhood
B-Pbs, socio-demographic data, and school performance as measured by ISAT scores.
All aspects of this study, including data linkage, analysis, and presentation of results,
were conducted according to a protocol approved by the Institutional Review Board
of the University of Illinois at Chicago.

Inclusion criteria and study design

The study population is a cohort of children who were: born in the six-county area
of metropolitan Chicago between 1994 and 1998; residents of Chicago during early childhood;
had a B-Pb test reported to the CDPH between 1996 and 2006; and were enrolled in a
CPS school with available ISAT scores between 2003 and 2006. There were 58,560 children
who met these inclusion criteria, representing 39% of the 149,768 children enrolled
in CPS in the third grade. We found no significant differences in the characteristics
of children in the study compared with those enrolled in Chicago Public Schools, p??0.0001
level, except that Hispanic children were under-represented in the study sample. This
is likely because Hispanic children are much more likely to have immigrated to the
US and do not have available birth registry data, thus making them ineligible for
the study. We also excluded 1,136 children who were missing race/ethnicity. An additional
74 children had missing data for all of the outcome variables, resulting in a final
sample size of 57,350 children. We used this full data set to investigate the shape
of the dose–response relationship between B-Pbs and ISAT math and reading failure
rates across the entire range of B-Pbs. Because the primary focus of this study, however,
was to investigate the effects of low-level lead exposure on school performance, we
restricted most analyses to the 47,168 children who had B-Pbs less than 10 ?g /dL.

Datasets

The dataset for this study was created by linking three administrative databases:
the Chicago Blood Lead Surveillance Program, the Chicago Birth Registry, and CPS performance
records.

The blood lead surveillance dataset

Under the Illinois Lead Poisoning Prevention Act, laboratories are required to report
the results of blood lead concentration tests for children 72 months or younger residing
in Illinois to the Illinois Department of Public Health (IDPH). The IDPH shares the
blood lead concentration tests with the CDPH Childhood Lead Poisoning Prevention Program
(Chicago CLPPP). The Chicago CLPPP database of B-Pbs for children living in Chicago
from 1996 to the present includes the following information: identification (child’s
name, sex, date of birth, race, ethnicity, and, more rarely, SSN, Medicaid ID, or
guardian’s name,), contact information (address, unit, zip code, and, more rarely,
phone number) and test information (physician, laboratory, date of sample, analysis,
age at the time of the test, type of test, and result). The Chicago CLPPP promoted
venous testing and required follow-up testing for elevated capillary tests.

Blood lead venous samples were analyzed by the IDPH lab and by private laboratories
using either ICPMS or atomic absorption spectroscopy. The Chicago Blood Lead Surveillance
System (CBLS) collects and manages child-specific data with unique child, lab test
result and address fields.

The birth registry

We linked the B-Pb database with Chicago Vital Statistics to account for socio-demographic
factors that might influence school performance. The CDPH Vital Statistics section
manages birth certificate data that includes child’s date of birth, weight at birth,
race/ethnicity, APGAR scores (a method of assessing the health of newborn children
based on Appearance, Pulse, Grimace, Activity, Respiration), parental race/ethnicity
and ages, maternal census tract, birthplace and maternal report of gestational smoking
and alcohol intake, birth order, information on prenatal care, and clinical age of
gestation.

CPS school performance records

We used ISAT reading and math scores as a measure of school performance. ISAT data
is collected as part of the Chicago Public Schools Student Information System (CPSIS),
which tracks the academic progress of each CPS student. That data source includes
the following data: identifying information (name, date of birth, sex, race, and ethnicity),
contact information (address, unit, zip code, and phone number), Independent Education
Plan (yes/no), Special Education (yes/no) and school performance information (reading,
math, and composite standardized test scores).

Data linking protocol

Linkage of these three datasets was performed by Chapin Hall, Center for Children
at the University of Chicago, which maintains a data sharing and data analysis agreement
with CDPH. Chapin Hall received B-Pb and birth registry datasets from CDPH. Chapin
Hall matched these files to CPS data by first name, last name and date of birth using
probabilistic matching methods. Exact matches were taken first, and then a SAS function
called SPEDIS was used to conduct “fuzzy matching” that provides a numeric value for
distance between matches and creates a list of probable matches. The probability of
commonly found names was included in the matching protocol. The list of possible matches
was then verified manually. Finally, we replaced all identifying information, including
names and dates of birth which were used to link the files, with unique coded identifiers
and deleted all individual identifiers.

Exposure measures

Blood lead tests were used as a measure of childhood lead exposure. Most children
in the study had only one recorded B-Pb in the matched blood lead surveillance dataset.
For children with more than one blood lead test, the most recent venous blood lead
concentration test was used (children with multiple B-Pbs accounted for less than
10% of the sample). Most children in the study (88%) had venous test results. For
the purposes of this study, results included all B-Pbs (capillary and venous) as reported
in the dataset. In secondary analyses, we examined the results using only venous B-Pbs
and there were no appreciable differences in the results; therefore, all B-Pb tests
were used.

Outcome measures

The main outcome variable for this study was 3^rd grade ISAT scores, a measure of individual student achievement relative to the Illinois
Learning Standard. The State of Illinois also uses the results to report student achievement
to the public. The ISAT measures the achievement of students for reading and mathematics
in grades three through eight and for science in grades four and seven. ISAT results
used were for the years 2003 through 2006. The 2006 ISAT had a grade adjusted scale,
allowing students from 3^rd through 8^th grade to be measured on the same underlying scale 8]. Additionally, the failure rate for reading and math tests was used as an outcome
variable. Each year the reading and math test scores are divided into four categories:
academic warning (failure), below standards, meets standards and exceeds standards.
Children who get a score in the academic warning (failure) category will be retained
and required to repeat the third grade unless they are retested and score higher.

Covariates

The covariates used in this study (Table 1) were shown in other studies to be associated with academic achievement, including
gender 9], poverty 10],11], race/ethnicity 12], maternal education 13], and the incidence of very low birth weight or preterm birth 14]. Gender was included as a potential confounder because previous literature has indicated
that girls have higher performance than boys on standardized reading tests and perform
slightly better on standardized math tests 9].

Table 1. Demographic characteristics, mean blood lead concentrations, reading and math ISAT
scores of Chicago cohort

Enrollment in the free or reduced-price lunch program was used as an indicator of
family income to control for poverty and as a proxy for the income of the child’s
family. Eligibility for the free or reduced-price lunch program, which is administered
by the USDA, is based on income data submitted to CPS each year, and ranges from 1.3
to 1.85 times the federal poverty rate (approximately equivalent to an annual income
of less than $45,000 for a family of four).

The race/ethnicity variable (data taken from the birth registry) includes Hispanic
sub-populations. The distribution of race/ethnicity in the study sample is representative
of the CPS student body, approximately two-thirds of which are African American, and
one-quarter of which is Hispanic (Table 1). Analyses were stratified by race/ethnicity to test for effect modification.

The maternal education variable (data taken from the birth registry) measures the
highest completed level of maternal education at the time of the child’s birth. The
maternal education variable was coded into five categories (some high school, high
school graduate, some college, college graduate and post college education), for compatibility
with previous pediatric lead studies 4].

The incidence of very low birth weight or early preterm birth has previously been
found to be the strongest birth outcome predictor of academic performance 14]. We created a dichotomous variable for the birth weight and preterm delivery based
on data from the birth registry. We combined children who either had a very low birth
weight (=1.5 kg) or were very premature (gestational age ?33 weeks) in one category
for this analysis 14]. We adjusted for child’s age at the time of the blood lead test to account for age-related
differences in blood lead concentrations. Children who had higher B-Pbs at 4 or 5 years
of age usually had had higher B-Pbs at age 2 or 3 years of age. The typical peak exposure
for children is 2 to 3 years of age, so a higher B-Pb at age 4 or 5 is usually indicative
of a higher exposure over early childhood 1]. The mean age of the children at the time of the blood lead test is 45 months. Because
a blood lead test is required for preschool, HeadStart, and kindergarten entrance
in Chicago schools, children are typically tested as part of the enrollment process
in an early education program. All of these blood lead tests occurred prior to the
ISAT test, which typically occurs in the late spring of the child’s third grade year,
between the child’s ninth and tenth birthday.

Methods of analysis

Standard descriptive methods, multivariable linear regression, and log binomial regression
were used to determine whether there is an association between childhood B-Pbs and
school performance. Among the sample of children with B-Pb of 10 ?g/dL (n?=?47,168),
we examined study variables using descriptive methods. Unadjusted analyses were conducted
to determine the relationship between covariates and exposure and outcome variables
using standard t-tests, analysis of variance, and chi-squares to test for differences
in means or proportions between groups. We constructed multivariable linear regression
models to determine the effect of B-Pbs??10 ?g/dL on reading and on math ISAT scores,
both of which were normally distributed dependent variables. We first treated B-Pb
as log-transformed B-Pb and secondly as a set of indicator variables representing
each B-Pb, consistent with previous studies 1],4],8]. The models were compared through several test statistics (including the adjusted
R² and the root mean square error (RMSE)). The results of both the log-transformed B-PB
analysis and the indicator variable version of B-Pb did not provide a better model
fit than simply using untransformed B-Pb for lead exposure levels of??10 ?g/dL. Regardless
of how we defined B-Pb, the estimated decline in ISAT scores was comparable. Therefore,
the untransformed version was used for ease of interpretation. Next, we examined interaction
terms to test for effect modification. The interaction terms between B-Pbs and race/ethnicity
were found to be significant (p??0.0001). Therefore, we stratified by race/ethnicity
for all racial groups with sufficient sample sizes (non-Hispanic whites, non-Hispanic
blacks, and Hispanics), and ran multivariable linear regression models that adjusted
for all of the covariates mentioned previously.

School performance in reading and math was also measured using a dichotomous pass/fail
outcome variable that was created by grouping those exceeding, meeting or falling
below the standard as passing and those who failed the standard as failing. Multivariable
log binomial regression models were used to produce relative risks and 95% confidence
intervals for the relationship between B-Pb and failing vs. passing math and reading
for the sample of children with B-Pbs 10 ?g/dL.

To estimate the percentage of reading and math failure attributable to lead exposure
of B-Pbs 5–9 vs. 0–4 ?g/dL, we calculated the adjusted population attributable risk
percent (PAR%), using the following equation:

where p_c is the prevalence of B-Pbs between 5–9 ?g/dL among those who failed the reading and
math test, respectively and adjRR is the adjusted relative risk from multivariable
log binomial regression models that treated B-Pb as a dichotomous variable (5–9 vs.
0–4 ?g/dL) and controlled for all covariates.

Finally, we repeated log binomial regression models for reading and math failure for
the larger sample of children across the whole spectrum of B-Pbs, this time using
the log transformed B-Pb, which provided a better fit to the data for the full range
of B-Pbs, ranging from non-detectable levels to 120 ?g/dL. Unadjusted and adjusted
probabilities of failure for each B-Pb were estimated from a crude and multivariable
model, respectively, and plotted to examine whether the slope of the relationship
is different for B-Pb values??10 ?g/dL vs. ? 10 ?g/dL. Adjusted probabilities were
produced at the mean values of the covariates. An interaction term between the continuous
B-Pb value and an indicator for values??10 ?g/dL vs. ? 10 ?g/dL was added to this
model to formally test the significance of a change in slope at 10 ?g/dL. Although
we used the full range of B-Pbs for models, we truncated the figures at 40 ?g/dL due
to high variance in unadjusted probabilities of failure for higher B-Pbs. SAS version
9.2 was used for all analyses 15].