RESEARCH

• General
• Screening & Diagnosis
• Prevention Measures
• Treatment Techniques & Results
• Environment & Product Monitoring
• National Policy,Standards, Legislation & Enforcement
• Studies & Findings
• Country Projects & Programs
• New Ideas & Recommendations
• Publications
• Unpublished Work

Studies & Findings

IMPACT OF LEAD ON BEHAVIOR AND

LEARNING OF CHILDREN

Dr. Herbert Needleman

Introduction

Childhood lead poisoning evolved through four phases of understanding and myth:

1. Long held belief that there was no such thing.
2. Once excepting that childhood lead poisoning did exist, a belief that there were only two possible outcomes – death or complete recovery.
3. Lead poisoning occurs only in symptomatic children.
4. Silent lead exposure has long term consequences and IQ is not the most important target.

Lead has adverse neurological effects, with decrease in IQ and reading ability in young adulthood. Reading disability due to poisoning was two or more grades below expected. Long term exposure of this metal results not only in school failure but also reduces life success.

Affected children are hyperactive and inattentive. The study by Wilson and Herrsteins show that the criminal behavior is of early onset, predominant in males, with aggressiveness, decreased IQ, and hyperactivity. The aggression, delinquency and attention were studied on children by both teachers and parents and were found to be increased in lead affected children.

The study on delinquency has shown that it is of two types:

1. Early onset/life persistent – where it shows high prevalence of neurological dysfunction in 6% of population committing 50% of the crime.
2. Late onset/transient – where there is no increase in neuropsychological dysfunction.

Increase in lead decreases the SAT scores and 8^th grade scores. Twelve studies related to lead levels and IQ were examined by Dr. Needleman and Dr. Constantine A. Gatsonis, and the results are presented below in this paper.

Low-Level Lead Exposure and the IQ of Children:

A Meta-analysis of Modern Studies

We identified 24 modern studies of childhood exposures to lead in relation to IQ. From this population, 12 that employed multiple regression analysis with IQ as the dependent variable and lead as the main effect and that controlled for nonlead covariates were selected for a quantitative, integrated review or meta-analysis. The studies were grouped according to type of tissue analyzed for lead. There were 7 blood and 5 tooth lead studies. Within each group, we obtained joint P values by two different methods and average effect sizes as measured by the partial correlation coefficients. We also investigated the sensitivity of the results to any single study. The sample sizes ranged from 75 to 724. The sign of the regression coefficient for lead was negative in 11 of 12 studies. The negative partial r's for lead ranged from - .27 to - .003. The power to find an effect was limited, below 0.6 in 7 of 12 studies. The joint P values for the blood lead studies were < .0001 for both methods of analysis (95% confidence interval for group partial r, -.15 ⁺ .05), while for the tooth lead studies they were .0005 and .004, respectively (95% confidence interval for group partial r, -.08⁺.05). The hypothesis that lead impairs children’s IQ at low dose is strongly supported by this quantitative review. The effect is robust to the impact of any single study.

The neurotoxic properties of lead at high doses have been recognized for at least a century and are not a matter of dispute. In 1943, Byers and Lord ¹ first suggested that childhood exposure to doses of lead insufficient to produce clinical encephalopathy was associated with deficits in psychological function. The question of low-level lead exposure has been studied widely over the past two decades and, in contrast to high-dose lead exposure, has been the source of considerable contention. Several methodological difficulties encountered in the conduct of these studies have contributed to the controversy. Among them are:

1. selecting adequate markers of exposure or internal dose,
2. measuring outcome with instruments of adequate sensitivity,
3. identifying, measuring, and controlling for factors that might confound the lead effect,
4. recruiting and testing a sample large enough to provide adequate statistical power to detect a small effect, and (5) designing a study that avoids biases in sample selection.

A number of reviews of studies on the effects of low-level lead exposure on the neuropsychological function of children have been published.^2-5 The outcome of major focus in these reviews has been psychometric intelligence. The general approach was to provide narrative summaries in which the epidemiologic and statistical issues often received limited critical attention. Where quantitative synthesis was attempted, it consisted of a simple tally of those studies showing statistically significant effects (at the .05 level) vs those that did not. This approach gives undue emphasis to the individual study's P value and attaches equal weight to all studies without regard to their specific merits or flaws. The size of the effect measured in each study is generally ignored in the process.

The statistical techniques that have been subsumed recently under the rubric of meta-analysis offer a framework within which formal research synthesis can be conducted with more clearly defined methods and criteria.^6-9 In this approach, individual studies are treated as data points in a larger "meta-study." Summary measures from each study are pooled by one of a number of techniques, and quantitative inferences are drawn about the research questions of interest. The difficulties entailed in combining dissimilar studies ("apples and oranges") is a concern for any meta-analysis.^6,8 It points to the need for some measure of commonality in the studies that are being combined. At the same time, the usefulness and novelty of meta-analysis lies in the fact that it enables the investigator to combine the results of studies that differ in several respects, while addressing the same research questions.

The first meta-analysis of six lead-IQ studies was reported by Schwartz et al¹⁰ in 1985. They used Fisher's aggregation technique to calculate a joint P value of .004 for the effect of lead on IQ in the six studies. Needleman and Bellinger¹¹ extended the analysis by Schwartz et al and also used Fisher's technique on pooled tooth and blood lead studies.

In the last few years, a substantial number of new epidemiologic studies from various nations, using more refined designs, larger sample sizes, and more sophisticated statistical techniques, have been reported. This presents an opportunity for a more comprehensive meta-analysis. Herein, 12 recent studies are reviewed and a quantitative synthesis of their results is presented. The major outcome of interest is full-scale IQ, although many studies also examined the effects of lead exposure on important functions such as school performance, reading ability, and classroom behavior. All studies reviewed employed multiple regression analysis in which the dependent variable (IQ) was treated as continuous. Lead exposure was classified by one of two methods: blood or tooth lead level. In contrast to earlier attempts, this analysis divides the studies by tissue analyzed and combines inferences within tissue groups. The question of possible bias in the obtained sample of studies (known as the "file drawer" problem) is addressed. Moreover, the aggregate effect of the exclusion criteria is assessed by performing an analysis that combines all of the initial 24 studies. The sensitivity of the results of this meta-analysis is further investigated by eliminating each of the included studies, one at a time, from the analysis and observing how this affects the conclusions. The statistical power of each study to find an effect is also computed.

This article presents a discussion of some methodological difficulties encountered in the studies reviewed, examines the critical question of effect assessment in pollutant studies, and concludes with comments on the difficulties entailed in drawing causal inferences from observational studies of lead exposure and intellectual development.

Table 1.--Candidate Studies for Meta-analysis*

Study	Year	No. of Subjects	Tissue	Lead Level †	Data Analysis	Included/Reason for Exclusion	Comments	Lead Effect,P<.05
Kotok12	1972	C=25; E=24	Blood	C=38; E=81	t test	No/G,D	A, B	No
Perino and Ernhart13	1974	C=5O; E=30	Blood	C<3O; E =40-70	Multiple regression	No/H	· · ·	Yes
Rummo et al14	1979	C=45; E=45	Blood	C=23; E=61-88	ANOVA	No/G,D	A	Yes
de la Burde and Choate15	1975	C=67; E=7O	Blood	R = 30-100	×2	No/D,E,G	C	Yes
Landrigan et al16	1975	C=78; E= 46	Blood	C< 40; E = 40-68	t test	No/D,E		Yes
McNeil et al17	1975	C=37; E=lOl	Blood	C=29; E=59	t test	No/D	E, G, H	No
Yamins18	1976	80	Blood	PbB--X;=33.2=9.1	Multiple regression	No/E	A, B	Yes
Ketok et al19	1977	C=36; E=24	Blood	C=38; E=81	t test	No/G,D	A, B	No
Ratcliffe20	1977	C=23; E=24	Blood	C=28; E=44	t test	No/E	A, B	No
Needleman et al21	1979	C = 100; E = 58	Tooth	PbC = 24; PbE = 36;	ANCOVA	No/H		Yes
					PbTC£ 10; PbTE³ 2O			Yule et al22
1981		166	Blood	C£ 13; E = 13-32	Multiple regression	Yes		Yes
Winneke et al23	1982	C = 26; E = 26	Tooth	PbTC-x = 2.4;	PbTE-x = 9 t test	No/D	A	No
McBride et al24	1982	108	Blood	C = 2-9; E = 19-29	ANOVA	No/D, E		No
Smith et al25	1983	402	Tooth	PbT-x= 5.1 = 2.8	ANCOVA	No/H		No
Winneke et al26	1983	115	Tooth	PbT-x= 6.2; PbB-x =	14Multiple regression	Yes		No
Harvey et al27	1984	48	Blood	R = 6.2-26.8	Multiple regression	No/F	A	No
Shapiro and Marecek28	1984	193	Tooth	R = 30-150	Multiple regression	No/E	· · ·	Yes
Needleman et al29	1985	218	Tooth	PbT-x = 12.7	Multiple regressaon	Yes	· · ·	Yes
Emhart et al30	1985	80	Blood	C = 30; E = 40-70	Multiple regression	Yes	A	No
Schroeder et al31	1985	104	Blood	Median = 30	Multiple regression	Yes	...	Yes
Hawk et al32	1986	75	Blood	PbB-x = 21; R = 6-47	Multiple regression	Yes	A	Yes
Lansdown et al33	1986	C = 80; E = 80	Blood	C = 7-12; E = 13-24	Multiple regressmon	Yes	· · ·	No
Hatzakis et al34	1987	509	Blood	PbB-x = 23; R = 7-63	Multiple regression	Yes	· · ·	Yes
Pocock et al35	1987	402	Tooth	PbT-x = 5.1 = 2.8	Multiple regression	Yes	...	Yes
Fergusson et al36	1987	724	Tooth	PbT-X = 6.2 = 3.8	Multiple regression	Yes	. ..	No
Fulton et al37'	1987	501	Blood	GM = 11.5; R = 3-34	Multiple regressmon	Yes	·..	Yes
Hansen et al38	1987	156	Tooth	PbT x = 10.7; PbBx = 5	Multiple regression	Yes		Yes

*A indicates small sample; B, weak outcome measures; C, poor exposure measures; D, inadequate data analvsis or reporting E, inadequate or no covariate control: F , overcontrol:G, clinical levels of lead exposure (blood lead level >3.86 m mol/L); H, later reanalysis substituted (Needleman et al²⁹ [1985] for Needleman et al²¹ [1979] , Pocock et al ³⁵ [1987] for Smith et al²⁵ [1983], and Ernhart et al³⁰ [1985] for Perino and Ernhart'³ [1974]; PbTx, mean tooth lead value; PbB x, mean blood lead value; R, range; PbTC, values for control group; PbTE, values for high-lead group; GM, geometric mean; ANOVA, analysis of variance; and ANCOVA, analysis of covariance.

† all tooth studies are measured in parts per million and all blood studies are measured in micrograms per deciliter.

Methods

Data Collection

All studies on lead exposure and children's neurobehavioral development that were published since 1972 were examined for eligibility. The sources of candidate studies were a computerized MEDLINE subject search and a search of programs of meetings on metals, neurotoxicology, lead, pediatrics, and public health. Dissertation abstracts were also searched. Table 1 lists the studies identified in the search^12-38 and presents summary data.

Studies were excluded for the following reasons: (1) Inadequate control of covariates reflecting socioeconomic and familial factors.^{12,16-20,24,28} (2) Overcontrol of factors that reflect exposure to the independent variable, lead. One study²⁷ controlled for pica and peeling paint. (3) Inclusion of subjects with defined clinical lead poisoning (ie, blood lead levels >3.9 m mol/L). ^12,14,19 (4) Reported data either did not permit any further quantification¹⁵ or did not enable us to calculate the coefficient of lead in a multiple regression model.^{12,14,16,17,23,24}

Some studies were excluded on the basis of more than one of the above criteria. The first criterion effectively excludes most of the early studies in this area since these simply compared high-and low-lead groups, with limited or no control for relevant covariates. The second criterion was selected to avoid over-control. The one study²⁷ that was excluded on this basis also involved a very small sample (multiple regression with 17 covariates and complete data on 48 subjects).

Two of the studies^21,25 originally analyzed the data by dichotomizing lead exposure. The data were later reanalyzed by regression, treating exposure as a continuous variable. We used the results reported in the reanalyses. Supplementary information about the regression analysis was obtained from the authors of two studies.^26,38

Data Analysis

To achieve an acceptable level of homogeneity, the studies were divided into two groups according to the type of tissue analyzed for lead (blood or tooth). The P values within each group were compared for homogeneity using the technique of Rosenthal,^8(p76) which is based on the sum of the squared deviations of the t values for lead from the group mean.

Joint P values for lead were calculated for each of the two groups using two different approaches proposed by Fisher and by Mosteller and Bush.^8(p94) In Fisher's procedure, the logarithm of the product of the individual P values is multiplied by - 2. The resulting quantity has a χ² distribution with 2N df. In the procedure by Mosteller and Bush, the weighted sum of the t values of the lead coefficient is computed, with each coefficient being weighted by its df. This method effectively weights each study by the number of subjects involved. It is particularly useful in this meta-analysis because of the wide range of sample sizes (75 to 724).

For each study, the partial correlation coefficient of lead was derived from the corresponding t value and was used as a measure of effect size. These coefficients were transformed to z scores using Fisher's transformation^8(p27) and were then compared via a χ² statistic.^8(p77) When the hypothesis of homogeneity was not rejected, the values of partial r from each study were treated as independent estimates of a common (group) partial correlation. Weighted z score averages were computed and were used to construct 95% confidence intervals for the group partial correlation coefficient.^7(p227) Power for each study to find a "small" effect was computed using the method (and program) described in Gatsonis and Sampson.³⁹ We used the definition by Cohen⁴⁰ of a "small" effect (partial r=.14).

Finally, to assess whether the exclusion of 12 of the original 24 studies had a biasing effect on our conclusions, we used Fisher's aggregation technique in an analysis that included all 24 studies. For most of the early studies, P values were either given in the published reports or derived on the basis of the published data. In the few cases where a P value was not available, we followed a conservative approach and assumed it was .5.

Results

All studies considered and reasons for exclusion are listed in Table 1. Of the 12 excluded studies, 5 reported an effect significant at the .05 level and 7 did not. Twelve studies were included in the meta-analysis; 7 of them measured exposure by blood lead and 5 by tooth lead values (Table 2). The two groups were analyzed separately. In 11 of the 12 studies reviewed, the t value of the regression coefficient for lead was negative, ranging from -3.86 to 0.48 in the blood lead group and from -3.0 to -0.03 in the tooth lead group. The partial correlation coefficient of lead ranged from -.27 to .05 and from -.2 to -.003, respectively, for the two groups.

The dependent variable (IQ) was measured by the Wechsler Intelligence Scale for Children-Revised in eight studies. Two studies employed the Stanford Binet IQ Scale, one employed the British Ability Scale, and one employed the McCarthy Scale. The comparison of the distributions of lead exposure was hindered by two difficulties: (1) methods for measuring lead level differed, particularly in the tooth lead group, and (2) summary descriptions of the distribution of lead exposure also differed. In the blood lead group, the lead exposure in the study by Landsdown et al³³ (mean, 0.62 m mol/L) was among the lowest, while the exposure in the study by Schroeder et al³¹ (median, 1.46 m mol/L) was among the highest. In the tooth lead group, where analytic methods were different, the lead exposure in the study by Smith et al²⁵ was among the lowest (248 of 402 children had tooth lead concentration <5.5 ppm), while the exposure in the study by Needleman et al²¹ (mean, 12.7 ppm) was among the highest. The sets of covariates included in the regression equations differed for each study, although most covariates purported to measure factors that were similar across studies. It is impractical to present herein a detailed list of the covariates for each study. A condensed form of this information is in Table 3, in which we classified the various covariates into groups on the basis of seven factors. Where available, the unadjusted coefficient of lead is also included in Table 3, along with the coefficient of lead in the final model. In some studies the logarithm of the lead measurement was used in the regression equations.

Table 2.--Studies Included in the Meta-analysis*

Exposure Publication Subjects'
Study Year Measure Outcome Measure Status Age, y Country
Yule et al²² 1981 Blood WISC-R V, F Journal 6-12 United Kingdom
Lansdown et al³³ 1986 Blood WISC-R V, F Journal Preschool United Kingdom
Winneke et al²⁶ 1983 Tooth WISC-R V, F Journal 7-12 Germany
Needleman et al²⁹ 1985 Tooth WISC-R V, F Journal 7-8 United States
Emhart et al³⁰ 1985 Blood McCarthy Scale Journal Preschool United States
Schroeder et al³¹ 1985 Blood Bayley/Stanford Binet IQ Scale Journal 1-6 United States
Hawk et al³² 1986 Blood Stanford Binet IQ Scale Journal 3-7 United States
Fergusson et al³⁶ 1987 Tooth WISC-R V, F Journal 8-9 New Zealand
Fulton et al³⁷ 1987 Blood British Ability Scale C Journal 6-9 United Kingdom
Hatzakis et al³⁴ 1987 Blood WlSC-R V, F PROC 7-12 Greece
Pocock et al³⁵ 1987 Tooth WISC-R F Journal 6 United Kingdom
Hansen et al³⁸ 1987 Tooth WISC-R V, F PROC 7-8 Denmark

*WISC-R indicates Wechsler Intelligence Scale for Children-Revised; V, verbal; F, full-scale; and PROC, proceedings of meeting.

Table 3.--Covariates Entered Into the Final Multiple Regression Model*

---

Lead Coefficients†

Parental Perinatal Physical Parent Parental
Study‡ SES Factors Factors Factors Gender IQ Rearing Unadjusted Final Model
Yule et al²² (2) * … Age NA -8.08 (4.63)
Lansdown et al³³ (2) * … Age NA 2.15 (4.48)
Winneke et al²⁶ (52) * ... * * * NA -0.125 (466)
Needleman et al²⁹ (5) * * * ... NA -0.21 (0.07)
Ernhart et a³⁰ (3) ... * * Age ... * ... NA NA
Schroeder et al³¹ (7) * … NA -0.199 (0.07)
Hawk et al³² (1) * … * * * -0.456 - 0.255 (0.15)
Fergusson et al³⁶ (7) * * * ... * NA - 1.46 (1.25)
Fulton et al³⁷ (21)§ * * ... * * * * -5.45 (1.5) -3.70( 1.31_)

Hatzakis et al³⁴ (10) * * * * -0.376 -0.266 (0.07)
Smith et al²⁵ (18) * * * ... * * * -2.66(0.86)_-0.77(0.63)_

Hansen et al³⁸ (6) * · · · * NA -4.27 (1.21)

*Asterisk (*) indicates those factors entered into the model; and SES, socioeconomic status.

†NA indicates not available. Where available, coefficients for lead are given for the unadjusted bivariate model and the final multivariate model.

‡The number of coefficients entered into the initial model is in parentheses.

§The SE of the coefficients was estimated from the data.

The P values for the common directional hypothesis that lead is negatively correlated with IQ were tabulated. Before combining the probabilities, the homogeneity of the P values was assessed. The χ² statistics were 11.02 (df=6, P = .09) and 5.13 (df= 4, P = .26) for the blood lead and tooth lead group, respectively. Thus, the hypothesis of homogeneity cannot be rejected for either group. Combined P values in the blood lead group were less than .0001 for both methods of combining probabilities. The corresponding combined P values for the tooth lead group were less than .0005 and .004, respectively.

Sensitivity Analysis

The sensitivity of the findings was examined by removing the studies one by one from the analysis and recalculating combined P values (Table 4). For the tooth lead group the highest combined P value was .025 and the lowest was .0001. The corresponding figures for the blood lead group were below .0001. The overall finding of a significant lead effect is supported by both methods of combining the data. No single study seems to be responsible for the significance of the final finding.

Effect Size

In the case of multiple regression/correlation studies, the usual measure of effect size is the partial correlation coefficient (partial r).^7,8,40 Derived partial r's for the 12 studies under review are given in Table 5.

Each partial r was converted to a z score using Fisher's z transform. The χ² statistics for homogeneity were 5.78 (df=6, P >.4) for the blood lead group and 6.44 (df= 4, P> .1) for the tooth lead group. The hypothesis of homogeneity of the effect sizes cannot be rejected for either of the two groups. The weighted z score averages were -.152 (SE = .027) and -.08 (SE =.025), respectively. In the original scale, the approximate 95% confidence intervals for the group partial r were -.15 ± .05 for the blood lead group and -.08 ± .05 for the tooth lead group.

The results of the analysis in terms of the partial r’s support those obtained from the analysis of the P values. Neither approach provides an overall estimate of the raw effect size, ie, of the average change in IQ units per unit change in lead exposure. A meaningful attempt to arrive at such an overall estimate is precluded by the substantial differences in model specification among the studies, as well as in units and methods of measuring lead exposure and outcome.

Selection Bias and the File: Drawer Problem

There were two basic steps in the selection of studies for this meta-analysis: (1) the retrieval of studies and (2) the formulation and application of exclusion criteria to the retrieved studies. The possibility of bias in both steps was investigated. In particular with respect to the second step, calculations with all the original 24 studies included showed that Fisher's statistic was 93.8 (df= 34, P<.0001) for the blood lead group, 42.5 (df=14, P<.001) for the tooth lead group, and 136.4 (df=48, P<.0001) for all studies together. This is evidence that the application of the exclusion criteria was not an important source of bias in this meta-analysis.

The possibility of bias resulting from the first step has been termed the file drawer problem.^8(p107) Such bias may result from at least two sources (beyond faults in the retrieval process): the failure of all investigators to report their results or the failure of journals to publish all results submitted. Studies that show a statistically significant result do tend to be published more frequently. We estimated the magnitude of the file drawer problem by calculating the number of unpublished nonsignificant studies that would be necessary to bring the overall P value to greater than .05. Using the procedure of Rosenthal,^8(p108) we found that 26 null result studies would be necessary to dilute the finding for the tooth lead group and that 67 would be necessary for the blood lead group. This procedure assumes that the mean z score of the unseen studies is 0. A more stringent procedure is suggested by Iyengar and Greenhouse,⁴¹ which assumes that all unseen studies simply are not significant at the .05 level. Under this assumption, it would require 16 and 35 studies to dilute the finding for the tooth lead group and the blood lead group, respectively. Given the expense of conducting human studies of lead exposure and the amount of attention directed to this question, it is unlikely that this number of negative studies have escaped notice.

Table 4.-Results of Synthesis of 12 Studies

---

Weighted t Values Fisher's Technique

P
Study z (One-Sided) Χ² p

Blood Lead Studies
All studies - 5.46 <.0001 61.29 <.0001

Eliminating one study at a time

(study eliminated)

Hatzakis et al³⁴ -3.88 <.0001 42.87 <.0001

Hawk et al³² -5.34 <.0001 55.3 <.0001
Schroeder et al³¹ - 5.15 <.0001 49.68 <.0001
Fulton et al³⁷ - 4.87 <.0001 49.68 <.0001
Yule et al²² - 5.25 <.0001 54.86 <.0001
Lansdown et al³³ -5.56 <.0001 60.52 <.0001
Emhart et al³⁰ -5.31 <.0001 54.86 <.0001

Combining studies using log-transformed values
(Fulton et al,³⁷ Yule et al,²² and Lansdown et al³³) 18.83 .005

---

Tooth Lead Studies
All studies - 2.65 .004 33.11 <.0005
Eliminating one study at a time
(study eliminated)
Needleman et al²⁹ -1.97 .024 19.29 <.025
Hansen et al³⁸ -2.3 .011 23.9 <.005
Winneke et al²⁶ - 2.67 .004 31.68 <.0005
Smith et al²⁵ - 2.36 .009 28.69 <.0005
Fergusson et al³⁶ - 3.04 .001 28.88 <.0005

Combining studies using log-transformed values

(Smith et al²⁵ and Fergusson et al³⁶) - 1.61 .001 8.66 <.0005

Table 5.--Lead Coefficients for Full-scale IQ Scores*

P Sample Partial Total
Study Coefficient SE t ( One-Sided) Size r R2

Blood Lead Studies
Hatzakis et al³⁴ -0.27 0.07† -3.86† .0001 509 -.17 0.25
Hawk et al³² - 0.25 0.15 - 1.67 .05 75 - .20 0.21

Schroeder et al³¹ - 0.2 0.07† - 2.78 .003 104 - .27 NA
Fulton et al³⁷‡ - 3.7 1.37 -2.77 .003 501 -.12 0.46
Yule et al²²‡ -8.08 4.63 - 1.75 .04 129 - .16 NA
Lansdown et al³³‡ 2.15 4.48† 0.48 .68 86 .05 NA
Emhart et al³⁰ NA NA - 1.8† .04 80 - .20 NA

(Average weighted partial r = - .152; 95% confidence interval, -.2 to -.1 )

________________________________________________________________ Tooth Lead Studies
Needleman et al^{2 9} -0.21 0.07 -3 .001 218 -.20 0.35
Hansen et al³⁸ -4.27 1.91 -2.23§ .01 156 -.18 0.2
Winneke et al²⁶ -0.13 4.66 -0.03§ .49 115 -.003 0.13
Pocock et al³⁵‡ - 0.77 0.63 - 1.22 .11 388 - .06 NA
Fergusson et al³⁶‡ - 1.46 1.25 - 1.17 .12 724 - .04 NA

Average weighted partial r = -.08; 95% confidence interval, -.13 to - .03)

_________________________________________________________

*NA indicates not available.

†Estimated from data in article.

‡Log Transforms.

§Obtained from the author.

Power Calculations

The studies included in this meta-analysis are observational. The values of the covariates cannot be fixed in advance by design but are themselves outcomes of the study. Any calculation of power must account for this extra variability.³⁹ Table 6 presents the a priori power of each study to detect a partial r of .14 (denoted as a "small" effect⁴⁰). "Small" in this sense does not mean biologically unimportant, it means difficult to identify. Cohen⁴⁰ has pointed out that a result of this size "all too frequently in practice represents the true order of magnitude of the effects being tested." As can be seen from Table 5, a partial r equal to -.14 is near the center of the values for the partial correlation coefficient that were derived from the studies under review. Of the 12 studies, 8 had power below 60% to detect an effect of this magnitude.

The power figures given here are optimistic: they are calculated on the number of covariates present in the final model reported in each study. Most studies, however, initially controlled for many more covariates than those in the final model. As few articles gave information about missing values in the data, it is possible that some of the sample sizes used to calculate power are larger than the effective sample sizes of the studies.

Some Methodological Issues

The inclusion criteria ensured that the studies analyzed provided an acceptable minimum control for relevant covariates. In two studies,^22,33 control was done only for social class. Multiple regression analysis was employed in all studies, usually in stepwise form. No study reported any analysis of residuals, model checking, and detection of possible outliers in the data. Only two studies attempted to select an "optimal" regression model in a formal way. No study addressed the issue of errors in measurement of the independent variables. The question of errors in variables is particularly relevant when measuring exposure at low levels. Other covariates that represent arbitrary constructs (eg, marital relationships, parental interest, parental involvement in school, and so on) are also particularly vulnerable to errors-in-variables problems.

Comment

The overall evidence from our meta-analysis establishes a strong link between low-dose lead exposure and intellectual deficit in children. A natural question that arises at this point is whether the link is a causal one. The answer to this question goes beyond the formal meta-analytic method. Some of the epistemological issues encountered in making causal inferences are discussed below.

The effects of lead on the central nervous system are embedded in a complex process involving biologic, environmental, familial, and socioeconomic factors. Epidemiologic studies cannot, by themselves, establish causal relationships. Causality is not subject to empirical proof, whether in the field or in the laboratory.⁴² Given that direct demonstration of proof of a low-dose lead effect in a naturalistic setting is not achievable, epidemiologists rely on canons⁴³ that, if satisfied, permit the conservative drawing of causal inferences. They are (1) time precedence of the putative cause, (2) biologic plausibility, (3) non-spuriousness, and (4) consistency.

Cross-sectional studies such as those reviewed herein cannot establish the time precedence of lead exposure; the level of lead was measured at the same time as IQ. The claim has been made that neurobehavioral deficits result in excess lead intake, ie, deficient children mouth more leaded substances. This assertion has been effectively refuted by forward studies of lead exposure beginning at birth. These studies have shown a clear relationship between umbilical cord blood lead levels and later development at 6 to 24 months.^44-46

Biologic plausibility demands that mechanisms at a lower biologic level have been demonstrated to explain the phenomenon under examination. Lead is a thoroughly investigated neurotoxin.⁴⁷ Among many effects that have been demonstrated, lead has been shown to affect neurotransmitter activity, brain adenyl cyclase activity, and dendritic complexity.^48-50 Demonstration of dose-response relationships strengthens the plausibility of the relationships studied. Convincing demonstrations of dose-related behavioral effects have been made in animal studies.^51,52 Epidemiologic demonstrations of association between dose (blood or tooth lead levels) and response (teachers' ratings of classroom behavior and reaction time under varying intervals of delay) also have been published.^21,34

Nonspuriousness means that the relationship put forth in the causal claim is not due to a confounder or a set of confounders. Complete confounder control is impossible in real world studies. In most studies reviewed, control of confounders has reduced the magnitude of the lead-IQ effect but has not obliterated it. The argument for nonspurious-ness is further strengthened by the evidence provided by animal studies in rodents and subhuman primates, which produced cognate outcomes in cross-fostered litter mates.^51,52

Finally, consistency requires that the phenomenon be demonstrated in different studies under similar but not identical circumstances. The statistical nature of these investigations requires an extended notion of consistency. Even if the effect under study exists in nature, the P values and effect sizes reported in investigations of the question will vary in magnitude, and not all studies will give a significant result.

A different type of evidence for consistency was offered in the study by Wallsten and Whitfield.⁵³ This evidence is based on the probabilistically encoded opinions of six lead experts of widely ranging viewpoints about the dose-response relation between lead exposure and IQ. Five of the six experts' estimates of the dose-response curve were convergent, leading the authors to state: "In view of the extensive debate concerning the effects of lead on IQ, the degree of consensus reflected in the study’s results is notable, especially since the experts were selected so as to span the full range of opinion."

 

Table 6.--Power Calculations for "Small" Effects of Lead ( =a .05; Partial r =.14)

No. of
Sample Covariates
Study Size (Final) Power

Blood Lead Studies

Fulton et al³⁷ 501 14 0.87
Hatzakis et al³⁴ 509 8 0.88
Hawk et al³² 75 5 0.21
Schroeder et al³¹ 104 7 0.28
Yule et al²² 129 2 0.35
Lansdown et al³³ 86 2 0.25
Ernhart et al³⁰ 45 4 0.23

---

Tooth Lead Studies

Needleman et al²⁹ 218 5 0.53
Fergusson et al³⁶ 724 8 0.96
Smith et al²⁵ 388 10 0.78
Winneke et al²⁶ 115 4 0.31
Hansen et al³⁸ 156 7 0.40

 

 

The four previously cited reviews of the studies of lead at low dose differed in their evaluation of essentially the same evidence. One review came to a qualified negative conclusion,⁵ one came to a positive conclusion,⁴ and two found the evidence inconclusive.^2,3 This difference of opinion partly proceeds from a limitation inherent in the method of narrative reviewing; it essentially evaluates each study in isolation and is unable to achieve a systematic synthesis. Meta-analysis avoids this limitation and includes all studies in a joint inference. Using this method, and incorporating into our review a number of recent studies that were not available to the earlier reviewers, we found that although the sample of studies varied widely in their individual power to find an effect, and not all found an effect by the conventional rule of P<.05, 11 of 12 studies reviewed reported a negative coefficient for lead. The joint probability of the findings reported occurring by chance under the null hypothesis was quite small, and this was not materially influenced by any single study. The estimated effect sizes for the two groups were both significantly different from zero. These findings, taken in sum, permit a strong inference that low-dose lead exposure is causally associated with deficits in psychometric intelligence.

References

1. Byers RK, Lord E. Late effects of lead poisoning on mental development. AJDC. 1943;66:471-494.
2. Rutter M. Raised lead levels and impaired cognitive/behavioral functioning a review of the evidence. Dev Med Child Naurol. 1980;2~(suppl 42):1-26.
3. Bornschein R, Pearson D, Reiter L. Behavioral effects of moderate lead exposure in children and animal models. CRC Grit Rev ~bzicol. 1980;8:43-99.

4. Needleman HL, Bellinger D. The developmental consequences of childhood

exposure to lead: recent studies and methodological issues. In: Lahey BE, Kaz~din

AE, eds. Advances in Clinical Child Psychology New York, NY: Plenum

Publishing Corp; 1984;7:195-220.

4. Pocock SJ, Ashby D. Environmental lead and children2~ intelligence: a review of recent epidemiologicalstudies. The Statistician. 1985;35:31-44.
5. Glass GV, McGaw B, Smith ML. Meta-analysis in Social Research. Beverly Hills, CaliE Sage Publications Inc;l981.
6. Hedges LV, Olldn I. Statistical Methods for Meta-analysis. Orlando, Fla: Academie Press Inc; 1985.
7. Rosenthal R. Meta-analytic Proceduresfor Social Research. Beverly Hills, Calif: Sage PublicationsInc;l984. Sacks HS, Berrier J, Reitman D, Ancoma-Berk VA, Chalmers TC. Mete-analyses of randomized controltrials. NEnglJ Med 1987;316:450-455.

10. SehwartzJ, PitcherH, Levin R, OstroB, Nicholas AL. Cost and Benefits of

Reducing Lead in Gasoline: Final Regulatory Impact Analysis Washington, DC:

Office of Policy Analysis, Environmental Protection Agency; 1985.

11. Needleman HL, Bellinger D. Type II errors in studies of low level lead exposure:

a critical and quantitative review. In: Smith M, Grant L, Sors A, eds. Lead

Ezposure and Child Development: An International Assessment. Lancaster,

England: MTP Press Ltd; 1988.

12. Kotok D. Development of children with elevated blood levels: a controlled study.

J Pediatr 1972;80:57-61.

13. PerinoJ, Ernhart C. The relation of subclinical lead level to cognitive and sensory

impairment in black preschoolers. JLearn DisaB. 1974;7:2630.

14. Rummo JH, Routh DR, Rum~o NJ, Brown JF. Behavioral and neurological

symptomatic and asymptomatic lead exposure in children. Arch Envlron Health.

1979;34:120-124.

15. de la Burde B, Choate MS. Early asymptomatic lead exposures and development

at school age. J Pediatr. 1975;87:63&664.

16. Landrigan P Baloh R, Whitworth R, Staeling N, Rosenblum B.

Neuropsychological dysfunction in children with chronic low level lead

absorption. Lancet.1975;1:708.

17. MeNeil J, Ptasnik J, Croft D. Evaluation of long-term effects of elevated blood

lead concentrations in asymptomatic children. Arch Ind HygieneIbzieol.

1975;26:97-118.

18. Yamins JG. The Relationship of Subclinical Lead Inttozication to Cognitive and

Langwcgeli~nction in Preschool Children. Long Island, NY: Hofstra University;

1976. Thesis.

19. Kotok D, Kotok R, Heriot T. Cognitive evaluation of children with elevated blood

levels. AJDC. 1977;131:791-793.

20. Ratcliffe JM. Developmental and behavioral functions in young children with

elevated blood lead levels. Br JPrev Soc Med. 1977;31:258-264.

21. Needleman H, Gunnoe C, Leviton A, et al. Deficits in psychologic and classroom

performance in children with elevated dentine lead levels. NEnglJMed.

1979;300:584-695.

22. Yule W, Lansdown R, Miller I, Urbanowicz M. The relationship between blood

lead concentration, intelligence, and attainment in a school population:

a pilot study. Dev Med Child Neurol. 1981;23:567- 576.

23. Winneke G, Kramer G, Brockhaus A. Neuro- psychological studies in children

with elevated tooth lead concentration, I: pilot study. Int ArchOcczlp Ennriron

Health. 1982;51:169183.

24. McBride WG, Black BP, English BJ. Blood lead levels and behaviour of 400

preschool children. MedJAust. 1982;2:26-29.

25. Smith M, Delves T, Lansdown R, Clayton B, Graham P. The effects o~lead

exposure on urban children: The Institute of Child Health/Southampton Study.

Dev Med Child Neurol. 1983;25(suppl47):1-54.

26. Winneke G, Kramer G, Brockhaus A, et al. Neuropsyehological studies in

children with elevated tooth lead concentration. Int Arch, Occul, Envi-

ron Health. 1983;51:231-252.

27. Harvey PG, Hamlin MW, Kumar R, Delves HT. Blood lead, behaviour and

intelligence test performance in preschool children. Sci Total Envron.

1984;40:4&60.

28. Shapiro IM, Marecek J. Dentine lead concenbation as a predictor of

neuropsychological functioning in inner-city children. Biol ~ace Elem Res.

19&1;6:69-78.

29. Needleman HL, Geiger SK, Frank R. Lead and I~ scores: a reanalysis. Science.

1985;227:701 704.

30. Emhart CB, Landa B, Wolf AW. Subclinical lead level and developmental deficit:

re-analyses of data. J Learn Disab. 1985;18:475479.

31. Schroeder SR, Hawk B, Oto DA, Mushak P, Hicks RE. Separating the effects of

lead and social factors on IQ. Elzviron Res. 1985;91:178-183.

32. Hawk BA, Schroeder SR, Robinson G, et al. Relation oflearl and social factors to

IQ of low SES children: a partial replication. Am. J Ment Def. 1986;91:178-183.

33. Lansdown R, Yule W Urbanowicz MA, Hunter J The relationship between blood

-lead concentrations, intelligence attainment and behavior in a school population:

the second London study. IntArch OcMlp Environ Health. 1986;57:225~235.

34. Hatzakis A, Kokevi A, Katsouyanni K, et al. Psychometric intelligence and

attentional performance deficits in lead exposed children. In: International

Conference on Heavy Metals in the Envronment. Edinburgh, Scotland: CEP

Consultants; 1987:204.

35. Pocock ST, Ashby D, Smith V. Lead exposure and children's intelligence. Int J

Epidemiol. 1987;16:57-67.

36. Fergusson DM, Fergusson JE, Harwood LJ, Ki~ett NG. A longitudinal study of

dentine lead levels, intelligence, school performance and behavior, III: dentine

lead levels and cognitive ability. J Child Psycltol Psychiatry. 1988;29:793-809.

37. Fulton M, Raab G, Thomson G, Laxen D, Hunter R, Hepburn W Influence of

blood lead on the ability and attainment of children in Edinburgh. Lancet.

1987;1:1221-1226.

38. Hansen ON, Trillingsgard A, Beese I, Lyngbye T, Grandjean P A

neuropsychologieal study children with elevated dentine lead levels. In: In-

ternational Conference on Heavy Metals in the Environment. Edinburgh, Scotland:

CEP Consultants; 1987:51-56.

39. Gatsonis C, Sampson A. Multiple correlation: exact power and sample size

calculations. PsychBull. 1989;106:516-524.

40. Cohen J Statistical Power Analysis for the Behavioral Sciences. Orlando, Fla:

Academic Press Inc; 1977.

41. Iyengar S, Greenhouse J. Selection models and the file-drawer problem. Stat Sci.

1988;3:109-135.

42. Rothman KJ Modem Epidemiology. Boston, Mass: Little Brown & Co Inc; 1986.
43. Kenny DA. Correlation and Causality. New York, NY: John Wiley & Sons Inc; 1979.
44. Bellinger D, Leviton A, Waternaux C, Needleman H, Rabinowitz M. Longitudinal analyses of prenatal lead exposure and early cognitive development. N Engl JMed. 1987;316:1037-1043.
45. Deitrich K, Krafft K, Beir M, et al. Early effects of fetal lead exposure: developmental findings at 6 months. In: International Conference on Heavy Metals in the Environment. Edinburgh, Scotland: CEP Consultants; 1987:63-65.
46. MeMichael AJ, Baghurst PA, Wigg NR, et al. Port Pirie cohort study:

environmental exposure to lead and children~ abilities at four years. NE~I J

MedA988;319:468-475.

47. Air &wclity Criteriafor Lead. Research Triangle Park, NC: Environmental Protection Agency; 1986. Publication EPAV600/83/028aa.
48. Hrdina PD, Hanin I, Dubas TC. Neurochemi cal correlates of lead toxicity. In: Singhal RL, Thomas JA, eds. Lead Zbzicity. Baltimore, Md: Urban & Schwarzenberg; 1980:273-300.
49. Nathanson JA. Lead-inhibited adenylate eyclase, a model for the evaluation of chelating agents in the treatment of lead toxicity. JPharrn Pharrnacol. 1977;29:511-513.
50. Averill D, Needleman HL. Neonatallead exposure retards cortical synaptogenesis in the rat. In:Needleman H, ed. Low Level Lead Ezposure: The Clinical Implications of Current Research. NewYork, NY: Raven Press; 1988.
51. Rice DC, Willes RF Neonatal low level lead exposure in monkeys (Maeacafasciculosis): effect on two-choice non-spatial form discrimination. J Environ It·Lthol Tozicol. 1979;2:1195-1203.
52. Cory-Schlecta DA, Bissen ST, Young AM, Thompson T. Chronic post-weaning lead exposure and response duration performance. Ibzicol Appl Phamnacol. 1981;60:78-84.
53. Wallsten TS, Whitfield RG. Assessing the risks to young children of three effects associated with elevated blood-lead levels. Argonne, Ill: Argonne National Laboratory; 1986. Report ANL/AA-32.

Reprint with permission from:

678 JAMA, February 2, 1SSO-Vol 263, No. 5 Lead Exposure and IQ--Needleman & Gatsonis.

TOP