Site Heterogeneity

TLC enrolled 780 children across four clinical centers:

The Baltimore site had a unique environmental intervention not available at the other three sites. This created systematic differences in the environmental context of treatment across centers.

Farfel et al. (2000) documented that Baltimore TLC homes received a tiered intervention system that was unavailable at the other three sites:

• Clean A: Professional cleaning only (standard across all sites)
• Clean B: Professional cleaning with enhanced techniques
• Minor Repair: Cleaning plus minor lead paint repair (e.g., window replacement, repainting)
• Moderate Repair: Cleaning plus moderate lead hazard reduction

The tiered system was part of a companion study on environmental intervention effectiveness. Baltimore participants were assigned to different intervention levels based on housing assessment.

This means Baltimore children, on average, had better environmental control than children at other sites. If chelation is more effective when environmental exposure is reduced, Baltimore would be expected to show a larger treatment effect.

The ICH E9 guideline on Statistical Principles for Clinical Trials (1998) recommends testing treatment-by-center interactions before pooling multi-site trial data:

“The consistency of treatment effects across centres should be examined.... A treatment-by-centre interaction should be assessed, and reasons for any significant interaction should be investigated.”

TLC did not pre-specify a site × treatment interaction test in the protocol. The only pre-specified interaction test was gender × treatment.

No published TLC analysis reports a site-by-treatment interaction test. The primary publications pooled data across all four sites without examining whether treatment effects were consistent.

If the environmental context differed by site, the treatment effect could have varied by site. Consider two scenarios:

Scenario 1: Homogeneous null effect
All four sites show similar null treatment effects. This would support the conclusion that succimer does not improve cognition regardless of environmental context.

Scenario 2: Heterogeneous effects
Baltimore (with better environmental control) shows a positive treatment effect, while other sites show null or negative effects. Pooling would average these effects, producing an overall null result that masks the Baltimore finding.

Without the site-by-treatment interaction analysis, we cannot distinguish between these scenarios. The published data does not allow us to determine whether succimer might work in the context of adequate environmental intervention.

If Baltimore showed a treatment effect that was diluted by pooling with other sites, the policy implication would be different: chelation might work, but only when combined with environmental remediation.

The site-level differences extended to how children were enrolled. The TLC Group (1998, pp. 316–318) describes recruitment through primary care clinics, community screening programs, health fairs, and other referral pathways — but these pathways varied by center based on local health department practices and screening infrastructure.

There was no defined sampling frame or population registry. It is unknown what proportion of eligible children in each community were screened, referred, or enrolled. The TLC Group (1998, Table 3) reports site-level enrollment figures but does not report the denominator of eligible children at each site. The enrolled cohort represents a convenience sample rather than a probability sample of lead-exposed children.

This matters for both internal and external validity. If recruitment pathways selected systematically different populations at each site — differing in lead exposure history, socioeconomic status, or health-seeking behavior — these unmeasured differences could confound the pooled analysis. The conclusion by Rogan et al. (2001) that succimer “should not be prescribed for children with blood lead levels between 20 and 44 µg/dL” was derived from this convenience sample at four urban sites but applied as a universal policy recommendation.

The primary publications do not report outcomes by site. No table or figure shows IQ results stratified by clinical center.

The Farfel et al. (2000) paper on environmental intervention reports dust lead levels by intervention group (Clean A, Clean B, Minor Repair, etc.), but does not link these to treatment arm or cognitive outcomes.

A re-analysis of the TLC data examining site-by-treatment interactions would be informative but requires access to the original dataset. Such an analysis has not been published.