March 12, 2018

lead and CDV deaths in US adults

Low-level lead exposure and mortality in US adults is examined in a population-based cohort study, published in The Lancet Public Health.

A roundup accompanied this analysis.

Title, Date of Publication & Journal

Title: ‘Historical lead exposure may be linked to 256000 premature deaths from cardiovascular disease in adults the USA each year’ by Lanphear et al.

Published: Monday 12 March

The Lancet Public Health

Study’s main claims – and are they supported by the data

This paper is a follow-up of a cohort of patients (of a previously published study) from the US which quantifies the number of cardiovascular deaths attributed to low levels of lead exposure.

It is critical to note that this analysis is from patients enrolled from 1988-1994 in the US. This is 25-30 years ago! i.e. when laws were a bit more lax than what they are now (both in the US and in the UK). During this period of enrolment, lead was still used in gasoline in the US and was phased out by 1996. The findings relate to a historical effect rather of exposure to lead rather than a current/ongoing effect. The authors clearly state themselves that “it is more accurate to view this study as estimating how many deaths might have been prevented if historical exposures to lead had not occurred.”

Given the evidence presented in this paper, this claim is plausible but may not be the sole cause of cardiovascular deaths. Particularly since the measurements to detect lead levels were only taken once. This matters because a patient’s social factor can change (e.g. no longer works at a gas station), so you cannot assume a patient’s lead level remains the same over time.

This paper does not claim that cardiovascular deaths are a direct consequence of low levels of lead exposure. The authors themselves point out that atherosclerosis (plaque build-up in arteries) and hypertension could (but are not proven) to be “underlying mechanisms for the cardiovascular toxicity of lead”, based on a randomised controlled trial. The authors themselves point out that the analysis does not adjust for air pollution or arsenic, both of which are known risk factors of cardiovascular disease mortality.

Lead poisoning is potentially a surrogate marker for “low socio-economic status”.

It would be wrong to conclude that 18% of all deaths in the USA may be a result of lead exposure.  That claim would be fine if it was true that the whole US population were exposed to low levels of lead 30 years ago. But that isn’t the case.

What the paper actually is saying is this: if patients in this study with high lead levels had them reduced to 1, then 18% of the deaths in this group (i.e. levels of lead exposure greater than 1) could have been prevented.

In other words, you cannot assume that 18% of deaths being avoidable in one group (i.e. low levels of lead exposure) is the same as 18% of ALL deaths across a country.

Strengths/Limitations

Strengths

This paper is perhaps one of the first to specifically look at the effects of low exposure of lead on mortality outcomes. They highlight that it is an important risk factor to look at in future studies.

This is a very large cohort, representative of the US population.

Limitations

Lead levels were only measured at baseline and not again i.e. over time.

The results didn’t take into account a family history of cardiovascular disease.

The authors claim that there was a higher risk of cardiovascular disease mortality for those who did not smoke vs. those that did, but a major limitation with this is that smoking was self-reported. People tend to lie about these sorts of things and so it’s not clear how accurate this might be. Secondly, it is unclear how many patients were included in this analysis (since they make no statement of how they dealt with those who were former smokers in the analysis) and so it is unclear how representative it is.

The previously published paper included patients that were ≥17. Given that this is a follow up of that study, it is not clear why they only include patients that are ≥20 years.

1150 patients (9%) had lead concentration level below 1 μg/dL, and therefore these patients’ levels were imputed with a value of 0.7 μg/dL. The reason this was done was because they couldn’t detect the exact concentration level but they knew it was somewhere between 0 and 1 (treated like missing data). The authors reference a paper which suggests calculating this value with the level of detection (which is 1 μg/dL in this study) divided by square root of 2 (for this analysis this would simply be 1/√2 = 0.7). The paper they reference with this method suggests calculating in this way if data are highly skewed with a geometric SD ≥ 3.

The authors only report a SE of 0.13 no SD, so it’s unknown if the method they chose is appropriate and it’s unclear whether the assumptions made for this method were appropriate.

It’s unclear how much impact using a value of 0.7 μg/dL has on the analysis. They could have done these using sensitivity analyses (which works by changing the assumptions to see how robust their conclusions are). For instance, had they also looked a low value of (say) 0.3 would that have changed the results/conclusions?

The authors claim that using a RCS model with 5 knots showed a steep rise in the risk of death for low concentrations of lead. Using 5 knots may be excessive and they do not state where the knots are placed. The choice of these may impact on how the curve is modelled.

Glossary

SE – standard error: a measure of how “wrong” we are about the estimate (in this case the geometric mean)

SD – standard deviation: a measure of how spread the data is from the mean

RCS – restricted cubic splines: a complex statistical model that is used to get a better fit of a model to the data, used to capture non-linear trends of the data.

Impute/imputation – statistical method used to deal with missing data. Simply speaking, data that are missing are replaced with a value (this is single imputation as done here since “missing” values are replaced once with a “known” value). A better method is multiple imputation, which replaces a missing value while accounting for the uncertainty of that imputed value (and that is done by replacing that missing value several times and averaging it).

* ‘Low-level lead exposure and mortality in US adults: a population-based cohort study’ by Bruce P Lanphear et al. published in The Lancet Public Health on Monday 12th March 2018. 

Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry (PSI) and experienced statisticians in academia and research. A list of contributors, including affiliations, is available here.