Editorial: Note About Inaccurate Results Published in the American Journal of Epidemiology and the American Journal of Public Health

This article was jointly published in the American Journal of Epidemiology (Am J Epidemiol. 2017;185(6):407–408) and the AJPH (Am J Pub Health. 2017;107(4):502).

In 2013, Masters et al. published articles in the American Journal of Public Health (AJPH)¹ and the American Journal of Epidemiology (AJE)² in which they reported results of analyses of data from the National Health Interview Survey that were linked to individual National Death Index mortality records from 1986 to 2006. The two papers, which were related, were about age variation in the association between obesity status and mortality risk in US adults. In the AJE article, Masters et al. reported that contrary to common knowledge, the association between obesity and mortality risk became stronger with age (Figure 2 of Masters et al.²). In the AJPH article, they computed the corresponding population attributable fractions and found that the percentage of mortality attributable to obesity increased with age and across birth cohorts (Figure 3 of Masters et al.¹).

In a letter to the editor in the AJE and another in the AJPH, Wang³ and Wang and Liu⁴ criticized Masters et al. for their approach to correcting for selection bias. According to these authors, Masters et al. underestimated the effect of obesity on mortality among younger people and overestimated it among older people. In both letters, the authors concluded that the bias was such that instead of increasing with age, as reported by Masters et al.,^1,2 the proportion of mortality attributable to obesity should have decreased with older age. Masters et al. responded to Wang in the AJE⁵; however, the letter by Wang and Liu in the AJPH⁴ was published alone because Masters et al. declined the invitation to respond.

BIASED RATES

When asked to examine the issue, Hanley wrote an article jointly published in the AJE⁶ and the AJPH⁷ in which he demonstrated that the criticisms of Masters et al. that were expressed in the letters to the editors were valid. Using two independent simulations based on similar data, Hanley showed that the approach used by Masters et al. not only did not correct for the simulated selection bias but also introduced a new bias in the opposite direction of the true association. Hanley’s report was reviewed internally by several editors, including three editors with strong expertise in biostatistics, and externally by an Ethical Advisory Board comprising three public health scientists.

This issue has important public health implications. The results of Masters et al. could be used by insurance companies to justify increasing the life insurance premiums of obese persons as they grow older. The results could also lead insurance companies to reduce rates for obese persons younger than 50 years of age because they spuriously suggested that younger obese persons are less likely to die than are nonobese persons.

TO WITHDRAW OR NOT

As editors, we are convinced by the consistent assessments of Wang³ and Wang and Liu,⁴ as well as the careful simulations by Hanley,^6,7 that the statistical strategy used by Masters et al. to remove selection bias was ill originated and that their hazard ratio estimates were severely biased. We have, however, no reason to suspect misconduct or fraud and will therefore not withdraw the articles if the authors do not take the initiative to do so. The authors seem to have made genuine errors in their original papers. The errors escaped the review process, and our Journals accepted their papers for publication. All relevant information is now available to the readers.