Formaldehyde Exposure and Asthma in Children: A Systematic Review

Gerald McGwin Jr.; Jeffrey Lienert; John I. Kennedy Jr.


Environ Health Perspect. 2010;118(3) 

In This Article


This review was conducted using a modified version of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines for the conduct of systematic reviews and meta-analysis of observational studies (Stroup et al. 2000). We identified studies through the PubMed/MEDLINE (National Library of Medicine 2009) and Google Scholar (2009) databases, employing a search strategy that combined text word (e.g., "formaldehyde and asthma and children") and medical subject headings to identify reports regarding formaldehyde exposure and asthma. The reference lists of the identified studies were also reviewed to identify other relevant studies. Studies were initially selected if they appeared to contain qualitative or quantitative estimates for the association between formaldehyde exposure and asthma in children. We were specifically interested in studies that compared children with and without asthma with respect to formaldehyde exposure. All of the studies initially selected were in English.

In total, we identified 18 articles that met the aforementioned criteria, and after a detailed review, determined that 10 articles contained information suitable for use in a systematic review. Three review articles were excluded (Burr 1999; Daisey et al. 2003; Mendell 2007). Three additional articles were excluded because they were not asthma-specific, but rather focused on respiratory symptoms (e.g., chest discomfort) or pulmonary function (Franklin et al. 2000; Symington et al. 1991; Wantke et al. 1996). Two studies were excluded because, although asthma-specific, they did not contain a reference or control group (Delfino et al. 2003; Erdei et al. 2003). For the 10 articles included, we abstracted information regarding study design and setting, subject response/participation rates, definition of asthma (e.g., physician diagnosis), sample size, average (minimum and maximum) formaldehyde levels, average age of study subjects, and quantitative estimates (or raw data) for the association between formaldehyde exposure and asthma as well as whether such estimates were adjusted and, if so, for what measures (Table 1). Three of 10 studies did not contain actual formaldehyde measurements, and attempts to obtain this information from the study authors have been unsuccessful to date (Doi et al. 2003; Pati and Parida 2005; Tavernier et al. 2006).

Once the relevant results from each study were extracted, we determined that homogenizing the individual study results using a single unit of formaldehyde measurement would be necessary. Because most studies reported their results as odds ratios (ORs) per 10-µg/m3 unit increase in formaldehyde, this was chosen as the common metric. Thus, results for those studies using different units were transformed. For example, if a study reported an OR reflecting a 1-µg/m3 increase in formaldehyde, the natural logarithm of the OR was calculated and multiplied by 10; this value was then exponentiated to obtain an OR for a 10-µg/m3 unit increase in formaldehyde. This process was repeated for the 95% confidence interval (CI). Thus, for each study, an OR and 95% CI for the association between asthma and a 10-µg/m3 unit increase in formaldehyde exposure was obtained. One study (Zhao et al. 2008) provided two estimates: one for indoor and another for outdoor exposure, both of which were used.

Pooled ORs and 95% CIs were obtained using inverse variance-weighted, fixed-effects, and random-effects models. We tested heterogeneity using the Q test and quantified with the I 2 statistic. Whereas the Q test only determines whether statistically significant heterogeneity exists, the I 2 statistic calculates the proportion of the variability that can be attributed to heterogeneity across the studies. I 2 values of 25%, 50%, and 75% have been suggested as indicators of low, moderate, and high heterogeneity, respectively. Fixed-effects models are considered appropriate for values of < 50%, whereas for values of ≥ 50%, random-effects models are preferred. To evaluate whether the observed results were unduly influenced by any individual study and to determine if there was any publication bias, an influence plot and a funnel plot, respectively, were used.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: