Household Air Pollution and Acute Lower Respiratory Infections in Adults: A Systematic Review

Introduction Household air pollution from solid fuel burning kills over 4 million people every year including half a million children from acute lower respiratory infections. Although biologically plausible, it is not clear whether household air pollution is also associated with acute lower respiratory infections in adults. We systematically reviewed the literature on household air pollution and acute lower respiratory infection in adults to identify knowledge gaps and research opportunities. Methods Ten bibliographic databases were searched to identify studies of household air pollution and adult acute lower respiratory infection. Data were extracted from eligible studies using standardised forms. Results From 4617 titles, 513 abstracts and 72 full-text articles were reviewed. Eight studies met the inclusion criteria of which 2 found a significant adjusted increased risk of acute lower respiratory infection, 2 identified a univariate association whilst 4 found no significant association. Study quality was generally limited. Heterogeneity in methods and findings precluded meta-analysis. Discussion A systematic review of the literature found limited evidence for an association between household air pollution and risk of acute lower respiratory infection in adults. Additional research, with carefully defined exposure and outcome measures, is required to complete the risk profile caused by household air pollution in adults. Registration number CRD42015028042.


Introduction
Household air pollution from solid fuel burning kills over 4 million people every year including half a million children from acute lower respiratory infections. Although biologically plausible, it is not clear whether household air pollution is also associated with acute lower respiratory infections in adults. We systematically reviewed the literature on household air pollution and acute lower respiratory infection in adults to identify knowledge gaps and research opportunities.

Methods
Ten bibliographic databases were searched to identify studies of household air pollution and adult acute lower respiratory infection. Data were extracted from eligible studies using standardised forms.

Results
From 4617 titles, 513 abstracts and 72 full-text articles were reviewed. Eight studies met the inclusion criteria of which 2 found a significant adjusted increased risk of acute lower respiratory infection, 2 identified a univariate association whilst 4 found no significant association. Study quality was generally limited. Heterogeneity in methods and findings precluded meta-analysis.

Discussion
A systematic review of the literature found limited evidence for an association between household air pollution and risk of acute lower respiratory infection in adults. Additional research, with carefully defined exposure and outcome measures, is required to complete the risk profile caused by household air pollution in adults.

Introduction
Exposure to household air pollution from domestic solid fuel use causes over 4 million deaths per year, and is the most important environmental risk factor for disability adjusted life years worldwide [1]. Household air pollution has been found to increase the risk of childhood acute lower respiratory tract infection (ALRI) by 78% (pooled odds ratio = 1.78 (95% CI 1.45 to 2.18) [2]) and is responsible for half a million deaths in young children every year. Over a third of the world's population are exposed to high levels of household air pollution, with women and children experiencing the greatest burden [3].
Disease outcomes in adults attributed to household air pollution from solid fuels include chronic lung and cardiovascular disease [1]. Based on the evidence for an association between household air pollution and child ALRI [2], and evidence from ecological and toxicology studies regarding the health effects ambient air pollution in all ages [4], an association between household air pollution and ALRI in adults is biologically plausible but lacks a convincing evidence base. Although the Global Burden of Disease study 2013 incorporated adult ALRI as an outcome for household air pollution exposure, the risk estimate was derived from an integrated exposure-response curve for PM 2.5 (particulate matter <2.5μm diameter) from ambient and household air pollution, and tobacco smoke which has limitations [1,5]. It is important to determine whether or not there is an increased risk of ALRI associated with household air pollution in adults since the attributable risk is potentially high and the burden greatest in people living in low and middle income countries (LMICs). ALRI is a common cause of hospitalisation among adults in LMICs with high associated morbidity and mortality [6]. Mortality among adults with pneumonia in sub-Saharan Africa is around 10%, with over 50% of deaths seen in people under 35 [7][8][9].
To ensure that public health interventions are appropriately directed, the true burden of disease needs to be clarified [10,11]. We systematically reviewed the literature, including studies of any design, to synthesise the evidence base for the relationship between household air pollution exposure from domestic solid fuel use and adult ALRI worldwide. We describe the current literature, identify knowledge gaps and future research opportunities.

Methods
This systematic review is registered with the Centre for Reviews and Dissemination (Registration number: CRD42015028042). The full protocol is available at http://www.crd.york.ac.uk/ prospero.

Eligibility criteria
Eligibility criteria for studies based on participants, exposures, outcomes and study designs were used to identify studies that provided an effect estimate for household air pollution on adult ALRI in the form of a relative risk, odds ratio (OR) or hazard ratio, or data allowing calculation of an effect estimate. English language papers and non-English language papers (including French and Spanish) where a translation was available were included. Translation was not available for Chinese literature.

Participants
All adults were eligible, without geographical restriction. Studies exclusively focussed on children (<18 years) were excluded, but to avoid the exclusion of potentially useful data on adults, studies that included low numbers of under 18-year olds within a primarily adult population were included.

Exposure
Exposure was defined as air pollution from indoor burning of any solid fuels-including wood, charcoal, animal dung, crop residues and coal-for household purposes. We included studies that quantified exposure through direct measurement of specific pollutants, questionnaires regarding exposure history, comparison of groups exposed to types of exposure (e.g. different stove types), or before and after an intervention to reduce exposure. Studies examining outdoor air pollution, occupational exposures, non-fuel combustion sources or non-solid fuels were excluded.

Outcomes
The primary outcome of interest was ALRI including pneumonia, acute bronchitis or bronchiolitis. However we included studies that defined the outcome as "acute" or specified duration of less than 14 days, even if infection was not confirmed, on the assumption that acute respiratory illnesses in the absence of underlying disease would likely be infectious in origin. Studies reporting illnesses lasting longer than 14 days were excluded, to avoid inclusion of chronic diseases. Studies examining exacerbations of chronic conditions such as asthma or COPD were included if an acute infectious exacerbation was defined. Studies that did not distinguish between upper respiratory tract infections and ALRI were excluded.

Study designs
Individually-and cluster-randomised control trials; controlled before-and-after trials; cohort studies; case-control studies and cross-sectional surveys were included.

Search strategy
The databases listed in Box 1 were searched on 17/12/2015 using the search terms and Boolean phrases listed in Box 2. An example search strategy is shown in Box 3. Searches were limited to papers published from 01/01/1960 onwards, and to human studies. Identified papers were imported into EndNote X7. Following removal of duplicates, titles of identified papers were reviewed for eligibility by two authors. The first 20% were all reviewed in duplicate by two independent authors to check for agreement. The remaining 80% were reviewed by one author with 10% of these being cross-checked by a second author to ensure ongoing agreement.
Abstracts of the selected titles and then full text of the selected articles were reviewed for eligibility, according to the selection criteria. Abstracts and full text articles were all reviewed in duplicate by two independent authors-HJ reviewed all abstracts and full text articles and the second review was performed by either HS, DH or GM. Where discordant decisions were not resolved by discussion, the decision was made by a third author. If the full text article was not available in English, translation was sought to determine eligibility.
Alternative sources were also searched for eligible studies: www.who.int/trialsearch, www. clinicaltrials.gov and EAGLE (European Association for Grey Literature Exploitation, www. opengrey.eu). Reference lists of all selected papers were reviewed for any potentially eligible titles. For studies where relevant details were not available, authors were invited to supply information if their contact details were available.

Data extraction
Data were extracted from the selected papers using a previously piloted data extraction form, from which a summary table and narrative synthesis were produced. Where effect estimates were not provided, these were calculated from the published data.

Quality assessment
Studies that quantified exposure by direct household or personal measurement of specific pollutants were considered to provide a higher quality of exposure assessment than those that relied on self-reported exposure. Studies using prospective assessment of ALRI by a physician or trained health care worker using predefined criteria or diagnostic investigations were considered to be of higher quality than those reliant of self-reported episodes of ALRI.

Risk of bias assessment
Risk of bias was assessed using the Liverpool Quality Assessment Tool, which reviewed the study methodology in four domains-subject selection, exposure assessment, outcome assessment and adjustment for confounders-and assigned quality ratings (low, moderate or high risk of bias) [12]. We adopted this instead of the Cochrane Collaboration's tool for assessing risk of bias due to the variety of study designs identified. Publication bias was not assessed due to the small number of papers identified.

Data analysis
Methodological heterogeneity in exposure and outcome assessment precluded a meta-analysis, but a narrative summary of the selected papers and a summary of findings table were produced.

Results
Database searches identified 8605 records and 331 records were identified from the alternative sources set out in the methods. After removal of duplicates 4617 records remained. The number of titles, abstracts and papers reviewed was 4617, 513 and 72 respectively, as summarised in the PRISMA flowchart (Fig 1), with the reasons for exclusion at the abstract and full text stages (see S1 File for PRISMA Checklist). Agreement between authors was 95-96% for the first 20% of titles reviewed, and 95-98% for the 10% overlap of remaining titles. It was not possible to locate 5 selected abstracts [13-17] and 2 selected full text papers-both published in Chinese [18,19]-for review. Two authors were contacted for further details about their recent or ongoing studies identified at www.clinicaltrials.gov. One study is a prospective cohort study of febrile adults in Tanzania for which recruitment has been completed but the data are not yet available (NCT01947075). The second study is a randomized trial of smoke reduction interventions in Native American populations, but we were unable to obtain further details (NCT02240069). In addition, we are conducting a case-control study of the effects of air pollution exposure and chronic respiratory disease on pneumonia risk in Malawian adults [20].
Ten papers met the criteria for inclusion; data from 2 of these were duplicated in other selected papers [21,22], and therefore not extracted. The remaining 8 papers included 4 crosssectional [23][24][25][26], 2 cohort [27,28] and 2 case control studies [29,30]. The articles covered study locations from 8 countries across 4 continents. The largest study was a retrospective cohort study from China, which reported on the causes of death in 42,422 farmers [28]. The main findings of all included studies are summarised in Table 1, which demonstrates the heterogeneity of study participants, exposure definitions and outcome definitions.  Most studies were based in rural settings and recruited participants from communities. There were 3 exceptions, which were hospital or health-centre based studies, conducted in urban areas of Canada, Mexico and Serbia [25,29,30].
Most studies predominantly or exclusively recruited females (accounting for 48-100% of participants within studies). In the study from Tanzania, the cooks were mainly female but the gender and age of the unexposed group were not reported [24].
Except for the study from Canada, which restricted recruitment to individuals over 65 [29], and the study from China [28], in which deaths occurred between 25 and 80 years of age, all studies recruited younger adults. Five studies restricted participants to adults over 18, whilst the 3 remaining studies-from Sierra Leone, Tanzania and Kenya-included younger participants but were conducted in predominantly adult populations.
One study evaluated coal use (Shen et al) and the 7 others evaluated biomass fuel use, predominantly wood. Four studies measured levels of household pollutants directly, but only 1 presented an estimate for the effect of pollutants on ALRI risk. In Kenya, Ezzati et al. measured particulate matter <10μm diameter (PM 10 ) for 14-15 hours per day for 200 days in 55 households, and demonstrated a dose-response relationship between exposure to PM 10 and ALRI risk [27]. The authors report that this finding was supported by analysis with a continuousexposure variable and inverse-quadratic relation, although have not published these results. All studies monitoring air pollution levels measured PM 10 rather than PM 2.5 . This may misrepresent the true harmful exposure, as smaller PM 2.5 particles are inhaled deeper in to the lungs than PM 10 and are therefore thought to be more pathogenic [31].
The 7 studies that did not provide an effect estimate for measured pollutants all used questionnaire data to classify exposure to household air pollution, based on a variety of selfreported measures including fuel or stove type, quantification of fuel use, ventilation, cooking frequency and location.
Two studies defined the outcome using prospective health care worker diagnosis of ALRI; in Canada, hospitalised cases had symptoms or signs, and radiological changes consistent with pneumonia [29]; in Kenya, the outcome was defined as nurse-diagnosed ALRI following review and physical examination at 1-2 weekly visits over a 2 year period [32]. A third study, from Nepal, included a physician's review to diagnose ALRI, but no further definition was provided and it is unclear whether retrospective diagnoses were included [23]. The study from China used retrospective review of death certificate records, which had been completed by physicians on the basis of clinical and radiological investigations [32]. Similarly, the study in Mexico used a retrospective review of hospital records by specialists to identify cases [30]. The remaining studies used self-reported symptoms or past diagnoses, with varying recall periods, to define the outcome [24][25][26].
Six of the 8 studies made some adjustment for potential confounders in their analysis of the effect of household air pollution on ALRI [23,25,26,28,30,32], but several studies failed to make adjustments for important potential confounders, such as economic status and smoking.
The effect estimates quantified by the studies are shown in Table 1 and in a Forest Plot (Fig  2). There were inconsistent findings across the studies. Whilst the 2 cohort studies have marked differences in study design, they demonstrated statistically significant harmful effects of household air pollution exposure on ALRI after adjustment for confounders. A dose-dependent relationship between PM 10 and ALRI was detected in Kenya, (OR = 7.10, (95% CI 2.26 to 22.32) for exposure to >7000μg/m 3 PM 10 (highest exposure category) vs. 200-500 μg/m 3 PM 10 ) [32], and a reduction in risk of pneumonia deaths was observed in participants using an improved coal stove instead of a traditional coal stove (OR = 0.49 (95% CI 0.31 to 0.80) and 0.53 (95% CI 0.32 to 0.88)) in Chinese men and women respectively [28]. The studies from Mexico and Tanzania both reported harmful effects of household air pollution exposure in univariate analysis [24,30], but these were no longer significant after adjustment for confounders in the Mexican study. The studies from Nepal, Serbia and Sierra Leone did not identify a significant association between household air pollution and risk of ALRI [23,25,26]. The study from Canada found that fireplace use was associated with a reduction in the risk of pneumonia, although this was unadjusted [29].
All of the case-control and cross-sectional studies were found to be at high risk of bias in at least 2 out of the 4 domains assessed, leading to an overall assessment of "moderate/high" or "high" risk of bias. Exposure assessments were weak. The cohort studies were of a generally higher standard, although their risk of bias was moderate in at least 2 out of 4 domains.

Discussion
This paper synthesises the current evidence base for the relationship between household air pollution and adult ALRI, identifying 8 eligible studies that have quantified this relationship. Despite a paucity of available studies, the available data provides some evidence for an increased risk of adult ALRI from exposure to household air pollution. The review has highlighted disparities between the findings of the studies, although these are unsurprising given the methodological heterogeneity seen. Methodological limitations were also noted, including poor exposure and outcome classifications, and potential biases including selection and recall biases. Although suggestive of an association, the current evidence is not sufficient to make a direct assessment of the potential impact of household air pollution on adult ALRI.
Methodological differences in exposure classification are a likely source of variation between study findings in this review. Indirect classification of exposure to household air pollution was common, and may have resulted in exposure misclassification. Accurate measurement of exposure to pollutants such as PM 2.5 is challenging in the field. Monitoring is becoming more practical, with a variety of devices now available, which will benefit future population studies of health impacts such as ALRI. Exposure, however, needs to be measured at repeated intervals to accurately classify levels for diseases with a long latency period from defined exposures and this remains a challenge.
With regards to major sources of bias and error, the quality of studies identified was varied. Two studies used retrospective records of patients, and none of the prospective studies reported response or refusal rates of participants, so may have been subject to selection bias. Several studies used some form of random sampling but the cross-sectional study from Serbia only recruited women who attended a health centre so the findings are not generalisable to the wider population. The case-control study from Mexico used patients with otolaryngeal disease as controls, but as household air pollution can also be a risk factor for this outcome [3], this may have biased the findings towards the null. Only 1 study directly measured exposure to household pollutants [32]. All other studies relied on questionnaire responses regarding exposures, which are prone to recall bias, with varying recall periods and often poor definitions of exposure. Similarly, several studies relied on potentially biased and inaccurate recall of previous illness; outcome misclassification may have diluted the true effect. Three studies used prospective assessment of patients by a health-care worker for the outcome classification, although the quality of definitions used varied. Two studies made no adjustment for confounders, and other studies failed to adjust for some important confounders such as smoking, and so their findings may be misleading. Four studies were cross-sectional and are therefore unable to establish temporality between exposure and outcome.
Although the literature was previously reviewed as part of the Comparative Risk Assessment conducted for the GBD 2010 [33], this review provides an update, adding 5 studies to the evaluation. This review also provides a narrative summary of the current evidence which was not previously available, and a critical appraisal of methodological limitations. The review included a comprehensive search of published bibliographic databases-including databases from Africa and Latin America-and available grey literature, including contact with known researchers in the field undertaking relevant research. A limitation of this review is that we were unable to include literature published in Chinese. Due to the paucity of evidence, publication bias was not assessed.
Diseases associated with household air pollution disproportionately affect those living in poverty, who rely on solid fuels to meet their energy needs [34]. To consolidate GBD estimates -which currently use modelled data from other sources of air pollution-we set out to quantify the association between household air pollution and adult ALRI. Whilst it is likely that the burden of disease from household air pollution in adults includes ALRI, the published literature does not provide definitive evidence of this. Further research, using robust direct exposure measurement and accurate classification of outcome, is required to improve the evidence base. Exposure to household air pollution is preventable but resources are limited in low-income populations with competing health priorities: with high quality evidence of the true scale of the problem and cost-effectiveness of interventions, resources to reduce the global burden of disease can be effectively allocated.