We found that several exposures within the housing and indoor home environment may increase the risks of RTIs in the adolescent population. In particular mold, dampness, and candle-burning, but also other housing-related indoor exposures, were associated with an increased risk of RTIs. Our results also indicate that children living in owned houses had several other favorable exposures that may decrease the risk of RTIs (i.e.increased number of rooms, lower household density and lower gas stove usage and mold and dampness). On the whole, we provide evidence for potential effects of microbial and pollutant exposures on rates of RTIs in the adolescent population.

We find associations between residential mold and dampness and RTIs, corroborating previous studies’ findings [9]. Associations between mold and dampness and most RTIs were observed regardless of model and analytic strategy. We believe this strengthens the body of evidence for the role of residential mold and dampness on respiratory infectious morbidity in children, but also suggests that at presumably lower residential exposure levels this potential effect is small. In this study we were able to address some of the research gaps in the role of mold/dampness in RTIs, by investigating RTIs specifically and adjusting for most relevant confounders [16]. Prospective data collection would increase certainty about the temporality of the observed associations. The biological mechanisms behind the findings are as of yet unclear, however, Fisk et al. suggest among several candidate mechanisms an immunomodulatory effect of mold that increases susceptibility to infection [9].

Our results generally indicate no increased risk of RTIs in children with fireplace usage, which may have several explanations. A recent randomized controlled trial (RCT) confirms previous reports of increased indoor pollution with wood-burning, demonstrating a dose-response effect on lower RTIs with increasing PM2.5 [24]. Most of the fireplaces included in the RCT were of low or medium quality. We did not have data on the quality of the fireplaces in our study, but high-grade fireplaces are expected to lead to lower indoor pollution while being a source of heating [25]. In a community intervention trial investigating changes in RTIs with substituting wood stoves for improved wood stoves or other heat sources, substantial decreases in PM2.5 at the neighborhood level were observed, but differences between homes with and without changed stoves were not observed [25]. This might indicate that the neighborhood ambient air pollution from wood stoves in the community is more influential than the in-home pollution. At the individual level, a similar lack of association to what we observed has been observed for pulmonary function in Danish adults, suggesting either minimal effects of particle pollution levels from fireplaces within the home on pulmonary function or potential interactions with other sources of particle pollution [26]. Most, if not all, sources of particle pollution in the home have no beneficial effect on the immune system, but fireplaces are during winter months for some a critical source of heating. The role of cold exposure on susceptibility to RTIs is potentially confounded by other factors, but if truly causal, averting cold exposure might counterbalance some of the negative effects of particle pollution from fireplaces [14, 27]. Additionally, high-grade fireplaces potentially enhance ventilation by mechanically extracting indoor air and replacing it with outdoor air. However, other negative health effects to residents within a residential area with particle pollution from fireplaces and the global warming effects of carbon emissions should continue to inform public health recommendations for safe and climate-friendly heating sources, even though these findings do not support a detrimental effect of the studied fireplaces on RTIs in children [28].

Our results suggest that candle-burning during the summer increases rates of several RTIs in a dose-response relationship, although these results must be seen in light of the inverse association and null associations for winter candle-burning. Interestingly, two studies in an adult Danish cohort also found no negative effects of candle-burning during winter on pulmonary function [26] or risk of respiratory disease events [29]. The lack of an exposure contrast, i.e. the fact that candles are used in almost all Danish households during winter time, may explain our finding [8].

Alternative explanations behind our findings regarding the conflicting results on candle-burning deserve elaboration. While we observed small decreased rates for winter candle-burning, the confidence intervals generally included the null and may suggest unknown confounding. At present, we suggest our findings for summer candle-burning are harder to explain in the context of unknown confounding, although this certainly cannot be excluded. The nearly linear dose-dependent relationship between summer candle-burning and RTIs may indicate a true causal effect, or a factor which closely correlates with summer candle-burning and increases rates of RTIs exists, which we have not identified in our data or literature. Although socioeconomic position is associated with candle-burning in general, in this population frequencies were substantively quite similar across strata, compared to other exposures [8]. It would be interesting to see studies replicating or refuting these associations with granular exposure data, prospective data collection, and more certain measures of infection (asymptomatic, symptomatic and severe).

While the ‘farming effect’ in the context of allergies has been proposed to be due to microbial exposures in the development of respiratory disease in children, [30] our results are also compatible with potential confounding or mediation by rurality. Children living in farmhouses are likely to live in more rural areas, go to smaller schools, and meet respiratory pathogens less frequently, compared to those in more densely populated cities. This might explain similar findings for children living in farmhouses with livestock as for those living without livestock, but does not necessarily exclude a common microbial cause that is unique to rural living. Indeed, previous studies suggest a greater microbial diversity and abundance in homes of pig and cow farmers that is not only attributable to direct microbial transfer from the farmer’s own stable [31, 32]. We would be interested to see if the common cold finding is confounded, as our primary results suggest, or points to a small, non-generalizable effect of microbial exposures unique to farm living. Interestingly, we found no inverse association with dog and cat exposure and an increased relative rate of self-reported influenza.

These results suggest that certain exposures in the home are relevant to consider in the primary prevention of RTIs in children. We report results for children at a point in life where their risks of severe RTIs are at a relatively lower point than earlier or later in life. Whether these potential environmental factors influence rates of RTIs during infancy, when the burden of RTIs is much higher, would be relevant to investigate. If these associations are causal effects, large reductions in absolute numbers of symptomatic RTIs could be achieved despite relatively small effects per exposure. We only considered the home and we could speculate that the exposome, of which we only measure a portion, could have even more meaningful impacts when addressed as a whole. The case for improved housing may include considerations of respiratory burden.

A major strength of our study is the large sample size with data on specific RTIs as opposed to respiratory symptoms alone – however it should be noted that these were self-reported. Many of the most relevant housing and indoor home environment exposures were also available and with sufficient granularity to investigate a dose-response relationship. We were also able to gather both self-reported and registry-based data on a large number of covariates.

Despite this, certain limitations of these data must be highlighted. First, these data are susceptible to misclassification due to misattribution of a set of symptoms to a specific illness, [33, 34] which would be less likely with a doctor-diagnosis – even then, a general practitioner (GP) who performs no diagnostic testing will not necessarily be free of misclassifying the responsible pathogen. The difficulties in distinguishing between illnesses causing influenza-like illness has recently been highlighted by Spencer et al. [34] and in a meta-analysis positivity for influenza viruses ranged only between 11 and 56% in children presenting with influenza-like illness [33]. Additionally, in our study, the one outcome which specified a doctor diagnosis (pneumonia), was also the most severe, and associations observed may not be generalizable to milder disease course. Our primary concern is misclassification with other respiratory pathogenesis unrelated to infection. This is a possibility with the self-reported nature of our outcomes, but may be less likely than misclassification between RTIs.

In a similar vein, we capture illness severity in the range most commonly experienced in this age range, but asymptomatic infections or mildly symptomatic infections lasting shorter than 3 days were not included in counts of RTIs. Asymptomatic infection with respiratory pathogens is common [35, 36]. A potential explanation for contrasting associations between self-reported common cold and influenza for the household density factor may be related to differences in effects of environment on mildly symptomatic and moderate-severe symptom RTIs. In this case, this may either illustrate true differences in determinants of disease severity or data limitations regarding our outcomes of interest.

Although most estimates are precise and with narrow CIs, consideration of several sources of bias is also warranted. We attempted to deal with confounding bias through the use of both data- and theoretically-driven selection of observable covariates. We additionally conducted exploratory analyses with an agnostic approach to exposure and covariate clustering. Several covariates of note that we did not adjust for were neighborhood characteristics like ambient air pollution and school indoor environment, since these were not available. Levels of ambient air pollution in a Danish study were inversely correlated with fireplace usage, positively correlated with gas stove usage without an exhaust hood, and not correlated with candle usage [26]. These correlations for gas stove usage and fireplace usage are likely partially related to socioeconomic position and building characteristics which we were able to adjust for. Unmeasured confounding may explain positive associations for gas stove usage and a lack of association for fireplace usage, although we expect much of this variation could be captured in the variables we adjusted for. We also did not adjust or include pollution from cooking practices themselves (e.g. interactions and independent effects of type of cooking and food type cooked). Previous research has shown that cooking itself (especially certain foods and preparation methods) is a significant source of indoor air pollution [37]. We presume that detailed data on socioeconomic position likely captures much of the neighborhood characteristics, school characteristics, and (albeit less so) food-related exposures of relevance (e.g. such as meat preparation, amount of cooking vs. frying). The role of ventilation and air exchange in RTIs is increasingly recognized, however, we did not have available data to examine natural or mechanical ventilation. There may be residual confounding due to incomplete covariate characterization, as well as unknown or unmeasured confounding variables we did not adjust for. Both negative and positive confounding might affect our results.

Since the main data in this study are of cross-sectional nature, a major concern is reverse causality. We think that the results of sensitivity analyses only including children that have lived in the same household, and therefore most likely with the same exposure pattern throughout childhood, were reassuring, as they showed a similar pattern to the main results. We believe it is rather unlikely that reverse causality would cause some of the observed associations, however – families with a high burden of RTIs in children would not be more likely to create a moist/damp home environment to compensate for this burden. We do not think that confounding by hereditary factors would explain our results, although we cannot fully exclude this possibility. It is possible that parents with respiratory diseases use an available fireplace less often, however, this mechanism would attenuate any real positive association between fireplace use and RTIs. Additionally, for parental asthma this explanation is not clearly reflected in our data. Candle-burning and gas stoves were not widely acknowledged at the time of the study as indoor pollutants, so reverse causality may be less likely for these associations. Pet ownership and farmhouse living may be related to asthmatic parents’ choices, however, this may go in both directions. To take such potential confounding into account, we adjusted for prevalent allergic and asthmatic disease in parents. Future studies with a better ascertainment of timing of exposure and changes in behavioural factors may clarify the temporal relationship between disease burden and rates of RTIs. It is an additional concern that both outcome and exposure were reported by the same respondents (parents), which could result in differential misclassification. Future studies with prospective data collection on ascertained RTIs will be important in corroborating our results.