Societal Protection and Population Vulnerability: Key Factors in Explaining Community-Level Variation in Fatal Fires Involving Older Adults in Sweden

We employed an ecological design with Swedish municipalities as the unit of analysis. Our outcome measure was the number of fatal fires involving at least one older adult (65 + years) recorded over the period 1999–2015 (16 years) per 100.000 person-years of observation. Our study relies on data from official sources. The outcome data was obtained from the Swedish Civil Contingencies Agency, and the population denominator used to quantify rates per person-years was obtained from Statistics Sweden. To collect relevant covariate data, we screened available databases and registers for municipality-level data relating to rescue service performance or organizations, the health of the older population and other municipal factors that may directly or indirectly affect fire mortality rates. Specifically, we began by reviewing the Municipality and County Council Database (Kolada by Swedish acronym), which collates data from multiple official sources and surveys conduct in Sweden. We also screened databases held by relevant authorities, including the Swedish Civil Contingencies Agency, The National Board of Health and Welfare and Statistics Sweden. After compiling a list of factors available in these databases, we discussed their potential relation to fire mortality rates at the municipal level with the reference group consisting of eight leading fire safety experts and practitioners familiar with the Swedish context. After this discussion, we arrived at the covariates, defined and detailed along with their sources, in Table 1. When possible, we extracted data for the period 1999–2015. In some cases, data was unavailable for the entire period or only for a period that partially overlaps with our outcome data. For these variables, we extracted data from as many years as possible within, or close to, the period 1999–2015 (see Table 1 for details).

Many covariates were highly collinear (see Appendix: Table 4), so we opted for a dimensionality reduction approach to reduce the number of variables into principal components. Specifically, we used principal components analysis (PCA), followed by Varimax rotation [25]. We selected the number of components to retain by visual inspection of a scree plot (see Supplementary Figure A1 for details), after confirming that the data was suitable for PCA (Kaiser–Meyer–Olkin’s measure of sampling adequacy: .783; Bartlett’s test of sphericity: χ²(45): 948.7, p < 0.001). We used Poisson regression to quantify incidence rate ratios for each covariate and to analyze how and if the identified principal components relate to the outcome. The PCA was conducted in SPSS Version 25, and the results were exported to Stata version 15.1 for the remaining analyses.

A total of 669 fatal fires involving at least one person over the age of 65 years was recorded in the data, which corresponds to an expected rate of 2.36 fatal fires 100.000 person-years of observation. However, the observed rate varies from 0 to 16.92 across municipalities (Fig. 1).

Table 2, Column I shows how each covariate is associated with the rate of fatal fires involving older adults without accounting for the other factors in the data. The unadjusted estimates show that the share of the population living in urban areas (“urbanization”) is negatively correlated with mortality rates. On the other hand, we observed positive correlations with rescue service response times and fall injury incidence rates, as well as the shares of the older population living alone, having home care or living in nursing homes. Together, these results imply that rescue service organizational factors and context (e.g., distance) matter and that the vulnerability of the population does as well.

The coefficients from a multiple regression including all covariates were generally imprecise. We present the results from this analysis in Table 2, Column II, but advise caution in interpreting the results for each factor individually given the strong collinearity between many of the factors in the data. For instance, Spearman’s rho for the correlation between urbanization and median response time is − 0.7, probably because the travel distance is smaller in urban areas. It is therefore difficult for the model to separate the effects of these factors in our data. We found correlations of similar magnitude for the population variables, indicating that they are proxies for the same underlying phenomena (e.g., the vulnerability of the population) (Appendix: Table 4), which motivates a dimensionality reduction approach.

Using the 256 municipalities with complete covariate and outcome data, the PCA resulted in two components which we interpreted as (1) (municipal) urbanization and (rescue service) efficiency and (2) population vulnerability (the factor loadings for each covariate is presented in Table 3, and the scree plot used to determine the number of components is presented in Supplementary Figure A1). Together, these components explain 56% of the variance in the covariate data.

Figure 2 shows how the two components, divided into quintile groups, relate to the outcome while keeping the other component constant at its median value (estimated using a Poisson regression model with the two [quintile-grouped] components as independent variables). As might be expected, the results show that higher urbanization and efficiency is associated with lower fire mortality rates, whereas the vulnerability of the older population is associated with higher rates. In terms of magnitude, the rate difference is − 0.8 per 100.000 person-years between the highest and lowest quintile group on the urbanization component (− 33% in relative terms) and + 1.1 per 100.000 person-years between the highest and lowest group on the vulnerability component (+ 56%). The difference is statistically significant (p = .008 in both cases).

As a final analysis, we plotted the components in a two-way scatter plot, including reference lines that divide the sample of municipalities into four groups: (1) high urbanization-low vulnerability, (2) high urbanization-high vulnerability, (3) low urbanization-high vulnerability and (4) low urbanization-low vulnerability (Fig. 2). We then calculated the observed mortality rates per group. As shown in the plot, we found the lowest rates in the high urbanization-low vulnerability group and the highest in the low urbanization-high vulnerability group. In fact, the rate reduction from the high-risk group to the low-risk group is larger than the rate reductions in the “medium-risk” (lower left and upper right) quadrants combined, which implies an interaction effect in the sense that societal protection efforts may be more efficient in low vulnerability populations (Fig. 3).

Overall, the results from this study imply that there are (at least) two separate, underlying components (individual vulnerability and societal protection) that determine fire-related mortality rates among older adults at the local level. While it is ill-advised to infer causality from observational data, prior evidence supports the plausibility of these two mechanisms. Importantly, the results suggest that while societal protection, through for example short response times, does lower the fire-related mortality rates, they do not eliminate the risk-increasing aspect of individual vulnerability.

From a public health perspective, the importance of sociodemographic risk factors and individual vulnerability in explaining inequality in mortality is well established, as are the necessary preventative measures, for example increasing the general educational level, increasing health literacy and improving access to healthcare. General public health measures, such as those aimed at increasing physical activity and sense of coherence in the population, may also be potentially important interventions, provided that they lead to lower individual vulnerability and increased co-living.

While decreasing individual vulnerability is an important long-term goal, not merely from a fire-related mortality perspective, such interventions take time. Therefore, in order to reduce fire-related mortality rates amongst older adults in the short term, other interventions need to be promoted. Previous studies have shown the importance of rescue service response times, [22] and although there is evidence that the presence of smoke alarms can compensate for slower response times [26], for many older adults escaping a fire can be exceedingly difficult. Therefore, in terms of the older adult population, response times is likely to be crucial, especially considering that a large percentage of older adults in Sweden have functioning smoke alarms [27].

Despite the importance, however, in order to accomplish a high level of service as required, the societal costs risk being high. This is further complicated by the fact that Sweden, in similarity to other high-income countries, has a high urbanization rate, particularly among younger, well-educated individuals [28]. This urbanization pattern has a dramatic effect on the financial status of rural municipalities [29] meaning that rural communities are therefore more likely to be forced into cost-cutting. The municipal organization of rescue services, as well as public health interventions, could thereby be an organizational barrier to reducing fire-related mortality among older adults.

There are some limitations to our study worth mentioning. Firstly, the urbanization process could also have other implications on the fire mortality risk that have not been possible to include in this study. For example, sprinkler systems are more likely to exist in newer and larger multi-home buildings than in one-family homes [30] and given that high-rise and newer buildings are predominantly built in urban areas, it could be more likely that passive safety is higher in urban areas than in rural areas. However, according to the Swedish Contingency Agency, only 0.4% of residential properties in Sweden have sprinkler systems [31]. As such, this will unlikely affect our results.

Secondly, a potential problem with our data is that it is collected from secondary sources and may therefore vary in quality. The mortality data also covers a longer period than many of the covariates, which may introduce some bias if there are strong municipality-specific time trends. However, the precision of the estimates would be too low to provide any meaningful information if we had used the same periods as the covariates. It also appears unlikely that the observed associations can be attributed entirely to such artefacts, especially given the theoretical plausibility of the observed associations. However, we cannot rule this out, and replication in other settings is therefore strongly warranted. Our results motivate further research and focus on societal efforts both in the short- and long-term perspective. On a broader scale, the results also imply that public health agencies and rescue service organizations could benefit from working together to decrease fire mortality rates at the local level, especially in communities with vulnerable populations.

Our analyses suggest that the level of individual vulnerability in a municipality and the level of societal protection are equally important in explaining municipal differences in fire mortality amongst older adults. From a societal prevention perspective, the results suggest that while improved societal protection would reduce fire-related mortality rates, merely focusing on these aspects will not eliminate the risk-increasing aspect of individual vulnerability. Not least, this is apparent in regards to fire-related deaths that are related to smoking and individuals who are alcohol-, drug- or disability-impaired. These cases are often the result of small, undramatic fires meaning that sprinklers and/or fire alarms (and thereby the response of rescue services) are relatively ineffective [32]. Therefore, both societal protection and individual vulnerability needs to be improved in order to reduce fire mortality rates.