An interrupted time series analysis of the cardiovascular health benefits of a coal coking operation closure

A growing body of research has documented associations between air pollution exposures and increased adverse health outcomes, providing a scientific basis for the setting of air pollution standards and air quality improvement policies (Rich 2017, Schraufnagel et al 2019b). The potential health benefits of these regulations are usually estimated based on concentration-response models from existing observational epidemiological studies at ambient levels, but the resulting health benefits of air quality improvements have not yet been as extensively validated with documented after-the-fact outcome evidence (Boogaard et al 2017). This quandary is largely due to the usually long time taken to implement such control programs, which can then be confounded by factors such as population composition and/or health care coverage changes. Therefore, to inform and support further regulatory actions, demonstration of human health benefits resulting from implementations of real-world control policies is of emerging interest and necessity (Bell et al 2011). Accountability studies, which quantify the air quality improvement and health benefits of abrupt air pollution interventions and regulatory actions, have therefore become increasingly important (Henneman et al 2017, Rich 2017). These studies, which leverage air and health outcome records during abrupt pollution interventions, or 'natural' experiments, provide opportunities to rigorously test the causal link between air pollution and health (Boogaard et al 2017, Henneman et al 2017). Our research provides such a novel real-world test of the hypothesis that reductions of fossil-fuel related industrial air pollution result in associated reductions in adverse cardiovascular health effects.

The closure of the Shenango, Inc. coke plant, located in Pittsburgh, PA on Neville Island in the Ohio River in Allegheny County, provides such an opportunity to conduct accountability research to investigate the pollution profile and health outcome changes associated with a major coking plant shutdown. The metallurgical coke used in iron and steel industry is produced by destructive heated distillation of coal in oxygen-free coke ovens (Weitkamp et al 2005, Mu et al 2013). Coke plant operations are known to generate high emissions of hazardous air pollutions including particulate matter, sulfur dioxide, volatile compounds, and trace elements related to coal combustion (Ghose et al 1999, Weitkamp et al 2005, Khare and Baruah 2011, Mu et al 2013, Saikia et al 2015). Past epidemiological studies of such sources have found that occupational exposures to coke oven emissions are associated with various adverse health outcomes, including chronic obstructive pulmonary diseases (Hu et al 2006), lung function decline (Wang et al 2016), hypertension (Yang et al 2017), cancers (U.S. Department of Health and Human Services 2016) and neurobehavioral function impairments (Qiu et al 2013). High levels of heavy metals and metalloids detected in the environment around coking plants could also potentially induce health risks in nearby local communities via multiple biological pathways (Cao et al 2014).

The general association between air pollution exposure and negative health effects is well-established, and studies have found that the fossil-fuel-combustion-related emission sources contribute to large health burden (Yang et al 2019, Vohra et al 2021). However, despite the known hazardous air pollutants released into the atmosphere by such coal-processing facilities, and past statistical associations between those industrial point source air pollutants and adverse health effects (Bhopal et al 1994, Aylin et al 2001, Parodi et al 2003, 2005), there is currently only limited documented epidemiological evidence for an association between elevated pollution exposures specifically from an operational coke works and negative health outcomes. This paucity of research can primarily be attributed to the difficulty in statistically differentiating the effects of the plant emissions from other air pollution sources, as well as health-impactful factors in the vicinity of the plants (such as the commonly low income status of nearby residents), as well as unavailability of spatially detailed health and/or source-specific air pollution exposure data (e.g. via source apportionment of particulate matter concentrations) (Thurston et al 2011). This abrupt coking plant closure analysis avoids these potentially confounding issues, allowing a clear discernment of the health effects of such a coking facility's air pollution.

Before its closure, the Shenango, Inc. coke plant was a significant point emission source of air pollution in Pittsburgh, PA (Kelly and Fischman). Based on the 2011 point source emission inventory (Allegheny County Health Department Air Quality Program 2019), this shutdown in January 2016 caused reductions of 97.3 tons yr⁻¹ of primary PM_2.5 and 901.6 tons yr⁻¹ of PM_2.5 precursors (e.g. SO₂), accounting for 3.5% of Allegheny County's annual point source emissions. With this precipitous alteration of the local air pollution profile, the plant's vicinity has therefore become an ideal case-study to discern whether the air pollution improvement benefited the health of the surrounding community. Already, one preliminary analysis found that, after the shutdown of the Shenango, Inc. coke plant, the annual rates of overall hospitalization and emergency department visits for both cardiovascular and respiratory diseases significantly decreased (Brink et al 2019). However, that simplistic analysis has been critiqued as insufficient evidence to conclude that the drop substantially resulted from the air pollution reduction at the coke plant, nor to identify the pollutant constituent that may have caused such a change, without a more detailed analysis of the health and air pollution data, including versus an unaffected control case during the same period (Marusic 2018). A more comprehensive analysis is needed to facilitate a fuller assessment of the potential health benefits of a coke plant air pollution cleanup.

In this study, we seek to address the plant closure evaluation challenges by applying a more rigorous interrupted time series (ITS) analysis of health data collected after, versus before, the closure. As one of the strongest quasi-experimental designs to allow the discernment of health impacts in such situations like dramatic air pollution emission changes, the ITS analysis has become increasingly frequently used for health outcome evaluations of exposures and public health interventions (Bernal et al 2017, 2018, Yinon and Thurston 2017, Burns et al 2019, Turner et al 2020). This study therefore applies statistically rigorous ITS models to test our hypothesis that the substantial reduction in local air pollution induced by the closure of the Shenango, Inc. coke plant was accompanied by immediate and/or longer-term declines in cardiovascular disease hospitalizations and emergency department visits.

2.1. Study period and sites

2.2. Air monitoring data

2.3. Inpatient hospitalizations and emergency department visits data

2.4. Study design and statistical approach

The ACHD local daily mean air quality data and health data were assessed for potential changes in trends before and after the closure of the coking plant using the ITS method. The size of the intervention effect is estimated using a segmented regression analysis of ITS (Bernal et al 2017):

$$\begin{equation*}{{\text{Y}}_t} = {\beta _0} + {\beta _1}T + {\beta _2}{X_t} + {\beta _3}P{X_t} + f\left( t \right)\end{equation*}$$

where Y_t is the health outcome at time t; T is the time elapsed since the start of the study; β₀ represents the baseline level at T = 0; β₁ is the trend before plant closure; β₂ indicates the step-level change in health outcome at the time of plant closure, as X_t is an indicator variable for plant closure (0 for pre-closure, 1 for post-closure); P indicates the time passed after the closure occurred, β₃ is the health outcome trend slope change during the time following the closure, and $f\left( t \right)$ is a term controlling for seasonality using splines (Bhaskaran et al 2013). Ambient air quality data from the ACHD and EPA CSN monitors were grouped based on sample date (before and after the coke plant closure) and compared using two-tailed t-tests for pre- vs. post-closure differences. Statistical analyses were performed using R Statistical Software version 4.2.2 (R Core Team 2022).

Residual autocorrelations were evaluated with the Breusch-Godfrey test. The 95% confidence intervals of the fitted interrupted time series regression model parameters were derived by assuming normality and using the R confint command. Models were tested in sensitivity analysis using both the imputed and non-imputed ED visit data and cyclic cubic splines with different degrees of freedom (DF) and stratification by month to adjust for seasonality. Further stratified analysis was performed to investigate the health outcome changes by sex and age subgroups; trends between cardiovascular outcome and traumatic injury control outcome were also compared in stratified subgroup analysis.

The study area populations remained stable over the study period at the case and the positive control sites, and the sociodemographic properties were comparable to county and state statistics (table 1). The annual mean daily ambient air PM_2.5 level monitored at the case site and negative control site were similar in 2015, the year before the Shenango coal coking plant closure, while the levels at the exposed control site, with an operating plant, was slightly higher (table 1).

A significant overall (3 year average, pre vs. post) reduction of ambient SO₂ is observed after the plant closure, both at the ACHD monitoring site (just across the river from the Shenango coal coking plant), and at the nearby Lawrenceville CSN monitoring site (table 2). In contrast, no significant SO₂ change was observed at the monitoring sites located near the still operating Clairton coke plant. The mean ambient SO₂ levels at the site next to the plant decreased by 90% after the closure, while at the nearest CSN monitor (at Lawrenceville, 10.6 km away) the SO₂ levels decreased by 50%.

As a test whether the noted SO₂ reductions were coal-related, measurements of trace elements with known characteristic sources at the nearest (Lawrenceville) CSN site were also assessed for potential changes in PM constituents after the closure (Thurston et al 2011). We found reductions in key coal-related PM_2.5 constituents: sulfate dropped from 2.22 (95% CI: 2.05, 2.38) μg m⁻³ to 1.32 (95% CI: 1.24, 1.39) μg m⁻³, while arsenic (As) dropped from 1.29 (95% CI: 1.07, 1.50) ng m⁻³ to 0.44 (95% CI: 0.29, 0.60) ng m⁻³) (p < 0.001), consistent with PM_2.5 composition changes expected from such a coking plant closure.

Also consistent with a beneficial effect by the plant closure, a statistically significant downward trend in the quarterly counts of cardiovascular inpatient hospitalization cases was seen post-closure in study zip codes near the Shenango plant (p < 0.05) (see table 3, figure 1). The difference from the pre-closure trend was −33.06 (95% CI: −59.08, −7.05) cases/year for cardiovascular inpatient hospitalization (from a 5.09 [95% CI: −13.84, 24.02] cases/year rate of increase before closure, to a −27.97 [95% CI: −46.90, −9.04] cases/year rate of decrease after the closure). Similar beneficial changes in hospitalization trends were also observed for ischemic heart diseases and cerebrovascular diseases after closure: We saw a change of −13.89 (95% CI: −26.42, −1.35) ischemic heart diseases hospitalizations/year (from a 6.26 [95% CI: −2.86, 15.38] cases/ year rate of increase before the closure to a −7.63 [95% CI: −16.75, 1.49] cases/year rate of decrease after the closure), and a change of −12.21 (95% CI: −20.25, −4.17) cerebrovascular diseases hospitalizations/year rate (from a 1.61 [95% CI: −4.24, 7.45] cases/year increase in rate before closure to a −10.60 [95% CI: −16.45, −4.76] cases/year rate of decrease after the closure). In sharp contrast to the near-plant study area, no significant health outcome changes were observed from the projected trend in either of the control populations over time after the plant closure.

Zoom In Zoom Out Reset image size

Similarly, while a statistically significant upward time trend was observed in cardiovascular emergency department visits at all three study locales before the intervention, there was an immediate drop, and then a longer-term downward trend, in the incidence of daily and weekly cardiovascular emergency department visits only in the coking plant neighboring community after the shutdown of the Shenango coke plant. Compared with the pre-closure emergency department visit trend, an immediate decrease of 18.61 visits (95% CI: 14.71, 22.52), 42% relative to the pre-closure weekly mean visits (N = 44.2 [95% CI: 42.2, 46.1]), was observed at the week after the shutdown. The difference in post-closure long-term trend from pre-closure trend was −0.31 (95% CI: −0.35, −0.27) cases/week, from a 0.17 [95% CI: 0.14, 0.20] cases/week increase in rate before closure to a −0.14 [95% CI: −0.17, −0.11] cases/week rate of decrease after the closure. In contrast, no change from the long-term upward trend was observed in study control populations, away from the Shenango plant (table 4, figure 2). Sensitivity analysis using the original unimputed data and all healthcare providers in the database yielded similar conclusions (see supplement).

Zoom In Zoom Out Reset image size

Significant immediate and longer-term cardiovascular outcome health improvements were especially found to be associated with the closure among sensitive subpopulations living near the facility. Notably, we found a statistically significant immediate reduction and downward trend of cardiovascular ED visits in those over 65 years of age in areas near the plant after the closure: Before Shenango coke plant closure, the weekly mean ED visit rate in population over 65 years of age at the case site was 16.4 visits per week (95% CI: 15.4, 17.3); the coke plant closure was associated with an immediate change of −5.04 [95% CI: −7.18, −2.90] visits followed by a long-term change of −0.13 [95% CI: −0.16, −0.11] visits/week, from a 0.06 [95% CI: 0.04, 0.07] cases/week rate of increase before closure to a −0.08 [95% CI: −0.09, −0.06] cases/week rate of decrease after the closure) (see table 5, figure 3). Moreover, such a reduction in medical visits after the closure was not seen in counts of visits for physical injuries, a control health outcome not thought to be affected by air pollution, confirming the specificity of the closure health benefits to biologically plausible outcomes.

Zoom In Zoom Out Reset image size

This study used the ITS method to evaluate the immediate and longer-term (cumulative) air pollution exposure change impacts on the incidence of cardiovascular outcomes after the closure of the Shenango, Inc. coal coking plant in January 2016. We confirmed simultaneous immediate and long-term declines in air pollution and cardiovascular disease rate in the neighboring resident community following the shutdown, when compared to expected trends from before the closure. The total cardiovascular ED visits in the coking plant neighboring community were 61% fewer than the predicted trend during the three years after the closure, while there were 13% fewer total cardiovascular hospitalizations. Residents over 65 years of age comprised 60% of the cardiovascular ED visit reduction. These results are clearly consistent with the conclusion that the plant closure was responsible for the reductions seen in the near-plant zip codes.

Our findings of the local air pollution trends are consistent with previous reports on this coal coking plant's point source emissions and local air quality impact (Huet-Vaughn et al 2018, Kelly and Fischman). The shutdown of the industrial emission source had the greatest estimated effect on ambient SO₂ levels directly related to coal processing at the monitors closest to the coke plant; the SO₂ change was smaller at monitoring site 10.6 km away, and no longer significant at the positive control site 25.9 km away with other emission sources. The measured PM_2.5 levels measured closest to the coke plant (from the ACHD monitoring network) decreased after the plant closure, becoming similar with the measurements 10.6 km away (from the EPA monitoring network). We also observed significant changes in coal-related trace element levels (As and S) at the CSN monitor nearest the coking plant. Similar air pollution findings have been reported in studies in Europe and Asia, identifying elevated levels of SO₂ and particulate matter arsenic, sulfate in the proximity of coking plants (Khare and Baruah 2011, Díaz-Somoano et al 2012, Mu et al 2012, Zajusz-Zubek et al 2017).

To the best of our knowledge, this is the first study using a statistically rigorous ITS analysis to investigate the health benefits in the local community after a coal coking plant closure. While past studies have suggested that living in the vicinity of coke plants can potentially increase health risks in resident populations (Parodi et al 2005, Porter et al 2014), our statistical findings of both short- and longer-term cardiovascular health improvements in local residents more definitively add to the growing body of evidence that policies implemented to regulate and reduce industrial emissions have real public health benefits (Henneman et al 2019, Martenies et al 2019, Rauner et al 2020). We found that both hospitalization and emergency department visit counts significantly decreased 40% immediately after the shutdown of the coke plant in the near-plant exposure population, and then statistically significantly continued downward in the months that followed. However, the near-plant community reduction in cardiovascular medical visits after the closure was not found in the positive or negative control populations during the same post-closure period, consistent with the interpretation of the near-plant reductions as due to the plant closure.

There are known air pollution health effect mechanisms that are consistent with the cardiovascular benefits we noted after the plant closure. Coal-related air pollution can cause oxidative stress, inflammation, and systemic health effects, and thus increase cardiovascular health risk (Schraufnagel et al 2019a, Maciejczyk et al 2021), so It is biologically plausible that the shutdown of the coal coking plant reduced both acute and chronic systemic inflammation among nearby residents. The immediate and longer-term benefits from a quantum drop in air pollution exposure are also consistent with steady reductions in ill-effects that have similarly been found over time following smoking cessation (Gratziou 2009, Polosa et al 2018). In contrast to the cardiovascular benefits of closure, control outcome traumatic injury ED visits, not biologically relevant to coke plant air pollution exposures, were unaffected: these control outcome counts followed the pre-closure cardiovascular visits trend in the same near-plant study population. Thus, the cardiovascular health benefits that coincided with the coking plant closure were only among residents in the exposure area, and also specific to disease categories that were plausibly affected by air pollution. The specificity of effects to both the exposed population and to plausible health outcomes is supportive of the conclusion of a causal inference of cardiovascular health benefits from the pre- vs. post-closure air pollution reductions (Hill 1965, Fedak et al 2015). Thus, our results also provide confirmation of the usefulness of epidemiological time series and cohort study results for the estimation of the potential health benefits from clean air policies.

Previous studies have ranked Pittsburgh as one of the cities with highest levels of air pollution and most air-pollution related deaths in the United States (American Lung Association 2022, Mckeon et al 2022). Historically, stationary source air pollution emissions in the Pittsburgh region have largely come from industrial sources, such as coking operations like Shenango and Clairton, so understanding the health outcomes of industrial air pollution is uniquely critical for improving public environmental health of communities near such emission sources (Lange et al 2022). Other studies have also found that low-socioeconomic-status communities experience higher levels of criteria air pollutants, partially from industrial sources (Hajat et al 2015, Boing et al 2022, Jbaily et al 2022). These communities, exposed to higher level of pollution, may also experience greater response to such pollution, and face higher overall risk for negative health outcomes (Gwynn and Thurston 2001). By demonstrating the public health benefits from the ending of these coal coking plant exposures, our study provides scientific inputs for the policy-making process at other industrial sites with high levels of air pollution emissions, thus further justifying action to control emissions, as well as the addressing of environmental injustices to communities living near such plants.

Despite its well-balanced design, there were certain limitations to this study. First, we were limited by the data release rules of the local health data agencies, and were therefore unable to acquire health data at a higher spatial-temporal resolution for inpatient hospitalization or with detailed diagnosis information for the emergency department visits. Consequently, we had more limited power to detect the health effects for specific disease categories such as ischemic heart diseases and cerebrovascular diseases in smaller subpopulations that may be at higher risk. Also, unlike the control sites, we were unable to completely eliminate statistical autocorrelation from the imputed daily ED visits data analysis at the Shenango exposure site, but we did find similar health effect results with the unimputed data, for which there was no significant autocorrelation.

Second, despite our extensive efforts to evaluate potential confounding via rigorous inclusion of both positive and negative control populations, as well as consideration of a control outcome of traumatic injury cases within the same population, the possibility of unaddressed confounding cannot be excluded. However, given our measures, and the geographic and administrative proximity between the exposure site and positive control site, we are not aware of any other particular policy changes that took place at the study sites during our study period that could have alternatively led to the robust observed trends we found.

Another potential study limitation is that, because our analysis employed aggregated count data, we were unable to adjust for health conditions and lifestyle factors at the individual participant level. There also were modest demographic differences between the study sites and the county and state population: our exposure population had a slightly higher poverty rate and percentage of non-white population and a smaller population growth rate. Caution should be given in the generalization of our findings to a population with very different population characteristics.

Overall, our research provides compelling scientific evidence that this intervention eliminating fossil-fuel related coal-coking air pollution emissions significantly improved both the air quality and cardiovascular health of the nearby community. In addition, this work provides rigorous validation of past policy applications of statistical associations found between acute air pollution exposures and adverse health to estimate clean air health benefits.

This work is supported by the Heinz Endowments Grant E8132. We also thank Jonah M Long (Pennsylvania Department of Health) for assistance with the EpiCenter data processing.

The data cannot be made publicly available upon publication because the cost of preparing, depositing and hosting the data would be prohibitive within the terms of this research project. The data that support the findings of this study are available upon reasonable request from the authors.