The source population included WTC-exposed retired firefighters who left the workforce between 9/11/2001 and 9/11/2021 (n=7,843). The population was restricted to non-Hispanic Black, non-Hispanic White, and Hispanic male participants who had at least three PFTs over the course of the study period. Race and sex exclusions were made due to small numbers in certain subgroups. These exclusions consisted of 16 female, 27 Asian, and 7 American Indian participants. Incident data were available electronically beginning in 2010. Participants who retired before 2010, prior to the period which incident response data were fully available electronically (n=4,132) and those who retired between 2010 and 2021 but did not respond to at least one incident in this period (n=540) were excluded. The analytic cohort included participants who retired between January 2010 and September 2021, and responded to at least one incident during this period (n=3,171). All study participants provided informed consent to research and this study was approved by the Albert Einstein College of Medicine Institutional Review Board. A flow of participants is described in Figure 1.

Demographic and anthropometric covariate data as well as dates of birth, hire and retirement were obtained from the FDNY employee database and from routine monitoring exams conducted through the WTC Health Program. Self-reported WTC exposure data were obtained from participants’ first post-9/11 medical monitoring exam, most within one year after the attacks.[20] Smoking status was assessed at each monitoring exam as current, former, or never.

Routine firefighting particle exposures (P) were estimated using fire incident characteristics, response data, and emission factors from the JEM.[12] Briefly, fire incident and response information were provided by the Bureau of Communications at FDNY.[12] Incident variables used in the JEM included fire alarm levels (i.e., 1–5, with level 5 indicating the greatest fire involvement, potential for spread and firefighter response) and incident type (e.g., structural commercial building, non-structural vehicle fire, etc.). Response data included tour roster information such as date/time of an incident and unique codes for each responding firehouse. This enabled linkage of firefighters to specific fire incidents since roster information also included firehouse and date which individuals worked. Roster data also included information regarding their role (i.e., position description). The position description was dichotomized into ‘highly exposed’ and ‘minimally exposed’ based on likelihood to be involved in first-line combat of a fire response. P, in kilograms (kg), were estimated for every combination of fire alarm level, incident type, and position description from 2010–2021 using US EPA air emissions methodological framework.[13, 14] Findings are presented per 1,000 kg of particles. For example, approximately one 3^rd-5^th alarm commercial dwelling fire corresponds to P=1162.94 kg, three brush fires of any alarm level correspond to P=323.86 kg × 3=971.58 kg, and nearly 29 second alarm private dwelling fires corresponds to P=35.23 kg × 29=1021.67 kg.

As part of the WTC Health Program at the FDNY routine medical monitoring exams are conducted approximately annually. During these evaluations, PFTs, specifically spirometry exams, are performed to assess lung diseases, and, for active-duty firefighters, to assess whether they are medically capable of carrying out job responsibilities. The specific spirometry protocol used at FDNY has been described elsewhere.[17, 21] EasyOne spirometers (NDD Medical Technologies) were employed as part of the medical monitoring regimen. These devices perform automatic quality grading in accordance with the American Thoracic Society standards. While PFTs measure several metrics including vital capacity, for this study the outcome of interest is FEV₁ because is the most reproducible, least effort-dependent, and is most commonly used in longitudinal analyses.[22] Spirometry data included in this study spanned from September 2001 through September 2021. Pre-9/11 data were not included in this study.

Demographic and other descriptive characteristics for the cohort were calculated as proportions/percentages and means/standard deviations (SD), as appropriate. Each component (i.e., incident type, fire alarm level/severity and highly/minimally exposed fire-runs) used for development of the JEM, and in turn estimation of P, was calculated as an annualized total per individual and was presented as medians and interquartile ranges (IQR: Q1-Q3). A cumulative sum at the end of participants’ career was computed for the study period of 2010–2021.

Linear mixed effects models were used to evaluate the association between exposure, P, and longitudinal FEV₁.[23] All covariates including P were included as fixed effects and the intercept was included as a random effect to account for intra-subject variability and correlations between repeated measures. The specific model is shown below: Eq. 1 Yij=β0+β1Xij+Z0i+ϵij where Y=FEV₁; X=P for each interval; i=each repeated observation; j=each individual participant; β₀=intercept; β₁=regression coefficient for the difference in FEV₁ per P; Z is the matrix of random effects; and ε is the vector of random errors.

Different outcomes were evaluated to assess the acute and chronic impact of particle exposure on lung function. This was done by calculating the sum of P at 15 pre-defined intervals prior to a PFT. At each exam, the sum of P was calculated during the previous 0.25 (3 months), 0.5 (6 months), and 0.75 (9 months), 1, 1.5, 2, 3, 4, 5, 6, 7, 8, 9, and 10 years before each repeated PFT as well as cumulatively for the entire follow-up period. For example, when estimating exposure for an exam on 12/1/2019, the 3-month outcome will include all P estimated from 9/1/2019 through 12/1/2019. The cumulative outcome for the same exam would include all P from 1/1/2010 (i.e., beginning of follow-up) and 12/1/2019. Mock data are presented in Supplemental Table 1 to illustrate how each outcome variable was ascertained for a given period. Using P as the exposure, unadjusted and adjusted models were fit. Adjusted models included age, race/ethnicity, height, smoking, and weight as covariates. Each participant contributed all available PFT measurements from 2010 to retirement, and their first exam within one year after retirement to allow for precise estimation of lung function in relation to P. In a secondary analysis the cohort was subdivided based on period of retirement: those who left the workforce between 2019–2021 (group a), were analyzed separately from those that retired between 2010–2018 (group b). This cutoff was chosen to allow for an adequate follow-up period (at least 9 years) in terms of data availability for the JEM. Different cutoffs were considered and produced similar results; however, 9 years provided an optimal balance of follow-up time and participant numbers. A sensitivity analysis excluding post-retirement exams was conducted.

To test for effect modification of routine firefighting exposure, P, by previous WTC exposure intensity, the model below was fit (Eq. 2) which included an interaction term between P and WTC arrival time (i.e., arrived at disaster site on 9/11/2001 to 9/12/2001 vs. arrived 9/13/2001 to closing of WTC site) for each outcome. Then, models were stratified by WTC arrival time to illustrate any potential differences in the association between P and lung function by level of WTC exposure intensity. Eq. 2 Yij=β0+β1X1ij+β2X2ij+β3X1ij∗X2ij+Z0i+ϵij where Y=FEV₁; X₁=P for each interval; X₂=WTC arrival time; i=each repeated observation; j=each individual participant; β₀=intercept; β₁=regression coefficient for the difference in FEV₁ per P; β₂=regression coefficient for the difference in FEV₁ for each level of WTC arrival time; β₃=regression coefficient for the interaction between P and WTC arrival time; Z is the matrix of random effects; and ε is the unobserved vector of random errors.

In a secondary analysis to understand potential differences between subgroups, linear mixed effects modelling was also used to calculate changes in FEV₁ for groups a and b. Rather than P, the exposure measure was years of service, and all covariates were included as fixed effects and intercepts as random effects. A piecewise linear mixed effects model was fit such that the slope could be estimated for two critical periods: i) before exposure data were available - 9/11/2001–12/31/2009; ii) active service years from 2010 through retirement when exposure data were available. Models controlled for the same confounders as above.

The specific model that allows for calculation of the time effects before and after the two critical periods can be found below: Eq 3. Yij=(β0+β1tij)+Zy+εij if t≤p1(β0+β2k)+(β1+β2)tij+Zy+εij if t>p1 Where y=FEV₁; p₁=2010; t=time; β₀=adjusted mean FEV₁ at beginning of follow-up; β₁=change in FEV₁ per unit overall elapsed during the follow-up period as a fixed effect; β₂=difference in the change in FEV₁ between the late (t>p₁) and early (t≤p₁) periods and for the slope using time elapsed where t>p₁;k=knot in 2010; Z is the matrix of random effects; ε is the unobserved vector of random errors; i=each repeated observation; and j=each individual participant.

R v. 4.1.3 and SAS were used for data harmonization, analysis, and graphics. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

The primary analytic cohort included 3,171 participants (N=14,954, PFTs). Demographic characteristics are presented in Table 1. The cohort was homogenous with respect to race/ethnicity, arrival time at the WTC disaster site, height, and weight. The mean age of the cohort was 38.3 (SD=6.1), consisted of predominantly White participants (n=3,008; 94.9%) and responded to the WTC attacks on 9/11/2001–9/12/2001 (n=2,767; 87.3%). In the post-2010 period, when fire incident data (i.e., P) were available, the cohort had an average of over 6 years of follow-up (mean=6.2; SD=3.1) and over 4 PFTs (mean=4.4; SD=2.8). As expected, group a, participants who retired between 2019–2021, had the highest number of post-2010 service years (mean=10.3; SD=0.8) and the most serial PFTs (mean=7.6; SD=1.9) as compared with group b, who retired between 2010–2018 and had fewer post-2010 years of follow-up (mean=4.9; SD=2.4) and PFTs (mean=3.4; SD=2.2). The overall cohort has an average of approximately 5 PFTs (mean=5.3; SD=1.4) from 9/11/2001 through 12/31/2009. A description of participants who retired before 2010 and thus were not included in any of the main analyses can be found in Supplemental Table 2.

On average, the cohort responded to more structural (median=56.5/year; Q1-Q3=30.9–86.0) than non-structural incidents (median=19.3/year; Q1-Q3=9.5–29.4). 1^st alarm fires were the most common type of incident (median=74.2/year; Q1-Q3=39.3–110.4) and comprised 17.5% of all 1^st alarm incidents which included medical responses. Firefighters had a higher frequency of responses in minimally exposed roles (median=131.7/year; Q1-Q3=4.2–423.0) than highly exposed roles (median=40.3/year; Q1-Q3=0.0–339.5). Overall, firefighters were exposed to 2,922.3 kg of P (Q1-Q3=922.8–10209.7) from routine fire incidents.

Group a responded to more events during their time of employment, annually, of all types and severities and worked in relatively more highly exposed positions, on average (Table 2). As anticipated, given the longer duration of employment during the study period, cumulative P for group a was nearly three times that of group b (median of 6,346 vs. 2,274). Additionally, group a responded to more structural, non-structural, highly exposed, minimally exposed and incidents of all severities (fire alarm levels), on average when compared to group b.

After controlling for age, race/ethnicity, height, smoking, and weight, a −27 mL/1,000 kg P (β=−26.87; 95% CI=−34.54, −19.20) was observed when analyzing exposures within 3 months before PFT measurements (Figure 2a). Over the entire cumulative 12-year follow-up period, every 1000 kg P was associated with a 13 mL lower FEV₁ (β=−13.34; 95% CI=−13.98, −12.70). Sensitivity analyses excluding post-retirement exams yielded similar findings.

In our secondary analysis, this result was substantiated among those who retired both earlier (i.e., before 2019) and later in the study period (i.e., 2019–2021) (Supplemental Figure 1). For groups a (i.e., those who retired between 2019–2021) and b (i.e., those who retired between 2010–2018) 19 mL/1,000 kg P (β=−18.82; 95% CI=−30.44, −7.20) and 27 mL/1000 kg (β=−27.49; 95% CI=−36.81, −18.16) were observed when analyzing exposures within 3 months before PFT measurements, respectively. Similar trends were observed for groups a and b when analyzing exposure within 6 months through 12 years before a PFT.

A significant interaction was observed between time of arrival at the WTC site and P (β₃=−12.90; 95% CI=−22.70, −2.89) for the analysis evaluating exposures within 3 months before PFTs, only. Models stratified by time of arrival at the WTC are presented in Figure 2b. When PFTs were analyzed 3 months after routine firefighter exposure, the association between P and FEV₁ was slightly stronger for those with higher levels of WTC exposure (i.e., between 9/11–9/12/2001) (β₁=−25.89; 95% CI=−33.91, −17.89) than those with lower levels of WTC exposure (i.e., after 9/12/2001) (β₁=−12.76; 95% CI=−30.35, +4.84). The confidence intervals for the two groups overlap, which could be due to a lack of precision. The association of cumulative routine exposures was similar for both those that arrived at the WTC site early and late.

Similar associations were found for early arrivers in both groups a (β₁=−23.05; 95% CI=−35.83, −10.26) and b (β₁=−31.47; 95% CI=−42.03, −20.91) when evaluating exposures within 3 months of a PFT (Supplemental Figure 2).

When evaluating annual rate of decline for the period where exposure data were not available (2001–2009) and active firefighting years during the period where exposure data were available (2010-retirement), differences were observed for the analytic cohort (Table 3). The overall rate of decline in the early period (β=−25.7 mL/year; 95% CI= [−26.7, −24.6]) was less than in the late period (β₁+β₂=−37.2 mL/year; 95% CI= [(−38.2, −36.2]), after controlling for age, race/ethnicity, height, smoking, and weight. Similar observations were observed for those that retired later (a) and earlier (b).