Among workers engaged in rescue and recovery efforts, occupational exposures to dust and fumes at the World Trade Center (WTC) disaster site have been associated with lower respiratory symptoms (1), worse lung function (2) and accelerated longitudinal lung function decline trajectories (3, 4), asthma (5) and other chronic airway diseases (6–8), and both qualitative (9) and quantitative (4, 8, 10–12) abnormalities on computed tomography of the chest. Common co-existing morbidities such as obesity are linked to greater severity of WTC-related respiratory symptoms, airway disorders, and morbidity (13–19).

The heterogeneity of the WTC-related lower airway diseases (LAD) poses a challenge for research studies and clinical care. Although stringent case definitions have been used to address this issue in clinical research studies (6), cluster analysis can be particularly helpful in classifying heterogeneous WTC-related chronic LAD (20) into groups without any a priori assumptions. To achieve this goal, cluster analysis examines multiple clinical, physiologic and imaging characteristics, including novel quantitative chest CT (QCT) features.

Identifying categories or clusters of WTC-related chronic LAD could help us better study and understand their pathogenesis, which should ultimately lead to improved diagnosis and management. We sought to, first, identify clusters of WTC-related chronic LAD and then examine their association with longitudinal lung function trajectories (4, 10) as clinical outcomes, in a cohort of WTC workers with QCT imaging data.

From the WTC Chest CT Imaging Archive cohort (n=1,630), 311 subjects had all quantitative chest CT measurements, as well as complete clinical and spirometric data from the surveillance visit closest to the CT scan. None of the standardized differences (StD) in the comparison of variables between the subjects included in this study, and those with incomplete or missing data and thus excluded (n=1319, see Table OS1) suggested an important difference in age, baseline BMI, smoking status, early WTC exposure arrival and cumulative duration, prevalence of dyspnea or evidence of bronchodilator response. Included subjects’ smoking intensity was slightly higher than among excluded subjects (7.8 SD 11.6 vs. 6.3 SD 12.8 pack-years, StD 0.22).

Table 1 summarizes the main characteristics of the 311 study participants included in the study and the results of the cluster analysis. Chest CT scans were performed 6.65 (standard deviation or SD 1.87) years and nearest spirometries 6.43 (SD 2.12) years after 9/11/2001. Consistent with the characteristics of the WTC General Responders Cohort (2), the participants’ mean age (SD) was 42.9 (8.5) years on September 11, 2001, most were male (~84%) and nearly half (49.8%) arrived at the WTC disaster site within 48 hours (early).

The cluster analysis showed that two, three, and five clusters generated the highest gap statistic (1.21 to 1.26), but that a five-cluster solution seemed most satisfactory. Except for HAV%, all variables were significantly different among the five clusters (see Table 1), which are displayed graphically in the heat map of Figure 1. Clusters 3 and 5 had a lower proportion of male participants (~68% to ~72%) than that in other clusters (ranging from ~86% to ~95%). In all clusters, the mean baseline BMI was ≥27.7 kg/m², with cluster 3 having the highest mean value (31.3 kg/m²). On the other hand, cluster 3 was the only one with a slight weight loss instead of weight gain over time. Regarding smoking status, former and current smoking were most frequent in clusters 1 (72.4%) and 4 (84.9%) and least so in clusters 2 (30.5%) and 5 (28.4%). Consistent with such smoking status, clusters 1 (15.8 pack-years) and 4 (14.1 pack-years) had the highest smoking intensity, while clusters 2 (3.6 pack-years) and 5 (2.0 pack-years) had the lowest smoking intensity. While early arrival to the WTC disaster site was most commonly reported among subjects in clusters 2 (96.3%) and 3 (72.4%) (2), subjects in cluster 3, 4, and 5 had longer cumulative WTC occupational exposure than those in clusters 1 and 2. With regard to spirometric patterns, airflow obstruction was most common in cluster 1 (65.5% of subjects), and a low FVC pattern prevailed in clusters 2 and 3 (respectively 54.9% and 75% of subjects). Conversely, a normal spirometry largely predominated in subjects in cluster 4 and 5 (93% and 87.8%, respectively), with relatively infrequent BDR (11.6 and 8%, respectively).

Based on the observed clinical, spirometric, and imaging characteristics, the five clusters could be defined as follows (Figure 1):

Of note, 39.2% of the Non-Latino White subjects were Polish workers, almost all of them ever-smokers[37] and, not surprisingly, mostly in clusters 1 (31%, 9 of 29) or cluster 4 (54.8%, or 46 of 84).

Lastly, we analyzed FEV₁ longitudinal trajectory groups as a clinical outcome for the five clusters (Table 2). For this longitudinal analysis, we had a mean of 5.0 (SD 1.7) good quality spirometries in each of 256 (82.3%) of our subjects. In this analysis, the highest proportion of subjects with rapid decline in FEV₁ was in cluster 1, while the highest proportion of subjects with intermediate decline in FEV₁ was in clusters 2 and 3. Although cluster 1 had the highest proportion of subjects with gain in FEV₁, this was based on a low count. Of note, subjects in cluster 4, who had one of the highest mean FEV₁%predicted and predominance of normal spirometry, but also the highest Pi10, had a proportion of subjects with rapid decline in FEV₁ only second to that of subjects in cluster 1.

Cluster analysis confirms a low FVC pattern (clusters 2 and 3) and chronic airflow obstruction (suggestive or consistent with COPD, cluster 1) (8, 2, 48) as distinctive chronic LAD phenotypes in WTC responders, while also supporting the presence of two distinct subgroups with unique clinical characteristics and longitudinal trajectories, seemingly non-susceptible smokers in cluster 4 and relatively healthy subjects in cluster 5. Our analysis also illustrated the adverse cross-sectional and longitudinal lung function effects of tobacco smoking, obesity and longitudinal weight gain trends.

Based on established diagnostic criteria, we previously noted the heterogeneity of the LAD among WTC responders (6), while most other studies have been based on symptom surveys (1, 3, 38), self-reported diagnoses (5, 39), or single lung function measurements (40, 41). In contrast, cluster analysis can integrate multiple variables, does not rely on a priori diagnostic assumptions and has been used to identify distinct phenotypes in heterogeneous diseases such as asthma (42–44), COPD (45), and obstructive sleep apnea (46, 47). To our knowledge, cluster analysis has not been previously conducted in the WTC cohorts, or in other inhalational disaster situations.

The functional and QCT characteristics of COPD predictably identified individuals that fit that disease phenotype in our population, as we demonstrated in a previous studies (8). COPD in these workers has been mild to moderate (48), with close to a third of affected subjects being never smokers (48). Our data in these workers with COPD thus far appear to suggest a predominance of the airway- over an emphysema-predominant subtype, as only 4% and 9.6%, respectively, had QCT evidence of emphysema, as indicated by low attenuation volume percent (LAV%) of ≥5% and ≥2.5%, respectively (8, 10), and half of the cases having functional features of asthma COPD overlap (48).

Low FVC (equivalent to a preserved ratio impaired spirometry, PRiSM) (49) has long been described as the most frequent abnormal spirometric pattern in the WTC occupational cohorts (1, 2, 6, 50). This airway disease pattern is likely heterogeneous (49), and 86.2% (75/87) of our subjects with it were in clusters 2 and 3. Unlike other cohorts (e.g., COPDGene^®), where subtypes of low FVC impairment have been recently identified (51), the WTC cohort was unselected for cigarette smoking. Our low FVC clusters 2 and 3 showed the strongest association (2) with early arrival at the WTC disaster site, when environmental exposures were likely highest (25). Low FVC subjects in cluster 3 were more dyspneic, obese, and also showed slightly more restriction (based on TLV_CT), and QCT evidence of proximal airway inflammation (WAP), the latter most closely matching (as in our previous study (8)) that of COPD subjects. In contrast, cluster 2 subjects had the highest proportion of never smokers (similar to the relatively healthy subjects of cluster 5) and were less dyspneic. Importantly perhaps for our long-term surveillance, clusters 2 and 3 had lower proportions of subjects with rapid decline in lung function.

Our overall findings confirm some previously identified functional and imaging traits associated with accelerated lung function decline in the WTC cohort, such as BDR, WAP (10), and LAV% (4). Consistent with our previous study (37), Latinos (predominantly laborers) tended to be slightly younger, never smoking, and late arrivals at the disaster site, and have less functional and imaging evidence of LAD (cluster 5). Of note, cluster 4 had a high proportion of subjects with rapid lung function decline (who may thus be at high risk for COPD) and Pi10 as the only significantly associated QCT marker. Given those results with cluster 4, Pi10, as well as other airway imaging metrics, deserve further investigation in the WTC cohort, as studies have shown its association with persistent chronic bronchitis (52), and possibly with more adverse but possibly preventable long-term functional outcomes. Cluster 4 also showed weight gain, and our previous studies had suggested the association of obesity (12) and weight gain (10) with quantitative CT markers of airway inflammation, and accelerated longitudinal expiratory flow decline, respectively. While cluster 4 had the highest proportion of current and former smokers, our studies have shown the remarkable success of smoking cessation interventions in this cohort (2, 8).

Our analysis benefited from an abundance of cross-sectional clinical, functional and respiratory structural data, and longitudinal functional trajectories to assess the potential adverse clinical impact of the identified clusters on respiratory health, in a population not selected for smoking as has been the case with other respiratory cohorts. Limitations on the other hand include the duration of the longitudinal follow up, which may have limited the detection of more severe lung diseases, such as COPD and interstitial lung diseases. Surveillance is, however, ongoing, and should provide additional information in the future. Sample size may have been limiting for only one of 21 variables examined, but the number of cases of low FVC and subjects with apparently normal spirometries allowed cluster analysis to suggest underlying pathophysiological features and indicators to investigate further as the surveillance continues. We used a sample of complete data cases but our comparison of included vs. excluded subjects suggests that our study sample is quite representative of the WTC Chest Imaging Archive cohort.

In conclusion, we identified five clusters of chronic lower respiratory disease in WTC responders, which have basic functional and QCT characteristics and clinical relevance for the treatment and longitudinal follow up of this population. Although much has been published about the management of disease types such as COPD and asthma, investigations like this shed light on the characteristics and evolution of other clusters such as our clusters 2, 3, and 4. Additional investigations are warranted of those subtypes, for further characterization of their clinical features, identification of additional QCT markers of early lung injury (and, possibly, repair) and longitudinal lung function trajectory.