Particulate Matter Air Pollution and Liver Cancer Survival

Abstract

Particulate air pollution (PM) exposure has been associated with cancer incidence and mortality especially with lung cancer. The liver is another organ possibly affected by PM due to its role in detoxifying xenobiotics absorbed from PM. Various studies have investigated the mechanistic pathways between inhaled pollutants and liver damage, cancer incidence, and tumor progression. However, little is known about the effects of PM on liver cancer survival.

20,221 California Cancer Registry patients with hepatocellular carcinoma (HCC) diagnosed between 2000–2009 were used to examine the effect of exposure to ambient PM with diameter less than 2.5µm (PM_2.5) on HCC survival. Cox proportional hazards models were used to estimate hazard ratios (HRs) relating PM_2.5 to all-cause and liver cancer-specific mortality linearly and non-linearly— overall and stratified by stage at diagnosis (local, regional, and distant)—adjusting for potential individual and geospatial confounders.

PM_2.5 exposure after diagnosis was statistically significantly associated with HCC survival. After adjustment for potential confounders, the all-cause mortality HR associated with a 1 standard deviation (5.0 µg/m³) increase in PM_2.5 was 1.18 (95% CI: 1.16 – 1.20); 1.31 (95% CI:1.26 – 1.35) for local stage, 1.19 (95% CI:1.14 – 1.23) for regional stage, and 1.05 (95% CI:1.01 – 1.10) for distant stage. These associations were nonlinear, with substantially larger HRs at higher exposures. The associations between liver cancer-specific mortality and PM_2.5 were slightly attenuated compared to all-cause mortality, but with the same patterns.

Exposure to elevated PM_2.5 after the diagnosis of HCC may shorten survival, with larger effects at higher concentrations.

Introduction

Air pollution is classified as a carcinogen by the International Agency for Research on Cancer (IARC).¹ While associations between ambient particulate matter with diameter less than 2.5µm (PM_2.5) and lung cancer have been well documented, associations with cancers at other sites have received less attention.² The liver may be a target as PM_2.5 can induce oxidative stress, inflammation, genotoxicity, and accelerate liver inflammation and steatosis, driving the development and progression of liver cancer.³ For example, PM_2.5 exposure has been linked to increased serum levels of hepatic enzymes such as alanine aminotransferase (ALT), which is a marker of liver damage and a predictor of hepatocellular carcinoma (HCC), the most common liver cancer.⁴ Long-term PM_2.5 exposures have recently been linked to an increased incidence of HCC,⁴ but the effects of PM_2.5 exposure on survival of HCC have not been investigated. We hypothesized that PM_2.5 exposure accelerates the progression of HCC and decreases survival after the diagnosis of liver cancer.

Materials and Methods

To determine whether PM_2.5 exposure is associated with survival in liver cancer patients, we combined data on patients newly diagnosed with primary HCC (ICD-O-3 site code of C22.0 and morphology codes of 8170–8176⁵) between 2000–2009 from the California Cancer Registry (CCR) (http://www.ccrcal.org) and PM_2.5 air pollution data collected by the U.S. Environmental Protection Agency’s (EPA) Air Quality System (AQS) database⁶ for the same time period, using the same methods as described in detail in a previous study.⁷ Briefly, the CCR data contain information on patient demographics, routine follow-up on vital status, tumor characteristics (including stage of diagnosis), and first course of treatment within 6 months of diagnosis for all HCC patients diagnosed in California. Cancer records were geocoded by longitude and latitude based on residential address at diagnosis, and then assigned into census tracts and block groups for assignment of area-based estimates of rural-urban commuting area (RUCA) codes and socioeconomic status (SES). Specifically, RUCA codes are census tract-level designations from 1 (metropolitan) to 10 (rural) based on the size and direction of primary commuting flows, using measures of population density, urbanization, and daily commuting (www.ers.usda.gov/data-products/rural-urban-commuting-area-codes.aspx).⁸ Socioeconomic status was calculated at the census block group level using validated area-level measures based on census 2000 data and American Community Survey 2007–2011 5-year estimates (online supplement).^9–11 Routine CCR follow-up of cancer patients monitored patient vital status through information sharing with reporting hospitals and linkage with a variety of administrative records.¹² There were 20,221 eligible HCC cases with PM_2.5 exposures included in the analysis. We censored the date of last follow-up to December 31, 2011.

Using hourly measurements of ambient PM_2.5 (in µg/m³) in the AQS database, monthly average PM_2.5 concentrations were calculated from hourly and daily measurements⁶, and spatially interpolated to residence locations from up to four closest air quality monitoring stations using inverse distance-squared weighting.^7,13 PM_2.5 exposures were not assigned to patient residences with unmatched geocodes or with the nearest monitor located > 25km away. Survival-period exposure summaries were calculated as the average ambient residential monthly PM_2.5 from date of diagnosis to date of death or loss to follow-up or the end of study.

Kaplan-Meier curves were used to calculate median survival stratified by stage at diagnosis and categorized PM_2.5 exposure. Cox proportional hazards models were used to estimate the PM_2.5 exposure (<10 µg/m³, 10–25 µg/m³, >25 µg/m³) and survival association. We considered both all-cause and liver cancer specific mortality, where liver cancer was the underlying cause of death on the death certificate (ICD-10 code C22.0-C22.9).¹⁴ Patient survival was censored due to loss to follow-up or study end (or, for liver cancer specific mortality, censored due to death by another cause). Hazard ratios (HRs)—scaled to a standard deviation (SD) increase in PM_2.5—were obtained from Cox models that adjusted for predetermined potential confounders: age, sex, race/ethnicity (non-Hispanic white, Hispanic, non-Hispanic black, Asian/Pacific islanders, other/unknown), marital status (single, married, formerly married, unknown), year of diagnosis, month of diagnosis, and initial treatment (surgery, radiation, and/or chemotherapy versus none), SES, dichotomized RUCA (metropolitan core, non-metropolitan core), categorized distance to primary interstate highways and primary US and State Highways (<300m, 300–1500m, >1500m). Adjusted Cox models were stratified by tumor stage at diagnosis (local, regional, distant). Sensitivity analyses were performed by further stratifying the stage-specific Cox models by distance to monitor, geocoding accuracy (street-level matching vs. city-level matching), regions (Los Angeles county, San Francisco Bay Area, San Diego county, and other regions in California), and socioeconomic status. Nonlinear associations with PM_2.5 were examined using a natural cubic spline with 2 degrees of freedom² and Cox models with categorized PM_2.5.

Analyses were conducted using SAS version 9.4 (SAS Institute Inc.) and R version 3.3.1 (http://www.R-project.org). Hypothesis tests were 2-sided with a 0.05 significance level.

Results

Overall, median survival was 0.64 years and only 324 patients were lost to follow-up (i.e., not known to be dead and with a last follow-up date 15 months or longer before the study end). HCC patients were on average 63.7 years old at diagnosis, racially diverse (38.9% non-Hispanic white, 25.9% Hispanic white, 26.1% Asian/Pacific islander) and predominantly male (75.0%), and living in metropolitan core areas (90.6%) (Table 1). The most common stage at diagnosis was local (44.8%) followed by: regional (27.5%), distant (17.6%), and unknown (10.0%). The number of patients diagnosed at local stage increased from 533 (5.9%) in 2000 to 1347 (14.9%) in 2009 while the number decreased from 548 (15.4%) to 395 (11.1%) in distant stage (data not shown). The most common initial treatments were chemotherapy (32.4%) and surgery (21.9%). In general, patients who lived in areas with high PM_2.5 tended to have lower socio-economic status and were more likely to be diagnosed with advanced stage disease (eTable1).

Table 1.

Summary of demographic, clinical, treatment characteristics, and distances of residential addresses at diagnosis from highways and PM_2.5 exposures for liver cancer patients newly diagnosed in California from 2000–2009, by stage at diagnosis.

Characteristics (Mean ± SD or %)	Local n=9,064 (44.8%)	Regional n=5,570 (27.5%)	Distant n=3,565 (17.6%)	Unknowna n=2,022 (10.0%)	Total n=20,221
Age at diagnosis (years)	64.1±11.9	62.9±12.3	62.8±13.3	66.0±12.7	63.7±12.4
Male	72.7	78.2	77.2	72.8	75.0
Race (%)
Non-Hispanic whites	39.1	37.9	38.5	41.5	38.9
Hispanic	26.2	25.6	25.9	25.5	25.9
Non-Hispanic blacks	6.9	8.7	9.1	6.7	7.8
Asian/Pacific islanders	26.8	26.2	25.2	24.2	26.1
Others/Unknown	1.0	1.5	1.2	2.0	1.3
Marital status (%)
Single	16.9	20.8	21.3	16.7	18.7
Married	57.2	55.8	53.5	50.2	55.5
Formerly married	23.7	20.9	22.5	29.0	23.3
Unknown	2.1	2.4	2.7	4.1	2.5
Rural-urban commuting area (%)
Metropolitan core	91.0	90.9	90.4	87.7	90.6
Non-metropolitan core	9.0	9.1	9.6	12.3	9.4
Socioeconomic status (SES, %)
Lowest	19.0	20.9	23.0	24.0	20.7
Lower-middle	21.8	22.0	22.1	22.2	22.0
Middle	20.8	21.0	21.0	20.3	20.8
Higher-middle	19.9	19.6	19.6	18.9	19.7
Highest	17.5	16.1	14.2	14.2	16.2
Unknown	0.9	0.4	0.1	0.3	0.6
First-course treatment types (%)
Surgery	36.0	16.6	4.5	4.3	21.9
Radiation	2.2	1.7	10.0	0.7	3.3
Chemotherapy	36.6	36.8	25.7	13.4	32.4
Geocode match quality (%)
Street address	91.8	90.8	91.4	91.2	91.4
Area-level	8.2	9.1	8.5	8.8	8.6
Other or missing	<0.1	0.1	0.1	<0.1	<0.1
Distance to primary interstate highwayb
< 300 m	10.3	10.4	9.6	10.6	10.2
300 – 1500 m	38.4	38.8	40.0	37.5	38.7
>1500 m	42.0	40.8	41.0	42.6	41.5
% missing	9.3	10.1	9.4	9.2	9.5
Distance to primary US and State highwaysb
< 300 m	4.0	4.5	3.8	4.1	4.1
300 – 1500 m	15.1	14.0	13.9	13.3	14.4
>1500 m	71.6	71.5	72.9	73.4	71.9
% missing	9.3	10.1	9.4	9.2	9.5
PM2.5 (µg/m3)	12.9±4.6	13.3±5.0	14.0±5.5	14.3±5.6	13.3±5.0
% missingc	9.0	9.3	9.9	12.0	9.5

^aInsufficient evidence available to assign a stage (e.g., patient dies before workup is complete, patient refuses diagnostic procedure, or limited workup is performed due to patient’s age or simultaneous contraindicating condition).

^bDistance values are primarily missing for participants with poor geocode matches (worse than street address match).

^cExposure assignments were excluded when the nearest monitor was located > 25km away or a geocode match was absent.

Median all-cause mortality times were higher in low PM_2.5 areas for patients diagnosed at either local or regional stage (Table 2). For example, median survival times for patients diagnosed at local stage living in lowest and highest PM_2.5 exposure areas were 2.16 and 0.07 years, respectively. After adjusting for potential confounders, PM_2.5 level was inversely associated with all-cause mortality with statistical significance. Allowing for nonlinear associations using splines or categorized exposure, the adjusted PM_2.5 HR for all-cause mortality were significantly higher in high exposure areas in both overall and in stage-specific models (Figure 1, Table 2). In stage-specific models using a linear term for continuous PM_2.5 exposure, the HRs associated with a 1 SD increase in PM_2.5 were 1.31 (95% CI: 1.26 – 1.35) for local stage, 1.19 (95% CI: 1.14 – 1.23) for regional stage, and 1.05 (95% CI: 1.01 – 1.10) for distant stage. This indicates the damaging effect of PM_2.5 is most substantial for patients with local disease and diminishes with worsened stage, which likely reflects the shortened survival/exposure time for patients with late stage disease. The associations between liver cancer specific survival and exposure to PM_2.5 were slightly attenuated, but the pattern that adjusted PM_2.5 HR was larger for patients diagnosed at early stage remained (eTable 2). Sensitivity analyses show that stage-stratified HRs were robust to stratification by distance to monitor, geocoding accuracy, and socioeconomic status (eTable 3). Stratification by various metropolitan areas (LA county, San Diego county, San Francisco Bay Area) showed a consistent pattern of larger PM_2.5 HR for patients with local stage at diagnosis as compared to patients with more advanced stage at diagnosis.

Table 2.

Median all-cause mortality time stratified by stage at diagnosis and categorized fine particulate matter (PM_2.5) exposure, and adjusted^a hazard ratios for all-cause mortality from pooled and stage-specific Cox proportional hazard models with categorized PM_2.5 and continuous PM_2.5^b.

Median survival (years)
		Number of patients	Local	Regional	Distant	Overall
Categorized PM2.5 (µg/m3)	< 10	5364	2.16	0.61	0.18	0.83
	10–15	7431	1.92	0.59	0.23	0.87
	15–20	3794	1.07	0.36	0.21	0.49
	20–25	1393	0.65	0.33	0.27	0.42
	25–30	386	0.25	0.15	0.12	0.19
	≥30	140	0.07	0.06	0.09	0.08
Overall			1.52	0.48	0.20	0.64
Hazard Ratioa (95% confidence interval)
Categorized PM2.5 (µg/m3)	< 10		REF	REF	REF	REF
	10–15		1.04 (0.98 – 1.11)	1.02 (0.95 – 1.10)	0.93 (0.85 – 1.02)	1.00 (0.96 – 1.04)
	15–20		1.35 (1.24 – 1.46)	1.28 (1.17 – 1.40)	0.97 (0.87 – 1.09)	1.18 (1.12 – 1.24)
	20–25		2.06 (1.83 – 2.32)	1.43 (1.25 – 1.64)	0.94 (0.81 – 1.09)	1.46 (1.36 – 1.57)
	25–30		3.85 (3.15 – 4.70)	2.52 (2.00 – 3.17)	1.68 (1.36 – 2.09)	2.40 (2.14 – 2.69)
	≥30		10.24 (7.40 – 14.16)	4.60 (3.28 – 6.45)	2.60 (1.82 – 3.73)	4.61 (3.87 – 5.50)
Continuous PM2.5b			1.31 (1.26 – 1.35)	1.19 (1.14 – 1.23)	1.05 (1.01 – 1.10)	1.18 (1.16 – 1.20)

^aAdjusted for age, sex, race/ethnicity, marital status, socioeconomic status, rural–urban commuting area, distance to primary interstate highway, distance to primary US and state highways, month of diagnosis, year of diagnosis and initial treatments.

^bHazard ratios are scaled to a 1 standard deviation increase in PM_2.5 (equivalent to 5.0 µg/m³).

Figure 1.

Nonlinear patterns of adjusteda associations between PM_2.5 exposure and all-cause mortality (solid line) and 95% confidence interval (dashed line) estimated using a natural cubic spline with 2 degrees of freedom. Points with confidence intervals are hazard ratios and 95% confidence intervals for all-cause mortality from pooled and stage-specific Cox proportional hazard models with categorized PM_2.5 exposure.b The distribution of PM_2.5 exposures are demarcated using ticks above the x-axis.c

^a Adjusted for age, sex, race/ethnicity, marital status, socioeconomic status, rural–urban commuting area, distance to primary interstate highway, distance to primary US and state highways, month of diagnosis, year of diagnosis and initial treatments.

^b The categories of PM_2.5 exposure were 0–10, 10–15, 15–20, 20–25, 30+ µg/m³. The hazard ratios for all-cause mortality from pooled and stage-specific Cox proportional hazard models with categorized PM_2.5 exposure are presented in Table 2 and plotted at 5, 12.5, 17.5, 22.5, and 35 µg/m³ in the Figure 1.

^c Three patients with PM_2.5 exposure > 50 µg/m³ were excluded from the plot (but not the model).

Discussion

Our study provides the first evidence that exposure after HCC diagnosis to a major ambient air pollutant, PM_2.5, is associated with shortened survival. These associations were strongest for patients diagnosed at local stage and showed larger adverse effects at higher PM_2.5 levels. To our knowledge, no study has related PM_2.5 to liver cancer survival, but several have examined other PM_2.5-liver cancer associations.^2,4 For instance, in a large cohort study (n=66,820) Wong et al found that PM_2.5 was associated with mortality from accessory organ cancers (liver, gall bladder, and pancreas) (overall: 1.35 (95% CI: 1.06–1.71), in males: 1.28 (95% CI: 0.83–1.96), in females: 1.37 (95% CI: 1.05–1.80), and in female never smokers: 1.36 (95% CI: 1.01–1.84)).² A key difference between our study and the Wong et al study is that we calculated the average PM_2.5 exposure after diagnosis, addressing the hypothesis that PM_2.5 exposure after diagnosis adversely affects survival. Wong et al assigned exposures based on the subjects' recruitment year average (between 1998 and 2001). Our study specifically targeted HCC and conducted stage-specific analysis, which limits carry-over effects at diagnosis.

Our population-based study design—consisting of all Californian HCC cases between 2000–2009—minimizes selection and survivorship bias. California also has one of the largest and longest running air pollution monitoring networks and a wide range of PM_2.5 exposures allowing large valid scale exposure assessment using standard methods. Residential addresses were available only at the date of diagnosis, so there could be exposure misclassification due to patients relocating. However, patients with HCC were less likely to relocate after diagnosis as the survival times were very short. An important limitation of our registry-based design is that personal-level data (e.g., alcohol consumption, Hepatitis A/B status, weight, and residential histories after diagnosis) were not available in the CCR. We did control for area-level SES and rural/urban factors, major confounders to the PM_2.5-HCC survival associations, and believe that factors other than SES and residential location were unlikely to be associated with the spatio-temporal distribution of ambient air pollution exposures. Thus, the associations could not have spuriously induced with the adjustment of these omitted factors. Nevertheless, future studies should consider additional personal-level factors.

Previous studies have suggested exposure-response relationships for PM_2.5 that were linear for lung cancer mortality or steeply increasing at low exposure levels and flattening out at higher exposures for cardiovascular mortality.^15–16 On the one hand, our finding of nonlinear (concave upward, quadratic curved) PM_2.5-HCC survival associations has significant public health implications, providing evidence that relatively small reductions in high pollution levels could have substantial health impacts. On the other hand, the magnitude of the hazard ratios at the high pollution levels were large. While these findings were plausible and consistent with the results from the analysis on the linear association, we recognize there might be unmeasured risk factors such as those personal-level confounders that are not available in our data. Future research is needed to confirm this finding.

In summary, we found adverse effects of PM_2.5 exposure after diagnosis on liver cancer survival. Not only were such effects more profoundly for those diagnosed with early stage, but they also increased strongly with concentration, suggesting that reductions in high PM_2.5 exposure could increase survival for a non-respiratory system cancer.

Novelty & Impact Statements.

This is the first registry-based study to link individual-level estimates of air pollution exposures after liver cancer diagnosis to survival on a population-based sample of 20,221 patients with newly diagnosed liver cancer during 2000–2009 in California. We found adverse effects of PM_2.5 exposure after diagnosis on liver cancer survival increased strongly with concentration, suggesting that reductions in high PM_2.5 exposure could increase survival for a non-respiratory system cancer.