Cancer Incidence Among Texas Publicly Funded Substance Abuse Treatment Clients Texas Medicine December 2017

Texas Medicine Logo(1)

The Journal — December 2017

Tex Med. 2017;113(12):e1. 

By Robert J. Reynolds, MPH, PhD; Steven M. Day, PhD; Alan Shafer, PhD; and Emilie Becker, MD

Dr Reynolds and Dr Day, Mortality Research & Consulting, Inc., City of Industry, California; Dr Shafer and Dr Becker, Texas Health and Human Services Commission, Austin, Texas. Send correspondence to: Alan Shafer HHSC Behavioral Health Services Decision Support 909 West 45th St, Austin, TX 78751; email: Alan.Shafer@hhsc.state.tx.us.

Authors' Note: The views and opinions expressed in this paper are those of the authors and do not represent any official view or policy of any government, university, or business.

Abstract

In this study, we examine the incidence of cancer among a large cohort of patients receiving Texas Department of State Health Services publicly funded substance abuse treatment services between 2005 and 2009. We hypothesized that substance abuse patients would have an increased incidence of cancer, especially cancers associated with alcohol misuse, tobacco use, and opiate dependence. We compared cancer incidence among 119,237 substance abuse patients with those in the Texas general population as reported to the Texas Cancer Registry.

The cohort was 60% male; and 50% white, 30% Hispanic, and 20% black. Mean age at the start of follow-up was 47.6 years (SD, 10.5 years), with mean follow-up time of 2.4 years (SD, 1.5 years). Primary drug dependency was 30% alcohol; 25% cocaine; 15% opiate; 13% amphetamine or methamphetamine; and 17% marijuana or other drugs. Almost 75% of the patients used tobacco regularly.

The overall age-adjusted incidence rate was lower for substance abusers than for the general population but not significantly so. This was true for several specific cancer sites as well, again not reaching statistical significance in most cases. Statistically lower rates of liver cancer were found for male and female substance abusers; higher rates of alcohol-related cancers were observed for males.

Multivariate models estimated significant increases in cancer incidence among substance abusers with increasing age. Hispanics were at greater risk of liver cancer but at decreased risk of lung cancer. The only significant substance-specific effect was for alcohol abusers, who were at significantly higher risk than all other groups for developing alcohol-related cancers.

Introduction

Substance abusers have elevated death rates in comparison with the general population. Many abusers face elevated risk of both infectious and noninfectious chronic disease depending on which drugs they use and their routes of administration.1,2

Many studies link sustained heavy alcohol use to increased cancer rates at multiple sites, including liver, esophagus, and breast.3-8 Most of these studies have demonstrated some version of a dose-response relationship between level of alcohol usage and cancer incidence. One study reported a standardized mortality ratio (SMR) of approximately 2.0 for cancer deaths among alcohol use disorder patients, while another found a J-shaped relationship between cancer mortality and drinking, wherein small amounts of alcohol lowered cancer risk, while heavier use increased risk.3,4 In contrast, a meta-analysis5 estimated a small, yet significantly elevated, risk of oropharyngeal, esophageal, and breast cancer for light drinkers. 

Researchers have also examined the risk of cancer among marijuana users. A review9 noted that while marijuana use does not seem to increase the risk for lung cancer, evidence is mounting for a link between marijuana use and testicular cancer. The authors further note that there are few studies from which to draw any firm conclusions on the associations of marijuana use and risk for other types of cancers. 

A large cohort study from Australia examined causes of death for opiate-dependent persons between 1985 and 2005.10 The results indicate that both the age-adjusted cancer mortality rate and the SMR were increasing for this population across the study period. The authors report opiate-dependent persons to be at 1.7 times the risk of death by cancer as the general population of New South Wales over the entire period of 1985 through 2005. The SMRs for 1995 through 1999 and 2000 through 2005 were reported as 2.2 and 2.0, respectively. The authors conclude that the increase in cancer mortality among opiate users may be due to declining opiate use in Australia, which in turn lowers the risk of death by competing causes at younger ages. The net result is that dependent persons who might have otherwise died before they could die from cancer are now living long enough to do so.10 The increased risk of mortality from cancer may also imply greater cancer incidence among this cohort and potentially among opiate-dependent persons in general. Another possibility is that as there has been a general decline in opiate use and as more people are surviving longer after a cancer diagnosis, opiate dependence that began during cancer treatment may be accounting for a larger proportion of the opiate-addicted population.11 This possible reversal of cause and effect may apply to other drugs as well (eg, alcohol and cannabis).

In this study, we examine the incidence of cancer among a large cohort of patients receiving publicly funded substance abuse (SA) treatment services from the Texas Department of State Health Services (DSHS), over a multiyear follow-up period. We hypothesize that SA patients have an increased incidence of cancer, especially cancers associated with alcohol misuse, tobacco use, and opiate dependence (due to risk of hepatitis C infection from intravenous administration).  

Methods

Study Population

The study population consisted of all adults older than 18 years receiving SA treatment services funded by DSHS during the years 2005 through 2009. To qualify for SA treatment by DSHS, clients must have incomes at or below the federal poverty level and have diagnoses consistent with abuse or dependence or both on alcohol and other substances as defined by the Diagnostic and Statistical Manual of Mental Disorders-IV.  Up to three substances, including alcohol, were available from clients’ treatment records (primary, secondary, and tertiary substances) for analysis. Study entry for each person in the DSHS population was either first contact with Texas DSHS or January 1, 2005, whichever came later. The end of follow-up was either the date of first cancer diagnosis or December 31, 2009, whichever occurred earlier. 

Data Sources

The Texas DSHS Behavioral Health Integrated Provider System is a web-based data collection and reporting system that provides records of service and assessment for publicly funded SA treatment clients. Service records were obtained for all patients aged 18 years and older treated at these clinics between January 1, 2005, and December 31, 2009. 

The Texas Cancer Registry (TCR) is a statewide population-based registry that is the primary data source for Texas cancer statistics. One of the largest cancer registries in the United States, the TCR currently meets the National Program of Central Cancer Registries, Centers for Disease Control and Prevention high quality data standards and is Gold Certified by the North American Association of Central Cancer Registries for the years included in this study.12  

Record Linkage 

We matched persons in the SA treatment database to their corresponding records in the TCR database using Registry Plus, Link Plus (Version 2.0) probabilistic matching software.13 First name, middle name, last name, Social Security number, and date of birth were used as matching variables, and matches were performed within gender groups. Variables describing substances used, treatment, risk factors, and outcomes were retained from the SA data, while cancer records included cancer site, stage, behavior, age at diagnosis, and more.

Analysis

We computed age-adjusted incidence rates (AAIRs) for various cancers in the study population, and compared them with AAIRs for the general population. The AAIRs for the study population were calculated as follows: 

  1. Incident tumors were counted separately for each stratum of calendar year, integer age, gender, and race;
  2. For those with incident cancer during the study period, time at risk was computed as the time from study entry to the time of first incident tumor;
  3. Because the study lacked information on censoring due to deaths for those not experiencing an incident cancer during the study period (non-cases), mortality was simulated. Evidence-based adjustments to general population mortality rates to account for excess mortality in substance abusers were derived from the study by Callaghan et al.2 The resulting mortality rates at all ages were used to simulate occurrences of death for non-cases throughout the study period. Exposure time for each non-case was then calculated as the time from first entry into the DSHS system to the time of simulated death, or to the end of the study period if no simulated death occurred.  For all non-cases, these simulated exposure times were summed within each calendar year, age, gender, and race or ethnicity category, and ultimately the stratum-specific totals for 500 such simulations were averaged and used as non-case exposure times in subsequent steps in the analysis.
  4. Incidence rates were calculated as the total number of cancers observed in a given category divided by the corresponding total exposure time for cases and non-cases in that category. 
  5. The rates determined in step 4 were multiplied by numbers of persons in the standard million population for the year 2000 for ages 15 years and above to get expected numbers of deaths within each age group.
  6. The sum of all expected numbers of deaths divided by the total of the standard million population for ages 15 and above yielded the AAIRs for the given category. 

Data processing and mortality simulation for the DSHS SA data were generated using SAS software.14 The calculation of AAIR for the SA data were completed using the epitools package in R.15,16 For the TCR data, AAIRs were calculated using SEER*Stat, version 8.1.2.17 The procedure for computing AAIRs with this tool are essentially the same as steps 4-6 above. Mid-year population estimates are used by SEER*Stat in place of actual exposure time to calculate crude incidence rates. These estimates take into account deaths and migration (as well as births) and are, therefore, a reasonable approximation of exposure time for the dynamic population during each year. Full details on the methods used by SEER to estimate annual population counts within age, race or ethnicity, and gender are available on the SEER website. The AAIRs were considered to be significantly different if the 95% confidence interval of TCR-based (general population) AAIR was not included in the 95% confidence interval of the AAIR for the DSHS SA data.

We also fit Poisson models to model the incidence rate of the various single-site and multisite cancers. In these models, main effects for the primary substance of persons, their age, and in some cases their gender or race or both were potentially included, with terms proving not to be significant dropped from final models. We also tested interaction between these factors, keeping any interaction term that proved to be statistically significant and modified the involved main effects by 10% or more. In the event that such an interaction was detected, we kept the main effects of that interaction regardless of their individual significance. All Poisson models were fit in R using the generalized linear model (glm) package.15  

Results

Table 1 displays the demographic and actuarial characteristics of the study sample. There was a total of 119,237 patients who accrued between 283,643 (adjusted for mortality) and 288,135.6 (unadjusted for mortality) person-years of follow-up time over the study period. Males composed between 58% and 62% of the clients entering the study in each year, and were 59% of the overall sample. The sample was approximately 50% white, but the proportion of whites trended downward over the course of the study. The proportion of blacks in the sample also declined slightly year over year, but the proportion of Hispanics rose in each year from under 28% in 2005 to over 31% in 2009. 

Tobacco use prevalence was fairly steady at approximately 73%, much higher than that of the US general population, which was approximately 20% across the period 2005 to 2009.18 The mean age at study entry was 47.6 (SD=10.5) years, with a mean age at diagnosis of 49.9 (SD=10.4) years.

Table 2 displays the categorization of SA clients in the cohort into substance use categories. Nearly one-third of the cohort (29.5%) were being treated for alcohol dependence/abuse, while 25.6% were being treated for cocaine abuse. The next largest group was the mixed group, consisting primarily of marijuana (87%, see Table 2) with a variety of other drugs in small amounts. Amphetamine users were 12.6% of the cohort, while opiate users were 14.7%. These groups were used both for applying evidence-based adjustments to mortality rates for the simulated survival times for non-cases of cancer for AAIRs and, also, for groupings for analysis in Poisson models.

Table 3 gives the counts and the resulting AAIRs for individual tumors and groups of related cancers for the DSHS cohort, as well as for the Texas general population. A total of 574 incident cancers were found in the SA cohort over the study period. The most common single-site cancer was lung cancer (104 cases), followed by liver cancer (57 cases). Of the composite cancer groups, alcohol-related cancers19 (inclusive of mouth and pharynx, larynx, esophagus, colon and rectum, liver, and breast) were most common with 351 incident cases, followed by tobacco-related cancers (lung, trachea, bronchus, oral cavity, pharynx [oro-, naso-, hypo-], larynx, esophagus, pancreas, bladder and renal pelvis, nasal cavities and sinuses, stomach, liver, kidney, uterine cervix, and myeloid leukemia) with 283 incident cases. (Note that the cancers in the alcohol-related and tobacco-related groups are not all mutually exclusive.) 

In general, the AAIRs were lower for the SA group than for the Texas general population (though in no case did this rise to the level of statistical significance), with the exception of liver cancer and alcohol-related cancers, for which the SA group AAIRs were higher than for the general population. For liver cancer, the AAIR for women in the SA group was 19.1 (9.1, 4132.6) cancers per 100,000 person-years, compared with 6.0 (5.8, 6.3) for the Texas general population; for men, the corresponding figures were 46.3 (26.9, 880.8) for the SA group, and 17.8 (17.4,18.2) for the Texas general population. Alcohol-related cancers among males had a significantly higher AAIR than the Texas general population, at 299.9 (229.5, 1080.3) cancers per 100,000 person-years compared with 126.2 (125.1, 127.4) in the Texas general population. For all other comparisons, no statistically significant difference was found between the SA population and the Texas general population.

Table 4 contains the results of Poisson regression models fit to the data assuming no competing risk from death. The incidence rate ratios (IRRs) are the relative risks associated with being in a category of primary substance preference, age, race, and gender. One model is presented for each of the major cancer types in Table 3, save for prostate cancer, for which no cases were observed. Models are also presented for the composite outcomes of alcohol-related and tobacco-related cancers. The categories of primary substance (drug of choice) are those of Table 2.

Significant age effects were seen in all of the models. The youngest ages experienced greatly reduced risk for all cancer types, with IRRs between 0.09 (0.03, 0.26) for 18- to 39-year-olds in the lung cancer model, and 0.35 (0.23, 0.54) for the same age group in the alcohol-related cancers model. In the model for colon cancer, risk was relatively uniform below age 50 years, then rose sharply in the age groups leading up to 60 years or older. In other all other models, risk tended to begin to rise at the age 45-49 years category.

Few significant race effects appeared in the models. Four of the models (breast, colon, alcohol-related tumors, and tobacco-related tumors) showed no significant race effects at all, and the terms were dropped from the final models. Hispanics were found to be at reduced risk of lung cancer in comparison with whites but at increased risk of liver cancer in comparison with whites. Gender was not significant in any models and, thus, was not included in any of the models in Table 4.

Interestingly, few significant primary substance effects were seen in the six models. A notable and significant effect is in the model for alcohol-related cancers. Unsurprisingly, all groups are at significantly lower risk of developing an alcohol-related cancer in comparison with alcohol abusers, with IRRs ranging from a low of 0.17 (0.07, 0.32) for amphetamine users to a high of 0.33 (0.20, 0.53) for abusers of marijuana and other drugs.

Discussion

The wide confidence intervals on most estimates in Table 3 reflect the fact that there were few tumors in most categories, likely because of the relatively young age of the cohort members and the relatively short follow-up period. As most cohort members enter treatment in their mid-30s to mid-40s and are followed on average about 2.5 years, many in the cohort are still in their early 50s or younger by the end of the study. The incidence for most cancers is still comparatively low at these ages, with the Texas Cancer Registry’s state profile indicating a Texas general population cancer AAIR of 92.9 for persons younger than 50 years.20 In addition, if an increased risk exists for developing various cancers from chronic drug use, this risk may have a long latency period that has yet to be observed in this relatively young cohort.

As evidenced by the AAIRs in Table 3, the incidence of liver cancers is elevated in the SA cohort compared with the general population of Texas, especially in men. This increase in risk is not reflected in the model for liver cancer in Table 4, as we found no significant gender effect in the multivariate model. Though not significant, the standardized incidence ratio greater than 1.0 for liver cancer among opiate users compared with the reference group of alcoholics may reflect the fact that liver cancer among substance abusers is often a result of contracting hepatitis C when sharing needles.1 It is unclear why Hispanics have a higher risk of liver cancer than blacks and whites; to the extent that these cancers are driven by intravenous drug use and hepatitis C, the increased risk for Hispanics may simply reflect a greater rate of needle-sharing among opiate users. It is also possible that the increased risk may reflect higher contact with immigrants from Mexico or a higher proportion of immigrants in this particular sample (nationality was not available for analysis) where the prevalence of hepatitis C is much greater among intravenous drug users than among intravenous drug users in Texas.21,22 In this scenario, equally risky behavior would be more detrimental as the baseline risk of contracting hepatitis C would be greater when needles are shared.

Alcohol-related cancers also had significant differences in age-adjusted incidence in Table 3, where again substance abusers were more likely to develop these cancers than were members of the general population in Texas. The model in Table 4 confirms that being alcohol dependent is the only significant risk factor other than age for developing alcohol-related cancers. Users of other primary substances were between 17% and 47% as likely to develop cancers of these types. Thus, the large numbers of alcohol-dependent persons being treated by DSHS are likely driving the elevated AAIRs for the entire cohort.

Substance abusers had lower point-estimate AAIRs for colon cancers than the Texas general population but not significantly so. Only 24 such tumors occurred in the cohort, and thus confidence intervals were wide. The risk model for colon cancer in Table 4 showed no relationship to any factors other than age and only at ages greater than or equal to 50 years. The low numbers of colon cancers and the steep elevation in risk associated with age suggest that either colon cancer is largely unaffected by the substances under study here or that the cohort was too young to observe any differences in incidence for a cancer type that is so heavily linked with age.

Since the models assume full survivorship in non-cases, the IRRs reported in Table 4 are likely to be underestimates of their true values, as the observed follow-up time is inflated. However, as this cohort is relatively young, the reduction in person-time that resulted from simulating mortality is not large and, thus, is not likely to push any of the insignificant effects in Table 4 to significance. This is especially true because significance tests and confidence intervals are driven largely by the ratio of events to follow-up time.

This study has several limitations stemming from the data sources used. One is the lack of mortality information for non-cases. As mentioned above, this leads to an overestimation of time at risk that, in turn, leads to a dilution of the incidence rates if the time at risk is not adjusted, and while we have made an evidence-based adjustment, it can only be an approximation, the precision of which is unknown. Another limitation is the lack of information about many social determinants of health that could further modify cancer risk, such as body mass index, family history, and more. In keeping with this theme, more information about the pattern of drug abuse for the clients under study ― particularly drug use history and duration ― was unavailable. Such data could have been key to better understanding the risk of the various cancers and would be a potentially valuable addition to the models in Table 4

Finally, the fact that these data originated from publicly funded SA treatment centers is likely to yield selection bias as many of the patients included here are of low socioeconomic status. As such, they may not be entirely representative of substance abusers in the general population, especially those who have their treatment funded by private insurance.  

As demonstrated by the prevalence of tobacco use (Table 1) and the percentage of the cohort being treated for alcohol dependence (Table 2), substance abusers have a high prevalence of tobacco and alcohol use. Thus, that alcohol-related and tobacco-related cancers are relatively common in this cohort is not a great surprise. That other cancers are not generally more common and are, in fact, rare is also not surprising given the age distribution. Nevertheless, that we observed no associations between other cancers and specific substances is interesting in its own right. Overall, our results do not offer support for a hypothesis of generally increased risk for cancer other than potentially for already-known mechanisms related to hepatitis C and exposure to tobacco and alcohol. Our results also fail to support hypotheses of decreased risk of cancer compared with the general population as the few instances where AAIR were lower for SA clients failed to reach statistical significance. Further research in populations of older substance abusers with better characterization of drug use history, including tobacco use, should be undertaken.

References  

  1. Mathers BM, Degenhardt L, Bucello C, Lemon J, Wiessing L, Hickman M. Mortality among people who inject drugs: a systematic review and meta-analysis. Bull World Health Organ. 2013;91(2):102–123.
  2. Callaghan RC, Cunningham JK, Verdichevski M, Sykes J, Jaffer SR, Kish SJ. All-cause mortality among individuals with disorders related to the use of methamphetamine: a comparative cohort study. Drug Alcohol Depend. 2012;125(3):290–294.
  3. Roerecke M, Rehm J. Cause-specific mortality risk in alcohol use disorder treatment patients: a systematic review and meta-analysis. Int J Epidemiol. 2014;43(3):906–919. 
  4. Jin M, Cai S, Guo J, et al. Alcohol drinking and all cancer mortality: a meta-analysis. Ann Oncol. 2013;24(3):807–816. 
  5. Bagnardi V, Rota M, Botteri E, et al. Light alcohol drinking and cancer: a meta-analysis. Ann Oncol. 2013;24(2):301–308. 
  6. Cai S, Li Y, Ding Y, Chen K, Jin M. Alcohol drinking and the risk of colorectal cancer death: a meta-analysis. Eur J Cancer Prev. 2014;23(6):532–539.
  7. Corrao G, Bagnardi V, Zambon A, Arico S. Exploring the dose-response relationship between alcohol consumption and the risk of several alcohol-related conditions: a meta-analysis. Addiction. 1999;94(10):1551–1573.
  8. Key J, Hodgson S, Omar RZ, et al. Meta-analysis of studies of alcohol and breast cancer with consideration of the methodological issues. Cancer Causes Control. 2006;17(6):759–770.
  9. Huang YH, Zhang ZF, Tashkin DP, Feng B, Straif K, Hashibe M. An epidemiologic review of marijuana and cancer: an update. Cancer Epidemiol Biomarkers Prev. 2015;24(1):15–31. 
  10. Degenhardt L, Larney S, Randall D, Burns L, Hall W. Causes of death in a cohort treated for opioid dependence between 1985 and 2005. Addiction. 2014;109(1):90–99. 
  11. Starr TD, Rogak LJ, Passik SD. Substance abuse in cancer pain. Curr Pain Headache Rep. 2010;14(4):268–275.
  12. NAACCR (North American Association of Central Cancer Registries). Who is certified. https://www.naaccr.org/Certification/WhoisCertified.aspx. Accessed August 30, 2016.
  13. US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion. Registry Plus, a suite of publicly available software programs for collecting and processing cancer registry data. Atlanta, GA; 2015. http://www.cdc.gov/cancer/npcr/tools/index.htm. Accessed September 1, 2011.
  14. SAS Institute Inc. SAS software. Cary, NC. Copyright 2012. 
  15. R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2016. 
  16. Aragon TJ. Epitools: Epidemiology Tools. R package version 0.5-7. 2017. https://CRAN.R-project.org/package=epitools. Accessed January 1, 2014. 
  17. National Institute for Health, National Cancer Institute, Surveillance Research Program. SEER*Stat software, version 8.1.2. 2013. www.seer.cancer.gov/seerstat. Accessed January 1, 2014.
  18. Jamal A, Homa DM, O'Connor E, et al. Current cigarette smoking among adults — United States, 2005–2014. MMWR Morb Mortal Wkly Rep. 2015;64(44):1233–1240. 
  19. International Agency for Research on Cancer Working Group on the Evaluation of Carcinogenic Risks to Humans. Tobacco smoke and involuntary smoking. IARC Monogr Eval Carcinog Risks Hum. 2004;83:1–1438.
  20. Texas Cancer Registry. State Profile. Austin, TX: Texas Department of State Health Services. https://statecancerprofiles.cancer.gov/quick-profiles/index.php?statename=texas. Accessed April 24, 2017.
  21. White EF, Garfein RS, Brouwer KC, et al. Prevalence of hepatitis C virus and HIV infection among injection drug users in two Mexican cities bordering the US. Salud Publica Mex. 2007;49(3):165–172.
  22. Baillargeon J, Wu H, Kelley MJ, Grady J, Linthicum L, Dunn K. Hepatitis C seroprevalence among newly incarcerated inmates in the Texas correctional system. Public Health. 2003;117(1):43–48.

December 2017 Texas Medicine Contents
Texas Medicine Main Page

 

 

Last Updated On

October 08, 2018

Related Content

Cancer | Substance abuse