Introduction

About 40% of households in the USA have a dog, roughly over 36 million households have a cat and almost 4 million have a pet bird. The steadily increasing popularity of pets indicates that Americans believe that pets enhance their lives (Pet Ownership: Data from American Pet Products Manufacturers’ 2005/2006 National Pet Owners Survey 2006). However, vigorous assessment of the overall health impact of having pets has been overlooked (Levine et al. 2013). A health benefit from pet ownership has been observed in numerous reports (Friedmann and Thomas 1995; Friedmann et al. 2011; Friedmann et al. 2003; Ogechi et al. 2016), and pet therapy is becoming popular for helping people cope with or recover from health problems, especially for cardiovascular diseases (Creagan et al. 2015; Levine et al. 2013). However, detrimental effects on health outcomes (Moody et al. 1996; Parker et al. 2010; Parslow and Jorm 2003), specifically increased risk of cancer, for pet owners over non-pet owners were also reported elsewhere (Franti et al. 1980; Holst et al. 1988; Laumbacher et al. 2006; Petridou et al. 1997).

To the best of our knowledge, as of today, only one study has longitudinally assessed the relationship between the risk of cancers and pet ownership in the general population of the USA. Almost all previous studies were cross-sectional (Franti et al. 1980) or case-control studies (Alavanja et al. 1996; Gardiner et al. 1992; Holst et al. 1988; Kohlmeier et al. 1992; Laumbacher et al. 2006; Modigh et al. 1996; Morabia et al. 1998; Petridou et al. 1997; Swensen et al. 2001; Tranah et al. 2008). Cancer generally takes a long time to develop, casting substantial concerns over the validity of data collected retrospectively in previous studies. In addition to recall biases, most previous studies also used hospital-based controls, bringing selection bias to the association in question (Kohlmeier et al. 1993). To address these limitations, and fill the gap between the increasing popularity of pets and scarcity of high-quality evidence on the association between pet ownership and cancer, we analyzed the data of a nationally representative cohort from the general population to examine the relationship between pet ownership and the risk of dying from colorectal cancer, one of the most common cancers diagnosed in both men and women in the US (Cronin et al. 2018).

Methods

Study population

We used the data collected in the third National Health and Nutrition Examination Survey (NHANES III) conducted in 1988–1994. The NHANES III consisted of a nationwide probability sample of non-institutionalized civilians. We restricted our analyses to 17,388 adults aged 19 and older whose information related to pet ownership was assessed at the baseline survey in 1988–1994, and the vital statuses were available by the end of 2010. In total, 13,929 participants were retained for the main analyses after exclusions due to various reasons (Fig. 1).

Fig. 1
figure 1

Flow chart of the study population included; 13,929 adults aged 19+ years, NHANES III follow-up study 1988–2010

Full size image

Baseline data collection

NHANES III baseline data were collected during an in-home interview and a subsequent visit to a mobile examination center (MEC). The demographic and health-related information was collected using a standardized questionnaire/protocol. Informed consent was obtained from all NHANES participants by the National Center for Health Statistics, which administrated the survey.

Pet ownership

In the housing characteristics section of the household interview, respondents were asked a two-part question about whether a pet resided in the home: “Does a pet live here? If yes, what type of pet is it?” The response choices included dog, cat, bird, fish, rodent, rabbit, reptile, farm pet and other. Given the relatively small number of persons who had pets other than dogs, cats or birds, the current study examined the association between death of colorectal cancer and the ownership of dogs, cats or birds when looking at the type of pets.

Physical activity level and body mass index (BMI)

Sedentary lifestyle has been documented to be associated with the risk of colorectal cancer (Kerr et al. 2017); therefore, physical activity level and body mass index were included as the potential confounders. Physical activity was assessed as the monthly frequency of these activities: walking, jogging, biking, swimming, calisthenics, gardening, weight lifting, aerobics and dancing. Participants listed up to four activities they had engaged in that lasted at least 10 min and involved moderate or vigorous effort. Monthly physical activity was converted to weekly physical activity. The level of physical activity was categorized as “vigorous” as participating in such activity every day, “moderate” if 4–5 times per week and “light” if 1–3 times per week, otherwise as “physically inactive.” Body mass index (BMI) was calculated as directly measured weight in kilograms divided by height in squared meters (kg/m2).

Cigarette smoking, alcohol drinking and atopic conditions

Behavioral risk factors play crucial roles in the development of colorectal cancer, especially alcohol drinking and cigarette smoking (Hamada et al. 2018; Rossi et al. 2018). We defined current alcohol drinking as “heavy” if the respondents reported that, in the last 12 months, five or more drinks were consumed on more than 10 days and “moderate” if the number of days on which five drinks or more were consumed was ≤ 10 days, otherwise as “rare or lighter.” Tobacco smoking status was categorized as “never/rare,” “former” and “current.” The formers and the currents were further classified as “moderate/heavy smokers” if the respondents reported that ten cigarettes or more were smoked per day, otherwise as “light smokers.” Serum cotinine, an alkaloid found in tobacco and also the predominant metabolite of nicotine, was used to control for underreporting of cigarette smoking or exposure to environmental tobacco smoke. Self-rated health condition has been associated with total mortality and may serve as a proxy of risk-adjustment effort (Lainscak et al. 2014) . The history of asthma was used as the proxy of atopic diseases and was assessed using the Medical Condition Module (MCM) of NHANES. The MCM interview was conducted at the MECs to collect self-reported data on a broad range of health conditions. The respondents were asked, “Has a doctor or other health professional ever told you that you had any of the following conditions?” This was followed by a list of various chronic and acute conditions, some described with vernacular expressions, representing prevalent conditions in the US population.

Socio-demographics

Race/ethnicity was coded as whites, blacks or Mexican Americans. Regardless of age, educational attainment was measured as the highest completed grade of school and categorized into high school/equivalent or below, some college years and college graduate or above. Family income was assessed using the poverty income ratio (PIR), calculated from the previous year’s family income and family size, and compared with the federal poverty line (PIR = 1) (Housing and Household Economic Statistics Division 2009; Lainscak et al. 2014). Marital status was classified into two categories: married or others, which included never married, widowed, divorced and separated.

Following up of vital status and death from colorectal cancer

A total of 12 identifiers (social security number, sex, date of birth, etc.) were used to link NHANES III participants with the National Death Index to ascertain vital status and the cause of death. Vital status ascertainment was based upon the results from a probabilistic match between NHANES III and death certificate records. More than 96% of deceased participants and all living participants were successfully followed up and correctly classified. The cause of death was determined using the underlying cause listed on the death certificate. The International Classification of Diseases, Injuries and Cause of Death (ICD) 10th revision was used to code deaths after 1999, and deaths in 1999 and before were originally reported in ICD 9th but recoded using ICD 10th. The codes for colorectal cancers included C18-C21 (malignant neoplasms of the colon, rectum and anus).

Statistical analysis

We used SAS (SAS 9.4, Research Triangle Park, NC) procedures for surveys with appropriate weighting and nesting variables to produce accurate national estimates and adjust for the over-sampling of specific populations. Using Cox proportional hazard regression models, we estimated the adjusted hazard ratios (HRs) for the relative risk and compared the risk among pet owners over non-pet owners (reference). The 95% confidence intervals (95% CIs) of HR were calculated, and the p values were two tailed. The Cox models were run to adjust for age, sex, race/ethnicity, family income, education attainment, alcohol drinking, cigarette smoking, marital status, body mass index, physical activity, self-evaluated health status, serum cotinine and history of asthma measured at the baseline survey. The person-year contributions from each participant were calculated as the time between their date of the baseline examination and the date of death (if it occurred), or 31 December 2010 (if still alive), whichever occurred first. To assess reverse causality, we repeated the steps described above but all deaths occurring in the first 2 years of the follow-up were excluded. The participants with cancer diagnosed before the baseline survey were also excluded from the sensitivity analyses. We created time-dependent covariates to test the proportional hazard assumption.

Results

As nationally representative samples, the study population reflected the demographic profile of the American population in the late 1980s and early 1990s (Table 1). Whites, African Americans, Mexican Americans and others accounted for 77.2%, 10.5%, 4.9% and 7.3%, respectively. Dogs were the most popular pet, owned by 25.8% of the study population, followed by cats (20.4%) and birds (4.6%). Pet ownership varied across socio-demographic strata (Table 2). Almost half of adults aged 20–64 years old, but about a quarter (25.8%) of adults ≥ 65 years, owned a pet. About 48% of white families had a pet, along with 22% of African American families. Current heavy or moderate smokers had the highest percentage of owning a pet (50.8%), while adults who never smoked had the lowest percentage (37.9%). Pet owners were generally more physically active and more likely to have a history of asthma than their non-pet owner counterparts.

Table 1 Selected characteristics of the overall weighted study population;13,929 adults aged 19+, NHANES III 1988–1994a
Full size table
Table 2 Pet ownership by socio-demographic and health behavioral characteristics; 13,929 adults aged 19+, NHANES III 1988–1994
Full size table

By the end of the 18- year follow-up (mean = 15 years), 70 colorectal cancer deaths were recorded, of which only 3 occurred among bird owners (Table 3). Compared with non-pet owners, the crude hazard of dying from colorectal cancer was significantly higher among neither the owners of any pets nor the owners of any specific type of pets; the crude HR was 1.37 (95% CI = 0.74–2.55) for owning any type of pets. After adjustment for age, the HRs of dying from colorectal cancer were significantly higher in pet owners than in non-pet owners; the HR was 2.58 (1.38–4.81) for owners of any pets and 2.30 (1.06–5.00) for cat owners. Owning a dog was not significantly associated with the hazard of dying from colorectal cancers; the age-adjusted HR was 0.90 (0.38–2.12).

Table 3 Rates and unadjusted HRs (95% CI) of dying from colorectal cancer; 13,929 adults aged 19+, NHANES III follow-up study 1988–2010
Full size table

It is important to note that adjustment for socio-demographics, risky health behaviors, serum cotinine and atopic conditions at baseline increased rather than decreased the HR estimates (Table 4). For example, the age-adjusted HR of colorectal cancer death for owning any pet was 2.58 (1.38–4.81). After comprehensively adjusting for all the covariates included, the HR increased to 2.83 (1.51–5.30) for owners of any pet. The association was mainly driven by owning a cat, and the fully adjusted HR for owning a cat was 2.67 (1.22–5.86). The adjusted HR for owning a dog was close to the null value, i.e., 0.89 (0.37–2.12). Excluding the study participants with prior diagnosed cancer at the baseline survey (n = 1232) made the association stronger, 4.34 (2.26–8.34) for owning any type of pets and 3.33 (1.45–7.64) for owning a cat (data not shown). Excluding the deaths occurring in the first 2 years of follow-up (n = 338) did not change the estimates. The adjusted HR was 2.85 (1.51–5.38) for owning any type of pet and 2.67 (1.19–6.02) for owning a cat, almost identical to the corresponding estimates obtained from the main analyses. The p value for the interaction between pet ownership and time was 0.24, indicating no strong evidence of an exponential trend over time in the hazard ratio of dying from colorectal cancer because of having a pet.

Table 4 Hierarchically adjusted HR of colorectal cancer deaths associated with pet ownership; 13,929 adults aged 19+, NHANES III follow-up study 1988–2010a,b,c
Full size table

Discussion

From a nationally representative cohort, we found that owning a pet, specifically a cat, was associated with an elevated hazard of dying from colorectal cancer. The detrimental effect that pets conferred on increasing the risk of colorectal cancer was primarily attributed to owning a cat and cannot be explained by the history of smoking and drinking, exposure to environmental tobacco smoke or atopic conditions of the owners. No such association was observed among dog owners.

There are a limited number of studies examining the relationship between having a pet in the home and the risk of developing cancers and dying from cancers. Ecologically, mortality from lung cancer was much higher in The Netherlands than in Sweden, and The Netherlands has a higher percentage of households with birds as pets compared with Sweden. At the time of interrogation, more than twice the number of breast cancer patients had kept dogs permanently in the last 10 years compared with the control individuals (Laumbacher et al. 2006). An increased risk of childhood leukemia was observed to be associated with pet ownership among 153 incident cases in Greece (Petridou et al. 1997). An elevated risk of lung cancer among bird keepers was reported 2 decades ago (Gardiner et al. 1992; Holst et al. 1988; Kohlmeier et al. 1992). A recent prospective analysis of 123,560 postmenopausal women enrolled in the Women’s Health Initiative revealed that among never-smokers, there were non-statistically significant associations between cat ownership and lung cancer (Garcia et al. 2016). It is worth noting that all previous studies were case-control studies and were conducted with participants > 40 years old (Gardiner et al. 1992; Holst et al. 1988; Modigh et al. 1996; Morabia et al. 1998) or only postmenopausal women (Garcia et al. 2016). In addition to the recall bias, misclassification of covariates caused by impaired memory among the elderly may also contribute to the inconsistent findings.

The mechanisms underlying the association observed in the current study are to be illustrated. The health hazards of mycotoxins to humans and animals have been examined extensively in recent years (Di et al. 2014; Wang et al. 2014) and may be a part of the biological mechanism. Ingesting foodstuffs contaminated with aflatoxin could be a source of aflatoxin exposure (Cortes et al. 2010) because aflatoxins can be excreted in the litter. Organic dusts present in unhygienic homes could be a source of molds, and aflatoxin exposure through these pollutants is very likely. Therefore, handling of contaminated feeds, litters and organic dust associated with pets may cause inhalation exposure to aflatoxin and pose a serious health threat to pets and their owners. Inhalation exposure of airborne contaminants can be absorbed from the lungs and translocated to many tissues, including the gastrointestinal tract (Bond et al. 1986; Harrison et al. 1993). Existing data support a role of food-associated mycotoxins in the induction and/or persistence of human chronic intestinal inflammatory diseases in genetically predisposed patients (Maresca and Fantini 2010). Certainly, a serious long-term complication of chronic inflammation is a part of the development of colorectal cancer (Song et al. 2015). It is difficult to interpret why cat owners rather than dog owners showed more vulnerability to colorectal cancer. In general, proteins released from cat dander and skin cells in the indoor environment are smaller and lighter than those of dogs and can easily penetrate into the human body and contaminate other household items. These proteins may serve as a carrier of aflatoxin from contaminated pet food and may increase the aflatoxin exposure for home occupants. Alternately, a lack of physical activity is causally related to colon cancer (Slattery 2004). Dog owners were relatively more physically active; the residual confounding from physical activity may counteract the detrimental effects of keeping a dog.

As an observational study, the current report cannot make any causal inferences about the association between having a cat and an elevated hazard of dying from colorectal cancer. However, by controlling for the socio-demographics and behavior-related carcinogenic risk factors, including cigarette smoking, exposure to environmental tobacco smoke and sedentary lifestyle, we obtained a stronger rather than weakened association. Therefore, the current study provides additional evidence to rule out the possibility that the relationship between pet ownership and hazard of dying from colorectal cancer was spuriously generated from confounding effects. The most interesting findings of the current report might be that the association was mainly driven by having a cat, or potentially a bird. The pet type-specific association sheds light on the direction for future research to identify the pet-harbored carcinogenic risk factors.

The relatively healthy study population made the current study unique, but it was also a major limitation. The small number of deaths (3 colorectal cancer deaths among birth owners) due to colorectal cancer among a healthy study population prevented further stratification by ethnicity and other potential effect modifiers. The small number of colorectal deaths also made the estimates less precise. Using deaths rather than incident cases presented an issue. Mortality data reflect the biological effects of risk factors as well as the quality of health services available to participants. Colorectal cancer is historically prevalent among socioeconomically disadvantaged populations who have limited access to healthcare. Finally, the risk factors, including pet ownership, were assessed from the baseline survey without repeated measurements; misclassification may occur and statistical power was consequently compromised—generating underestimations instead of a spurious association. The duration (i.e., how long they kept pets in their household) and intensity (i.e., how many pets were kept in household) are especially important in determining associations between environmental exposures and chronic outcomes, such as colorectal cancer (White et al. 1998). However, these data were not available from the baseline survey, preventing us from assessing the dose-response association.

Despite the limitations, this study has strengths as well. To the best of our knowledge, the current study might be the first cohort study performed among a national cohort and therefore provides evidence with better generalizability. The comprehensive information collected made it possible to control for various socio-demographic and behavioral factors. Equally important, NHANES collected data using a rigorous protocol with extensive quality control procedures. The current study adds to the existing literature, which mainly consists of case-control studies derived from hospital-based studies. Future work should concentrate on the etiology and preventive strategies to mitigate the detrimental effects of having a pet in households, and larger studies with adequate statistical power are needed to investigate this issue in a more robust manner.