Objective allergy markers and risk of cancer mortality and hospitalization in a large population-based cohort

The Committee on Human Subjects in Research of the University of Groningen reviewed the study and affirmed the safety of the protocol and study design and specifically approved this study. All participants gave their written informed consent.

We studied objective allergy markers, cancer mortality and hospitalization using the Vlagtwedde–Vlaardingen cohort study. The Vlagtwedde–Vlaardingen study was set up as a general population-based cohort study on the epidemiology of pulmonary diseases in exclusively Caucasian individuals of Dutch descent [12, 13]. This study started in 1965 and participants had medical examinations every 3 years until the last survey in 1989/1990. In Vlaardingen, only participants who were included at baseline (1965 or 1969) were approached for follow-up, whereas in Vlagtwedde new subjects aged between 20 and 65 were invited to participate at every survey. We updated the vital status of all participants on 31 December 2008 and evaluated five main cancer mortality outcomes, i.e., mortality from all types of cancer, lung, colorectal, prostate, and breast, either as primary or secondary cause of death. The causes of death were coded according to the International Classification of Diseases (ICD). Hospitalization data since 1 January 1995 were obtained using probabilistic matching based on date of birth, gender, and postal code. The probabilistic matching method was used because of privacy regulations. A match was defined if date of birth, gender, and postal code in our source-file (the Vlagtwedde–Vlaardingen cohort data) were exactly equal to those in the hospital admission registry file. Diagnosis at discharge was used to identify the reason for the hospitalization. The endpoints used for the current study were having at least one hospitalization due to any cancer or due to a specific type of cancer (i.e., lung, colorectal, prostate, and breast cancer). Subjects who were lost to follow-up or died within 2 years after the start of the hospitalization data (in 1995) and who were not hospitalized during these years were excluded from these analyses. However, subjects with a shorter registration period but with a hospital admission for cancer in this period are included in the group ‘at least one hospitalization due to any cancer.’

We collected data on age, gender, and smoking habits using the Dutch version of the British Medical Research Council questionnaire [12, 13]. We used the data of a subject’s first available survey. We defined smoking habits as follows: Never smoker and ever (i.e., ex and current) smoker (including pipe/cigar smokers).

The body mass index (BMI) was calculated as weight in kilograms divided by the square of the height in meters (kg/m²).

Peripheral blood eosinophil counts were assessed in a 1:11 dilution of peripheral blood with a Bürker counting chamber [12, 13].

Skin prick tests were performed at the first available survey. Four common aeroallergens (house dust, mixed pollen, epidermal products, and mixed molds) were applied intracutaneously to the forearm (Diephuis, Groningen, the Netherlands) [14]. Wheal diameters for each allergen were measured to the nearest half millimeter and coded on a six-point scale (0 = 0–5.0 mm, 1 ≥ 5.0–7.5 mm, 2 ≥ 7.5–10.0 mm, 3 ≥ 10.0–12.5 mm, 4 ≥ 12.5 mm, 5 ≥ 15.0 mm). Scores for the four allergens were added to a skin test sum score (minimum 0, maximum 20). Skin test positivity was defined as a skin test sum score ≥3 [15].

Serum total Immunoglobulin E (IgE) was determined at only one survey, i.e., the final survey, with the CAP system (Pharmacia, Woerden, the Netherlands) and expressed in kU/L [15].

Cancer was classified according to the ICD-coding system: Any type of cancer (ICD 7: 140–239 and 294; ICD 8: 140–239; ICD 9: 140–239 and 288; ICD10: C00-C97, D00-D48), lung cancer (cancer of trachea, bronchus, and lung) (ICD 7: 162, 163; ICD 8: 162, 163; ICD 9: 162, 163, 165; ICD10: C33, C34, C38, C39), cancer of colon and rectum (further referred to as colorectal cancer) (ICD 7: 153, 154; ICD 8: 153, 154; ICD 9: 153, 154; ICD 10 C18-C21), breast cancer (ICD 7: 170; ICD 8: 174; ICD 9:174, 175 and ICD10: C50), and prostate cancer (ICD 7: 177; ICD 8: 185; ICD 9: 185 and ICD 10: C61).

Descriptive analyses of the subject characteristics and the mortality and hospitalization data were performed. Differences between groups were tested with independent samples t test and Chi-square test for continuous and categorical variables, respectively. Blood eosinophil counts and serum total IgE were log-transformed to obtain normality of the distribution. Multivariate Cox regression (for mortality) and logistic regression (for hospitalization) with adjustment for age, gender, Forced Expiratory Volume in 1 s (FEV₁) as % of predicted, BMI (all at the first survey), and place of residence were used to estimate the effect of the allergy markers on the cancer outcomes. To determine whether the association of the three allergy markers was different for males and females, or for ever and never smokers, stratified analyses were performed, and interactions between allergy markers and gender, or smoking were tested. In the Cox regression analyses, censoring took place when the subjects were still alive, were lost to follow-up, or died of causes other than cancer or the specific cancer under study [16]. Time was defined from the first available survey until cancer mortality or until censoring in the analyses of eosinophils and skin test positivity. Similarly, in the analyses of IgE, time was defined from the only available survey, i.e., the final survey, until cancer mortality or until censoring. Finally, to investigate the robustness of our results, we conducted several sensitivity analyses. All analyses were performed at Statistics Netherlands (The Hague, the Netherlands). p values <0.05 (two sided) were considered to be statistically significant.

Among all 8,465 subjects, 4,505 (53.2 %) were alive, 1,194 (14.1 %) died due to cancer, 2,473 (29.2 %) died due to another reason than cancer, 158 subjects (1.9 %) died due to external causes such as an accident, suicide or homicide, in 13 (0.1 %) subjects the cause of death could not be determined, and 122 (1.5 %) subjects were lost to follow-up (Table 1). Of those subjects who died due to cancer, most died of lung cancer (n = 275, 23.0 %), followed by colorectal cancer (n = 134, 11.2 %), prostate cancer (n = 83, 7.0 %), and breast cancer (n = 117, 9.9 %) (Fig. 1).

Details on the associations between allergy markers and baseline characteristics of the subjects in 1965/1967/1969 and vital status in 2008 are presented in the Online Resource Tables 1 and 2, respectively.

Among all 8,465 subjects, 7,085 (83.7 %) subjects had data available on peripheral blood eosinophil counts and on all included covariates, 6,193 (73.1 %) had data available on skin test positivity and on all included covariates, and for 2,324 (27.5 %) subjects, data on IgE and on all included covariates could be obtained (Online Resource Table 3).

Subjects who died due to cancer were more often men, older, had a higher BMI, were more often smokers, and had a lower FEV₁ % predicted at the first survey than those who were alive. Subjects who died due to cancer had a higher level of peripheral blood eosinophils compared with subjects who were alive. Subjects who died due to cancer had less skin test positivity compared with subjects who were alive (p = 0.001). There were no significant differences in the level of serum total IgE between subjects who died due to cancer and those who were alive (Table 1).

Of the total number of 8,465 subjects, 6,174 subjects were successfully matched to the hospital admission file (in 20 subjects, a successful match could not be found, and 2,271 subjects were lost to follow-up or died before the start of registration of hospitalization in 1995). Of these 6,174 subjects, we excluded 91 subjects because they had a follow-up period shorter than 2 years and no hospital admission in these 2 years after the start of registration of hospitalization (Fig. 1). Among all 6,083 subjects with data on hospitalization, 1,022 (16.8 %) subjects were hospitalized for any type of cancer. Subjects, who had at least one hospitalization due to cancer were significantly older, were more often smokers, had a higher BMI, and had a lower FEV₁ % predicted compared with subjects who had no hospitalization. Subjects with at least one hospitalization due to cancer were less often skin test positive compared with subjects who had no hospitalization (p = 0.02). Subjects with at least one hospitalization due to cancer were older compared with subjects who had at least one hospitalization but not due to cancer (p = 0.03). There were no significant differences in the level of eosinophils, and the level of serum total IgE, between subjects who had at least one hospitalization for any type of cancer and those who were not hospitalized at all or those who were hospitalized but not due to cancer (Table 2).

Among all 6,083 subjects with data on hospitalization, 5,449 (89.6 %) subjects had data available on peripheral blood eosinophil counts and on all included covariates, 4,611 (75.8 %) had data on skin test positivity and on all included covariates, and 2,299 (37.8 %) subjects had data on IgE and on all included covariates (Online Resource Table 3).

In the total population, we found no significant association between number of eosinophils and cancer mortality or cancer hospitalization (Table 3; Fig. 2a). A higher number of eosinophils was significantly associated with decreased risk of colorectal cancer mortality in ever smokers (Hazard ratio (HR) (95 % confidence interval (CI)) = 0.61 (0.45–0.83); see Table 4) and males (0.60 (0.42–0.83); Table 5) (Online Resource Figure 1 and 2). The interaction between the number of eosinophils and smoking and gender was significant. To assess whether this association is gender or smoking dependent, we investigated the interaction between eosinophils and smoking separately in males and females and the interaction between eosinophils and gender separately in never and ever smokers. We observed a significant interaction between eosinophils and ever smoking within males and a significant interaction between eosinophils and male gender within ever smokers (Online Resource Table 4). There were no significant interactions between number of eosinophils and gender or smoking in the analyses on hospitalization due to cancer (Online Resource Table 5).

Skin test positivity was not associated with cancer mortality or cancer hospitalization in the total population (Table 3; Fig. 2b). Within females, skin test positivity was associated with a decreased risk of mortality from any type of cancer (0.59 (0.38–0.91); Table 5), and the interaction between skin test positivity and gender on any type of cancer mortality was significant. There were no significant interactions between skin test positivity and gender on any type of cancer hospitalizations (Online Resource Table 6).

Serum total IgE was not associated with cancer mortality or cancer hospitalization in the total population (Table 3; Fig. 2c). The association between total IgE and cancer mortality risk was not significantly different between ever and never smokers (Table 3). The association between serum total IgE and cancer mortality risk was significantly different between males and females, with a significantly increased risk of mortality from lung cancer among females (4.64 (1.04–20.70)) (Table 4). Higher levels of serum total IgE were associated with a lower chance of hospitalization for all types of cancer among males (0.76 (0.59–0.98) (Online Resource Table 6).

A sensitivity analysis on cancer mortality, excluding the subjects who were lost to follow-up or died within 2 years of the visit with the assessment of the allergy marker [(eosinophils: total n = 44, cancer mortality n = 10), (skin test positivity: total n = 41, cancer mortality n = 7), and (total IgE: total n = 30, cancer mortality n = 12)], gave similar results as the main analysis (results not shown).

In addition, we performed separate analyses on cancer as primary cause of death and as secondary cause of death. The results of these analyses were comparable to the results of the presented analyses where we analyzed cancer as either the primary or secondary cause of death (results not shown). Since FEV₁ could be on a causal path from allergy to cancer mortality, we performed a sensitivity analyses excluding FEV₁ from our Cox regression model. These analyses gave the same results as our main analyses (results not shown).

The different inclusion strategy of Vlagtwedde and Vlaardingen may introduce bias. We, therefore, stratified our analyses by place of residence. In addition, we performed a meta-analysis with these two studies and demonstrated evidence of association between both datasets. The meta-analysis showed the same result as the original pooled analysis (results not shown).

Furthermore, additional adjustment for years of recruitment gave the same results as our main analyses (results not shown).