Study area and population

Our study setting was Kaohsiung City, a harbor city located on the southwestern coast of Taiwan (153.6 km² and has 11 administrative districts), belonging administratively to Taiwan's National Health Insurance Program Kao-Ping District, which administers to 3,234,941 persons in Taiwan's Kaohsiung and Pingtung areas in 1999. During that year, 1,065,624 Kaohsiung City residents had applied for health insurance compensation. Ninety-six percent of Taiwan's citizens are enrolled in Taiwan's National Health Insurance Scheme, and 93% of the physicians there have contracts with the National Health Insurance Bureau [29], making its database nationally representative.

According to the Kao-Ping District insurance database for 1999, a total of 221,965 female Kaohsiung residents 40 years of age or older made health insurance claims. Kaohsiung municipal records confirmed that they had lived in the city for 5 years or more. These women were defined as study cases if they had received a diagnosis of definite or suspected chronic bronchitis (ICD-9 code: 491) at least twice in 1999. Those who had sought medical attention for traffic accidents (ICD-9 code: E800-E848) or acute gastroenteritis (ICD-9 code: 008.8; 009.1; 558.3; 558.9) in the same year and had never been diagnosed as having chronic bronchitis were used as study controls. We excluded both study cases and controls if they had been diagnosed with other pulmonary-associated diseases, including asthma (ICD-9 code: 493), pulmonary tuberculosis (ICD-9 code: 011.9 or 010-018), bronchiectasis (ICD-9 code: 494), fibrotic cyst (ICD-9 code: 277), pulmonary tumor (ICD-9 code:162), emphysema (ICD-9 code: 492), extrinsic allergic alveolitis (ICD-9 code: 495), or chronic airway obstruction, not elsewhere classified (ICD-9 code: 496). We were left with 1,846 study cases and 4,624 controls (Additional file 1).

Between June, 2000 and March, 2005, about one-third of the study cases (n = 622) were randomly selected, interviewed and administered pulmonary function tests in their homes. Once a study case was successfully recruited, interviewed and administered a pulmonary function test, we found one study control age-matched to within 3 years. The matched control was chosen from the same administrative area to adjust for the possible influence from external environmental hazards such as air pollution from large factories or traffic. After excluding those who refused to participate or those we could not locate, etc., we had 210 study cases and 210 controls successfully complete the interview and pulmonary function tests, making a response rate of 67.3% and 68.4% respectively. The average age (±SD) (years) of responders (n = 210) and non-responder study cases were 64.6 ± 9.6 and 63.2 ± 10.8, which was not significantly different (p = 0.11); of responders and non-responder study controls were 64.6 ± 9.7 and 64.7 ± 11.2, which was also not significantly different (p = 0.91). This study was approved by IRB at Kaohsiung Medical University, and all potential subjects signed informed consent forms.

SHS exposure measurement

Validation of SHS exposure

To verify the participant's response that she was a cigarette smoker or had been exposed to SHS exposure reported on the questionnaire, we randomly selected 4 smokers, 23 second-hand smokers, and 44 non-smokers to provide one-spot urine specimens so that we could measure cotinine levels. Urinary cotinine was extracted by a liquid-liquid extraction method and measured by liquid chromatography tandem mass spectrometry (LC/MS/MS) equipped with a triple-quadruple mass spectrometer and TurboIonSpray™ (API 3000™, Applied Biosystems, Foster City, CA, USA) [30]. The urine specimen was co-spiked with cotinine-d ₃ as an internal standard. The detection limit for cotinine was 0.07 ng/ml. Urinary creatinine was measured by spectrophotometer (U-2000, Hitachi, Tokyo, Japan) with a wavelength set at 520 mm.

Since cotinine can only be used to test recent exposure to cigarette smoke, we could only use it to confirm the questions regarding the smoking of cigarettes and exposure to cigarettes in the past three days prior to the interview, when the urine specimens were collected. The mean urinary cotinine levels in the four active smokers were 0.79 ± 0.13 mg/g creatinine, significantly higher than that found for the 23 second-hand smokers (0.006 ± 0.006 mg/g creatinine) and 44 non-smokers (0.002 ± 0.002 mg/g creatinine) (both p < 0.01). The urinary cotinine levels in the second-hand smokers were also significantly higher than those in the non-smokers (p < 0.01) (Additional file 2). Nineteen of 23 second-hand smokers had information about the number of cigarettes they had been exposed to in the past three days prior to the interview. We found a high correlation between the number of cigarettes they reported being exposed to and the urinary cotinine levels (Spearman correlation coefficients, r = 0.55, p = 0.02, n = 19) (Figure 1), confirming the validity of our SHS questionnaire.

Outcome measurements

Statistical Analysis

Paired t-statistics and univariate conditional logistic regression were first used to compare the demographics of the 210 study cases and their age- and district-matched controls. After we further crosschecked using criteria set by the ATS into three disease groups and severity of COPD, χ² and Fisher's exact tests, or ANOVA statistics, when appropriate, were used to compare the differences of different variables. We used polytomous logistic regressions to elucidate the association between cigarette smoking (yes vs. no) and SHS exposure (yes vs. no), including cumulative SHS exposure categories (<32 and ≥ 32 years), and health status outcomes (based on physician diagnosis and ATS criteria, and spirometry by GOLD criteria) in three age periods and then combined these into a lifetime period, adjusting for age, height, educational level, cooking status, burning incense, and tea consumption. Statistical significance of trend in the adjusted odds ratios (AORs) across cigarette smoking and SHS exposure and different health status outcomes were calculated by categorizing variables and treating scored variables as continuous. Multiple linear regressions were used to investigate the relationship between the same individual variables and pulmonary function (FEV₁, FVC and FEV₁/FVC). In addition, we calculated the population attributable risk percentages (PAR%) of cigarette smoking and SHS exposure for chronic bronchitis and probable chronic bronchitis, also for severity of COPD [34]. The data was analyzed using the SAS statistical package; all p-values were two-sided.

Characteristics of subjects on different health outcomes

Firstly, we had 210 study cases and 210 age- and district-matched controls (Additional file 1). There were no significant differences in the age, height, weight, education levels, and cigarette smoking between the two groups (data not shown). Only 3% of women had ever worked in a restaurant. The mean FEV₁ levels (L) in the study cases and controls were 1.48 ± 0.42 and 1.64 ± 0.49 (p < 0.01), and the mean FVC levels (L) were 1.90 ± 0.53 and 2.00 ± 0.61 (p = 0.02) respectively.

Impact of SHS exposure on health status based on physician diagnosis and ATS criteria

Impact of SHS exposure on health status based on spirometry by GOLD criteria

Impact of SHS exposure on pulmonary function status

Main findings of this study

This study indicates that, besides cigarette smoking, SHS exposure significantly increases the risk of chronic bronchitis in non-smoking Taiwanese women by using different outcome indicators, including physician-diagnosis/ATS criteria, the severity of COPD [6] and pulmonary function (FEV₁ or FEV₁/FVC ratio) impairment. We also used the urinary cotinine levels for the validation of self-report to SHS exposure. These results were also seen when exposure assessment was based on duration of SHS exposure at home, rather than household-reported cigarette smoke use alone.

Our results of increased risk of chronic bronchitis with SHS exposure in women concur with some previous studies [24, 26, 27]. However, some of the studies have shown mixed results (Additional file 3). The definitions of COPD/chronic bronchitis may be one of the reasons. We used two sets of disease classification of chronic bronchitis concurrently, to reduce the possibility of misclassification of our surveyed subjects and illustrate the effect of chronic bronchitis more clearly. One case-control study conducted by Kalandidi and his colleagues found that women married to smokers and exposed to ≤ 1 pack/day were at 2.5 times (95% CI: 1.3-5.0) a greater risk of having COPD than those married to non-smokers, but did not find a similar increase in women whose spouses smoked > 1 pack/day [24]. Simoni and his colleagues, performing a similar study, found female spouses of smokers to be at 2.24 times (95%CI: 1.40-3.58) more risk to have obstructive lung disease than those married to non-smokers. Unfortunately, this study population covered asthma or emphysema disease and this might reflect the complex nature of the disease [27]. In a longitudinal study, Sandler and his colleagues followed 14,783 healthy subjects exposed to SHS for 12 years and found that the estimated relative risk (RR) of death for emphysema or bronchitis was 5.65 (95% CI: 1.19-26.8); this result may be confounded by the small number of deaths (13) in that study [26].

A dose-response relationship was noted between duration of SHS exposure at home by the cut-point of 32 SHS exposure years and COPD severity based on spirometry, although it was present based on the criteria of physician diagnosis and crosschecked with ATS criteria. The probable reason is: the category of disease phenotype by ATS criteria, which was subjective, was not as accurate as by pulmonary function tests, which was objective. One study reported that only those with long duration of SHS exposure (42 years or more at home) were at increased risk of COPD (AORs:1.55; 95%CI:1.09-2.21) [17]. Thus, our findings add additional information about the association of cumulative SHS exposure with COPD risk in non-smokers.

Cotinine in urine is widely used as a biomarker of SHS exposure in epidemiology studies [35]. Although our study relied on self-reports of SHS exposure, we measured urinary cotinine levels in a subset of study subjects to verify the questionnaire data, as similarly done with other studies [23, 27, 28]. We found the urinary cotinine levels in smokers, second-hand smokers and non-smokers to agree well with a report that urinary excretion of the nicotine metabolites in smokers is approximately one hundred times higher than that observed in second-hand smokers and non-smokers [36]. In our study, the more exposure to SHS our second-hand smokers self-reported per day, the higher their cotinine levels, confirming the validity of response to the questionnaire item. This agreement has been found by others [37–39]. Eisner and his colleagues reported the first study indicating that the highest urine cotinine tertile was associated with poorer COPD severity scores both in cross-sectional or 1-year follow-up study [18], which adds substantive additional evidence that SHS exposure is deleterious for patients with COPD by objective measures of urinary cotinine levels.

What this study adds

PAR% has been used to judge priorities for public health action [34, 40]. Because there is a high prevalence of cigarette smoking in Caucasian men and women, several studies in those countries have considered the PAR% with regard to smoking and COPD [8, 41, 42]. The smoking attributable fraction of COPD mortality in the United States has been reported to be 69.4% for women. In that same time period, smoking was prevalent in 28.3-31.6% of the women and 31.8-40.6% of the men in the United States [8]. In Taiwan, the gap in smoking prevalence between women and men (3-4% vs. 55-60%) is much wider, so using PAR% would be helpful to study the contribution of SHS exposure to the risk of chronic bronchitis. Eisner and his colleagues reported that the PAR% was eleven percent for the highest quartile of home SHS exposure and seven percent for work exposure, though it is not known how they calculated these percentages [17].

The results of the above studies are, except for age, based on a single risk factor without regard to other factors [27, 41]. Using a multivariate model to calculate summary PAR% [34], we found although the AORs for chronic bronchitis for smokers is high, the prevalence of cigarette smoking among women is low. This made the PAR% for chronic bronchitis caused by cigarette smoking to be quite low (23.2%) (Table 2). In contrast, we found the AORs for chronic bronchitis showed that women exposed to SHS were 3.65 times more likely than those not exposed to SHS to have chronic bronchitis. This risk combined with a very high prevalence rate for exposure to SHS yielded a 47.3 PAR%, higher than the report from Simoni and his colleagues (adjusted PAR% = 12%) [27], making SHS a more important public health issue in non-smoking women. Similar results of PAR% occurred when we analyzed the SHS-attributable fraction of COPD severity (Table 3). The above results suggest that SHS exposure may be the underlying reason for the high prevalence of chronic bronchitis among non-smoking Taiwanese women. As shown in our female population, who smoke relatively rarely (3-4%), but live with high male smoking prevalence (55-62%) [9], this provides a good opportunity to add information on poorly investigated respiratory disease effects of lifetime SHS exposure in Taiwanese non-smoking women.

Limitations of this study

Our study has some limitations. First, the prevalence rate for chronic bronchitis in women was relatively low, probably because we required that each woman must be diagnosed as having chronic bronchitis at least twice that year in order to be included as an experimental case from our national claims data. This would cause an underestimation. Second, we have only randomly selected about 17% study cases for interview to date and the response rates in the groups of study cases and controls were about sixty-seven percent, which was not very high. This might increase the likelihood of selection bias; however, the age of responders and non-responders was comparable, and unlikely to bias our results. In addition, subjects were blinded to the purpose of our study, so their concern about lung function with their exposure or disease outcome should not cause bias in our study. Third, in this study, there was probably some recall bias regarding estimates of lifetime exposure status. Still, the misclassification of exposure is likely to be random, which would have null effect on our results. Regardless, we were still able to find significant risk, suggesting that this bias might not exert much influence on our results.