Agreement of offspring-reported parental smoking status: the RHINESSA generation study

Study population

The study population is from the RHINESSA generation study - Respiratory Health in Northern Europe, Spain and Australia (www.rhinessa.net). One parent of each offspring in RHINESSA had previously participated in the Respiratory Health in Northern Europe (RHINE) study (www.rhine.nu), part of the European Community Respiratory Health Survey (ECRHS) initiated in 1989–1992 with three study waves (www.ecrhs.org) performed ten years apart. RHINE data was collected from seven different centres in five countries: Reykjavik (Iceland), Bergen (Norway), Umeå, Uppsala and Göteborg (Sweden), Aarhus (Denmark) and Tartu (Estonia). Furthermore, offspring with parents from three additional ECRHS centres, namely Melbourne (Australia), and Huelva and Albacete (Spain), were included in the RHINESSA population. Written consent was obtained from each participant in all study centres. In total, 8260 offspring from the RHINESSA cohort were matched with one of their parents (n = 6045). We used self-reported questionnaire information about smoking status from the parents themselves (collected in RHINE III/ECRHS III (2010–2012) or RHINE II/ECRHS II (1999–2000)) as well as offspring-reported parental smoking status (collected, among adult offspring (> 18 years), 2013–2016). Smoking during pregnancy was assessed from a specific women’s questionnaire in the RHINE III study. The questionnaires used have been developed in the framework of ECRHS, RHINE and RHINESSA and are used in several studies before. All questionnaires are available online from the study webpages listed above.

We used information from 7185 adult offspring (18-51 years) with information on one of their parents, n = 5307 (27-67 years), obtained in prioritised order from either RHINE III, ECRHS III, RHINE II or ECRHS II with full information as the eligibility criterion. In the centres of Melbourne, Huelva and Albacete, all smoking data were assessed by interview-administered questionnaires while data from all other centres were mostly assessed by a self-administrated questionnaire.

A flowchart of the study population is provided in Fig. 1. Additionally, in the analysis of smoking during pregnancy, we used data from women participating in Rhine III, who also answered the questionnaire on women’s health, 807 offspring and their mothers (n = 679).

Reporting of smoking status

Parental smoking status during offspring’s childhood was defined based on the parents’ date of birth, age of smoking debut or current age and number of years they had smoked (current smokers) or the year they had quit (ex-smokers). Offspring’s childhood was defined as the age period 0–10 years.

Predictors for disagreement

The different tobacco products were converted to cigarettes. One cigarillo was converted to three cigarettes, one cigar to five cigarettes, and one gram pipe tobacco equalled one cigarette [12]. The amount of smoking was dichotomized as 10 or more cigarettes per day or less than 10 cigarettes per day on average.

Statistical procedures

Sensitivity and specificity were calculated with 95% confidence intervals [CIs]. Cohen’s Kappa estimate [K], with 95% CIs, was calculated to estimate the agreement of the offspring-reported parental smoking status during childhood and pregnancy with the parent reported smoking status during the same periods. Parents’ own report was considered the gold standard. Cohen’s Kappa interpretation was based on the following categories: poor agreement, < 0.2; fair, 0.21–0.40; moderate, 0.41–0.60; good, 0.61–0.80; and very good, 0.81–1.00 [13]

Multivariate logistic regression models were conducted to estimate the odds ratio [OR] of whether offspring’s sex, age, educational level, asthma status, own smoking status or own parental status and parents’ sex and amount of smoking were predictors for disagreement. The regression models were performed with repeated measurements due to multiple offspring from the same parent using proc. GENMOD function in SAS. The model was mutually adjusted for the included variables and further adjusted for study centre. Disagreement was defined as discrepant answers between offspring and their parents. The significance level was set at a p-value of < 0.05 (two-sided) and 95% CIs were calculated.

We performed sensitivity analyses where those excluded due to missing co-variables were included in the calculation of sensitivity, specificity and K. Furthermore, we performed sensitivity analyses where childhood was expanded to cover 0–18 years. We performed analysis by sex of the parents, and by the offspring’s own parental status. Subgroup analyses were conducted to examine if parents’ amount of smoking affected the sensitivity, specificity, or agreement. Further subgroup analyses were performed to examine whether false answers from the main analyses were false positives or false negatives. Additionally, sensitivity analysis with a reversed priority line (1. RHINE II, 2. ECRHS II, 3. RHINE III, and 4. ECRHS III) of the included data from the parents was performed to examine if the period of the data collection and perception of smoking at that time had an impact on the estimates. Separate analyses of centres were conducted for sensitivity, specificity and K, and adjustment for study centre was applied in the logistic regression models. All statistical analyses were performed using SAS (SAS Institute Inc., Cary, NC, USA) version 14.1.

Sensitivity analyses

Sensitivity analyses including those who were excluded due to missing covariates produce similar results, sensitivity 0.82 (95% CI 0.81–0.84), specificity 0.95 (95% CI 0.95–0.96) and κ 0.79, (95% CI 0.77–0.80). When childhood was expanded to cover 0–18 years, similar results were also observed, sensitivity 0.82 (95% CI 0.80–0.83), specificity 0.96 (95% CI 0.95–0.96) and κ 0.79, (95% CI 0.77–0.80).

Analyses by sex showed largely similar results with a slightly lower sensitivity 0.81 (95% CI 0.79–0.83) and κ 0.78, (95% CI 0.76–0.80) and similarly specificity 0.96 (95% CI 0.95–0.97) of the fathers compared to mothers, sensitivity 0.84 (95% CI 0.82–0.86), κ 0.80, (95% CI 0.78–0.82) and specificity 0.95 (95% CI 0.94–0.96), Additional file 1: Table S1. Analyses by offspring’s own parental status showed a lower sensitivity 0.78 (95% CI 0.76–0.80), and κ 0.77, (95% CI 0.75–0.79) and similar specificity 0.96 (95% CI 0.96–0.97) of the offspring without children, compared to offspring with children, sensitivity 0.87 (95% CI 0.85–0.88), κ 0.80, (95% CI 0.77–0.82) and specificity 0.93 (95% CI 0.92–0.94).

In subgroup analyses of parents’ smoking amount, the sensitivity 0.89 (95% CI 0.88–0.91) and κ 0.85, (95% CI 0.83–0.86) were increased when the parents had smoked more than 10 cigarettes per day during their offspring’s childhood compared to the main analysis and decreased if the parents had smoked less than 10 cigarettes per day, sensitivity 0.74 (95% CI 0.72–0.76) and κ 0.72, (95% CI 0.70–0.74). The specificities remained unchanged. Separate analysis of offspring and parental predictors for disagreement produced similar results as the main analyses with offspring and parental predictors in the same model.

Further analyses of the 715 false answers in the main analysis showed a higher prevalence of false negative answers (n = 511, 71%) compared to false positive answers (n = 204, 29%), meaning that the parents had reported smoking more often than the offspring. Characteristics of offspring themselves who reported false negative answers were similar to the total study population except higher educational level (University/College 66% vs 60%), whereas offspring who reported false positive answers were slightly older and more often females (59% vs 58%), smokers (17% vs 12%), and asthmatics (18% vs 16%). Offspring who reported false negative were less likely to report regarding their mothers (49% vs 55% females) compared to the total population, whereas offspring who reported false positive answers were more likely to report regarding their mothers (60% vs 55% females). Parents’ mean age was slightly higher for offspring who reported false negative as well as false positive compared to the total population (54 and 55 vs 53 years).

Sensitivity analysis of the reversed data priority produced similar sensitivity, specificity and Cohen’s Kappa as for the main analysis.

Analyses stratified by study centre with respect to sensitivity, specificity and Cohen’s Kappa showed largely the same result as for the combined analysis; however, with wider confidence intervals with kappa ranging from 0.71 to 0.83 (Additional file 2: Table S2).

Sex of the parents

The offspring were more likely to give a correct answer regarding their mothers’ than their fathers’ smoking status, which may be explained by the fact that mothers in general spend more time with their children, especially in early life, and consequently the offspring are more aware of their mothers’ smoking status. A similar argument was used by the “National Heart, Lung, and Blood Institute - Family Heart Study” of offspring-reported parental history of chronic heart diseases and diabetes that demonstrated high agreement with parental self-reported history; however, a lower agreement for asthma and hypertension [10], where they suggested that conditions which affect “daily routines such as medication use or diet changes” also give relatives more contact and awareness of the diseases. However, gender roles have changed over the years, especially in Northern Europe, and today, it is not always true that women spend more time than men at home with their children. However, data in the present study are based on offspring born between 1963 and 1998, and women are today still more likely to have custody of their children after a divorce [14], which could be part of the explanation of why the offspring are more aware of the mothers’ smoking status.

Amount of parental tobacco consumption

A low amount of parental tobacco consumption predicted discrepant answers confirming the results of Simard et al. [4]. The more the parents smoked, the more likely it is that children noticed their parents’ smoking habits. A limitation of the questions on smoking amount is that it is not specific for the offspring’s childhood, but reported as a long-term average, which may have varied over the lifetime of the parent’s smoking which could introduce a potential for misclassification.

We found a quite high sensitivity and an even higher specificity indicating a low risk of misclassifying non-smokers, while the risk of misclassifying smokers is higher but still limited. In line with these findings, we could see a higher prevalence of false negative answers compared to false positive answers among the offspring which may be due to the increasing awareness of smoking and second-hand smoking’s negative influence on health [15], and parents’ influence on children’s later smoking behaviour [16]. The use of indirect information could, to a higher extent, introduce misclassification of smokers compared to non-smokers and therefore underestimate the true prevalence of smokers.

Offspring

Offspring’s own parental status was shown to be a predictor for discrepant answers. A higher sensitivity and agreement was found among offspring with children themselves. These results could indicate that offspring with children are more aware of exposures in the past, and may start to ask their parents about their own childhood. Even though younger participants had a shorter recall period, the younger the offspring, the higher the odds ratio of discrepant answers, however, this was not significant. This could possibly be explained by the fact that perception of smoking has changed over time in society, e.g. greater knowledge about the harmful impacts of smoking on health [15] and parents could have avoided smoking in the vicinity of their children to a greater extent for the younger offspring. We chose to prioritize the use of the most recent data collected from parents so that answers from both parents and children reflected similar societal attitudes with respect to smoking. The sensitivity analysis of a reversed priority in preferred data provided similar results.

Limitations and strengths

Using parents’ own self-reported information as the gold standard probably introduces a risk of underestimation of the true prevalence of smoking [2–4]. The numbers of parents who smoked during the offspring’s childhood and pregnancy were based on participants from numerous Western countries as well as Australia and from several different decades which makes it difficult to compare the percentages of smokers with the known prevalence of specific countries or time periods.

In our study, the analysis of the discrepant answers between offspring and their parents was primarily false negative answers. The underestimation of the offspring’s report in addition with the lack of a true gold standard (e.g. saliva/urinary cotinine) will enhance the risk of underestimating the true prevalence of smoking. A consequence is that smoking information given on behalf of others is not ideal if the purpose of a study is to determine prevalence of smoking, but it can be useful in investigations researching associations between smoking and diseases, even though some misclassification can be expected.

A limitation of the study is that the wording for parents own report of smoking (which picks up a measure of “current” use) is different to the wording for the offspring-reported parental smoking (which picks up a measure of “regular” use), which is not identical concepts. For example, a parent who smoked very intermittently may consider themselves a smoker, but the offspring may not see this as regular use.

We do not have information on whether the offspring who completed self-administrated questionnaires consulted their parents to provide answers about parental smoking. However, we are reassured by finding no clear evidence of difference from centre specific analysis. The measures of agreement from centres based solely on interview data (where parental consultation was not possible) (Melbourne, Huelva, and Albacete) did not substantially differ from centres that collected data from self-administrated questionnaires, Additional file 2: Table S2.

A major strength of our study is the large population-based RHINESSA population, which enabled us to examine both male and female offspring as well as both male and female parents from a general population. A further strength is the ability to determine predictors for discrepant answers based on the large amount of available information from both parents and offspring.