Study design and data sources
This cross-sectional research is nested in the Lifelines cohort, a multidisciplinary, prospective, population-based study using a unique three-generation design to examine the health and health-related behaviors of 167,729 people living in the north of the Netherlands. Lifelines employs a broad range of investigative procedures to assess the biomedical, sociodemographic, behavioral, physical, and psychological factors contributing to the health and disease of the general population, focusing on multi-morbidity and complex genetics [16,17,18]. Between 2007 and 2019, Lifelines conducted two in-person assessments and three online follow-up questionnaires. A third in-person assessment is currently ongoing. However, these assessments and follow-up times do not completely match the period evaluated for the screening rounds evaluated. Therefore, for the current study, we retrieved data on the most recent sociodemographic, reproductive, and lifestyle factors available from the Lifelines cohort and linked them to data from the Dutch Nationwide Pathology Databank (PALGA) for 2000–2020 to determine participation in cervical cancer screening.
Setting: Dutch cervical cancer screening
In the Netherlands, primary screening for cervical cancer changed from cytology-based to high-risk human papillomavirus (hrHPV)-based testing in 2017 [19]. Before the change, women aged 30–60 years were invited every 5 years to undergo primary screening by cytology testing [19]. Since the change, women aged 30, 35, 40, 50, and 60 years have been invited to undergo primary hrHPV testing [20], with women aged 45, 55, and 65 years only invited if they had a hrHPV positive result or missed the last round of screening [20]. However, all women were tested in the first round of the hrHPV-based program (2017–2021) because their hrHPV statuses at ages 40, 50, and 60 years had not yet been established [20].
Population
To address regular participation in cervical cancer screening, we included only women from the Lifelines cohort who were eligible for all the four cervical cancer screening rounds between 2000 and 2019 (i.e., born between 1955 and 1974) [21]. As age is the main factor to invite women for screening in the Netherlands, the birth year was used to define the eligibility year for each screening round (e.g. A woman who was born in 1970 is eligible for her first screening in 2000 when she turns 30, and in 2005 when she turns 35, and so on) [6]. Women were excluded if they had undergone hysterectomy (based on self-report in the Lifelines questionnaire before 2000 and their PALGA records thereafter) or if they died before screening (based on Lifelines questionnaires).
Outcome
Data on participation in the cervical cancer screening were retrieved from PALGA records. Four screening rounds were evaluated: 2000–2004, 2005–2009, 2010–2014, and 2015–2019. In each screening round, a woman was considered participant when she had a primary screening test recorded within 36 months of the start of the eligibility year (except for women eligible in 2019, when we allowed a maximum time of 24 months). Otherwise she was considered non-participant [22]. Participation regularity was defined as follows: “regular” if women attended all four screening rounds, “irregular” if they attended one to three screening rounds, and “never” if we found no record of screening in any of the four rounds. Analyses on a second definition of regularity were also performed and are presented in the supplementary data.
Exposures and confounders
All the exposure and confounders were retrieved from lifelines study. To ensure the use of the most recent Lifelines data, we only included data from the last questionnaire or assessment with the variables of interest; if missing, we used the next most recent questionnaire. The following health-related behavioral factors were used as the main exposures: smoking habits, alcohol consumption, Lifelines Diet Score (LLDS), BMI, physical activity, television (TV) watching (as a proxy for sedentarism), sleep duration, hormonal contraception use, number of children, and age first childbirth. In addition, we used country of birth/ethnicity, educational level, income, and marital status as confounders that have known associations with participation in cervical cancer screening.
Smoking status was categorized as never, former, and current. Never smokers answered “no” to the question “Have you ever smoked for as long as 1 year?” Former smokers had to report being smokers for ≥ 1 year or having stopped for at least 1 month before questioning. Current smokers answered “yes” to the question “Do you smoke now, or have you smoked in the last month?” [23].
Alcohol consumption was calculated by dividing the average number of alcohol glasses consumed per drinking day by the number of drinking days per month. It was then categorized as high (> 1.5 drinks per day), light to moderate (> 0 and ≤ 1.5 drinks per day on average), or none [24].
We calculated the LLDS from a food frequency questionnaire, considering the relative intakes of different food groups with known positive (e.g., vegetables) or negative (e.g., red or processed meat) health effects on a scale from 0 (lowest diet quality) to 48 (highest diet quality) [23]. The LLDS was then categorized as low (2–23), middle (24–28), and high (29–46) based on minimum and maximum scores of 2 and 46, respectively.
BMI was grouped into underweight (< 18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25–29.9 kg/m2), and obese (≥ 30 kg/m2) categories [25].
Physical activity was evaluated by the Short Questionnaire to Assess Health-enhancing Physical Activity (SQUASH), although we only considered moderate-to-vigorous physical activity (MVPA) related to commuting and leisure time [23]. Based on the physical activity guideline set out by the Dutch Health Council, we categorized MVPA as low (<150 min/week), medium (150–299 min/week), and high (≥ 300 min/week). We then based sedentary behavior on the number of hours watching TV per day [26], categorized as low (≤ 2 h/day), medium (3–4 h/day), and high (≥ 5 h/day).
Total self-reported sleep per day was categorized according to the recommendations of the American National Sleep Foundation into adequate (7–9 h), marginal (6 or 10), and inadequate (< 6 or > 10 h).
For hormonal contraception, respondents could answer “yes” or “no” to the question “Have you ever used hormonal contraception?”.
We primarily used the child’s year of birth, as reported by mothers at the baseline and follow-up questionnaires, to estimate the number of births during the study. When this was absent, we used responses to the baseline question “how many children do you have?” All women who still had missing data were assumed to have no children if at least one questionnaire response indicated no pregnancies. Women were then grouped by the number of children (0, 1–2, and ≥ 3). The year of birth of the oldest child was used to estimate the age of the first child, and the mother’s age at this birth (≤ 26 years, 27–30 years, and ≥ 31 years) was estimated as the difference between her birth year and that of her oldest child.
Sociodemographic confounders included country of birth/ethnicity, educational level, and income, as reported previously [22], with the inclusion of marital status categorized into three groups: no partner, relationship without cohabiting, and relationship with cohabiting (including marriage).
Statistical analysis
Sociodemographic and behavioral factors are presented by participation regularity, using the chi-squared test for linear trend to estimate the association between each exposure/confounder and the outcome. To evaluate the association of behavioral factors with participation regularity, we performed univariate analysis by multinomial logistic regression for all participants. Missingness was treated as an additional category for each variable in the univariable model and addressed by multiple imputation in the multivariable model. The multivariable model included all statistically significant variables and presented in a forest plot. Since the rate of missing data was slightly higher among never participants compared to regular and irregular participants, we conducted a sensitivity analysis by running two additional multivariate models. One used missingness as an additional category for each variable and the other included only participants with complete data for each variable in the model. Odds ratios (ORs) are reported with their 95% confidence intervals (95%CIs). All analyses were conducted using IBM SPSS Version 25.0 (IBM Corp., Armonk, NY, USA).
Due to the small number of missing values for BMI, they were not reported to protect the confidentiality of the participants.
