Study population and participants

EstBB is a volunteer-based population biobank with data from more than 210,000 individuals who were 18 years or older at recruitment, of whom 65% were female [24]. The cohort represents approximately 20% of the total adult population of Estonia (n_{total_adult} = 1,077,308 in 2025) [35]. Nearly all participants (99.5%) are insured by the EHIF [24], which covers 95% of the total Estonian population [36], suggesting the potential for selection bias due to insurance status is minimal. The comprehensive description of the cohort has been published previously [24]. Health-related data are linked from the Estonian Health Information System, EHIF, Tartu University Hospital Database, and many others. The information about medical diagnoses dates to 2004 and is stored according to the ICD-10 Version:2016 system. The EstBB 206K data freeze—with follow-up until the end of 2022—was used in this study.

All female EstBB participants were included in the study, but the focus was on those with at least one prescribed HC purchase record between 2004 and 2022 and aged 15–55 at the time of purchase. Basic descriptive characteristics of HC users, such as the age of the first observed HC usage period, the age at the start of each HC usage period, and the body mass index (BMI) closest to the start of each usage period, were used for analysis. BMI values originated from self-reported data or have been extracted from EHRs and were categorised by Centers for Disease Control and Prevention criteria: (a) underweight (BMI < 18.5), (b) healthy weight (18.5 ≤ BMI < 25.0), (c) overweight (25.0 ≤ BMI < 30.0), and (d) obese (BMI ≥ 30.0).

Hormonal contraceptive usage

In Estonia, all HCs except for emergency contraceptives require a prescription. Data on HC prescriptions and purchases originates from the EHIF, which subsidises medical expenses for people covered by national health insurance. The purchase invoice records the purchase of prescribed HC in a pharmacy. The linking procedure has been previously described in Leitsalu and colleagues [37]. Specifically, our dataset includes the purchase invoices from 2004 to 2022. For analysis, we used the Anatomical Therapeutic Chemical (ATC) Classification System code and the active substance name of the prescribed and purchased drug, the purchased dosage, the package contents, and the purchase date.

HCs were defined as purchased prescriptions of ATC level 3 G02B and G03A, and ATC level 5 G03HB01 (Fig 1). There was no information about emergency contraceptives (ATC level 4 G03AD), as a prescription is not needed for their purchase in Estonia. All entries where prescribed and purchased package ATC codes did not match were excluded from the analysis (n_purchases = 30,930 (1.8%)). In the following filtering step, all entries of short-acting HCs where the purchased dosage was less than 0.3 or greater than 6 were excluded (n_purchases = 653 (0.04%)). This had a negligible impact on prevalence estimates and analyses, and ensured the dataset was free from mistype errors. In the Estonian healthcare system, buying less than a third of a HC package is not possible, and gynaecologists or midwives prescribe, per visit, a maximum of three prescriptions, each for two short-acting HC packages (i.e., for a six-month coverage period). Finally, purchases for which the individual’s age at purchase was outside the 15–55 range (n_purchases = 1,306 (0.1%)) and individuals for whom BMI values closest to the HC usage start date were outside the range of 14.0–43.0 (n_individuals = 851 (1.1%)) were excluded.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of input data preparation for the construction of hormonal contraceptive (HC) usage periods and classification of HC user types using linked Estonian Biobank (EstBB) and Estonian Health Insurance Fund (EHIF) medication prescription and purchase data.

Several filtering steps were applied to identify eligible purchases and users: (i) selecting purchases of relevant Anatomical Therapeutic Chemical (ATC) codes (G02Bxxx, G03Axxx, G03HB01); (ii) matching prescribed and purchased ATC codes, excluding mistyped dosages, and restricting purchases to the relevant purchase age range; and (iii) restricting to individuals with body mass index (BMI) values within the eligible range, recorded closest to the purchase date. The final number of purchases (n_purchases) and users (n_users) after each step is shown. The final HC purchase data were converted into HC usage periods, which were used to classify HC users. The manually defined data points are for illustrative purposes and are not real observations.

https://doi.org/10.1371/journal.pmed.1005086.g001

The purchasing data was converted into HC usage periods. Specifically, the purchased dosage and package contents for each HC method were used to calculate the total number of covered days per purchase, assuming that the contraceptive was used consistently and according to guidelines. The default number of covered days per purchase for pills, transdermal patches and vaginal rings was set to 28 or 35 (latter only for G03AC03), for implants to 1,095 days (corresponding to three years), while the coverage days for intrauterine devices (IUDs) depended on the formulation dosage (1,095, 1,835 or 2,190 days corresponding to 3, 5 or 6 years for 13.5, 19.5 mg or 52 mg levonorgestrel-releasing IUD (LNG IUD), respectively). Inferred HC usage periods were additionally censored with ICD-10 pregnancy codes (O00–O99, excl. O85–O92 (complications predominantly related to the puerperium)) that occurred within the original coverage period. For example, if a pregnancy code appeared within an inferred HC usage period, we ended that period at the pregnancy diagnosis date. We did not backdate before the pregnancy date, as pregnancy-related diagnoses vary widely in gestational timing. This approach was applied across all HC types to avoid differential bias. Regular purchases (<90 days of pause between two purchase dates) of the same HC were considered one usage period since the fecundity is generally restored in two to three months after discontinuation [38,39]. Furthermore, for HC formulations with intermittent dosing, the drug-free days were also considered as part of the usage period. Any purchase of a different HC formulation (different ATC level 5 code) would mark the start of a new usage period. There was no differentiation between brands, dosages, or regimes for the same HC formulation.

To answer different research questions, HCs were grouped into ATC levels 4 or 5 accordingly and categorised by hormonal composition, route of administration, duration of action and contraceptive method type. Combined oral contraceptives (COCs) were classified by pharmacological properties and progestin generations as: (a) anti-androgenic (cyproterone/ethinylestradiol, chlormadinone/ethinylestradiol), (b) second-generation (levonorgestrel/ethinylestradiol), (c) third-generation (desogestrel/ethinylestradiol, gestodene/ethinylestradiol, norgestimate/ethinylestradiol, norelgestromin/ethinylestradiol), (d) fourth-generation (drospirenone/ethinylestradiol, dienogest/ethinylestradiol), and (e) estradiol preparations (nomegestrol/estradiol, dienogest/estradiol valerate). While grouping COCs by generation is informal [40], it is a common practice in the literature, where different COC generations are associated with different side effects profiles, especially thrombosis risk levels [41,42]. However, as it also varies across studies, we explicitly define our grouping criteria for clarity.

Finally, all HC users were classified as broad switchers, rapid switchers, non-switchers and rapid discontinuers following these criteria: (a) broad switchers were individuals who used two or more different HC formulations during the study observation period, (b) rapid switchers were individuals who had short gap (≤90 days) between two different formulation usage periods and short (<90 days) pre-switch formulation usage period, (c) non-switchers were individuals who had at least one long (>180 days) usage period and used only one formulation during the study observation period, and (d) rapid discontinuers were individuals who had only one short (<90 days) usage period and used only one formulation during the study observation period. Rapid discontinuers were additionally verified if they had any recorded diagnoses after the first purchase date, as evidence of not moving out of Estonia (which could otherwise result in missing purchases), and the purchase was made more than 90 days before the end of the study period (to rule out artefacts created by the end of the observation period). Overall, broad switchers are a wider class, with rapid switchers being its subset, while the remaining classes are mutually exclusive of each other and switchers (S1 Appendix). While the terminology was adopted from existing research [43,44], the thresholds were defined based on the clinical guidelines recommending an initial three-month ‘trial’ period and the Estonian prescription system.

Longitudinal diagnosis history and thromboembolism risk factors

To identify clinical diagnoses before, during and after HC use, EHRs with ~4,500 ICD-10 three-character category and four-character subcategory diagnosis codes were used. The illustration of the strategy is shown in Fig A of S2 Appendix. First, the records of clinical diagnoses in the three-month window/90 days preceding the first-time-observed HC usage period start date were used to identify potential non-contraceptive reasons for HC initiation (Table A in S3 Appendix).

Second, when evaluating diagnoses during the HC usage period and up to three months after the usage period end date to identify diagnoses related to potential side effects and possible reasons for HC discontinuation, we only considered first-time-ever diagnoses to avoid misinterpretation from pre-existing conditions. The same approach was used for diagnoses during the gaps between two different HC usage periods of rapid switchers. Our list of the curated selection of diagnoses related to potential side effects, based on side effects listed in patient information leaflets (Table A in S3 Appendix).

Third, the prevalence of VTE/ATE was evaluated across the entire study period. VTE and ATE were defined using ICD-10 diagnosis codes (Table A in S3 Appendix).

Expanding on the approach of Kurvits and colleagues [13], in the subset of women who purchased HCs in 2022, thromboembolism risk factors were assessed in a two-year period ending on the first purchase date in 2022 (Fig B in S2 Appendix). All thromboembolism risk factors were divided into three classes: (a) health conditions, (b) medication use, and (c) genetic risk factors. Health conditions were defined using the ICD-10 codes (Table A in S3 Appendix) [13,45]. Previous thrombotic events, family history of thrombotic events in first-degree relatives (parent–offspring or full siblings), and age >35 in 2022 were considered additional risk factors. The familial relationship between participants was inferred centrally for the EstBB. It is estimated based on coefficients calculated from the genetic data (kinship inference and identical-by-descent segment inference for close relatives) using the KING software, which implements a robust relationship inference algorithm [46].

The concomitant use of other medications associated with an increased risk of thromboembolism during the HC usage period was identified using purchasing data of ATC codes H02 (corticosteroids), N05A (antipsychotics), and N06A (antidepressants) [13].

Additionally, the imputed genotype data were assessed for the presence of two known VTE-associated variants (FVL (rs6025, MAF = 0.02) in the F5 gene and PTM (rs1799963, MAF = 0.01) in the F2 gene). The genotyping and imputation procedure of the EstBB cohort has been described previously [24]. Briefly, all samples have been genotyped centrally using the Global Screening Array (Illumina) and imputed using a population-specific reference panel of 2,695 whole-genome sequencing samples [47]. Imputation INFO scores for both variants were >0.96, confirming the good quality of the genotypes.

Statistical analyses

The categorical data were summarised using frequency calculations, and the continuous data were summarised by calculating the mean, standard deviation (SD), median and interquartile range (IQR [Q1, Q3]) where relevant. The annual use of HCs was calculated by summing up all active users in a given year. Annual HC users were stratified into five age groups (15–19, 20–29, 30–39, 40–49, and 50–55) using age at purchase, calculated by subtracting the birth year from the purchase year. The age at the start/end of the HC usage period was calculated similarly to the age at purchase. The first-ever observed HC usage period during the study was considered the initiation period, and users were referred to as initiators. Annual prevalence was calculated by dividing the number of active users in the specific age group in the given year by the total number of women in the EstBB of that age group in that year. The proportion of each HC type for each calendar year and age group was calculated as the total number of covered days by HC type divided by the total number of covered days by all HCs in that year and age group. The prevalence ratio (PR) with 95% confidence interval (CI) was used to compare thromboembolism risk factor prevalence between HC groups, for consistency, as these provide more interpretable effect estimates for common outcomes (>10%). For other categorical associations, the odds ratio (OR) with 95% CI was reported. Fisher’s exact test was used to assess statistical significance for all comparisons between categorical variables. The significance threshold was set to p < 0.05, except for risk factor comparisons, where Bonferroni correction was applied to account for multiple testing across 15 factors (p < 0.003). The data were analysed and visualised using the following packages: R base, data.table (v1.15.4) [48], dplyr (v1.1.4) [49], tidyr (v1.2.0) [50], stringr (v1.5.1) [51], lubridate (v1.9.3) [52], formattable (v0.2.1) [53], epitools (v0.5.10.1) [54], ggplot2 (v3.5.1) [55], ggalluvial (v0.12.5) [56], viridis (v0.6.5) [57], scales (v1.3.0) [58] and patchwork (v1.2.0) [59] in the R Statistical Software v4.1.3 [60].

Ethics approval and consent to participate

The ethical approval 1.1–12/624 for the study was granted by the Estonian Committee on Bioethics and Human Research (Estonian Ministry of Social Affairs), using data according to release application number 6–7/GI/16011 from the Estonian Biobank. All participants have signed a broad written informed consent form.

General use trends

The study cohort included 132,731 women, and we focussed on 73,071 who had purchased HC at least once between the ages of 15 and 55. HC user characteristics are summarised in Table 1. Women contributed 1,040,298 person-years of follow-up (from the first purchase date), with an average follow-up of 14.2 years per person. The average age at the first HC purchase was 27.9 (SD = 9.2) years, with the highest number of initiators at the age of 19 (n_users = 4,842). The BMI distribution at the first purchase date was comparable to that reported for the Estonian female population of similar age [61].

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Demographic characteristics of Estonian Biobank (EstBB) women (N = 73,071) who purchased a prescribed hormonal contraceptive (HC) at least once between ages 15 and 55 in the period from 2004 to 2022.

https://doi.org/10.1371/journal.pmed.1005086.t001

Across 19 years, 20 different HC formulations with five distinct routes of administration (intrauterine device, intravaginal ring, oral pill, transdermal patch and subdermal implant) were used. The most widely used HCs were COCs containing gestodene/ethinylestradiol (GSD/EE, n_users = 27,651) and drospirenone/ethinylestradiol (DRSP/EE, n_users = 26,030), levonorgestrel-releasing IUD (LNG IUD, n_users = 24,465), desogestrel-only pill (DSG, n_users = 16,557) and CHC ring containing etonogestrel/ethinylestradiol (ENG/EE, n_users = 13,376). Detailed information about all purchased HCs across the study period is presented in Table 2.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Overview of hormonal contraceptives (HCs) used by Estonian Biobank participants from 2004 to 2022.

https://doi.org/10.1371/journal.pmed.1005086.t002

The demographic structure of our study cohort has shifted notably over the years, reflected in a decline in the proportion of individuals aged (constructed age, see Methods) 15–29, from 40.6% in 2004 to 15.3% in 2022 (Fig 2A, Table B in S3 Appendix). Accordingly, the annual prevalence of CHC users was much higher than that of POC users, but has decreased with time (Fig 2B, Table C in S3 Appendix), while the prevalence of POC users steadily increased. Although our estimated user prevalence was systematically higher than that reported by Kurvits and colleagues [13], who estimated the annual prevalence of HC users aged 15–49 in Estonia from 2005 to 2019, the pattern of yearly changes was the same. Specifically, the annual prevalence of POC users in the EstBB increased across all age groups (15–19: from 0.3% in 2004 to 10.9% in 2022, 20–29: from 3.9% in 2004 to 17.2% in 2022, 30–39 from 3.2% in 2004 to 21.3% in 2022, 40–49: from 0.9% in 2004 to 27.6% in 2022 and 50–55: from 0.03% in 2004 to 18.0% in 2022). The annual prevalence of CHC users in the adolescent and the oldest age groups of EstBB increased (15–19: from 17.0% in 2004 to 38.2% in 2022, and 50–55: from 1.0% in 2004 to 2.0% in 2022), while for the other age groups it decreased (20–29 from 39.4% in 2004 to 26.6% in 2022, 30–39 from 21.5% in 2004 to 14.8% in 2022, and 40–49: from 9.6% in 2004 to 8.5% in 2022). Similar trends were observed by Kurvits and colleagues [13], except that our dataset showed an increase in CHC user prevalence in groups 15–19 and 20–29 from 2015 onward, and their study lacked information for the 50–55 group.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Overview of hormonal contraceptive (HC) use trends from 2004 through 2022.

A: Age dynamics in the Estonian Biobank (EstBB) study sample (HC users and non-users combined) during the study period. B: Annual prevalence (%) of HC users in different age groups across studied years in EstBB (darker colour) and Kurvits and colleagues [13] (lighter colour), shown separately for two main HC classes, progestin-only hormonal contraceptives and combined hormonal contraceptives. C: Distribution of total covered days by HC type across studied years, stratified by age groups (15–19, 20–29, 30–39, 40–49, 50–55). Users using more than one HC type within the same calendar year contribute to each relevant group. Abbreviations: IUD, intrauterine device; COC, combined oral contraceptive; CHC, combined hormonal contraceptive.

https://doi.org/10.1371/journal.pmed.1005086.g002

The changes in the distribution of total covered days (i.e., active use) by HC type among women from different age groups throughout the study observation period are presented in Fig 2C and Table D in S3 Appendix. While generally there was a continuous decrease in anti-androgenic, 2nd and 3rd generation COC use, the proportion of 4th generation COC use was increasing, then remained steady in more recent years. An exception to this trend was observed in the 15–19 and 20–29 age groups, where the proportion of 2nd generation COC use began increasing again from 2015 (Fig 2C and Table D in S3 Appendix). The use of estradiol COCs and HC implants began in 2010 and 2014, respectively. The proportions of different HC types varied slightly between years, reflecting the availability of specific formulations, as some multiphasic COCs and progestin-only pills (POPs), such as gestodene CHC (GSD/EE(s)), levonorgestrel-only pill (LNG) and drospirenone-only pill (DRSP), were only available on the market and used for 2–4 years.

While levonorgestrel-releasing IUDs (LNG-IUD) were increasingly used and primarily by older women (accounting for 89.4% of active HC use in the 50–55 age group in 2022), CHC dominated among younger and middle-aged women, especially in the early 2000s and 2010s, with the highest usage of 3rd and 4th generation COCs. On the other hand, POPs accounted for the highest proportion of active HC use among the 20–29 and 30–39 age groups (6.9% and 8.0% in 2022, respectively), with these percentages remaining stable over the years. Estradiol COCs were the least prevalent type of HCs and had minimal variation in their share of active HC use across the studied period (up to 0.9% of active HC use across all age groups throughout the years). CHC rings and patches were almost equally prevalent across all age groups and displayed a declining trend (Fig 2C). Etonogestrel-releasing (ENG) implants were more common over the years among adolescents aged 15–19 (8.2% in 2022) and women aged 20–29 (7.2% in 2022), whereas among women in older age groups, the proportions remained steadily low.

First observed hormonal contraceptive usage period

The first purchased prescriptions in 83.4% of women were CHCs, notably 36.1% were 3rd generation COCs and 24.9% were 4th generation COCs, irrespective of BMI category (S4 Appendix). However, each increase in the year of the first purchase was associated with lower percentages of 3rd generation COCs (decreasing from 55.6% of initiators in 2004 to 14.8% in 2022) and higher percentages of LNG-IUDs (increasing from 1.7% of initiators in 2004 to 36.0% in 2022). The 4th generation COCs peaked in 2011 with 38.7% of all initiators choosing this method. POPs were most commonly used by the 30–39 age group (11.4%) and by women from the obese category (9.4%), while LNG-IUDs were chosen by 39.5% of women from the 50–55 age group. Estradiol COC and ENG implant were the least used options for HC initiation (0.2% and 0.3%, respectively). The HC formulation abbreviations and their definitions are presented in Box 1.

Box 1. Hormonal contraceptive abbreviations used in the study.

Download:

• PNG
  larger image
• TIFF
  original image

https://doi.org/10.1371/journal.pmed.1005086.t003

The first contraceptive period length of specific HC formulations varied from less than a month (<30 days) to more than three years (>1,095 days), with a median of approximately 5.5 months (168 days) as the most common for the majority of HCs, except for long-acting methods, such as LNG-IUD and ENG implant, and multiphasic CHCs, such as LNG/EE(s) and GSD/EE(s) with higher medians (2,190, 1,027, 384, and 279 days, respectively, Fig 3). Some usage period lengths were underestimated when the first purchase was made near the end of the observation period or when the woman continued using the method past the age of 55, which exceeded the study age limits.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Length of first observed hormonal contraceptive (HC) usage period for different HC formulations.

Specific HC formulations are listed on the y-axis (see Box 1 for definitions), while the HC formulation usage period length in days (days shown on the logarithmic scale) is plotted along the x-axis. The colour illustrates the HC type. The box represents the interquartile range (25th to 75th percentile), with the central diamond shapes marking the median, and the black dots marking outliers. Abbreviations: IUD – intrauterine device; COC, combined oral contraceptive; CHC, combined hormonal contraceptive; (s), CHC pill with sequential preparations; (p), transdermal patch.

https://doi.org/10.1371/journal.pmed.1005086.g003

Looking into the ICD-10 three-character category diagnosis codes recorded up to three months before the first HC usage period start date, we found 1,233 different codes, where 53.1% (n_users = 38,820) of HC initiators had a Z30 diagnosis code related to contraceptive management, and 11.2% (n_users = 8,151) did not have any record of the medical condition. Additionally, 4.8% (n_users = 3,531) had heavy menstrual bleeding (N92), 4.0% (n_users = 2,922) medical abortion (O04), 2.4% (n_users = 1,744) dysmenorrhoea (N94), 1.6% (n_users = 1,202) acne (L70), and 0.7% (n_users = 472) PCOS (E28.2), which were from the list of curated diagnoses we considered as a potential reason for starting HCs.

Previous studies have shown changes in prescribing and purchasing trends during the Coronavirus Disease 2019 (COVID-19) pandemic, driven by lockdowns and limited access to healthcare services [62]. We investigated the impact of the COVID-19 pandemic and pandemic-induced lockdowns on HC initiation, comparing 2019 (pre-COVID-19 period) to 2020 and 2021 (COVID-19 periods). Overall, up to 55.4% of first-ever initiators were consistently from the 20–29 age group, and up to 50.0% of initiators with a history of prior HC use were from the 30–39 age group. Our longitudinal cohort’s yearly number of HC users gradually declined from 2019 to 2021. While the population-weighted average rate of first-ever initiators decreased from 1.3 to 1.1 to 0.8 during this period, other initiators decreased from 9.4 to 8.6 to 8.3, which likely also reflects the ageing dynamics of our study cohort. Furthermore, the overall monthly purchase trends differed notably in 2020 (Fig A in S5 Appendix), with March 2020 20% above the grand mean and April 2020 20% below it. Specifically, March had higher purchase numbers of CHC pills, patches, rings, and POPs than March 2019, with the peak on March 13th (Fig D in S5 Appendix). Conversely, LNG-IUDs and implants had their lowest record numbers in March and April 2020. Following June 2020, only 2nd-generation COCs and LNG-IUDs had a slight increase in purchases compared to the corresponding periods in 2019. Also, compared to 2021, only 2nd-generation COCs, estradiol COCs, implants, LNG-IUDs, and POPs returned to numbers similar to those recorded in the pre-COVID-19 period.

Hormonal contraceptive switching

While 35.8% (n_users = 26,151) of women used only one HC formulation during the observed study period, 64.2% (n_users = 46,920) switched to a different one (ranging between 2 and 11 different HC formulations, Fig 4A). Specifically, out of all HC users, 23.9% (n_users = 17,453) were classified as non-switchers, 46.5% (n_users = 33,991) as broad switchers, 17.7% (n_users = 12,929) as rapid switchers, 7.3% (n_users = 5,362) as rapid discontinuers, while 4.6% (n_users = 3,336) did not fit into any of the defined categories. The latter group included, for example, users of one HC formulation for less than 180 days, but more than 90 days (see Methods). From these types of users, we were interested in non-switchers, rapid switchers and rapid discontinuers (user characteristics presented in Table E in S3 Appendix). Non-switchers and rapid switchers were predominantly women from the 20–29 age group (33.8% and 48.8%, respectively), while rapid discontinuers were from the 30–39 age group (32.2%). The non-switcher usage periods ranged from 182 to 6,934 days, with an average of 1,658 days (corresponding to 4.5 years), whereas in rapid switchers, the average time to switch from one formulation to another was 49 days.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Hormonal contraceptive (HC) usage behaviour.

A: Number of different HCs purchased per individual during the study observation period. B: Graphical illustration of switching pathways of rapid switchers (pre-switch formulation is the HC method individuals switched from, while the new formulation is the HC method individuals switched to; see Box 1 for HC formulation abbreviations). The colour illustrates the HC type. C: Main ICD-10 diagnoses recorded between the first purchase of pre-switch HC formulation and the new formulation in rapid switchers. The coloured bars indicate the type of pre-switch HC method. Abbreviations: IUD, intrauterine device; COC, combined oral contraceptive; CHC, combined hormonal contraceptive.

https://doi.org/10.1371/journal.pmed.1005086.g004

The illustration of the HC switching behaviour of rapid switchers with complex multidirectional flows is presented in Fig 4B. The HC formulations with the highest number of rapid switch events were short-acting HCs with the highest usage volume, such as GSD/EE, DRSP/EE, norelgestromin and ethinylestradiol (NGMN/EE) patch, DSG, and CHC ring. Each of these accounted for 14.6%–20.4% of rapid switchers who had at least one rapid switch from that formulation, and around 9% of users of given HC, except 17.3% of NGMN/EE users, and around 14% of DSG, and CHC ring users. The subsequent HC formulations for these switchers were similar and overlapping, where GSD/EE rapid switchers mostly switched to DRSP/EE (25.2%), dienogest/ethinylestradiol (DNG/EE, 14.7%), and NGMN/EE patch (11.0%), DRSP/EE users switched to GSD/EE (23.3%), DNG/EE (16.4%), and CHC ring (13.2%), and NGMN/EE users switched to GSD/EE (23.7%), CHC ring (20.6%) and DRSP/EE (19.4%). Most notably, switching to DNG/EE, LNG-IUD, and ENG implant were more common than switching from them. Among rapid switchers who switched from COCs (n_users = 7,800 (60.3%)), 66.5% switched to a different COC, 11.3% switched to a CHC patch, and 11.1% to the CHC ring. Those who switched to the ring used the new method for approximately 4.9 months (median value) compared with 3.7 months for a patch.

In the period between the first purchase date of the pre-switch HC formulation and the first purchase date of the new formulation, no ICD-10 diagnosis codes were identified for 50.2% (n_users = 6,486) of rapid switchers, while 23.1% (n_users = 2,982) had diagnoses from our curated list of potential side effects (see Methods). Notably, we observed diagnoses related to reproductive system disorders (e.g., excessive/frequent menstruation (n_users = 556)), neurological/pain conditions (migraine and other headaches (n_users = 233), acne (n_users = 180), menstrual pain (n_users = 147)), mood disorders (depressive episode (n_users = 140), anxiety (n_users = 93)), and mastodynia (n_users = 28) (Fig 4C and Table F in S3 Appendix). In addition to this, HC switching was previously associated with contraceptive failure and subsequent abortion [63]. In our cohort, medical abortion (ICD-10 O04) overlapped with inferred HC usage periods in 3.2% (n_users = 2,359, n_periods = 2,440) of all users, with COCs being the most used contraceptive before abortion. The majority of these women (77.3% (n_users = 1,823)) used HCs again (median time to the new HC formulation = 684 days/1.9 years (11–6,132 days)), where other COCs (n_periods = 792 (32.5%)), LNG-IUD (n_periods = 125 (5.1%)) and POPs (n_periods = 122 (5.0%)) were the most common choices following abortion. Similarly, in 2.2% (n_rapid = 288, n_periods = 291) of rapid switchers, abortion diagnosis was recorded between purchases of two HC formulations. The most common switches were from one short-acting method to another of comparable effectiveness and safety, such as from one COC to a different COC (n_periods = 79 (27.1%)), from a CHC patch to a COC (n_periods = 43 (14.7%)), and from a COC to the CHC ring (n_periods = 28 (9.6%)).

Presence of thromboembolism risk factors

Among 13,610 HC users with purchases in 2022, which is the most recent year with complete data, 65.0% (n_users = 8,840) were CHC users (mean age at purchase (SD) = 33.9 (8.1) years; median = 33, IQR [27,40]) and 35.0% (n_users = 4,770) POC users (mean age at purchase = 37.0 (7.8) years; median = 37, IQR [31,43]), with 590 individuals using both types. Individuals who used different HCs in 2022 were classified based on their first HC. Thromboembolism was diagnosed in 43 women (VTE/ATE diagnosis recorded after the first HC purchase date in 2022; mean age at diagnosis = 40.0 (7.5) years; median = 40, IQR [35,45]), of whom 38 were diagnosed during the inferred HC usage period. Next, to characterise the thromboembolism risk profile of POC and CHC user groups, we calculated the prevalence ratio (PR) for 15 different risk factors (Fig 5 and Table G in S3 Appendix, with POC users as the reference category). The most common risk factors were age above 35 (55.1% of POC users and 40.6% of CHC users; PR = 0.74, 95% CI [0.71, 0.77]; p_Bonferroni < 0.001), purchase of risk-increasing medications (14.8% of POC and 17.2% of CHC; PR = 1.16, 95% CI [1.07, 1.26]; p_Bonferroni = 0.006), obesity (14.5% of POC and 10.0% of CHC; PR = 0.69, 95% CI [0.63, 0.76]; p_Bonferroni < 0.001), and migraine (8.2% of POC and 5.6% of CHC; PR = 0.68, 95% CI [0.60, 0.78]; p_Bonferroni < 0.001). The prevalence for carrier status of thrombosis risk-increasing genetic variants was slightly higher among POC users (5.7% of POC versus 4.7% of CHC; PR = 0.82, 95% CI [0.71, 0.96]; p_Bonferroni = 0.167), while the prevalence of personal history of VTE/ATE was twice as high (3.4% of POC and 1.7% of CHC; PR = 0.50, 95% CI [0.40, 0.62]; p_Bonferroni < 0.001). The prevalence of family history of thrombosis, other thrombosis risk conditions, hyperlipidaemia, diabetes, and pneumonia did not differ significantly between POC and CHC users after Bonferroni correction (p ≥ 0.003). The same analysis, stratified by estrogen dose (low ethinylestradiol, moderate ethinylestradiol, and estradiol CHC user groups), showed largely similar patterns across low and moderate ethinylestradiol groups (Table H in S3 Appendix and S6 Appendix). The estradiol group showed slightly higher prevalence ratios across several risk factors, though estimates should be interpreted cautiously given the small sample size and wide confidence intervals. Out of 13,610 analysed HC users, 61.7% (n_users = 8,397) had at least one first-degree relative in the EstBB, and 7.5% (n_users = 1,022) had a first-degree relative diagnosed with VTE or ATE before HC purchase. Women using HCs who had at least one first-degree relative diagnosed with thromboembolism had 1.7-fold higher odds of being a FVL or PTM carrier (76/1,022 (7.4%) versus 327/7,375 (4.4%); OR = 1.73, 95% CI [1.32, 2.25]; p < 0.001). The sensitivity and specificity of family history of thrombosis for identifying carrier status were 18.9% (95% CI [15.0, 22.7]) and 88.2% (95% CI [87.5, 88.9]), respectively.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Prevalence ratios of venous and arterial thromboembolism (VTE and ATE) risk factors in progestin-only hormonal contraceptive (POC) vs. combined hormonal contraceptive (CHC) users with purchases in 2022.

On the y-axis are risk factors and within-group percentages, while on the x-axis is the prevalence ratio (PR) with 95% confidence interval (CI). Abbreviations: FVL, Factor V Leiden; PTM, Prothrombin G20210A Mutation.

https://doi.org/10.1371/journal.pmed.1005086.g005

Across the entire study group of all HC users (n_users = 73,071), we found that 5.3% (n_users = 3,886) carried at least one VTE high-risk genetic variant (3.5% (n_users = 2,556) only FVL, 1.8% (n_users = 1,276) only PTM, 0.07% (n_users = 54) both FVL and PTM). While in 1.5% (n_users = 1,060) of users, thrombosis diagnosis occurred during the inferred HC usage period, of those, only 17.6% purchased antithrombotic agents (ATC group B01A) after the diagnosis and while HCs were in active use according to purchasing data. The top five ICD-10 thrombosis codes were I80 (unspecified phlebitis and thrombophlebitis, n_users = 330), I80.2 (deep vein thrombosis, n_users = 212), I26.9 (pulmonary embolism, n_users = 130), I80.3 (embolism or thrombosis of lower extremity, n_users = 125) and I82.8 (embolism and thrombosis of vena cava, n_users = 78). Among the 1,060 cases during HC usage, 8.6% carried either the FVL or PTM variant. Overall, we observed a significantly higher percentage of thrombosis cases at any time point in carriers using HC (6.5%) compared to the non-carriers using HC (4.2%; OR = 1.61, 95% CI [1.40, 1.84], p < 0.001).