Cross-sectional and longitudinal metabolomics-based profiles associated with oral contraceptive and progestin-only pill use: A Finnish population-based study.

Elena Toffol, Jari Haukka, and Oskari Heikinheimo contributed to the study conception and design; Elena Toffol performed statistical analyses and drafted the first version of the manuscript. All authors contributed to the interpretation of results and revised the manuscript for important intellectual content. All authors approved the final version to be published.

The Health 2000 and Health 2011 surveys were approved by the Coordinating Ethics Committee of the Helsinki and Uusimaa Hospital District (HUS Regional Committee on Medical Research Ethics) on May 31, 2000 and June 17, 2011, respectively (reference 45/13/03/00/11). All participants gave written informed consent. The current study has been approved by THL Biobank on November 3, 2020 (study number THLBB2020_21).

This work was supported by the Finnish Cultural Foundation (E.T., grant number #00211101 and grant number #00230159).

Table  1 reports the baseline characteristics of the 1721 participants with available information on their current use of OC and metabolomics data. A total of 299 individuals (17.4%) were using an OC at the time of the surveys, the most common being third generation COCs (37.1%), followed by OCs containing cyproterone and estrogen (24.1%). OC users were younger (mean (SD), 34.8 (6.3) vs 39.5 (6.1) years), more likely to use alcohol (95.6% vs 86.9%) and had a lower BMI (mean (SD), 24.8 (4.6) vs 25.4 (5.1)) than those not using HC. Background characteristics of the study population comparing oral contraceptive OC users vs non‐users of hormonal contraception (HC). Pooled Health 2000 and Health 2011 data, n  = 1721 with metabolomics data. Abbreviations: BMI, body mass index; COC, combined oral contraceptive; HC, hormonal contraception; OC, oral contraceptive; POP, progestin‐only pill. Norgestimate + Ethinylestradiol, Drospirenone + Ethinylestradiol, Dienogest + Estradiol. Range: 30–49 years in Health 2000; 18–49 in Health 2011. Insulin, oral hypoglycemics, antithrombotic therapy, lipid medications, combination products of estrogens and progestins, estrogens, progestogenes, systemic corticosteroids, systemic antibiotics, systemic antimycotics, cytostatics, anti‐inflammatory analgesics, opioids, other analgesics (no NSAID group), migraine medications, epilepsy medications, antipsychotics, anxiolytics, hypnotics and sedatives, antidepressants, psycholeptics and psychoanaleptics in combination, long and short acting beta agonists, combination product of beta agonist and corticosteroids, inhaled corticosteroids, anticholinergics. Heart diseases, hypertension, venous thrombosis, diabetes, psychological or mental illnesses, rheumatoid arthritis, osteoarthritis, cancer. In analyses adjusted for age, BMI, duration of use, study cohort, disease, and medication use (Model 2), use of COCs was significantly associated with a number of altered metabolic measures compared to non‐use of HC (second generation COCs: 51 of the 212 metabolic measures, median difference in biomarker concentration: 0.46 SD; third generation COCs: 110 out of 212 metabolic measures, median difference in biomarker concentration: 0.43 SD; other COCs: 76 out of 212, median difference in biomarker concentration: 0.56 SD; cyproterone and estrogen: 59 of the 212 metabolic measures, median difference in biomarker concentration: 0.38 SD) (Table  S5 ). On the contrary, the metabolic profiles of POP users did not significantly differ from those of non‐users of HC (26 out of 212 metabolites, mostly in the opposite directions than the other contraceptive types, median difference − 0.34 SD) (Figure  1 ). The results did not substantially differ from those of the unadjusted model (Model 1) or after adjustment for lifestyle habits (Model 3) (Figures  S3 and S4 ). In particular, use of all different groups of COCs was significantly associated with higher concentrations and ratios of monounsaturated fatty acids (MUFAs) but lower ratios of polyunsaturated fatty acids (PUFAs), and with higher concentrations and ratios of triglycerides in lipoproteins. In addition, users of third generation COCs and, to a lesser extent, other COCs and cyproterone and estrogen, had higher levels of inflammation markers and higher concentrations of HDL and lipids in HDL lipoproteins, but lower percentage of cholesterol and higher percentage of triglycerides in non‐HDL lipoproteins (Figure  1 , Figures  S3 and S4 ). Associations between use of different types of oral contraceptives (OCs) and 212 metabolic measures, Model 2; reference category: Non‐users of hormonal contraception (HC). Analyses conducted in pooled Health 2000 and Health 2011 surveys. Model 2 is adjusted by age, body mass index, duration of use, study cohort, disease, and medication use. Results are in SD units of difference in metabolite concentrations; bars indicate 95% CI. Closed circles indicate significant associations at p ‐value adjusted for False Discovery Rate. The reference category corresponds to the straight line set at 0.0 SD. The metabolic profiles of COC users were similar when the reference group was set to users of LNG‐IUD rather than non‐users of HC (Table  2 ; Figure  2 , Figures  S5 and S6 ). The associations were characterized by a slightly larger effect size, with median differences in biomarker concentration in Model 2 of 0.62 SD (second generation COCs), 0.50 SD (third generation COCs), 0.64 SD (other COCs), and 0.46 SD (cyproterone and estrogen) (Table  S5 ). The metabolic profile of POP users did not substantially differ from that of the reference group (i.e., LNG‐IUD users). Associations between use of different types of oral contraceptives (OCs) and 212 metabolic measures, Model 2; reference category: Users of levonorgestrel intrauterine device (LNG‐IUD). Analyses conducted in pooled Health 2000 and Health 2011 surveys. Model 2 is adjusted by age, body mass index, duration of use, study cohort, disease, and medication use. Results are in SD units of difference in metabolite concentrations; bars indicate 95% CI. Closed circles indicate significant associations at p ‐value adjusted for False Discovery Rate. The reference category corresponds to the straight line set at 0.0 SD. Background characteristics of the study population comparing oral contraceptive (OC) vs levonorgestrel intrauterine device (LNG‐IUD) users. Pooled Health 2000 and Health 2011 data. Population with metabolomics data. Abbreviations: COC, combined oral contraceptive; LNG‐IUD, levonorgestrel intrauterine device; OC, oral contraceptive; POP, progestin‐only pill. Norgestimate + Ethinylestradiol, Drospirenone + Ethinylestradiol, Dienogest + Estradiol. Range: 30–49 years in Health 2000; 18–49 in Health 2011. Insulin, oral hypoglycemics, antithrombotic therapy, lipid medications, combination products of estrogens and progestins, estrogens, progestogenes, systemic corticosteroids, systemic antibiotics, systemic antimycotics, cytostatics, anti‐inflammatory analgesics, opioids, other analgesics (no NSAID group), migraine medications, epilepsy medications, antipsychotics, anxiolytics, hypnotics and sedatives, antidepressants, psycholeptics and psychoanaleptics in combination, long and short acting beta agonists, combination product of beta agonist and corticosteroids, inhaled corticosteroids, anticholinergics. Heart diseases, hypertension, venous thrombosis, diabetes, psychological or mental illnesses, rheumatoid arthritis, osteoarthritis, cancer. The patterns of associations varied in different age groups (Figures  S7– , S9 ). While associations in the older (40–49 years, OC users n  = 62, HC non‐users n  = 744) and younger (18–29 years, OC users n  = 32, HC non‐users n  = 44) age groups did not generally reach statistical significance, most of the associations were driven by the middle age group (30–39 years, OC users n  = 184, HC non‐users n  = 634). Among middle‐aged women, use of all types of preparations, except for POPs, was associated with metabolic alterations comparable to those identified in the entire sample; in this age group, the metabolic alterations were especially evident for those using third generation COCs (126 out of 212 metabolic measures, median difference in biomarker concentration: 0.44 SD, range −0.80 SD to 1.08 SD) (Table  S5 ). Among younger and older women, most of the associations were related to the use of other COCs. In the longitudinal analyses of the 11‐year change in metabolic profiles of starters of LNG‐IUD ( n  = 59), continuers of LNG‐IUD ( n  = 33), stoppers of second or third generation COC ( n  = 19) and switchers from second or third generation COC to LNG‐IUD ( n  = 24) (Table  S4 ), no significant differences in metabolic changes were found among women who continued, nor among those who started to use LNG‐IUD, compared to never‐users of HC ( n  = 142). On the other hand, significant changes in metabolic profiles of switchers (49 out of 212 metabolic measures) and stoppers of second and third generation COC (39 metabolic measures) were found. In detail, during the 11‐year follow‐up time, women who switched from second or third generation COC to LNG‐IUD had an increase in the percentage of unsaturation of fatty acids (PUFA: omega 6, specifically LA) and a decrease of saturated fatty acids and MUFA, along with a decrease of total and lipoprotein triglycerides, lipids, and phospholipids, and of particle concentration in lipoproteins. In addition, women who stopped using second or third generation COC had a decrease of cholesterol, total lipids, and phospholipids in HDL lipoproteins (Figure  3 ). Eleven‐year changes in molecular concentrations of 212 metabolic measures by starting (LNG‐IUD), continuing (LNG‐IUD), switching (from COC to LNG‐IUD) and stopping (COC) use of HC (II‐ and III generation COC and LNG‐IUD); reference category: Never‐users of hormonal contraception. Analyses are adjusted by age at baseline and 11‐year change in BMI. Results are in SD units of change in metabolite concentrations; bars indicate 95% CI. Closed circles indicate significant associations at p ‐value adjusted for False Discovery Rate. The reference category corresponds to the straight line set at 0.0 SD. COCs, combined oral contraceptives; HC, hormonal contraception; LNG‐IUD, levonorgestrel intrauterine device.

The main finding of this study is that the use of OCs, and particularly the use of COCs, is associated with a large number of metabolic alterations in comparison to both non‐users of any HC and to users of LNG‐IUD, with the magnitude of the associations ranging between −1.2 and +1.3 SD. The detected associations were indicative of an unfavorable cardiometabolic profile, with higher levels of saturated fatty acid concentrations and ratios, triglycerides, and, to some extent, cholesterol in lipoproteins, but lower levels of amino acids. Conversely, the use of POPs was not associated with substantial metabolic differences compared to non‐use of HC or use of the LNG‐IUD. An additional finding is that in longitudinal analyses, while the continuation or starting of use of LNG‐IUD was not related to 11‐year changes in metabolic profiles (as in never‐users of HC), women who changed from COC to LNG‐IUD or stopped using second or third generation COC had a change towards more favorable profiles, with greater unsaturation levels and lesser total and lipoprotein triglycerides and other lipids. These results are in line with the notion of metabolic alterations and adverse events in relation to the use of COCs, while supporting the metabolic safety of progestin‐only contraception irrespective of the route of administration, and align with those of a previous metabolomics study of HC. 13 Interestingly, in our study, the alterations were mostly evident for third‐generation COCs and other COCs, including anti‐androgenic progestins drospirenone and dienogest, combined with EE and estradiol, respectively. Similarly, results of a metabolomics study on two cohorts of OC users vs non‐users of HC found that OC use (mostly COCs) is associated with higher levels of triglycerides and phosphatidylcholines, but lower lysophosphatidylcholines and amino acids such as glutamine, glycine, and tyrosine. 15 Lower levels of amino acids (specifically, alanine, glutamine, glycine, tyrosine, and proline) in relation to COC use are a common finding in metabolomics studies of different COC types, 12 , 16 , 25 possibly related to increased oxidative stress status and protein turnover. 16 , 25 The effects of COCs on lipid profile are likewise well‐known, although they appear to vary with the type of estrogens and progestins. In our study, COC use was associated with increased levels of total HDL cholesterol, particularly those of HDL3, which in turn is related to an increased risk of coronary heart disease. 26 Although Ruoppolo et al. 12 showed that women on third generation OCs had higher total and HDL cholesterol and triglycerides than non‐users, newer molecules are considered to have a safer lipid profile than second generation preparations (containing, e.g., LNG). For example, a previous randomized study of more than 100 women found no significant changes in total, HDL, and LDL cholesterol or total triglycerides after 6 months of nomegestrol acetate + estradiol, while 6 months of LNG + EE led to decreased HDL and increased LDL cholesterol and total triglycerides. 27 In line with Morin‐Papunen et al., 6 we found that third generation and other COCs in particular were associated with increased levels of HDL cholesterol, HDL triglycerides, triglycerides in LDL lipoproteins, as well as phosphatidylcholines, phosphoglycerides, and sphyngomyelines. As, contrary to natural estrogens, EE typically increases both HDL and other lipoprotein levels, 5 this finding could be attributable to the effects of EE contained in all the third generation preparations and most of the other COC preparations. In addition to changes in lipoprotein composition and concentrations, we observed a general increase in absolute levels of all fatty acids, along with higher proportions of MUFAs and saturated fatty acids, but lower proportions of PUFAs (e.g., omega 6 and linoleic acid) in users of all types of COCs, especially third generation and other COCs. Higher saturation levels in users of hormonal contraception have been reported previously. 13 , 28 Although MUFA‐rich diets are considered metabolically healthy, 29 the different effects of different MUFAs and PUFAs in terms of cardiovascular and metabolic risk appear more complex. 30 Previous studies reported cardiovascular disease risk positively associated with the percentages of MUFA but negatively with percentages of PUFA and PUFA/MUFA ratio, 31 further confirming the metabolically unfavorable profile of COC users. The interpretation of these findings is, however, limited by the lack of information, in our data, on nutritional intake. As estrogens modulate the bioavailability of PUFA absorption, our findings may in fact reflect lifestyle and eating habits, including fatty acid supplement use. Reciprocally, omega‐3 fatty acids are tightly linked to ovulatory processes and risk of infertility, as they contribute to the regulation of estrogen and progesterone production. 32 Further research is needed to clarify the link between dietary intake, fatty acids, gonadal hormones, and hormonal contraceptives. Moreover, use of third generation COCs was also associated with increased levels of glycoprotein acetyls, an inflammation marker predictor of diabetes, cardiovascular diseases, and all‐cause mortality. 33 The finding of a pro‐inflammatory status especially in users of third generation COCs confirms previous findings of high levels of C‐reactive protein, an independent marker of cardiovascular risk. 6 , 34 Our findings add to previous knowledge, indicating that metabolic alterations may be evident also irrespective of, or before, clinical events, and irrespective of the duration of use. In fact, we controlled our results for factors such as chronic or severe diseases, including a history of cardiovascular events and venous thrombosis, and for the duration of use. This suggests that, although clinical events such as venous thrombosis are known to occur in the first year after starting HC, metabolic alterations may occur and persist irrespective of the duration of use. Additionally, our controlling variable was set to include all previous periods of use, thus also for women currently not using HC. In this way, we were able to take a possible healthy user bias into account. To the best of our knowledge, the only previous study to longitudinally examine the metabolomic‐based effects of OCs found that metabolic changes of current use were confirmed in continuers and starters of COCs, but they normalized in those who stopped using COCs. 13 The lack of changes in LNG‐IUD starters in our study confirms the marginal effects of the LNG‐IUD on the metabolic profile and, if any, mostly in a protective direction. 10 , 11 This is further supported by the observation that, when changing the reference category to use of LNG‐IUD, the metabolic alterations of COCs did not change but became slightly larger in magnitude. In addition, the similar metabolic profiles in POP users compared to both non‐users and LNG‐IUD users confirm the metabolic safety of progestin‐only preparations, irrespective of their route of administration. On the other hand, the finding of favorable metabolic changes in women who switched from second or third generation COCs to LNG‐IUD or stopped using COCs is an encouraging one, as it supports the notion that metabolic alterations of COC are reversible to a never use, or even to a more favorable profile after interruption of use. Thus, the main practical conclusion from this longitudinal part of the study is that discontinuation of COCs is likely associated with favorable changes in metabolic parameters commonly related to cardiometabolic risk; on the other hand, discontinuation or initiation of LNG‐IUD use is not substantially associated with such metabolic alterations. This is clinically reassuring as women typically use several different contraceptive methods over their fertile years. This study has a number of limitations. First, the sample size was especially small in some of the subgroups and in longitudinal analyses; thus, we cannot rule out that null findings in certain groups (e.g., POP users, as well as users in younger and older age groups) may be due to a lack of power. The small sample size also precluded any analyses on preparation subgroups to account for different potencies of different forms of estrogens/progestins; similarly, longitudinal analyses could not be conducted for the group of COC users. Additionally, the younger group was under‐represented, as metabolomics data for women of less than 30 years of age were available only from the Health 2011 survey. Further, we lacked information on the exact composition of the category of cyproterone and estrogen, which may have contained both EE‐ and estradiol‐based combinations. Additionally, as only information on cumulative duration of use of HC was available, we could not account for parameter changes in relation to the time of use of different contraceptives. Because of the study design (a population‐based study rather than a clinical trial) we did not have complete information on conditions possibly affecting the indication, type, and use of contraception, as well as the individual metabolomic profile, such as endometriosis, premenstrual syndrome, and polycystic ovary syndrome. However, during the process of population selection, all women using hormone replacement therapy for menopause, menstrual problems, or other reasons were excluded from the study, and the results were adjusted for covariates covering a large set of diseases and lifestyle habits. On this regard, the current medication and disease status variables were binominal variables, each of them encompassing a rather wide range of drugs or conditions, with very different potential effects on the metabolic profile. However, because this is an untargeted metabolomics study aimed at differentiating broad metabolic profiles of users of different contraceptives, the clinical interpretation of the effects of single medications or disease conditions, as well as of absolute concentrations of metabolites, is out of the scope of the study. The long interval between the baseline and follow‐up assessments, with no information on HC use and on any intermediate events (such as comorbidities, medications and lifestyle habits) between these points, further limits the interpretation of the longitudinal results. However, because the results of the analyses in the cross‐sectional part of the study did not substantially vary across different levels of adjustments, it is plausible that inclusion of covariates would have not changed the results of longitudinal analyses. We cannot exclude that some of the identified longitudinal changes were in fact due to natural aging. However, all our results were controlled for baseline age. In addition, changes related to age are expected to equally influence all contraceptive user groups. Finally, because of the likely high correlations between several metabolites, interpretation of the implications and importance of single metabolites may have been biased. Strengths of the study include the large number of available metabolomics measures and the extensive information on possible confounders. Although HC use was self‐reported, potentially introducing a recall bias, a self‐report of contraception use has been shown reliable in specifically focused studies. Furthermore, the availability of follow‐up data for a subgroup of women allowed a longitudinal examination of the associations.

Results of our study confirm previous observations on various metabolic alterations related to the use of COCs, especially third generation and other COCs, while reinforcing the notion of metabolic safety of POPs and LNG‐IUD. These associations appeared to be mostly reversible after interruption of use or switch to different preparations.

Use of hormonal contraception (HC) is a safe and effective option for birth control for a large proportion of fertile‐aged women. 1 Although an increasing number of women opt nowadays for long‐acting reversible contraception, such as hormonal intrauterine devices, oral contraceptives (OC) are still among the most chosen options, 1 given their ease of use. In the past decades, there has been a continuous development of new molecules and their combined preparations, aimed at improving OC safety and side‐effect profile. This has resulted in an important reduction of adverse events related to OC use. Still, the use of OC is known to lead to various metabolic changes, and combined oral preparations, especially those containing high doses of synthetic estrogens and third and fourth generation progestins, carry a risk for deep venous thrombosis and other cardiovascular accidents. 2 , 3 According to a recent meta‐analysis, the use of combined OCs (COC) containing different progestins in combination with ethinylestradiol (EE) is associated with increased levels of high‐density lipoprotein (HDL) cholesterol, triglycerides, and low‐density lipoprotein (LDL) cholesterol. 4 While the estrogenic component of COC, EE in particular, increases the levels of lipoproteins, the progestin component modulates the metabolic effects of EE and estrogens. 5 In addition, progestins can exert androgenic or glucocorticoid effects, possibly leading to weight gain, insulin resistance, and hyperglycemia, increased levels of plasma triglycerides and LDL cholesterol, and decreased levels of HDL cholesterol. Thus, nonandrogenic or anti‐androgenic progestins may have a safer cardiometabolic profile, with minimal influence on the lipid profile and carbohydrate metabolism. 5 On the contrary, progestin‐only contraceptives, including progestin‐only pills (POP), progestin‐releasing implants, and intrauterine systems, appear to have a safer metabolic profile. 6 , 7 , 8 , 9 , 10 , 11 To date, only a few studies have extensively examined the metabolic profiles associated with the use of different types of OCs using the metabolomics technique. 12 , 13 , 14 , 15 , 16 , 17 These studies have consistently found alterations in lipid and amino acid profiles related to the use of OCs. However, a detailed examination of how different combinations and generations of OCs impact metabolism, and how their impact persists in the long term, has not been conducted in detail. Thus, the aim of the current study is to comparatively examine cross‐sectional metabolic profiles associated with the use of different types of COCs and POPs separately, and longitudinal changes related to the interruption of use, or change to different preparations. In addition, this study aims to explore how different preparations perform relative to the use of a metabolically safer contraceptive option, such as the levonorgestrel‐releasing intrauterine device (LNG‐IUD), in a large population‐based study carried out in Finland in 2000 and its 11‐year follow‐up study.

Oskari Heikinheimo serves occasionally on advisory boards for Bayer AG and Gedeon Richter, and has designed and lectured at educational events of these companies. The rest of the authors have nothing to declare.

Material for this study is based on two population‐based surveys carried out in Finland in 2000–2001 (Health 2000 study) and 2011–2012 (Health 2011 study). 18 , 19 Data were collected through interviews, questionnaires, and an extensive health examination inclusive of blood sampling carried out across 80 regions in Finland. The Health 2000 survey covered more than 8000 people aged 30 years and over, who were thereafter invited to participate in the Health 2011 follow‐up study, along with an additional sample of nearly 2000 young adults (18–29 years). The population for this study was selected to include all women, aged 18–49 years, who had self‐reported on their current and past use of HC, and had metabolomics data obtained from serum samples drawn in connection with the health examination. In each survey the following exclusion criteria were applied: current pregnancy, menopause (“Did your periods end?”, with answers “naturally with the menopause” or “because of an operation or radiotherapy”), current use of menopausal hormonal therapy (“During the past month have you used hormone replacement therapy as tablets, gel or patches because of menopause, menstrual problems or some other reason?”) and hysterectomy. To compare OC users vs non‐users of HC, additional exclusion criteria were current use of LNG‐IUD (or of vaginal ring or transdermal patch, information available only in Health 2011) and missing Anatomical Therapeutic Chemical (ATC) code for the used OC, resulting in a final population of 1721 individuals with available metabolomics data (299 currently using an OC, and 1422 HC non‐users). In secondary analyses, only current users of OC or LNG‐IUD users were retained, resulting in a final population of 640 individuals with available metabolomics data (299 currently using an OC, and 341 users of LNG‐IUD). Detailed sample size, background information, and selection procedure for each survey are reported in Tables  S1 and S2 and Figure  S1 . Of the population obtained after applying the initial general exclusion criteria, 327 women had complete information of HC use and metabolomics data in both surveys (Health 2000 and Health 2011) and thus contributed to the sample used for the longitudinal part of the study. Information on the current and previous use of HC (pills and LNG‐IUD) and its duration (“For many years altogether have you been taking ‐contraceptive pill/a hormonal intrauterine device [all usage periods included]?”) was obtained through questions during the health interview. Respondents were additionally asked to report the name of the specific contraceptive preparation they were currently using (Health 2000) and/or the name of any prescription medicine they were using at the moment of the interview (Health 2000 and Health 2011). Based on the ATC codes obtained through this information, the following categories of OCs were defined: second generation COCs (i.e., those containing ethinylestradiol and levonorgestrel [ATC: G03AA07, G03AB03]), third generation COCs (containing ethinylestradiol, and desogestrel or gestone [ATC: G03AA09, G03AA10, G03AB05, G03AB06]), other COCs (i.e., those containing ethinylestradiol and norgestimate or drospirenone, or estradiol and dienogest [ATC: G03AA11, G03AA12, G03AB08], POPs [ATC: G03AC01, G03AC02, G03AC03, G03AC09]), and cyproterone and estrogen pills (ATC G03HB01) (Table  S3 ). The 52 mg LNG‐IUD was the only hormonal intrauterine device commercially available in Finland until 2013, with an approved maximum duration of 5 years of use at the time. Covariates obtained from the health examination, self‐administered questionnaires, and clinical measurements included the following: age; alcohol use (yes, no/quit); frequency of smoking (every day, sometimes, never); physical activity (regular physical activity vs no regular physical activity); current use (during the past 7 days) of prescription drugs; chronic/severe diseases (heart diseases, hypertension, venous thrombosis, diabetes, psychological or mental illnesses, rheumatoid arthritis, osteoarthritis, cancer); and body mass index (BMI) based on the measured height and weight. To take into account any previous use of HC among both current users and non‐users, and a possible healthy user bias, for all respondents, an additional categorical variable was created to describe the cumulative duration of use of contraception (none, up to 1 year, two to 5 years, and more than 5 years). In each survey, the metabolomics measures were obtained from blood samples drawn during the health examination, after at least 4‐h fasting. The serum and plasma samples were centrifuged at 1600–1800 G for 10 min and immediately frozen to −20°C on site, normally within 45–60 min but no later than 90 min from sampling; thereafter, they were transferred to their final storage location (−70°C), no later than 1–2 weeks after sampling. Serum sections were analyzed with a high‐throughput serum nuclear magnetic resonance (NMR) metabolomics platform ( 1 H NMR Spectroscopy, Nightingale Health 2018–2019). Biomarkers were quantified independently for each serum sample. Nightingale's biomarker analysis technology applies a single experimental setup, and spectral information is converted to absolute concentrations (in molar units) of the metabolic measures. The platform allows for the simultaneous quantification of almost 250 metabolic biomarkers per sample, including 12 lipid measures of 14 lipoprotein subclasses (6 very‐low‐density lipoproteins VLDLs, 4 high‐density lipoproteins HDLs, 3 low‐density lipoproteins LDLs, intermediate‐density lipoprotein IDL), other detailed molecular information on serum lipids (e.g., sphingomyelin, fatty acids, etc.), or low molecular weight metabolites (e.g., amino acids). An additional set of metabolite ratios is also computed. 20 , 21 The study consists of a cross‐sectional and a longitudinal part. After preliminary inspections of the metabolomics data in each cohort, 16 metabolites with more than 100 missing observations were excluded from the analyses, resulting in a total of 212 metabolic measures; metabolic measures with value “zero” were replaced with 0.25× the minimum observed value for that metabolite, and, after log transformation, remaining missing metabolomics data were imputed through random forest imputation. 22 The overall missing and imputed metabolomics data represented 12% of all observations, with proportions of missing data observed within each individual case ranging between 2.4% ( n  = 5 metabolites, 28 cases) and 21.7% ( n  = 46 metabolites, 1 case). Specifically, the percentages of imputed missing metabolomics data were 0.12% among HC non‐users, 0.23% among users of second generation COCs, 0.14% in users of third generation COCs, and 0.28% in users of cyproterone and estrogen pills. Given the 11‐year time‐interval between the two surveys, in the cross‐sectional part of the study the two datasets were pooled together on common variables. Cross‐sectional analyses of associations were carried out via generalized estimating equations method with exchangeable correlation structure (with “gee” R‐package), with each metabolic measure as the outcome variable, and use of each separate group of OCs (vs current non‐use of any HC) as the predictor of interest. Three models were fitted: Model 1, unadjusted; Model 2, controlled for age, BMI, duration of use, study cohort, diseases, and medication use (which corresponds to the minimal sufficient adjustment, please see the Direct Acyclic Graph in Figure  S2 ); Model 3, which is Model 2 further adjusted for alcohol use, smoking, and physical activity. Model 2 was additionally repeated in age‐stratified groups (18–29 years, 30–39 years, 40–49 years). In secondary analyses, the three above‐described models were conducted in the population inclusive of all current OC vs LNG‐IUD users. In these analyses, the metabolic profiles of each separate group of OC users were compared to those of current users of LNG‐IUD (reference category). To allow the comparison across multiple measures, association magnitudes are reported in SD units of difference in biomarker concentration compared to the reference group. In the longitudinal part of the study, of the 327 women with complete data at both time points, users of POPs, other COCs, and cyproterone and estrogen were excluded due to small group sizes. The remaining participants ( n  = 277) were divided into five groups: (1) continuers, that is, women who were using LNG‐IUD at both time points ( n  = 5 COC continuers were excluded due to small number); (2) stoppers, that is, women who used second or third generation COC in 2000, but no HC in 2011; (3) starters, that is, women who used no HC in 2000, but used LNG‐IUD in 2011; (4) switchers, that is, women who changed between second or third generation COC to LNG‐IUD between 2000 and 2011; and (5) never‐users, who did not use any HC neither in 2000 nor in 2011 (Table  S4 ). Changes in levels of each metabolite were computed as the difference between levels in 2011 and in 2000. Linear regression models were further performed using the “ggforestplot” R‐package 23 to compare changes in each metabolite levels in the four groups of users of HC, never‐users of HC being the reference category. Analyses were controlled for age at baseline and change in BMI. To take the multiple testing into account, the false discovery ratio procedure was applied. All the analyses were performed with R software version 4.2.3. 24

Figure S1. Figure S2. Figure S3. Figure S4. Figure S5. Figure S6. Figure S7. Figure S8. Figure S9. Table S1. Table S2. Table S3. Table S4. Table S5. Legends