Epidemiology and Psychiatric Population-based cohort study of oral

Sciences
contraceptive use and risk of depression
cambridge.org/eps
T. Johansson1 , 2 , S. Vinther Larsen3 , 4 , M. Bui5 , W. E. Ek1 , T. Karlsson1 and
Å. Johansson1
1
Department of Immunology, Genetics and Pathology, Science for Life Laboratory, Uppsala University, Uppsala,
Original Article 2
Sweden; Centre for Women’s Mental Health during the Reproductive Lifespan – Womher, Uppsala University,
3
Cite this article: Johansson T, Vinther Uppsala, Sweden; Neurobiology Research Unit, Copenhagen University Hospital, Rigshospitalet, Denmark;
4 5
Larsen S, Bui M, Ek WE, Karlsson T, Department of Clinical Medicine, University of Copenhagen, Copenhagen, Denmark and Centre for
Johansson Å (2023). Population-based cohort Epidemiology and Biostatistics, Melbourne School of Population and Global Health, University of Melbourne,
study of oral contraceptive use and risk of Melbourne, VIC, Australia
depression. Epidemiology and Psychiatric
Sciences 32, e39, 1–8. https://doi.org/10.1017/
S2045796023000525 Abstract
Aim. Research on the effect of oral contraceptive (OC) use on the risk of depression shows
Received: 16 November 2022
Revised: 1 May 2023
inconsistent findings, especially in adult OC users. One possible reason for this inconsistency
Accepted: 10 May 2023 is the omission of women who discontinue OCs due to adverse mood effects, leading to healthy
user bias. To address this issue, we aim to estimate the risk of depression that is associated with
Keywords: the initiation of OCs as well as the effect of OC use on lifetime risk of depression.
depression; epidemiology; mental health;
Methods. This is a population-based cohort study based on data from 264,557 women from
women
the UK Biobank. Incidence of depression was addressed via interviews, inpatient hospital or
Corresponding author: Therese Johansson; primary care data. The hazard ratio (HR) between OC use and incident depression was esti-
Email: [email protected] mated by multivariable Cox regression with OC use as a time-varying exposure. To validate
causality, we examined familial confounding in 7,354 sibling pairs.
Results. We observed that the first 2 years of OC use were associated with a higher rate of
depression compared to never users (HR = 1.71, 95% confidence interval [CI]: 1.55–1.88).
Although the risk was not as pronounced beyond the first 2 years, ever OC use was still asso-
ciated with an increased lifetime risk of depression (HR = 1.05, 95% CI: 1.01–1.09). Previous
OC use were associated with a higher rate of depression compared to never users, with ado-
lescent OC users driving the increased hazard (HR = 1.18, 95% CI: 1.12–1.25). No significant
association were observed among adult OC users who had previously used OCs (HR = 1.00,
95% CI: 0.95–1.04). Notably, the sibling analysis provided further evidence for a causal effect
of OC use on the risk of depression.
Conclusions. Our findings suggest that the use of OCs, particularly during the first 2 years,
increases the risk of depression. Additionally, OC use during adolescence might increase the
risk of depression later in life. Our results are consistent with a causal relationship between OC
use and depression, as supported by the sibling analysis. This study highlights the importance
of considering the healthy user bias as well as family-level confounding in studies of OC use
and mental health outcomes. Physicians and patients should be aware of this potential risk
when considering OCs, and individualized risk–benefit assessments should be conducted.

Introduction
An estimated 151 million women of reproductive age use oral hormonal contraceptives
(Haakenstad et al., 2022), of which many women benefit in terms of avoiding abortions and
unwanted pregnancies, as well as menstrual bleeding and pain disturbance (David and Kling,
2020). Clinical evidence indicates that hormonal contraception can affect some women’s mood
(Payne, 2003), yet the link between using hormonal contraception and depression remains inad-
equately addressed (Robakis et al., 2019; Schaffir et al., 2016). Several studies have identified
an association between hormonal contraception use during adolescence and an increased risk
© The Author(s), 2023. Published by of depression (Anderl et al., 2022, 2020; de Wit et al., 2020; Skovlund et al., 2016; Zettermark
Cambridge University Press. This is an Open et al., 2018). The effects of hormonal contraceptive use on depression risk in adults are less clear,
Access article, distributed under the terms of with some suggesting either no increased risk (Cheslack-Postava et al., 2015; Duke et al., 2007;
the Creative Commons Attribution licence
(http://creativecommons.org/licenses/by/4.0),
Lundin et al., 2022) or a decreased risk of depression (Keyes et al., 2013; Toffol et al., 2012).
which permits unrestricted re-use, Within the last decade, large-scale population-based studies have explored the association
distribution and reproduction, provided the between hormonal contraceptive use and depression. A Danish study on more than one million
original article is properly cited. women found a higher risk of depression for all types of hormonal contraception across all age
groups, with the largest risk among adolescents (Skovlund et al., 2016). Similarly, a Swedish
study on 800,000 women found a positive association between hormonal contraceptive use and
the use of any type of psychotropic drug but only among adolescents (Zettermark et al., 2018).

https://doi.org/10.1017/S2045796023000525 Published online by Cambridge University Press

 2 Johansson et al.

A Swedish study conducted on 900,000 women discovered that dif- from questionnaires, interviews, physical health measures, biolog-
ferent types of hormonal contraceptives were linked to increased ical samples, and imaging. Participants are also linked to health
use of antidepressants among adolescents, but in adults, this associ- records, including hospital inpatient data, primary care data, can-
ation was only observed for progestin-only compounds. Moreover, cer, and death registry data. In the present study, we included all
the use of combined contraceptives appeared to have a protec- female participants of UKB (N = 264,557).
tive effect and was associated with a lower risk of depression
(Lindberg et al., 2012). Similarly, a more recent study based on
740,000 Swedish women found that combined oral contraceptives Assessment of exposure
(OCs) were associated with a lower risk of depression when current
OC users were compared to nonusers (never and previous users). During the initial assessment visit, information on OC use, includ-
However, when never users were used as the reference group, this ing the age when first initiated and last discontinued, was obtained
association was no longer significant (Lundin et al., 2022). through a touch screen questionnaire. The relevant UKB data fields
In contrast to the observational studies, randomized clinical tri- include 2784 (ever taken OCs), 2794 (age started OCs) and 2804
als have shown little or no effect of hormonal contraception on (age when last used OCs). The majority of women in UKB initiated
mood (de Wit et al., 2021). However, most of these studies did OC during the 1970s/beginning of the 1980s (Fig. 1). During this
not consider the previous use of hormonal contraception. As high- period, the second-generation pills were predominantly used in
lighted in several studies (de Wit et al., 2021; Skovlund et al., 2016; the United Kingdom. At the end of the 1960s, OCs that contained
Zettermark et al., 2018), one limitation in most previous studies is levonorgestrel with dosages ranging from 100 to 150 μg, in com-
the potential influence of a healthy user bias. Mood effects of OCs bination with 20, 30, or 50 μg of ethinyl estradiol, were introduced
can lead to discontinuations and is a contraindication for their use to the market (Dhont, 2010). For women who were still using OCs
(Larsson et al., 1997), which may result in decreased participation (N = 4,766), age of last use was set to the age at assessment. Women
of affected women in subsequent clinical trials and underestima- unsure about OC discontinuation (N = 16,223), were excluded in
tion of effects. Similar underestimation of effects can be seen in the time-dependent analysis
observational studies not considering previous OC use, as exem-
plified in the Netherlands Study of Depression and Anxiety cohort,
which found that current OC use in a between-person analysis Assessment of depression and covariates
was associated with lower risk of depression, while the within- Incident depression was addressed as the first diagnosis of depres-
analysis showed that the time during OC use was associated with an sion (Table S1). This information was obtained from either
increased risk of depression. This discrepancy is likely explained by the verbal interview during the assessment at the clinic or the
a healthy user bias as those negatively affected by OC use had dis- International Classification of Disease code F32 recorded in the
continued and thus represented nonusers in the between-person inpatient hospital or primary care data, as described in more
analysis (Morssinkhof et al., 2021). A study that specifically focused detail in the Supplementary material and on the UKB website
on long-term effects observed that adolescent OC users had a (UK Biobank, 2019). In order to select suitable confounders for
higher prevalence of depression several years after initial exposure, our main analysis, we applied the directed acyclic graph approach
compared to never and adult OC users. These findings indicate that (VanderWeele et al., 2008) (see Figure S1). Information on poten-
adolescence could be a susceptible phase for OC use to heighten the tial confounders was assessed from data obtained during the ini-
lifetime depression risk (Anderl et al., 2020). tial assessment center visit and included year of birth, Townsend
Inconsistent findings in previous studies may be explained by deprivation index (TDI; used as a proxy for socioeconomic sta-
healthy user bias. Research that accounts for this type of bias by, for tus), number of live births, number of stillbirths, polycystic ovary
example, considering the temporality between the first initiation of syndrome (PCOS), age at menarche, age at sexual debut, and fam-
OCs and depression incidence is needed. Using a “new user” design ily history of severe depression (defined as having one or more
approach (Yoshida et al., 2015), as opposed to a “prevalent user” first-degree relatives with depression). Adjustment for potential
approach that includes both current and new users, avoids under- population stratification was achieved by including the first five
estimating the effect of exposure. The present study was therefore genetic principal components, as described in the Supplementary
designed to estimate both the incidence rate of depression associ- methods. For details of each covariate identified in UKB,
ated with the first initiation of OC use but also the lifetime risk that see Table S2.
is associated with OC use.
In this study, we utilize medical information from more than
250,000 UK Biobank (UKB) women. As most studies conducted
on this topic, this one is observational, which limits the ability to Alternative outcome assessment
make inferences about causality. Therefore, we aimed to provide To account for women who may have experienced depression but
supporting evidence of a causal relationship between OC use and did not seek treatment or receive a diagnosis, we conducted a sec-
depression through the examination of familial confounding in ondary outcome analysis on the subset of women who completed
sister pairs (Li et al., 2020). the UKB mental health questionnaire (MHQ: Table S3) (Davis
et al., 2020). The MHQ was designed to assess lifetime depressive
disorder using the Composite International Diagnostic Interview
Methods Short Form. Relevant data fields from the UKB MHQ include
the following: 20446 (ever had prolonged feelings of sadness or
Study population
depression), 20441 (ever had prolonged loss of interest in normal
UKB is a population-based cohort that recruited 500,000 par- activities) and 20433 (age at first episode of depressive symptoms).
ticipants, aged 37–71 years, from across the United Kingdom Participants who answered “yes” to either 20446 or 20441 were
(UK) between 2006 and 2010. The study collected extensive data asked to report their age at the onset of symptoms (20433).

https://doi.org/10.1017/S2045796023000525 Published online by Cambridge University Press

 Epidemiology and Psychiatric Sciences 3

Figure 1. Year of initiating oral contraceptive use. The

histogram shows the number of women who first initiated
oral contraceptive use each year. The numbers are based
on the 205,858 women in the UK Biobank who reported
their age at the time of initiation. The year at initiation has
been calculated using information on year of birth.

Statistical analysis to OC use was coded as “never use” for all women from birth and
changed to “initial use” when a woman started using OCs. After 2
Women were followed from birth until the first occurrence of
years of use, the OC exposure variable was reclassified as “remain-
depression or until the end of follow-up (age at initial UKB assess-
ing years of use”, followed by “recent use” and “previous use”
ment visit), whichever came first. OC use was modelled as a
(see Supplementary material).
time-varying variable using Cox modelling for counting processes
(Therneau and Grambsch, 2000). The reference group comprised
never users, defined as those who never used OCs, and non- Sensitivity analyses
exposed users prior to OC initiation. In the main analyses, we
Sensitivity analyses were performed to test whether the OC-
estimated the associated risk of depression within 2 years after OC
associated depression risk remained similar when (1) parous
initiation, in all women, as well as in groups stratified by age at
women were censored 1 year before their first live birth to avoid
initiation: (1) adolescents (women who initiated OCs before or
the possible influence of postpartum depression, (2) restricting the
at their 20th birthday) and (2) adults (women who initiated OCs
sample to women who identified as “white Irish”, “white British” or
at age 20 or older). Here the time-varying OC use exposure was
“other white” (N = 257,185) to minimize the risk of confounding
coded as “never use” in all women from birth and changed into
due to population stratification, (3) excluding women with other
“initial use” at age of initiation for women who initiated OCs.
psychiatric disorders or medical indications for OC use to ensure
After 2 years of use, the women were censored (see Supplementary
that incident depression was identified and to reduce the risk of
method for more information). In addition, we estimated the life-
confounding by indication and (4) limiting the analysis only to OC
time risk of depression, with the time-varying exposure coded as
users to eliminate the possibility that never users of OC differ from
“never use” for all women from birth. This exposure status changed
users in ways that may affect disease risk. Specifically, we compared
to “ever use” if a woman initiated OC use and continued to be
the hazard rates within 2 years after initiation of OC use with those
classified as “ever use”, regardless of the age at discontinuation.
before initiation.
While age was analyzed as the primary time scale, year of birth
was included as a covariate in the models (Cologne et al., 2012) to
account for cohort effects. To include only women in their repro- Sibling analysis
ductive years, women were censored if they reached menopause,
underwent a hysterectomy, or bilateral oophorectomy, whichever To assess the possible causal relationship between OC use and
came first. We estimated the hazard (rate) ratio (HR) of incident depression, we analyzed a subcohort of female siblings in UKB
depression among users versus never users and its 95% confidence (see Supplementary methods identification details). Inference
interval (CI). All analyses were performed using R version 4.1.1. about Causation from Examination of Familial Confounding is a
regression-based approach for determining causality through the
use of paired observational data collected from related individ-
uals (Li et al., 2020). The statistical model considers both direct
Time-dependent analysis
and indirect causes between the exposure and the outcome, as
Apart from the main analyses, we conducted additional time- well as the impact of shared familial factors. If there is an associa-
dependent analyses to compare the effect of OC use at initiation, tion between a person’s outcome and the person’s own exposure
to the effect (i) during remaining years of use, (ii) among recent that remains unchanged after adjusting for their relative’s expo-
users, as well as (iii) in previous users. The time-varying exposure sure, this would indicate a cause-and-effect relationship between

https://doi.org/10.1017/S2045796023000525 Published online by Cambridge University Press

 4 Johansson et al.

Table 1. Distribution of general characteristics in oral contraceptive initiators and never users

Oral contraceptive initiators (ever users) Never usersa

Number (%) 205,860 (80.6) 49,645 (19.4)

Depression, N (%) 20,454 (9.94) 4,296 (8.65)

Year of birth, median (full range) 1,952 (1936−1970) 1,946 (1936−1970)

Age, median (1st–3rd quartile) 56 (49−62) 62 (56−66)

TDI, median (1st–3rd quartile) −2.22 (−3.68 to 0.3) −1.99 (−3.54 to 0.86)

Number of live births, median (1st–3rd quartile) 2 (1−2) 2 (1−3)

Age at first birth, median (1st–3rd quartile) 26 (22−29) 25 (22−28)

Age of primiparous women at birth of child, median (1st–3rd quartile) 29 (24−34) 28 (24−32)

Number of still births, N (%) 4,744 (0.02) 1,693 (0.03)

Polycystic ovary syndrome, N (%) 1,603 (0.78) 291 (0.59)

Age at menarche, median (1st–3rd quartile) 13 (12−14) 13 (12−14)

Age at sexual debut, median (1st–3rd quartile) 18 (17−20) 20 (18−23)

Age at menopause, median (1st–3rd quartile) 50 (45−52) 50 (45−53)

Family history of depression, N (%) 229 (0.11) 115 (0.23)

Postmenopausal – Yes, N (%) 119,231 (57.92) 35,470 (71.45)

Postmenopausal – No, N (%) 53,568 (26.02) 6,986 (14.07)

Postmenopausal – Not sure hysterectomy, N (%) 23,175 (11.26) 5,889 (11.86)

Postmenopausal – Not sure other, N (%) 9,777 (4.75) 1,168 (2.35)

Age when initiated oral contraceptives, median (1st–3rd quartile) 21 (18−24) NA

Age when discontinued oral contraceptives, median (1st–3rd quartile) 32 (27−40) NA

Duration of oral contraceptive use, median (1st–3rd quartile) 10 (5−18) NA

a
Never users are defined as those who never initiated oral contraceptives during the study period (from birth until the recruitment visit).

the exposure and the outcome. On the other hand, if the associa- never users (Fig. 2 and Table S4). In the age-stratified analyses,
tion between a person’s outcome and their own exposure, as well adolescents had an increased rate of depression (HR = 1.95, 95%
as the association between the person’s outcome and the exposure CI: 1.64–2.32) 2 years following initiation, adults also experienced
of their relative, both are attenuated towards the null after adjust- an increased rate (HR = 1.74, 1.54–1.95: Fig. 2 and Table S5).
ing for each other, this would not support the existence of a direct Although not as pronounced as close to the initiation, also the life-
causal relationship between the exposure and the outcome. We time risk of depression was higher (HR = 1.05, 95% CI: 1.01–1.09)
examined two causal situations: (1) OC use (X) and depression (Y) among ever users compared to never users (Fig. 2 and Table S6).
are associated due to familial confounding only and (2) X and Y are To capture women with symptoms of depression, which might
associated due to a causal effect of X on Y. For technical details on not have come to clinical attention, we analyzed the subcohort of
the method, see Supplementary methods. 82,232 women who completed the online MHQ. Of those, 44,605
reported experiencing at least one of the core depressive symp-
Results toms. OC initiation was associated with an increased hazard rate
of depressive symptoms (HR = 2.00, 95% CI: 1.91–2.10 during the
The study population comprise a total of 264,557 women. Among first 2 years) compared to never users (Fig. 2 and Table S4). OC ini-
the women included, 80.6% were ever users. The median time from tiators who completed the online MHQ and began using OC before
first initiation to last use of OC use was 10 years, and the median or at the age of 20 had 130% higher rate of depressive symptoms
age at initiating and discontinuing use was 21 and 32 years, respec- (HR = 2.30, 95% CI: 2.11–2.51), compared to never users, while
tively. At the initial recruitment visit, the ever users were younger, the corresponding increase in adult initiators was 92% (HR = 1.92,
had a lower TDI (higher socioeconomic status), had less often a 1.81–2.04: Fig. 2 and Table S5). Ever use of OC was also associated
family history of depression, and had an earlier sexual debut, com- with an increased rate of depressive symptoms (HR = 1.27, 95%
pared to the never users. During follow-up, a total of 24,750 women CI: 1.23–1.30) compared with never users (Fig. 2 and Table S6).
received a diagnosis of depression. For participant characteristics,
see Table 1.
Time-dependent analysis
OC use and depression
In the time-dependent analysis, continued use of OCs was not
During the first 2 years of OC use, there was an increased rate associated with an increased rate of depression (HR = 0.94, 95%
of depression (HR = 1.79, 95% CI: 1.63–1.96), compared with CI: 0.89–0.99: Fig. 3 and Table S7). However, both recent (2 years

https://doi.org/10.1017/S2045796023000525 Published online by Cambridge University Press

 Epidemiology and Psychiatric Sciences 5

Figure 2. Short-term and lifetime HR of initiation of oral contraceptive use on depression. All estimates are adjusted for year of birth, TDI (used as proxy for socioeconomic
status), number of live births, number of still births, PCOS (defined as having ICD10 code E28), age at menarche, age at sexual debut and family history of severe depression
(defined as having one or more first-degree relatives with depression). Adolescents are defined as women who initiated oral contraceptives either before or at the age of 20,
while adults are defined as women initiating oral contraceptives after turning 20 years old. Lifetime risk = among all women who initiated oral contraceptives at some time
point during the follow-up. *Secondary outcome measurement on the subcohort of women (N = 82,232) who answered in mental health questionnaire.

Figure 3. Time-dependent HR of oral contraceptive use and depression. All estimates are adjusted for year of birth, TDI (used as proxy for socioeconomic status), number of
live births, number of still births, PCOS (defined as having ICD10 code E28), age at menarche, age at sexual debut and family history of severe depression (defined as having
one or more first-degree relatives with depression). *Recent users defined as those who discontinued within 2 years. *Secondary outcome measurement on the subcohort of
women (N = 82,232) who answered the mental health questionnaire. **Parous women were followed until 1 year before delivery.

since cessation) and previous OC users (more than 2 years since Sensitivity analyses
cessation) had an increased hazard of depression (HR = 1.17,
Sensitivity analysis on the subcohort of women who self-identified
95% CI: 1.08–1.27 and 1.07, 1.03–1.11), respectively, compared
as white did not yield any marked changes to the risk estimates
with never users. The increased hazard of depression in previous
(Tables S4, S6 and S8). Similar estimates were also found when
users was driven by adolescent OC users (HR = 1.18, 95% CI:
women with a medical indication for OC use and with other
1.12–1.25), whereas no significant association was found for adult
psychiatric disorders were excluded (Tables S4 and S9). In the
OC users 2 years after discontinuing (HR = 1.00, 0.95–1.04: Table
sensitivity analysis where parous women were followed until 1
S10). Our secondary outcome analysis (MHQ) revealed that even
year before giving birth or depression (whichever came first), the
after using OC for more than 2 years, an increase in the hazard
HRs were slightly higher or similar (Fig. 3 and Tables S6–S8).
rate (HR = 1.13, 95% CI: 1.09–1.17: Fig. 3 and Table S7) was still
Among those who had ever initiated OCs, the hazard rate of receiv-
observed. In line with our primary outcome measure (i.e., a depres-
ing a first depression diagnosis within 2 years after initiation was
sion diagnosis), an increased association of depressive symptoms
higher (HR = 1.93, 95% CI: 1.71–2.17: Table S11) than before
was found among recent and previous OC users (HR = 1.40, 95%
initiating OCs.
CI: 1.33–1.48 and 1.13, 1.10–1.17), respectively.

https://doi.org/10.1017/S2045796023000525 Published online by Cambridge University Press

 6 Johansson et al.

Sibling analysis attributed to a greater susceptibility to gonadal hormones, includ-

ing hormonal contraception, during crucial developmental periods
Causal inference analysis was performed on a total of 7,354 first-
that affect the organization of brain structures and may lead to
degree sister pairs. Among these, 81% had initiated OCs. The
long-lasting changes (Anderl et al., 2022, 2020; Cahill, 2018; de Wit
within-sibling pair correlation for OC use was 0.20 (Pearson’s
et al., 2020).
correlation coefficient, P < 0.001), and the corresponding odds
Residual confounding, due to familial disposition, early menar-
ratio (OR) was 3.08 (95% CI: 2.80–3.38). The within-sibling pair
che (Karina and Sivakumaran, 2017) and sexual debut (McKetta
correlation for depression was 0.03 (Pearson’s correlation coeffi-
and Keyes, 2019), has been suggested to explain the increased
cient, P = 0.002), and the corresponding OR was 2.16 (1.28–3.40).
risk of depression associated with OC use. These factors were all
A sibling’s OC use was positively associated with a depression diag-
adjusted for in the current study. However, other potential con-
nosis (model 1: 𝛽self = 0.51, 95% CI: 0.23–0.80: Table S12). The
founders include medical indications for hormonal contraception
co-sibling’s OC use was also associated with the sibling’s depres-
use (Duke et al., 2007). The present study accounted for medical
sion diagnosis (model 2: 𝛽co-sibling = 0.29, 95% CI: 0.001–0.58).
indication by excluding women with dysmenorrhoea, endometrio-
Adjusting for the co-sibling’s OC use (model 3), 𝛽′ self remained
sis, and PCOS, but as the premenstrual dysphoric disorder diagno-
unchanged (𝛽′ self = 0.48, 95% CI: 0.19–0.76) compared with 𝛽self
sis did not exist in the ICD10, it was not possible to directly adjust
in model 1 (P for difference 0.16), while 𝛽′ co-sibling = 0.19 (95%
for this. However, the heritability of premenstrual dysphoric dis-
CI: −0.10 to 0.48) was attenuated towards null as compared with
order is high (Condon, 1993), and therefore the sibling analysis
𝛽co-sibling in model 2. These results support the hypothesis of a
should partly account for any such confounding.
causal relationship between OC use and depression, such that OC
Our findings must be interpreted in the light of several lim-
use increases the risk of depression.
itations. First, the main limitation of this study is the potential
recall bias in the self-reported data, particularly regarding the age
of OC use initiation and discontinuation. Second, the study is sub-
Discussion ject to a certain sample selection bias as the UKB consists of a
In this study, including 264,557 women, we showed that OC use is healthier population compared to the general population of the
associated with an increased risk of depression shortly after ini- UK (Fry et al., 2017), which hampers the generalizability of our
tiation. The increased risk declined with continued OC use, but findings. In addition, white Europeans are overrepresented in the
the lifetime risk associated with ever OC use remained signifi- UKB. The small number of observations in non-white participants
cantly increased. Our findings are comparable to what was found precluded us from performing analyses stratified by ethnic back-
in a Danish study (Skovlund et al., 2016), which identified that the ground. Third, we were not able to evaluate different formulations
risk peaked half a year after initiation and declined with contin- or routes of administration as we did not have detailed information
ued use. These results could be explained by hormonal fluctuations on the OC type used. Hence, our results might not be generalizable
induced by OC initiation, which can affect women who are par- to all types of OCs used today. Nonetheless, given the birth year
ticularly sensitive to changes in the levels of hormones and their of the women included in the study, our results are mainly based
metabolites, such as allopregnanolone (Hantsoo and Epperson, on the second-generation OCs containing a combination of both
2015). These fluctuations could alter GABAergic regulation of the estrogen and progesterone, which are still used by many women
hypothalamic–pituitary–adrenocortical in this group of women today. Fourth, as we only had information about age at first and
(Gordon et al., 2015). Our results are also comparable to what was last use, we were not able to capture if they stopped and restarted in
seen in a study estimating the risk of suicidal behaviour, which was between, which makes the time-dependent association less precise.
found to be higher during the initial use of OC (Edwards et al., Fifth, there is a potential for recall bias also for the self-reported
2022). family history data used as covariate information. However, miss-
Unlike most previous studies, we estimated the time-varying ing family history data is minimal (88% report parental history and
effects of OC use using a “new user” design approach (Yoshida 93% report sibling history) (Hujoel et al., 2022), and it has been
et al., 2015). This allowed us to capture events occurring in differ- shown that self-reported family history is accurate (∼80% correla-
ent time windows during follow-up. Using a prevalent user design, tion between true and self-reported family history, based on sibling
which assumes the effect is similar in current and new users, would concordance) (Hujoel et al., 2020). Last, some confounders were
miss the increased risk seen early in the treatment course. This can only measured once, which could impact our estimates. However,
explain why some previous studies, where the rate of depression we incorporated time-varying covariates when possible to account
among current OC users is compared to the rate among never or for changes during follow-up.
previous users, have not identified a significant effect (Cheslack-
Postava et al., 2015; Lundin et al., 2022; McKetta and Keyes, 2019).
Our study found higher depression rates in the first years after
Conclusion
discontinuing OCs. This may reflect that women who get mood-
related problems discontinue OC use, but are not diagnosed with Our findings support that OC use is causally associated with an
depression until after cessation. increased risk of depression in adolescents as well as in adults,
Our results, consistent with the Danish study (Skovlund et al., especially shortly after the initiation. It is important to emphasize
2016), suggest that the risk of depression is increased not only that most women tolerate OCs well without experiencing adverse
among adolescents initiating OCs but also among women older mood effects, making them a great option for many. However,
than 20 years. However, our findings showed that women who used educating OC users, screening for depression, informing pri-
OCs during adolescence remained at a heightened risk even after mary healthcare practitioners regarding the OC–depression
they discontinued, whereas such a risk was not apparent among relationship and conducting further research to determine the
adult OC users. It has been hypothesized that the increased risk cause of hormone contraceptive-precipitated depression are
later in life among those who used OCs during adolescence may be warranted.

https://doi.org/10.1017/S2045796023000525 Published online by Cambridge University Press

 Epidemiology and Psychiatric Sciences 7

Supplementary material. The supplementary material for this article can Duke JM, Sibbritt DW and Young AF (2007) Is there an association between
be found at https://doi.org/10.1017/S2045796023000525. the use of oral contraception and depressive symptoms in young Australian
women? Contraception 75(1), 27–31.
Availability of data and materials. The data used for this study is avail- Edwards AC, Lonn ̈ SL, Crump C, Moscicki ́ EK, Sundquist J, Kendler KS
able for bona fide researchers from the UK Biobank Resource (http://www. and Sundquist K (2022) Oral contraceptive use and risk of suici-
ukbiobank.ac.uk/about-biobank-uk/) and can be accessed by an application to dal behavior among young women. Psychological Medicine 52(9),
the UK Biobank. 1710–1717.
Fry A, Littlejohns TJ, Sudlow C, Doherty N, Adamska L, Sprosen T,
Acknowledgements. We acknowledge the participants and staff involved in Collins R and Allen NE (2017) Comparison of sociodemographic and
UKB for their contribution. The computations were performed on resources health-related characteristics of UK Biobank participants with those
provided by the Swedish National Infrastructure of Computing (SNIC) of the general population. American Journal of Epidemiology 186(9),
through Uppsala Multidisciplinary Center for Advanced Computational 1026–1034.
Science (UPPMAX) under projects SNIC 2018/8-372 and sens2017538. Gordon JL, Girdler SS, Meltzer-Brody SE, Stika CS, Thurston RC, Clark CT,
Prairie BA, Moses-Kolko E, Joffe H and Wisner KL (2015) Ovarian
hormone fluctuation, neurosteroids, and HPA axis dysregulation in peri-
Financial support. This work was primarily funded by the Swedish Research
menopausal depression: A novel heuristic model. The American Journal of
Council, the Swedish Brain Foundation and the Uppsala University center for
Psychiatry 172(3), 227–236.
Women’s mental health during the reproductive lifespan.
Haakenstad A, Angelino O, Irvine CMS, Bhutta ZA, Bienhoff K, Bintz C,
Causey K, Dirac MA, Fullman N, Gakidou E, Glucksman T, Hay SI,
Competing interests. None.
Henry NJ, Martopullo I, Mokdad AH, Mumford JE, Lim SS, Murray CJL
and Lozano R (2022) Measuring contraceptive method mix, prevalence, and
Ethical standards. All participants provided informed consent to partic- demand satisfied by age and marital status in 204 countries and territories,
ipate in UKB. Our study was conducted under generic approval from the 1970–2019: A systematic analysis for the Global Burden of Disease Study
research ethics committee (reference 11/NW/0382) under the UKB applica- 2019. The Lancet 400(10348), 295–327.
tion no. 41143. The present research was also approved by the Swedish Ethical Hantsoo L and Epperson CN (2015) Premenstrual dysphoric disorder:
Review Authority (dnr: 2020-04415). Epidemiology and treatment. Current Psychiatry Reports 17(11), 87.
Hujoel MLA, Gazal S, Loh P-R, Patterson N and Price AL (2020) Liability
threshold modeling of case–control status and family history of disease
increases association power. Nature Genetics 52(5), 541–547.
References
Hujoel MLA, Loh P-R, Neale BM and Price AL (2022) Incorporating family
Anderl C, de Wit AE, Giltay EJ, Oldehinkel AJ and Chen FS (2022) history of disease improves polygenic risk scores in diverse populations. Cell
Association between adolescent oral contraceptive use and future Genomics 2(7), 100152.
major depressive disorder: A prospective cohort study. Journal Karina IM and Sivakumaran P (2017) Hormonal contraception and its asso-
of Child Psychology and Psychiatry, and Allied Disciplines 63(3), ciation with depression. JAMA Psychiatry 74(3), 301.
333–341. Keyes KM, Cheslack-Postava K, Westhoff C, Heim CM, Haloossim M,
Anderl C, Li G and Chen FS (2020) Oral contraceptive use in adolescence Walsh K and Koenen K (2013) Association of hormonal contraceptive use
predicts lasting vulnerability to depression in adulthood. Journal of Child with reduced levels of depressive symptoms: A National Study of Sexually
Psychology and Psychiatry, and Allied Disciplines 61(2), 148–156. Active Women in the United States. American Journal of Epidemiology
Cahill L (2018) How does hormonal contraception affect the develop- 178(9), 1378–1388.
ing human adolescent brain? Current Opinion in Behavioral Sciences 23, Larsson G, Blohm F, Sundell G, Andersch B and Milsom I (1997) A longi-
131–135. tudinal study of birth control and pregnancy outcome among women in a
Cheslack-Postava K, Keyes KM, Lowe SR and Koenen KC (2015) Oral con- Swedish population. Contraception 56(1), 9–16.
traceptive use and psychiatric disorders in a nationally representative sample Li S, Bui M and Hopper JL (2020) Inference about causation from exam-
of women. Archives of Women’s Mental Health 18(1), 103–111. ination of familial confounding (ICE FALCON): A model for assessing
Cologne J, Hsu W-L, Abbott RD, Ohishi W, Grant EJ, Fujiwara S and causation analogous to Mendelian randomization. International Journal of
Cullings HM (2012) Proportional hazards regression in epidemiologic Epidemiology 49(4), 1259–1269.
follow-up studies: An intuitive consideration of primary time scale. Lindberg M, Foldemo A, Josefsson A and Wiréhn A-B (2012) Differences in
Epidemiology (Cambridge, Mass.) 23(4), 565–573. prescription rates and odds ratios of antidepressant drugs in relation to indi-
Condon JT (1993) The premenstrual syndrome: A twin study. The British vidual hormonal contraceptives: A nationwide population-based study with
Journal of Psychiatry: The Journal of Mental Science 162, 481–486. age-specific analyses. The European Journal of Contraception & Reproductive
David PS and Kling JM (2020) Spotlight on the noncontraceptive benefits of Health Care 17(2), 106–118.
contraceptives. Journal of Women’s Health (2002) 29, 887–888. Lundin C, Wikman A, Lampa E, Bixo M, Gemzell-Danielsson K, Wikman P,
Davis KAS, Coleman JRI, Adams M, Allen N, Breen G, Cullen B, Dickens C, Ljung R and Sundstrom ̈ Poromaa I (2022) There is no association
Fox E, Graham N, Holliday J, Howard LM, John A, Lee W, McCabe R, between combined oral hormonal contraceptives and depression: A Swedish
McIntosh A, Pearsall R, Smith DJ, Sudlow C, Ward J, Zammit S and register-based cohort study. BJOG: An International Journal of Obstetrics &
Hotopf M (2020) Mental health in UK Biobank – Development, imple- Gynaecology 129(6), 917–925.
mentation and results from an online questionnaire completed by 157 366 McKetta S and Keyes KM (2019) Oral contraceptive use and depression among
participants: A reanalysis. BJPsych Open 6(2), e18. adolescents. Annals of Epidemiology 29, 46–51.
de Wit AE, Booij SH, Giltay EJ, Joffe H, Schoevers RA and Oldehinkel AJ Morssinkhof MWL, Lamers F, Hoogendoorn AW, de Wit AE, Riese H,
(2020) Association of use of oral contraceptives with depressive symp- Giltay EJ, van den Heuvel OA, Penninx BW and Broekman BFP
toms among adolescents and young women. JAMA Psychiatry 77(1), (2021) Oral contraceptives, depressive and insomnia symptoms in adult
52–59. women with and without depression. Psychoneuroendocrinology 133,
de Wit AE, de Vries YA, de Boer MK, Scheper C, Fokkema AA, Schoevers RA 105390.
and Giltay EJ (2021) Hormonal contraceptive use and depressive symptoms: Payne JL (2003) The role of estrogen in mood disorders in women. International
Systematic review and network meta-analysis of randomised trials. BJPsych Review of Psychiatry 15(3), 280–290.
Open 7(4), e110. Robakis T, Williams KE, Nutkiewicz L and Rasgon NL (2019) Hormonal con-
Dhont M (2010) History of oral contraception. The European Journal of traceptives and mood: Review of the literature and implications for future
Contraception & Reproductive Health Care 15(sup2), S12–S18. research. Current Psychiatry Reports 21(7), 57.

https://doi.org/10.1017/S2045796023000525 Published online by Cambridge University Press

 8 Johansson et al.

Schaffir J, Worly BL and Gur TL (2016) Combined hormonal contracep- UK Biobank (2019) First Occurrence of Health Outcomes Defined by
tion and its effects on mood: A critical review. The European Journal 3-character ICD10 code https://biobank.ndph.ox.ac.uk/ukb/ukb/docs/first_
of Contraception & Reproductive Health Care: The Official Journal of the occurrences_outcomes.pdf.
European Society of Contraception 21(5), 347–355. VanderWeele TJ, Hernán MA and Robins JM (2008) Causal directed acyclic
Skovlund CW, Mørch LS, Kessing LV and Lidegaard O (2016) Association graphs and the direction of unmeasured confounding bias. Epidemiology
of hormonal contraception with depression. JAMA Psychiatry 73(11), (Cambridge, Mass.) 19(5), 720–728.
1154–1162. Yoshida K, Solomon DH and Kim SC (2015) Active-comparator design and
Therneau TM and Grambsch PM (2000) Modeling Survival Data: Extending new-user design in observational studies. Nature Reviews Rheumatology
the Cox Model New York: Springer. 11(7), 437–441.
Toffol E, Heikinheimo O, Koponen P, Luoto R and Partonen T Zettermark S, Perez Vicente R and Merlo J (2018) Hormonal contraception
(2012) Further evidence for lack of negative associations between increases the risk of psychotropic drug use in adolescent girls but not in
hormonal contraception and mental health. Contraception 86(5), adults: A pharmacoepidemiological study on 800 000 Swedish women. PLoS
470–480. One 13(3), e0194773.

https://doi.org/10.1017/S2045796023000525 Published online by Cambridge University Press