Risk of breast cancer with prolactin elevating antipsychotic drugs: An observational study of U.S. women (ages 18–64)

Abstract

Purpose/Background:

Antipsychotic drugs are well established to alter circulating prolactin levels by blocking dopamine D-2 receptors in the pituitary. Prolactin activates many genes important in the development of breast cancer. Prior studies have found an association with antipsychotic use and risk of breast cancer.

Methods/Procedures:

The IBM MarketScan Commercial and Medicaid Databases were used to establish a large, observational cohort of women taking antipsychotics drugs compared to anticonvulsants or lithium. A new user design was used that required 12 months of insurance enrollment prior to the first antipsychotic or anticonvulsant/lithium prescription. Invasive breast cancer was identified using diagnostic codes. Multivariable Cox proportional hazards models were used to evaluate the risk of breast cancer with antipsychotic drug exposure controlling for age and other risk factors.

Findings/Results:

A total of 914 (0.16%) cases of invasive breast cancer were identified among 540,737 women. Exposure to all antipsychotics was independently associated with a 35% increased risk of breast cancer (aHR 1.35 95% CI (1.14–1.61). Category 1 drugs (high prolactin) were associated with a 62% increased risk (aHR 1.62, 95%CI (1.30–2.03), category two drugs a 54% increased risk (aHR 1.54 95% CI (1.19–1.99), while category three drugs were not associated with breast cancer risk.

Implications/Conclusions:

In the largest study of antipsychotics taken by U.S. women, a higher risk between antipsychotic drug use and increased risk for breast cancer was observed, with a differential higher association with antipsychotic categories that elevate prolactin. Our study confirms other recent observational studies of increased breast cancer risk with antipsychotics that elevate prolactin.

Introduction

The incidence of breast cancer in the United States is 268,600 cases per year. Approximately 13% of women will be diagnosed with breast cancer at some point during their lifetime.¹ Endocrine evidence suggests that prolactin, in addition to other factors, plays an important role in the development of breast tumorigenesis. Prolactin receptor (PRLr) is overexpressed by 95% of breast cancers and activates many genes responsible for proliferation and metastatic spread of breast cancer through PRL-PRLr-JAK2-STAT5 signaling.^2,3 Many antipsychotic drugs are well known to produce elevated serum prolactin and can induce breast neoplasms in rodent studies, as mentioned in their package inserts.^2–4 Hyperprolactinemia from antipsychotic drug use is caused by blockade of dopamine (D-2) receptors in the tuberoinfundibular tract by removing inhibitory influence on lactotroph cells in the anterior pituitary. Common side effects in humans include amenorrhea, oligomenorrhea, osteoporosis, gynecomastia, early telarche, sexual dysfunction, and infertility.² Mouse models that closely mimic human breast cancer initiation, have shown that hyperprolactinemia-inducing antipsychotics accelerate early lesion progression to cancerous cells.³

Circulating Prolactin and Breast Cancer Risk

Epidemiologic evidence suggests that circulating prolactin concentrations are associated with increased breast cancer risk. Prolactin levels are measured in the blood (normal levels are typically 10–28 μg/L in women).² Several large prospective studies, in which blood was collected prior to breast cancer diagnosis, have observed positive associations between prolactin and risk.^5–9 In a pooled analysis of ~80% of the world’s prospective data, the relative risk (RR) comparing women in the top vs bottom quartile of prolactin levels was 1.3 (95% confidence interval (CI): 1.1, 1.6, p-trend = 0.002). The results were similar for both premenopausal and postmenopausal women.⁵

Antipsychotic Database Studies of Breast Cancer Risk

Recent antipsychotic drug database studies have shown growing evidence for increased risk for breast cancer with prolactin elevating antipsychotics, often with a dose-response pattern. Older studies found mixed associations, but also had insufficient sample sizes and weak study designs, discussed later. An often cited observational study by Wang, et al examined over 100,000 women and compared 52,819 women on dopamine antagonists to 55,289 women not treated with them. Women who used antipsychotic dopamine antagonists had a 16% greater risk (adjusted HR 1.16, 95% CI, 1.07–1.26) of developing breast cancer compared to women taking a random drug. The study found a dose–response relationship between larger cumulative dosages and greater risk.⁴ Another population based cohort study in Taiwan stratified drugs by prolactin elevating potential and found higher rates of breast cancer in women with schizophrenia compared to women without a mental illness that were not taking antipsychotics. Women treated with risperidone, paliperiodone or misulpride had a higher breast cancer risk. (adjusted HR 1.96, 95% CI, 1.36–2.82).¹¹ A recent Danish study of first time breast cancer diagnosis found that long-term use of antipsychotics (≥10 000 mg olanzapine equivalents) was associated with an increased risk of breast cancer, with an adjusted OR of 1.18 [95% confidence interval (CI), 1.06, 1.32]. A weak dose–response pattern was seen, with ORs increasing to 1.27 (95% CI 1.01, 1.59) for ≥50 000 mg olanzapine equivalents.¹³ An insurance database cohort study in South Korea compared women prescribed second generation antipsychotics to age matched controls. A higher risk of breast cancer was observed in subjects prescribed ≥10,000 mg of olanzapine equivalent dose (HR=1.29, 95% CI 1.14–1.46) than those with <10,000 mg (HR=1.05, 95% CI 1.00–1.11). The increased risk in the case group became significant after six years of the observation period (≥6yrs: HR=1.24, 95% CI 1.14–1.35, 3-<6yrs: HR=1.06, 95% CI 0.97–1.15, <3yrs: HR=1.02, 95% CI 0.95–1.09).

Laboratory studies have shown that antipsychotic induced prolactin exerts a dichotomous effect on mammary cells in mouse models that mimic human breast tissue. Prolactin-JAK2-STAT5 signaling weakens the apoptosis anticancer barrier in precancerous lesions and thus may increase breast cancer risk. However, the same signaling pathway can also cause the mammary cells to undergo differentiation potentially leading to lower cell proliferation rates and thus may reduce the chance of gaining mutations and consequently breast cancer risk.³ A cautious approach to utilizing antipsychotics by avoiding prolactin elevation is currently advised for women with a previously detected breast cancer.^2,3

Antipsychotics, Prolactin and Drug Categorization

In order to examine potential biological effects of circulating prolactin induced by antipsychotics, we aimed to observe breast cancer risk by first concisely categorizing antipsychotic drugs according to their prolactin elevating properties.^2,19 The terms “first and second generation” were arbitrarily assigned to many antipsychotics drugs depending on their first release to the market and potential to cause extrapyramidal side effects, and can cause confusion, because they are not based solely on propensity to raise prolactin.^20,21 For instance, typical or “first generation” antipsychotics, as well as the “second generation” drugs, risperidone and paliperidone have the highest propensity to raise prolactin (45-to >100 ng/ml), whereas “atypical” antipsychotics are associated with more modest elevations, while aripirazole/ brexpriprazole are not associated with it.^2,19–29 Utilizing prolactin based taxonomy², we conducted the largest epidemiologic study to date investigating the risk of breast cancer and antipsychotics from an administrative claims database of over 150 million American women. We hypothesized that women prescribed prolactin elevating antipsychotic drugs would have higher risk of breast cancer compared to women taking anticonvulsants or lithium as the active comparator group. We further hypothesized that relative to control drugs, the risk of breast cancer would follow the order of high, mid and low prolactin elevating categories of antipsychotics.

Methods

Data Sources

We performed a large, retrospective, observational cohort study of breast cancer risk in women 18–64 years of age exposed to antipsychotic agents, anticonvulsants, and/or lithium using administrative claims data from the IBM® MarketScan® Commercial and Multi-State Medicaid Databases. Women 65 years of age and older were not included in this database as they are often covered by Medicare and would not have complete claims data. The MarketScan Databases include medical and outpatient prescription drug claims from 150 million privately insured persons (Commercial), and 20 million Medicaid insured persons. Variables were defined using ICD-9-CM diagnosis and procedure codes and CPT codes on medical claims, and outpatient prescription drug claims submitted for reimbursement. The databases contain no patient identifiers and the study was considered exempt by the Washington University in St. Louis School of Medicine, Human Research Protection Office.

Study population

We identified women between the ages of 18 to 64 years with an outpatient prescription drug claim with days’ supply > 0 days for an antipsychotic, anticonvulsant, or lithium from 1/1/2007 through 6/30/2016 (1/1/2012 through 6/30/2016 from the Medicaid Multi-State database). Supplemental Table 1 contains the generic drug names for each category of drug in the study. We identified new users as those women with at least 12 months of continuous medical and prescription drug insurance enrollment prior to the first claim for a study drug, the fill date of which was used as the index date.

Quantification of Exposure to Antipsychotics and Control Drugs

We separated antipsychotic drugs into three categories (category 1 highest to category 3 lowest) based on their propensity to elevate prolactin. (Supplemental Table 1). These categories were derived a priori from previously published antipsychotic clinical guides.^2,19 We chose anticonvulsants and lithium as comparator drugs since they are not known to raise prolactin, but are also prescribed to patients with mental illnesses, and thus could mitigate some potential confounders related to having a mental health diagnosis such as parity and age at first birth.^30–33 Exposure within the drug categories was measured using drug-specific defined daily dose (DDD), a quantity developed by the World Health Organization’s Collaborating Centre for Drug Statistics Methodology.³⁴ The strength, quantity, and days’ supply listed on each claim were used to compute per-day DDD, with upper limits applied based on typical prescribing patterns determined by one author (TR, Supplemental Table 1). To determine each woman’s cumulative exposure to each category of drug, we summed DDDs across all claims beginning with the index fill and ending with the first outcome or censoring event (i.e., end of enrollment, six years after index, or evidence of prevalent disease [see below]). We then divided cumulative exposure by the length of this time period to obtain the average exposure during the observed treatment time. The average DDD per day of observed treatment was calculated as the cumulative DDD/total days of observed treatment period. For example, a woman with prescriptions for 300 mg/day of chlorpromazine (category 1 medication) covering every day of her study observation would have an average DDD of 1.00 for chlorpromazine (300 mg defined by the WHO as the average daily maintenance dose). If the same woman had prescriptions for the same dose of chlorpromazine for 10% of her study observation time, the average DDD for chlorpromazine would be 0.10 per day. Women treated solely with prochlorperazine were not included, since it is also an antiemetic, but it was included in the calculation of the total DDD exposure to Category 1 antipsychotics with other study drug exposures.

Identification of Invasive Breast Cancer

We identified invasive breast cancer using hierarchical definitions based on the quality of available evidence from diagnosis/ procedure codes (Supplemental Table 2). We first excluded likely prevalent cases indicated by a prescription drug claim for tamoxifen or a diagnosis of breast cancer or history of breast cancer on any claim prior to the index date. Invasive breast cancer was identified primarily by International Classification of Diseases, 9th/10th edition, Clinical Modification, (ICD-9/10) codes for invasive breast cancer on a claim with a Current Procedural Terminology (CPT-4) code for surgical pathology microscopic examination (Supplemental Table 2), indicating pathologic verification.³⁵ If this was not present, a diagnosis of invasive breast cancer on an inpatient facility claim or on at least two provider/outpatient claims separated by 30–180 days was required, with the later of the two dates for outpatient diagnoses used as the date of invasive cancer. For potential cases of invasive breast cancer without pathologist confirmation, evidence of surgical treatment (mastectomy or breast-conserving surgery within one month before through six months after diagnosis), or chemotherapy (within six months after diagnosis, chemotherapy administration coded for invasive breast cancer) was required to establish the diagnosis of invasive breast cancer. Diagnoses without associated surgical or chemotherapeutic treatment were assumed to represent prevalent/ unverified cases and were treated as censoring events.

Identification of Covariates

Covariates of interest were identified using ICD-9/10 diagnosis codes and ICD-9/10 procedure and CPT-4 codes from inpatient and outpatient claims, and prescription drug claims for some covariates. Baseline covariates assessed using all available claims³⁶ prior to the index date included potential risk factors for invasive breast cancer, including benign breast disease, smoking and smoking-related diseases, diabetes, alcohol abuse, obesity, and additional proxies for obesity, including sleep apnea, gastroesophageal reflux disease, lipodystrophy, and other lipidemias (Supplemental Table 2). The additional proxies for obesity were included in the definition of obesity due to the low sensitivity of coding for obesity in claims data, including in breast cancer patients.³⁷ We used prescription drug claims to identify lipid-lowering medications, statins, smoking cessation therapy, and antidiabetic agents, to improve sensitivity of their identification. Hormone replacement therapy was identified using prescription drug claims in the year prior to the index date in the therapeutic class ‘Estrogens & Combinations’, categorized as estrogen only vs estrogen plus progesterone. Hormone replacement therapy was restricted to women aged 50 years and older or younger women with a diagnosis of acquired absence of genital organs. Preexisting benign breast disease was identified by diagnosis or procedure codes (Supplemental Table 2), but was not considered if it was first coded within 180 days of the date of invasive breast cancer, since it was most likely concurrent with the invasive breast cancer. For all variables defined by diagnosis or procedure codes, the absence of codes for the condition in the claims data was interpreted as lack of the condition of interest (and the variable thus coded 0), as is typically done in analyses of claims data. Demographics included as covariates in the models including age, sex, and commercial vs. Medicaid health insurance were determined at the index date. U.S. region and race were not included in statistical analyses, since geography was only available for commercially insured patients and race was only available for Medicaid patients.

Statistical Analyses

Bivariable analyses of the association between average exposure to antipsychotics and risk of invasive breast cancer were performed using Cox proportional hazards models. An analysis of all pooled antipsychotic drugs as well as an analysis of antipsychotic category based (1–3) exposures were performed. The primary exposure of antipsychotic medication utilization was modeled as described above by the average DDD, with the hazard ratio (HR) quantifying the risk associated with a 1.00-unit increase in the DDD. Personal history of breast cancer was a censoring event during follow up through 30 days after an invasive breast cancer diagnosis, since coding of personal history in a short time frame after diagnosis likely indicates prior malignancy. Additional censoring, as described above, was performed at the earliest date of the following: death date, end of enrollment, or six years after the index date. Total follow-up was truncated at six years because of the small number of women remaining in the population. Three multivariable Cox proportional hazards models were used to adjust for known breast cancer risk factors that were determined a priori and included in the models regardless of significance. One model adjusted for known risk factors for breast cancer (e.g., obesity, hormone replacement therapy), another adjusted for age alone (modeled as a cubic spline), and the third model adjusted for age plus known risk factors for breast cancer. In addition to primary models, a sensitivity analysis was performed including additional adjustment for mental health conditions plus known risk factors for breast cancer, to account for potential confounding by severity of mental illness. Identical models were developed stratified by age less than or equal to or greater than 50 years as additional sensitivity analyses. We assessed ascertainment bias by counting the number of office/outpatient evaluation and management (E&M) encounters to providers other than psychiatrists/ psychologists during the baseline period for women with and without breast cancer.

Results

Demographic characteristics of the 540,737 commercially and Medicaid-insured women who were new users of antipsychotics, anticonvulsants and/ or lithium and met inclusion criteria for the study are in Table 1. There were 441,750 commercially insured women from 2007–2016 with a median age of 42 years and median duration of enrollment of 1,734 days, and 98,987 Medicaid insured women from 2012–2016 with a median age of 36 years and median duration of enrollment of 1,461 days. During the observation period of post-index enrollment in the MarketScan Databases, 914 (0.2%) women were newly diagnosed with invasive breast cancer. Slightly over 40% of the women resided in the Southern U.S. census region, consistent with the regional geographic representation in the U.S. In the Medicaid database, approximately 60% of the women newly filling a prescription for study medications were White.

Table 1.

Demographics of the Population of Women Aged 18–64 Years Stratified by Initial Exposure to Antipsychotic Medications or Anticonvulsants/Lithium

Characteristic	New Users of Antipsychotic Medicationsa 312,702 (57.8%) n (%)	New Users of Anticonvulsants/Lithium 228,035 (42.2%) n (%)
Age in years, median (IQR)	41 (30, 52)	39 (29, 50)
Duration of total enrollment in days, mean (SD)	1,849 (982)	1,881 (1,005)
Duration of total enrollment in days, median (IQR)	1,614 (1,096, 2,406)	1,643 (1,096, 2,496)
Medicaid enrollment	64,747 (20.7)	34,240 (15.0)
Regionb
Northeast	34,680 (14.0)	25,909 (13.4)
North Central	58,740 (23.7)	44,607 (23.0)
South	108,704 (43.8)	84,106 (43.4)
West	42,345 (17.1)	36,651 (18.9)
Unknown	3,486 (1.4)	2,522 (1.3)
Racec
White	37,297 (57.6)	21,794 (63.7)
Black	17,838 (27.6)	6,422 (18.8)
Hispanic	1,044 (1.6)	613 (1.8)
Other	6,703 (10.4)	4,483 (13.1)
Unknown	1,865 (2.9)	928 (2.7)

Abbreviations: IQR, interquartile range; SD, standard deviation.

^aIncludes 15,391 (4.9%) patients with exposure to antipsychotics and anticonvulsants/lithium at index.

^bAvailable only for commercially insured patients.

^cAvailable only for Medicaid insured patients.

Of the women in the study population, approximately 52% filled at least one prescription for a category 3 antipsychotic agent, while 15% filled at least one prescription for a category 1 agent. Forty-nine percent of women filled at least one prescription for an anticonvulsant medication during their period of study observation. The median average DDD exposure to the three categories of antipsychotic medication was 0.1 or less, reflecting intermittent utilization of these medications.

Bivariable comparison of factors during the baseline period associated with invasive breast cancer among women in the cohort are summarized in Table 2. Obesity, diabetes, hormone replacement therapy (in women 50 years and older, or younger women coded for surgical menopause), and preexisting benign breast disease were significantly more common in women newly diagnosed with invasive breast cancer than in women not diagnosed with breast cancer during the time period of their health insurance enrollment. The median age of women with breast cancer was 12 years older than women not diagnosed with breast cancer (53 vs 41 years, respectively, p < 0.0001). The median number of outpatient E&M encounters during the baseline year prior to the index date was slightly higher in women with subsequent breast cancer than in those without invasive breast cancer during the follow-up observation period (median 4 vs. 3 visits, p < 0.0001). Over 84% of women had at least one E&M encounter during the baseline period.

Table 2.

Bivariate Analysis of Factors During the Baseline Period Associated with Invasive Breast Cancer in Women with at least One Filled Prescription for Antipsychotic, Anticonvulsant, and/or Lithium Medications in the MarketScan Commercial and Multi-State Medicaid Databases

Characteristic	Invasive Breast Cancer N = 914 (0.2%) n (%)	No Breast Cancer N = 539,823 (99.8%) n (%)	P
Age in years, median (IQR)	53 (47, 58)	41 (30, 51)	<.0001
Duration of total enrollment in days, median (IQR)	2,557 (1,827, 3,472)	1,643 (1,096, 2,496)	<.0001
Number of Evaluation & Managementa encounters during baseline period, median (IQR)	4 (2, 7)	3 (1, 6)	<.0001
At least 1 Evaluation & Management encounter during baseline (%)	798 (87.3)	455,281 (84.3)	0.01
Mental health conditions
Bipolar disorder	67 (7.3)	66,716 (12.4)	<.0001
Schizophrenia	22 (2.4)	13,983 (2.6)	0.72
Major depression	339 (37.1)	216,113 (40.0)	0.07
Medicaid enrollment	74 (8.1)	98,913 (18.3)	<.0001
Hormone replacement therapyb	122 (13.3)	26,428 (4.9)	<.0001
Estrogen/progesterone-based HRT	19 (2.1)	3,073 (0.6)	<.0001
Estrogen-only HRT	103 (11.3)	23,355 (4.3)	<.0001
Obesity	545 (59.6)	253,509 (47.0)	<.0001
Diabetes	141 (15.4)	58,733 (10.9)	<.0001
Alcohol abuse	43 (4.7)	33,784 (6.3)	<.0001
Preexisting benign breast disease	34 (3.7)	12,430 (2.3)	0.004

Abbreviations: IQR, interquartile range; SD, standard deviation.

^aEvaluation and management encounters include office/outpatient clinic or consultation visits.

^bHormone replacement therapy assessed only in women 50 years and older or in younger women with evidence of surgical menopause.

Multivariable and Bivariable Models for Risk of Invasive Breast Cancer

Figure 1 provides the results of the bivariable and multivariable analyses of the outcome of invasive breast cancer in women treated with pooled antipsychotic drugs (categories 1–3) compared to women treated only with anticonvulsants/ lithium (for full model results see Supplemental Table 3). Antipsychotic treated women had a higher overall risk of breast cancer than anticonvulsant/ lithium users (aHR 1.40; 95% CI 1.19–1.64), with the hazards ratio corresponding to a one-unit increase in the average daily DDD for antipsychotic agents. There was little change in the risk of breast cancer associated with exposure to the combined categories of antipsychotics after adjustment for the known breast cancer risk factors enumerated in Supplemental Table 3 (with only known risk factors but not age, aHR 1.45; 95% CI, 1.23–1.70).The risk of breast cancer associated with exposure to the antipsychotic agents was slightly lower after adjustment for age alone (aHR 1.30; 95% CI, 1.08–1.55), and after adjustment for age plus known risk factors (aHR 1.35; 95% CI, 1.14–1.61). The HRs for breast cancer associated with the pooled antipsychotic agents were similar in the sensitivity analyses stratified by age (Supplemental Tables 4 and 5).

Figure 1. Bivariate and Multivariable Analyses of Risk of Invasive Breast Cancer in Women Treated with Pooled Antipsychotic Agents, Compared to Women Treated Only with Anticonvulsants or Lithium.

Known risk factors for breast cancer included in the model were as follows: estrogen/progesterone-based hormone replacement therapy, estrogen-only hormone replacement therapy, diabetes, obesity, alcohol abuse, and preexisting benign breast disease. Medicaid enrollment was included in the model as a proxy for younger age at first birth and parity. Mental health diagnoses included were bipolar disorder, schizophrenia, and major depression. When noted, age was included using a cubic spline with seven knots. The HR represents the increase in breast cancer risk per one defined daily dose (DDD) in average daily DDD to the pooled antipsychotic medications.

Figure 2 shows results of analyses considering exposure to antipsychotic categories 1–3 individually (for full model results see Supplemental Table 6). In bivariable analysis, category 1 (high prolactin drugs) and category 2 drugs (mid prolactin) were associated with significantly increased risk of breast cancer: category 1 drugs (HR 1.50; 95% CI, 1.25–1.81) and category 2 drugs (mid prolactin), (HR 1.65; 95% CI, 1.25–2.17). However, category 3 (low prolactin) drugs were not associated with significantly increased risk of breast cancer (HR 1.11; 95% CI, 0.83–1.50). The results were similar when adjusted for known risk factors: category one (aHR 1.63; 95% CI, 1.38–1.92), category 2 (aHR 1.69; 95% CI, 1.32–2.18), and category 3 drugs (aHR 1.09; 95% CI, 0.81–1.47). The significant associations of drug categories 1 and 2 remained after adjustment for known risk factors plus mental health conditions; category 1 (aHR 1.59; 95% CI, 1.34–1.90), category 2 (aHR 1.70; 95% CI, 1.32–2.19) and also remained after adjustment for age plus known risk factors: category 1 drugs,(aHR 1.62; 95% CI, 1.30–2.03) and category 2 drugs, (aHR 1.54; 95% CI, 1.19–1.99). The category 3 drugs observed null finding remained after adjustment. We also conducted sensitivity analyses stratified by age (18–50 years and 51–64 years). This revealed similar findings for category one and category 3 drugs as in the primary analyses, but a higher risk of breast cancer for younger women aged 18–50 taking category 2 drugs adjusted for known risk factors plus age, (aHR 1.91; 95% CI 1.21–3.01) compared to women 51–64 years, (aHR 1.43; 95% CI 1.01–2.03) (Supplemental Tables 7 and 8).

Figure 2. Bivariate and Multivariable Analyses of Risk of Invasive Breast Cancer in Women Treated with Antipsychotic Agents by Category of Propensity to Elevate Prolactin, Compared to Women Treated Only with Anticonvulsants or Lithium.

Known risk factors for breast cancer included in the model were as follows: estrogen/progesterone-based hormone replacement therapy, estrogen-only hormone replacement therapy, diabetes, obesity, alcohol abuse, and preexisting benign breast disease. Medicaid enrollment was included in the model as a proxy for younger age at first birth and parity. Mental health diagnoses included were bipolar disorder, schizophrenia, and major depression. When noted, age was included using a cubic spline with seven knots. The HR represents the increase in breast cancer risk per one defined daily dose (DDD) in average daily DDD to the individual category of antipsychotic medications.

Discussion

In the largest observational study evaluating risk of breast cancer in U.S. women (18–64 years old) taking antipsychotic drugs, we observed that women who filled prescriptions for prolactin-elevating category 1 and 2 antipsychotic drugs had significantly elevated (54–62%) risk of breast cancer per one-unit increase in the defined daily dose, compared to women with prescriptions only for comparator medications (anticonvulsants and/or lithium), after adjusting for age and other known breast cancer risk factors. This is the first report of elevated breast cancer risk in U.S. women taking commonly prescribed newer antipsychotic drugs. Many of the drugs in category 2 are considered to have moderate prolactin elevating potential, yet were still associated with higher breast cancer risk. Given the sporadic duration of use of antipsychotic medications in this real-world population, it may be more informative to instead compare the increased risk associated with moving from the 25^th percentile to the 75^th percentile of DDD drug exposure. For the category 1 antipsychotic medications, the increased risk associated with moving from the 25^th to 75^th percentile (0.01 to 0.11 DDD) of utilization would result in a 6.2% increased risk of invasive breast cancer in a maximum 6 years of follow-up. Similarly, for the same increased utilization of category 2 antipsychotics medications (0.03 DDD for the 25^th percentile to 0.30 DDD for the 75^th percentile) the associated increase in breast cancer in this population would be 14.6% increased risk.

Study Strengths

The study included a large population of women and utilized three separate antipsychotic categories with an active comparator control group (treated with only anticonvulsants and/or lithium). By using commonly prescribed psychotropic drugs as active comparators, a comparison of populations with a likely psychiatric diagnoses was done to reduce confounding by indication.³⁸ We controlled for known risk factors for invasive breast cancer available in the administrative claims data, and used a rigorous approach to model exposure to the drugs by standardizing all drugs to DDD and calculating the average daily exposure to the individual drugs over the observed treatment period with a maximum follow-up of 6 years. We observed that Category 1 (highest potential to elevate prolactin) and category 2 drugs were associated with significant risk, while category 3 drugs with no known potential to elevate prolactin had no association with increased risk of breast cancer. There were sufficient numbers of patients in each drug category. The data are consistent with prior studies with similar designs that suggest the possibility that antipsychotic drugs contribute to breast cancer risk, but also builds on prior research by suggesting that this risk may be limited to drugs with higher propensity to elevate prolactin.^2–4,6 The null finding of category 3 drugs provides further evidence that the observed increased risk of breast cancer in categories 1 and 2 may be due to elevated serum prolactin levels.

Study Limitations

The median duration of follow up in each group was ~4 years. It is plausible that it might take many years for breast cancer to develop in the setting of elevations in serum prolactin, thus additional studies with longer duration of follow-up are warranted to verify our findings. The average daily drug category exposure during the study observation period was used to quantify exposure due to the intermittent utilization of antipsychotic mediations. We required 12 months of insurance enrollment without a study drug to establish a “new user” cohort, however given episodic use of these drugs and the average duration of insurance enrollment, it is possible that some women included in our study as “new users” had prior experience with study drugs. Although we were unable to control for residual confounding due to parity, menopausal status, family history of breast cancer, alcohol, sedentary lifestyle, diet, obesity, age at first birth, and race, adjustment for diagnosed mental health conditions may have reduced residual confounding due to co-occurrence of some of these conditions. Actual serum prolactin levels may not be reflective of the three antipsychotic drug categories and all patients taking antipsychotics may not have the hypothesized prolactin response.

The analysis population did not include women aged 65 and older, since our version of the Commercial and Multi-state Medicaid databases are limited to persons younger than 65 years. Although the number of outpatient evaluation and management visits in the baseline period were very similar in women with and without breast cancer, the lack of clinical detail does not allow us to rule out some degree of ascertainment bias. It is also possible that clinicians could favor one antipsychotic drug category over another depending on severity of mental illness, insurance coverage, etc. and that those other factors might be associated with confounding risk factors. Despite these limitations, the adjustments we performed resulted in associations consistent with the literature concerning known risk factors for invasive breast cancer.³⁹ Finally, we must consider important potential confounding when comparing women taking anticonvulsants or lithium to those taking antipsychotics. Epidemiologic data regarding parity in women with severe mental illness is lacking. However, we note that one study reported a lower parity rate in women with bipolar disorder,⁴⁰ and another study reported that women with schizophrenia had similar parity to the general population.⁴¹ This strengthens our observed findings, as higher parity exerts a protective effect on breast cancer risk.

Clinical Implications

Mentally ill patients face disparities in screening/ prevention of common cancers and are more likely to die from breast cancer, and women with psychotic disorders are about half as likely as the general population to receive mammography screening.⁴² Our data and the other publications reviewed above suggest that the use of antipsychotic drugs that elevate prolactin may contribute to incident cases of breast cancer. We also report a surprisingly elevated association of breast cancer risk with category 2 drugs modestly elevate prolactin, particularly in younger women, aged 18–50 years. We note that there are two new European studies that found evidence of increased breast cancer risk with cumulative dose exposure with high prolactin drugs. A nationwide Danish cohort study of all residents and their use of antipsychotics for up to 20 years found a low risk for breast cancer with first and second generation antipsychotics compared with women with breast cancer who did not receive such prescriptions (HR 1.18; 95% CI 1.06–1.32). A dose response pattern was seen in the Danish study with slightly stronger associations with higher cumulative exposure.¹³ A Finnish nationwide study of women with schizophrenia observed that exposure to prolactin elevating vs prolactin sparing antipsychotics for 5 or more years (but not 1–4 years) was associated with an increased breast cancer risk (aOR 1·56; 95% CI 1·27–1·92). Our findings in U.S. women provide observational support that increased circulating prolactin is likely the plausible biological mechanism for the observed increased risk of breast cancer.⁴³

Any risks of all medications must be weighed against potential benefits. Antipsychotic drugs have life-saving properties and should not necessarily be avoided due to potential risk for breast cancer.¹⁰ Our findings contribute additional knowledge of the safety profile of antipsychotic medications and may be useful to patients considering use of these medications. Further prospective studies aimed at studying predictive biomarkers (prolactin levels, prolactin gene expression and mammographic breast density) are needed to design appropriate intervention studies for women taking antipsychotic drugs. In the meantime, we recommend vigilance to mitigate breast cancer risk by monitoring serum prolactin levels, particularly in those treated with category 1 and 2 drugs, including in younger women. Moreover, women at high risk for breast cancer who are prescribed category 1 or 2 drugs may be considered for more frequent mammography screening. The study findings of elevated risk of breast cancer in younger women raises questions regarding the long term effects of antipsychotics in children and adolescents. Somewhat reassuring is that hyperprolactinemia can be mitigated by lowering antipsychotic doses, switching antipsychotics to a category 3 drug, adding dopamine agonists, or by adding a low dose of aripiprazole.^6,29 Further development of novel antipsychotics that are prolactin sparing or studies of combination strategies with dopamine partial agonists (particularly with long acting injectable drugs) are needed.

In conclusion, psychiatrists should utilize a patient-centered approach to breast cancer risk reduction by carefully considering the totality of risks and benefits of antipsychotic drugs in female patients that are started and maintained on them. Further prospective investigation of antipsychotic related breast cancer risk, including studies of prolactin effects on breast tissue density (a strong correlate for breast cancer risk) are needed. Clinicians should consider factors such as genetic mutations (BCRA), high breast tissue density, family history, parity and exogenous hormone use when treating patients needing antipsychotics.¹ We agree with the conclusions from a recent Finnish study that antipsychotics without prolactin-increasing properties should be considered as a first-line long-term treatment for women.⁴³ Many prolactin elevating dopamine antagonists, including anti-emetics⁴ are used for a wide variety of general hospital care issues including chemotherapy induced nausea as well as for agitation and delirium. It is reassuring that short term use of antipsychotics has not been reported to increase breast cancer risk.^13,43 For long term use, baseline and maintenance dose monitoring of prolactin concentrations (along with the usual lipid, BMI and metabolic parameters) should be considered⁴³ and, in cases of hyperprolactinemia, switching to a category 3 drug or mitigating high prolactin with concomitant drugs (aripiprazole/ brexpiprazole²⁹ or dopamine agonists) can also be considered to mitigate breast cancer risk.

Figure 3. Bivariate and multivariable analyses of risk of invasive breast cancer in women treated with antipsychotic agents by category of propensity to elevate prolactin, compared to women treated only with anticonvulsants or lithium.

Known risk factors for breast cancer included in the model were as follows: estrogen/progesterone-based hormone replacement therapy, estrogen-only hormone replacement therapy, diabetes, obesity, alcohol abuse, and preexisting benign breast disease. Medicaid enrollment was included in the model as a proxy for younger age at first birth and parity. Mental health diagnoses included were bipolar disorder, schizophrenia, and major depression. When noted, age was included using a cubic spline with seven knots. The HR represents the increase in breast cancer risk per 1 DDD in average daily DDD to the individual category of antipsychotic medications.

Table 3.

Bivariate Analysis of Factors During the Baseline Period Associated with Invasive Breast Cancer in Women with at least One Filled Prescription for Antipsychotic, Anticonvulsant, and/or Lithium Medications in the MarketScan Commercial and Multi-State Medicaid Databases

Characteristic	Invasive Breast Cancer N = 2,708 (0.1) n (%)	No Breast Cancer N = 1,560,131 (99.9) n (%)	P
Age in years, median (IQR)	53 (46, 58)	41 (30, 51)	<.0001
Number of Evaluation & Managementa encounters during baseline period, median (IQR)	3 (1, 6)	3 (1, 6)	<.0001
At least 1 Evaluation & Management encounter during baseline (%)	2,290 (84.6)	1,266,234 (81.2)	<.0001
Mental health conditions
Bipolar disorder	340 (12.6)	245,069 (15.7)	<.0001
Schizophrenia	141 (5.2)	60,677 (3.9)	0.0004
Major depression	800 (29.5)	472,411 (30.3)	0.40
Medicaid enrollment	354 (13.1)	313,534 (20.0)	<.0001
Hormone replacement therapyb	375 (13.8)	82,003 (5.3)	<.0001
Estrogen/progesterone-based HRT	67 (2.5)	10,549 (0.7)	<.0001
Estrogen-only HRT	308 (11.4)	71,454 (4.6)	<.0001
Obesity	1,289 (47.6)	573,689 (36.8)	<.0001
Diabetes	338 (12.5)	137,312 (8.8)	<.0001
Alcohol abuse	77 (2.8)	70,204 (4.5)	<.0001
Preexisting benign breast disease	59 (2.2)	18,684 (1.2)	<.0001

Abbreviations: IQR, interquartile range; SD, standard deviation.

^aEvaluation and management encounters include office/outpatient clinic or consultation visits.

^bHormone replacement therapy assessed only in women 50 years and older or in younger women with evidence of surgical menopause.