Impact of major depressive disorder on breast cancer outcomes: a national retrospective cohort study

Abstract

Background

Establishing whether women with major depressive disorder who develop breast cancer have poor outcomes is key to optimizing care for this population. To this end, we examined associations between major depressive disorder and breast cancer recurrence and mortality.

Methods

Using medical record data from the US Department of Veterans Affairs health-care system, we established a retrospective cohort of women with local or regional stage invasive breast cancer between 2010 and 2019 and followed them through 2022. We used a 2-year window to identify women diagnosed with major depressive disorder before breast cancer diagnosis. We used multivariable Cox-proportional hazards regression to estimate associations between major depressive disorder and breast cancer recurrence and mortality while accounting for competing risks and adjusting for sociodemographic, clinical, lifestyle, and tumor characteristics.

Results

We identified 6051 women with breast cancer, of whom 1754 (29%) had major depressive disorder. The mean (SD) age at breast cancer diagnosis was 57 (11) years. In multivariable analyses, women with major depressive disorder had a 37% (hazard ratio = 1.37, 95% CI = 1.19 to 1.57) higher risk of recurrence and a 30% (hazard ratio = 1.30, 95% CI = 1.02 to 1.64) higher risk of breast cancer mortality. The association between major depressive disorder and recurrence was stronger among women with estrogen receptor–positive breast cancer. In secondary analyses, there were statistically significant interactions between major depressive disorder and multiple exposures with respect to recurrence, including current smoking, substance abuse, and nonreceipt of screening mammography.

Conclusions

Women with major depressive disorder had inferior breast cancer outcomes compared with women without a history of major depressive disorder. Research is needed to investigate underlying mechanisms linking depression to breast cancer progression and evaluate interventions to improve outcomes in this high-risk population.

Introduction

Depression is reported in up to 60% of women both before breast cancer diagnosis and throughout the disease course, and it affects women of all races and ages.^1-4 Despite the high prevalence of depression, the direction and magnitude of the relationship between depression and breast cancer outcomes still needs to be clearly defined.⁵ A few observational studies have reported that women with breast cancer and depression have risks of breast cancer recurrence and mortality that are up to 2 times higher than women with breast cancer alone, which is concerning,^2,3,6,7 although other studies have reported null or negative associations.^3,4,6

Several limitations in previous research may have contributed to these inconsistent findings. The few studies that measured the impact of a depression diagnosis before developing breast cancer on recurrence or mortality lacked statistical power due to small sample sizes and limited numbers of events.^2-4,6,7 These studies also had limited data on potential confounders (ie, factors that influence both depression and breast cancer outcomes), such as sociodemographic, clinical, lifestyle, and prognostic factors, which could lead to an underestimation or overestimation of the results.^2-4,6,7 Further, the ascertainment of depression is inconsistent across studies because many health-care systems do not routinely screen for depression or evaluate depression severity.^2,3,6,7 We aimed to address some of these limitations in the current study.

In this retrospective cohort study, we investigated associations between a history of major depressive disorder and both recurrence and mortality among women with breast cancer in the United States Department of Veterans Affairs (VA) health-care system. We hypothesized that women with prediagnostic major depressive disorder would experience worse breast cancer outcomes compared with women without prediagnostic major depressive disorder. We chose to study major depressive disorder given that it is a more severe form of clinical depression and a relatively homogeneous condition with well-established diagnostic criteria. The VA health-care setting provided a unique opportunity to study these associations because of its mandated annual depression screening, systematic evaluation of major depressive disorder by health-care professionals, and ability to provide equal-access health-care services. These features allowed for comprehensive data collection and long-term patient follow-up.^8,9

Methods

Data source

The VA is the largest nationally integrated and equal-access health-care system in the United States, with more than 170 medical centers and 1250 community-based outpatient clinics. It provides comprehensive medical care to more than 9 million veterans, all of whom receive health insurance and other accommodations.^10,11 Veterans enroll in VA health insurance upon retirement, and they are unlikely to go outside the VA for health-care services, particularly for treatment of chronic diseases such as cancer and major depressive disorder.^10,11 This tendency allows the VA to follow patients closely for recurrence and mortality.

The VA has an electronic health record (EHR) system with a centralized corporate data warehouse containing longitudinal information about sociodemographic and lifestyle factors and receipt of all inpatient and outpatient services provided by VA facilities nationally. Some of this information is available as structured data elements. Information about all patients with cancer occurring within the VA health-care system, including their treatments, is mandated to be reported to the VA Central Cancer Registry (VACCR). Each VA facility uses cancer registrars to manually abstract data on cancer diagnoses, characteristics, treatments, and outcomes (eg, cancer recurrence) from pathology and clinical notes in the EHRs of identified patients with cancer using OncoTraX software in compliance with data requirements set by the North American Association of Central Cancer Registries.¹² Vital status (ie, alive or date of death, if applicable) is available from the VA Vital Status File, and cause of death is available from the joint VA–US Department of Defense Suicide Data Repository, which matches VA decedents to the National Death Index.¹³ Mortality data from the Veterans Health Administration Vital Status File is estimated to be 99% complete and accurate at the time of use.¹⁴

Study design and population

From the VACCR, we assembled a retrospective cohort of all VA-enrolled women (aged 18 years and older) with newly diagnosed local or regional invasive breast cancer (based on International Classification of Diseases, Ninth Revision [ICD-9] codes 174.0-174.9 or International Statistical Classification of Diseases, Tenth Revision [ICD-10] codes C50.0-C50.9) between 2010 and 2019 at a VA facility nationally (Figure 1). Women were indexed at breast cancer diagnosis and followed up through 2022. We excluded women who were diagnosed with breast cancer after 2019 due to the possibility of incomplete data resulting from health-care access challenges during the COVID-19 pandemic or who had a prior cancer diagnosis at the VA within the past 10 years. To ensure that women were active users of VA medical services, we required them to have at least 1 health-care encounter at a VA facility within 2 years before their breast cancer diagnosis (baseline period). Since 1998, patients accessing care at the VA are screened with standardized depression questionnaires annually (eg, Patient Health Questionnaire-2, Patient Health Questionnaire-9, Beck Depression Inventory); patients with a positive screen are evaluated for major depressive disorder by a clinician.^8,9 The VA reports screening rates nationally of 98%, thereby reducing the risk of undiagnosed major depressive disorder.^8,9

Figure 1.

Flowchart of patient selection for women with early-stage breast cancer in the US Veterans Health Administration (2010-2019). Abbreviation: VA = US Department of Veterans Affairs.

Institutional review board approval

This project was approved by the Veteran’s Institutional Review Board of the White River Junction VA Medical Center in White River Junction, Vermont. All study procedures were carried out in compliance with federal and institutional ethical guidelines. The requirement to obtain informed consent from study participants was waived because the Institutional Review Board deemed this study to involve no more than a minimal risk to the privacy of individuals.

Primary exposure

Our primary exposure was clinically diagnosed major depressive disorder during the 2-year baseline period before breast cancer diagnosis. The depression screening ensured that all patients were assessed for depression. We identified women with a major depressive disorder diagnosis by using a validated algorithm to extract major depressive disorder diagnoses from medical records based on ICD-9 (codes 296.2x and 296.3x) or ICD-10 (codes F32.0–F32.5, F32.9, and F33.x).^15,16 Women are classified as having a major depressive disorder diagnosis if they had at least 1 inpatient or 2 outpatient visits with a major depressive disorder diagnosis code during the 2 years before breast cancer diagnosis.^15,16

Covariates

Information about year of breast cancer diagnosis, tumor characteristics, and receipt of initial cancer treatment for the primary breast cancer came from the VACCR. Tumor characteristics included cancer stage, estrogen receptor status, progesterone receptor status, and ERBB2 (formerly HER2) status. Initial cancer treatments included surgery, radiation, chemotherapy, hormone therapy, and immunotherapy. We also classified breast cancers into luminal A-like (estrogen receptor–positive and/or progesterone receptor–positive/ERBB2-negative), luminal B-like (estrogen receptor–positive and/or progesterone receptor–positive/ERBB2-positive), triple-negative (estrogen receptor–negative/progesterone receptor–negative/ERBB2-negative), and ERBB2-positive (estrogen receptor–negative/progesterone receptor–negative/ERBB2-positive) subtypes.¹⁷

We included additional covariates that could influence major depressive disorder and breast cancer outcomes a priori based on published literature. For static variables (eg, race, ethnicity), we used all available data from EHRs. All other covariates were ascertained from EHRs within 2 years before breast cancer diagnosis. Clinical and lifestyle variables included patients’ smoking status, substance abuse (ICD-9 codes 291.x, 303.x-305.0, and 305.2-305.9; ICD-10 codes F10.x-F16.x, and F18.x-F19.x), body mass index (BMI), receipt of screening mammography (Current Procedural Terminology code 77067), and comorbidities based on the Charlson Comorbidity Index.¹⁸ For comorbidities, we identified active conditions by applying a validated algorithm of ICD-9 or ICD-10 codes to extract diagnoses for each Charlson Comorbidity Index condition from EHRs.^15,16 This approach minimizes potential misclassification by excluding inactive or historical conditions that may appear on patients' problem lists but may no longer inform their current health status. Sociodemographic variables included patients’ age, race, Hispanic ethnicity, marital status, rurality of primary residence, any problems related to housing and economic circumstances (eg, homelessness or inadequate housing; low income or poverty; inadequate food, water, or other material/social resources; ICD-9 code V60.x; ICD-10 code Z59.x), and VA priority group rating.^19-22 The VA priority group ratings were an indirect proxy for socioeconomic status because it is a composite measure based on patients’ income, financial security, Medicaid eligibility, receipt of VA assistance benefits (eg, pension or assisted living), and capacity for gainful employment.^20,23 The ratings also account for health-related factors, including severity of service-connected disabilities.^20,23 The ratings range from 1 to 8, and the VA assigns a lower rating a higher priority.^20,23

Breast cancer outcomes

Our outcomes of interest were breast cancer recurrence and mortality. Local recurrence, defined as recurrence in the same breast, or distant recurrence is abstracted biannually from pathology and clinical notes in patients’ EHRs by the VACCR. Cause and date of death were determined from death certificates using both the underlying cause and all causes of death (ICD-10 cause-of-death code C50).²⁴

Statistical analysis

We used descriptive statistics to summarize patients’ sociodemographic, clinical, and cancer characteristics overall and by prediagnostic major depressive disorder status. We defined patients’ survival time as the years from breast cancer diagnosis (index date) to the first event (death or recurrence) or censoring, whichever occurred first. Women who died from causes other than breast cancer, who developed a second primary cancer before the first outcome event, or who were alive at study end were treated as censored observations for both outcomes. We used multivariable Cox proportional hazards regression to estimate adjusted, cause-specific hazard ratios (HRs) with 95% confidence intervals (CIs) for the associations between major depressive disorder status and both recurrence (local and distant) and mortality while accounting for competing risks through censoring. Our primary goal was to estimate the relative effect of prediagnostic major depressive disorder on breast cancer outcomes, which is best suited to cause-specific hazard modeling. We used complementary log-log plots of the estimated survival function for each outcome and Schoenfeld residuals test to confirm that the underlying proportional hazards assumption was met. To examine the extent of potential bias introduced by censoring competing events, we repeated our analyses using Fine and Gray competing-risks regression. The results from our Cox models were identical to those produced by Fine and Gray models, suggesting that censoring competing events did not introduce significant bias in our study. Therefore, we opted to present findings from our Cox models.

To determine the independent associations between major depressive disorder and each breast cancer outcome and to identify the most influential set of covariates, we developed 4 models, with progressive adjustment for sociodemographic, clinical, lifestyle, and tumor characteristics. Adjustment for prognostic factors (ie, tumor characteristics) primarily served to improve the precision of our estimates by accounting for the strong influence these variables have on the outcomes. Model 1 included patients’ age and year of breast cancer diagnosis. Model 2 included model 1 variables and all tumor characteristics. Model 3 included model 1 variables and all sociodemographic, clinical, and lifestyle characteristics. Model 4 (fully adjusted) included all variables from models 1, 2, and 3. We modeled age and BMI as continuous variables to maintain the most granular level of data.

We encountered missing data for smoking status (19%), estrogen receptor status (15%), progesterone receptor status (15%), and ERBB2 status (20%). Therefore, we used multiple imputation, with chained equations to impute these missing data values. Before imputation, we compared observed data patterns with those expected under the missing-at-random assumption using the Little Missing Completely at Random test.²⁵ In addition, we fitted logistic regression models for each variable with missing data, where the presence of missingness was regressed on other covariates in the dataset. This approach enabled us to examine whether the missing data were associated with the observed variables, thereby supporting the missing-at-random assumption. We performed 20 imputations based on literature recommendations for moderate levels of missing data, which suggested that the number of imputations should be at least equal to the percentage of missing data to achieve stable estimates.²⁵ Each imputation model included all covariates, including outcome variables, to improve the accuracy of imputed values (ie, ensuring that the imputed values maintained the same statistical properties as the observed values).²⁵ Following imputation, we reassessed the distribution of imputed and observed values for smoking status and tumor biomarkers to verify that the imputation produced plausible values. We analyzed each of the 20 imputed datasets separately, and then pooled the results using Rubin rules.²⁵ This method combines parameter estimates and SEs across the imputed datasets, accounting for variability both within and between imputations to ensure the robustness of our final models. Finally, we conducted sensitivity analyses using complete cases to evaluate the robustness of our imputation results.

Reported P values were based on 2-sided tests and were considered statistically significant if P  was less than .05. We used Stata/MP, version 18.0, software (StataCorp LP) for all statistical analyses.

Secondary analysis

We used model 4 (fully adjusted) from our primary analysis to estimate the independent and joint effects of major depressive disorder and each variable using a single reference group. For the interactions evaluated, we statistically tested any significant departures from multiplicativity by including an interaction term in the model (Wald test). To examine potential interactions between major depressive disorder and BMI, we categorized BMI based on the US Centers for Disease Control and Prevention’s guidelines (BMI <25 kg/m², BMI 25-29 kg/m², BMI ≥30 kg/m²).²⁶ To examine potential interactions between major depressive disorder and age, we categorized age as younger than 50 years or 50 years of age or older, given that 50 years of age is generally considered the average age of menopause in the United States, which can significantly influence breast cancer outcomes.

Sensitivity analysis

We performed 3 sensitivity analyses to understand potential bias arising from misclassification of depression and differences in major depressive disorder severity. First, some women in our unexposed group may have screened positive for depressive symptoms on standardized depression questionnaires without receiving a formal major depressive disorder diagnosis. These cases may represent milder forms of depression that did not meet the diagnostic criteria for major depressive disorder. To understand potential bias arising from misclassification of depression in our unexposed group, we excluded women who screened positive on depression questionnaires but did not receive a clinical diagnosis of major depressive disorder. Second, it is possible that the measurement and rates of chronic conditions may have been affected by the transition from the ICD-9 to ICD-10 coding system in 2015. ICD-10 codes have greater specificity than ICD-9 codes do and can improve the accuracy of measuring diagnoses. To better understand whether there was misclassification of depression related to the transition from ICD-9 to ICD-10 coding, we performed stratified analyses before (2010-2014) and after (2015-2019) the transition to ICD-10 coding. Third, patients who received care in an inpatient setting may have more severe major depressive disorder than patients who received care in an outpatient setting. To address potential differences in major depressive disorder severity, we further distinguished major depressive disorder diagnoses based on the setting in which these diagnoses were made (ie, outpatient visits for major depressive disorder only or both outpatient and inpatient visits for major depressive disorder).

Results

Baseline characteristics

Descriptive characteristics of the study population are outlined in Table 1. We identified 6051 women with early-stage invasive breast cancer, of whom 1754 (29%) had prediagnostic major depressive disorder. The average (SD) age at breast cancer diagnosis was 57 (11) years overall and 56 (10) years among women with major depressive disorder. Sixty-eight percent of tumors were localized, 94% were estrogen receptor positive, and 84% were luminal A. Nearly 90% of women received any cancer treatment in the VA health-care system following their diagnosis. Eighty-four percent of patients underwent surgery.

Table 1.

Patient characteristics among female veterans with early-stage breast cancer in the US Veterans Health Administration, by prediagnostic major depressive disorder status (2010-2019)

Characteristic	No major depressive disorder	Major depressive disorder	Overall
	(n = 4297)	(n = 1754)	(N = 6051)
Sociodemographics
Age at diagnosis, mean (SD), y	57 (11)	56 (10)	57 (11)
Race, No. (%)
Black	1293 (30.1)	512 (29.2)	1805 (29.8)
White	2856 (66.5)	1194 (68.1)	4050 (66.9)
Othera	148 (3.4)	48 (2.7)	196 (3.2)
Hispanic ethnicity, No. (%)	202 (4.7)	91 (5.2)	293 (4.8)
Married/long-term partner, No. (%)	1209 (28.1)	447 (25.5)	1656 (27.4)
Rural primary residence, No. (%)	1191 (27.7)	481 (27.4)	1672 (27.6)
Any homelessness or economic problems, No. (%)	220 (5.1)	266 (15.2)	486 (8.0)
Priority group rating, No. (%)
0-4	1333 (31.0)	722 (41.2)	2055 (34.0)
5-6	1268 (29.5)	561 (32.0)	1829 (30.2)
7-8	1696 (39.5)	471 (26.9)	2167 (35.8)
Clinical characteristics
Charlson Comorbidity Index, mean (SD)	0.51 (1.0)	0.75 (1.1)	0.58 (1.0)
BMI (kg/m2) at diagnosis, median (IQR)	28.5 (6.4)	30.7 (9.2)	29.0 (7.4)
Smoking status, No. (%)
Current	1055 (24.5)	604 (34.4)	1659 (27.4)
Former	772 (18.0)	402 (22.9)	1174 (19.4)
Never	2470 (57.5)	748 (42.7)	3218 (53.2)
SUD, No. (%)	839 (19.5)	670 (38.2)	1509 (24.9)
Screening mammography, No. (%)b	1629/3257 (50.0)	715/1318 (54.3)	2344/4575 (51.2)
Cancer characteristics, No. (%)
Cancer stage
Localized	2894 (67.4)	1202 (68.5)	4096 (67.7)
Regional	1403 (32.7)	552 (31.5)	1955 (32.3)
Estrogen receptor positive	4056 (94.4)	1641 (93.6)	5697 (94.2)
Progesterone receptor positive	3977 (92.6)	1605 (91.5)	5582 (92.3)
ERBB2 positive	510 (11.9)	227 (12.9)	737 (12.2)
Molecular subtype
Estrogen receptor+/progesterone receptor+/ERBB2‒	3570 (83.1)	1423 (81.1)	4993 (82.5)
Estrogen receptor+/progesterone receptor+/ERBB2+	487 (11.3)	220 (12.5)	707 (11.7)
Estrogen receptor‒/progesterone receptor‒/ERBB2‒	182 (4.2)	85 (4.9)	267 (4.4)
Estrogen receptor‒/progesterone receptor‒/ERBB2+	58 (1.4)	26 (1.5)	84 (1.4)
Cancer treatment (initiated)
Surgery	3631 (84.5)	1469 (83.8)	5100 (84.3)
Chemotherapy	1466 (34.1)	634 (36.2)	2100 (34.7)
Radiation	1885 (43.9)	775 (44.2)	2660 (44.0)
Hormone therapy	1949 (45.5)	806 (46.0)	2755 (45.5)
Immunotherapy	225 (5.9)	102 (5.8)	357 (5.9)
Any cancer treatment	3714 (86.4)	1512 (86.2)	5226 (86.4)
Year of cancer diagnosis
2010-2011	727 (16.9)	211 (12.0)	938 (15.5)
2012-2013	950 (22.1)	271 (15.5)	1221 (20.2)
2014-2015	962 (22.4)	297 (16.9)	1259 (20.8)
2016-2017	840 (19.6)	444 (25.3)	1284 (21.2)
2018-2019	818 (19.0)	531 (30.3)	1349 (22.3)
Breast cancer outcomes
Breast cancer recurrence, No. (%)	641 (14.9)	322 (18.4)	963 (15.9)
Time to cancer recurrence, median (IQR), y	2.6 (3.9)	2.1 (3.4)	2.4 (3.7)
Follow-up time, median (IQR), y	5.0 (4.7)	3.7 (4.3)	4.6 (4.7)
Breast cancer mortality, No. (%)	258 (6.0)	114 (6.5)	372 (6.2)
Time to death, median (IQR), y	3.3 (3.8)	2.7 (2.4)	3.0 (3.6)
Follow-up time, median (IQR), y	5.5 (4.7)	4.2 (4.4)	5.0 (4.7)

Abbreviations: BMI = body mass index; SUD = substance use disorder.

a

“Other” race includes individuals who identify as Asian, American Indian or Alaska Native, or Native Hawaiian or Other Pacific Islander.

b

Screening mammography estimated among women aged 50 years or older.

Women with prediagnostic major depressive disorder had a higher average Charlson Comorbidity Index score (mean [SD] = 0.75 [1.1] vs 0.51 [1.0]) compared with women without major depressive disorder. In addition, a higher proportion of women with major depressive disorder experienced homelessness or economic problems (15% vs 5%), reported currently smoking (34% vs 25%), and had a substance use disorder (SUD; 38% vs 20%). There were no differences in other patient characteristics by major depressive disorder status.

There were a total of 963 breast cancer recurrences and 372 breast cancer deaths during the follow-up period. The median (IQR) length of follow-up from breast cancer diagnosis to recurrence or censoring was 3.7 (4.3) years among women with major depressive disorder and 5.0 (4.7) years among women without major depressive disorder. The median (IQR) length of follow-up from breast cancer diagnosis to death or censoring was 4.2 (4.4) years among women with major depressive disorder and 5.5 (4.7) years among women without major depressive disorder. The median (IQR) time to recurrence (2.1 [3.4] years vs 2.6 [3.9] years) and mortality (2.7 [2.4] years vs 3.3 [3.8] years) was shorter among women with major depressive disorder than among women without major depression disorder.

Breast cancer recurrence and mortality

The associations between prediagnostic major depressive disorder and breast cancer recurrence and mortality are shown in Table 2. Major depressive disorder was associated with a 48% (HR = 1.48, 95% CI = 1.30 to 1.70) higher hazard of recurrence compared with women without major depressive disorder after adjusting for age and year of breast cancer diagnosis (model 1). This estimate remained consistent in model 2 (HR = 1.47, 95% CI = 1.28 to 1.68) with the addition of tumor characteristics and treatments. In model 3, major depressive disorder was associated with a 38% (HR = 1.38, 95% CI = 1.19 to 1.58) higher hazard of recurrence compared with women without major depressive disorder after adjusting for model 1 variables and sociodemographic, clinical, and lifestyle characteristics. The model 4 estimate (fully adjusted) was similar to that for model 3 (HR = 1.37, 95% CI = 1.19 to 1.57).

Table 2.

Adjusted hazard ratios for the associations between prediagnostic major depressive disorder and breast cancer recurrence and mortality among women with early-stage breast cancer in the US Veterans Health Administration (2010-2019)

Characteristic	No.	Events/person-years	Model 1a	Model 2b	Model 3c	Model 4d
			HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)
Breast cancer recurrence
No major depressive disorder	4297	641/22 198	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	1754	322/7547	1.48 (1.30 to 1.70)	1.47 (1.28 to 1.68)	1.38 (1.19 to 1.58)	1.37 (1.19 to 1.57)
Breast cancer mortality
No major depressive disorder	4297	258/23 728	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	1754	114/8275	1.36 (1.09 to 1.71)	1.35 (1.08 to 1.69)	1.31 (1.04 to 1.65)	1.30 (1.02 to 1.64)

Abbreviations: HR = hazard ratio; CI = confidence interval.

a

Adjusted for age and year of breast cancer diagnosis.

b

Adjusted for age, year of breast cancer diagnosis, cancer stage, molecular subtype, and cancer treatment.

c

Adjusted for age, year of breast cancer diagnosis, race, Hispanic ethnicity, marital status, priority group rating, rurality of primary residence, any homelessness or economic problems, Charlson Comorbidity Index, smoking status, substance use disorder, body mass index, and receipt of screening mammography.

d

Adjusted for age, year of breast cancer diagnosis, cancer stage, molecular subtype, cancer treatments, race, Hispanic ethnicity, marital status, priority group rating, rurality of primary residence, any homelessness or economic problems, Charlson Comorbidity Index, smoking status, substance use disorder, body mass index, and receipt of screening mammography.

Major depressive disorder was associated with a 36% (HR = 1.36, 95% CI = 1.09 to 1.71) higher hazard of breast cancer mortality compared with women without major depressive disorder after adjusting for age and year of breast cancer diagnosis (model 1). The estimates were similar for models 2 and 3. In the fully adjusted model (model 4), major depressive disorder was associated with a 30% (HR = 1.30, 95% CI = 1.02 to 1.64) higher hazard of mortality.

Secondary analyses

Figure 2, A shows the independent and joint effects of prediagnostic major depressive disorder and tumor characteristics on breast cancer recurrence. Notably, compared with women with localized breast cancer, the joint effect of major depressive disorder and regional disease (HR = 1.65, 95% CI = 1.33 to 2.05) was significantly higher than the independent effect of regional disease alone (HR = 1.15, 95% CI = 0.97 to 1.38), P less than  .001 for interaction. In addition, women with major depressive disorder and estrogen receptor–positive cancer had a significantly higher hazard of recurrence (HR = 1.42, 95% CI = 1.23 to 2.65) compared with women with estrogen receptor–positive cancer alone, P less than .001 for interaction.

Figure 2.

The independent and joint effects of prediagnostic major depressive disorder and tumor characteristics on breast cancer recurrence (A) and mortality (B) among women with early-stage breast cancer in the US Veterans Health Administration (2010-2019). Abbreviation: supr. = suppressed due to too few counts (<11) and/or back-calculation possible.

Figure 2, B shows the independent and joint effects of prediagnostic major depressive disorder and tumor characteristics on breast cancer mortality. The interactions between major depressive disorder, cancer stage (P < .001 for interaction), and estrogen receptor–positive disease (P = .020 for interaction) remained consistent for mortality.

Figure 3, A shows the independent and joint effects of prediagnostic major depressive disorder and sociodemographic, clinical, and lifestyle characteristics on breast cancer recurrence. Notably, current smokers with major depressive disorder demonstrated a 91% (HR = 1.91, 95% CI = 1.55 to 2.37) higher hazard of recurrence than never-smokers without major depressive disorder, whereas current smokers without major depressive disorder exhibited a 33% (HR = 1.33, 95% CI = 1.11 to 1.60) higher hazard of recurrence (P < .001 for interaction). Similarly, women with major depressive disorder and co-existing SUD had a higher hazard of recurrence (HR = 1.63, 95% CI = 1.32 to 2.00) than either exposure independently when compared with women with breast cancer alone (P < .001 for interaction). In addition, women with major depressive disorder who did not undergo screening mammography had a significantly higher hazard of recurrence (HR = 1.62, 95% CI = 1.30 to 2.02) than either exposure independently when compared with women with breast cancer alone who received screening mammography (P < .001 for interaction).

Figure 3.

The independent and joint effects of prediagnostic major depressive disorder and sociodemographic, clinical, and lifestyle characteristics on breast cancer recurrence (A) and mortality (B) among women with early-stage breast cancer in the US Veterans Health Administration (2010-2019). Abbreviation: BMI = body mass index.

Figure 3, B shows the independent and joint effects of prediagnostic major depressive disorder and sociodemographic, clinical, and lifestyle characteristics on breast cancer mortality. The interaction between major depressive disorder and receipt of screening mammography remained consistent for mortality (P = .035 for interaction).

Sensitivity analyses

Findings from our sensitivity analyses are shown in Table 3. Estimates from all 4 models were consistent in both direction and magnitude with estimates from our primary analysis for recurrence and mortality. Notably, our analysis examining differences by prediagnostic major depressive disorder severity revealed a significantly higher hazard of recurrence (model 4 HR = 1.87, 95% CI = 1.48 to 2.36) and mortality (model 4 HR = 1.99, 95% CI = 1.38 to 2.88) among women who had both outpatient and inpatient visits for major depressive disorder compared with women without major depressive disorder.

Table 3.

Adjusted hazard ratios from sensitivity analyses for the associations between prediagnostic major depressive disorder and breast cancer recurrence and mortality among women with early-stage breast cancer in the US Veterans Health Administration (2010-2019)

Characteristic	No.	Events/person-years	Model 1a	Model 2b	Model 3c	Model 4d
			HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)
Breast cancer recurrence
Complete cases
No major depressive disorder	2074	345/10 438	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	854	183/3604	1.54 (1.29 to 1.85)	1.55 (1.29 to 1.86)	1.41 (1.17 to 1.71)	1.43 (1.18 to 1.72)
Excluding self-reported depression
No major depressive disorder	3521	531/18 147	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	1754	322/7547	1.49 (1.30 to 1.72)	1.47 (1.28 to 1.69)	1.35 (1.16 to 1.56)	1.34 (1.15 to 1.55)
Diagnosed 2010-2014
No major depressive disorder	2159	427/15 112	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	632	173/4266	1.47 (1.23 to 1.75)	1.44 (1.20 to 1.72)	1.32 (1.10 to 1.59)	1.30 (1.08 to 1.56)
Diagnosed 2015-2019
No major depressive disorder	2138	214/7086	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	1122	149/3281	1.54 (1.24 to 1.90)	1.55 (1.25 to 1.92)	1.49 (1.19 to 1.86)	1.49 (1.19 to 1.86)
Prediagnostic major depressive disorder severity
No major depressive disorder	4297	641/22 198	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Outpatient visits only	1413	236/6156	1.34 (1.15 to 1.56)	1.32 (1.14 to 1.54)	1.26 (1.08 to 1.47)	1.25 (1.07 to 1.46)
Outpatient and inpatient visits	341	86/1391	2.12 (1.69 to 2.65)	2.08 (1.66 to 2.61)	1.88 (1.49 to 2.38)	1.87 (1.48 to 2.36)
Breast cancer mortality
Complete cases
No major depressive disorder	2074	122/11 310	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	854	59/4026	1.48 (1.08 to 2.02)	1.45 (1.06 to 2.00)	1.42 (1.03 to 1.95)	1.39 (1.00 to 1.92)
Excluding self-reported depression
No major depressive disorder	3521	220/19 438	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	1754	114/8275	1.35 (1.08 to 1.70)	1.34 (1.07 to 1.69)	1.27 (1.01 to 1.62)	1.25 (0.98 to 1.60)
Diagnosed 2010-2014
No major depressive disorder	2159	188/16 372	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	632	69/4767	1.34 (1.02 to 1.77)	1.31 (0.99 to 1.74)	1.24 (0.93 to 1.66)	1.21 (0.90 to 1.63)
Diagnosed 2015-2019
No major depressive disorder	2138	70/7356	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Prediagnostic major depressive disorder	1122	45/3508	1.47 (1.01 to 2.15)	1.51 (1.03 to 2.22)	1.43 (0.96 to 2.13)	1.48 (0.99 to 2.22)
Prediagnostic major depressive disorder severity
No major depressive disorder	4297	258/23 728	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)	1.00 (Referent)
Outpatient visits only	1413	79/6722	1.18 (0.91 to 1.52)	1.17 (0.90 to 1.51)	1.14 (0.88 to 1.48)	1.12 (0.86 to 1.47)
Outpatient and inpatient visits	341	35/1553	2.18 (1.52 to 3.12)	2.22 (1.55 to 3.18)	1.99 (1.38 to 2.86)	1.99 (1.38 to 2.88)

Abbreviations: HR = hazard ratio; CI = confidence interval.

a

Adjusted for age and year of breast cancer diagnosis.

b

Adjusted for age, year of breast cancer diagnosis, cancer stage, molecular subtype, and cancer treatment.

c

Adjusted for age, year of breast cancer diagnosis, race, Hispanic ethnicity, marital status, priority group rating, rurality of primary residence, any homelessness or economic problems, Charlson Comorbidity Index, smoking status, substance use disorder, body mass index, and receipt of screening mammography.

d

Adjusted for age, year of breast cancer diagnosis, cancer stage, molecular subtype, cancer treatments, race, Hispanic ethnicity, marital status, priority group rating, rurality of primary residence, any homelessness or economic problems, Charlson Comorbidity Index, smoking status, substance use disorder, body mass index, and receipt of screening mammography.

Discussion

To our knowledge, this retrospective cohort study in the United States VA health-care system represents the first study to examine the associations between major depressive disorder and breast cancer recurrence among women with early-stage breast cancer. The prevalence of prediagnostic major depressive disorder was high at 29%, and the median time to recurrence was 5 months shorter for women with major depressive disorder than for women without the condition. We observed a 37% increase in recurrence among women with major depressive disorder compared with women without major depressive disorder, after adjusting for key sociodemographic, clinical, lifestyle, and tumor characteristics. The association between major depressive disorder and recurrence was statistically significantly stronger among patients with estrogen receptor–positive disease. Consistent with our recurrence findings, major depressive disorder was associated with a 30% increase in breast cancer mortality. Our secondary analyses revealed novel statistically significant interactions between a major depressive disorder diagnosis and multiple exposures with respect to recurrence, including current smoking, SUD, and nonreceipt of screening mammography.

Prior research on depression and breast cancer mortality has been limited in scope and often lacked adjustment for key prognostic factors, such as cancer stage, molecular subtype, and cancer treatments.^2-4,6,7 Prior research often failed to account for certain lifestyle and behavioral factors that emerged as statistically significant in our study, including current smoking status, substance abuse, and nonreceipt of screening mammography.^2-4,6,7 Therefore, it is difficult to compare our findings with these studies. Nevertheless, risk estimates from prior studies ranged from null to positive (6%-150%).^2-4,6,7 Our study is also unique in its investigation of recurrence, which has historically been difficult to study because of limited long-term follow-up of women with breast cancer and a high loss to follow-up.

One possible explanation for inferior breast cancer outcomes among women with estrogen receptor–positive disease in our study could be low initiation and long-term adherence to effective cancer treatments, including endocrine therapies. Endocrine therapies have been shown to reduce the risk of recurrence, breast cancer–specific, and all-cause mortality by approximately 50%, 33%, and 22%, respectively.^27-30 Although 70% to 80% of women initiate endocrine therapy, 50% to 70% are not adherent by the third year of treatment, depending on the adherence measure (eg, prescription refills, self-report), with depression frequently cited as a risk factor for nonadherence.^31-36 A meta-analysis of 9 small studies revealed that women with self-reported depressive symptoms were 89% more likely to be nonadherent to endocrine therapy compared with their counterparts without depressive symptoms.³³

Our secondary analyses highlight the importance of assessing behavioral and lifestyle factors among women with major depressive disorder. We found that current smokers, women with SUD, and women who did not undergo screening mammography were at elevated risk of breast cancer recurrence and mortality. Although not a strong risk factor for breast cancer, smoking has been associated with biological changes related to tumor progression, such as metabolic abnormalities, cell proliferation, and migration.^37,38 The biological mechanisms by which SUD may affect tumor progression in women with breast cancer warrants further study. Abuse of alcohol and other drugs can affect breast cancer progression through complex biological pathways, including stress-response mechanisms, immune function, and metabolic processes.³⁹ Individuals with SUD may engage in behaviors that promote tumor progression, such as reduced medication adherence.³⁹ The high prevalence of these modifiable risk factors among women with major depressive disorder and breast cancer provides an opportunity for targeted interventions to improve survival, such as smoking-cessation programs, SUD treatment, and efforts to enhance adherence to cancer screening and treatments. Integrating comprehensive mental health support into cancer care could help address these risk factors and improve outcomes for this vulnerable population.

Various biological pathways can be affected by major depressive disorder that may also affect tumor progression. Dysregulation of stress-related neuroendocrine pathways in individuals with major depressive disorder may promote tumor progression through effects on immune function, inflammation, and tumor biology.^40-42 Depression and chronic stress have been associated with alterations in cortisol levels, immune suppression and increased production of proinflammatory cytokines, all of which could contribute to tumor growth and metastasis.^40-42 It is also possible that these biochemical factors may be more pronounced among women with severe major depressive disorder. Our sensitivity analyses indicated a possible dose-response association with major depressive disorder severity whereby women with severe major depressive disorder (evidenced by both inpatient and outpatient visits for major depressive disorder) had nearly double the hazards of breast cancer recurrence and mortality than did women with breast cancer alone or with less severe major depressive disorder (outpatient visits only). Future preclinical and clinical research is needed to understand more about the biological underpinnings of our observations.

Limitations

Although our study contributes valuable insights to the associations between prediagnostic major depressive disorder and breast cancer outcomes, several limitations should be acknowledged. Our reliance on EHRs and cancer registry data alone could potentially affect the accuracy of our findings due to possible incomplete documentation. In addition, although the retrospective cohort design of our study enabled us to establish the temporal sequence of major depressive disorder and breast cancer outcomes, our ability to make strong causal inferences is still limited because this was not a prospective study and we cannot rule out residual or unmeasured confounding in any observational study. Our study was underpowered to assess interactions between major depressive disorder and ethnicity because of the composition of our cohort, where 95% of women identified as non-Hispanic. The lack of ethnic diversity may limit the generalizability of our results. Strengths of our study include the use of a large and well-defined cohort of women with breast cancer and major depressive disorder, long-term follow-up from an equal-access health-care system, robust data-collection methods for our primary exposure and outcomes, and comprehensive adjustment for an extensive set of covariates that could influence major depressive disorder and breast cancer outcomes.

Several factors support the generalizability of our findings to civilian women with breast cancer and major depressive disorder. First, although veterans may have unique military-related exposure histories (eg, combat, posttraumatic stress disorder, traumatic brain injury, chronic pain, amputations), there is limited evidence to suggest that these factors are specifically associated with breast cancer outcomes. Second, the VA cohort is comparable to the general population with respect to race, comorbidity burden (as measured by the Charlson Comorbidity Index), and breast cancer stage at diagnosis. Compared with the Surveillance, Epidemiology, and End Results (SEER) Program, a national database of patients with cancer, the racial makeup is similar, with 67% of our VA cohort and 71% of the SEER population self-reporting their race as White. Further, studies of women with breast cancer in the general population reported that the majority of women younger than 65 years of age have a Charlson Comorbidity Index between 0 and 1, which is similar to the average Charlson Comorbidity Index of 0.58 in our VA cohort.⁴³ The distribution of breast cancer stage was also comparable between our VA cohort (68% localized) and the general SEER population (65% localized), but the age at breast cancer diagnosis was slightly younger in our VA cohort than in the general SEER population (median = 62 years). Third, the VA uses the National Comprehensive Cancer Network guidelines in decision making regarding breast cancer screening and treatment. Therefore, breast cancer recurrence and mortality patterns do not differ between VA and non-VA populations. According to data from large population-based studies, recurrence rates range from 10% to 20% within 5 years for localized and hormone-positive cancers and can be higher for more aggressive subtypes (ERBB2-positive or triple-negative subtypes).^44-46 Our recurrence estimate of 16% over a median follow-up of 4.6 years is aligned with expectations for a mixed cohort. For breast cancer mortality, the SEER database reports an average 5-year relative survival rate of approximately 91%, meaning 9% die from breast cancer within this time frame. Our breast cancer mortality estimate of 6.2% over a median follow-up of 5.0 years is consistent with national averages. Finally, the diagnosis of major depressive disorder is based on standardized diagnostic criteria and is consistently applied across the VA and non-VA health-care settings,⁹ which means that the diagnosis should be comparable between VA and general-population cohorts. The prevalence of major depressive disorder may be higher in our VA cohort (29%) than among women in the general population (lifetime prevalence of 10%-20%) because of more regular screening for depression in VA settings, which may not happen routinely in non-VA settings.^47,48 This factor should not affect the generalizability of our findings, however, and would instead help minimize misclassification of major depressive disorder based on underreporting.

Our study provides compelling evidence of an association between prediagnostic major depressive disorder and inferior breast cancer outcomes among female veterans, particularly for women with estrogen receptor–positive tumors, current smokers, women with SUD, and women who did not undergo screening mammography. Next steps include confirming our findings in a well-designed prospective study and investigation of interventions that could improve outcomes for women with breast cancer and major depressive disorder.

Acknowledgments

The study sponsor had no role in the study design or conduct; data collection, management, analysis, or interpretation; manuscript preparation, review, or approval; or the decision to submit the manuscript for publication.

Author contributions

Maya Aboumrad, MPH (Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Software; Visualization; Writing—original draft; Writing—review & editing), Corinne Joshu, PhD, MPH (Conceptualization; Investigation; Methodology; Supervision; Writing—review & editing), Kala Visvanathan, MD, MHS (Conceptualization; Funding acquisition; Investigation; Methodology; Supervision; Writing—review & editing)

Funding

This research was intramurally supported by the VA Medical Centers (White River Junction, Vermont) and grants from the National Cancer Institute, National Institutes of Health (T32-CA009314 and P30CA006973) and the Breast Cancer Research Foundation.

Conflicts of interest

M.A. has no conflicts of interest to disclose; C.J. is a recipient of an American Cancer Society Honorarium; K.V. reports grant funding from Cepheid, nonfinancial support from Optra Health Inc, and an issued patent to C11625. K.V., a JNCI associate editor and coauthor of this article, was not involved in the editorial review or decision to publish the manuscript. The contents of this manuscript have not been previously presented, posted, or published.

Data availability

Data cannot be shared publicly because they involve sensitive human subject data. Data may be available for researchers who meet the criteria for access to confidential data after evaluation from the White River Junction Institutional Review Board and VA Research and Development Committees. As a VA national legal policy (Veterans Health Administration directive 1605.01), VA will share patient data only if there is a fully executed contract [Cooperative Research and Development Agreement] in place for the specific project. These contracts are typically negotiated in collaboration with VA national Office of General Council and lawyers from the collaborating institution. These national sharing policies and standards also apply to deidentified data. In addition, if a contract is in place allowing sharing of deidentified data outside of VA, then VA national policy (Veterans Health Administration directive 1605.01) states that deidentification certification must be met by expert determination. The expert determination requires independent assessment from an experienced master or PhD in biostatistics, from a third party not involved in the project, and may require outside funding to support. In addition, for an outside entity to preform research on VA patient data, institutional review board and VA Research and Development Committee approval is required for the specific project. Data requests may be sent to White River Junction VA Medical Center, 215 N Main St, White River Junction, VT 05009.