This cross-sectional analysis utilized data from a prospective cohort study examining psychological distress screening tools among metastatic breast cancer patients in southwestern Nigeria. The parent study was conducted at NSIA-LUTH Cancer Centre, a specialized oncology unit within Lagos University Teaching Hospital, one of Nigeria’s premier tertiary healthcare institutions. The center provides comprehensive cancer services including surgery, chemotherapy, radiotherapy, and supportive care to patients from Lagos and surrounding states. Data collection occurred between September 2020 and February 2022, spanning the COVID-19 pandemic period.

Consecutive patients presenting with metastatic breast cancer were enrolled if they met the following criteria: (i) histopathologically confirmed breast carcinoma with clinical or radiological evidence of distant metastases; (ii) age ≥ 18 years; and (iii) ability to complete study instruments in English or with translation assistance in Yoruba, Igbo, or Hausa languages. Patients with severe cognitive impairment precluding informed consent or questionnaire completion were excluded. The study enrolled 313 patients, of whom 274 (87.5%) had complete depression assessment data and comprised the analytical sample.

Depression was assessed using the Beck Depression Inventory-II (BDI-II), a 21-item self-report instrument measuring depression severity over the preceding two weeks [18]. Each item is rated on a 4-point scale (0–3), yielding total scores ranging from 0 to 63. The BDI-II has demonstrated robust psychometric properties in oncology populations, including validation studies in sub-Saharan African cancer cohorts with Cronbach’s alpha values exceeding 0.85 [19]. Standard severity categories were applied: minimal depression (0–13), mild depression (14–19), moderate depression (20–28), and severe depression (29–63) [20]. Clinically significant depression (CSD), our primary exposure, was defined as BDI-II ≥ 20, corresponding to moderate-to-severe depression warranting clinical intervention [21]. This threshold is consistent with oncology-specific validation studies, including Warmenhoven et al. [22], who validated the BDI-II ≥ 20 cutoff in advanced cancer patients, and aligns with NCCN-recommended screening thresholds for depression in oncology populations [23]. Sensitivity analyses using continuous BDI-II scores and alternative cutoffs (Distress Thermometer ≥4) provided convergent evidence supporting the robustness of this operationalization.

Depression was assessed at initial cancer center presentation, representing a cross-sectional snapshot. Therefore, temporal ordering between depression onset and treatment initiation cannot be established; CSD may represent pre-existing depression, reactive depression to the cancer diagnosis, or both, and this distinction carries important implications for interpretation. The BDI-II was administered in English, with translation assistance available in Yoruba, Igbo, and Hausa for patients with limited English proficiency. While the BDI-II has demonstrated adequate psychometric properties in sub-Saharan African cancer populations [19,22], formal linguistic validation for Nigerian languages was not performed for this study. The Distress Thermometer, as a single-item visual analog scale, is inherently less susceptible to linguistic translation challenges but may be subject to cultural interpretation differences. These considerations are discussed further in the limitations.

Treatment uptake: A composite binary variable indicating receipt of any cancer-directed therapy was created, including systemic chemotherapy (anthracycline-based, taxane-based, or combination regimens), targeted therapy (trastuzumab for HER2-positive disease), hormonal therapy, and surgical intervention (mastectomy or breast-conserving surgery). This composite approach acknowledged the complexity of treatment decision-making in resource-limited settings where optimal multimodal therapy may not be universally accessible.

Survival status: Vital status was ascertained through medical record review and direct patient/family contact at follow-up. Time-to-event was calculated from the date of cancer diagnosis to death from any cause or last known follow-up for censored observations.

Comprehensive clinical data were extracted from medical records and study questionnaires. Demographic variables included age and sex. Tumor characteristics encompassed histological subtype and immunohistochemical profile (estrogen receptor, progesterone receptor, HER2 status). Metastatic burden was characterized by sites of distant spread (lung, spine, liver, pleura, brain). Clinical examination findings included presence of skin ulceration, peau d’orange, and axillary lymph node status (no palpable nodes, mobile nodes, or matted nodes). Psychological distress was additionally assessed using the National Comprehensive Cancer Network Distress Thermometer (DT), a validated single-item visual analog scale (0–10) with established cutoffs for clinically significant distress (≥4) [23].

Analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC). Descriptive statistics characterized the study population. For the primary analysis of treatment uptake, multivariable logistic regression was employed with CSD as the primary exposure variable. Potential confounders were identified through univariable analysis using a liberal p < 0.20 threshold and clinical judgment. To avoid overfitting, the number of covariates was limited to approximately one per 10 outcome events [24]. Quality of life demonstrated significant interaction with CSD (p = 0.015) and was therefore excluded from multivariable models. This decision was based on the rationale that quality of life likely lies on the causal pathway between depression and outcomes (i.e., it functions as a mediator rather than a confounder), and its inclusion would attenuate the very association under investigation, constituting inappropriate adjustment per epidemiologic guidance on distinguishing confounders from mediators [25].

Survival analysis employed Kaplan-Meier methods with log-rank testing for unadjusted comparisons and Cox proportional hazards regression for adjusted analyses. The proportional hazards assumption was verified using Schoenfeld residuals and log-log plots. Statistical significance was defined as two-sided p < 0.05. Post-hoc power calculations are presented for interpretive context rather than prospective justification, acknowledging the well-documented limitations of retrospective power analyses [26]. These indicated 80% power to detect odds ratios ≥2.0 for the treatment uptake analysis given the observed CSD prevalence and event rates. Notably, the observed effect sizes were small (OR 0.95, HR 1.12) with confidence intervals spanning both protective and harmful effects, indicating genuine uncertainty rather than a definitive null finding.

The 39 patients (12.5%) with incomplete BDI-II data were compared with complete cases on all available baseline characteristics. These patients did not differ significantly in age (53.8 vs. 54.1 years, p = 0.89), treatment receipt (60.0% vs. 64.6%, p = 0.58), or mortality (11.4% vs. 14.6%, p = 0.61), suggesting that missingness was likely random rather than systematic, though non-random missingness cannot be definitively excluded [27,28].

The study was approved by the Lagos University Teaching Hospital Health Research Ethics Committee (LUTH-HREC, ADM/DCST/HREC/APP/7058). All participants provided written informed consent. Data were de-identified prior to analysis, and confidentiality was maintained throughout in accordance with the Declaration of Helsinki. This study represents a secondary analysis of de-identified patient data from the original study. All identifying information was removed prior to data sharing, and researchers had access only to anonymized data. Data were first accessed for this secondary analysis on October 27, 2024.

Of 313 enrolled patients, 274 (87.5%) had complete BDI-II data and comprised the analytical cohort. The study population was predominantly female (99.4%) with a mean age of 54.0 years (SD 13.0, median 53.0). The near-exclusive female composition reflects the disease epidemiology of breast cancer but limits generalizability to male breast cancer patients, who may experience distinct psychosocial and treatment-seeking patterns. Invasive ductal carcinoma was the dominant histology (93.6%), and triple-negative breast cancer was the most common immunophenotype (52.1%), followed by hormone receptor-positive/HER2-negative disease (21.9%). Lung metastases were most prevalent (53.2%), followed by spine (48.4%), pleura (18.3%), liver (13.5%), and brain (12.8%). Physical examination revealed ulceration in 8.3%, peau d’orange in 9.9%, and significant axillary lymphadenopathy in 53.2% (mobile nodes: 29.2%; matted nodes: 24.0%). Complete demographic and clinical characteristics are presented in Table 1.

Depression was highly prevalent in this metastatic breast cancer population. Overall, 144 patients (52.6%) exhibited depressive symptoms of any severity, and 49 (17.9%) met criteria for CSD (BDI-II ≥ 20). The severity distribution was: minimal/no depression (BDI-II < 14), 130 (47.4%); mild depression (BDI-II 14–19), 95 (34.7%); moderate depression (BDI-II 20–28), 45 (16.4%); and severe depression (BDI-II 29–63), 4 (1.4%). The predominance of mild-to-moderate depressive symptoms (34.7% mild, 16.4% moderate) with relatively few cases of severe depression (1.4%) suggests a preponderance of adjustment-related depressive responses rather than severe depressive disorder, consistent with psychological adaptation models in cancer populations [29]. Mean BDI-II scores differed substantially between groups: 8.2 ± 4.1 in the non-CSD group versus 23.1 ± 3.8 in the CSD group (p < 0.001). Concordantly, significant psychological distress (DT ≥ 4) was present in 46.0% of participants, with a trend toward higher distress scores in the CSD group that did not reach statistical significance (57.1% vs. 43.6%, p = 0.10). Depression characteristics stratified by CSD status are presented in Table 2.

Overall, 177 patients (64.6%) received cancer-directed treatment. Treatment modalities included surgery (n = 161, 51.6%), anthracycline-based chemotherapy (AC: n = 45; CAF: n = 28; FEC: n = 19), cyclophosphamide-based regimens (CMF: n = 10), taxane-based therapy (n = 27), and targeted therapy with trastuzumab (n = 12). Many patients received multimodal therapy. Strikingly, treatment receipt was nearly identical between depression groups: 65.3% (32/49) of CSD patients and 64.4% (145/225) of non-CSD patients received treatment (unadjusted OR 1.04, 95% CI 0.52–2.07, p = 0.92).

In multivariable logistic regression adjusting for age, distress, and clinical factors, CSD remained non-significantly associated with treatment receipt (adjusted OR 0.95, 95% CI 0.36–2.52, p = 0.92). Depression assessment and treatment status were ascertained contemporaneously at the cross-sectional evaluation; thus, the temporal sequence between depression onset and treatment initiation cannot be determined. In contrast, clinical indicators of disease extent and resectability demonstrated strong associations with treatment receipt. Patients with lung metastases were significantly more likely to receive treatment (OR 3.63, 95% CI 1.59–8.29, p = 0.002), as were those with ulcerating tumors (OR 20.39, 95% CI 2.39–173.92, p = 0.006). The very wide confidence interval for ulceration reflects sparse data for this predictor (n = 26, 8.3%) and should be interpreted as indicating the direction and approximate magnitude of association rather than a precise point estimate, consistent with known small-sample bias in logistic regression [30]. Axillary lymph node status was the strongest predictor: mobile nodes (OR 20.33, 95% CI 7.18–57.58, p < 0.001) and matted nodes (OR 7.84, 95% CI 3.05–20.15, p < 0.001) were both strongly associated with treatment receipt compared to patients without palpable lymphadenopathy. The model demonstrated good discrimination (c-statistic = 0.84). Full results are presented in Table 3.

During follow-up (median 756 days, IQR 45–1095), 40 patients (14.6%) died. Mortality was similarly distributed between depression groups: 7/49 (14.3%) in the CSD group and 33/225 (14.7%) in the non-CSD group. Kaplan-Meier survival analysis revealed virtually superimposable survival curves between CSD and non-CSD patients throughout the observation period (log-rank p = 0.95; Fig 1). Survival probabilities at 200, 400, 600, and 800 days were nearly identical between groups (S1 Table).

Cox proportional hazards analysis confirmed the absence of association between CSD and mortality. The unadjusted hazard ratio for CSD versus non-CSD was 0.97 (95% CI 0.43–2.19, p = 0.95), and the adjusted hazard ratio was 1.12 (95% CI 0.48–2.61, p = 0.80) after controlling for age, treatment, distress, and clinical covariates. With only 40 deaths (7 in the CSD group), the analysis had limited statistical power to detect modest hazard ratios, and the wide confidence interval (0.48–2.61) spans both protective and harmful effects, indicating that a clinically meaningful association cannot be excluded. Clinical factors demonstrated stronger associations: ulceration (HR 19.26, 95% CI 2.48–149.85, p = 0.005) and lymph node involvement predicted increased mortality, while receipt of treatment, particularly combined chemotherapy and surgery, was strongly protective (HR 0.25, 95% CI 0.10–0.65, p = 0.004). Lung metastasis was associated with a 91.4% lower hazard of mortality (adjusted HR 0.086, 95% CI 0.019–0.397, p = 0.002). This counterintuitive finding may reflect multiple mechanisms: (i) ascertainment bias, wherein patients with lung metastases may have been identified through routine imaging, indicating earlier detection within the metastatic cascade; (ii) differential access, as pulmonary symptoms may prompt earlier presentation to oncology services; and (iii) immortal time bias, as patients surviving long enough for lung metastasis documentation may represent inherently better-prognosis individuals. This finding warrants cautious interpretation and prospective validation. Complete Cox regression results are presented in Table 4.

Several sensitivity analyses were conducted to test the robustness of findings. When depression was modeled as a continuous variable, BDI-II scores showed no significant association with treatment receipt (OR per 5-point increase: 1.02, 95% CI 0.89–1.17, p = 0.78) or mortality (HR per 5-point increase: 0.96, 95% CI 0.79–1.16, p = 0.66). Analysis using the Distress Thermometer as an alternative psychological measure yielded consistent results: high distress (DT ≥ 4) was not significantly associated with treatment uptake (64.3% vs. 64.9%, p = 0.93) or survival (HR 1.28, 95% CI 0.65–2.52, p = 0.47). Missing data analysis revealed that the 35 patients with incomplete BDI-II data had similar baseline characteristics to complete cases, including age (53.8 vs. 54.1 years, p = 0.89), treatment receipt (60.0% vs. 64.6%, p = 0.58), and mortality (11.4% vs. 14.6%, p = 0.61), suggesting minimal selection bias.

The 17.9% prevalence of CSD and 52.6% prevalence of any depression aligns with global estimates from meta-analyses reporting depression rates of 15–25% among breast cancer patients [31,32]. These rates are also broadly consistent with emerging data from the African continent, including Calys-Tagoe et al.’s [12] report of 84.2% depression prevalence among Ghanaian breast cancer patients and Ohaeri et al.’s [13] documentation of substantial psychosocial morbidity in Nigerian oncology populations. However, several factors suggest these estimates may underestimate true depression burden in this population. Mental health stigma remains substantial in Nigeria, with studies documenting that over half of healthcare providers hold stigmatizing attitudes toward mental illness [15]. Patients may consequently underreport symptoms on self-administered instruments. Additionally, depression in cancer patients often manifests with somatic symptoms, fatigue, appetite disturbance, sleep disruption, that overlap with cancer symptoms and treatment effects, potentially leading to misattribution and underrecognition [22]. The predominance of triple-negative disease (52.1%) in our cohort, higher than typically observed in Western populations, may reflect both biological differences in African breast cancer epidemiology and the aggressive disease biology associated with delayed presentation.

It is important to note that the BDI-II identifies symptom severity levels exceeding validated thresholds rather than formal psychiatric diagnoses per DSM-5 or ICD-11 criteria. The absence of structured clinical interviews (e.g., SCID) means that our CSD classification captures self-reported symptom burden, which may differ from clinical depression diagnosed by mental health professionals. This distinction is relevant because self-report instruments may both over-identify (through somatic symptom overlap with cancer) and under-identify (through stigma-related underreporting) clinically meaningful depression in this population.

Our finding that CSD was not associated with treatment uptake contrasts sharply with evidence from HICs, where systematic reviews document significantly reduced treatment receipt among cancer patients with mental illness [4]. We propose that this divergence may reflect differences in treatment allocation mechanisms between resource-abundant and resource-constrained settings. In HICs, where treatment capacity generally exceeds demand, psychosocial factors may influence patient preferences, provider recommendations, and treatment adherence in ways that translate to measurable disparities. In contrast, in settings like Nigeria where treatment capacity is severely limited, fewer than ten radiotherapy centers serve over 220 million people [16], clinical triage based on tumor biology, disease stage, and anticipated treatment response may predominate, potentially overriding psychosocial considerations.

The strong associations observed between clinical factors and treatment receipt support this hypothesis. The very large odds ratios for lymph node status (OR 20.3 for mobile nodes) and ulceration (OR 20.4) likely reflect clinical decision algorithms that heavily weight tumor resectability and disease extent. The conceptual framework illustrated in Fig 2 depicts these hypothesized relationships. In effect, the clinical imperative to allocate scarce treatment resources optimally may create conditions that are paradoxically more equitable access across psychosocial strata, though at the cost of overall treatment availability. This resource-constraint hypothesis represents a tentative explanatory framework requiring prospective validation with health system data, including provider decision-making processes and treatment allocation criteria, rather than an established mechanism.

The absence of survival differences between CSD and non-CSD patients (14.3% vs. 14.7% mortality) contrasts with population-based studies from HICs demonstrating significant excess mortality among cancer patients with severe mental illness [5,33]. Several factors may explain this finding. First, the cross-sectional assessment of depression at cancer presentation captures a mixture of reactive depression (temporally proximate to cancer diagnosis) and pre-existing depression. Evidence suggests that chronic or pre-existing depression has stronger associations with cancer outcomes than acute reactive depression [4], but our design could not distinguish these constructs. Second, in a population with high baseline mortality from metastatic disease, the marginal effect of depression may be difficult to detect, particularly with limited statistical power. Third, competing risks from the aggressive tumor biology characteristic of this population, predominantly triple-negative and late-stage disease, may overwhelm any depression-attributable mortality increment.

Notably, treatment access emerged as a dominant survival determinant. The 75% mortality reduction associated with combined chemotherapy and surgery (HR 0.25) underscores the critical importance of multimodal treatment access, rather than psychosocial optimization, as the proximate driver of survival improvement in this setting. This finding aligns with the broader literature on cancer care disparities in LMICs [3] and suggests that interventions targeting treatment access may yield greater survival gains than psychosocial interventions alone.

Several alternative explanations for the null depression–outcome associations warrant consideration beyond the resource-constraint hypothesis. First, measurement bias from potential underreporting due to stigma and cultural response patterns may have attenuated true associations. Second, misclassification bias is a concern; if patients with genuine CSD scored below the BDI-II ≥ 20 threshold due to symptom underreporting, they would be classified into the non-CSD group, diluting the exposure contrast and biasing results toward the null. Third, residual confounding from unmeasured variables, including socioeconomic status, insurance coverage, social support networks, health literacy, and traditional medicine use, cannot be excluded. Fourth, the BDI-II cutoff of ≥20, while validated in Western oncology populations, may not optimally capture the construct of clinically significant depression in this Nigerian cohort, where somatic symptom expression and cultural manifestations of distress may differ from Western presentations [34,35]. These alternative explanations collectively suggest caution in interpreting the null findings as definitive evidence of absence.

This study has several strengths. It represents one of the largest investigations of depression and cancer outcomes from sub-Saharan Africa, utilizing validated instruments with established psychometric properties in African populations. The comprehensive clinical data enabled robust adjustment for clinicopathological confounders. The single-center design, while limiting generalizability, ensured consistency in clinical practice patterns and data collection.

Important limitations warrant consideration. The cross-sectional design precludes causal inference regarding the temporal relationship between depression and outcomes. Statistical power was limited for detecting modest effect sizes, and larger multicenter studies are needed to exclude clinically meaningful associations. The relatively short median follow-up (756 days) may have been insufficient to detect survival differences that emerge over longer horizons [33]. Cultural factors may affect symptom expression and reporting on Western-developed instruments, despite validation efforts. Formal linguistic validation of the BDI-II for Nigerian languages (Yoruba, Igbo, Hausa) was not performed, and cross-cultural equivalence of the instrument cannot be assumed. Finally, unmeasured confounding, including socioeconomic status, social support, and health literacy, cannot be excluded.

A particularly important limitation relates to potential misclassification bias arising from mental health stigma and underreporting. In the Nigerian sociocultural context, where mental health stigma remains substantial [15], patients may underreport depressive symptoms on self-administered instruments. This would result in differential misclassification, whereby truly depressed patients scoring below the BDI-II ≥ 20 threshold would be classified into the non-CSD group, effectively diluting the CSD group and attenuating any true association between depression and outcomes toward the null. This non-differential misclassification of the exposure represents a bias toward the null hypothesis and may partly explain the absence of observed associations. The potential magnitude of this bias is difficult to quantify without a gold-standard clinical interview comparison, but the documented prevalence of mental health stigma in Nigerian healthcare settings suggests it may be non-trivial [34].

Despite the null findings for depression-outcome associations, the substantial depression prevalence observed carries important implications for cancer care delivery. First, depression affects quality of life and patient-reported outcomes regardless of effects on survival, warranting attention on humanitarian and patient-centered care grounds [36]. Second, simple screening tools such as the Distress Thermometer can feasibly be implemented in resource-constrained settings to identify patients requiring psychosocial support [23]. Third, our findings suggest that in severely resource-constrained settings, interventions to expand treatment capacity and reduce structural barriers to care may yield greater survival benefits than targeted psychosocial interventions, though these approaches need not be mutually exclusive. Importantly, our findings should not be interpreted as evidence that psychosocial interventions lack survival benefit in LMIC settings; rather, they suggest that the relationship between depression and cancer outcomes may be context-dependent and mediated by health system characteristics. The substantial depression prevalence underscores the need for integrated mental health services on humanitarian, quality-of-life, and patient-centered grounds, regardless of measured survival associations.

Practical implementation of mental health integration within Nigerian oncology settings warrants consideration. Brief validated screening tools (DT, PHQ-2) require minimal training and can be administered in under 5 minutes, making them feasible for routine use in busy tertiary cancer centers. Task-shifting models utilizing trained oncology nurses for initial screening and psychoeducation have demonstrated effectiveness in similar LMIC contexts [37]. At NSIA-LUTH and comparable Nigerian tertiary centers, mental health screening could be integrated into existing treatment workflows at the point of initial oncology consultation, with referral pathways to psychiatry or clinical psychology for patients exceeding screening thresholds. Digital health innovations, including mobile-based psychoeducation and remote psychological support platforms, offer additional scalable approaches to addressing the substantial mental health burden in settings with limited specialized mental health workforce [38].

Future studies should employ prospective longitudinal designs with serial depression assessments to capture the temporal dynamics of the depression–cancer outcome relationship. Multicenter collaborations across African cancer centers could provide adequate power to detect modest associations and improve generalizability. Research examining healthcare system factors mediating these relationships, including provider decision-making processes, treatment allocation criteria, and patient navigation, would inform intervention development. Finally, implementation studies evaluating integrated mental health and cancer care models adapted for LMIC contexts are needed to establish evidence-based approaches to addressing the substantial psychosocial burden we observed.