This prospective cohort study included a nationally representative sample from the National Health and Nutrition Examination Survey (NHANES), which is a biannual and ongoing series of surveys to track the health and nutritional status of the US population, and used a complex, multistage, probability sampling design, with oversampling of various subpopulations to increase estimate accuracy. Participants were selected through a four-stage probability sampling design with primary sampling units selected at the county level, census tract level, and household level in the 50 states and the District of Columbia. Considering the complicated survey design, including oversampling, survey nonresponse, and poststratification, sample weights are provided. All the NHANES protocols used were approved by the Centers for Disease Control and Prevention National Center for Health Statistics Ethics Review Board, and all participants provided written consent after being fully informed. Each participant was invited to participate in a face-to-face interview, a series of physical examinations, and laboratory tests in a mobile examination facility. This study included 2947 cancer survivors aged 20 years or older who participated in the NHANES (2007–2018) survey, and a flowchart of the survivor inclusion process is shown in Fig. S1 (Additional file 1). Informed consent and institutional review board approval were not necessary for the current investigation because we used published data sets from the NHANES that included no personally identifiable information. This prospective cohort study followed the Strengthening in the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Data on the cancer diagnosis and the number of types, including age at diagnosis, and up to three recorded diagnoses, were gathered from self-reported cancer assessments. Cancer survivors were defined as individuals who provided an affirmative response to the question, “Have you ever been told by a doctor or other health professional that you had cancer or a malignancy of any kind?” and were asked, “What kind of cancer was it?” and “How old were you when cancer was first diagnosed?” The number of years since the first cancer diagnosis was calculated as the difference between the participant’s current age and the age at which they were first diagnosed with cancer.

The Patient Health Questionnaire-9 (PHQ-9) was used to measure the severity of depressive symptoms in cancer survivors during the 2 weeks before the survey, and its validity and performance have been validated in cancer patients [16, 17]. The PHQ-9 consists of nine items on depressive symptoms (lack of interest, depressed mood, trouble sleeping, fatigue, appetite problems, worthlessness, lack of concentration, psychomotor agitation or retardation, and suicidal thoughts). Each item is scored on a scale ranging from 0 (not at all) to 3 (almost daily), adding to a total score ranging from 0 to 27, with higher PHQ-9 scores indicating more severe depressive symptoms. Depressive symptom categories were defined as none (score, 0–4), mild (score, 5–9), or moderate-severe (score, ≥ 10) [18]. Individuals with a total score ≥ 10 points were considered to suffer from major depression; the sensitivity of this threshold was 88% and the specificity was 88% [18].

Survivors’ sleep disturbances in the 2 weeks before the survey were assessed through the self-report question, “How often have you been bothered by trouble falling or staying asleep, or sleeping too much over the last 2 weeks?”, with response options of “not at all,” “several days,” “more than half the days,” and “nearly every day”. Survivors who responded "not at all" were considered to have no sleep disturbances; otherwise, they were considered to have sleep disturbances [19].

The primary outcomes included all-cause mortality, cancer-specific mortality, and noncancer mortality. We used the NHANES public-use linked mortality file as of December 31, 2019, which was linked to the National Death Index (NDI) using a probabilistic matching method, to ascertain the mortality status of the follow-up population. The primary cause of death was recorded using the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10). All-cause mortality is defined as death from all causes; cancer-specific mortality is defined as ICD-10 codes C00–C97; deaths from other causes are referred to as noncancer mortality. The number of months from the interview date to the date of death or, for individuals who did not suffer an event, through December 31, 2019, was defined as the follow-up period.

The choice of covariates was made using previous literature and substantive reasoning. Sociodemographic data on age, sex, race and ethnicity, educational level, marital status, the family income to poverty ratio, and work status were gathered using a standardized questionnaire. Participants self-reported their racial and ethnic backgrounds using the National Center for Health Statistics categories of Mexican American, other Hispanic, non-Hispanic White, non-Hispanic Black, and other races or ethnicities. Other races or ethnicities include American Indian/Alaskan Native/Pacific Islander, Asian, and multiracial. The categories for education level were as follows: “less than high school diploma,” “high school diploma or general equivalency diploma,” and “some college or above.” The four categories of married, never married, living with a partner, and other (including widowed, divorced, or separated) were used to classify people’s marital statuses. The family income to poverty ratio was divided into three groups: 1.30 or less, 1.31 to 3.50, and more than 3.50 [20]. The following work statuses were created using the Occupation Questionnaire: nonemployed (including unemployed individuals, retirees, students, and individuals who are not actively looking for work), part-time (working 1–34 h per week), and full-time (working 35 h per week) [21]. Sleep duration was assessed using the self-report question “How much sleep do you get (in hours)?”. Diabetes/hypertension/hypercholesterolemia was self-reported by participants who had been diagnosed by a doctor or other health professional.

The NHANES Prescribing Information Document was consulted for information regarding antidepressant use. During household interviews, data on medication use were gathered. Participants were asked about medication use in the past 30 days. When participants said “yes,” they were requested to present all medication containers or, if none were accessible, to report the name of the medication. The Lexicon Plus database was used to process and classify all prescription medication data. “Psychotherapeutic medications” was the first level category, while “antidepressants” was the second. The use of at least one antidepressant within the previous 30 days was considered antidepressant use in our study.

A complex sampling design and sampling weights were considered to ensure that the results were nationally representative, and all analyses accounted for the unequal probability of selection, oversampling of specific subpopulations, and nonresponse adjustments by the NHANES analytic guidelines [22]. Data were analyzed using R version 4.3.0. A two-sided P < 0.05 was considered to indicate statistical significance.

The baseline characteristics of survivors with varying degrees of depressive symptoms and sleep disturbances are described. Hazard ratios (HRs) and 95% CIs for the correlations of single depressive symptoms or sleep disturbances (adjusted for covariates that did not include each other) with all-cause, cancer-specific, and noncancer mortality were estimated using multivariable Cox proportional hazards regression models. Participants were divided into groups based on depressive symptoms and sleep disturbances to estimate mortality risks and examine joint associations using multivariable Cox proportional hazards regression models adjusted for the same set of covariates. Final-stage multivariable models were adjusted for age, sex, race and ethnicity, educational attainment, marital status, the family income to poverty ratio, work status, diabetes status, hypertension status, hypercholesterolemia status, NHANES cycles, number of cancer types, number of years since the first cancer diagnosis, use of antidepressants, and sleep duration. In addition, subgroup analyses were performed according to age, sex, educational attainment, work status, and use of antidepressants using Cox proportional hazards regression models.

Several sensitivity analyses were conducted to assess the robustness of our findings. First, we performed a sensitivity analysis and excluded participants who died during the initial 2-year follow-up to reduce the likelihood of reverse causation [23]. Second, because of the high cancer mortality rate among Black individuals, we conducted a sensitivity analysis excluding non-Hispanic Black participants [24]. Third, to test the effect of missing variables, multiple interpolation was used to infer all missing independent variables [25].

Mediation studies were performed to determine whether sleep disturbances mediated the association between the exposure variable (depressive symptoms) and the outcome (mortality). In our analysis, thousands of bootstraps were used. The results display p values for mediated effects, the proportion of mediating effects, and indirect pathway effect sizes.