Evaluating the psychosocial effects of cancer on sleep quality and mental health in elderly populations | BMC Psychiatry

Study design and data source

This cross-sectional study utilized data from the 2018 wave of the Mexican Health and Aging Study (MHAS), also known as Estudio Nacional de Salud y Envejecimiento en México (ENASEM) [30,31,32]. The MHAS is a nationally representative, longitudinal population-based study of community-dwelling adults aged 50 years and older in Mexico, initiated in 2001 and designed to prospectively evaluate the impact of disease on health, function, and mortality of older adults in both urban and rural areas. The study protocols and survey instruments are highly comparable to the U.S. Health and Retirement Study (HRS), enabling cross-national comparisons of aging processes and serving as the first longitudinal study of older adults in Mexico with a broad socioeconomic perspective. This analysis follows the STROBE (Strengthening the Reporting of Observational studies in Epidemiology) guidelines for reporting cross-sectional studies.

The MHAS employs a complex, multi-stage stratified sampling design based on Mexico’s National Employment Survey (Encuesta Nacional de Empleo, ENE) conducted by the Instituto Nacional de Estadística y Geografía (INEGI). The baseline sample (2001) targeted a nationally representative sample of individuals born in 1951 or earlier (aged 50+ in 2001), with the sampling frame consisting of households from the National Employment Survey that had at least one resident aged 50+ years. Geographic coverage included all 32 states of Mexico, encompassing both urban and rural areas, with special consideration given to households in six states accounting for 40% of migrants to the United States, which were over-sampled to ensure adequate representation of this population. When multiple age-eligible persons resided in selected households, one was randomly selected as the primary respondent, and spouses or partners of selected individuals were recruited regardless of age. The 2018 wave included all surviving participants from previous waves (2001, 2003, 2012, 2015) plus refreshment samples to maintain national representativeness, accounting for attrition and demographic changes.

Participants and eligibility

From the 2018 MHAS wave, we included adults aged 50 years and older at the time of interview with complete data on cancer diagnosis status and key demographic and health variables. Specific inclusion criteria were: (1) age 50 years or older at time of 2018 interview, (2) complete data on cancer diagnosis status, and (3) complete data on primary outcome variables. Exclusion criteria included individuals with incomplete data on cancer status, missing data on primary outcome variables, and institutional residents such as those in nursing homes or hospitals. After applying these criteria, our analytical sample comprised 9,302 participants, including 855 individuals with physician-diagnosed cancer (9.2%) and 8,447 controls without cancer diagnosis (90.8%).

While MHAS sample sizes were predetermined based on achieving national representativeness rather than specific hypothesis testing for our research question, we conducted post-hoc power analysis to evaluate our ability to detect meaningful differences between cancer survivors and controls. With our achieved sample size of 9,302 participants (855 cancer cases, 8,447 controls), we had greater than 80% power to detect effect sizes of 0.2 or larger (small to medium effects) for continuous variables and odds ratios of 1.3 or greater for binary outcomes, using a two-sided alpha level of 0.05.

Data collection procedures and instrumentation

Data were collected through computer-assisted personal interviews (CAPI) conducted by trained INEGI interviewers who underwent standardized training protocols. Direct interviews were prioritized, with proxy interviews conducted only when participants were unable to respond due to illness, hospitalization, or temporary absence. Comprehensive quality control measures included supervisor oversight during data collection, standardized interviewer training, and systematic data validation procedures implemented by INEGI.

The 2018 MHAS survey included comprehensive questionnaires covering multiple domains: demographic characteristics and socioeconomic status, health conditions and self-rated health assessment, functional status evaluation including Activities of Daily Living (ADL) and Instrumental Activities of Daily Living (IADL), cognitive assessment and mental health screening using the CESD-9 depression scale, social support and family network evaluation, healthcare utilization and insurance coverage, and detailed housing characteristics and wealth indicators. Interview duration typically ranged from 90 to 120 minutes, depending on the participant’s health status and complexity of responses.

Variable definitions and measurements

Cancer status, our primary exposure variable, was defined through self-reported physician diagnosis using the question “Has a doctor ever told you that you have cancer?” with responses categorized as binary (Yes/No). While this approach does not distinguish between cancer types, stages, or treatment status, it identifies individuals who have received a formal cancer diagnosis from a healthcare provider, consistent with epidemiological surveys in aging populations.

Primary outcome variables included comprehensive socioeconomic indicators: total household wealth measured as a continuous variable in Mexican pesos, including all financial assets, real estate, business investments, and durable goods; employment status categorized as currently working versus not currently working; detailed labor force status including working, not working, homemaker, student, or looking for work; public pension receipt as a binary indicator; and home ownership status categorized as owned/currently paying, rented, or borrowed/free transfer.

Social support variables encompassed family and friend contact frequency, measured both as a binary indicator (any regular contact vs. none) and as a continuous frequency scale ranging from 1 (never) to 9 (daily); participation in social activities including religious services, community groups, volunteer work, or recreational activities; frequency of social activity participation on a 1-9 scale; and financial assistance received from children as a binary indicator.

Health and functional status measures included self-rated health on a 5-point Likert scale (excellent, very good, good, fair, poor); Body Mass Index calculated from self-reported height and weight; ADL limitations assessed through six standard items (bathing, dressing, eating, getting in/out of bed, walking, using toilet) with responses summed to create a continuous count of limitations [33]; IADL limitations evaluated through four items (managing money, shopping, preparing meals, managing daily activities) similarly summed [34, 35]; depression symptoms measured using the CESD-9 scale, a validated 9-item version of the Center for Epidemiologic Studies Depression Scale with scores ranging from 0-9 [36, 37]; multiple sleep disturbance indicators including difficulty falling asleep, frequent night awakenings, early morning awakening, and feeling rested; pain assessment as a binary indicator of any pain experience; and falls in the past 2 years including both occurrence and frequency.

Additional covariates included age in continuous years, gender as binary (male/female), years of formal education as a continuous variable, smoking status (current/former smoker vs. never smoker), alcohol consumption (any current drinking vs. none), and history of falls in the previous 2 years.

Statistical analysis

The study employed a series of statistical methods to explore the associations between sleep disturbances, mental health, and cancer:

Descriptive statistics

Initial descriptive statistics summarized participant characteristics, including means, standard deviations, and frequencies. Differences between cancer and control groups for continuous variables were assessed using independent t-tests, while chi-square tests were employed for categorical variables to highlight demographic disparities.

Multiple Logistic Regression: To evaluate the relationship between sleep disturbances and cancer, multiple logistic regression models were constructed to assess the influence of various covariates. The models offered a gradual increase in complexity:

• Model 1 focused on analyzing the direct impact of sleep disturbances (e.g., difficulty falling asleep, frequency of night awakenings) along with demographic factors like age, gender, years of education, and various physical difficulties experienced by participants (e.g., difficulty walking, dressing, bathing).

• Model 2 extended the first model by adding variables representing additional sleep-related difficulties, such as managing daily activities and other lifestyle factors.

• Model 3 incorporated factors related to quality of life, including participation in social activities and frequency of contact with family and friends, alongside the previously mentioned variables.

• Model 4 included comprehensive socioeconomic variables, such as total household wealth and employment status, alongside all prior factors. This allowed for a thorough examination of how these diverse influences interacted with sleep disturbances to impact cancer outcomes.

OR were calculated for each model along with 95% CIs to determine the significance and strength of these relationships. Statistical significance was ascertained with a p-value < 0.05.

Bayesian statistics

In this study, we employed Bayesian statistical methods to evaluate the impact of cancer on various dimensions of sleep disorders. The Bayesian approach allows for the incorporation of prior knowledge and updating beliefs based on observed data, providing a comprehensive framework for understanding the complexities of our dataset. Specifically, we utilized Bayesian regression analysis to model the relationships between cancer diagnosis, sleep disturbances, and psychological factors such as depression.

The analysis involved estimating effect sizes and credible intervals to quantify the uncertainty around model parameters. This approach enhances our understanding of the direct effects while facilitating the exploration of indirect relationships through mediation analysis. The Bayesian framework allows for a nuanced interpretation of the data, accommodating the complexities inherent in the relationships between cancer, psychological factors, and sleep outcomes.

Propensity score matching

To control for potential confounding variables in our analysis, we employed propensity score matching (PSM) to create a balanced comparison between cancer patients and healthy controls. PSM is a statistical technique that estimates the effect of a treatment—in this case, cancer diagnosis—by accounting for the covariates that predict receiving the treatment.

Initially, we fitted a logistic regression model to estimate the propensity score for each participant based on demographic characteristics, including age, gender, education levels, and health status indicators. This score represents the probability of being diagnosed with cancer given the observed characteristics and allows for the matching of individuals from the cancer group with comparable individuals in the control group who share similar baseline characteristics.

After matching, we evaluated the balance between the two groups on key variables to ensure that the groups were comparable and that bias from confounders was minimized. This process provided a robust analytical framework, enabling us to focus on causal inference related to the effects of cancer on sleep disturbances while controlling for other influencing factors.

By integrating Bayesian statistics and propensity score matching into our methodology, we aimed to provide a rigorous analytical foundation for understanding the intricate relationships between cancer, mental health, and sleep disturbances among older adults.

Mediation analysis

To investigate potential mediating effects of sleep disturbances on cancer outcomes through psychological factors, two mediation analyses were conducted using SEM. This approach allows for a comprehensive assessment of the direct and indirect relationships between sleep disturbances, psychological factors, and cancer outcomes.

First Mediation Analysis: This analysis aimed to examine whether psychological health (measured with CES-D scores) mediated the relationship between sleep disturbances and cancer. The steps included:

Estimating the direct effect of sleep disturbances (e.g., difficulty falling asleep, frequency of night awakenings) on the diagnosis of cancer.

Assessing the relationship between sleep disturbances and the mediating variable (CES-D scores), allowing for the identification of any significant effects of sleep on psychological well-being.

Evaluating the indirect effect through the mediating pathway by examining how CES-D scores impacted the likelihood of cancer diagnosis.

Bootstrap methods were employed to obtain confidence intervals regarding the indirect effects, enhancing the robustness and accuracy of the mediation analysis results.

Second Mediation Analysis: The second analysis focused on whether life satisfaction and social engagement mediated the relationship between cancer and sleep disturbances. This analysis followed a similar procedure:

Estimating the direct path from cancer to sleep disturbances.

Evaluating the impact of cancer on life satisfaction and social engagement levels.

Examining how changes in life satisfaction and social engagement influenced sleep disturbances.

The evaluation utilized the same bootstrapping technique to establish confidence intervals for the indirect effects of the mediators, thus reinforcing the reliability of the findings.

Software used

All statistical analyses were performed using IBM SPSS Statistics (version 26) and R Statistical Software (version 4.0). SPSS was primarily utilized for descriptive statistics, basic comparative analyses (DESCRIPTIVES, FREQUENCIES, CROSSTABS, T-TEST), and initial data exploration, while R was particularly utilized for advanced statistical modeling, including mediation analysis and multiple logistic regression. Key R packages included survey (version 4.2-1) for complex survey data analysis, mediation (version 4.5.0) for mediation analyses, ggplot2 (version 3.4.2) for data visualization, and tableone (version 0.13.2) for descriptive statistics tables. Graphical representations were created with R to provide visual summaries of results, enhancing the interpretability and communication of statistical findings. The integration of both software platforms ensured comprehensive analytical capabilities while maintaining reproducibility through cross-validation of key results between platforms.

Ethical considerations and data access

The MHAS protocol has received comprehensive ethical approval from relevant institutional review boards in both Mexico and the United States, including the Instituto Nacional de Estadística y Geografía (INEGI), Instituto Nacional de Geriatría (INGer) in Mexico, and the University of Texas Medical Branch Institutional Review Board where applicable. All participants provided written informed consent prior to participation, with consent procedures conducted in Spanish and following ethical principles outlined in the Declaration of Helsinki regarding human subjects research.

The study adhered to principles of voluntary participation, with participants informed of their right to withdraw at any time or refuse to answer specific questions. Privacy and confidentiality were maintained through data de-identification procedures, with no personal identifiers retained in the analytical dataset. MHAS data are publicly available through the official study website (www.MHASweb.org) following established data use agreements that require researchers to agree to specific terms regarding data protection, appropriate use, and acknowledgment of the MHAS study team.