This study analyzed proxy-reported exit interviews from the Korean Longitudinal Study of Aging (KLoSA) for adults aged 45 and older who died in hospital between 2006 and 2020 (N = 1,731). Eleven symptoms occurring in the last month of life were assessed and dichotomized. The cohort had a mean age at death of 74.2 years and a mean symptom count of 4.76. The three most prevalent symptoms were loss of appetite (66.5%), severe fatigue (59.4%), and depression (54.3%). Other assessed symptoms included dyspnea, persistent coughing, frequent vomiting, difficulty with urination and defecation, extremity discomfort, loss of consciousness, periodic mental confusion, and difficulty controlling emotions.
Demographic and socioeconomic factors captured in KLoSA included age at death, sex, education, marital status, religious affiliation, number of living children, and health insurance type. These variables were included as covariates in multivariable logistic models for each symptom. The dataset described that 52.85% of decedents were men, about 31.03% had elementary school education as the highest level completed, and 90.08% were covered by national health insurance. The analysis tested associations of these sociodemographic variables with each symptom outcome, with results indicating varying influence by symptom type.
Decedents averaged four physician-diagnosed chronic illnesses, most commonly hypertension (52.85%), cancer (45.63%), and diabetes (41.56%). Falls and pain in the year before death were captured as binary variables; 13.10% had experienced at least one fall, and 35.99% had experienced pain in the year prior to death. Functional dependency was measured as continuous scores for activities of daily living (ADLs; range 0–7, mean 3.76) and instrumental ADLs (IADLs; range 0–10, mean 5.53) during the three months preceding death. Across nearly all symptom models, greater ADL and IADL dependency and reported pain in the year before death were consistently associated with higher odds of symptom occurrence.
Duration of illness prior to death was collected in detailed categories and reclassified into five groups for analysis (sudden death without prior illness; >1 day–<6 months; 6 months–<1 year; 1–<5 years; ≥5 years). The study found that prolonged illness duration was one of the most consistent predictors of symptom presence across the assessed symptoms. Final hospitalization length was recorded (1–730 days); hospital stays of 31 days or more were treated as hospitalizations started more than 30 days before death for analytic purposes.
Hospice admission since baseline was recorded, including the number of admissions and duration of the most recent stay (1–730 days), with non-admitted decedents assigned 0 days. Only 4.51% of decedents had accessed hospice care before death. The authors highlight a critical gap between high symptom burden and limited hospice utilization, suggesting unmet need for palliative services among hospitalized decedents.
KLoSA is a nationwide, biennial panel survey of community-dwelling adults aged 45 and older that uses multistage, stratified sampling. Between 2008 and 2020, 3,545 deaths were recorded among panel participants; exit interviews were conducted for 2,727 decedents (76.9%). For this study, cases were excluded if death occurred outside hospital (n = 940), place of death was unspecified (n = 11), sampling weight was missing (n = 22), or key covariate data were missing (n = 23), yielding the analytic sample of 1,731 hospital decedents. Symptom outcomes were assessed by single-item proxy questions; “don’t know” responses were recoded as “no.”
Analyses were performed using Stata 19 SE with specification of the primary sampling unit, stratum, and KLoSA sampling weights. Descriptive statistics reported weighted means and percentages. Survey-weighted linear regression compared total symptom counts across categorical subgroups, with Bonferroni-corrected pairwise comparisons when applicable. Because symptom counts exhibited overdispersion, survey-weighted negative binomial regression was used to identify factors associated with total number of symptoms. Eleven separate multivariable survey-weighted logistic regression models were fitted—one per symptom—using the same covariate set. Multicollinearity was assessed by variance inflation factors (all within acceptable range). As a sensitivity analysis, models were re-estimated excluding cases with explicit “don’t know” responses for each symptom. The authors did not apply multiple-testing correction across the 11 distinct symptom models, noting each addressed a prespecified outcome. A two-tailed p < 0.05 denoted statistical significance.
The study reports a high burden of end-of-life symptoms among hospital decedents in South Korea, with a mean of 4.76 symptoms in the last month of life. Loss of appetite, severe fatigue, and depression were the most commonly reported. Across nearly all symptom outcomes, three factors were the most consistent predictors of symptom presence: prolonged illness duration, greater functional dependence in ADLs and IADLs, and pain during the year before death. The low hospice admission rate (4.51%) underscores limited engagement with specialized palliative services among this inpatient population. The authors recommend earlier recognition of symptom indicators, integration of palliative care into hospital practice, and expanded hospice access to address the gap between symptom burden and available support.
The authors acknowledge the use of proxy-reported data and the potential for recall bias. Symptom measurement relied on single-item proxy questions and treated “don’t know” responses as absence, which may underestimate true prevalence. The cross-sectional secondary analysis design precludes causal inference. Further details on model-specific effect sizes and p-values for each covariate–symptom association were reported in the original article but are not reproduced here; readers should consult the source for full tables and inferential statistics.