Examining the network dynamics of core symptoms in Chinese patients with advanced cancer: a diary study

Traditional approaches examining isolated symptoms fail to capture dynamic network structures in palliative care. This study aimed to decompose physiological and psychological symptom networks across multiple levels and identify central symptoms that may inform hypothesis-generating targets for future intervention studies in advanced cancer patients. This secondary retrospective longitudinal study utilized diary-based design and multilevel network analysis at a tertiary cancer hospital in mainland China. 166 advanced cancer patients receiving palliative care completed ≥ 20 consecutive daily assessments using standardized PCOC tools: Symptom Assessment Scale (SAS), Palliative Care Problem Severity Score (PCPSS), and Resource Utilisation Group-Activities of Daily Living (RUG-ADL). Multilevel vector autoregressive (mlVAR) models estimated temporal networks (lag-1 effects), contemporaneous networks (same-day associations), and between-subjects networks. Network centrality was quantified using expected influence (EI), with stability evaluated through bootstrap procedures. Time-series stationarity was confirmed via Augmented Dickey-Fuller (ADF) tests, and autocorrelation function (ACF) analysis was conducted to assess temporal persistence patterns across symptom domains. Four distinct network patterns emerged. In SAS networks, nausea showed highest temporal centrality (temEI = 0.102) with cross-lagged associations with appetite loss and bowel distress, while sleep distress (conEI = 0.836) and appetite (conEI = 1.018) were contemporaneously central. PCPSS networks revealed psychological/spiritual issues were associated with subsequent family/carer problems ( r = 0.059) with highest between-subjects centrality (0.891). ADL networks identified toileting dependency as temporally central (temEI = 0.199), associated with subsequent changes in functional decline. The comprehensive network highlighted psychological/spiritual problems with strongest temporal influence (temEI = 0.251) and six significant connections, particularly associated with ADL deterioration. Autocorrelation analysis revealed distinct persistence patterns: ADL measures showed strongest persistence; SAS demonstrated moderate decay; PCPSS exhibited scale-specific patterns with psychological/spiritual concerns showing greater persistence (0.814 → 0.258) than pain (0.544 → 0.020). This study identified distinct symptom network patterns across temporal, contemporaneous, and between-subjects levels. Findings provide preliminary evidence that may guide future intervention studies targeting central symptoms to investigate potential cascading effects and optimize monitoring strategies based on symptom persistence characteristics.