Various aspects of social relationships have been examined as risk factors for mortality. In particular, most research has focused on either loneliness or social disengagement. We aimed to extend the current research by adding a group-level segregation measure utilizing the whole social network of one entire village in South Korea. The analyses were based on the Korean Social Life, Health and Aging Project data collected over eight years across five waves. Of the 679 old adults who participated throughout the entire project (to wave 5), 63 were confirmed as deceased. All three aspects of social relationships examined, loneliness, social disengagement, and group-level segregation, were associated with mortality in the traditional Cox proportional hazard model without considering health-related time-varying covariates. However, a Cox marginal structural model, a counterfactual statistical measure that is designed to control for censoring bias due to sample attrition over the eight years and time-varying confounding variables, revealed that only group-level segregation was associated with mortality. Our results strongly suggest that more attention is needed on group-level segregation for mortality studies, as well as on well-known individual-level risk factors, including social disengagement and loneliness. All methods were carried out in accordance with relevant guidelines and regulations.