Page 95 - 南京医科大学自然版

第46卷第8期 陈芳怡，王 威，胡馨茹，等. 心理韧性与中国老年人全因死亡率关联强度的异质性：一项前瞻性队列
2026年8月 研究［J］. 南京医科大学学报（自然科学版），2026，46（8）：1197-1210 ·1201 ·

able and multivariable Cox proportional hazards regres⁃
2 Results
sion models，as well as Cox models applied to data ad⁃
justed by propensity score matching（PSM）and inverse 2.1 Participants’characteristics and between⁃group
probability weighting（IPW）. The proportional hazards comparisons
assumption was assessed using the Schoenfeld test. A total of 7 481 participants were included，com⁃
（HR）and 95% confidence interval（CI）were estimat⁃ prising 3 214（42.96%）in the low resilience group and
ed to quantify the magnitude of these associations. Ad⁃ 4 267（57.04%）in the high resilience group. The over⁃
ditionally，Kaplan⁃Meier survival curves and log⁃rank all mean age was（84.44±10.77）years. Slightly more
tests were used to compare cumulative all⁃cause mor⁃ than half of the participants were female（53.05%）or
tality between the high and low resilience groups across residing in rural areas（51.25%）. The majority were of
the raw，PSM⁃adjusted，and IPW⁃adjusted datasets. Han ethnicity（94.28%），had insurance（89.97%），did
In consideration of the impact of obesity，the het⁃ not live alone（82.62%），were never⁃smokers（64.06%），
erogeneity in the association between psychological re⁃ were never⁃drinkers（66.86%），and reported no depres⁃
silience and all⁃cause mortality risk was analyzed us⁃ sive symptoms（81.87%）. The mean BMI was（21.05±
2
ing the STEPP method in PSM data. STEPP requires 4.37）kg/m ，and the mean psychological resilience
the independent variable to be binary. The analysis em⁃ score was 13.99±3.02. During a median follow⁃up of 1
ployed four obesity indices（ABSI，BRI，WHtR，and 408.0（858.0，2 482.0）days，statistical significance
BMI）as covariates for splitting of subpopulations. The was observed in all baseline characteristics between the
STEPP analysis followed four steps：① The sample low and the high resilience groups（P < 0.05，Table 1）.
size of subpopulations and overlapping samples were 2.2 Association of psychological resilience and all ⁃
set，and all participants were divided into multiple cause mortality
overlapping subpopulations according to the continu⁃ The RCS curve showed a non⁃linear relationship
ous covariation；② In each subpopulation，the associa⁃ between psychological resilience and all⁃cause mortali⁃
tion and its strength were estimated by Cox proportion⁃ ty（Figure 2）. Significant differences in all⁃cause mor⁃
al hazards regression models；③ The heterogeneity of tality rates were observed between the high and low re⁃
the association strength among subpopulations was ana⁃ silience groups（all P < 0.05）（Figure 3A）. The as⁃
lyzed；④ The median of obesity indices in each sub⁃ sumption for the Cox proportional hazards model hasn’t
population was plotted on the horizontal axis，and the been violated（P > 0.05）. In the univariable Cox pro⁃
corresponding HR in each subpopulation was plotted portional hazards regression model，high psychological
on the vertical axis. To meet the sample size of subpop⁃ resilience was associated with a 34.6% lower risk of all⁃
ulations and partition as many subpopulations as possi⁃ cause mortality compared with low psychological resil⁃
ble，a sliding ⁃ window subpopulation partition method ience（HR=0.654，95% CI：0.612-0.698，P < 0.001）
was used in STEPP（10）. The subpopulation size was （Figure 3B）. After multivariable adjustment，high psy⁃
set as 2 692（60% of the total PSM sample），with an chological resilience remained associated with a re⁃
overlap of 2 422 participants（90% of the subpopulation duced mortality risk（HR=0.911，95% CI：0.846-
size）. Additionally，sensitivity analyses were conducted 0.980，P=0.012），corresponding to an 8.9% risk reduc⁃
to evaluate the robustness of our findings. The random tion（Figure 3B）. Similar associations were observed in
forest method was employed to impute covariates with the PSM⁃adjusted model（HR=0.883，95% CI：0.813-
a missing proportion below 30% to test the results of 0.960，P=0.003）and the IPW ⁃ adjusted model（HR=
the multiple imputation method. 0.916，95% CI：0.848- 0.990，P=0.027）（Figure 3B）.
Statistical analyses and data visualization were Additionally，Kaplan⁃Meier curves with log⁃rank tests
performed by R 4.3.0 and Excel 2021. A two ⁃ tailed based on the full cohort，PSM，and IPW samples all
test was used，and the P value < 0.05 was considered demonstrated significantly lower all⁃cause mortality in
statistically significant. the high resilience group compared with the low resil⁃