Fixed-effects model | Cox proportional hazard model | |||
β (95% CI) | P value | HR (95% CI) | P value | |
Main analysis | 0.015 (0.011 to 0.019) | <0.001 | 1.96 (1.49 to 2.56) | <0.001 |
Multiple imputed† | 0.013 (0.008 to 0.018) | <0.001 | 1.95 (1.61 to 2.36) | <0.001 |
With participants answering less than 45 items included‡ | 0.019 (0.011 to 0.026) | <0.001 | 1.91 (1.47 to 2.48) | <0.001 |
Marginal structural model§ | – | – | 1.65 (1.26 to 2.15) | <0.001 |
For the frail respondents in the baseline¶ | 0.015 (0.004 to 0.027) | 0.009 | 0.61 (0.41 to 0.93) | 0.020 |
*The reference group was the well-controlled BP group.
†The imputing method was from the posterior predictive distribution. Five sets of data were generated, and the regression coefficients were pooled.
‡We included participants who answered fewer than 45 items on the frailty-related index and repeated the analysis. There were 3349 respondents in 2013, 3121 in 2015 and 2868 in 2018. For the Cox regression, 2010 respondents were included in the analysis.
§Antihypertensive treatment and comorbidities were adjusted as time-varying confounders, and other covariates were adjusted as fixed confounders. Inverse probability of treatment weighting was used.
¶A total of 1322 respondents categorised into the frail group at baseline (2013) were included in this sensitivity analysis. In the Cox regression model, the improvement of the frailty state (from frail to non-frail) was set as the outcome.