Table 3.
Univariate and multivariate logistic regression models exploring the relationship between frailty (i.e.: Rockwood score >4) and comorbidity (i.e.: Charlson Score ≥4).
| Logistic regression models – Outcome : to be frail (ie having CSHA Clinical Frailty Scale score >4) We analyzed 2280 elderly people assisted by doctors with good quality of data-input (see main text, see Appendix D) | |||
|---|---|---|---|
| |
I.Monovariate logistic models OR (CI 95%) Z test P |
II.Multivariate logistic model without interactions OR (CI 95%) Z test P |
III.Multivariate logistic model with interactions OR (CI 95%) Z test P |
| To be female | OR = 1.83 (1.49–2.25) p < .0001 |
OR = 0.46 (0.22–0.69) p < .0001 |
OR = 1.58 (1.25–2.01) p < .0001 |
| Having a Charlson Score ≥4 | OR = 3.64 (2.93–4.51) p < .0001 |
OR = 1.18 (0.94–1.43) p < .0001 |
OR = 4.21 (3.20–5.53) p < .0001 |
| To be aged ≥85 | OR = 12.64 (9.77–16.34) p < .0001 |
OR = 2.42 (2.16–2.69) p < .0001 |
OR = 15.78 (11.44–21.76) p < .0001 |
| Age ≥ 85 X Clarlson ≥4 | – | – | ROR** = 0.33 (0.19–0.58)p < .0001 |
| Model diagnostics | |||
| Pregibon test | – | Z -3.07 p = .002 |
Z = 0.32 p = .748 |
| Hosmer-Lemeshow test | – | Chi2 = 18.18 df = 4 p = .0011 |
Chi2 = 3.05 df = 4 p = .54 |
| AIC statistic | – | 1952.257 | 1939.972 |
*Note: the exponentialized coefficient of the interaction variable is a ratio of odds ratios ROR.
We explored the details of the relationship between frailty and comorbidity through three models of logistic regression in which the condition of frailty was the outcome and age, sex and comorbidity were the predictors.
The multivariate model without interactions (II) does not fit well: this can be seen from the outpouts of the Pregibon test (p = .002) and the Hosmer-Lemeshow test (p = .0011); the AIC statistic (1952.27 versus 1939.972) shows also that the informative contribute is worst respect that of model III (i.e. that with the interaction).
The multivariate model with interactions (III) shows the better goodness of fit (Hosmer-Lemeshow test p = .54) and the better pattern of covariates (Pregibons test p = .748). To be female shows to be a significant predictor of frailty status and Age shows to be a significant confounder in the relations between comorbidity and frailty.
In detail, multimorbidity has been shown to be an independent predictor of frailty only for patients under the age of 85. The linear combination of the nonexponentialized coefficients of the variables involved in the interaction (third model of Table 4) allowed in fact to calculate for patients with multimorbidity (i.e. with Charlson score ≥4) compared to those without multimorbidity (i.e. with Charlson score <4) an Odds Ratio of frailty correspondent to OR = 4.21 (3.20–5.53) p < .0001 in subjects under 85 years and to OR = 1.42 (0.88–2.29) p = .149 in subjects 85+ years old, respectively. So, a serious comorbidity shows to be a prognostic factor for frailty only under 85 years of age.