Page 101 - GTM-3-4

Global Translational Medicine Prediction of in-stent restenosis

To illustrate the results of the evaluation, Kaplan-Meier
curves were plotted on the survival function graph
(with restenosis as the endpoint), where the x-axis is the
observation period (before the development or absence of
restenosis; defined in months). The log-rank test was used
to compare the time intervals before restenosis in groups
with or without a certain variable (e.g., stent type [DES
vs. BMS]), with the null hypothesis stating no differences
between the groups. The significance level for rejecting the
null hypothesis was P < 0.05.
To confirm the influence of potential risk predictors
identified in the first stage of the study on restenosis
development, we performed survival analysis using Cox
regression with multiple variables:
λ t () = λ t ⋅() exp ( β x +…+ β x + β ) (II)
1
i
0
ki
0
i 1
k
where λ (t) is the risk of restenosis in the i-th patient
i
during the observation period t; λ (t) is the baseline
0
risk of restenosis in each patient by default; x ,… x is
ki
1i
the potential risk factor(s) of restenosis; β ,…β is the
1
k
coefficient(s) of the regressors identified in the first stage
Figure 1. Retrospective analysis design of the study and evaluated using the partial likelihood
method; and β is the intercept. The coefficients of the
0
requiring continuous glucocorticosteroids and another regressors in the Cox model were estimated using the
controller therapy; (vi) familial hypercholesterolemia; and maximum partial likelihood method with the partial
(vii) cancer that required chemotherapy and radiation likelihood function according to the Efron and the Breslow
therapy after percutaneous coronary intervention. formulas. For both formulas (Equations I and II), the
Akaike and Schwartz information criteria values (AIC and
2.3. Statistical analysis BIC) were calculated. The model was estimated using the
Patients were divided into two groups based on the absence Efron partial likelihood method, as it yielded lower Akaike
(control group) or presence (restenosis group) of re-stenosis and Schwartz information criteria values compared to the
requiring repeat percutaneous coronary intervention. Breslow method (for Breslow: AIC=3282.33, BIC=3313.58;
The restenosis group included 516 patients, while the for Efron AIC=3280.96, BIC=3312.21). The form of the
control group included patients without hemodynamically partial likelihood function was selected based on the
significant restenosis. The endpoints (death, acute smallest information criteria values.
myocardial infarction, acute cerebrovascular accident, Statistical significance of risk predictors was tested
repeat hospitalization, and repeat revascularization) were according to the Wald test at a significance level of
determined for all patients within 5 years. R Studio was p < 0.05; the null hypothesis was the assumption that the
used in packages (“survival,” “survMisc,” and “survminer”). regressor coefficient = 0, that is, there was no impact of
Single- and multivariate survival analyses (e.g., Cox the investigated factor on the risk of restenosis. The Wald
regressions) and Kaplan-Meier analysis were performed. statistic (Z ) was calculated as follows:
The latter was used to estimate the differences in function w
S(t) before the development or absence of restenosis in β
ˆ
different stent types: Z W = SE β ( ) (III)
ˆ
( )
ˆ
ˆ
St [ (nj ) ] (I) where SE β () is the estimated standard error β .
nj
( −+ )
To assess the quality of the Cox model, the likelihood
where n is the total number of observations; d is equal to ratio (LR) test was performed with the null hypothesis of
j
1 if the event occurred during the considered observation no significance in the general model. The hypothesis was
period or 0 – if the event was censored. rejected in favor of the alternative at P < 0.05. Harrell’s
Volume 3 Issue 4 (2024) 3 doi: 10.36922/gtm.4957