Page 102 - GTM-3-4

Global Translational Medicine Prediction of in-stent restenosis

concordance (C)-index (CIH), the measure of explained collected all data required based on the retrospective
2
randomness ( R mer ), and the measure of explained deviation analysis from 2015 to 2020 (5-year sample). After excluding
2
( R mev ) were considered as the quality metrics of the Cox incomplete and low-quality data, the total sample size
model parameterization. The measures are defined as was 798 patients and nine risk factors (variables). The
follows: number of risk factors was reduced from 12 to 9, as three
manuscripts lack qualitative information for quantitative
 2 
1 exp
R 2 =−  ( l − l)  (IV) analysis. Potential risk predictors were selected based on
mer
 n  the scientific articles reviewed. In 516 cases, restenosis
developed within 5 years (60 months) after stent
R 2
R 2 mev = R ( 2 +π / 61⋅ −( R ) ) (V) implantation. In 282 cases, no hemodynamically significant
mer
2
restenosis was reported. Table 2 presents the comparison
mer
mer
of two groups (patients with vs. without restenosis) based
where l is the logarithm of the partial likelihood on the frequency of coronary restenosis risk factors. The
function, l is the logarithm of the restricted (0 for all median (Me), as well as the first (Q ) and third quartiles
1
regressors) partial likelihood function, and n is the number (Q3), were calculated for continuous attributes, while the
of patients. It was believed that the explanatory power of incidence (%) and relative frequency (%) were calculated
the Cox model is higher if the quality metrics are close to 1. for frequency attributes. The P-level for frequency
attributes was calculated based on X (the criterion with
2
Simulation results were interpreted based on the likelihood correction for continuous attributes) using the
calculation using estimated hazard ratio (HR) regression Wald-Wolfowitz runs test.
coefficients:
Group comparisons indicated that the following
tx )
λ (| predictors are statistically significant (P < 0.05): male sex,
x )
HR x() = λ t () i = exp( β (VI) history of myocardial infarction, nominal stent diameter,
i
i
0
and the presence of stent coating.
In addition, we assumed that the baseline risk function
λ (t) depends on time t, but the risk factors were not Figure 3 displays the survival curves (before restenosis)
i
independent of the time t. The 95% confidence interval of patients with DES/BMS. The graph illustrates that the
(CI) was calculated using the Greenwood formula: type of stent significantly affects restenosis development.
BMS in the first few months leads to a dramatic increase in
(
( )) (
ˆ
ˆ
D St ~ St 2 jt ≤t nn d j −d ) (VII) the incidence of restenosis, and the likelihood of no adverse
(
events occurring within 60 months is drastically reduced.
( )) ∑ : j
j
j
j
This is confirmed by the log-rank test, which rejects the null
hypothesis of no difference in the likelihood of restenosis
3. Results at 60 months after BMS/DES stenting (P < 0.001).
3.1. Systematic analysis results In 2016, Buccheri et al. performed a systematic review
13
In the first stage, we analyzed publications in the PubMed of multiple studies, evaluating risk factors, survival, and
and eLibrary databases. A total of 13,775 full-text articles the frequency of adverse events in patients who underwent
were identified using the keyword “coronary in-stent implantation of different stents. Our results are consistent
restenosis” (PubMed: 11,950; eLibrary: 1,825). At the with the findings of this review.
primary screening stage, 997 articles remained after Using the selected risk factors as potential predictors for
excluding duplicates and articles that did not correspond to coronary restenosis, a Cox model was plotted. The model
the topic. After a detailed review of the abstracts, 61 full-text was estimated using the Efron partial likelihood method, as
publications were selected for further analysis. The aim of it yielded lower Akaike and Schwartz information criteria
the systematic analysis was to identify potential risk factors values compared to the Breslow method. Table 3 presents
for coronary restenosis. In the final selection, only six full- Cox regression coefficient values for each predictor,
text articles remained (Table 1), enabling the identification calculated based on the HR, the 95% CI limits (determined
of 12 restenosis variables. A detailed analysis of publications using Greenwood’s formula), and the corresponding
matching the search criteria is presented in Figure 2. p-level of error for rejecting the null hypothesis (i.e., model
coefficient estimate = 0). Model quality is relatively high:
3.2. Retrospective analysis results CIH (Harrell’s C-index) = 0.615; R 2 = 0.71; R 2 = 0.74.
mer
mev
In the second stage, we analyzed the identified risk factors Moreover, the LR test confirmed the significance of the
for coronary restenosis from the systematic review. We overall Cox model equation (LR = 15.84; P < 0.001).
Volume 3 Issue 4 (2024) 4 doi: 10.36922/gtm.4957