Anzeige
Erschienen in:
06.06.2023 | Author's Reply
Answer to the Letter to the Editor of Hadi Raeisi Shahraki concerning “Predictive factors for respiratory failure and in-hospital mortality after surgery for spinal metastasis” by Jaipanya P, et al. (Eur Spine J [2023]: doi: 10.1007/s00586-023-07638-z)
Erschienen in: European Spine Journal | Ausgabe 8/2023
Einloggen, um Zugang zu erhaltenExcerpt
We thank for the interest in our paper [1] and for your queries and comments. Regarding the statistical analyses of the present study, we would like to clarify that the statistic was performed by a biomedical statistician. We would like to explain our statistics as the details below.
1.
Using the logistic regression for rare events modeling unplanned postoperative intubation (UPI), mechanical ventilation dependence (MVD) and in-hospital mortality only occurred for a few percent of patients, indicating the presence of rare events. But in rare event settings, the results from logistic regression estimates can be biased and may not be valid.
Response The authors understand your concern about the possible bias that may occur due to the presence of rare events. The frequency of events in this present study was around 5.5–5.9%. We considered this as small number of events. Some variables that were shown to have association with postoperative unplanned intubation (UPI), requirement of postoperative mechanical ventilation > 48 h (MVD), and in-hospital mortality have displayed wide range of confidence interval. This may be due to small number of events of the present study. Thus, we recommend to carefully interpret the results of the present study and meticulously clinical application.
2.
Statistically significance is considered as being a predictor. Although there is a fundamental difference between explanation models and prediction models, these two types of modeling are often conflated. Due to the fact that significant variables in a regression model do not guarantee prediction performance, the term “significant predictor” is not correct.
Response Thank you for your consideration of the fundamental of predictors. Our study aims to identify the factors those associate with the occurrence of UPI, MVD, and in-hospital mortality based on logistic regression model. The primary objective did not yield a predictive model, which is beyond the scope of this present study. In our opinion, the factors those were identified in the multivariate logistic regression analysis were clinically relevant and comparable to previous studies [2‐4]. We hope this publication encourage researchers to pursue further larger cohort studies, particularly for predictive model for clinical applications, and contribute to the growing body of literature.
3.
Lack of any goodness-of-fit test or indicator. To show the prediction capability of a variable, a variety of indicators is applicable. But lack of related findings to show the prediction accuracy of the proposed variables and models is questionable.
Response To show the prediction accuracy of the proposed variables, the models were evaluated using the Hosmer–Lemeshow goodness-of-fit test. The p values for the Hosmer–Lemeshow goodness-of-fit test of associated factors for requirement of postoperative mechanical ventilation > 48 h, associated factors for postoperative unplanned intubation and associated factors for in-hospital mortality were 0.1773, 0.8944 and 0.2065, respectively, those were greater than the significant threshold. The area under receiver operating characteristic curve of associated factors for requirement of postoperative mechanical ventilation > 48 h, associated factors for postoperative unplanned intubation and associated factors for in-hospital mortality were 0.8505, 0.9531 and 0.9146, respectively.
4.
Variables with the largest OR were introduced as the most associated factors. Larger coefficients don’t necessarily identify more important predictor variables and lower p values don’t necessarily identify independent variables that are practically important. So, to identify the most associated factors the standardized regression coefficients must be computed.
Response We agree that larger coefficients do not necessarily identify more important predictor variables. We computed the standardized regression coefficients of the associated factors for requirement of postoperative mechanical ventilation > 48 h, associated factors for postoperative unplanned intubation and associated factors for in-hospital mortality. The standardized regression coefficients were shown in the below table.
|
Variables
|
Standardized regression coefficients
|
|---|---|
|
Associated factors for requirement of postoperative mechanical ventilation > 48 h
|
|
|
Measured blood loss > 2000 mL
|
1.097
|
|
Surgery involves cervical spine
|
0.862
|
|
ASA classification ≥ 4
|
0.659
|
|
Associated factors for postoperative unplanned intubation
|
|
|
Postoperative sepsis
|
1.008
|
|
Lung metastasis
|
0.977
|
|
Postoperative pulmonary complications
|
0.642
|
|
CNS metastasis
|
0.512
|
|
Associated factors for in-hospital mortality
|
|
|
Postoperative sepsis
|
0.860
|
|
Postoperative pulmonary complications
|
0.764
|
|
CNS metastasis
|
0.519
|