class: center, middle, inverse, title-slide .title[ # 免疫相关性肺炎预测项目复盘 ] .author[ ### Lin Yu ] .date[ ### 2023-07-31 ] --- ### Timeline milestone: - 2023/01/05: 投稿 - 2023/02/22: 审稿人1首次反馈 - 2023/05/11: 审稿人2首次反馈 - 2023/06/12: 接收 - 2023/06/29: 见刊
--- ## Overview - **outcome of interest**: IRP - **sample size**: 190 (48 cases & 142 non-cases) - **study design**: retrospective case control study - **method(s)**: - **Elastic Net** model was constructed using a **repeated k-fold cross-validation** framework (repeat = 10; k = 3). - 100 combination of hyper-parameters were tuned: 10 alphas + 10 lambdas - missing rate < **15%** included - CV ratio: **8:2** (37 pts for 'external' validation) **result(s)**: 11 predictors included; **"external"** AUC of 0.81 (95% CI 0.58–0.90), AP of 0.76, scaled Brier score of 0.31, and Spiegelhalter-z of −0.29 (P-value:0.77). --- ## Methods <img src="data:image/png;base64,#f1.png" width="100%" /> --- ## Feedback - C1.1: It is unclear how the cohort was selected. What is meant with the ratio of 1:4 (**Figure1**) - C1.2: The results of the univariate tests are not corrected for multiple testing, and it is unclear which statistical test was used. - C1.3: Pre-selecting features based on p-values for differences between IRP and non-IRP is **overfitting** when done on all samples. - C1.4: It is unclear how the "final prediction model" used on the external test data was established. - C2.1: In general, the cohort size used for modeling is **quite small** – a total of 190 patients only 42 of which had immune related pneumonitis, and while they did perform some cross validation tests during their machine learning analysis, it is unclear how their algorithm would **generalize to other populations of patients**. - C2.2: there is a significant **skew in the sexes of the patients**. almost 80% are male It is unlikely that their model is relevant for female patients given their low representation. - C2.3: There is **no definition** of when the laboratories were performed and what time points were selected for the modeling. --- ## Summary - 代码公开:https://github.com/gongli0707/IRP-prediction/tree/main/IRPproject ; - 借助流程图:纳排图、模型步骤图(+说明文字); - 变量的收集需要描述清楚,如体温 (首次就诊?最近一次就诊?多次就诊平均值?) - 不足点应主动在讨论部分指出; - rebuttal letter引用其他文献支撑自己的观点! - 写作习惯: manuscript_main_text.docx + manuscript_main_displays.docx + manuscript_supplement.docx - response letter:
Download Response to reviewer comments0428.docx
Download Response to reviewer comments0516.docx
Download Response to reviewer3 comments_0607.docx