ICM: 经胸超声和脓毒症的病死率：MIMIC-III期数据库分析_复现

Abstract

Purpose: While the use of transthoracic echocardiography (TTE) in the ICU is rapidly expanding, the contribution of TTE to altering patient outcomes among ICU patients with sepsis has not been examined. This study was designed to examine the association of TTE with 28‑day mortality specifically in that population.

目的：尽管经胸超声（TTE）的应用在ICU中正在快速地普及，但是，在ICU的脓毒症患者中，TTE是否能改变其预后仍未得到证实。该研究的目的是验证该人群中TTE的应用与28天病死率之间的相关性。

Methods and results: The MIMIC ‑III database was employed to identify patients with sepsis who had and had not received TTE. The statistical approaches utilized included multivariate regression, propensity score analysis, doubly robust estimation, the gradient boosted model, and an inverse probability‑weighting model to ensure the robustness of our findings. Significant benefit in terms of 28‑day mortality was observed among the TTE patients compared to the control (no TTE) group (odds ratio = 0.78, 95% CI 0.68–0.90, p < 0.001). The amount of fluid administered (2.5 vs. 2.1 L on day 1, p < 0.001), use of dobutamine (2% vs. 1%,p = 0.007), and the maximum dose of norepinephrine (1.4 vs. 1 mg/min, p = 0.001) were significantly higher for the TTE patients. Importantly, the TTE patients were weaned off vasopressors more quickly than those in the no TTE group (vasopressor‑free days on day 28 of 21 vs. 19, p = 0.004).

方法和结果： MIMI-III期数据库纳入了已经进行和未进行TTE检查的脓毒症患者。为确保研究结果的可靠性，涉及的统计方法包括：多变量回归、倾向性分析、双重稳健估计、梯度提升模型和逆概率加权模型。与对照组（未应用TTE的患者）相比，我们观察到，就28天病死率这个指标来说，干预组组明显获益于TTE的应用（OR=0.78.95%CI 0.68-0.90，p <0.001）。对照组的第一天液体入量（2.5 vs 2.1/L），多巴酚丁胺的应用比例（2% vs 1%，p=0.007）和去甲肾上腺素的最大应用剂量（1.4 vs 1.mg/min，p=0.001）均高于对照组。更重要的是，与对照组相比，干预组停用血管活性药更快（28天内无血管活性药物的应用天数分别为21 vs 19，p=0.004）。

Conclusion: In a general population of critically ill patients with sepsis, use of TTE is associated with an improvement in 28‑day mortality.

结论：在脓毒症患者中，TTE的应用与患者28天病死率的改善具有相关性。

本文想复现论文中的所有数据，文章中已提及所有代码可在https://github.com/nus-mornin-lab/echo-mimiciii找到。可是发现需要用到R语言，不会呀。所以使用本方法把所有sql语言都跑出来作为实体化视图。先产生以下几张视图化视图。

1、Study cohort

上述流程按照代码完全可以复现出来，最终形成cohort这张表，选取echo=1则有3262条记录。这张cohort我们可以看到有以下信息：icustay_id,subject_id, hadm_id, 第一次入icu的科室（MICU、SICU）,开始和结束的时间，年龄，什么时候做的超声。

2、Table 1

table1的资料主要来源于表merged_data，苗黄部分证实一致，使用mean()；绿色的计量占比可先用value_counts()，a = df1[‘icd_chf’].value_counts()，然后print(a[1]/(a[0]+a[1]))。通过这张表格基本可以得到所有数据。

3、三天的液体出入量

先有fluid_input.xls、fluid_output.xls、merged_data.xls这三张表，然后合并出final_fluid.xlsx。大致差不多。

import numpy as np 
import pandas as pd
from pathlib import Path
import matplotlib.pyplot as plt 


fuildinput = Path("./fluid_input.xls")
fuildoutput = Path("./fluid_output.xls")
mergeddata = Path("./merged_data.xls")

df_1 = pd.read_excel(fuildinput)
df_2 = pd.read_excel(fuildoutput)
df_3 = pd.merge(df_1, df_2, on= ['icustay_id','nday'], how = 'outer')
df_4 = pd.read_excel(mergeddata)

df_3['amount'] = df_3['amount_x'] - df_3['amount_y']

df_day1 = df_3[df_3['nday'] == 1][['icustay_id','amount']]
df_day2 = df_3[df_3['nday'] == 2][['icustay_id','amount']]
df_day3 = df_3[df_3['nday'] == 3][['icustay_id','amount']]

df_day1_day2 = pd.merge(df_day1,df_day2,on=['icustay_id'],how = 'outer')
df_final = pd.merge(df_day1_day2,df_day3,on=['icustay_id'],how = 'outer')
df_final = pd.merge(df_4[['icustay_id','echo']],df_final,how = 'left')

#df_final = df_final.rename(columns={'amount_x':'day1','amount_y':'day2','amount':'day3'})
new_names =['icustay_id','echo','day1','day2','day3']
df_final.columns = new_names



df_final.sort_values(by=['icustay_id'], ascending=True, inplace=True)
df_final.index = range(len(df_final))

df_final.to_excel('final_fluid.xlsx')