orcid：<https://orcid.org/0000-0003-4737-7975> 邮箱：liyd39@mail.sysu.edu.cn

基本信息

心血管内科-重症医学 (2006至今)

生于80年代的苦逼内宅小医生一枚，一直在繁忙的科室工作。

Github： github.com/shaou77 | 个人博客： shaou77.github.io

专业技能

  • 心内科知识：有一点积累，心电图略懂一二，但是最新知识可能已经过时。

  • 重症知识：半路出家，根基不牢，半懂不懂，呼吸和抗感染是短板。

  • 编辑知识：40岁考了一个国家出版专业技术人员中级，略懂一二。但不可否认，从地级市医院到教学医院，在临床医生向研究型医生转型早期，审稿工作对于思维训练获益很大。

  • 医学研究：擅长用别人的数据做自己的研究（即所谓大数据研究）。对R和回顾性研究有一点心得。在AI年代，善于复现别人的研究（就是自己没什么想法），但往往复现的研究发表不出来啊。(╯°Д°)╯︵┻━┻

研究经历

GitHub - https://github.com/shaou77/

  • 本人做的基于公共数据库的大数据研究，都会放在我的GitHub上。毕竟在升级早期，站上的大神们的代码对我帮助很大，我也做一点微小的工作，把自己微小的火花传下去，只是质量不高罢了。

Paper1 - Association Between Red Cell Distribution Width and Hospital Mortality in Patients with Sepsis

  • DOI: 10.1177/03000605211004221

  • 代码地址：https://github.com/shaou77/rdw-and-sepsis

  • 第一篇SCI，从0到1的突破。只是当时心里承受能力轻，投了一两家被拒了，就投了一个大家说好发的。——结果怎么投的时候还没事，发的时候就成了预警期刊呢？你一个Sage出版社，比我年纪还大的期刊啊……晋升没用上，版面费也没报到。o(▼皿▼メ;)o

  • 但是文章质量我觉得还可以。也是我首篇被引用数达两位数的文章。样本量大，传统统计方法做的很全面，亚组分析完善，甚至当时时髦的NRI，IDI都做了。只是开始不会用Rstudio的r或者rmd文件，代码都是用文本保存的，可能有一些偏差。

  • 补充：现在已经不是预警了。

Paper2 - Albumin Level at Admission to the Intensive Care Unit Is Associated With Prognosis in Cardiac Arrest Patients

  • DOI: 10.7759/cureus.14501

  • 代码地址 ：无

  • 其实是第一篇的延伸和模仿，当时也是脸皮薄，拒了几次就投了一个E-SCI，很快发了，无心插柳，晋升反而用到了。非SCI，代码也懒得维护了。

Paper3 - Red blood cell distribution width predicts in-hospital mortality in patients with a primary diagnosis of seizures in the ICU: a retrospective database study

Paper4 - Development and validation of a nomogram for predicting atrial fibrillation in patients with acute heart failure admitted to the ICU: a retrospective cohort study

Paper5 - Interpretable machine learning models for predicting 90-day death in patients in the intensive care unit with epilepsy

  • DOI: 10.1016/j.seizure.2023.11.017

  • 代码地址：https://github.com/shaou77/rdw_seizures （没错，和paper3一个仓库，发表的时候忘记改仓库地址了，只好硬着头皮把它维护进去了）

  • 参加了张敬信老师mlr3寒假学习班的作品。（感谢当时二区开区，很多课我都能按时上）学习机器学习的作品。模范了一篇也是使用mlr3的文章。从数据清洗，模型调参，模型比较，模型验证到模型解释的全套。不得不说，mlr3包确实强大。

  • 太太发得比我好系列。3区，癫痫领域老牌期刊，epilepsy的预备役。

  • 有GPT帮忙，代码已经简洁美观多了。

Paper6 - Glycemic Comparison Index (GCI): a retrospective analysis of its prognostic value in ICU patients with AMI and diabetes

看到有一个一区期刊，分数9分多，糖尿病冠心病交叉领域，数据库文章也能发。我就冲着杂志主题做了一个研究，还是用我熟悉的数据库，花了一个春节假期的时间，做出了初稿。但是仍然被桌拒。然后投了几个糖尿病杂志，辗转一年，最终还是被老熟人，Springernature旗下的BMC系列接收了。

Paper7 -Real-world efficacy and safety of perampanel versus levetiracetam monotherapy for focal epilepsy: A single-center retrospective study

终于有了一篇非数据库的文章，太太的横向课题，我做了所有的统计分析，撰写了结果和方法部分。代码不开源。
*DOI:[10.1016/j.yebeh.2025.110630] (https://doi.org/10.1016/j.yebeh.2025.110630)

Paper 8 - A Nomogram Prediction Model and Validation of Hypoglycemia Risk in Patients With Decompensated Cirrhosis and Type 2 Diabetes

这是继 Paper7 之后的第二篇非公共数据库文章——也别问是不是"完全独立"完成的。数据是合作者在广东几家医院，一个病人一个病人翻病历扒出来的，真正的脏活。我负责全部统计和 nomogram 这一摊。

*DOI:[10.1155/jdr/6812038] (https://doi.org/10.1155/jdr/6812038)

Paper 9 - Early albumin administration and mortality in ICU patients with heart failure and hypoalbuminemia: a target trial emulation

MIMIC 数据库，模范了别人做 TTE（target trial emulation，目标试验仿真）的文章。这两年 TTE 逐渐被接受，自己也想试一把。没想到屡投屡拒——不是内容的问题，编辑/审稿人一看是公共数据库来源就直接放一边。最后还是投到了 BMC Cardiovascular Disorders——自己熟悉的栏目、熟悉的编辑流程，2026 年 8 月总算接收。

正好赶上医院大变革的尾巴。2万多银子啊！

*DOI:[10.1186/s12872-026-06514-0] (https://doi.org/10.1186/s12872-026-06514-0)

  • 代码地址：https://github.com/shaou77/AlbinHF_tte

以后还有发表，本文将持续更新。（我巴不得月月更新）

Basic Information

Cardiovascular Medicine - Critical Care (Since 2006)

A hardworking doctor born in the 80s, always working in a busy department.

Github: github.com/shaou77 | Personal Blog: shaou77.github.io

Professional Skills

  • Cardiology Knowledge: Some accumulation, with a basic understanding of ECG, but the latest knowledge may be outdated.

  • Critical Care Knowledge: An amateur in this field with a shaky foundation. Respiratory and anti-infection areas are my weak points.

  • Editing Knowledge: Obtained a national intermediate-level publishing professional qualification at the age of 40. A bit of knowledge, but it's undeniable that during the early transition from a clinical doctor to a research-oriented doctor, the review work greatly benefited my thinking training.

  • Medical Research: Proficient in using other people's data for my research (so-called big data research). I have some experience with R and retrospective studies. In the AI era, I'm good at reproducing others' research (just lack original ideas), but reproduced studies are often hard to publish. (╯°Д°)╯︵┻━┻

Research Experience

GitHub - https://github.com/shaou77/

  • All my big data research based on public databases is available on my GitHub. After all, in the early stages of my career, the codes from experts on the platform were very helpful to me. I also do a little work to pass on my tiny sparks, although the quality is not high.

Paper1 - Association Between Red Cell Distribution Width and Hospital Mortality in Patients with Sepsis

  • DOI: 10.1177/03000605211004221

  • Code: https://github.com/shaou77/rdw-and-sepsis

  • My first SCI, a breakthrough from 0 to 1. Initially, I was not very resilient and submitted to a few journals, got rejected, and finally submitted to one that was said to be easy to publish in. Ironically, it was fine when I submitted, but when it was published, it became a warning journal. You, a journal from Sage Publications older than me... It wasn't useful for my promotion, and the publication fee wasn't reimbursed. o(▼皿▼メ;)o

  • However, I think the quality of the article is okay. It was also my first paper to be cited double-digit times. Large sample size, comprehensive traditional statistical methods, detailed subgroup analysis, and even the then-fashionable NRI and IDI were done. I just didn't know how to use Rstudio's r or rmd files, so the code was saved in text, which might have some deviations.

  • Note: It is no longer a warning journal now.

Paper2 - Albumin Level at Admission to the Intensive Care Unit Is Associated With Prognosis in Cardiac Arrest Patients

  • DOI: 10.7759/cureus.14501

  • Code: None

  • It's actually an extension and imitation of the first paper. Back then, I was thin-skinned, got rejected a few times, and submitted to an E-SCI journal, which got published quickly. Surprisingly, it was useful for my promotion. Non-SCI, and I was too lazy to maintain the code.

Paper3 - Red blood cell distribution width predicts in-hospital mortality in patients with a primary diagnosis of seizures in the ICU: a retrospective database study

Paper4 - Development and validation of a nomogram for predicting atrial fibrillation in patients with acute heart failure admitted to the ICU: a retrospective cohort study

  • DOI: 10.1186/S12872-022-02973-3

  • Code: https://github.com/shaou77/AFinAHF

  • Successful financial management series (Q4 when published, Q3 the next year). A work while learning prediction models, standard and compliant, containing everything a standard prediction model article should have, and nothing it shouldn't.

Paper5 - Interpretable machine learning models for predicting 90-day death in patients in the intensive care unit with epilepsy

  • DOI: 10.1016/j.seizure.2023.11.017

  • Code: https://github.com/shaou77/rdw_seizures (Yes, same repository as paper3, forgot to change the repository address when publishing, so I had to maintain it under pressure)

  • A product of Professor Zhang Jingxin's mlr3 winter course. (Thanks to the temporary ICU ward II, I could attend many classes on time) A work while learning machine learning. Modeled after an article also using mlr3. From data cleaning, model tuning, model comparison, model validation to model interpretation, a full set. I must say, the mlr3 package is indeed powerful.

  • My wife published better than me series. Q3, an established journal in the field of epilepsy, a reserve for epilepsy.

Paper 6 - Glycemic Comparison Index (GCI): a retrospective analysis of its prognostic value in ICU patients with AMI and diabetes

I came across a Q1 journal with an impact factor above 9 that focuses on the intersection of diabetes and coronary heart disease, and it also accepts database-based studies. Motivated by the journal’s scope, I designed this study using a database I was already familiar with. I spent the entire Spring Festival holiday completing the first draft. The manuscript was desk-rejected at first. I then submitted it to several diabetes-related journals, and after nearly a year of revisions and resubmissions, it was eventually accepted by the BMC series under Springer Nature, a journal family I know well.

Paper 7 - Real-world efficacy and safety of perampanel versus levetiracetam monotherapy for focal epilepsy: A single-center retrospective study

This is finally a study that does not rely on a public database. It originated from my wife’s horizontal research project. I was responsible for all statistical analyses and for writing the Methods and Results sections. The code for this study is not publicly released.

Paper 8 - A Nomogram Prediction Model and Validation of Hypoglycemia Risk in Patients With Decompensated Cirrhosis and Type 2 Diabetes

This paper is my second non-public-database study, after Paper 7 — though don't ask whether it was done "entirely by me." The data was painstakingly collected by our collaborators across several hospitals in Guangdong, one patient at a time, by digging through both paper and electronic charts — proper grunt work. I was responsible for all statistical analyses and the nomogram construction.

Paper 9 - Early albumin administration and mortality in ICU patients with heart failure and hypoalbuminemia: a target trial emulation

A MIMIC database study, modeled after existing TTE (target trial emulation) papers. TTE has become more accepted in recent years, and I wanted to try it once. What I didn't expect was being rejected over and over — not because of the content, but because editors and reviewers tend to dismiss public-database studies on sight. Eventually it landed at BMC Cardiovascular Disorders — a journal and editorial flow I'm familiar with — and was finally accepted in August 2026.

It came in just as our hospital is going through major administrative changes. Whether the publication fee can be reimbursed is anyone's guess — given the gap between invoicing and the reimbursement window, the odds are not great. But it got published, and that is something. Having TTE run end-to-end, from data extraction to a finished manuscript, means the next person in our team can take this path more smoothly.

Code: https://github.com/shaou77/AlbinHF_tte