Jiazheng Chen
统计课教你算 Cox 模型，没人教你进了公司之后每天在干什么。这本书补的就是这一段。全书跟着两条完整的链走。一条是临床试验：从 protocol 和 SAP，到数据库锁定、SDTM、ADaM、TLF、QC，直到提交。另一条是真实世界数据：从可行性提数、病例定义、队列漏斗，到混杂、倾向评分、交付给研究者。三个贯穿案例--420 人的三期试验 LUMINA-3，6,000 人的 PRISM-HTN，以及在四百多万会员里找出 92 名患者的 BEACON-SM--从第 1 章一直走到第 16 章，同一批受试者、同一份数据、同一套编号。技术之外，这本书用同样的篇幅讲那些没人系统写过的事：怎么读懂 SAP 与 programming spec、怎么把结果讲给不懂数据的人听、怎么在受监管的环境里留痕、怎么准备简历与面试、入行头一年真实的样子。全书代码只用 SAS，共 118 段，已在 SAS OnDemand for Academics 上整册执行，输出与书中印出的每一张表逐格核对。配套 56 个 CSV 数据集与一份可运行版代码，读者可以自己跑出书里的每一个数字。书中另有一百多个「踩坑」块，记录的是这一行真实的失败方式--不是点缀，是这本书最想传达的部分。写给三种人：还在读书、想进这一行的学生；从数学、计算机、生物、医学转行的人；以及刚入行、只看得见自己那一小格的新人。所有数据与案例均为原创虚构。正文 500 页，附术语表、SAS 速查、法规查询指引、programming spec 与 shell 样例。How Biostatisticians Work: The Real Work, Communication, and Getting Hired (in Chinese)Statistics courses teach you to fit a Cox model. Nobody teaches you what the job is actually like once you are inside a company. This book fills that gap.It follows two complete chains end to end. One is a clinical trial: protocol and SAP, database lock, SDTM, ADaM, TLF, QC, submission. The other is real-world data: feasibility pull, case definition, cohort funnel, confounding, propensity scores, delivery to the investigator. Three running cases - the 420-subject phase III trial LUMINA-3, the 6,000-patient PRISM-HTN cohort, and BEACON-SM, which finds 92 patients among 4.18 million members - carry the same subjects, the same data and the same numbering from Chapter 1 to Chapter 16.Alongside the technical craft, the book gives equal room to the things rarely written down: reading a SAP and a programming spec, explaining results to people who do not work with data, leaving an audit trail in a regulated environment, building a portfolio, interviewing, and what the first year is really like.All code is SAS - 118 programs, executed end to end on SAS OnDemand for Academics, with the output checked cell by cell against every table printed in the book. Fifty-six CSV datasets and a paste-and-run edition of the code ship with it, so readers can reproduce every number themselves. More than a hundred PITFALL boxes record the ways this work actually fails.Written for students heading into the field, career changers from mathematics, computing, biology or medicine, and new hires who can see only their own small square of the process. All data and cases are original fiction.500 pages, with a glossary, a SAS quick reference, a guide to regulatory sources, and worked spec and shell examples.
