R语言临床预测模型实战"
本书从R语言临床预测模型基本原理讲起，逐步深入到进阶实战，并配合实战案例，重点介绍临床预测模型的构建、评价、验证，让读者可以快速领悟 3~5分预测模型 SCI（science citation index，科学引文索引）的套路与精髓，为后续冲击10分以上SCI打基础。
本书分为13章，主要内容有线性回归、Logistic回归、Cox回归、竞争风险模型等；自变量筛选方法有传统方法、逐步回归法、Lasso法、随机森林法、**子集法、主成分分析法等；模型可视化涉及多种形式的列线图、Calibration校准曲线、ROC、DCA 曲线等图形绘制，不仅涉及单模型的可视化，还涉及单模型多时点、多模型同时点的可视化；模型评价指标涉及C指数、AUC、NRI、IDI 等；模型验证主要涉及简单交叉验证、K折交叉验证、留一法交叉验证及Bootstrap法。
本书内容通俗易懂，案例丰富，实用性强，适合R语言临床预测模型的入门读者和进阶读者，尤其是临床、护理、公共卫生、药学等专业的硕士研究生和博士研究生或从事相关研究的科研人员阅读。另外，本书还可以作为相关培训机构的教材。
R语言是常用的数据分析处理工具之一，尤其是在临床预测模型领域，目
前尚无其他语言可替代。虽然一些软件也可进行一定程度的临床预测模型分析，
但业内对这些软件的接受程度较低。虽然有一些小众软件可进行临床预测模型
分析，但是分析结果完全偏离真实结果，因此笔者不建议各位读者使用R语
言之外的软件进行临床预测模型分析。
近年来，随着R语言的逐渐完善，只需要少量代码即可完成各种常用的
临床预测模型的构建、评价/比较、验证及可视化。学习本书内容，可在较短
时间内完成3~5分的临床预测模型类的SCI，较为迫切地解决了广大读者的燃
眉之急。
笔者的使用体会
R语言是开源的，生态完整，有大量的现成数据分析包。R语言在临床预
测模型的数据分析领域中是一个理想的工具，有丰富的模型构建、评价/比较、
验证、可视化的代码，整合了多种形式的预测模型，可处理高维大数据，开发
方便、快捷，且可将模型转换成应用程序或网页，更加高端地展示学术成果。
Rstudio是R语言的一种免费的集成开发环境。相对于在文本文件中编
写R语言代码，Rstudio调试代码更加方便，对新手更加友好。Rstudio支持
Markdown，可以较为容易地编辑代码文档。利用Rstudio可以很好地记录数据
探索分析的过程，便于回顾、修改。
本书的特色
●从零开始：从模型构建开始，详细介绍模型构建的变量筛选方法，入
门门槛低。
●版本新颖：书中的大部分软件包采用官网的最新版本（截至本书撰写
时）进行讲解。
II
R语言临床预测模型实战
●经验总结：全面归纳和整理作者多年的R语言临床预测模型培训教学实践经验。
●内容实用：结合大量案例进行讲解，并对实现同一结果的多种代码进行对比。
本书的内容
本书可以分为三部分：第一部分是预测模型的构建，第二部分是预测模型的评价/比较，
第三部分是预测模型的验证。
第一部分主要介绍线性回归、Logistic回归、Cox回归、竞争风险模型的构建，以及常
用的变量筛选方法，如先单后多、逐步回归、Lasso、随机森林、最优子集、主成分分析等。
另外，本书对拟合模型中的常用统计指标均进行深度解读，如OR、HR、sHR等指标。
第二部分主要介绍如何进行模型的评价/比较，涉及多种形式的列线图、Calibration校
准曲线、ROC曲线、DCA曲线等的绘制，不仅涉及单模型的可视化，也包括单模型多时点、
多模型同时点的可视化；模型评价指标涉及C指数、AUC、NRI、IDI等。
第三部分主要介绍模型验证，涉及简单交叉验证、K折交叉验证、留一法交叉验证及
Bootstrap。
本书读者对象
●R语言零基础入门人员；
●医学类院校本科生、硕士研究生、博士研究生；
●高校及医疗机构科研人员；
●培训机构讲师；
●对临床预测模型感兴趣的人员。
作 者
2022年7月
第1章 临床预测模型概述 ············································1
1.1 如何构建预测模型 ·································································2
1.1.1 先单后多 ······················································································2
1.1.2 逐步回归 ······················································································2
1.1.3 Lasso ···························································································2
1.1.4 随机森林 ······················································································3
1.1.5 最优子集 ······················································································3
1.2 如何评价预测模型 ·································································3
1.2.1 拟合优度检验 ················································································3
1.2.2 ROC ····························································································4
1.2.3 Calibration ·····················································································4
1.2.4 DCA曲线 ·····················································································4
1.3 如何验证预测模型 ·································································4
1.3.1 交叉验证 ······················································································5
1.3.2 Bootstrap法 ···················································································5
1.4 小结 ···················································································5
第2章 线性回归 ························································6
2.1 线性回归概述 ·······································································7
2.1.1 相关的概念 ···················································································7
2.1.2 回归的概念 ···················································································9
2.2 线性回归实战 ······································································11
2.2.1 简单线性回归 ···············································································11
IV
2.2.2 多重线性回归 ···············································································17
2.3 小结 ··················································································25
第3章 Logistic回归 ··················································27
3.1 概述 ··················································································28
3.1.1 二分类Logistic回归的基本原理························································28
3.1.2 多分类Logistic回归的基本原理························································30
3.1.3 有序Logistic回归的基本原理···························································30
3.1.4 1∶m匹配条件Logistic回归的基本原理···········································31
3.2 Logistic回归分析实战 ···························································31
3.2.1 二分类Logistic回归 ·······································································31
3.2.2 多分类Logistic回归 ·······································································41
3.2.3 有序Logistic ················································································49
3.2.4 1∶m匹配条件Logistic ·································································53
3.3 小结 ··················································································56
第4章生存资料分析 ················································59
4.1 概述 ··················································································60
4.1.1 Kaplan-Meier生存曲线的基本原理····················································60
4.1.2 Cox比例风险模型的基本原理 ··························································61
4.2 生存资料分析实战 ································································62
4.2.1 Kaplan-Meier生存曲线 ···································································62
4.2.2 Cox比例风险模型 ·········································································72
4.3 小结 ··················································································81
第5章竞争风险模型 ················································83
5.1 概述 ··················································································84
5.2 竞争风险模型实战 ································································85
5.2.1 Fine-Gray检验（单因素分析） ·························································85
5.2.2 crr()函数法（多因素分析） ·····························································95
V
5.3 小结 ··················································································97
第6章 自变量筛选 ···················································99
6.1 传统方法 ··········································································100
6.1.1 基于连续性资料 ··········································································100
6.1.2 基于二分类资料 ··········································································105
6.1.3 基于生存资料 ·············································································112
6.2 逐步法 ·············································································116
6.2.1 基于连续性资料 ··········································································117
6.2.2 基于二分类资料 ··········································································121
6.2.3 基于多分类资料 ··········································································124
6.2.4 基于有序资料 ·············································································126
6.2.5 基于1∶m匹配资料 ···································································128
6.2.6 基于生存资料 ·············································································130
6.2.7 基于竞争风险资料 ·······································································131
6.3 Lasso法 ···········································································134
6.3.1 基于连续性资料 ··········································································135
6.3.2 基于二分类资料 ··········································································144
6.3.3 基于多分类资料 ··········································································151
6.3.4 基于生存资料 ·············································································160
6.4 随机森林法 ·······································································167
6.4.1 基于连续性资料 ··········································································168
6.4.2 基于二分类资料 ··········································································172
6.4.3 基于多分类资料 ··········································································177
6.4.4 基于生存资料 ·············································································180
6.4.5 基于竞争风险资料 ·······································································187
6.5 最优子集法 ·······································································191
6.5.1 基于连续性资料 ··········································································192
6.5.2 基于二分类资料 ··········································································193
6.5.3 基于多分类资料 ··········································································193
6.5.4 基于有序资料 ·············································································196
6.5.5 基于生存资料 ·············································································196
VI
6.5.6 基于竞争风险资料 ·······································································198
6.6 主成分分析法 ····································································200
6.7 小结 ················································································204
第7章列线图·························································205
7.1 列线图简介 ·······································································206
7.2 基于连续资料 ····································································206
7.2.1 普通静态列线图 ··········································································206
7.2.2 网页动态列线图 ··········································································210
7.2.3 花样列线图 ················································································212
7.2.4 静态列线图的变种 ·······································································215
7.3 基于二分类资料 ·································································216
7.3.1 普通静态列线图 ··········································································216
7.3.2 网页动态列线图 ··········································································222
7.3.3 花样列线图 ················································································223
7.3.4 静态列线图的变种 ·······································································227
7.4 基于有序资料 ····································································228
7.4.1 普通静态列线图 ··········································································228
7.4.2 花样列线图 ················································································233
7.4.3 静态列线图的变种 ·······································································236
7.5 基于生存资料 ····································································237
7.5.1 普通静态列线图 ··········································································237
7.5.2 网页动态列线图 ··········································································246
7.5.3 花样列线图 ················································································248
7.5.4 静态列线图的变种 ·······································································253
7.5.5 生存概率列线图 ··········································································255
7.6 基于竞争风险模型 ······························································256
7.6.1 普通静态列线图 ··········································································256
7.6.2 静态列线图的变种 ·······································································264
7.6.3 生存概率列线图 ··········································································265
7.6.4 花样列线图 ················································································266
7.7 小结 ················································································270
第8章 Calibration校准曲线 ·································271
8.1 Calibration校准曲线简介 ·····················································272
8.2 基于二分类资料 ·································································272
8.2.1 单一模型校准曲线 ·······································································272
8.2.2 多模型校准曲线 ··········································································281
8.3 基于生存资料 ····································································284
8.3.1 单模型单时点校准曲线 ·································································285
8.3.2 单模型多时点校准曲线 ·································································292
8.3.3 多模型同时点校准曲线 ·································································294
8.4 基于竞争风险模型 ······························································299
8.4.1 单模型单时点校准曲线 ·································································299
8.4.2 多模型同时点校准曲线 ·································································301
8.5 小结 ················································································304
第9章 C指数计算 ···················································305
9.1 C指数简介 ·······································································306
9.2 基于二分类资料 ·································································306
9.3 基于生存资料 ····································································310
9.4 基于竞争风险模型 ······························································315
9.5 小结 ················································································318
第10章 ROC曲线 ···················································319
10.1 ROC曲线简介 ·································································320
10.2 基于二分类资料 ·······························································320
10.2.1 单一模型ROC曲线 ····································································320
10.2.2 多模型ROC曲线 ·······································································328
10.3 基于生存资料 ··································································332
10.3.1 单模型单时点ROC曲线 ······························································332
10.3.2 单模型多时点ROC曲线 ······························································336
10.3.3 多模型同时点ROC曲线 ······························································339
VIII
10.4 基于竞争风险模型····························································343
10.4.1 单模型单时点ROC曲线 ······························································343
10.4.2 单模型多时点ROC曲线 ······························································345
10.4.3 多模型同时点ROC曲线 ······························································348
10.5 小结 ··············································································349
第11章 DCA曲线 ···················································350
11.1 DCA曲线简介 ·································································351
11.2 基于二分类资料·······························································351
11.2.1 单一模型DCA曲线····································································351
11.2.2 多模型DCA曲线·······································································356
11.3 基于生存资料··································································359
11.3.1 单模型单时点DCA曲线 ······························································359
11.3.2 单模型多时点DCA曲线 ······························································363
11.3.3 多模型同时点DCA曲线 ······························································365
11.4 小结 ··············································································368
第12章 NRI、IDI计算 ·············································369
12.1 NRI、IDI简介 ·································································370
12.2 基于二分类资料 ·······························································370
12.3 基于生存资料 ··································································378
12.4 小结 ··············································································384
第13章交叉验证及Bootstrap ·····································385
13.1 概述 ··············································································386
13.2 简单交叉验证 ··································································389
13.3 K折交叉验证 ··································································391
13.4 留一法交叉验证 ·······························································393
13.5 Bootstrap ·········································································395
13.6 小结 ··············································································397
本书从R语言临床预测模型基本原理讲起，逐步深入到进阶实战，并配合实战案例，重点介绍临床预测模型的构建、评价、验证，让读者可以快速领悟 3~5分预测模型 SCI（science citation index，科学引文索引）的套路与精髓，为后续冲击10分以上SCI打基础。
本书分为13章，主要内容有线性回归、Logistic回归、Cox回归、竞争风险模型等；自变量筛选方法有传统方法、逐步回归法、Lasso法、随机森林法、**子集法、主成分分析法等；模型可视化涉及多种形式的列线图、Calibration校准曲线、ROC、DCA 曲线等图形绘制，不仅涉及单模型的可视化，还涉及单模型多时点、多模型同时点的可视化；模型评价指标涉及C指数、AUC、NRI、IDI 等；模型验证主要涉及简单交叉验证、K折交叉验证、留一法交叉验证及Bootstrap法。
本书内容通俗易懂，案例丰富，实用性强，适合R语言临床预测模型的入门读者和进阶读者，尤其是临床、护理、公共卫生、药学等专业的硕士研究生和博士研究生或从事相关研究的科研人员阅读。另外，本书还可以作为相关培训机构的教材。
彭献镇，南京医科大学康达学院流行病与统计学系主任，研究方向为慢性病流行病学， 担任江苏省预防医学会青年委员，在国内外发表20余篇学术论文，主持各级科研课题7项； 擅长SPSS、Stata、SAS、GraphPad Prism、PASS、R、Amos 等多种软件的统计分析建模工作， 实战经验丰富，创建大鹏统计工作室至今，已为5000余名硕士研究生、博士研究生提供统计分析、科研设计服务。
