传统统计法与稳健统计法评价医学检验实验室能力验证结果比较

doi:10.3969/j.issn.1673-8640.2017.05.020

摘要/Abstract

摘要：

目的探讨传统统计法（Dixon检验法剔除离群值后计算x和s）与稳健统计法[计算中位值和标准化四分位间距（NIQR）]医学检验实验室能力验证结果评价的适用范围。方法对2016年上海市临床检验中心快速C反应蛋白检测能力验证计划上报结果按检测系统分为A组（29家实验室）、B组（15家实验室）和C组（9家实验室）。应用传统统计法和稳健统计法分别对各组的统计量进行比较分析。结果对于A组结果,稳健统计法和传统统计法的指定值基本相同,但前者计算出的NIQR和指定值的标准不确定度（u_x）都小于后者计算出的s和指定值的u_x;对于B组结果,Dixon检验法剔除离群值后用传统统计法计算的指定值与稳健统计法计算所得较为接近,但前者所得到的u_x<0.3σ,而后者所得到的u_x>0.3σ;对于C组结果,Dixon检验法剔除离群值后用传统统计法计算的指定值与稳健统计法计算所得较为接近,2种方法计算所得的u_x>0.3σ,但前者计算所得的s均小于后者计算所得的NIQR。结论当实验室的数量较多（>18）时,可采用稳健统计法给出严格的评价结论;当利用Dixon检验法剔除离群值后,参加实验室的数量较少（12~18）的情况下,应对余下的数据采用传统统计法进行统计和分析;当剔除离群值后实验室的数量更少（≤12）时,除了对余下的数据采用传统统计法进行统计和分析,还需考虑指定值的u_x的影响。

关键词: 统计方法, 医学检验实验室, 能力验证, 稳健统计法, 传统统计法, Dixon检验法, 不确定度

Abstract:

Objective To investigate the scopes of conventional statistics（x and s after removal of outliers by Dixon method） and robust statistics [median and normalized interquartile range（NIQR）] for proficiency testing in laboratory medicine. Methodse Participants of the Proficiency Testing Program 2016 on C-reactive protein organized by Shanghai Center for Clinical Laboratory were classified into group A（29 laboratories）,group B（15 laboratories）and group C（9 laboratories）according to determination systems. The results of conventional statistics and robust statistics were analyzed and compared. Results As for group A,the designated values were same for the 2 methods. However,the NIQR and standard uncertainty （u_x）of designated values by conventional statistics were smaller than s and u_x by robust statistics. As for group B,the designated values were close between the 2 methods. Conventional statistics had u_x<0.3σ,and robust method had u_x>0.3σ. As for group C,the designated values were close between the 2 methods. Both methods had u_x>0.3σ. The s of conventional statistics was smaller than the NIQR of robust statistics. Conclusions When the number of participants is >18,robust statistics can be used. When the number of participants is 12-18,conventional statistics may be used. When the number is ≤12,besides conventional statistics should be used,the u_x of designated values should be considered as well.

Key words: Statistical methods, Clinical laboratories, Proficiency testing, Robust statistics, Conventional statistics, Dixon method, Uncertainty

中图分类号:

R446.1

郭晓俊, 朱岭峰, 徐翀, 居漪. 传统统计法与稳健统计法评价医学检验实验室能力验证结果比较[J]. 检验医学, 2017, 32(5): 435-439.

GUO Xiaojun, ZHU Lingfeng, XU Chong, JU Yi. Conventional statistics and robust statistics for proficiency testing in laboratory medicine[J]. Laboratory Medicine, 2017, 32(5): 435-439.

图/表 6

参考文献 9

[1]	International Organization for Standardization. Accredition criteria for proficiency testing provider[S]. ISO 17043,ISO,2010.
[2]	周琦,李少男,李小鹏,等. 临床检验室间质量评价数据中离群值的研究[J]. 中华检验医学杂志,2007,30(2):214-216.
[3]	中国合格评定国家认可委员会. CNAS-GL02:2014能力验证结果的统计处理和能力评价指南[S].2014
[4]	张德然. 统计数据中异常值的检验方法[J]. 统计研究,2003,20(5):53-55.
[5]	International Organization for Standardization. Statistical methods for use in proficiency testing by interlaboratory comparisons[S].ISO 13528,ISO,2015.
[6]	王承忠. 实验室间比对的能力验证及稳健统计技术第五讲稳健统计技术(一)[J].理化检验(物理分册),2004,42(11):589-593.
[7]	尹江伟,吴礼康,肖惠贞. 能力验证的结果评价和存在问题探讨[J]. 现代测量与实验室管理,2012,20(5):39-40.
[8]	肖亚玲,王薇,王治国. ISO15189:2012与室间质量评价[J]. 现代检验医学杂志,2014,29(5):161-163.
[9]	刘洪,黄燕. 我国统计数据质量的评估方法研究——趋势模拟评估及其应用[J]. 统计研究,2007,24(8):17-21.

编号	样本1	样本2	样本3	样本4	样本5
1	27.19	27.33	14.98	48.35	34.49
2	28.20	27.50	15.20	47.70	34.60
3	28.01	27.38	15.69	47.63	34.99
4	28.20	28.60	15.40	48.60	35.60
5	27.21	29.36	15.12	47.52	35.22
6	27.50	26.32	12.63*	42.19^*	34.96
7	27.48	29.11	15.92	46.87	34.53
8	28.53	27.98	15.80	48.81	35.33
9	25.30	28.30	15.30	49.00^*	32.70
10	28.00	27.33	15.25	47.39	34.39
11	27.35	26.01	16.76	52.54^*	35.54
12	26.87	26.26	15.22	48.83	36.51
13	26.19	28.86	16.85	50.90	35.04
14	29.45	27.50	15.91	52.62^*	33.39
15	28.09	28.38	15.29	48.17	35.28
16	28.50	28.64	16.02	49.04	36.20
17	29.02	28.61	15.81	48.9	35.36
18	27.10	28.02	14.98	47.11	34.25
19	26.50	26.40	15.10	45.80	34.80
20	29.29	29.74	16.03	48.67	35.48
21	29.10	27.84	15.60	48.30	35.51
22	30.76	27.53	17.54^*	53.91^*	35.85
23	27.59	27.24	15.35	47.27	35.69
24	27.75	28.24	16.23	50.68	35.76
25	30.64	26.99	17.26	48.98	36.92
26	27.49	24.29^*	15.00	44.72^*	32.67^*
27	29.36	28.27	15.60	48.49	35.51
28	24.14	21.87^**	13.41^*	40.21^**	30.60^*
29	31.05	29.88	17.18	51.26	36.68

编号	样本1	样本2	样本3	样本4	样本5
1	27.19	27.33	14.98	48.35	34.49
2	28.20	27.50	15.20	47.70	34.60
3	28.01	27.38	15.69	47.63	34.99
4	28.20	28.60	15.40	48.60	35.60
5	27.21	29.36	15.12	47.52	35.22
6	27.50	26.32	12.63*	42.19^*	34.96
7	27.48	29.11	15.92	46.87	34.53
8	28.53	27.98	15.80	48.81	35.33
9	25.30	28.30	15.30	49.00^*	32.70
10	28.00	27.33	15.25	47.39	34.39
11	27.35	26.01	16.76	52.54^*	35.54
12	26.87	26.26	15.22	48.83	36.51
13	26.19	28.86	16.85	50.90	35.04
14	29.45	27.50	15.91	52.62^*	33.39
15	28.09	28.38	15.29	48.17	35.28
16	28.50	28.64	16.02	49.04	36.20
17	29.02	28.61	15.81	48.9	35.36
18	27.10	28.02	14.98	47.11	34.25
19	26.50	26.40	15.10	45.80	34.80
20	29.29	29.74	16.03	48.67	35.48
21	29.10	27.84	15.60	48.30	35.51
22	30.76	27.53	17.54^*	53.91^*	35.85
23	27.59	27.24	15.35	47.27	35.69
24	27.75	28.24	16.23	50.68	35.76
25	30.64	26.99	17.26	48.98	36.92
26	27.49	24.29^*	15.00	44.72^*	32.67^*
27	29.36	28.27	15.60	48.49	35.51
28	24.14	21.87^**	13.41^*	40.21^**	30.60^*
29	31.05	29.88	17.18	51.26	36.68

方法	统计量	样本1	样本2	样本3	样本4	样本5
传统统计法		28.00	27.58	15.60	48.29	34.96
	s	1.52	1.63	1.02	2.81	1.31
	CV（%）	5.44	5.90	6.51	5.81	3.75
	u_x	0.28	0.30	0.19	0.52	0.24
	0.3σ	0.46	0.49	0.31	0.84	0.39
稳健统计法	med	28.00	27.84	15.60	48.49	35.28
	NIQR	1.38	1.10	0.64	1.25	0.84
	RCV（%）	4.93	3.95	4.10	2.58	2.38
	u_x	1.38	1.10	0.64	1.25	0.84
	0.3σ	0.41	0.33	0.19	0.38	0.25

方法	统计量	样本1	样本2	样本3	样本4	样本5
传统统计法		28.00	27.58	15.60	48.29	34.96
	s	1.52	1.63	1.02	2.81	1.31
	CV（%）	5.44	5.90	6.51	5.81	3.75
	u_x	0.28	0.30	0.19	0.52	0.24
	0.3σ	0.46	0.49	0.31	0.84	0.39
稳健统计法	med	28.00	27.84	15.60	48.49	35.28
	NIQR	1.38	1.10	0.64	1.25	0.84
	RCV（%）	4.93	3.95	4.10	2.58	2.38
	u_x	1.38	1.10	0.64	1.25	0.84
	0.3σ	0.41	0.33	0.19	0.38	0.25

编号	样本1	样本2	样本3	样本4	样本5
1	30.70	28.80	17.40	56.60	37.40
2	33.80	31.30	21.60^*△	53.30	36.30
3	31.80	27.90	18.70	49.10	35.60
4	31.00	28.90	18.70	48.80	34.70
5	30.50	29.10	18.10	48.50	35.90
6	32.10	29.80	19.40	51.10	36.10
7	31.60	28.40	19.00	50.20	35.80
8	30.30	27.80	18.40	45.80	33.60
9	31.20	27.90	18.10	49.50	35.20
10	31.40	28.00	18.20	49.80	35.30
11	31.80	28.20	18.60	50.40	35.60
12	31.60	28.50	18.50	50.10	35.30
13	33.50	30.10	19.30	53.10	37.00
14	30.13	28.05	17.81	48.17	35.10
15	42.69^*△	45.32^*△	32.1^*△	63.47^*△	69.41^*△