Logistic回归及ROC曲线综合评价动脉粥样硬化危险因素对雄激素性秃发的影响

Logistic Regression and ROC Curve for Comprehensive Evaluation of the Risk Factors of Atherosclerosis on Androgenetic Alopecia

目的采用Logistic回归及ROC曲线综合评价动脉粥样硬化危险因素对雄激素型秃发的影响。方法选取2018年1月—2018年12月就诊于我院皮肤科雄激素型秃发患者80例作为试验组,同期80例健康受试者作为对照组,结合以往研究的影响因素并统计临床一般统计学资料及动脉粥样硬化危险因素。采用Logistic回归分析筛选雄激素型秃发的指标。然后得出通过相关动脉粥样硬化危险因素预测该病患者的预测模型,最后采用ROC曲线评价不同影响因素对该病患者相关性的预测能力并找到最佳分界值及效能。结果两组性别、高血脂、LDLC、Lpa及DHT差异有统计学意义(P <0. 05),SRD5A2的基因型与AR上GGN序列重复数目对雄激素性秃发可能有影响(P <0. 05),LDL-C、Lpa、DHT、SRD5A2的基因型与AR上GGN序列重复数目相关性分别为0. 048、0. 057、238. 719、5. 395及1. 025,均具有统计学差异(P <0. 05),最终雄激素型秃发的概率预测模型为P=1/[1+e^((34. 091+3. 027SRD5A2+2. 857AR上GGN序列重复数目-5. 475LDLc-1. 685Lpa-0. 024DHT))],新变量P的AUC为0. 992,此时诊断效能最好。敏感性为95. 00%,特异性为96. 20%。结论采用动脉粥样硬化危险因素联合AGA相关因素的回归模型能够提高对雄激素型秃发患者的预测效能,可提高其诊断敏感性和特异性,具有较高的临床应用价值,值得进一步推广使用。

Objective To evaluate the effect of atherosclerosis risk factors on androgen-induced alopecia by Logistic regression and ROC curve. Methods 80 patients with androgenic alopecia who were treated in our hospital from January 2018 to December 2018 were selected as the experimental group and 80 healthy subjects as the control group during the same period. Combined with the influencing factors of previous studies, clinical general statistical data and risk factors of atherosclerosis were analyzed. Logistic regression analysis was used to screen the indicators of androgenic alopecia. Finally, ROC curve was used to evaluate the predictive ability of different influencing factors on the correlation of male androgenic alopecia patients and to find the best dividing value and effect. Results There were significant differences in gender, hyperlipidemia, LDL-C, Lpa and DHT between the two groups ( P <0.05). The genotype of SRD5A2 and the number of GGN sequences on AR may have an effect on androgenetic alopecia ( P <0.05). The correlations between the genotypes of LDL-C, Lpa, DHT and SRD5A2 and the number of GGN sequences on AR were 0.048, 0.057, 238.719, 5.395 and 1.025, respectively, which were statistically significant ( P <0.05), and finally androgenetic baldness. The probabilistic prediction model of the hair is P =1/[1+e^(the number of GGN sequence repeats on the 34.091+3.027SRD5A2+2.857AR-5.475LDLc- 1.685Lpa -0.024DHT)], and the AUC of the new variable P is 0.992. The sensitivity was 95.00% and the specificity was 96.20%. Conclusion The regression model of risk factors of atherosclerosis combined with AGA related factors can improve the predictive efficiency of androgen alopecia patients, and improve the sensitivity and specificity of diagnosis. It has high clinical value and is worthy of further promotion and application.