将SNOT-22映射到EQ-5D-3L指数：在哮喘合并鼻息肉人群中复现既往开发的模型结构并对系数进行部分验证

文献信息

中文摘要

引言： 22项鼻窦结局测试（SNOT-22）广泛用于评估呼吸系统疾病的症状负担；然而，它没有相关的效用值。Crump等人开发了一种映射算法，用于在慢性鼻窦炎（CRS）中将SNOT-22应答转换为EQ-5D-3L效用值；我们在哮喘合并伴鼻息肉的慢性鼻窦炎（CRSwNP）人群中复现了该分析。
方法： 本研究使用了NAVIGATOR试验期间收集的SNOT-22和EQ-5D-5L数据。使用van Hout交叉walk算法生成相应的EQ-5D-3L效用值。按照Crump等人的方法拟合了四个普通最小二乘（OLS）回归模型，以估计效用值作为SNOT-22评分的函数。计算了自举法95%置信区间。
结果： OLS模型的调整R2范围为0.22至0.35，RMSE为0.16至0.18。在Crump等人中，以单个SNOT-22条目为变量的OLS回归模型是表现最佳的模型；在我们的人群中，具有相同变量的模型达到了相似的解释力（调整R2=0.33，我们的模型 vs. Crump等人的0.34）和可比的预测力（RMSE=0.16 vs. 0.23）。只有两个SNOT-22域与效用评分显著相关（p<0.002）。将“年龄”和“性别”作为变量加入后，调整R2相似（0.34），RMSE可比（0.16），AIC改善（-366.2 vs. -357.8）。进一步加入“访视次数”作为变量改善了所有拟合参数；然而，该模型被认为普遍性有限，因此更倾向于较简单的模型。应用Box-Cox变换或拟合更复杂的模型均未显著改善拟合。所有系数均在自举法95%置信区间内。
结论： Crump等人开发的效用映射算法和模型结构在哮喘合并CRSwNP人群中得到复现，达到了相似的拟合质量。该算法可用于在类似疾病领域从SNOT-22评分生成效用值。

英文摘要

INTRODUCTION: The 22-item sino-nasal outcomes test (SNOT-22) is widely used to evaluate symptom burden in respiratory disease; however, it does not have associated utilities. Crump et al. developed a mapping algorithm for converting SNOT-22 responses to EQ-5D-3 L utilities in chronic rhinosinusitis (CRS); we replicated this analysis in a population with asthma and CRS with nasal polyps (CRSwNP).
METHODS: This study used SNOT-22 and EQ-5D-5 L data collected during the NAVIGATOR trial. The van Hout crosswalk algorithm was used to generate corresponding EQ-5D-3 L utilities. Four ordinary least squares (OLS) regression models were fitted, following Crump et al., to estimate utilities as a function of SNOT-22 scores. Bootstrapped 95% CIs were computed.
RESULTS: Adjusted R2 for the OLS models ranged from 0.22 to 0.35, and RMSE from 0.16 to 0.18. The OLS regression model with individual SNOT-22 items as variables was the best-performing model in Crump et al.; in our population, a model with the same variables achieved similar explanatory power (adjusted R2 = 0.33 vs. 0.34 for our model vs. Crump et al.) and comparable predictive power (RMSE = 0.16 vs. 0.23). Only two SNOT-22 domains were significantly associated with utility score (p < 0.002). Adding 'age' and 'sex' as variables resulted in similar adjusted R2 (0.34), comparable RMSE (0.16), and improved AIC (-366.2 vs. -357.8). Adding 'visit number' as a further variable improved all fit parameters; however, this model was judged to have limited generalisability, and simpler models were preferred. Neither applying Box-Cox transformations nor fitting more complex models substantially improved fit. All coefficients were within bootstrapped 95% CIs.
CONCLUSION: The utility mapping algorithm and model structure developed by Crump et al. was replicated in a population with asthma and CRSwNP, achieving similar quality of fit. This algorithm could be used to generate utilities from SNOT-22 scores across similar disease areas.