Patient-Reported Testing Burden of Breast Magnetic Resonance Imaging Among Women With Ductal Carcinoma In Situ

Key Points Question Is there a short-term reduction in health-related quality of life associated with breast magnetic resonance imaging (MRI) in patients with ductal carcinoma in situ? Findings In this cohort study of 244 women diagnosed with ductal carcinoma in situ, a short-term reduction in quality of life associated with MRI was revealed, primarily owing to fear before the test and fear and physical discomfort during the test. Meaning Understanding the potential reduction in quality of life associated with MRI in patients with ductal carcinoma in situ may allow development of targeted interventions to improve the patient’s experience.


Development of the TMI
Using the general techniques for development of a multi-attribute utility-based index, Swan et al developed the Testing Morbidities Index (TMI) by applying preference weights to a health classification for measuring short term health related quality of life related to diagnostic testing. 1 The conceptualization of the TMI is consistent with the wait-tradeoff (WTO) concept, since the burden of testing represents a toll on the process of medical care. WTO scaling for TMI allows the burden of the testing experience to be deductible from quality-adjusted life years and has been applied to a wide range of diagnostic tests [2][3][4] . The TMI surveys used for diagnostic mammogram and breast MRI are provided below.

Modified TMI questionnaire
The following questions are about how you felt before, during and after your mammogram used to diagnose DCIS. Think only about that mammogram. Component TMI scores were also computed for the before, during and after components, where the above equation

Calculation of the joint utility scores
Computation of the joint utility score of breast MRI after diagnostic mammography followed Thompson et al 5 (refer to Appendix 1 of that paper), and utilized the three most common methods of estimating joint utility, namely additive, multiplicative and minimum models. For breast MRI and mammography TMI summated scale scores, the general equation for the additive model is: The general equation for the multiplicative model is: The general equation for the minimum model is: UMinimum = minimum of (TMIMRI , TMIMammo)

Sensitivity analysis incorporating multiple imputation for missing covariate data
Using complete case analysis with listwise deletion resulted in 16% of records being dropped from the multivariable regression models (Table 3 of the main paper). To assess the impact of missing data on the reported multivariable models, multiple imputation by chained equations 6,7 , otherwise known as fully conditional specification, was used to impute missing covariate data for the N=244 subjects with TMI summated scale scores available for both modalities. A key assumption of multiple imputation is that the data are missing at random (MAR); or, that the probability of data being missing does not depend on the unobserved data, conditional on the observed data. Separate imputations were performed for the multivariable breast MRI TMI summated scale score regression model, and the multivariable joint utility score regression models.
For the breast MRI TMI summated scale score, the imputation model included the response variable (TMI summated scale score), as well as all covariates from the analysis model (age, race, ethnicity, insurance status, PROMIS-10 physical and mental T scores [T0], the revised 5-item ASC cancer worry subscale score [T0], and decision autonomy preference [T0]). In addition, given that the PROMIS-10 physical and mental T scores at T0 had the most missing data, the PROMIS-10 physical and mental T score collected at the first post-op visit following breast surgery (time point T2) were added to the imputation model as auxiliary variables (correlation=0.61 for physical T score between T0 and T2, and 0.67 for mental T score between T0 and T2); the mammography TMI summated scale score was also added as an auxiliary variable. In the imputations, a regression model was used for continuous variables, and a discriminant function model was used for categorical variables. Imputations were performed using SAS 9.4 PROC MI, with a total of 50 imputation samples. The same multivariable model was then fit to each imputation sample, with parameter estimates combined across samples using Rubin's rules 8 via SAS 9.4 PROC MIANALYZE.
For the joint utility scores, the imputation model included the response variables for the three models (joint (additive) utility score, joint (multiplicative) utility score, and joint (minimum) utility score), as well all covariates from the analysis models (age, race, ethnicity, insurance status, PROMIS-10 physical and mental T scores [T0], the revised 5-item ASC cancer worry subscale score [T0], and decision autonomy preference [T0]). In addition, given that the PROMIS-10 physical and mental T scores at T0 had the most missing data, the PROMIS-10 physical and mental T score collected at T2 were again added to the imputation model as auxiliary variables; the breast MRI TMI summated scale score was also added as an auxiliary variable. In the imputations, a regression model was used for continuous variables, and a discriminant function model was used for categorical variables. Imputations were performed using SAS 9.4 PROC MI, with a total of 50 imputation samples. The same multivariable models were then fit to each imputation sample, with parameter estimates combined across samples using Rubin's rules 8 via SAS 9.4 PROC MIANALYZE.
The amount of missing data by covariate is as follows:  Figure 1 of the main paper). 3 Continuous covariates were centered. Thus, the intercept can be interpreted as the mean joint utility score for patients who are at the mean age (59.1 years), mean ASC cancer worry level (2.41), mean physical and mental T scores (52.40 and 51.76, respectively), and who are at the reference level of the categorical covariates (White, non-Hispanic, private insurance, patient-driven decision preference).