Preferences of patients for benefits and risks of insomnia medications using data elicited during two phase III clinical trials

Abstract Study Objectives To elicit the trade-offs patients are willing to make between benefits and risks of medications for chronic insomnia, with the purpose of allowing a patient-centric interpretation of clinical trial data. Methods A discrete choice experiment (DCE) was included in the two placebo-controlled phase III trials that evaluated the efficacy and safety of daridorexant. The DCE design was informed by a two-phase qualitative study, followed by qualitative and quantitative pilot testing before fielding. Relative attribute importance (RAI) and acceptable trade-offs between benefits and risks were obtained using a mixed logit model. Results Preferences were elicited from 602 trial participants (68.1% female, aged 58.6 ± 14.5 years). Preferences were most affected by daytime functioning (RAI = 33.7%) as a treatment benefit and withdrawal symptoms (RAI = 27.5%) as a risk. Patients also valued shorter sleep onset (RAI = 6.4%), longer sleep maintenance (RAI = 5.4%), reduced likelihood of abnormal thoughts and behavioral changes (RAI = 11.3%), reduced likelihood of dizziness/grogginess (RAI = 9.2%), and reduced likelihood of falls at night (RAI = 6.5%). Patients were willing to make trade-offs between these attributes. For example, they would accept an additional 18.8% risk of abnormal thoughts and behavioral changes to improve their daytime functioning from difficult to restricted and an additional 8.1% risk of abnormal thoughts and behavioral changes to avoid moderate withdrawal effects. Conclusions Patients with insomnia were willing to make trade-offs between multiple benefits and risks of pharmacological treatments. Because patients valued daytime functioning more than sleep latency and duration, we recommend that functional outcomes and sleep quality be considered in treatment development and evaluation.


Contents
 For female subjects: pregnant, lactating or planning to become pregnant during projected duration of the study;  History or clinical evidence of any disease or medical condition or treatment, which may put the subject at risk of participation in the study or may interfere with the study assessments.
 Any circumstances or conditions, which, in the opinion of the investigator, may affect the subject's full participation in the study or compliance with the protocol.

Preference elicitation
5][36][37] Utility was defined as: where the systematic utility component ( ) is a function of the DCE attributes, and is an extreme value distributed random error, also referred to as psychometric noise.The error term, captures random choice influences above and beyond the included attributes and allows the estimation of the utility function as a logit model.The baseline utility, ( ) was defined as: (2) where is constant for the left alternative in the DCE, controlling for any left-right bias.The remaining parameters are marginal utilities that are assumed to be randomly distributed in the population with a mean and a standard deviation (SD) to be estimated: The utility function (equation 2) was estimated as a mixed logit (MXL).The MXL is the most general choice model, and accounts for various types of heterogeneity and correlation in the data. 38The baseline model included patients from both phase III trials with data collected at visit 4.

Relative Attribute Importance (RAI)
Given the ordinal nature of the marginal utilities (i.e., parameters have an arbitrary scale), RAI scores were used to normalize estimates and facilitate the interpretation of the findings.RAI of an attribute (=) was defined as: where max{D ̅ A D} is the largest marginal utility of any level of attribute = and, given the dummy coding, is equivalent to the maximum affect that the attribute can have on a treatments' utility.Thus, the RAI can be interpreted as a percentage, and measures the proportion of changes in treatment utility that can be assigned to changes in a particular attribute.

Maximum acceptable Risk (MAR)
To obtain insights into attribute trade-offs, MAR estimates were obtained.MAR measures the value of each attribute level, relative to its reference, in its equivalent level of risk of abnormal thoughts and behavior changes.Thus, MAR is a risk trade-off and expresses how much additional risk of abnormal thoughts and behavior changes patients were willing to accept for changes in other attributes.