Outcomes

The primary outcome is measured through the identification of the responder patients, defined as those showing an improvement of at least 20% in the ALSFRS-R slope during the 18-month randomization period compared to the 3-month lead-in period. Slope coefficients for ALSFRS-R decline will be calculated separately for the lead-in period and the treatment period using the linear regression model.

The secondary efficacy outcomes are presented in Table 2. The safety and tolerability of TUDCA will be evaluated through the assessment of adverse events, concomitant treatments, physical examinations, vital signs, and routine laboratory tests (hematology and biochemistry).

Table 2 Secondary and exploratory outcomesAnalysis method for primary and secondary outcomes

In the primary analysis, the effect of treatment on the primary endpoint will be measured by means of the odds ratio estimated through an unadjusted logistic regression model including a dummy variable for treatment group. Moreover, as part of the sensitivity analyses, a multivariable logistic regression model will be performed. Survival outcome will be analyzed using Kaplan–Meier survival analysis and log-rank test. Univariate and multivariable Cox proportional hazards models will also be applied. Additionally, the joint modeling omnibus test [13] will be used to assess the simultaneous effect on ALSFRS-R and survival. The longitudinal secondary endpoints (ALSFRS-R, ALSAQ-40, FVC, EQ-5D scale, MRC scale, and biomarkers) measured on a continuous scale will be analyzed using a multivariable linear mixed effect model (LME) to estimate the mean difference in the rate of decline between TUDCA and placebo over 18 months. The following covariates will be included in the multivariable models: age, gender, type of ALS, and region of initial site of diagnosis.

Country will be included as a random factor in the LME models. Clustered standard errors by country will be used in logistic and Cox regression models to obtain robust estimates. All primary and secondary analyses will be carried out on both ITT and per-protocol populations.

Subgroup analyses

Exploratory subgroup analyses will be performed for the primary outcome in the ITT population, according to the following subgroup variables measured at randomization (month 0): site of onset, type of ALS (familial, sporadic), age (categorized according to the median), duration of the disease since symptom onset (categorized according to the median), gender, country. Moreover, an exploratory analysis will be conducted stratifying patients according to neurofilaments level [14]. The subgroup effect will be estimated including an interaction term for each subgroup variable at time in the logistic regression. The forest plot will be used to summarize subgroup analyses. The subgroup analyses will also be carried out for the change from baseline on ALSFRS-R by adding the treatment-subgroup interaction term in the MLE model.

Missing dataImputation rules and key sensitivity analyses

In the primary analysis, for each patient, all available ALSFRS-R scores will be included in the linear regression model to estimate the monthly decline. The following rules for handling missing data will be applied: (i) for deceased patients, as death is clearly a poor outcome and there is no consensus on an alternative score, the lowest score (zero) will be assigned at the time of death, and (ii) for withdrawals or LTF patients without a post-randomization assessment of the ALSFRS-R score, the monthly decline of ALSFRS-R will be imputed using the “nearest neighbor” procedure [15]. For each of these subjects, we will identify the five other participants within the same treatment group whose baseline ALSFRS-R scores were closest to the baseline score of the patient. Then, the largest monthly ALSFRS-R decline observed among the identified neighbors (worst-case imputation) will be used for imputation. All missing data will be imputed within treatment groups defined by the randomized treatment assignment. As sensitivity analysis, within the nearest neighbor procedure, the “bestcase imputation” will also be carried out imputing the smallest observed monthly decline.

Multiple imputation-based sensitivity analyses

We intend to calculate the post-randomization slope only after imputing missing ALSFRS-R data in the 18 months of treatment period and then categorize participants into responders/non-responders and finally conduct logistic regression analysis to estimate the effect of TUDCA. This approach is supported by simulation studies conducted by Floden and Bell [16]. A first sensitivity analysis will be carried out using the multiple imputation (MI) technique under the missing at random (MAR) assumption to deal with missing ALSFRS-R data. The imputation model will include the treatment group, the prognostic baseline characteristics, and ALSFRS-R at baseline. For patients deceased or withdrawn without a post-randomization assessment, the same rules specified for primary analysis will be applied. Moreover, the imputation of missing data under the missing not at random (MNAR) assumption considering the reason for missingness will be carried out using the delta-based MI method. A range of penalties by reason of missingness (study termination due to adverse events AE and death of patients) will be considered. All other missing data will be imputed under MAR assumption, applying a null penalty. For withdrawals without a post-randomization assessment, the same rules specified for primary analysis will be used. A total of 50 imputations for each analysis will be performed, which should provide adequate precision since the amount of missing data is not expected to be large [17]. Results from different imputed datasets will be combined using Rubin’s rule.

Supporting analysis

Based on suggestions received from the EMA Scientific Advice Working Party on 28 February 2017, we will also consider a definition of responders with a threshold of 25%. The same model specified for the primary analysis will be applied.

Additional analyses to account for the COVID-19 pandemic

Prior to the COVID-19 pandemic, the primary analysis was planned using an ITT approach, by properly accounting for missing data and premature study termination. Despite the occurrence of the COVID-19 pandemic, the original primary objective of TUDCA-ALS study remains unchanged, implying that the estimate of TUDCA effect is not confounded by COVID-19 pandemic related disruptions. However, COVID-19-related intercurrent events need to be properly accounted for to ensure an appropriate interpretation of the trial results [18]. Following the recommendation received on December 1, 2022, after the mid-term review of the project by the European Commission, additional sensitivity and supplementary analyses have been planned in accordance with the ICH E9 (R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials [19].

In application of the COVID-19 amendment, changes have been made to adapt the electronic case report forms for the study visits performed after February 3, 2020. The main changes include the implementation of telemedicine visits, the specification of missing data due to COVID-19, collecting information on COVID-19 protocol deviations, COVID-19 vaccination, and study termination due to COVID-19. Information on COVID-19-related infections, therapy, and death are expected to be reported in the descriptions of protocol deviations, as well as in the form on concomitant medication and adverse events. All these data will be used to perform additional analyses in accordance with the specific EMA guidelines on the implications of coronavirus disease (COVID-19) on methodological aspects of ongoing clinical trials [20].

Assessing the impact of COVID-19 pandemic on trial conduction

To describe the impact of COVID-19 pandemic on trial conduction, with the aim of excluding treatment-specific data patterns by reason due to COVID-19, the following analysis will be performed by treatment group. Baseline characteristics will be summarized by the enrolment period categorized in pre-pandemic, pandemic with restrictive measures, and post-pandemic. The number of visits conducted via telemedicine (including audio-visual connections and telephone contacts), missing data, protocol deviations, end of study by reason, COVID-19 infections, concomitant medications for COVID-19, and death due to COVID-19, will be reported overall and by treatment group. A time to occurrence analysis will be conducted for COVID-19-related protocol deviations using Kaplan-Maier curves and the log-rank test.

Assessing the impact of the COVID-19 pandemic on treatment effect

Even if self-administered ALSFRS-R collected through telemedicine is a validated approach [21,22,23], an assessment of the association between on-site and telemedicine visits will be evaluated by comparing the distributions of ALSFRS-R measurements by the modality of visit conduction (in-person or telemedicine) stratifying for treatment group and study visit. The modality of visit conduction will be included in the multivariable LME model for ALSFRS-R as a fixed effect. To assess the impact of the pandemic on the ALSFRS-R slope, the pandemic time-period and the infection status will also be included in the multivariable LME model as fixed effects. The interaction term between treatment and pandemic time-periods will be included, and the treatment effect by pandemic periods will be reported. Additionally, an estimate of treatment effect will be obtained through inverse probability treatment weighting to assess the potential confounding effect of COVID-19 intercurrent events [24].

Sensitivity analyses for handling missing data in the primary endpoint

Missing data are expected to increase during the trial due to the COVID-19 pandemic. In the primary and key sensitivity analyses, pandemic-related missing data will be treated as all the other missing data.

In these additional analyses, according to the estimand framework [19], the treatment policy strategy is adopted for COVID-19 intercurrent events. In such approach, we consider the pandemic irrelevant in the effect estimate, and all collected data are used in the analyses. All missing data will be imputed by MI under MAR assumption, by including also COVID-19 infection status and pandemic time-period in the imputation model [25]. Moreover, under the assumption that the pandemic (restrictive measures, lockdown and infections) could negatively affect the outcome, the analysis under MNAR assumption will be repeated, conducting a tipping point analysis for COVID-19-related missing data. MI analysis will be conducted using the number of imputations and the pooling rule previously specified.

Sensitivity analysis for the secondary survival endpoint

As the pandemic may affect the survival probability, the proportional hazards assumption will be checked by visual assessment of Kaplan–Meier curves, log(− log) plots, and testing of scaled Schoenfeld residuals. If the assumption of proportional hazards between treatment groups is not verified, the ratio of restricted mean survival time between groups will be provided [18, 26, 27]. The infection status and the pandemic time-period will be included in the multivariable Cox model. Moreover, in case the number of censoring due to COVID-19 is not negligible, the treatment effect will be evaluated by using the inverse probability of censoring weighting method. The weights will be calculated by a logistic regression model considering baseline characteristics and time-dependent covariate such as the infection status and pandemic time-period [28, 29].

Finally, due to the reduction in sample size and consequently the decrease in power, an analysis assessing the association of the survival time with the disease progression (ALSFRS-R slope) will be conducted. We will perform the Cox multivariable regression, in which the primary exposure is the slope of ALSFRS-R score. The analysis will also be repeated considering the responder status as the main exposure.

Safety/harms

All AEs, distinct by severity and relationship with the randomized drug, will be documented and listed. The number and percentage of AEs will be reported, per treatment arm and overall, along with the number and percentage of participants who experienced at least one AE. Moreover, the number of concomitant medications and changes from baseline in vital signs, routine biochemistry and hematology analyses, and in respiratory endpoints will be analyzed. Discrete safety endpoints will be compared using a chi-squared test. The denominator will be the safety population. Continuous safety endpoints will be compared using a t-test for parametric variables or the Mann–Whitney test for non-parametric variables. The assumption of normal distribution will be checked using Shapiro–Wilk’s test.

Furthermore, AEs will be described according to the different pandemic periods [18].

Statistical software

Analyses will be carried out by the STATA 17 and R software (version 4.3.0).