Missing Data. The modified Xxxxxx Scale assigns a worst outcome score, 6, to deceased individuals, obviating the need for separate adjustments to the primary analysis to handle death as an outcome. For the BI, NIHSS, GOS, and SIS, missing 90-day endpoint values will be replaced with the worst case value if the patient died, e.g. BI = 0, INIHSS=42, GOS = 5. If the patient did not die, patients with data from a visit after day 7 but missing data on day 90 will be analyzed employing the last observation carried forward (LOCF). Patients with no data available from any visit after day 7 will have will have worst-case values assigned for the day 90 datapoint, e.g. BI = 0, NIHSS = 42, GOS = 5.
Missing Data. [...***...]
Missing Data. Years 2001 and 2002 were excluded from this analysis due to low country response numbers compared to the remaining nine years included in the dataset. Observations with a negative or zero value for number of people prosecuted were also excluded from the analysis—countries with zeros recorded tended to have missing values for other self-reported variables and seem to have been coded as an alternate to missing, while many were missing across the board (including country and year) with a zero value for number prosecuted. For the nine-year window, there were 45 countries with more than 5 years of missing data for the number of people prosecuted for human trafficking offenses, most of which are from the Tier 2/Tier 2 Watch List category and 21 of which are located in Africa. This suggests that missing data could be informative to the research question and is a limitation to this analysis. Future work should explore methods to most appropriately address the missing data issue.
Missing Data. No imputations of missing data will be performed. However, the following rules will be applied to ensure that all patients can be included in the final analysis: • Patients who are withdrawn from the study prior to Week 8 because of safety concerns or poor efficacy will be classified as non-responders from the time of their withdrawal in all analyses of response status, and their data will be censored at time of withdrawal in all time-to-event analyses. For continuous endpoints in such patients, all analyses for time points beyond the point of withdrawal will exclude missing data for these patients. • Patients who do not reach Week 8 because of early transplant will be classified as responders beyond their time of withdrawal in all analyses of response status, and their data will be censored at time of withdrawal.
Missing Data. Missing data was scrutinized for patterns (Xxxx & Xxxx, 1991) and then were treated based on level of missingness, patterns of missingness, and potential associations between the missing values and the outcome and other study variables of interest (Xxxxxxxx et al., 2006). Multiple imputation (Xxxxx, 1987) has been suggested by the FFCW research team as an acceptable method for dealing with missing data in the FFCW dataset (Xxxxxxxx, 2017) and was considered as an option. However, we instead handled missing data using full information maximum likelihood (FIML) because missing data on predictor variables was limited (range 0% -4%) (Xxxxxxx, 2012).
Missing Data. As is characteristic of RW data, the availability and completeness of relevant data may be limited. Missing data for the primary analysis will not be imputed, and the data will be analysed as recorded in the study; the mixed model for repeated measurements offers a simple alternative to handle missing data without requiring imputation. However, a sensitivity analysis may be performed using a method such as multiple imputation using the missing at random assumption to assess the impact of missing data on the primary objective. Missing data methods will be further defined in the SAP.
Missing Data. There was very little missing data across all self-report questionnaire measures (less than 1%). Case mean substitution technique was used when data were missing on less than 30% of items (Xxx-Xxxxxxxxxx & El-Xxxxx, 2005). This method ascribes the participant’s mean score based upon items that are present for that participant (Xxxxxxx, 1986). When missing data for a subscale of a measure was less than 30% for a particular participant on a specific subscale, it was replaced with the mean for that particular participants subscale. For Posttraumatic Cognitions Inventory (PTCI) and RAND 36-item Health Survey Questionnaire (RAND-36) subscale scores, the mean of all available items was taken even when missing items exceeded 30%, as per scoring instructions. This occurred only in three instances: for one participant within the social functioning (one out of two items missing) and the general health (four out of five items missing) subscales of the RAND 36; and for another participant on the Self-Blame subscale on the PCTI (one out of two items missing). If missing data exceeded 30% within a subscale where items were summed to obtain subscale scores, then the particular case was excluded for analyses for which the missing data was required. One carer did not complete any items on the Hospital Anxiety and Depression Scale (HADS) or on three of the RAND-36 subscales (energy and fatigue, emotional well-being, and pain); therefore this participant was excluded from specific analyses where these data were required. The participant who did not complete any of the interviews was excluded from specific analyses where interview data was required.
Missing Data. Thirteen cases were not included in the analysis due to neither the wellbeing or carer burden data being complete i.e. no pre measures, no post-measures or in 3 cases no outcome measures at all – see Table 1. Reasons for incomplete datasets include dropping out of the programme for personal reasons, a failure to complete and return measures, and in one case a participant declined to complete the post-group outcome measures as she said it made her feel worse to answer the questions. In addition, for 4 cases wellbeing data was available but not carer burden and in 5 cases carer burden was available but no measure of wellbeing was administered (this was the first group delivered in Dec 2005). Group 1: 2005 8 5 N = 3 no post measures Group 2: Summer 2006 4 4* Group 3: Oct 2006 6 4 N = 2 no post measures Group 4: Spring 2007 3 3 - Group 5: Oct 2007 7 6+ N = 1 declined to complete measure Group 6: Spring 2008 5 4 N = 1 no post measures Group 7: Autumn 2008 4 4 Group 8: Autumn 2009 7 6+ N = 1 no post measures Group 9: Spring 2010 1 1 - Group 10: Autumn 2011 5 4* N = 1 did not complete group Group 11: Jan 2012 7 4 N = 2 did not complete group N = 1 no post measures Group 12: Summer 2012 4 3 N = 1 no post measures Group 13 Winter 2012 6 6 - Total: N = 67 N = 54 N = 13 * 2 participants, one in each group, did not complete the Xxxxx Xxxxxx Interview pre- and post- intervention + 2 participants, one in each group, did not complete the coping questions. Anonymised evaluation questionnaires were also administered and data was available for 42 participants. The majority (n=34) completed a 14 item questionnaire which probed practical issues alongside questions relating to outcomes. Six questions relating to content were coded: i) number of sessions ii) were expectations met iii) what did the participant learn? iv) what did they like most? v) what did they like least? and vi) would they recommend the group? The remaining questions focussed on how the participant heard about the group, information provided before the group, timing of group, number of participants in the group, length of sessions, small or large group exercises and an open question. These data were coded into recurring themes. 8 participants completed one of two versions of a different questionnaire with slightly amended questions and a Likert scale for responding. As these datasets were in the minority, data was extracted in accordance with the 6 questions relating to content detailed above.
Missing Data. Missing observations caused by dropouts or uncompleted responses might cause a problem in studies of longitudinal data. When the analysis is restricted to complete cases and the missing data depend on previous responses, the generalized estimating equation (GEE) approach, which is commonly used when population-averaged effects are of primary interest, can lead to biased parameter estimates. There are also different approaches to address such problem, i.e. Multiple Imputation (Xxxxxx & Xxxxx, 1987) or Multiple Imputation by Chained Imputation (Xxxxx, Xxxxxxx, & Xxxx, 2011). However, if only a relatively small proportion of the data contain missing values the records containing missing data can be deleted using only the complete cases in the analyses. Further, in the statistical literature missing data are usually classified according to the underlying reasons of being missing. If data is missing completely at random (MCAR), the probability of a value being missing is independent of both the observed data and the unobserved data, e.g. by tossing a dice, comparisons are generally not subject to bias. When, in a function Y= f(X) relating an outcome with exposure X, the probability of a particular value y being missing depends only on the observed data (Y or X), then the missing data is considered to be missing at random. If the missing data can be considered missing at random, the estimates obtained for the quantile regression, are unbiased. If this assumption is false, the missing data are not ignorable and the missing mechanisms should be modelled (Xxxxxx & Xxxxx, 1987). In the work presented here using the large cohort of Swedish women the data used are essentially complete. Inverse-probability weighting (IPW) is more sophisticated method for handling missing data, which make the weaker assumption that the data are missing at random. We will apply this method to determine whether this method changes the conclusions of the original analysis.
Missing Data. Importantly, the survivor function for an individual can be calculated for any time within the study period, including those where they have missing data. The growth curve model tackles both sporadic missing observations and the right censoring by assuming that the ignorable missing data property of maximum likelihood estimation (Little and Xxxxx 2019) is applicable, allowing missingness to depend upon exogeneous covariates and both earlier and later observed depression scores (Xxxxxx and Xxxxxxx 1994) . Furthermore, predictors of missing data can be included in the growth curve model to give unbiased estimates used in the calculation of survival. However, data that are missing not at random require more complex procedures.
