Date and Time Imputation Flag Variables

When a date or time is imputed, it is required that the variable containing the imputed value be accompanied by a date or time imputation flag variable. The variable fragments to be used for these variables are DTF and TMF, as defined in ADaMIG Table 3.1.5.1 (https://www.cdisc.org/standards/foundational/adam/). DF and TF can be used as abbreviated forms of DTF and TMF, respectively, when the use of DTF or TMF would create a variable name longer than 8 characters. These additional imputation flag variables are conditionally required. The root, identified by "*", of the names of each pair of variables, *DT and *DTF (or *DF), should be identical. The same is true for the corresponding time and imputation flag variables *TM and *TMF (or *TF). Thus it is good practice to limit roots to 5 characters in length.

Section 2.9.5.6 Predefined Standard Variables for ADSL contains sets of timing variables which share a common prefix, such as TRTSDT, TRTSTM, and TRTSDTM. It should be noted that in many instances in Section 2.9.5.6 Predefined Standard Variables for ADSL, specific DTF and TMF flags are defined within sets of timing variables. However, imputation flags should be created for all date or time variables when imputation has been performed, even if there is not a specific imputation flag variable mentioned in Section 2.9.5.6 Predefined Standard Variables for ADSL. For example, the imputation flag variable has not been listed for EOSDT, but EOSDTF must be present if date imputation was performed.

1. As described in Section 2.9.3.5, Variable Naming Fragments, variables whose names end in DTF are date imputation flags. *DTF variables represent the highest level of imputation of the *DT variable based on the source SDTM dataset DTC variable. *DTF = Y if the year is imputed. *DTF = M if year is present and month is imputed. *DTF = D if only day is imputed. *DTF = null if *DT equals the SDTM dataset DTC variable date part equivalent. If a date was imputed, *DTF must be populated and is required. Both *DTF and *TMF may be needed to describe the level of imputation in *DTM if imputation was done.
2. As described in Section 2.9.3.5, Variable Naming Fragments, variables whose names end in TMF are time imputation flags. *TMF variables represent the level of imputation of the *TM (and *DTM) variable based on the source SDTM dataset DTC variable. *TMF = H if the entire time is imputed. *TMF = M if minutes and seconds are imputed. *TMF = S if only seconds are imputed. *TMF = null if *TM equals the SDTM DTC variable time part equivalent. For a given SDTM DTC variable, if only hours and minutes are ever collected, and seconds are imputed in *DTM as 00, then it is not necessary to set *TMF to "S". However if seconds are generally collected but are missing in a given value of the DTC variable and imputed as 00, or if a collected value of seconds is changed in the creation of *DTM, then *TMF should be set to "S". If a time was imputed *TMF must be populated and is required. Both *DTF and *TMF may be needed to describe the level of imputation in *DTM if imputation was done.

Note that using SDTM --DTC source variables for comparison purposes in analysis algorithms may be problematic in the presence of missing date or time elements. SDTM --DTC variables containing date, time, and datetime values are character strings that, in the presence of missing elements (i.e., year, month, day, hour, minute, second), sort or compare in a manner that may be equivalent to imputation of missing elements with the lowest possible value. For example, if in a given --DTC variable in a dataset, dates are present on all records but time is missing on some records, then within any given date, the records with missing time may sort or compare before the records that contain a value of time. Thus the --DTC variable would sort or compare in a manner that is equivalent to imputing midnight when time is missing. The sort or comparison may work mechanically, but imputing midnight may not be the most appropriate thing to do for statistical analysis. Further, the effective imputation of midnight would be hidden and not made explicit. It is important to consider the implications of implicit or explicit imputation whenever dates, times, or datetimes are compared or sorted.