*Copyright @ www.mycsg.in;

Writing to multiple output datasets in a single DATA step

Why write to multiple datasets in one pass

• The previous DATA step lesson covered explicit OUTPUT statements, which control when an observation is written to the output dataset
• SAS DATA steps can name more than one output dataset in the DATA statement and use conditional OUTPUT statements to route each observation to the appropriate destination
• This allows a single pass through the input data to produce several separate output datasets based on conditions
• Writing to multiple outputs in one step is more efficient than reading the same data multiple times with separate subsetting steps
• It also ensures that each observation is classified and routed consistently from one pass of the data

Create sample data

• We create a dataset of clinical adverse events with varying severity grades
• The goal is to route each event to a severity-specific dataset in one DATA step

data adverse_events; input subject $ event $ severity$ grade; datalines; 001 Headache Mild 1 001 Nausea Moderate 2 002 Fatigue Mild 1 002 Rash Severe 3 003 Dizziness Mild 1 003 Vomiting Moderate 2 003 Liver_Issue Severe 3 004 Insomnia Mild 1 005 Hypertension Moderate 2 005 Seizure Severe 3 ; run; Copy Code View Log

• Confirm that `adverse_events` has 10 rows with three severity levels: Mild (4), Moderate (3), and Severe (3)

View Data

Naming multiple output datasets in the DATA statement

List all output dataset names on the DATA statement

• To write to multiple datasets, list them all on the DATA statement separated by spaces
• When multiple datasets are listed, the default implicit output at the end of the DATA step writes to ALL named datasets unless you include explicit OUTPUT statements
• To route observations selectively, you must use explicit OUTPUT statements that name the specific target dataset
• Once you use any explicit OUTPUT statement in a DATA step, the implicit output is suppressed — you must explicitly OUTPUT every observation you want written

data ae_mild ae_moderate ae_severe; set adverse_events; if severity='Mild' then output ae_mild; else if severity='Moderate' then output ae_moderate; else if severity='Severe' then output ae_severe; run; Copy Code View Log

• Three output datasets are created in one pass of `adverse_events`
• Each observation goes to exactly one output dataset based on its severity value
• Verify: `ae_mild` should have 4 rows, `ae_moderate` 3 rows, `ae_severe` 3 rows
• Because explicit OUTPUT statements are used, no observation is written implicitly — every row must be handled by an IF-THEN branch

View Data

Route observations to more than one output dataset simultaneously

• An observation can be written to more than one dataset by including multiple OUTPUT statements for it
• In this example, we create a dataset of all events, plus a separate dataset containing only the events from subjects who had at least one severe event
• An observation for a subject with a severe event appears in both `ae_all` and `ae_severe_subjects`

proc sql noprint; select distinct subject into :severe_subjects separated by ' ' from adverse_events where severity='Severe'; quit; %put NOTE: Subjects with severe events: &severe_subjects; data ae_all ae_severe_subjects; set adverse_events; output ae_all; if index("&severe_subjects", subject) gt 0 then output ae_severe_subjects; run; Copy Code View Log

• `ae_all` receives every row — all 10 adverse events
• `ae_severe_subjects` receives every event from subjects 002, 003, and 005 because those subjects had at least one severe event — even their mild and moderate events are included
• This pattern is useful when you need a complete profile for a subject based on them meeting a specific criterion in at least one observation

View Data

Split data into train and test sets by row number

• Another practical use of multiple output datasets is splitting a dataset into two portions based on a row counter
• This is a common requirement in model building, where you need a training set and a validation set
• We use the automatic `_N_` variable, which holds the current iteration number of the DATA step

data ae_first_half ae_second_half; set adverse_events nobs=total; if _n_ le (total/2) then output ae_first_half; else output ae_second_half; run; Copy Code View Log

• The `nobs=total` dataset option in the SET statement populates `total` with the number of observations before the DATA step begins reading
• `_N_` is 1 for the first iteration, 2 for the second, and so on
• Observations 1 through 5 go to `ae_first_half` and observations 6 through 10 go to `ae_second_half`

View Data

Combine multiple output routing with additional variable creation

• Multiple-output DATA steps can also create or modify variables just as any other DATA step can
• In this example, we add a numeric severity score and a flag variable before routing each observation
• All output datasets receive the newly created variables because they are assigned before the OUTPUT statements

data ae_low ae_high; set adverse_events; select (severity); when ('Mild') severity_score=1; when ('Moderate') severity_score=2; when ('Severe') severity_score=3; otherwise severity_score=.; end; high_flag = (severity_score ge 2); if severity_score lt 2 then output ae_low; else output ae_high; run; Copy Code View Log

• Both `ae_low` and `ae_high` contain the `severity_score` and `high_flag` variables created during the step
• `ae_low` has 4 rows (grade 1 events) and `ae_high` has 6 rows (grade 2 and 3 events)
• Confirm that `high_flag` is 0 in `ae_low` and 1 in `ae_high`

View Data

Key points to remember

• List multiple dataset names on the DATA statement to create more than one output dataset in a single step
• Once you use any explicit OUTPUT statement in a DATA step, SAS suppresses the implicit end-of-loop output — every observation you want written must have an explicit OUTPUT call
• An observation can be written to multiple datasets by including more than one OUTPUT statement for it
• The `nobs=` dataset option on a SET statement provides the total observation count before the loop starts — use it with `_N_` for row-position-based routing
• Variables created or modified before the OUTPUT statements are included in all target datasets that receive that observation
• One-pass multiple-output DATA steps are more efficient than reading the same source dataset multiple times in separate subsetting steps