Table of contents
At a glance
How do you validate a SAS analysis for GMP use?
Define the SAS analysis, including data sources, programs, products and operating environment. Assess existing installation and developer evidence and identify missing process evidence. For an independent comparison, specify keys, completeness rules, missing-value handling and appropriate tolerances. Record the executed versions and assessed outcomes. A successful job or IQ/OQ report alone does not validate the entire customer application.
- SAS 9.4, Viya, Compute, and CAS must not be considered as a single unified functionality.
- Distinguish installation and component evidence from the correctness of a customer analysis.
- A results comparison must consider key parameters, completeness, outliers, and numerical criteria.
- The executed program, data timestamp, protocol, and evaluated result must demonstrate a clear and traceable relationship.
Which SAS analysis must function reliably?
Start with the decision derived from the results. A quality analysis can filter data, calculate key performance indicators, and generate a report for further evaluation. This is not solely dependent on formulas. A flawed data state or misassigned identity can render a mathematically correct value functionally unusable.
This article examines statistical programming and data analysis in the context of human medicinal product GMP. Annex 11 is a GMP guideline; together with Chapter 4, it forms the regulatory basis here. Clinical research, medical devices, and US requirements necessitate a separate assessment. The use of SAS or its prevalence in regulated industries does not automatically establish a specific application scenario.
Describe inputs, processing, professional criteria, and result recipients. Define which errors would preclude intended use. These specifications assist in limiting audit effort. This guide does not address a blanket validation of all SAS products and does not assign a fixed software category to any product name.
Capture SAS 9.4 and Viya separately
The SAS installation page mentions SAS 9.4M9, delivered in June 2025. In addition, there are product-specific maintenance stands and hotfixes. An entry "SAS 9.4" alone does not describe the complete inventory. Determine which products and interfaces are actually installed and licensed and which of them uses your evaluation.
Viya operates under a different versioning model. The vendor distinguishes between Stable and Long-Term Support cycles, annual and monthly versions, and specific releases. The September 2026 Support Overview extends through 2026.09. The cited Viya Programming Handbook is stated in the title to apply from 2020.1 to 2026.06. For later features, use the documentation that is relevant to your current version.
Use the following internal matrix as a baseline assessment. A development environment, a statistical add-on module, and an execution server perform distinct functions. Although source code can be reused, specific aspects such as data access, permissions, runtime conditions, and result storage must be individually verified.
| Level | Record the details | Open inspection question |
|---|---|---|
| SAS 9.4 | Products, Maintenance, Hotfixes, and Platform | Which evidence applies to this specific inventory? |
| Viya | Kadenz, Version, Release, and Operating Mode | What update and operating conditions apply? |
| Execution | Compute, CAS and relevant data references | Where are the employed steps performed? |
| Data access | Sources, Interfaces, Authorizations, and User Roles | Is the complete and accurate dataset available? |
| Analysis | Program, parameters and technical method | What independent expectation applies? |
| Result | Report, dataset, log and storage location | What remains after a run is retained and assessable? |
Do existing IQ/OQ reports actually cover the intended use?
The SAS-9.4 Qualification Tools manual describes two different tests. The IQ tool compares installed files based on checksums. Changes by configuration or license maintenance can create explainable differences. A conspicuous finding must therefore be evaluated in its specific context.
The OQ tool executes SAS programs provided by the component developer and reports the results of these executions. The scope described applies to Windows and UNIX environments; OQ for z/OS is not supported herein. The handbook used here dates from 2021 and does not confirm the suitability of each test suite described therein for a current M9 installation.
Request the actually available and appropriate third-party documentation for your initiative. Review version, configuration, test scope, results, and open deviations. Suitable external documentation can support your work; it does not need to be automatically duplicated. At the same time, component testing does not substitute for evidence of your filter performance, calculations, or result assignments. The remaining gap must be explicitly identified.
Understand the dataset prior to the value comparison
A professional data set contains more than one row. For quality evaluation, sample identity, time point, unit, and data date may be critical. Determine which combination uniquely identifies a data set. Then verify whether expected entries are missing, duplicated, or unexpectedly present.
PROC COMPARE assigns observations without an ID specification according to their positional order. When ID variables are specified, the assignment can be aligned to key values. The documentation addresses requirements regarding variable types and sorting. The presence of an ID field alone does not necessarily indicate that the comparison properly utilizes this field.
Prepare both inventories appropriately and verify uniqueness. A technically equivalent file with a different line order shall not be considered modified due to the order. Conversely, a comparison of remaining matching lines must not obscure missing data sets. Present completeness and value checks as separate results.
Define tolerance and outlier rules prior to run execution
A comparison result requires a clear expectation. For identities or categories, a different rule is appropriate than for numerical results. PROC COMPARE supports various comparison methods. When METHOD=ABSOLUTE, the absolute difference is compared against CRITERION; a difference greater than this threshold is deemed unequal.
Select the tolerance based on the significance of the size and the calculation method. A limit must not be expanded retrospectively, except to eliminate an observed difference. Also consider the actually implemented method: a number alone in CRITERION does not sufficiently define the intended comparison metric.
Missing values require a separate evaluation. The NOMISSING option may set a missing value equal to any arbitrary comparison values. Do not use this feature implicitly as evidence of completeness. Whether a missing value is professionally justified must be derived from your data and analysis rules. Additionally, review the log messages; the completion of a run does not constitute a documented business decision.
Test example: three IDs and an absolute limit
This is a worked example and was not executed as a SAS test. The reference data contain the distinct IDs 101, 102 and 103 with values of 10.00, 20.00 and 30.00 in the same unit. Expect precisely those three IDs, suitable data types and an absolute difference of at most 0.01 for each value. The tolerance applies only to this example.
Both records are sorted appropriately before the ID-based comparison. A NOMISSING exception is not used. The comparison file may first have a different order. After correct allocation all values match; their sum is 60.00. If the value is changed to 103 to 30.02, the difference is 0.02 and thus more than the specified 0.01.
Every error scenario begins anew with a copy of the complete normal inventory. Preserve the initial system state and predefine the expected response. Independent decimal calculation verifies the numerical values; it does not claim a SAS run. For a numerically precise case at the boundary, additional evaluation under the specified representation and computational conditions is required.
| Case | Targeted input | Independently established expected result | Expected evaluation |
|---|---|---|---|
| Other sequence | 103/30.00; 101/10.00; 102/20.00 | After sorting and ID assignment, no value deviation; total 60.00 | Normal operation exists |
| Altered value | 103 receives 30.02 | Absolute difference 0.02 > 0.01; Total 60.02 | Assess numerical deviations |
| Missing ID | 103 is completely waived | 2 expected IDs instead of 3; ID 103 is missing | No complete result substantiation |
| Duplicate ID | Additional line 102/20.00 | 4 rows, 102 appears twice, total 80.00 | Resolve ambiguous identifiers before release |
| Missing value | 103 present, value missing | ID inventory complete; a value not assessable | Apply the missing-value rule before treating values as equivalent |
Verify Viya deployment and data storage explicitly.
The Viya documentation distinguishes between SAS Compute Server and Cloud Analytic Services, abbreviated as CAS. Not every step is executed within the same context. Code, data references, and supported functions influence the execution path. Therefore, transferring a program from SAS 9.4 does not constitute evidence of identical runtime and memory conditions.
The Programming Handbook describes CAS as an environment for in-memory processing and parallel processing. It also distinguishes session-specific data from explicitly stored or otherwise provided results. Review your workflow to determine which data set was read and what remains available after session termination.
A practical review case concludes the intended session and allows an authorized person to subsequently retrieve the evaluated result. Are the data status, assignment, and content consistent? The review case links the calculation to the intended pathway for future result utilization. A technically accessible dataset is not automatically equivalent to a released report.
Program, log, and report to provide a clear justification for a binding traceability record
Maintain the executed program version, key parameters, input reference, and results. A log can be valuable for evaluating a run's execution, but it does not automatically replace a traceable history of all relevant data changes. The additional controls required depend on the specific documentation process.
Chapter 4 requires structured documentation and the definition of relevant raw data in electronic records. Therefore, determine what is needed for the subsequent evaluation of the data analysis. This may include, in addition to the report, user inputs, technically significant intermediate results, assessment notes, and controlled corrections. A general retention of every technical intermediate object is therefore not justified.
Define in addition the handling of interrupted or repeated runs. Why was the calculation repeated? Which run was used and what deviation was resolved? The original information must not be lost when a permissible correction is made. Verify access and retention measures on the actual platform, without inferring signature or audit trail guarantees from the name SAS.
Assess maintenance, Viya updates, and program modifications
A change to data access, program, or execution environment may require different validation evidence than the absence of a newly implemented feature. Document the reason and affected processes. Assign manufacturer information and existing regressions; add cases for newly identified failure pathways.
In Viya, version and specific release must be distinguished. The support overview additionally lists the Kubernetes platform requirements and supported update paths. However, a technically supported path does not constitute a professional suitability assessment of your analysis. Conversely, every patch does not automatically necessitate a full revalidation.
In the case of migration, data significance, key, format and use of results are to be investigated. A comparison of selected numbers is not enough if units or assignments have been changed. Plan also how previous evaluated results remain achievable. The conclusion determines which tests have been performed and which open points influence the intended use.
Start with a specific program execution
Select a relevant and manageable analysis as a starting point. The following checklist makes visible which information is already available and who can provide missing data. A comprehensive collection of general SAS documentation does not replace this specific entry point.
The department defines the intended use and acceptance criteria; programming specifies processing and comparison logic, IT manages system operation, and QA ensures integration into the quality management system. Key drivers for effort include heterogeneous product formulations, numerous data sources, ambiguous key identifiers, missing reference data, and historical results lacking clear program linkage. These factors enable the derivation of work packages; generic project duration estimates would be difficult to justify in advance.
- SAS products, maintenance/releases, hotfixes, and operational responsibility to be documented.
- Select an actual data pathway including the used programs and parameters.
- Verify the existing IQ/OQ and developer records against their respective scope.
- Unique keys, completeness, erroneous values, and comparison metric defined.
- Prepare independent target values as well as controlled normal and error inventories.
- Assignment of results logging, evaluation, corrections, and restoration.
Project experience and collaboration
Trace the analysis through to a usable results report
With analysis owners, we follow a known dataset from import to the results report. We compare keys, missing values and agreed calculation rules against independently established expectations. Existing IQ/OQ and developer evidence may cover some elements. Custom programs and data preparation may require additional business cases. With IT, we record the program and data versions actually run and check later retrieval of the evaluated report. A successful job alone does not demonstrate the result.
Frequently asked questions about SAS
Is SAS already validated because it is used in regulated companies?
Distribution is not a validation of your specific application. Existing supplier documentation can be valuable but must be relevant to the product, version, and intended use. For your data, software, parameters, and operating conditions, separate validation evidence may still be required. The key factor is a justified demonstration of coverage of your specific requirements, not a general statement about the SAS brand.
Can a validation approach cover SAS 9.4 and Viya together?
The overarching methodology may be shared. However, the technical environment and specific testing activities must clearly differentiate the configurations. SAS 9.4 employs maintenance and product versions, while Viya additionally includes cadence and specific releases. Execution servers, data access, and storage must be separately assigned. A common document is feasible, provided that the evidence does not become ambiguously intermingled across different configurations.
Does a successful SAS-IQ/OQ run replace our technical tests?
It may cover relevant portions of the evidence. Verify the actual test batch, its version, and the results. The described IQ checks files; the OQ runs provided component programs. This does not imply validation of your customer-specific filter or sample assignments. Conversely, sufficient matching supplier assessments do not need to be re-performed solely due to their origin.
Why can PROC COMPARE show differences despite identical data?
Without appropriate assignment rules, observations are compared based on their positional order. A changed sequence may then lead to inappropriate row comparisons. With suitable ID variables, the comparison can be aligned to keys; the types and sorting criteria must be compatible. First assess the comparison requirements and then evaluate actual value deviations, rather than prematurely concluding any result to be a calculation error.
Sources: [3]
How do you select an acceptable numerical comparison tolerance?
Justify the approach based on the size, calculation method, and intended use. Clearly define the reference method. An absolute difference is distinct from a relative measure. The example limit of 0.01 must not be accepted without evaluation. Assess anomalies in accordance with the pre-defined rule and do not alter this rule unilaterally to render an undesirable result acceptable.
Can a comparison inadvertently overlook missing values?
Yes, the specified options are relevant. The documented NOMISSING effect can assign missing values to arbitrary comparison values. Furthermore, a comparison of existing rows does not automatically address the question of missing IDs. Therefore, value content, expected identities, and uniqueness must be assessed separately. Whether a erroneous value is acceptable must derive from your subject-matter data rules.
Sources: [3]
Are Compute and CAS interchangeable in Viya?
The documentation outlines different implementation scenarios. Code and data references determine where processing occurs and how results are delivered. A general equivalence across all functions or numerical outputs cannot be inferred. Review the intended workflow together with its associated data and criteria, particularly when the execution path differs from a previous process.
Sources: [4]
What documentation should be available prior to a SAS project?
A concrete evaluation, the product and version list, data description, program status and well-known professional expectations are helpful. Complete existing installation or test reports and the rules for operation and result evaluation. Missing information is recorded as work orders. A clear representative process is usually more meaningful for the start than an uncommented collection of all manufacturer documents.
Who must collaborate when internally only a limited number of SAS specialists are available?
The following expertise is required: understanding of program evaluation, comprehension of the software programs, platform knowledge, and quality assurance. These responsibilities can be assigned according to organizational structure. In particular, independent assessment of the expected outcomes must not be based solely on the observed program results. Engage these individuals early in the planning process and clearly define the specific tasks to be performed by the external provider, as well as which decisions remain within the organization.
Sources: [8]
What drives the effort and duration of validation or migration?
Key factors include the number of relevant programs, data sources, and platform variants, as well as the quality of existing reference records. Ambiguous key parameters, historically accumulated metrics, and unclassifiable results may require additional analysis. The planned transition to Viya introduces further questions regarding execution and data storage. Only after this scope definition can reliable work packages and timelines be established; the product name alone does not provide a general figure.
Sources and review date
As of 6 October 2026. SAS-9.4 Reference: M9, along with the Procedure Handbook dated 15 September 2026; Qualification Tools Handbook from 2021, including a separate fit-for-purpose assessment. Viya Function Reference: Programming Handbook up to 2026.06; Support Overview up to 2026.09. Regulatory Example Scope: Human Medicinal Product-GMP. Tables and calculated values are standalone work aids and do not represent completed product testing.
- SAS 9.4 Maintenance Releases and Product Upgrades
- SAS 9.4 Qualification Tools User’s Guide
- Base SAS 9.4 Procedures Guide, Seventh Edition
- SAS Viya Platform Programming: Getting Started
- Updating SAS Viya — SAS Viya Forge
- SAS Viya Platform Operations: Versions in Standard Support and Update Paths
- European Commission: EudraLex Volume 4, Human Medicinal Products - GMP
- EU GMP Annex 11: Computerised Systems, January 2011 revision
- EU GMP Chapter 4: Documentation, January 2011 revision
Sources checked: .
Build the justification for your SAS analysis validation
We support you in defining platform, data, and analysis boundaries, as well as in evaluating existing evidence and implementing specific testing protocols.
- Computer system validation →
Bring your GxP systems into validated operation, from laboratory software, MES and eQMS to ERP. We define the scope, plan risk-based tests using GAMP 5 and hand the evidence to QA for release.

