Table of contents
At a glance
How do you conduct a risk assessment for computer system validation?
Describe the intended process and the consequences of possible failures. Assess existing controls against evidence of how effectively they work. Record uncertainties. Identify additional measures and targeted tests, including relevant failure cases. Reassess risk control using the results. A numerical score alone does not justify the scope of testing or a decision to use the system.
- Assess specific process consequences of failures, not just software functions or module names.
- A described control is only considered valid upon demonstrable suitability and evidence relevant to the considered situation.
- Missing information remains uncertainty; it is not downgraded to a low-risk category.
- Test planning, actual findings, and remaining usage limits fall under the same decision.
What the risk assessment for your validation should determine
The central question is: Which error could occur at which stage and lead to an incorrect quality decision, and what must be demonstrated to mitigate that risk? Therefore, do not start with an empty FMEA template. Begin with the workflow: Who enters the data, what processing follows, who uses the result, and which decision depends on it?
Annex 11 §1 links the scope of validation to a justified risk assessment taking into account patient safety, product quality, and data integrity. In the EudraLex registry, which becomes accessible on 6 October 2026, the revision from January 2011 remains referenced for human medicinal products. Consultation drafts are not considered valid requirements.
The following matrices are internal work aids. They neither replace your process knowledge nor the decisions governed by the quality system. The article examines the depth and alignment of the evidence. The URS defines the required "as-designed" expectations; traceability later demonstrates which records actually support this "as-designed" requirement. A risk assessment should integrate these documents rather than redundantly reiterating their content.
Define process boundary and failure sequence first
Select a professionally coherent segment, for example, the import of a laboratory result up to its use in the test report. Document the involved systems, interfaces, master data, and user roles. A technical import can be correctly completed even if the wrong sample was assigned. Therefore, "interface works" would be an overly simplistic evaluation criterion.
Separate root cause, error, and consequence. A changed field mapping may be a potential root cause. Incorrect sample identification during import constitutes the error. The consequence could be a misjudgment of the material. Furthermore, the same error may lead to delays. However, processing time alone does not reflect the potential quality impact.
Also verify whether the error is detected before any party relies on the dataset. A warning issued after report submission provides protection at a different point in time than a block placed prior to release. Inquire about special cases: partial imports, duplicate files, manual corrections, and proxy roles. Document excluded workflows with clear justification to ensure the scope under consideration remains transparent and traceable.
Assess existing controls with their actual evidence of validation
“Second-person check in place” tells you little about which error the check detects. Describe what the reviewers compare, the independent information available to them and what happens when they find a difference. Two people reading the same incorrectly generated report do not perform two independent checks.
Distinguish prevention, detection, and response. A clear assignment rule prevents a specific misclassification pathway. Comparison with an independent sample submission can identify any remaining misassignments. A documented hold until clarification prevents unauthorized use. These effects must not be credited multiple times for the same mechanism without proper evaluation.
As evidence, appropriate configuration records, actually performed tests, and meaningful operational observations may be used. Verify their version, process linkage, and scope. A training signature confirms participation or acknowledgment according to the training process; it alone does not demonstrate that an unusual error in daily operations has been identified. Organizational controls can be effective if execution, time windows, responsibility, and effectiveness are aligned with the associated risk.
Manage uncertainty without generating false precision
ICH Q9(R1) describes various methods and levels of formality. It does not require a FMEA or a universal risk priority number for each Computerised System Validation (CSV) assessment. The traceability of your decision is more important than an apparently precise decimal figure. A well-justified qualitative assessment may be more appropriate than three independently estimated numbers that are subsequently multiplied together.
When using a scale, define its criteria prior to individual assessment. For example, specify what constitutes a serious outcome within the considered process. Differentiate between estimated occurrence frequency and the probability that a deviation is detected in a timely manner. Do not alter the scale to achieve a desired release threshold.
If the data is missing, do you write down the specific question: Is the detection of duplicate sample identifiers configured? Has it been checked with different files? "No known incident" does not follow any reliable frequency without reliable detection and recording. Get specifically the missing information. A tight appointment is no proof of a lower risk; additional uncertainty can require just more technical clarification.
ICH Q9(R1) Section 5.1 clarifies that resource constraints do not justify a lower level of formality in quality risk management. Section 5.3 identifies, among other factors, poorly defined risk questions and inappropriate evaluation scales as sources of subjectivity. Therefore, self-assessments are prepared with supporting evidence and justified criteria, rather than being adjusted to meeting schedule or budgetary preferences.
Sources: [3]
A concise reference guide for justified test decisions
The following structure can be represented as a table or implemented within your existing system. Assign a stable identifier for each continuous error scenario. Link requirements and supporting evidence through references. Separate observations and assumptions to prevent the next reviewer from assuming a control exists for a verified function.
The evaluation should generate a specific action item: Which question is addressed by which test or other appropriate validation evidence? Additionally, identify who will professionally assess the results. A scheduled test date does not constitute a control verification. Until completion, the underlying uncertainty remains visible in the evaluation status.
| Field | Entry with decision utility |
|---|---|
| Process and Failure Scenario | Result import: A value is assigned to another sample. |
| Possible consequence | A material is evaluated based on an external result. |
| Control and Evidence | Comparison of identifiers; precise configuration, document version, and review to be specified. |
| Uncertainty | Unknown whether repeated imports override an already verified assignment. |
| Additional action | Limit overrides and exception handling to be professionally defined. |
| Decision on evidence | Check for conflict file and duplication; verify expected lock and data state. |
| Reassessment | Record the finding, remaining limit, responsible decision and reason for change. |
Targeted inspections based on risk instead of deriving from test quantities
A successful standard import alone is insufficient for a critical mapping exercise. Purposefully evaluate the pathways through which the identified error scenario may arise. This includes missing identifiers, duplicate data records, incompatible units, and data retransmission following a connection loss. The selection is derived from your interface and data structure; it is not a mandatory list applicable to every application.
Annex 11 §4.7 identifies appropriate test scenarios including relevant boundaries and error handling. Practically, this means: define the expected data state in addition to the visible alert. For example, after a rejected import, a partial value should not remain available unnoticed for evaluation if the process requires a complete, continuous data set.
Supplier tests can serve as a justified source of evidence. Verify that the specific functions, version, and prerequisites they cover are indeed aligned. Internal testing efforts should then focus on remaining questions, such as customer-specific assignments and role configurations. Similarly, document a justified decision against conducting an additional test. "Standard software" or "low score" alone does not establish that an appropriate basis of evidence already exists.
Example: A plausible value is obtained for the wrong sample
This example does not represent a completed product test. Two test assignments are designated TEST-A and TEST-B. The values are 98.4 and 101.2, expressed in the same agreed unit. Both results fall within a hypothetical range of 95.0 to 105.0. A simple plausibility check therefore fails to detect the swap of the two values.
The assumed source control consists of a review of the imported report. Since this report contains the same incorrect assignment, its independence remains unverified. For the example, an additional comparison of order, original export, and target assignment prior to result usage is defined. A unique composite key composed of order identifier and inspection parameter shall also eliminate ambiguous assignments.
The following matrix describes planned checks and expected responses. It intentionally includes no claimed actual results. For actual implementation, test environment, software version, source file, configuration, actual findings, and result review would need to be added. The controls must function correctly in the event of a failure; their presence in the concept does not automatically negate their evaluation.
| Test scenario | Expected behavior in the example | Why this validation is required |
|---|---|---|
| TEST-A = 98.4; TEST-B = 101.2 correctly transmitted | Both values remain correctly linked to the order and inspection parameter. | Documents the intended standard operating procedure. |
| Values at target reception are swapped; the unaltered original export remains for comparison. | Independent comparison with the original export identifies the deviation; usage remains restricted. | Verifies the control against plausible, but incorrectly assigned values. |
| Order identifier for two distinct target samples offered | No automatic selection of an arbitrary target sample; conflict will be resolved. | Tests the specified response to an ambiguous match. |
| Import the file with already processed result again | No silent overwriting of a validated result; defined repeat processing. | Covers the alternative failure path of a repeated import. |
| Comparison data for the reviewer not available | No unsupported release of results; predefined escalation. | Verifies the dependence of the control on its information source. |
After assessment, evaluate the remaining risk control.
Assuming the test shows: A ambiguous identifier is blocked, but the reviewer only sees the target report. Then, one error pathway is controlled and another remains open. Passing a partial criterion does not justify a green overall assessment. Separate documentation should be maintained for what is supported, which control assumption is invalidated, and which portion of the planned usage remains affected.
After correction, review the updated control and the affected neighboring procedures. A new key may inadvertently prevent previously authorized multiple checks. Therefore, the reevaluation also considers new failure pathways introduced by the corrective action. The original dataset and prior findings remain traceable; a repeat test does not replace the history of the decision.
The responsible functions decide on this basis regarding remaining actions and, if applicable, usage limitations. Describe a practical restriction: which procedure, which data, which roles, and which prerequisites? Risk management does not permit otherwise inappropriate practices. Therefore, a signed risk acceptance neither replaces required controls nor the fulfillment of applicable requirements.
Prepare evaluation in SMEs and keep up to date on changes
Bring a realistic process flow, selected example data, known errors, and existing documentation to the meeting. The department explains when a result is used. IT outlines data pathways and potential points of intervention. QA aligns quality implications and traceability requirements. Suppliers can clarify technical questions; the decision on the intended use requires approval from the relevant responsible personnel within the company.
The effort arises primarily from unknown interfaces, inconsistent master data, customer-specific logic, and unclear control procedures. A few complex error pathways can require more work than many well-understood standard functions. Gather open questions with responsible parties and provide concrete closure documentation. This transforms a lengthy workshop into a manageable validation plan.
Review the assessment upon changes, new findings, or relevant operational experiences. A change in the import source may invalidate previous comparisons. An additional shift may affect the reviewer's availability. Periodic review complements these circumstances; it is not a justification for failing to evaluate known relevant changes until the next scheduled entry.
Checklist: Is the risk assessment fit for decision-making?
Review a critical entry end-to-end from process through to result. If the rationale is only discernible from verbal explanations, precisely complete the missing linkage. The objective is to establish a decision that is usable for future changes, not a comprehensively detailed table.
- The process, system version, data flow and use under assessment are clearly defined.
- Cause, failure scenario, and potential quality impact are distinctly described.
- Existing controls specify the time, responsible party, and information basis.
- Documents are aligned to version and process; boundaries remain visible.
- Unknown circumstances are recorded as uncertainties with a task for clarification.
- Evaluation criteria are clear; numbers do not replace professional justification.
- Additional measures address specific failure pathways and potential new risks.
- Tests include relevant failure scenarios, expected data states, and appropriate validation evidence.
- Actual findings lead to a logical re-evaluation and decision.
- Changes, control failures, and new insights require a defined handling procedure.
Project experience and collaboration
Trace the consequence of failure through to a test
Start with a critical process. What happens if a value is missing, assigned to the wrong record or arrives late? IT shows us the existing controls, and the team examines which effects the evidence already demonstrates. A well-understood change may need a focused assessment. For a new process, first establish the intended use and the possible consequences of failure. Use the unanswered questions to define actions and tests, then evaluate the results.
Frequently asked questions about Risk assessment in validation
Is a FMEA required for the CSV risk assessment?
No. ICH Q9(R1) identifies several appropriate methods; FMEA is one option. Select a structure that clearly communicates failure modes, controls, uncertainties, and decisions. If your quality system includes FMEA, apply it systematically. For a manageable import process, a justified qualitative matrix may be sufficient, provided it actually addresses the relevant questions.
Sources: [3]
Which risk priority number allows for fewer tests?
There is no universal regulatory threshold here. An internal threshold is only viable when defined criteria and appropriate technical assessment are applied. First assess which error pathways are already covered by relevant evidence. Reduce additional testing based on this justified coverage, not merely because of a computational result. A serious open defect must not be rendered invisible due to low values in other parameters.
Can we classify an unknown failure probability as low?
Not solely due to missing reports. Clarify whether errors would be detected and documented. Identify gaps in knowledge and obtain targeted evidence, for example regarding assignment logic or repeat treatment protocols. Any preliminary assumption must have clearly defined boundaries. If uncertainty is materially relevant to decision-making, it must be considered in corrective actions, inspection depth, and usage decisions.
Sources: [3]
Does a second-person check automatically reduce risk?
No. Describe the error that the second person detects prior to use and based on which source. Two individuals reviewing the same erroneous target report do not independently verify the original assignment. Test or demonstrate the specific control using an appropriate error scenario. Additionally consider representation, available time, and the response in the absence of comparative data.
Must each requirement have its own risk matrix row?
Not mandatory. Related requirements may address the same failure scenario. Nevertheless, ensure that their linkage remains traceable. Separate entries when consequences, controls, or test decisions differ. A combined line item "Data Integrity" would be overly broad if sample assignment, result modification, and restoration each require distinct validation evidence. Granularity should facilitate decision-making, not increase the number of rows.
Can we use a supplier test as a control evidence?
Yes, if the scope, version, prerequisites, and statement of the document align with your question. A test of the import function can be valuable without exposing your own field mapping. Document the imported portion and remaining questions. The actual evidence is paramount, not the label "validation package." Supplement targeted checks for customer-specific components.
Sources: [2]
How many negative tests are required for a critical process?
The number of cases follows the relevant error pathways and controls. During result import, ambiguity, repetition, and missing comparison data may raise different questions. Three variants of the same simple error do not eliminate this breadth. For each selected case, describe the professional rationale and the expected state. Justify which variants are adequately covered by existing evidence.
Sources: [2]
When is the residual risk following a corrective action re-evaluated?
When the action has been implemented and its relevant impact can be assessed based on appropriate results, a planned restriction is not yet an effective restriction. Evaluate, in addition to the intended effect, potential new failure pathways. If actions remain open, distinguish the documented current state from the intended target state. The responsible decision must be based on the current state.
Sources: [3]
Who must participate in a CSV risk assessment?
Engage individuals who can assess process flows, technical implementation, and quality requirements. For complex processes, this typically includes functional, IT, and QA personnel; additional roles depend on the specific application. A supplier may provide technical evidence. Define within the quality system who evaluates and makes decisions. There is no single, identical list of participants or signatories applicable to every system.
Must a risk analysis be completely rewritten for each update?
No. Review which previous assumptions, controls, and justifications are affected. A modified interface may require a renewed assessment and targeted regression testing, while unaltered justifications remain valid. Document a clear, rationale-based decision without additional testing. An unchanged version label or absence of visible user interface changes alone does not prove the absence of technical impact.
Sources and review date
Primary sources were opened on October 6, 2026. Consideration is given to computerized system validation within human medicinal product GMP. Tables, test decisions, and result imports are standalone work aids; they do not define general risk thresholds.
- European Commission: EudraLex Volume 4 – Current Directory for Human Medicinal Product GMP
- EU-GMP Annex 11: Computerised Systems – Sections 1, 4 and Lifecycle Controls
- ICH Q9(R1): Quality Risk Management – EMA Step 5, 2023
- EMA: GMP/GDP Questions and Answers – Annex 11, Questions 5 and 7
Sources checked: .
Integrate risk assessment and concrete evidence.
Daniel Herrmann Consulting supports QA, IT, and functional departments in evaluating critical failure paths and developing a traceable documentation plan for the intended use.
- GAMP 5 consulting →
Understand how GAMP 5, EU GMP Annex 11 and 21 CFR Part 11 apply to your systems. We translate the relevant requirements into proportionate validation work and evidence your team can defend in an audit.

