The work system, not the worker: SEIPS and the sociotechnical turn in patient safety

If error is a consequence of the conditions of work rather than a property of the person, the conditions are what must be modelled. SEIPS, the Systems Engineering Initiative for Patient Safety, makes that move, and it is this module's deepest explanatory step.

The claim is that performance emerges from the work system, and that the person is one component inside it rather than the cause outside it. At that generality it is nearly unfalsifiable, so its value turns on whether the model can become instruments applied and tested across settings rather than restated. The reframing has changed how investigations are framed and the tools exist. That applying them improves care across settings is not yet established. This seventh of ten units generalises the frame the six before it each assumed, and asks whether a claim that general can be tested at all.

The move: the unit of analysis is the work system

Carayon and colleagues set out SEIPS as a framework relating the structures, processes and outcomes of care (Carayon et al., 2006). The framework decomposes the work system into person, tasks, tools and technologies, organisation, and internal environment. The model adds no further cause to the list that ends in error. It specifies the components whose fit produces the performance observed, so an investigation has somewhere to go after naming the person who acted.

Hignett and colleagues reviewed human factors in healthcare through four of them: occupational ergonomics, surgical safety, design for patient safety, and organisational and socio-technical systems (Hignett et al., 2013). Two are domain applications and two are systems approaches, so the commentary shows what an analysis looks like at each level and sets the range of work the model sits among.

A remedy aimed at one component shifts the load onto another

Smith and Carayon-Sainfort proposed that stress arises from imbalance among the elements of the work system rather than from any one element (Smith and Carayon-Sainfort, 1989). Design must therefore arrange the components together rather than repair one of them. A remedy aimed at one component can make the system worse by shifting load onto another. A ward that answers an interruption problem by adding a recording field has moved the cost rather than removed it.

Karsh and colleagues proposed a human factors engineering paradigm that designs systems both to support healthcare professional performance and to reduce hazards (Karsh et al., 2006). They argue that each goal alone is insufficient. That is why this unit rejects hazard-only remedies: a design that removes a hazard while degrading performance has not made the service safer.

Each restatement widened the model and made testing harder

The 2006 paper sets the structure (Carayon et al., 2006). SEIPS 2.0 adds configuration and engagement, so that patients and clinicians shape care rather than appear only as its recipients (Holden et al., 2013). It also states the redesign principles: balance the work system, and treat workers as active and adaptive rather than as components (Carayon et al., 2014). The carer it engages is often an older person doing a clinical task without training. SEIPS 3.0 expands the process component, using the patient journey to describe interactions with several care settings distributed over space and time (Carayon et al., 2020). SEIPS 101, with seven tools a team can apply off the shelf, is the practitioner version of the same account (Holden and Carayon, 2021). That operationalisation makes testing harder, since the wider the unit of analysis, the harder it is to hold anything constant.

A single application yields a diagnosis, not a test

The clearest recent application concerns the surgical count that prevents a retained foreign object. Meilan and colleagues observed 22 general and gynaecological surgical cases and interviewed staff while building SEIPS 101 tools for the count (Meilan et al., 2025). A journey map found six phases of counting in four environments and three operating room team members. A people map found eight people influencing the process, and a scan produced 56 barriers and 31 facilitators. The analysis preceded any intervention, so it is a diagnosis rather than a test of improvement. Its value is its specificity, since the count fails at the boundaries between phases and people rather than inside a single step. One boundary is a language barrier, where a team works through an interpreter.

The contested question: a tested theory or a restated vocabulary?

The disagreement is not about whether the work-system framing is right but about what it has demonstrated. The systematic review closest to answering it identified 12 projects representing 23 studies of human factors based healthcare system redesign across physical, cognitive and organisational issues (Xie and Carayon, 2014). Its finding is carefully worded: some evidence exists for effectiveness on process and outcome measures of quality and safety. Risk of bias was assessed in the 16 studies reporting an impact and the quality varied. One position holds that the model is applied and productive. On this reading the applications establish its explanatory reach, because SEIPS accounts for quality and safety outcomes a person-level view cannot (Carayon et al., 2014). A recent application produced findings a task-level analysis would not (Meilan et al., 2025).

The other position holds that the field has applied the vocabulary better than it has tested the propositions, and its strongest evidence comes from outside the SEIPS papers. Waterson and colleagues reviewed the methods available for sociotechnical systems and safety, worked through six case studies, and rated them against theoretical and practical questions (Waterson et al., 2015). They found gaps in coverage and applicability, including poor capture of influences external to the studied system. The parallel criticism concerns a proposition set with 15 tenets. A three-round Delphi study pruned it to 14, and kept only 10 for a unified causal model. On that reading the propositions are still being negotiated rather than settled (Salmon et al., 2022).

A third position separates the two questions. Gurses and colleagues treat slow improvement as an integration failure. All five of their recommendations concern capacity, market demand, practitioner numbers and research infrastructure rather than the model's validity (Gurses et al., 2011).

What would settle the disagreement is a study design rather than another application. The review's own call is for mechanisms to be defined, and the methods evaluation implies the criteria. The work-system analysis and the predicted effect are specified before the intervention, the study is repeated across settings, and the outcome could falsify the claim. If a specified work-system factor changes an outcome across sites, the model is working as a theory. If the analyses keep producing rich local accounts and the improvements stay local, it is a method for seeing, which is useful and is not the same thing.

What to do while the dispute is open

The evidence on the two sides is unequal. That the model is productive rests on its applications and on tools anyone can take off the shelf. That it is untested rests on a methods review, on a Delphi that pruned its own tenets, and on the tradition's own review, which claims only some evidence of varying quality. The strongest objection to this unit's thesis is that a claim this general cannot be falsified, so an explanatory success may be a restatement. While the dispute is open, a designer should specify the analysis and its predicted effect before intervening, repeat it across sites, and name the outcome that would falsify the model.

Where this is heading

SEIPS places part of the explanation outside the ward through its organisation and environment components, and Rasmussen states how far outside it runs. Risk management in a dynamic society is a control problem whose structure spans every level from operator to regulator for each hazard category (Rasmussen, 1997). Accidents arise from a mismatch between the demands on actors and the boundaries they work within, so the levers that produced the conditions for error sit above the wards.

SEIPS 3.0 moves the unit of analysis to the patient journey and names the problems that follow. These are multiple perspectives, genuine participation, and work at the boundaries between settings (Carayon et al., 2020). The question for the next decade is whether the turn produces comparisons. A tool applied in one hospital yields a diagnosis. The same tool applied in twenty, with the analysis specified in advance, yields a test. A patient journey also crosses a remote site, where the record is read over a slow connection.

Sources

• Carayon, P., Schoofs Hundt, A., Karsh, B. T., et al. (2006). Work system design for patient safety: the SEIPS model. Quality in Health Care. doi:10.1136/qshc.2005.015842 Sets out SEIPS as a framework relating the structures, processes and outcomes of care, with the work system decomposed into person, tasks, tools, organisation and environment, and applies it in outpatient surgery. The decomposition is what gives an investigation somewhere to go after it has named the person who acted.
• Smith, M. J. and Carayon-Sainfort, P. (1989). A balance theory of job design for stress reduction. International Journal of Industrial Ergonomics. doi:10.1016/0169-8141(89)90051-6 Proposes that stress arises from imbalance among the elements of the work system rather than from any one element. It is the ancestor of the SEIPS balance principle and supplies the rule that a fix aimed at one component can shift load onto another.
• Karsh, B. T., Holden, R. J., Alper, S. J. and Or, C. K. L. (2006). A human factors engineering paradigm for patient safety. Quality in Health Care. doi:10.1136/qshc.2005.015974 Reviews the paradigms that directed patient safety effort and proposes a human factors engineering paradigm that designs to support professional performance and to reduce hazards together, arguing that each goal alone is insufficient. It is the reason this unit rejects hazard-only remedies.
• Carayon, P., Wetterneck, T. B., Rivera-Rodriguez, A. J., et al. (2014). Human factors systems approach to healthcare quality and patient safety. Applied Ergonomics. doi:10.1016/j.apergo.2013.04.023 Sets out principles for redesigning healthcare systems with the SEIPS model, chiefly balancing the work system and encouraging the active and adaptive role of workers. It reviews applications without comparing outcomes across sites.
• Holden, R. J., Carayon, P., Gurses, A. P., Hoonakker, P., Hundt, A. S., Ozok, A. A. and Rivera-Rodriguez, A. J. (2013). SEIPS 2.0: a human factors framework for studying and improving the work of healthcare professionals and patients. Ergonomics. doi:10.1080/00140139.2013.838643 Reviews the SEIPS model and introduces SEIPS 2.0, which adds three concepts to the original: configuration, the dynamic and interactive properties of sociotechnical systems, engagement, the work done by patients, family caregivers and other non-professionals, and adaptation. It is the source of the configuration and engagement concepts this unit attributes to SEIPS 2.0.
• Carayon, P., Wooldridge, A., Hoonakker, P., Hundt, A. S. and Kelly, M. M. (2020). SEIPS 3.0: Human-centered design of the patient journey for patient safety. Applied Ergonomics. doi:10.1016/j.apergo.2019.103033 Expands the process component of the model, using the patient journey to represent a patient's interactions with several care settings distributed over space and time. It names the problems that follow, including multiple perspectives, genuine participation and work at the boundaries between settings.
• Holden, R. J. and Carayon, P. (2021). SEIPS 101 and seven simple SEIPS tools. BMJ Quality & Safety. doi:10.1136/bmjqs-2020-012538 Offers a simplified, practice-oriented version of the model and, for the first time, seven tools that users with no prior familiarity can apply off the shelf. It is the clearest sign that the model's value is being tested through use rather than restatement.
• Meilan, J., Lusk, C., Galinsky, E. K., et al. (2025). Using SEIPS 101 Tools to Model Surgical Count Processes. Journal of Patient Safety. doi:10.1097/pts.0000000000001386 Observed the count process in 22 general and gynaecological surgical cases and interviewed staff while building SEIPS 101 tools for the measure that prevents a retained foreign object. The journey map found six phases in four environments, the people map eight people, and the scan 56 barriers and 31 facilitators.
• Xie, A. and Carayon, P. (2014). A systematic review of human factors and ergonomics (HFE)-based healthcare system redesign for quality of care and patient safety. Ergonomics. doi:10.1080/00140139.2014.959070 Identified 12 projects representing 23 studies of human factors based redesign, assessed risk of bias in the 16 that reported an impact, and found the quality varied. It concludes that some evidence exists for effectiveness and that future work should define the mechanisms of improvement.
• Gurses, A. P., Ozok, A. A. and Pronovost, P. J. (2011). Time to accelerate integration of human factors and ergonomics in patient safety. BMJ Quality & Safety. doi:10.1136/bmjqs-2011-000421 Argues that progress in patient safety has been slow because human factors has been integrated too little, and gives five recommendations covering capacity, market demand, practitioner numbers, investment and interdisciplinary research. It locates the problem in uptake rather than in the model's validity.
• Waterson, P., Robertson, M. M., Cooke, N. J., et al. (2015). Defining the methodological challenges and opportunities for an effective science of sociotechnical systems and safety. Ergonomics. doi:10.1080/00140139.2015.1015622 Reviews the methods available for sociotechnical systems and safety, presents six case studies, and rates the methods against theoretical and practical questions. It finds gaps in coverage and applicability, including how far the methods capture influences external to the system studied.
• Salmon, P. M., Hulme, A., Walker, G. H., Waterson, P. and Stanton, N. A. (2022). Towards a unified model of accident causation: refining and validating the systems thinking safety tenets. Ergonomics. doi:10.1080/00140139.2022.2107709 Ran a three-round Delphi study with experts to validate the 15 systems thinking tenets. The exercise reduced them to 14 and identified 10 as important for a unified model of accident causation, showing the propositions still being negotiated.
• Rasmussen, J. (1997). Risk management in a dynamic society: a modelling problem. Safety Science. doi:10.1016/s0925-7535(97)00052-0 Argues that risk management is a control problem whose structure spans every level of society for each hazard category, modelled by functional abstraction rather than decomposition into parts. Accidents arise from a mismatch between the demands a system places on actors and the boundaries within which they work.
• Hignett, S., Carayon, P., Buckle, P. and Catchpole, K. (2013). State of science: human factors and ergonomics in healthcare. Ergonomics. doi:10.1080/00140139.2013.822932 Reviews human factors in healthcare through four worked examples: occupational ergonomics, surgical safety, design for patient safety, and organisational and socio-technical systems. Two are domain applications and two are systems approaches, so it shows the level each analysis works at.