Key research papers on patient safety

How the literature is organised

The founding documents of the field are book-length reports rather than journal articles, so the oldest works listed here are from 2003. The journal literature is organised by the part of the system it examines.

One group studies organisations and culture. Nieva and Sorra (2003) set out safety culture assessment as a tool for improvement and described the instruments then available; Azyabi, Karwowski and Davahli (2021) name the ones in use now — the Hospital Survey on Patient Safety Culture, the Safety Attitudes Questionnaire, the Patient Safety Climate in Health Care Organizations, the Modified Stanford Instrument and the Scottish Hospital Safety Questionnaire. A second group applies human factors engineering: Carayon et al. (2006) present the SEIPS model as a framework for relating structures, processes and outcomes in health care, worked through for outpatient surgery. Wiegmann et al. (2022) do the same service for the Swiss cheese model, arguing that the underlying theory of active and latent failures is often misunderstood by both its users and its critics. A third group studies teams: Manser (2009) draws on three lines of evidence — incident analyses, staff perceptions and observational studies of team behaviour — for teamwork in operating rooms, intensive care and emergency medicine, and Aggarwal et al. (2010) review training and simulation.

The fourth group, and the most useful one for nursing students, concerns the workforce. Rogers et al. (2004) had 393 hospital staff nurses keep logbooks and found that about 40% of the 5,317 shifts logged ran beyond twelve hours, with significantly increased error risk on those shifts, on overtime and above forty hours a week. Aiken et al. (2012) surveyed 33,659 nurses and 11,318 patients in 12 European countries plus a US sample and found that nurses in better work environments were half as likely to give their hospital a poor or failing safety grade, while each additional patient per nurse raised those odds. Hall et al. (2016) reviewed 46 studies and found poor wellbeing and burnout associated with worse patient safety in most of them, while noting that the lack of prospective studies makes causality hard to establish. Specific hazards have their own entry points: Burke (2003) on health care-associated infection and Trzeciak and Rivers (2003) on emergency department overcrowding.

Main debates

The first argument is about the systems view itself. Wiegmann et al. (2022) note that some authors dismiss the Swiss cheese model as an oversimplification of how accidents happen while others try to extend it, and their review is an attempt to settle what the theory actually claims before either side proceeds.

The second is about measurement, and it matters for any thesis built on survey data. Vikan et al. (2023) mapped 34 studies of the association between patient safety culture scores and adverse event rates: 76% found that higher culture scores went with fewer adverse events, but the studies differ in the tools they use, the units they measure and what they report, and the review found too few comparable studies to pool. Kakemam et al. (2021), surveying 2,295 nurses in 32 Iranian teaching hospitals, is an example of how the two are usually measured together — both by asking the same nurses. Janes et al. (2021) meta-analysed staff engagement against safety outcomes and found the relationship real but small (r = 0.22), and say plainly that causality is not established.

Where recent work is heading

Recent well-cited work covers clinician welfare after incidents: Vanhaecht et al. (2022) reviewed 83 publications and put a consensus definition of the second victim to an international expert group. Nijor et al. (2022) reviewed 28 studies on information overload in electronic health records and found irrelevant information, poor data display and excessive alerting raising cognitive load and error rates. Dietl et al. (2023) surveyed 137 staff before and after a four-hour communication intervention in two obstetric units and found perceived patient safety risks lower afterwards, though not communication or team performance. Bates et al. (2021) reviewed 392 studies on artificial intelligence across eight harm domains and expect the largest gains where current strategies fail and unstructured data have to be combined — adverse drug events, decompensation and diagnostic errors. Ocloo et al. (2021) reviewed 42 reviews on patient and public involvement and found theory and equality and diversity to be the neglected corners of it.