The inpatient ward environment is the basic unit of a healthcare facility. Both intrinsic and extrinsic influences on this unit often dictate the quality and safety of care. For surgical patients, although significant focus has been given to care quality in the peri-operative phase, it has become increasingly evident that the overall outcome is determined during the post-operative period of care on the surgical ward. This is demonstrated through the concept of failure to rescue, where stark differences in mortality rates between institutions are seen despite similar complication rates. Research to identify the drivers of these variations often focus on specific themes, rather than evaluating the surgical ward as a system. Furthermore, much of the research examines large administrative datasets with analysis conducted at the institutional, rather than unit level.

This thesis assesses the degree of variation that exists on surgical wards and identifies contributors to error that span the Donabedian model; this considers how processes are performed and how structural factors may influence outcomes. A close examination at the grassroots level has facilitated the identification of granular new metrics with direct relevance to day-to-day care at the ward level. With this approach, the potential for real-time risk assessment of this environment has been proposed, along with future directions to realise this objective.

This has been achieved via a sequence of studies that utilise a range of methods. Following a review of the current literature, a semi-structured interview study was conducted across multiple sites. The experiences of surgical patients, nurses, doctors and managers was explored through 51 interviews. Stakeholders were acutely aware that some surgical wards were safer than others and were able to identify errors within a number of processes, such as the conduction of ward rounds, communication among healthcare staff and medication administration. Furthermore, the development of potential errors was seen as embedded in complex structural influences; the effective performance of processes was impacted by factors such as staffing shortages, organisational bed pressures (i.e., leading to outlier patients) and a potentially challenging physical environment, with layout and lack of space presented as a potential obstacle to safe care. Participants were also able to propose a range of quality markers that reflected the range of influences at play on the ward.

This was followed with a Delphi Consensus study which organised the wide range of factors identified in the previous study and prioritised those deemed to have the most influence on the delivery of safe care on the surgical ward. An international panel of experts in patient safety and patient advocates considered multiple facets of this environment. Sixty-four of the 85 statements in the final questionnaire achieved consensus, highlighting the inherent complexity of the surgical ward. This led to an ethnographic observational study of surgical wards, with the aim to assess the degree of variability and measurability of these prioritised factors. Three broad domains were observed – processes of care, the care environment and organisational health. Alongside this observation, patients and nurses also completed validated questionnaires that measure safety culture.

There was a high degree of disparity with respect to how a ward behaves as a system from day-to-day. Variation in timings and features of the ward round as well as timeliness of clinical and nursing task completion was demonstrated. Organisational influences (e.g., staffing levels, skill mix, use of temporary staffing, ward occupancy, outlier patients etc) were highly dynamic.

The final study establishes an association between measurable factors identified in the observational study and patient outcomes and presents the feasibility of using these as real-time measures of safe care on the surgical ward. Many of these risk factors are retrievable from routinely collected data and were extracted from electronic health records and duty rostering programmes. The patient outcomes identified were also available from the same data sources, namely wrong time medication errors and clinical deterioration. Preliminary statistical models of harm are presented in this study, thus demonstrating that local routinely collected data may have a role in predictive modelling of the risk of harm within a specific setting. Local teams may be able to harness their own data to predict their own risk. This could help guide future policies and improvement strategies.

In conclusion, this thesis has comprehensively explored the entirety of the surgical ward as a system of care delivery, examined the complex array of factors at play as well as their potential interactions with one another and proposed new granular safety metrics that have a role for predictive modelling of the risk of harm at the local level. Further work is needed to develop these predictive models further, such as establishing methods to measure those factors that are not currently available through routinely collated data. This will allow future iterations of the predictive model to incorporate a wider range of factors that are potentially influencing care quality on the surgical ward, with the aim of enhancing sensitivity and applicability of the final model.