More information about text formats
Conventional statistical process control (SPC) has limitations when used with hospital averse event (AE) data. Much data, especially hospital infections like bacteraemias, arise in complex systems.1 These differ from the simple or complicated industrial systems that produce data that are analysed with such success with conventional SPC. AE data arising in complex systems are often nonlinear. Expected values are often unknown. There is often delay in obtaining the AE data e.g. with bacteraemia data – the patient has symptoms, a blood sample is obtained, it is sent to pathology for culture, analysis and reporting and is finally placed in a suitable database then analysed (one of the benefits of conventional SPC is in providing rapid feedback so an industrial process that is going out of control is promptly identified). Most hospital AEs are relatively uncommon and alert staff such as those in Infection Management will frequently detect a change well before a statistical analysis. However, analysis using a time-series chart is still desirable. It can add confirmation to the observations of Infection Management and Quality Improvement staff. A hospital department can summarise its performance with a chart. Management and the public can be informed. A problem is devising control limits about an often non-existent expected value using a linear mean value that may be atypical of much of the data.
How may this dilemma be overcome? The often changing predicted mean value can...
How may this dilemma be overcome? The often changing predicted mean value can be shown well using a generalised additive model (GAM). Its confidence limits can be useful as they delimit the range of the predicted mean supported by the GAM analysis. Trends and change-points may be identified. However, this fails to address potentially outlying individual observations such as large monthly counts and rates. It is suggested that the supported ranges identified by the confidence intervals about the individual monthly values can help identify outliers. Finally, a seasonal trend test can help identify the trends and change-points that frequently accompany determined efforts to improve or that complicate a deteriorating process. Admittedly, these observations apply chiefly to rate data such as bacteraemias and new isolates of antibiotic-resistant organisms. However, there are also binary data to which they apply.
1. Morton, A., Whitby, M., Tierney, N., Sibanda, N. and Mengersen, K. 2016. Statistical Methods for Hospital Monitoring. Wiley StatsRef: Statistics Reference Online. 1–8.