Search

BackgroundTraditional approaches for surgical site infection (SSI) surveillance have deficiencies that delay detection of SSI outbreaks and other clinically important increases in SSI rates. We investigated whether use of optimised statistical process control (SPC) methods and feedback for SSI surveillance would decrease rates of SSI in a network of US community hospitals.MethodsWe conducted a stepped wedge cluster randomised trial of patients who underwent any of 13 types of common surgical procedures across 29 community hospitals in the Southeastern United States. We divided the 13 procedures into six clusters; a cluster of procedures at a single hospital was the unit of randomisation and analysis. In total, 105 clusters were randomised to 12 groups of 8-10 clusters. All participating clusters began the trial in a 12-month baseline period of control or "traditional" SSI surveillance, including prospective analysis of SSI rates and consultative support for SSI outbreaks and investigations. Thereafter, a group of clusters transitioned from control to intervention surveillance every three months until all clusters received the intervention. Electronic randomisation by the study statistician determined the sequence by which clusters crossed over from control to intervention surveillance. The intervention was the addition of weekly application of optimised SPC methods and feedback to existing traditional SSI surveillance methods. Epidemiologists were blinded to hospital identity and randomisation status while adjudicating SPC signals of increased SSI rates, but blinding was not possible during SSI investigations. The primary outcome was the overall SSI prevalence rate (PR=SSIs/100 procedures), evaluated via generalised estimating equations with a Poisson regression model. Secondary outcomes compared traditional and optimised SPC signals that identified SSI rate increases, including the number of formal SSI investigations generated and deficiencies identified in best practices for SSI prevention. This trial was registered at ClinicalTrials.gov, NCT03075813.FindingsBetween Mar 1, 2016, and Feb 29, 2020, 204,233 unique patients underwent 237,704 surgical procedures. 148,365 procedures received traditional SSI surveillance and feedback alone, and 89,339 procedures additionally received the intervention of optimised SPC surveillance. The primary outcome of SSI was assessed for all procedures performed within participating clusters. SSIs occurred after 1171 procedures assigned control surveillance (prevalence rate [PR] 0.79 per 100 procedures), compared to 781 procedures that received the intervention (PR 0·87 per 100 procedures; model-based PR ratio 1.10, 95% CI 0.94-1.30, p=0.25). Traditional surveillance generated 24 formal SSI investigations that identified 120 SSIs with deficiencies in two or more perioperative best practices for SSI prevention. In comparison, optimised SPC surveillance generated 74 formal investigations that identified 458 SSIs with multiple best practice deficiencies.InterpretationThe addition of optimised SPC methods and feedback to traditional methods for SSI surveillance led to greater detection of important SSI rate increases and best practice deficiencies but did not decrease SSI rates. Additional research is needed to determine how to best utilise SPC methods and feedback to improve adherence to SSI quality measures and prevent SSIs.FundingAgency for Healthcare Research and Quality.

Shewhart X ¯ control charts commonly used for monitoring the mean of a process may be inaccurate or perform poorly when the subgroup size is small or the distribution of the process variable is skewed. Truncated saddlepoint distributions can increase the accuracy of estimated control limits by including higher order moments/cumulants in their approximation, yet this distribution may not exist in the lower tail, and thus the lower control limit may not exist. We introduce a novel modification in which some usually truncated higher-order cumulants are re-expressed as functions of lower-order cumulants estimated from data in a manner that ensures the existence of the truncated saddlepoint distribution over the complete domain of the random variable. The accuracy of this approach is tested in cases where the cumulants are assumed either known or estimated from sample data, and demonstrated in a healthcare application. [ABSTRACT FROM AUTHOR]

Clusters of nosocomial infection often occur undetected, at substantial cost to the medical system and individual patients. We evaluated binary cumulative sum (CUSUM) and moving average (MA) control charts for [...]