Efficient national surveillance for health-care-associated infections

Abstract

Background: Detecting novel healthcare-associated infections (HCAI) as early as possible is an important public health priority. However, there is currently no evidence base to guide the design of efficient and reliable surveillance systems. Here we address this issue in the context of a novel pathogen spreading primarily between hospitals through the movement of patients.

Methods: Using a mathematical modelling approach we compare the current surveillance system for a HCAI that spreads primarily between hospitals due to patient movements as it is implemented in Scotland with a gold standard to determine if the current system is maximally efficient or whether it would be beneficial to alter the number and choice of hospitals in which to concentrate surveillance effort.

Results: We validated our model by demonstrating that it accurately predicts the risk of meticillin-resistant Staphylococcus aureus bacteraemia cases in Scotland. Using the 29 (out of 182) sentinel hospitals that currently contribute most of the national surveillance effort results in an average detection time of 117 days. A reduction in detection time to 87 days is possible by optimal selection of 29 hospitals. Alternatively, the same detection time (117 days) can be achieved using just 22 optimally selected hospitals. Increasing the number of sentinel hospitals to 38 (teaching and general hospitals) reduces detection time by 43 days; however decreasing the number to seven sentinel hospitals (teaching hospitals) increases detection time substantially to 268 days.

Conclusions: Our results show that the current surveillance system as it is used in Scotland is not optimal in detecting novel pathogens when compared to a gold standard. However, efficiency gains are possible by better choice of sentinel hospitals, or by increasing the number of hospitals involved in surveillance. Similar studies could be used elsewhere to inform the design and implementation of efficient national, hospital-based surveillance systems that achieve rapid detection of novel HCAIs for minimal effort.

Fig. 1

Hospitals included and their connections. Red open circles indicate a hospital and blue arrows indicate a connection (at least one patient transferred) between the two hospitals

Fig. 2

The ROC curve for the comparison between bacteraemia cases in Scotland and predictions of the stochastic network model. The x-axis shows the false positive rate and the y-axis shows the true positive rate. The AUC = 0·97

Fig. 3

a. Sentinel surveillance system performance. Average detection time of a novel HCAI, following emergence in a single randomly selected hospital versus number of hospitals participating in surveillance. The solid black line corresponds to the gold standard ‘greedy’ algorithm; the coloured points indicate the average detection time after including all hospitals in that particular list. Dashed coloured lines are plotted as reference lines. b. As 3a but showing the average number of affected hospitals

Fig. 4

Costs associated with the different prioritising methods. Total cost associated with the gold standard selection of hospitals (lines) is compared to three other selection criteria (symbols). Costs are calculated either as a fixed cost per hospital (dashed line, open symbols) or variable costs that depend on the number of patient in-movements (see main text) (solid line, closed symbols). Costs are in arbitrary units but fixed and variable costs are scaled to give the same total cost if all hospitals participate in surveillance