Figure

Figure. The timing and nature of data collection. *Interview questions available from author on request. †Types of data collected from patient hospital records available from author on request. ‡Variables collected from patient...

We recruited, in consecutive order, adults (>18 years of age) admitted to 3 Australian hospitals in 2004 for knee or hip prostheses, cardiovascular procedures, femoropopliteal bypass grafts, or abdominal procedures, including abdominal hysterectomies and lower segment caesarean sections. Four infection-control research nurses recruited participants and collected data during the hospital admission process and on 4 separate occasions after surgery by visiting the patients in their homes (data collection is illustrated in the Figure). Monetary estimates of all costs were made by multiplying frequency with a cost vector for the item of service (9–12). Production losses were estimated by comparing the presurgery level of (unwaged and waged) productive activity with the actual level of (unwaged and waged) productive activity achieved during the 4 weeks postdischarge. These losses were converted to a monetary value by using market prices for labor, approximated by average pretax earnings (13).

The question we address is whether postdischarge SSIs independently affect costs. The specific cost outcomes we seek to explain are listed in Table A1. Adjustment was made for other factors believed to influence these cost outcomes (i.e., confounding factors); these included the type of surgical procedure, duration of surgery, American Society of Anesthesiologists score, wound class, number of coexisting conditions, length of hospital stay, whether patient was funded by the public sector or private insurance, admitting hospital, sex, age, ethnicity, patient's socioeconomic status (14), whether the patient was in waged employment, salary level and health-related quality of life as measured by the SF-12v2 Health Survey (available from http://www.sf-36.org/tools/sf12.shtml) scores at baseline and 4 weeks postdischarge. The complete set of explanatory variables available for analyses and the summary statistics are presented in Table 1 and Table A2. Because the outcome variables were continuous and linear, ordinary least squares regression was chosen to model the independent effect of SSI on cost outcomes (Table 2). See Table A3 for a description of the statistical analyses.

The mean age of the 449 patients included in the analyses was 63.65 years (SD 14.34), and 50.56% were women. The mean length of hospital stay for the sample was 7.8 days (SD 8.68, median 6 days, interquartile range 4–8). Thirty-eight of the 449 patients included in the study had a diagnosis of SSI postdischarge, which indicates an incidence of 8.46% for the 8-month period during which patients were recruited. A higher proportion of persons with SSI (18.24%) compared to those without SSI (2.43%) were readmitted to the hospital, but the mean lengths of stay of the readmitted persons were similar, 16.57 days versus 15.72 days, respectively. Summary statistics for all variables are included in Table 1 and in Table A2, and the ICD-10 procedures for the 38 cases of SSI are described in Table A4.

No evidence was found of multicollinearity or interactions between variables. However, none of the outcome variables were normally distributed, and variance of the error term was not constant (i.e., heteroscedastic), so all models were estimated by using the Huber–White covariance matrix (15). Results of the ordinary least squares regressions are summarized in Table A1. Strong statistical evidence shows that postdischarge SSI independently causes the following: 1.36 extra contacts with community-based services with increased costs of $47.78; 6.46 days of additional antimicrobial drug therapy with increased costs of $14.44; and an increase in total health service costs of AU $74 (US $57) when the costs of readmission to the hospital are excluded and AU $123 (US $94) when the costs of readmission to the hospital are included. The strength of the relationship between SSI and all other cost outcomes was not significant with the 95% confidence interval crossing zero for all other models.

These results support the view that most SSIs first appear after discharge from hospital, but we did not find any evidence that postdischarge SSI causes substantial economic costs even when costs are viewed from a societal perspective. These findings contradict Perencevich et al. (5), who found the economic cost of a case of SSI diagnosed after discharge was almost 50-fold the estimate we report here. Thus, what might explain this extreme discrepancy in attributed costs? The study designs and research methods differed. Compared to Perencevich et al. (5), we used more control variables (described in Table 1 and Table A2 and listed below the table in Table A1). Might this extended set of control variables reduce bias from omitted variables and so reduce the cost attributed to SSI? Another factor might be the surveillance method. Perencevich et al. (5) used automated record screening that relied on accurate documentation of diagnostic, testing, or treatment codes and pharmacy records. This process resulted in 89 diagnoses among 4,571 patients, an incidence rate of 1.9%. For our study, patients were recruited before surgery and infection-control research-nurses visited the patients in their homes on 4 occasions after discharge, during which time the wound was examined and the definition of the Centers for Disease Control and Prevention definition, modified by the Australian Infection Control Association Inc., was applied (8). This method yielded a much higher infection rate of 8.38%. One interpretation is that the surveillance method used by Perencevich et al. was not sensitive to all cases of postdischarge SSI. Instead, only those that generated certain data items in the downstream electronic records were flagged, and these may have been the most serious cases of SSI that generated the greatest costs. This theory might be supported by the higher rate of readmission among the patients with cases of SSI in the Perencevich data (34%) compared to the rate in our study (18%).

Of course, other factors may have an influence, such as the case mix and socioeconomic characteristics of the participants, the costs of the inputs to healthcare services (i.e., salaries for doctors and nurses), consumer preferences (i.e., for more or less postdischarge care), and predefined care protocols.

Also, our data only describe a 4-week period after surgery and not the 8-week period considered by Perencevich et al. (5). We recommend that readers interpret our results carefully but nevertheless suggest that the economic costs of SSIs that occur after hospital discharge are real but not substantial.