Effect of Healthcare-Acquired Infection on Length of Hospital Stay and Cost

Pages 12
Views 10

Please download to get full document.

View again

of 12
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
infection control and hospital epidemiology march 2007, vol. 28, no. 3 original article Effect of Healthcare-Acquired Infection on Length of Hospital Stay and Cost Nicholas Graves, PhD; Diana Weinhold,
infection control and hospital epidemiology march 2007, vol. 28, no. 3 original article Effect of Healthcare-Acquired Infection on Length of Hospital Stay and Cost Nicholas Graves, PhD; Diana Weinhold, PhD; Edward Tong, BSc (Hons); Frances Birrell, M App Epi; Shane Doidge, G Dip PH; Prabha Ramritu, MPH; Kate Halton, MSc; David Lairson, PhD; Michael Whitby, MPH objective. To estimate the independent effect of a single lower respiratory tract infection, urinary tract infection, or other healthcareacquired infection on length-of-stay and variable costs and to demonstrate the bias from omitted variables that is present in previous estimates. design. Prospective cohort study. setting. A tertiary care referral hospital and regional district hospital in southeast Queensland, Australia. patients. Adults aged 18 years or older with a minimum inpatient stay of 1 night who were admitted to selected clinical specialities. results. Urinary tract infection was not associated with an increase in length of hospital stay or variable costs. Lower respiratory tract infection was associated with an increase of 2.58 days in the hospital and variable costs of AU$24, whereas other types of infection were associated with an increased length of stay of 2.61 days but not with variable costs. Many other factors were found to be associated with increased length of stay and variable costs alongside healthcare-acquired infection. The exclusion of these variables caused a positive bias in the estimates of the costs of healthcare-acquired infection. conclusions. The existing literature may overstate the costs of healthcare-acquired infection because of bias, and the existing estimates of excess costs may not make intuitive sense to clinicians and policy makers. Accurate estimates of the costs of healthcare-acquired infection should be made and used in appropriately designed decision-analytic economic models (ie, cost-effectiveness models) that will make valid and believable predictions of the economic value of increased infection control. Infect Control Hosp Epidemiol 2007; 28: Healthcare-acquired infections (HAIs) are thought to generate substantial economic burdens. Patient morbidity and mortality risks are increased, hospital stay is prolonged, and additional cash costs arise for consumable items used to treat the infection. 1-3 The risk of acquiring an infection is also believed to increase with increased length of hospital stay. 4,5 A recent review suggested that 10%-70% of HAIs are preventable 6 with appropriate infection control. The decision to invest in additional infection control programs should be informed by the expected changes to both cost and health outcomes, and only efficient (ie, cost-effective) strategies should be used Cost outcomes will change because of increased expenditure on infection control and the cost savings of avoided cases of HAI. Health outcomes will change (improve) because excess morbidity and mortality risks are reduced. Although economic arguments for additional infection control programs have been made, few data on the costs and health benefits of these programs have been published. 11 These data are vital for decision makers who face an increasing pressure to tackle the problem of HAI from politicians and from journalists who regularly inform the public about the dangers of HAIs, especially those caused by methicillinresistant Staphylococcus aureus. Research-based models that describe the economics of additional infection control programs therefore rely on valid estimates of the independent effect of HAI on length of hospital stay and cost, but it is difficult to make bias-free estimates. An unadjusted comparison of the cost outcomes for patients with HAI and for those without HAI is not useful because of other differences, unrelated to HAI, between the two groups. For example, those with HAI might have more comorbid conditions and so might generate quite different (greater) cost outcomes regardless of the type of HAI. The challenge is to tease out the independent effect of HAI on From the Centre for Healthcare Related Infection Surveillance and Prevention, Princess Alexandra Hospital (N.G., E.T., F.B., S.D., P.R., K.H., M.W.), and the Institute of Biomedical and Health Innovation, School of Public Health, Queensland University of Technology (N.G., K.H.), Brisbane, Australia; the London School of Economics, London, United Kingdom (D.W.); and the School of Public Health, University of Texas Health Science Center at Houston, Houston (D.L.). Received November 28, 2005; accepted June 27, 2006; electronically published February 20, by The Society for Healthcare Epidemiology of America. All rights reserved X/2007/ $ DOI: /512642 healthcare-acquired infection and cost 281 cost outcomes by making allowances for all observable confounders. Haley 1 and Graves and Weinhold 12 review the existing methods of direct attribution and comparative attribution. Direct attribution requires an expert reviewer to assess the extra cost from HAI. This method has been criticized as being subjective and not reproducible, 13 and comparative attribution studies have been preferred by the research community. Researchers undertaking comparative attribution studies use data collected from a cohort of hospitalized patients and either (1) select a subset of infected patients who are then matched with uninfected controls for variables thought likely to affect cost outcomes (eg, age, sex, and comorbidities) or (2) build multivariable statistical regression models that describe the relationship between HAI and cost outcomes, while controlling for other factors thought likely to affect cost outcomes. 14 The disadvantage of matching is that infected patients can only be matched to uninfected controls for a limited number of variables. Matching more than, say, 5 or 6 variables requires a substantial increase in the size of the pool of controls, which makes the research process costly and inefficient. Matching too few variables might cause bias from omitted variables because important factors that explain the variation in cost outcomes are excluded. The source of the bias from omitted variables is described in Appendix A. The consequence of this bias is that the cost attributed to HAI is either overstated or understated, although our belief before performing the analysis was that the bias will be positive (ie, that the cost of HAI is overstated). 15 If case patients are subsequently excluded from the study to match more variables (ie, to mitigate bias from omitted variables), then a selection bias arises because not all case patients have the same opportunity to be included in the comparison of cost outcomes. Those undertaking matched cohort studies are forced to trade bias from omitted variables with bias from the selection of individuals and cannot escape both sources of bias at the same time. The use of statistical regression analysis for a cohort of patients can avoid selection bias completely and presents an opportunity to reduce bias from omitted variables. A correctly specified statistical regression model will summarize the association between the outcome variable (ie, length of stay or cost) and the independent variables (ie, HAI and other observable factors that might explain variation in outcomes). A statistical regression model might take the form of this equation: LOS p f(hai,other controls,u), where LOS is the outcome we wish to explain (length of stay in the hospital), HAI is occurrence of an infection (1 p infection and 0 p no infection) and other controls represents all other factors (eg, age, sex, primary diagnosis, comorbidities, and underlying state of health) that we believe are associated with the variation in length of stay across the sample. The term u represents the residuals or error terms that capture the residual variation in length of stay not explained by the independent variables (ie, HAI and other controls ). An important requirement for regression analysis is that the residuals are evenly distributed around the fitted regression line (ie, error terms are homoskedastic), and the smaller the residuals, the better the model. A model like this will completely avoid selection bias, because every individual from the cohort can be included in the analysis, and it will reduce bias from omitted variables, because many independent variables can be included within the set called other controls. It is better to control for confounding in the analysis stage (regression analyses) than in the design stage (matching). Another source of bias arises from the relationship between the variables HAI and LOS. Although we know that HAI increases the length of stay, there is good evidence 4,5,16 that length of stay also increases the risk of HAI. This reverse causality, or feedback ( LoS p HAI and HAI p LoS), induces a correlation between the error terms and the independent variables, leading to biased estimates and tests of hypotheses. 17 This problem is called endogenous variables bias and has been discussed in the context of HAI. 3,18 Graves and Weinhold 12,19 describe the problem in detail and report preliminary attempts at a solution, using an instrumental variables method. Controlling bias from endogenous variables and interpreting the results of an unbiased model is a methodological challenge for future research. The impact of different types of HAI on cost outcomes has been studied since the 1950s, when Clarke et al. 20 investigated whether S. aureus in surgical wounds extended the length of stay of patients admitted to an English hospital. Since then, studies using different methods of estimating costs have produced quite different results. Direct attribution methods have reported 3.7, 0.6, and 5.7 days of extra hospitalization due to lower respiratory tract infection (LRTI), urinary tract infection (UTI), and HAIs at other sites (hereafter, other HAIs ), respectively, 2 and an extra cost of US $589 for a case of UTI. 21 Studies with a matched-cohort design have produced larger estimates, with 10 studies of LRTI reporting an increase in length of stay between 5.33 and days, with a median increase of 9.35 days; 11 studies of UTI revealed a median increase of 3.6 days (range, days) 24,25,28,31-36 ; and 2 studies of other infections reported increases of 0 days 24 and 2.5 and 7 days. 2 One study 37 that used statistical regression methods attributed 8.4, 5.1, and 12.4 extra days of hospital stay to the occurrence of LRTI, UTI, and other infections, respectively, whereas a second study 38 reported an unadjusted difference of 25 days in the hospital for each LRTI but used statistical regression to estimate an adjusted additional cost of US $11,897 per LRTI. The aim of the research reported here is to use statistical regression to estimate the independent effect of HAI on length-of-stay and cost outcomes in a cohort of hospitalized patients. The method will avoid selection bias completely and will address bias from omitted variables by including a com- 282 infection control and hospital epidemiology march 2007, vol. 28, no. 3 prehensive set of explanatory terms in a statistical regression model. We also aim to demonstrate that omitting important variables from the statistical model causes bias in the costs attributed to HAI. We conduct separate analyses for patients who received a diagnosis of LRTI, UTI, or other HAI. We do not, however, attempt to address the bias from endogenous variables. methods Study Sites and Participants We recruited participants from a 712-bed tertiary care referral hospital and a 312-bed district hospital in southeast Queensland, Australia. Inclusion criteria were an age of 18 years or older and a minimum inpatient stay of 1 night for the clinical specialities listed in Table B1 in Appendix B. Patients were identified from a routinely generated list of all admissions, and patients with consecutive admissions were recruited from both hospitals between October 13, 2002, and January 16, 2003, by 5 registered research nurses who worked at the tertiary referral hospital and were seconded to collect the data from both hospitals. Data Collection All variables for data collection were selected on the basis of previous experience, expert opinion, and a review of the literature (the literature reviewed is listed in a Supplemental Reading List that is available in the online edition of the journal). Data collection tools were developed and then were tested during a 10-week pilot study, and criteria for selecting the values for all variables were established. This process involved the research nurses, a senior infectious diseases physician, the project coordinator, and an epidemiologist with a background in acute hospital services and infection control. The result was an extensive data dictionary that summarized definitions agreed to by the research team. This document was the reference for any decision to assign a value to a variable and is available from the authors on request. Demographic data were collected directly from the bedside by use of personal digital assistants that linked to a customdesigned Access database (Microsoft). After recruitment, data collection was completed by a review of the patient s medical record, the hospital-based corporate information systems, and the hospital pathology system, Auslab. 39 Variables were collected that described the demographic characteristics of the sample, including the primary or most recent occupation and education level, and a score of socioeconomic status was derived via an algorithm described by Jones and McMillan. 40 The length of hospital stay for each patient admitted was calculated, and data were collected that described the consumable items used by the patient. Market prices were then applied to estimate the variable costs incurred during the hospital stay. The type of admission (ie, elective, emergency, or transfer), admission to a clinical unit, and all diagnosis codes from the International Classification of Diseases and Related Health Problems, 10th Revision, Australian Modification 41 were recorded. Any adverse events that occurred during the hospital stay that might extend the stay were recorded, as were all observable risk factors for HAI. Data cleaning was undertaken alongside data collection by a separate researcher. The data cleaner and research nurses worked together to ensure consistent application of definitions. All cases of HAI were diagnosed using the Centers for Disease Control and Prevention definitions modified for Australia 42 ; when ambiguity existed, the senior infectious diseases physician was consulted. The medical records of all patients with HAI and a length of stay greater than 40 days were reviewed to identify when, during the hospital stay, the infection was diagnosed, was treated, and had cleared. Statistical Methods We used the generalized linear modeling (GLM) approach, which employs maximum-likelihood estimation to summarize the relationship between HAI and cost outcomes. Model selection was based on testing whether the distributional assumptions fit the data and which approach yielded the best residuals; model selection is described in Appendix C. Six models were specified to describe the relationship between the 3 types of infection (UTI, LRTI, and other HAIs) and the 2 outcome variables (length of stay and variable costs). To model length-of-stay outcomes, we chose a gamma distribution to characterize the outcome and a log link function to specify the relationship with the explanatory variables. To model variable cost, we chose a gamma distribution with a square-root link function. The gamma distribution is similar in shape to the log-normal distribution 45 and is robust to the nonnormal distributions typical of length-of-stay and cost outcomes All coefficients were retransformed back to the original units of length of stay (days) and variable costs (AU$). We began with general models that included all available variables as explanatory terms, and variables were excluded because of multicollinearity, as assessed by nested auxiliary regression, where each variable in turn was dropped from the model and the R 2 values were compared with those of a complete model. We then sought a parsimonious specification for each model by further reducing the model, using backward stepwise regression with a 5% threshold for statistical significance. SEs were made heteroskedasticity consistent via the Huber-White covariance matrix, which was applied in all estimation procedures. 48 StataTM (Stata) was used for all analyses. We undertook further analyses of the parsimonious models that described length-of-stay outcomes, to demonstrate the bias from omitted variables. We removed variables from each model, one by one, in no particular order, and compared the parsimonious model with each restricted model. The likelihood ratio (LR) test was used to assess the goodness of fit between the 2 models and to make an inference about whether the parsimonious model represented a better spec- healthcare-acquired infection and cost 283 ification of the relationship between HAI and the length-ofstay outcome than did the artificially restricted models. results Overview of the Data A total of 4,488 admissions were included in the study; 2,971 were admissions to the university teaching hospital, 1,640 were admissions for a surgical procedure, and 2,848 were admissions for nonsurgical specialties. The mean age of patients was 58 years (range, years), and 51% were male. There were 228 cases of HAI diagnosed, giving an overall incidence rate of 5.08% for the 95 days during which patients were recruited. This included 37 LRTIs (incidence rate, 1.76%), 79 UTIs (incidence rate, 0.82%), and 49 other HAIs (6 in the digestive system, 2 in the ear, 6 in the mouth and/ or esophagus, 1 in pleural fluid, 10 at an intravenous catheter insertion site\, 18 involving skin, and 6 at an unknown site) (incidence rate, 1.09%). The remaining 63 cases of infection were excluded from the analyses and comprised surgical site infections, bloodstream infections, and multiple infections. Patients with HAI and Length of Stay Greater Than 40 Days Of the 4,488 patients, 98 (2.2%) had a length of stay greater than 40 days. Of these 98 patients, 10 received a diagnosis of LRTI, 20 received a diagnosis of UTI, and 8 received a diagnosis of other HAIs. For all but 1 of the patients with LRTI and 2 of the patients with UTI, the HAI cleared early in the stay (ie, within 12 days after admission), and other factors were found that caused the long stay in hospital. In particular, most of these patients were transferred to a rehabilitation facility within the hospital to wait for placement in a long-term residential care center; other reasons for long stay were community-acquired bacteremia, serious medical conditions, hypoxic brain injury, prolonged stay in the intensive care unit, and slow recovery after surgery. Because the extended hospital stay (140 days) was unrelated to the episode of infection, those patients with a length of stay 140 days were excluded from the data set before the analyses were done. Unadjusted Comparison of Patients With HAI and Patients Without Patients without HAI were compared with patients with LRTI, UTI, and other infections for selected variables, and the results are presented in Table 1. A list of all variables available for analysis is included in Table B2 in Appendix B. Compared with patients without HAI, patients with HAI were older, stayed in the hospital longer, and incurred higher variable costs. Fewer of them were discharged home, had elective admissions, or were self-caring before hospital admission. A higher proportion died in the hospital, experienced an adverse event (eg, a fall, cardiac arrest, pressure ulcer, or gastrointestinal bleeding), presented with comorbidities (eg, anemia, chronic obstructive pulmonary disease, congestive heart failure, diabetes, o
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!