Advanced Data Analytics for Improved DecisionMaking at a Veterans Affairs Medical Center Downloaded from https://journals.lww.com/jhmonline by IhLYIClvaT9x3iiejbNueDP3RPWmWo+9k2n/Re5JJZn9uG1OYfZx1DrFJW95FukutFaMikJr3CzCTd85Wn/gTkBCXOiSv2Ew+6WNseVZHsYe1AExJQF23K6aotar6Rf/ on 02/18/2019 Ajay Mahajan, PhD, professor, mechanical and biomedical engineering, University of Akron, Akron, Ohio; Parag Madhani, MD, chief of cardiology and pulmonology, Marion VA Medical Center, Marion, Illinois; Sanjeevi Chitikeshi, PhD, assistant professor, electrical technology, Old Dominion University, Norfolk, Virginia; Padmini Selvaganesan, College of Engineering (biomedical), University of Akron; Alex Russell, College of Engineering (mechanical), University of Akron; and Preeti Mahajan, managing partner, Clipius Analytics, North Canton, Ohio EXECUTIVE SUMMARY This article reports on a data-driven methodology for decision-making at a Veterans Affairs medical center (VAMC) to improve patient outcomes, specifically the 30-day standardized mortality ratio (SMR30). The quarterly strategic analytics for improvement and learning (SAIL) reports are used to visualize the data, study trends, provide actionable recommendations, and identify potential consequences. A case study using more than 4 years of data demonstrates the power of the methodology. After reviewing data and studying trends at other VAMCs, a decision is made to reduce the SMR30 value by 1%. In running correlation algorithms, in-hospital complications (IHC) are shown to be most closely correlated with SMR30. Modeling the results from 17 quarters’ worth of data shows that a desired 1% change in SMR30 would require a targeted 18.6% decrease in IHC. This change, if accomplished, would yield good consequences on methicillinresistant Staphylococcus aureus mitigation but potentially unintended consequences with catheter-associated urinary tract infections and patient safety indicators that would need to be monitored. This knowledge could enable healthcare leaders to make informed decisions of both potentially positive and unintended consequences that can be monitored and minimized. This study lays the groundwork for a healthy discussion among leaders, staff, and clinicians on the path forward, resources required, and—most importantly—a dashboard that reflects the progress each week rather than a quarterly SAIL report. For more information about the concepts in this article, contact Padmini Selvaganesan at ps120.@zips.uakron.edu padminisg@gmail.com The authors declare no conflicts of interest. © 2019 Foundation of the American College of Healthcare Executives DOI: 10.1097/JHM-D-17-00164 54 Volume 64, Number 1 • January/February 2019 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Advanced Data Analytics for Improved Decision-Making INTRODUCTION Veterans Affairs medical centers (VAMCs) provide services to all U.S. veterans through the Veterans Health Administration. VAMCs differ significantly from private health providers, as they are in a federal system funded by the public and led by a cabinet-level secretary. It is the nation’s largest integrated healthcare system, stretching across all 50 states, the Caribbean, and the Pacific Rim. Managing this large enterprise is a challenge, and using data analytics to improve patient outcomes and reduce healthcare costs is a priority (Byers, 2015). Given the VA system’s scale, there is a critical need to design systems to evaluate healthcare performance, and to discover and improve potential inefficiencies (Caballer, Moya, Vivas, & Barrachina, 2010). The strategic analytics for improvement and learning value model (SAIL) is a system for summarizing hospital system performance in the VA system (U.S. Department of Veterans Affairs, n.d.). SAIL assesses 26 quality measures in areas such as death rate, hospital complications, and patient satisfaction, as well as overall efficiency at individual VAMCs. The data for each facility, spanning more than 6 years, are downloaded and viewed in spreadsheets. SAIL tables are updated every quarter. The 152 VAMCs are divided into four groups for comparisons based on standards for hospital complexity and intensive-care-unit level (U.S. Department of Veterans Affairs, 2014; Shulkin, 2017). SAIL is now an established internal quality improvement system, and it is one of the most robust and comprehensive systems of its kind in the healthcare industry. It was designed for comparing VA hospitals to one another and creating benchmarks to www.ache.org/journals guide continuous improvements in patient outcomes. This article takes SAIL a step further by helping VA leaders to compare metrics to those of peer VAs and identify inefficiencies in their own hospitals. 30-Day Standardized Mortality Ratio The most critical metric on the SAIL report may be the 30-day standardized mortality ratio (SMR30), which is the actual number of patients admitted to acute-care wards who died within 30 days of hospital admission, divided by the sum of the expected deaths of all acute-care-ward patients using the risk-adjusted mortality model that predicts death at 30 days (U.S. Department of Veterans Affairs, n.d.). The SMR30 included in this article is a rolling 12-month measurement. Identifying which in-hospital deaths are preventable and which are expected has never been easy (Ayaz, Sahin, Sahin, Bilir, & Rakıcı, 2014). However, the mathematical correlations between SMR30 and other important metrics can be determined, and we focus on them here. SMR30 depends on various metrics, and finding a correlation among those can provide a roadmap for lowering SMR30 by lowering targeted metrics such as preventable in-hospital complications (IHC) and healthcare-associated infections (HAI). Following are some of the metrics that can be targeted to make changes. Preventable In-Hospital Complications Preventable IHCs are conditions that arise after hospital admission. These conditions are potentially preventable if healthcare is delivered appropriately. The SAIL model tracks numerous preventable IHCs, which are identified using secondary diagnoses following algorithms published in the 55 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Journal of Healthcare Management literature (Needleman, Buerhaus, Mattke, Stewart, & Zelevinsky, 2002). These include surgical conditions (e.g., wound infection) and medical conditions (e.g., hospitalacquired pneumonia, shock or cardiac arrest, upper gastrointestinal bleeding, hospital-acquired sepsis, deep vein thrombosis, central nervous system complications). Healthcare-Associated Infections Healthcare-associated infections (HAIs) are caused by a wide range of common and unusual bacteria, fungi, and viruses during the course of medical care, according to the Centers for Disease Control and Prevention (CDC, n.d.). HAIs are associated with significant mortality and morbidities (Al-Tawfiq and Tambyah, 2014). The CDC defines HAIs as infections acquired while in the healthcare setting (e.g., inpatient hospital admission, hemodialysis unit, same-day surgery), with a lack of evidence that the infection was present or incubating at the time of entry into the healthcare setting (Horan, Andrus, & Dudeck, 2008). Infection brings increases in prolonged hospital stay, long-term disability, antimicrobial resistance, socioeconomic disturbance, and mortality rate (Khan, Baig, & Mehboob, 2017). There are many types of HAI metrics; the SAIL model uses the following: 1. Catheter-associated urinary tract infection (CAUTI). This is the most common nosocomial infection. It accounts for up to 40% of infections reported by acute-care hospitals (Klevens et al., 2007). Up to 80% of urinary tract infections are associated with the presence of an indwelling catheter (Apisarnthanarak et al., 2007). 56 2. Central line–associated bloodstream infections (CLABIs). These are deadly HAIs, with reported mortality rates of 12%–25% (CDC, 2002). The CDC defines a CLABI as recovery of a pathogen from a blood culture (a single blood culture for organisms not commonly present on the skin and two or more blood cultures for organisms commonly present on the skin) in a patient who had a central line at infection or within the 48-hour period before development of infection. 3. Methicillin-resistant Staphylococcus aureus (MRSA). This is a nosocomial pathogen. Established risk factors include recent hospitalization or surgery, residence in a long-term-care facility, dialysis, and indwelling percutaneous medical devices and catheters (Naimi et al., 2003). Patient Safety Indicators Patient safety indicators (PSIs) have been developed by the Agency for Healthcare Research and Quality (2015). These metrics are widely used to reflect quality of care inside hospitals, as well as across geographic areas, to focus on potentially avoidable complications and iatrogenic events. Previous research shows that PSIs may be useful for patient screening in VA facilities (Rosen et al., 2005). The SAIL values model includes 10 PSIs to derive an overall index value. For each indicator, the number of actual events is divided by the expected number of events, given the risk of having an event for each patient. These ratios are then normalized and averaged to derive an index score (U.S. Department Volume 64, Number 1 • January/February 2019 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Advanced Data Analytics for Improved Decision-Making of Veterans Affairs, 2014). The PSIs in the SAIL values model include pressure ulcer; death among surgical inpatients with serious, treatable complications; iatrogenic pneumothorax; selected infections due to medical care; perioperative hemorrhage or hematoma; postoperative physiologic and metabolic derangement; postoperative respiratory failure; perioperative pulmonary embolism or deep vein thrombosis; postoperative sepsis; and postoperative wound dehiscence. METHODS Our objective is to present data from the SAIL report so they are intuitive and actionable. The data are extracted and presented in a bar graph to make it easy to see which metrics need immediate attention. The visualization tool plots the VA’s ranking for each metric at one VAMC as compared to all other VAMCs, and then normalizes the data so users do not have to worry about which metric needs to go up or down (based on the arrows in the SAIL report). Figure 1 shows a plot from the visualization tool that provides a gray tone for those in the top 50%, white for those between 50% and 10%, and black for those in the lowest 10% in the country (and therefore in need of immediate action). It can be seen that this VAMC is doing very well in most metrics but has serious issues with a few. From here, it is easy to identify the most critical metrics, so we picked the most important—V2 → SMR30—to see how this VAMC is doing relative to other VAMCs (Figure 2). We consider the variables listed on the next page as the primary influencers of SMR30, and a mathematical model (Figure 2) shows how the system would work. The model performs a series of steps and determines which variables are highly correlated with SMR30 and can affect it the most. FIGURE 1 Plot From the Visualization Tool Note. SMR30 = 30-day standardized mortality ratio. www.ache.org/journals 57 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Journal of Healthcare Management FIGURE 2 Modeling Graph Note. CAUTI = catheter-associated urinary tract infection; CLABI = central line–associated bloodstream infection; IHC = in-hospital complications; MRSA = methicillin-resistant Staphylococcus aureus; PSI = patient safety index; SMR30 = 30-day standardized mortality ratio. 1. IHC 2. HAI a. CAUTI b. CLABI c. MRSA infection 3. PSI With data from 17 quarters for all VAMCs, we can look for these correlations. In this study, we review the 17 quarters of SAIL data for one VAMC. The objective is to develop a methodology to use data analytics to find insight from the data; if we can reach actionable conclusions on which metric to target for a particular VAMC, we can easily scale the study to any VAMC in the country. It must also be noted that some VAMCs do not have data for all the metrics in all quarters, so any approach must consider missing data. Here, we use the correlation coefficient, which is a measure that determines the degree to which two variables are associated. The range of values for the correlation coefficient is −1.0 to 1.0, and it cannot be greater than 1.0 or less than −1.0. A 58 Volume 64, Number 1 • January/February 2019 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Advanced Data Analytics for Improved Decision-Making correlation of −1.0 indicates a perfect negative correlation, while a correlation of 1.0 indicates a perfect positive correlation. Any correlation coefficient above 0.5 shows high correlation between the two metrics. The most common calculation is the Pearson product-moment correlation, which is determined using the following equation: rxy = ∑ ∑ n i =1 n i =1 ( x i − x )( yi − y ) ( x i − x )2 ∑ i =1 ( yi − y )2 n ∑ is sigma, the symbol of “sum up”; (xi− x¯) is the difference between each x value and the mean of x; and ( y i− y¯) is the difference between each x value and the mean of y. To show the relationship between various metrics, we perform regression analysis to model the relationship between two or more variables, with the response variable using the data available, and provide a linear equation. A multiple regression model with k predictor variables X1, X2, X3 ... Xk and a response Y is written as: Y = β0 + β1 X1 + β 2 X 2 +…+ β k X k + ε Y is the response variable, X1 to Xk are predictor variables, β0, β1, β2, ... βk are the regression coefficients, and ε refers to the residual terms of the multiple linear regression model (Bremer, 2012). RESULTS The correlation values between SMR30 and the selected metrics at one VAMC are shown in Table 1. As can be seen, SMR30 is most correlated with IHC (correlation coefficient of 0.667). This is a very reliable outcome, as we have data for all quarters. There is some positive correlation with HAI-MRSA, and we disregard all negative values because they are clinically not viable. We then create a model from IHC-SMR30, IHC, and other metrics (CAUTI, CLABI, MRSA) to see how changes to IHC affect other metrics, if we select IHC as the major metric to change to improve SMR30. The next step is to create a statistical model. This by far is the most significant contribution of this analysis—that is, the ability to model the impact of change in any metric on other metrics. The Minitab statistical tool is used to perform the stepwise regression on the data (Table 2). In stepwise regression analysis, the response metric is SMR30 and the continuous predictors are the other metrics such as IHC, CAUTI, MRSA, and PSI. The Minitab results show that stepwise regression has been performed on the response SMR30 and the four predictors IHC, CAUTI, MRSA, and PSI. The alpha values to enter and remove have been set at 0.05 to yield results at the conventional confidence level. The regression analysis results in retaining IHC and removing TABLE 1 Correlation Values Between SMR30 and Selected Metrics at One VAMC IHC 0.667 HAI-CAUTI −0.162 www.ache.org/journals HAI-CLABI Missing data HAI-MRSA 0.192 PSI −0.065 59 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Journal of Healthcare Management TABLE 2 Stepwise Selection of Terms Constant Step 1 Coefficient 0.973 IHC p 0.2471 .013 Note. Candidate terms: IHC, CAUTI, MRSA, PSI. α to enter = 0.05, α to remove = 0.05. TABLE 3 Forward Selection of Terms Constant IHC Step 1 Coefficient p 0.973 0.2471 .013 using the alpha value of 0.05 obtain the results shown in Tables 3 and 4. The Minitab results obtained from both forward selection and backward elimination of terms are the same as stepwise selection results. In both cases, the final step retained only IHC (p value < .05); other predictors can be eliminated based on the p value of each predictor (p value > .05). This shows that SMR30 is highly correlated to IHC and that the regression results are statistically significant. The consequences of making this change in IHC on other critical metrics such as MRSA, CAUTI, and PSI are shown in simple linear regression and described as follows: Note. Candidate terms: IHC, CAUTI, MRSA, PSI. α to enter = 0.05. other metrics. The regression equation obtained is shown here: SMR30 = 0.973 + 0.2471 IHC This confirms that the metric IHC is statistically significant with a p value of .013 (< .05) and is entered in the model, whereas the other was excluded. With the same data, both forward selection and backward elimination regression analysis CAUTI = 1.918 − 0.519 IHC MRSA = 0.1039 + 0.0639 IHC PSI = 0.767 − 0.108 IHC Figure 2 plots the four models. The x-axis shows values for IHC ranging from 0.0 to 2.0. The solid black line is the plot for SMR30, and the desired direction for it is shown (high to low). This is the driver for all changes. ILLUSTRATION Following is an illustration of what happens when SMR30 value is reduced by 1%. TABLE 4 Backward Elimination of Terms Step 1 Coefficient Constant IHC CAUTI MRSA PSI 0.905 0.270 0.0106 −0.141 0.081 p .044 .863 .820 .707 Step 2 Coefficient p Step 3 Coefficient p 0.937 0.265 .029 0.934 0.2526 .017 −0.158 0.068 .784 .720 0.051 .766 Step 4 Coefficient 0.973 0.2471 p .013 Note. Candidate terms: IHC, CAUTI, MRSA, and PSI. α to remove = 0.05. 60 Volume 64, Number 1 • January/February 2019 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Advanced Data Analytics for Improved Decision-Making Current SMR30 value: 1.207 Corresponding IHC value: 1.103 (vertical line on plot labeled “current”) New desired SMR30 value: 1.207 * 0.99 = 1.1949 Targeted IHC value from model: 0.898 (decrease of 18.57% from 1.103) Consequence on MRSA: 0.161 (decrease of 7.507% from 0.174) Consequence on CAUTI: 1.452 (increase of 7.902% from 1.351) Consequence on PSI: 0.67 (increase of 3.415% from 0.649) As can be seen by the results, a desired 1% change in SMR30 would require a targeted 18.57% decrease in IHC. If this change is accomplished, it would come with good consequences on MRSA, but may have unintended consequences of slightly increasing CAUTI and PSI. This does not mean that this will happen; rather, it simply points to the need to monitor these metrics closely. CONCLUSION This article reports on a data-driven methodology for decision-making at one VAMC to improve patient outcomes, specifically the SMR30. A case study using more than 4 years of data demonstrates the power of the methodology. The article outlines a roadmap for decreasing SMR30 by 1% and the possible consequences that should be monitored closely. This knowledge can allow healthcare leaders to make informed decisions on where to start making changes and how to explore the consequences of potential actions. The intent is not to manipulate the data and show all positive consequences, but instead to show possible consequences in trying to change www.ache.org/journals some metrics so that healthcare leaders can prepare for the consequences. With this knowledge, they can make changes, monitor progress weekly, and tweak efforts to achieve their desired goals. This is a proactive approach to making informed decisions rather than reacting to the SAIL report when it is released each quarter. REFERENCES Agency for Healthcare Research and Quality. (2015). Patient safety indicators [Fact sheet]. Retrieved from https://www.ahrq.gov/sites/ default/files/wysiwyg/professionals/systems/ hospital/qitoolkit/combined/a1b_combo_ psifactsheet.pdf Al-Tawfiq, J. A., & Tambyah, P. A. (2014). Healthcare associated infections (HAI) perspectives. Journal of Infection and Public Health, 7(4), 339–344. doi:10.1016/j.jiph.2014.04.003 Apisarnthanarak, A., Rutjanawech, S., Wichansawakun, S., Ratanabunjerdkul, H., Patthranitima, P., Thongphubeth, K., ... Fraser, V. J. (2007). Initial inappropriate urinary catheters use in a tertiary-care center: Incidence, risk factors, and outcomes. American Journal of Infection Control, 35(9), 594–599. doi:10.1016/j.ajic.2006.11.007 Ayaz, T., Sahin, S. B., Sahin, O. Z., Bilir, O., & Rakıcı, H. (2014). Factors affecting mortality in elderly patients hospitalized for nonmalignant reasons. Journal of Aging Research, 2014, 1–7. doi:10.1155/2014/584315 Bremer, M. (2012). Multiple linear regression [Class notes, Math 261A, Cornell University, Department of Biological Statistics and Computational Biology, Ithaca, NY]. Retrieved from http:// mezeylab.cb.bscb.cornell.edu/labmembers/ documents/supplement%205%20-%20multiple%20 regression.pdf Byers, M. (2015). Using big data to benefit veterans. FCW. Retrieved from https://fcw.com/articles/2015/01/12/comment-big-data-vha.aspx Caballer, T. M., Moya, C. I., Vivas, C. D., & Barrachina, M. I. (2010). A model to measure the efficiency of hospital performance. Mathematical and Computer Modelling, 52(7–8), 1095–1102. doi:10.1016/j.mcm.2010.03.006 61 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Journal of Healthcare Management Centers for Disease Control and Prevention. (n.d.). Healthcare-associated infections. Retrieved from http://www.cdc.gov/hai/index.html Centers for Disease Control and Prevention. (2002). Guidelines for the prevention of intravascular catheter-related infections. Morbidity and Mortality Weekly Report, 55(RR-10). Retrieved from https://www.cdc.gov/mmwr/PDF/rr/rr5110.pdf Horan, T. C., Andrus, M., & Dudeck, M. A. (2008). CDC/NHSN surveillance definition of health care-associated infection and criteria for specific types of infections in the acute care setting. American Journal of Infection Control, 36, 309–332. Khan, A. H., Baig, K. F., & Mehboob, R. (2017). Nosocomial infections: Epidemiology, prevention, control and surveillance. Asian Pacific Journal of Tropical Biomedicine, 7(5), 478–482. doi.org/10.1016/j.apjtb.2017.01.019 Klevens, R. M., Edwards, J. R., Richards, C. L., Horan, T. C., Gaynes, R. P., Pollock, D. A., & Cardo, D. M. (2007). Estimating health care-associated infections and deaths in U.S. Hospitals, 2002. Public Health Reports, 122(2), 160–166. Naimi, T. S., LeDell, K. H., Como-Sabetti, K., Borchardt, S. M., Boxrud, D. J., Etienne, J., ... & Lynfield, R. (2003). Comparison of community- and health care–associated methicillin-resistant staphylococcus aureus infection. Journal of the American Medical Association, 290(22), 2976–2984. doi:10.1001/jama.290.22.2976 Needleman, J., Buerhaus, P., Mattke, S., Stewart, M., & Zelevinsky, K. (2002). Nurse-staffing levels and the quality of care in hospitals. New England Journal of Medicine, 346(22), 1715–1722. Rosen, A., Rivard, P., Zhao, S., Loveland, S., Tsilimingras, D., Christiansen, C., ... Romano, P. (2005). Evaluating the patient safety indicators: How well do they perform on Veterans Health Administration data? Medical Care, 43(9), 873–884. Retrieved from http://www.jstor.org/stable/4640888 Shulkin, D. (2017). Ideas and opinions, understanding veteran wait times. Annals of Internal Medicine, 167(1), 52–54. U.S. Department of Veterans Affairs. (n.d.). Strategic analytics for improvement and learning (SAIL) value model measure definitions. Retrieved February 10, 2017 from https://www.va.gov/QUALITYOFCARE/ measure-up/SAIL_definitions.asp U.S. Department of Veterans Affairs (2014, November). Strategic analytics for improvement and learning (SAIL) [Fact sheet]. Retrieved from https://www.blogs.va.gov/VAntage/wp-content/ uploads/2014/11/SAILFactSheet.pdf PRACTITIONER APPLICATION: Advanced Data Analytics for Improved Decision-Making at a Veterans Affairs Medical Center Tiffany A. Love, PhD, FACHE, GNP, ANP-BC, CCA, CRLC, regional chief nursing officer, Coastal Healthcare Alliance, Rockport, Maine I n 2014, Secretary of Veterans Affairs (VA) Robert A. McDonald developed the Blueprint for Excellence (Moloney, 2014). This strategic document outlined the vision for the VA’s transformation into an exemplary national healthcare delivery system. The stated goals were to rebuild veterans’ trust and provide high-quality and The author declares no conflicts of interest. © 2019 Foundation of the American College of Healthcare Executives DOI: 10.1097/JHM-D-18-00238 62 Volume 64, Number 1 • January/February 2019 © 2019 Foundation of the American College of Healthcare Executives. All rights reserved. Reproduced with permission of copyright owner. Further reproduction prohibited without permission. Administrative Data Clinical Data example example example example example example example example Compose a one-to two-page paper in APA Style. Your paper should include a cover sheet and reference list. Submit the completed table with your written paper. You must cite and reference at least two sources. You may use your textbook as a source.
Purchase answer to see full attachment