In 2010, AHRQ estimated the attributable inpatient cost and excess inpatient mortality for HACs. Since then, these numbers have been used to quantify progress on the PfP goals of reducing hospital-acquired conditions, most recently in the National Scorecard on Rates of Hospital-Acquired Conditions, 2010 to 2015.88 This report is an update to those estimates, primarily relying on the most current literature, which included more than 4,000 studies across 10 HACs. There are some categorical departures from the 2010 report. First, the definition of venous thromboembolism events has been expanded beyond post-operative events to include any deep vein thrombosis or pulmonary embolism occurring in hospital. Second, C. difficile infections were not in the 2010 list, but are included here.
The approach to study selection in this report favored studies using high-quality design and robust statistical techniques to control for confounding and, therefore, possibly better estimate excess or attributable mortality and costs. We focused on studies from 2000s and later, which were deemed timely, but we did not employ strict restrictions on time frame for data included, especially in cases where there were few published studies. Exhibit 9 presents the comparison between the 2010 report results, recalibrated to 2015 dollars, and the current report results are in Exhibit 9.
|Additional Cost||Excess Mortality|
|Current Study Estimate (95% CI)||2010 AHRQ Estimate*||Current Study Estimate (95% CI)||2010 AHRQ Estimate|
|ADE||$5,746 (-$3,950–$15,441)||$5,452||0.012 (0.003–0.025)||0.020|
|CAUTI||$13,793 ($5,019–$22,568)||$1,090||0.036 (0.004–0.079)||0.023|
|CLABSI||$48,108 ($27,232–$68,983)||$18,537||0.150 (0.070–0.270)||0.185|
|Falls||$6,694 (-$1,277–$14,665)||$7,888||0.050 (0.035–0.070)||0.055|
|OBAE||$602 (-$578–$1,782)||$3,271||0.005 (0.003–0.013)||0.0015|
|Pressure Ulcers||$14,506 (-$12,313–$41,326)||$18,537||0.041 (0.013–0.093)||0.072|
|SSI||$28,219 ($18,237–$38,202)||$22,898||0.026 (0.009–0.059)||0.028|
|VAP||$47,238 ($21,890–$72,587)||$22,898||0.140 (-0.110–0.730)||0.144|
|VTE||$17,367 ($11,837–$22,898)||$8,723||0.043 (0.040–0.078)||0.104|
|CDI||$17,260 ($9,341–$25,180)||N/A||0.044 (0.028–0.064)||N/A|
*Note: 2010 AHRQ estimates have been converted to 2015 dollars for comparison to the meta-analysis-based estimates.
While the estimation of additional cost per HAC remained stable for ADE, falls, and SSI, we did observe differences in estimated additional cost for other HACs: some increases (CAUTI, CLABSI, VTE, VAP) and some reduction (OBAE). The most notable increase was for CAUTI, where the prior estimate was based on a single study using cost data from two small studies conducted outside the United States and prior to 1995.89 When compared to AHRQ estimates of excess mortality, only VTE decreased when compared to 2010 AHRQ estimates. This difference in excess mortality for VTE may be related to the change in definition to include all hospital-acquired VTE—not just postoperative VTE events. In all other HACs, while we observed some increase (CAUTI) and some decrease (ADE, pressure ulcers) compared to the 2010 estimated excess mortality rates, the differences were not statistically significant at the 0.05 significance level.
When looking at these differences by HACs, we observe some interesting variations. For instance, since AHRQ last estimated cost and mortality associated with CLABSI, the mortality has decreased from more than 18 percent of cases to 15 percent. While mortality has tended to decrease (although in statistically insignificant ways), cost associated with each CLABSI case has increased substantially from $18,537 to more than $48,000. Similarly, the mortality associated with VAP remained at approximately 14 percent in both 2010 and 2017 estimates, but the additional cost associated with VAP doubled from $22,898 in 2010 to $47,238 in 2017. CAUTI also shows a similar pattern of higher costs alongside comparable excess mortality.
Several potential etiologies may account for this pattern of similar to improved mortality and increased cost. First, prevention efforts may have disproportionately eliminated the least severe and least costly infections. For example, CAUTI has been the focus of several national prevention initiatives, including the AHRQ-sponsored CUSP Stop CAUTI program.90 One strategy used by these programs has been to reduce exposure to urinary catheters by focusing on the medical need for initial placement and maintenance of a catheter throughout a hospital stay. At the same time that prevention efforts were ramping up, the CDC National Healthcare Safety Network updated their CAUTI definition, clarifying the criteria for defining a CAUTI event.91 Together these factors likely lead the remaining HAC cases to be among patients with more severe infections and, thus, likely more costly.
Second, prevention efforts may have been more effective at reducing the risk of developing these HACs early in the exposure to the related devices (e.g., central lines, urinary catheters, invasive mechanical ventilation).92,93,94 This disproportionate reduction would thus shift the overall makeup of patients with these infections toward those requiring longer lengths of stay and higher costs. Third, HAC-prevention efforts may have been particularly effective at reducing these complications in severely ill patients at a high risk for early mortality but who still died. Thus, the remaining patients are those requiring longer-term devices (e.g., central venous access for dialysis, Foley catheters, or invasive mechanical ventilation) with a lower baseline risk of mortality but longer average hospital stays and associated costs.
Since the estimate of additional cost and excess mortality are obtained from combining, via meta-analysis, individual estimates for the published literature, the quality of the estimates depends on the quality of the underlying studies. In conducting this review of the literature, several concerns about the underlying studies arose. These are detailed in subsections below.
Competing Risk and Double Counting Issues
Hospitalized patients are often suffering multimorbidity, and in many cases are also physically frail. This poses problems when estimating additional cost and mortality for a specific HAC; the issue also permeates the research literature on HACs.
Conceptually, any number of specific clinical conditions or events might result in death, with each condition considered a competing risk in comparison to the others. Essentially, death from one cause precludes death from another cause. In studies of attributable mortality for patient populations who develop a HAC, there is often not a readily available and reasonably similar comparison group of patients to assess the counterfactual of what would have happened to a similarly ill patient population in the absence of the HAC under study.
The competing risk issue is further compounded by the fact that the probability of these events is often higher among those with more clinical conditions, or more severe manifestation of any given condition. However, most studies focus on the outcomes of patients where a particular HAC is documented and compare it to those without the specific HAC, with less-than-adequate accounting for differences in latent health. Similarly, some patients with a HAC may have been more likely to die (even absent the HAC) than those in the comparison population. As such, much of the current literature tends to overstate attributable costs and mortality associated with HACs.
Relatedly, many patients with a given HAC may have other HACs. This means that the sum of deaths from studies that focus on the effects of a single HAC, and do not exclude patients with other HACs from each study, will inevitably double-count some death as being attributable to HACs more generally. In fact, the majority of underlying studies we found focus on one HAC without considering the presence of other HACs—at best, studies focus on those with only one particular HAC documented. Because of this, the sum of deaths for each individual HAC exceeds the sum of deaths from any HAC. In the extreme, summing across excess deaths from a list of HACs could lead to implied death rates exceeding the actual overall in-hospital death rate.
Conversely, the few studies that estimate “any HAC”-related cost and mortality circumvent the competing risk concern by studying the effect of the presence of any HAC. While these types of studies garner headlines because of the magnitude of the overall concern, the tradeoff is that they provide little insight into potential points of intervention. This is because HACs have different underlying causes both in the healthcare system and from a biological standpoint.
The competing-risk and double-counting concerns may be addressed through better study designs that decompose (statistically) causes of death within a patient population. They also may benefit if studies assign weights to numerous potential causes of death. Well-constructed studies using approaches that account for numerous potential causes of death are not readily available in the literature. A lack of these studies represents a large knowledge gap that should be addressed.
Underlying Data Concerns
Many of the studies used for this meta-analysis conducted their analysis using AHRQ’s Healthcare Cost and Utilization Project National Inpatient Sample (HCUP-NIS). These administrative data have appeal in that they are well validated, centrally collected and curated, and collected using sampling frames that can generate national estimates.
However, these data are administrative billing data. They are not collected with the express purpose of studying HACs. As such, they are less reliable than clinical record data in distinguishing between present-on-admission and hospital-acquired conditions. They also lack clinical information used to define some HACs and may under-report HACs.
Additional research using electronic health record databases would substantially add to the literature on HAC incidence and consequences. The challenge with current studies that use these types of electronic health record data sources is questions about generalizability and scaling to national estimates. Emerging data sources that link clinical record data to billing data may facilitate substantially improved estimates of additional cost and excess mortality arising from HACs.
Another concern is that, although the QSRS Common Formats for Surveillance definitions use clinical information in addition to ICD codes for defining HACs, much of the literature still relies on definitions of HACs consisting primarily of ICD codes. Definitions relying on ICD codes can miss cases resulting in lower estimates of incidence and prevalence of HACs. Since the cases captured are true cases, additional cost and excess mortality estimates made using ICD-code-based definitions can likely be relied upon. In light of the changing definitions used in QSRS, future research should study HACs using these new definitions to better understand the incidence and consequences of HACs.
A final data concern is that many of the studies included in the meta-analysis did not directly report costs. Since a large proportion of the literature relies on national claims databases, they report on charges instead of costs. These charges are converted to costs using cost-to-charge ratios. To the extent that this methodology is used consistently over time, this would not drive observed differences in attributable cost from previous reports. Of course, a more accurate method would be to estimate costs directly using hospital records; however, the localized nature of such data opens the door to idiosyncratic center-specific and system-specific costing approaches, which is more difficult to address across studies than the issue of adjusting charges to arrive at an approximate measure of cost.
Opportunities for Future Research
There are a few distinct opportunities to improve the research on the attributable mortality and cost from HACs. First, some effort should focus on estimating the incidence of multiple HACs to address the concerns arising from double-counting and competing risks from other HACs. Second, there is room for improvement in understanding of the patient-level factors that raise the risk for HACs and using this understanding to construct more valid comparison populations. Both of these improvements combined with methods to weight likely causes of deaths would generate more valid estimates of the attributable mortality and costs from HACs than currently exist in the research literature.
Third, there are emerging sources of data that combine some form of administrative claims data with at least partial electronic medical record information across broader patient populations than have been studied using single-system data. Use of such data has the potential to leverage the strengths of the studies using the HCUP-NIS data (including standardization and wider generalizability) and those employing data from a single hospital or system (better identification of conditions and potentially the genesis of their onset). Estimates from such studies might provide a more rigorous estimate of the extent to which HACs increase mortality and costs.
Finally, the study undertaken here focused only on inpatient mortality and costs during the index hospitalization. Future work needs to account for the full effect of HACs on mortality and costs beyond the inpatient setting. Such research would provide insight into the extent to which HACs compress mortality and the full direct and indirect costs to the system. The challenge here would, of course, be access to data sources that facilitate these types of analyses.