Recognizing a growing need to justify expenditures on quality improvement programs that have a focus on avoidable mortality, our study sought to understand the association of mortality with overall LOS and ICU LOS. Specifically, we evaluated patients admitted to lower level of care settings, a population targeted by many mortality reduction programs including those using EWS5.
We found that patients who expired had significantly longer LOS, both overall and in the ICU. We also found that patients with urgent and emergency admissions, as well as surgical procedures, had longer LOS, and the latter two factors were also associated with longer ICU LOS. Complexity of correlations between mortality and LOS have been reported elsewhere, and causal relationships and interacting clinical considerations are not simple to delineate.32 In light of the range of factors associated with LOS measures, a strength of our analysis is its patient-level nature33 and use of the RI to ensure cohorts were of comparable acuity at the time of admission to account for the relation between baseline physiological acuity and mortality risk.1,34
Individual hospitals may find that the magnitude of the LOS difference between expired and non-expired patients in their populations is different from what we report here, and indeed additional matching criteria may be considered for deriving comparison cohorts. However, the magnitude of the difference we have found in both overall and ICU LOS for patients with comparable physiologic acuity on admission as measured by the RI patient condition score strongly suggests that a non-negligible LOS difference exists.
Hospital-based quality improvement initiatives face a growing need to demonstrate a robust business case justification and an argument for cost effectiveness. In particular, quality initiatives that require significant investment, such as operationalizing EWS for more timely detection and intervention on deteriorating patients, can have laudable quality goals yet struggle to translate even unambiguous success into a clear financial benefit. Hence, we posit that a rigorously established association between mortality and LOS may provide a means for organizations to extrapolate a reduction in both overall and ICU LOS from reductions in inpatient mortality. This would then allow readily available cost-per-day models to be brought to bear for the purpose of computing a financial ROI attributable to a reduction in avoidable mortality.
Furthermore, we note that the ability to infer a reduction in LOS from a reduction in avoidable mortality circumvents two associated LOS measurement challenges. First, while the LOS savings from a program focused on reducing avoidable mortality may be substantial within the specific target population, the volume of this target population (deaths and avoided deaths) is small (typically a few percent) relative to the total acute-care population, making it quantitatively challenging to discern an LOS change due to mortality reduction initiatives within total population-level LOS statistics. Second, variations in population acuity over time (which may be as simple as a more severe flu season from one year to another), variations in populations served (e.g. shifts in regional acute-care capacity, staffing and specialty service capacity, inter-hospital transfer patterns, ambulance diversion rates, etc.) and the fact that multiple contemporaneous operational changes are frequently made which have the potential, if not explicit aim, of influencing LOS, make causal attribution of changes to population-level LOS an uncertain proposition. Identifying and isolating initiatives likely to bear on mortality rates is a more manageable task, while possible acuity changes in a focused population can be accounted for using risk adjustment or acuity matching methodologies.
Our study has limitations. We lacked data on patient goals of care, comfort measure status, and advanced directives which, if included as covariates, could nuance the populations used in the matched cohorts. We did not include diagnosis in the analysis, and while the Rothman Index was built and validated as a diagnosis agnostic patient condition acuity score, we recognize that at a granular patient level the course of care, decisions related to treatment in the ICU, and overall LOS may have a dependence on diagnosis and comorbidities. Additionally, while we conducted the matching process separately for each health system, combining data in the final analysis may have obscured facility-level differences in LOS measures between cohorts.
We anticipate future work will entail expanding the methodological approach presented in this work to include segmenting populations by clinical diagnosis and/or comorbidities. Extending our analysis to evaluate the costs of care (or cost-per-day) specific to the patient sub-populations in our matched cohort groups may also elucidate important differences between generalized cost-per-day figures and the costs pertaining to our analysis populations (i.e. average costs are likely to be different in selected sub-population as a function of variations in acuity, treatment plans, goals of care etc.).